xref: /openbmc/linux/net/packet/af_packet.c (revision 15e47304)
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 
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 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
162 		int closing, int tx_ring);
163 
164 
165 #define V3_ALIGNMENT	(8)
166 
167 #define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
168 
169 #define BLK_PLUS_PRIV(sz_of_priv) \
170 	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
171 
172 #define PGV_FROM_VMALLOC 1
173 
174 #define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
175 #define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
176 #define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
177 #define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
178 #define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
179 #define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
180 #define BLOCK_PRIV(x)		((void *)((char *)(x) + BLOCK_O2PRIV(x)))
181 
182 struct packet_sock;
183 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
184 
185 static void *packet_previous_frame(struct packet_sock *po,
186 		struct packet_ring_buffer *rb,
187 		int status);
188 static void packet_increment_head(struct packet_ring_buffer *buff);
189 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
190 			struct tpacket_block_desc *);
191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
192 			struct packet_sock *);
193 static void prb_retire_current_block(struct tpacket_kbdq_core *,
194 		struct packet_sock *, unsigned int status);
195 static int prb_queue_frozen(struct tpacket_kbdq_core *);
196 static void prb_open_block(struct tpacket_kbdq_core *,
197 		struct tpacket_block_desc *);
198 static void prb_retire_rx_blk_timer_expired(unsigned long);
199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
200 static void prb_init_blk_timer(struct packet_sock *,
201 		struct tpacket_kbdq_core *,
202 		void (*func) (unsigned long));
203 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
204 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
205 		struct tpacket3_hdr *);
206 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
207 		struct tpacket3_hdr *);
208 static void packet_flush_mclist(struct sock *sk);
209 
210 struct packet_skb_cb {
211 	unsigned int origlen;
212 	union {
213 		struct sockaddr_pkt pkt;
214 		struct sockaddr_ll ll;
215 	} sa;
216 };
217 
218 #define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
219 
220 #define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
221 #define GET_PBLOCK_DESC(x, bid)	\
222 	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
223 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
224 	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
225 #define GET_NEXT_PRB_BLK_NUM(x) \
226 	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
227 	((x)->kactive_blk_num+1) : 0)
228 
229 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
230 static void __fanout_link(struct sock *sk, struct packet_sock *po);
231 
232 /* register_prot_hook must be invoked with the po->bind_lock held,
233  * or from a context in which asynchronous accesses to the packet
234  * socket is not possible (packet_create()).
235  */
236 static void register_prot_hook(struct sock *sk)
237 {
238 	struct packet_sock *po = pkt_sk(sk);
239 	if (!po->running) {
240 		if (po->fanout)
241 			__fanout_link(sk, po);
242 		else
243 			dev_add_pack(&po->prot_hook);
244 		sock_hold(sk);
245 		po->running = 1;
246 	}
247 }
248 
249 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
250  * held.   If the sync parameter is true, we will temporarily drop
251  * the po->bind_lock and do a synchronize_net to make sure no
252  * asynchronous packet processing paths still refer to the elements
253  * of po->prot_hook.  If the sync parameter is false, it is the
254  * callers responsibility to take care of this.
255  */
256 static void __unregister_prot_hook(struct sock *sk, bool sync)
257 {
258 	struct packet_sock *po = pkt_sk(sk);
259 
260 	po->running = 0;
261 	if (po->fanout)
262 		__fanout_unlink(sk, po);
263 	else
264 		__dev_remove_pack(&po->prot_hook);
265 	__sock_put(sk);
266 
267 	if (sync) {
268 		spin_unlock(&po->bind_lock);
269 		synchronize_net();
270 		spin_lock(&po->bind_lock);
271 	}
272 }
273 
274 static void unregister_prot_hook(struct sock *sk, bool sync)
275 {
276 	struct packet_sock *po = pkt_sk(sk);
277 
278 	if (po->running)
279 		__unregister_prot_hook(sk, sync);
280 }
281 
282 static inline __pure struct page *pgv_to_page(void *addr)
283 {
284 	if (is_vmalloc_addr(addr))
285 		return vmalloc_to_page(addr);
286 	return virt_to_page(addr);
287 }
288 
289 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
290 {
291 	union {
292 		struct tpacket_hdr *h1;
293 		struct tpacket2_hdr *h2;
294 		void *raw;
295 	} h;
296 
297 	h.raw = frame;
298 	switch (po->tp_version) {
299 	case TPACKET_V1:
300 		h.h1->tp_status = status;
301 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
302 		break;
303 	case TPACKET_V2:
304 		h.h2->tp_status = status;
305 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
306 		break;
307 	case TPACKET_V3:
308 	default:
309 		WARN(1, "TPACKET version not supported.\n");
310 		BUG();
311 	}
312 
313 	smp_wmb();
314 }
315 
316 static int __packet_get_status(struct packet_sock *po, void *frame)
317 {
318 	union {
319 		struct tpacket_hdr *h1;
320 		struct tpacket2_hdr *h2;
321 		void *raw;
322 	} h;
323 
324 	smp_rmb();
325 
326 	h.raw = frame;
327 	switch (po->tp_version) {
328 	case TPACKET_V1:
329 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
330 		return h.h1->tp_status;
331 	case TPACKET_V2:
332 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
333 		return h.h2->tp_status;
334 	case TPACKET_V3:
335 	default:
336 		WARN(1, "TPACKET version not supported.\n");
337 		BUG();
338 		return 0;
339 	}
340 }
341 
342 static void *packet_lookup_frame(struct packet_sock *po,
343 		struct packet_ring_buffer *rb,
344 		unsigned int position,
345 		int status)
346 {
347 	unsigned int pg_vec_pos, frame_offset;
348 	union {
349 		struct tpacket_hdr *h1;
350 		struct tpacket2_hdr *h2;
351 		void *raw;
352 	} h;
353 
354 	pg_vec_pos = position / rb->frames_per_block;
355 	frame_offset = position % rb->frames_per_block;
356 
357 	h.raw = rb->pg_vec[pg_vec_pos].buffer +
358 		(frame_offset * rb->frame_size);
359 
360 	if (status != __packet_get_status(po, h.raw))
361 		return NULL;
362 
363 	return h.raw;
364 }
365 
366 static void *packet_current_frame(struct packet_sock *po,
367 		struct packet_ring_buffer *rb,
368 		int status)
369 {
370 	return packet_lookup_frame(po, rb, rb->head, status);
371 }
372 
373 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
374 {
375 	del_timer_sync(&pkc->retire_blk_timer);
376 }
377 
378 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
379 		int tx_ring,
380 		struct sk_buff_head *rb_queue)
381 {
382 	struct tpacket_kbdq_core *pkc;
383 
384 	pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
385 
386 	spin_lock(&rb_queue->lock);
387 	pkc->delete_blk_timer = 1;
388 	spin_unlock(&rb_queue->lock);
389 
390 	prb_del_retire_blk_timer(pkc);
391 }
392 
393 static void prb_init_blk_timer(struct packet_sock *po,
394 		struct tpacket_kbdq_core *pkc,
395 		void (*func) (unsigned long))
396 {
397 	init_timer(&pkc->retire_blk_timer);
398 	pkc->retire_blk_timer.data = (long)po;
399 	pkc->retire_blk_timer.function = func;
400 	pkc->retire_blk_timer.expires = jiffies;
401 }
402 
403 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
404 {
405 	struct tpacket_kbdq_core *pkc;
406 
407 	if (tx_ring)
408 		BUG();
409 
410 	pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
411 	prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
412 }
413 
414 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
415 				int blk_size_in_bytes)
416 {
417 	struct net_device *dev;
418 	unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
419 	struct ethtool_cmd ecmd;
420 	int err;
421 	u32 speed;
422 
423 	rtnl_lock();
424 	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
425 	if (unlikely(!dev)) {
426 		rtnl_unlock();
427 		return DEFAULT_PRB_RETIRE_TOV;
428 	}
429 	err = __ethtool_get_settings(dev, &ecmd);
430 	speed = ethtool_cmd_speed(&ecmd);
431 	rtnl_unlock();
432 	if (!err) {
433 		/*
434 		 * If the link speed is so slow you don't really
435 		 * need to worry about perf anyways
436 		 */
437 		if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
438 			return DEFAULT_PRB_RETIRE_TOV;
439 		} else {
440 			msec = 1;
441 			div = speed / 1000;
442 		}
443 	}
444 
445 	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
446 
447 	if (div)
448 		mbits /= div;
449 
450 	tmo = mbits * msec;
451 
452 	if (div)
453 		return tmo+1;
454 	return tmo;
455 }
456 
457 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
458 			union tpacket_req_u *req_u)
459 {
460 	p1->feature_req_word = req_u->req3.tp_feature_req_word;
461 }
462 
463 static void init_prb_bdqc(struct packet_sock *po,
464 			struct packet_ring_buffer *rb,
465 			struct pgv *pg_vec,
466 			union tpacket_req_u *req_u, int tx_ring)
467 {
468 	struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
469 	struct tpacket_block_desc *pbd;
470 
471 	memset(p1, 0x0, sizeof(*p1));
472 
473 	p1->knxt_seq_num = 1;
474 	p1->pkbdq = pg_vec;
475 	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
476 	p1->pkblk_start	= pg_vec[0].buffer;
477 	p1->kblk_size = req_u->req3.tp_block_size;
478 	p1->knum_blocks	= req_u->req3.tp_block_nr;
479 	p1->hdrlen = po->tp_hdrlen;
480 	p1->version = po->tp_version;
481 	p1->last_kactive_blk_num = 0;
482 	po->stats_u.stats3.tp_freeze_q_cnt = 0;
483 	if (req_u->req3.tp_retire_blk_tov)
484 		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
485 	else
486 		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
487 						req_u->req3.tp_block_size);
488 	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
489 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
490 
491 	prb_init_ft_ops(p1, req_u);
492 	prb_setup_retire_blk_timer(po, tx_ring);
493 	prb_open_block(p1, pbd);
494 }
495 
496 /*  Do NOT update the last_blk_num first.
497  *  Assumes sk_buff_head lock is held.
498  */
499 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
500 {
501 	mod_timer(&pkc->retire_blk_timer,
502 			jiffies + pkc->tov_in_jiffies);
503 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
504 }
505 
506 /*
507  * Timer logic:
508  * 1) We refresh the timer only when we open a block.
509  *    By doing this we don't waste cycles refreshing the timer
510  *	  on packet-by-packet basis.
511  *
512  * With a 1MB block-size, on a 1Gbps line, it will take
513  * i) ~8 ms to fill a block + ii) memcpy etc.
514  * In this cut we are not accounting for the memcpy time.
515  *
516  * So, if the user sets the 'tmo' to 10ms then the timer
517  * will never fire while the block is still getting filled
518  * (which is what we want). However, the user could choose
519  * to close a block early and that's fine.
520  *
521  * But when the timer does fire, we check whether or not to refresh it.
522  * Since the tmo granularity is in msecs, it is not too expensive
523  * to refresh the timer, lets say every '8' msecs.
524  * Either the user can set the 'tmo' or we can derive it based on
525  * a) line-speed and b) block-size.
526  * prb_calc_retire_blk_tmo() calculates the tmo.
527  *
528  */
529 static void prb_retire_rx_blk_timer_expired(unsigned long data)
530 {
531 	struct packet_sock *po = (struct packet_sock *)data;
532 	struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
533 	unsigned int frozen;
534 	struct tpacket_block_desc *pbd;
535 
536 	spin_lock(&po->sk.sk_receive_queue.lock);
537 
538 	frozen = prb_queue_frozen(pkc);
539 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
540 
541 	if (unlikely(pkc->delete_blk_timer))
542 		goto out;
543 
544 	/* We only need to plug the race when the block is partially filled.
545 	 * tpacket_rcv:
546 	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
547 	 *		copy_bits() is in progress ...
548 	 *		timer fires on other cpu:
549 	 *		we can't retire the current block because copy_bits
550 	 *		is in progress.
551 	 *
552 	 */
553 	if (BLOCK_NUM_PKTS(pbd)) {
554 		while (atomic_read(&pkc->blk_fill_in_prog)) {
555 			/* Waiting for skb_copy_bits to finish... */
556 			cpu_relax();
557 		}
558 	}
559 
560 	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
561 		if (!frozen) {
562 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
563 			if (!prb_dispatch_next_block(pkc, po))
564 				goto refresh_timer;
565 			else
566 				goto out;
567 		} else {
568 			/* Case 1. Queue was frozen because user-space was
569 			 *	   lagging behind.
570 			 */
571 			if (prb_curr_blk_in_use(pkc, pbd)) {
572 				/*
573 				 * Ok, user-space is still behind.
574 				 * So just refresh the timer.
575 				 */
576 				goto refresh_timer;
577 			} else {
578 			       /* Case 2. queue was frozen,user-space caught up,
579 				* now the link went idle && the timer fired.
580 				* We don't have a block to close.So we open this
581 				* block and restart the timer.
582 				* opening a block thaws the queue,restarts timer
583 				* Thawing/timer-refresh is a side effect.
584 				*/
585 				prb_open_block(pkc, pbd);
586 				goto out;
587 			}
588 		}
589 	}
590 
591 refresh_timer:
592 	_prb_refresh_rx_retire_blk_timer(pkc);
593 
594 out:
595 	spin_unlock(&po->sk.sk_receive_queue.lock);
596 }
597 
598 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
599 		struct tpacket_block_desc *pbd1, __u32 status)
600 {
601 	/* Flush everything minus the block header */
602 
603 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
604 	u8 *start, *end;
605 
606 	start = (u8 *)pbd1;
607 
608 	/* Skip the block header(we know header WILL fit in 4K) */
609 	start += PAGE_SIZE;
610 
611 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
612 	for (; start < end; start += PAGE_SIZE)
613 		flush_dcache_page(pgv_to_page(start));
614 
615 	smp_wmb();
616 #endif
617 
618 	/* Now update the block status. */
619 
620 	BLOCK_STATUS(pbd1) = status;
621 
622 	/* Flush the block header */
623 
624 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
625 	start = (u8 *)pbd1;
626 	flush_dcache_page(pgv_to_page(start));
627 
628 	smp_wmb();
629 #endif
630 }
631 
632 /*
633  * Side effect:
634  *
635  * 1) flush the block
636  * 2) Increment active_blk_num
637  *
638  * Note:We DONT refresh the timer on purpose.
639  *	Because almost always the next block will be opened.
640  */
641 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
642 		struct tpacket_block_desc *pbd1,
643 		struct packet_sock *po, unsigned int stat)
644 {
645 	__u32 status = TP_STATUS_USER | stat;
646 
647 	struct tpacket3_hdr *last_pkt;
648 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
649 
650 	if (po->stats.tp_drops)
651 		status |= TP_STATUS_LOSING;
652 
653 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
654 	last_pkt->tp_next_offset = 0;
655 
656 	/* Get the ts of the last pkt */
657 	if (BLOCK_NUM_PKTS(pbd1)) {
658 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
659 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
660 	} else {
661 		/* Ok, we tmo'd - so get the current time */
662 		struct timespec ts;
663 		getnstimeofday(&ts);
664 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
665 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
666 	}
667 
668 	smp_wmb();
669 
670 	/* Flush the block */
671 	prb_flush_block(pkc1, pbd1, status);
672 
673 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
674 }
675 
676 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
677 {
678 	pkc->reset_pending_on_curr_blk = 0;
679 }
680 
681 /*
682  * Side effect of opening a block:
683  *
684  * 1) prb_queue is thawed.
685  * 2) retire_blk_timer is refreshed.
686  *
687  */
688 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
689 	struct tpacket_block_desc *pbd1)
690 {
691 	struct timespec ts;
692 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
693 
694 	smp_rmb();
695 
696 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
697 
698 		/* We could have just memset this but we will lose the
699 		 * flexibility of making the priv area sticky
700 		 */
701 		BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
702 		BLOCK_NUM_PKTS(pbd1) = 0;
703 		BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
704 		getnstimeofday(&ts);
705 		h1->ts_first_pkt.ts_sec = ts.tv_sec;
706 		h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
707 		pkc1->pkblk_start = (char *)pbd1;
708 		pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
709 		BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
710 		BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
711 		pbd1->version = pkc1->version;
712 		pkc1->prev = pkc1->nxt_offset;
713 		pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
714 		prb_thaw_queue(pkc1);
715 		_prb_refresh_rx_retire_blk_timer(pkc1);
716 
717 		smp_wmb();
718 
719 		return;
720 	}
721 
722 	WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
723 		pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
724 	dump_stack();
725 	BUG();
726 }
727 
728 /*
729  * Queue freeze logic:
730  * 1) Assume tp_block_nr = 8 blocks.
731  * 2) At time 't0', user opens Rx ring.
732  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
733  * 4) user-space is either sleeping or processing block '0'.
734  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
735  *    it will close block-7,loop around and try to fill block '0'.
736  *    call-flow:
737  *    __packet_lookup_frame_in_block
738  *      prb_retire_current_block()
739  *      prb_dispatch_next_block()
740  *        |->(BLOCK_STATUS == USER) evaluates to true
741  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
742  * 6) Now there are two cases:
743  *    6.1) Link goes idle right after the queue is frozen.
744  *         But remember, the last open_block() refreshed the timer.
745  *         When this timer expires,it will refresh itself so that we can
746  *         re-open block-0 in near future.
747  *    6.2) Link is busy and keeps on receiving packets. This is a simple
748  *         case and __packet_lookup_frame_in_block will check if block-0
749  *         is free and can now be re-used.
750  */
751 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
752 				  struct packet_sock *po)
753 {
754 	pkc->reset_pending_on_curr_blk = 1;
755 	po->stats_u.stats3.tp_freeze_q_cnt++;
756 }
757 
758 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
759 
760 /*
761  * If the next block is free then we will dispatch it
762  * and return a good offset.
763  * Else, we will freeze the queue.
764  * So, caller must check the return value.
765  */
766 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
767 		struct packet_sock *po)
768 {
769 	struct tpacket_block_desc *pbd;
770 
771 	smp_rmb();
772 
773 	/* 1. Get current block num */
774 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
775 
776 	/* 2. If this block is currently in_use then freeze the queue */
777 	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
778 		prb_freeze_queue(pkc, po);
779 		return NULL;
780 	}
781 
782 	/*
783 	 * 3.
784 	 * open this block and return the offset where the first packet
785 	 * needs to get stored.
786 	 */
787 	prb_open_block(pkc, pbd);
788 	return (void *)pkc->nxt_offset;
789 }
790 
791 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
792 		struct packet_sock *po, unsigned int status)
793 {
794 	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
795 
796 	/* retire/close the current block */
797 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
798 		/*
799 		 * Plug the case where copy_bits() is in progress on
800 		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
801 		 * have space to copy the pkt in the current block and
802 		 * called prb_retire_current_block()
803 		 *
804 		 * We don't need to worry about the TMO case because
805 		 * the timer-handler already handled this case.
806 		 */
807 		if (!(status & TP_STATUS_BLK_TMO)) {
808 			while (atomic_read(&pkc->blk_fill_in_prog)) {
809 				/* Waiting for skb_copy_bits to finish... */
810 				cpu_relax();
811 			}
812 		}
813 		prb_close_block(pkc, pbd, po, status);
814 		return;
815 	}
816 
817 	WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
818 	dump_stack();
819 	BUG();
820 }
821 
822 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
823 				      struct tpacket_block_desc *pbd)
824 {
825 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
826 }
827 
828 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
829 {
830 	return pkc->reset_pending_on_curr_blk;
831 }
832 
833 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
834 {
835 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
836 	atomic_dec(&pkc->blk_fill_in_prog);
837 }
838 
839 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
840 			struct tpacket3_hdr *ppd)
841 {
842 	ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
843 }
844 
845 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
846 			struct tpacket3_hdr *ppd)
847 {
848 	ppd->hv1.tp_rxhash = 0;
849 }
850 
851 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
852 			struct tpacket3_hdr *ppd)
853 {
854 	if (vlan_tx_tag_present(pkc->skb)) {
855 		ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
856 		ppd->tp_status = TP_STATUS_VLAN_VALID;
857 	} else {
858 		ppd->hv1.tp_vlan_tci = 0;
859 		ppd->tp_status = TP_STATUS_AVAILABLE;
860 	}
861 }
862 
863 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
864 			struct tpacket3_hdr *ppd)
865 {
866 	prb_fill_vlan_info(pkc, ppd);
867 
868 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
869 		prb_fill_rxhash(pkc, ppd);
870 	else
871 		prb_clear_rxhash(pkc, ppd);
872 }
873 
874 static void prb_fill_curr_block(char *curr,
875 				struct tpacket_kbdq_core *pkc,
876 				struct tpacket_block_desc *pbd,
877 				unsigned int len)
878 {
879 	struct tpacket3_hdr *ppd;
880 
881 	ppd  = (struct tpacket3_hdr *)curr;
882 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
883 	pkc->prev = curr;
884 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
885 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
886 	BLOCK_NUM_PKTS(pbd) += 1;
887 	atomic_inc(&pkc->blk_fill_in_prog);
888 	prb_run_all_ft_ops(pkc, ppd);
889 }
890 
891 /* Assumes caller has the sk->rx_queue.lock */
892 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
893 					    struct sk_buff *skb,
894 						int status,
895 					    unsigned int len
896 					    )
897 {
898 	struct tpacket_kbdq_core *pkc;
899 	struct tpacket_block_desc *pbd;
900 	char *curr, *end;
901 
902 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
903 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
904 
905 	/* Queue is frozen when user space is lagging behind */
906 	if (prb_queue_frozen(pkc)) {
907 		/*
908 		 * Check if that last block which caused the queue to freeze,
909 		 * is still in_use by user-space.
910 		 */
911 		if (prb_curr_blk_in_use(pkc, pbd)) {
912 			/* Can't record this packet */
913 			return NULL;
914 		} else {
915 			/*
916 			 * Ok, the block was released by user-space.
917 			 * Now let's open that block.
918 			 * opening a block also thaws the queue.
919 			 * Thawing is a side effect.
920 			 */
921 			prb_open_block(pkc, pbd);
922 		}
923 	}
924 
925 	smp_mb();
926 	curr = pkc->nxt_offset;
927 	pkc->skb = skb;
928 	end = (char *)pbd + pkc->kblk_size;
929 
930 	/* first try the current block */
931 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
932 		prb_fill_curr_block(curr, pkc, pbd, len);
933 		return (void *)curr;
934 	}
935 
936 	/* Ok, close the current block */
937 	prb_retire_current_block(pkc, po, 0);
938 
939 	/* Now, try to dispatch the next block */
940 	curr = (char *)prb_dispatch_next_block(pkc, po);
941 	if (curr) {
942 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
943 		prb_fill_curr_block(curr, pkc, pbd, len);
944 		return (void *)curr;
945 	}
946 
947 	/*
948 	 * No free blocks are available.user_space hasn't caught up yet.
949 	 * Queue was just frozen and now this packet will get dropped.
950 	 */
951 	return NULL;
952 }
953 
954 static void *packet_current_rx_frame(struct packet_sock *po,
955 					    struct sk_buff *skb,
956 					    int status, unsigned int len)
957 {
958 	char *curr = NULL;
959 	switch (po->tp_version) {
960 	case TPACKET_V1:
961 	case TPACKET_V2:
962 		curr = packet_lookup_frame(po, &po->rx_ring,
963 					po->rx_ring.head, status);
964 		return curr;
965 	case TPACKET_V3:
966 		return __packet_lookup_frame_in_block(po, skb, status, len);
967 	default:
968 		WARN(1, "TPACKET version not supported\n");
969 		BUG();
970 		return NULL;
971 	}
972 }
973 
974 static void *prb_lookup_block(struct packet_sock *po,
975 				     struct packet_ring_buffer *rb,
976 				     unsigned int previous,
977 				     int status)
978 {
979 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
980 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
981 
982 	if (status != BLOCK_STATUS(pbd))
983 		return NULL;
984 	return pbd;
985 }
986 
987 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
988 {
989 	unsigned int prev;
990 	if (rb->prb_bdqc.kactive_blk_num)
991 		prev = rb->prb_bdqc.kactive_blk_num-1;
992 	else
993 		prev = rb->prb_bdqc.knum_blocks-1;
994 	return prev;
995 }
996 
997 /* Assumes caller has held the rx_queue.lock */
998 static void *__prb_previous_block(struct packet_sock *po,
999 					 struct packet_ring_buffer *rb,
1000 					 int status)
1001 {
1002 	unsigned int previous = prb_previous_blk_num(rb);
1003 	return prb_lookup_block(po, rb, previous, status);
1004 }
1005 
1006 static void *packet_previous_rx_frame(struct packet_sock *po,
1007 					     struct packet_ring_buffer *rb,
1008 					     int status)
1009 {
1010 	if (po->tp_version <= TPACKET_V2)
1011 		return packet_previous_frame(po, rb, status);
1012 
1013 	return __prb_previous_block(po, rb, status);
1014 }
1015 
1016 static void packet_increment_rx_head(struct packet_sock *po,
1017 					    struct packet_ring_buffer *rb)
1018 {
1019 	switch (po->tp_version) {
1020 	case TPACKET_V1:
1021 	case TPACKET_V2:
1022 		return packet_increment_head(rb);
1023 	case TPACKET_V3:
1024 	default:
1025 		WARN(1, "TPACKET version not supported.\n");
1026 		BUG();
1027 		return;
1028 	}
1029 }
1030 
1031 static void *packet_previous_frame(struct packet_sock *po,
1032 		struct packet_ring_buffer *rb,
1033 		int status)
1034 {
1035 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1036 	return packet_lookup_frame(po, rb, previous, status);
1037 }
1038 
1039 static void packet_increment_head(struct packet_ring_buffer *buff)
1040 {
1041 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1042 }
1043 
1044 static void packet_sock_destruct(struct sock *sk)
1045 {
1046 	skb_queue_purge(&sk->sk_error_queue);
1047 
1048 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1049 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1050 
1051 	if (!sock_flag(sk, SOCK_DEAD)) {
1052 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1053 		return;
1054 	}
1055 
1056 	sk_refcnt_debug_dec(sk);
1057 }
1058 
1059 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1060 {
1061 	int x = atomic_read(&f->rr_cur) + 1;
1062 
1063 	if (x >= num)
1064 		x = 0;
1065 
1066 	return x;
1067 }
1068 
1069 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1070 {
1071 	u32 idx, hash = skb->rxhash;
1072 
1073 	idx = ((u64)hash * num) >> 32;
1074 
1075 	return f->arr[idx];
1076 }
1077 
1078 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1079 {
1080 	int cur, old;
1081 
1082 	cur = atomic_read(&f->rr_cur);
1083 	while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1084 				     fanout_rr_next(f, num))) != cur)
1085 		cur = old;
1086 	return f->arr[cur];
1087 }
1088 
1089 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1090 {
1091 	unsigned int cpu = smp_processor_id();
1092 
1093 	return f->arr[cpu % num];
1094 }
1095 
1096 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1097 			     struct packet_type *pt, struct net_device *orig_dev)
1098 {
1099 	struct packet_fanout *f = pt->af_packet_priv;
1100 	unsigned int num = f->num_members;
1101 	struct packet_sock *po;
1102 	struct sock *sk;
1103 
1104 	if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1105 	    !num) {
1106 		kfree_skb(skb);
1107 		return 0;
1108 	}
1109 
1110 	switch (f->type) {
1111 	case PACKET_FANOUT_HASH:
1112 	default:
1113 		if (f->defrag) {
1114 			skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1115 			if (!skb)
1116 				return 0;
1117 		}
1118 		skb_get_rxhash(skb);
1119 		sk = fanout_demux_hash(f, skb, num);
1120 		break;
1121 	case PACKET_FANOUT_LB:
1122 		sk = fanout_demux_lb(f, skb, num);
1123 		break;
1124 	case PACKET_FANOUT_CPU:
1125 		sk = fanout_demux_cpu(f, skb, num);
1126 		break;
1127 	}
1128 
1129 	po = pkt_sk(sk);
1130 
1131 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1132 }
1133 
1134 DEFINE_MUTEX(fanout_mutex);
1135 EXPORT_SYMBOL_GPL(fanout_mutex);
1136 static LIST_HEAD(fanout_list);
1137 
1138 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1139 {
1140 	struct packet_fanout *f = po->fanout;
1141 
1142 	spin_lock(&f->lock);
1143 	f->arr[f->num_members] = sk;
1144 	smp_wmb();
1145 	f->num_members++;
1146 	spin_unlock(&f->lock);
1147 }
1148 
1149 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1150 {
1151 	struct packet_fanout *f = po->fanout;
1152 	int i;
1153 
1154 	spin_lock(&f->lock);
1155 	for (i = 0; i < f->num_members; i++) {
1156 		if (f->arr[i] == sk)
1157 			break;
1158 	}
1159 	BUG_ON(i >= f->num_members);
1160 	f->arr[i] = f->arr[f->num_members - 1];
1161 	f->num_members--;
1162 	spin_unlock(&f->lock);
1163 }
1164 
1165 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1166 {
1167 	if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1168 		return true;
1169 
1170 	return false;
1171 }
1172 
1173 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1174 {
1175 	struct packet_sock *po = pkt_sk(sk);
1176 	struct packet_fanout *f, *match;
1177 	u8 type = type_flags & 0xff;
1178 	u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1179 	int err;
1180 
1181 	switch (type) {
1182 	case PACKET_FANOUT_HASH:
1183 	case PACKET_FANOUT_LB:
1184 	case PACKET_FANOUT_CPU:
1185 		break;
1186 	default:
1187 		return -EINVAL;
1188 	}
1189 
1190 	if (!po->running)
1191 		return -EINVAL;
1192 
1193 	if (po->fanout)
1194 		return -EALREADY;
1195 
1196 	mutex_lock(&fanout_mutex);
1197 	match = NULL;
1198 	list_for_each_entry(f, &fanout_list, list) {
1199 		if (f->id == id &&
1200 		    read_pnet(&f->net) == sock_net(sk)) {
1201 			match = f;
1202 			break;
1203 		}
1204 	}
1205 	err = -EINVAL;
1206 	if (match && match->defrag != defrag)
1207 		goto out;
1208 	if (!match) {
1209 		err = -ENOMEM;
1210 		match = kzalloc(sizeof(*match), GFP_KERNEL);
1211 		if (!match)
1212 			goto out;
1213 		write_pnet(&match->net, sock_net(sk));
1214 		match->id = id;
1215 		match->type = type;
1216 		match->defrag = defrag;
1217 		atomic_set(&match->rr_cur, 0);
1218 		INIT_LIST_HEAD(&match->list);
1219 		spin_lock_init(&match->lock);
1220 		atomic_set(&match->sk_ref, 0);
1221 		match->prot_hook.type = po->prot_hook.type;
1222 		match->prot_hook.dev = po->prot_hook.dev;
1223 		match->prot_hook.func = packet_rcv_fanout;
1224 		match->prot_hook.af_packet_priv = match;
1225 		match->prot_hook.id_match = match_fanout_group;
1226 		dev_add_pack(&match->prot_hook);
1227 		list_add(&match->list, &fanout_list);
1228 	}
1229 	err = -EINVAL;
1230 	if (match->type == type &&
1231 	    match->prot_hook.type == po->prot_hook.type &&
1232 	    match->prot_hook.dev == po->prot_hook.dev) {
1233 		err = -ENOSPC;
1234 		if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1235 			__dev_remove_pack(&po->prot_hook);
1236 			po->fanout = match;
1237 			atomic_inc(&match->sk_ref);
1238 			__fanout_link(sk, po);
1239 			err = 0;
1240 		}
1241 	}
1242 out:
1243 	mutex_unlock(&fanout_mutex);
1244 	return err;
1245 }
1246 
1247 static void fanout_release(struct sock *sk)
1248 {
1249 	struct packet_sock *po = pkt_sk(sk);
1250 	struct packet_fanout *f;
1251 
1252 	f = po->fanout;
1253 	if (!f)
1254 		return;
1255 
1256 	mutex_lock(&fanout_mutex);
1257 	po->fanout = NULL;
1258 
1259 	if (atomic_dec_and_test(&f->sk_ref)) {
1260 		list_del(&f->list);
1261 		dev_remove_pack(&f->prot_hook);
1262 		kfree(f);
1263 	}
1264 	mutex_unlock(&fanout_mutex);
1265 }
1266 
1267 static const struct proto_ops packet_ops;
1268 
1269 static const struct proto_ops packet_ops_spkt;
1270 
1271 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1272 			   struct packet_type *pt, struct net_device *orig_dev)
1273 {
1274 	struct sock *sk;
1275 	struct sockaddr_pkt *spkt;
1276 
1277 	/*
1278 	 *	When we registered the protocol we saved the socket in the data
1279 	 *	field for just this event.
1280 	 */
1281 
1282 	sk = pt->af_packet_priv;
1283 
1284 	/*
1285 	 *	Yank back the headers [hope the device set this
1286 	 *	right or kerboom...]
1287 	 *
1288 	 *	Incoming packets have ll header pulled,
1289 	 *	push it back.
1290 	 *
1291 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1292 	 *	so that this procedure is noop.
1293 	 */
1294 
1295 	if (skb->pkt_type == PACKET_LOOPBACK)
1296 		goto out;
1297 
1298 	if (!net_eq(dev_net(dev), sock_net(sk)))
1299 		goto out;
1300 
1301 	skb = skb_share_check(skb, GFP_ATOMIC);
1302 	if (skb == NULL)
1303 		goto oom;
1304 
1305 	/* drop any routing info */
1306 	skb_dst_drop(skb);
1307 
1308 	/* drop conntrack reference */
1309 	nf_reset(skb);
1310 
1311 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1312 
1313 	skb_push(skb, skb->data - skb_mac_header(skb));
1314 
1315 	/*
1316 	 *	The SOCK_PACKET socket receives _all_ frames.
1317 	 */
1318 
1319 	spkt->spkt_family = dev->type;
1320 	strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1321 	spkt->spkt_protocol = skb->protocol;
1322 
1323 	/*
1324 	 *	Charge the memory to the socket. This is done specifically
1325 	 *	to prevent sockets using all the memory up.
1326 	 */
1327 
1328 	if (sock_queue_rcv_skb(sk, skb) == 0)
1329 		return 0;
1330 
1331 out:
1332 	kfree_skb(skb);
1333 oom:
1334 	return 0;
1335 }
1336 
1337 
1338 /*
1339  *	Output a raw packet to a device layer. This bypasses all the other
1340  *	protocol layers and you must therefore supply it with a complete frame
1341  */
1342 
1343 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1344 			       struct msghdr *msg, size_t len)
1345 {
1346 	struct sock *sk = sock->sk;
1347 	struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1348 	struct sk_buff *skb = NULL;
1349 	struct net_device *dev;
1350 	__be16 proto = 0;
1351 	int err;
1352 	int extra_len = 0;
1353 
1354 	/*
1355 	 *	Get and verify the address.
1356 	 */
1357 
1358 	if (saddr) {
1359 		if (msg->msg_namelen < sizeof(struct sockaddr))
1360 			return -EINVAL;
1361 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1362 			proto = saddr->spkt_protocol;
1363 	} else
1364 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1365 
1366 	/*
1367 	 *	Find the device first to size check it
1368 	 */
1369 
1370 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1371 retry:
1372 	rcu_read_lock();
1373 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1374 	err = -ENODEV;
1375 	if (dev == NULL)
1376 		goto out_unlock;
1377 
1378 	err = -ENETDOWN;
1379 	if (!(dev->flags & IFF_UP))
1380 		goto out_unlock;
1381 
1382 	/*
1383 	 * You may not queue a frame bigger than the mtu. This is the lowest level
1384 	 * raw protocol and you must do your own fragmentation at this level.
1385 	 */
1386 
1387 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1388 		if (!netif_supports_nofcs(dev)) {
1389 			err = -EPROTONOSUPPORT;
1390 			goto out_unlock;
1391 		}
1392 		extra_len = 4; /* We're doing our own CRC */
1393 	}
1394 
1395 	err = -EMSGSIZE;
1396 	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1397 		goto out_unlock;
1398 
1399 	if (!skb) {
1400 		size_t reserved = LL_RESERVED_SPACE(dev);
1401 		int tlen = dev->needed_tailroom;
1402 		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1403 
1404 		rcu_read_unlock();
1405 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1406 		if (skb == NULL)
1407 			return -ENOBUFS;
1408 		/* FIXME: Save some space for broken drivers that write a hard
1409 		 * header at transmission time by themselves. PPP is the notable
1410 		 * one here. This should really be fixed at the driver level.
1411 		 */
1412 		skb_reserve(skb, reserved);
1413 		skb_reset_network_header(skb);
1414 
1415 		/* Try to align data part correctly */
1416 		if (hhlen) {
1417 			skb->data -= hhlen;
1418 			skb->tail -= hhlen;
1419 			if (len < hhlen)
1420 				skb_reset_network_header(skb);
1421 		}
1422 		err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1423 		if (err)
1424 			goto out_free;
1425 		goto retry;
1426 	}
1427 
1428 	if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1429 		/* Earlier code assumed this would be a VLAN pkt,
1430 		 * double-check this now that we have the actual
1431 		 * packet in hand.
1432 		 */
1433 		struct ethhdr *ehdr;
1434 		skb_reset_mac_header(skb);
1435 		ehdr = eth_hdr(skb);
1436 		if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1437 			err = -EMSGSIZE;
1438 			goto out_unlock;
1439 		}
1440 	}
1441 
1442 	skb->protocol = proto;
1443 	skb->dev = dev;
1444 	skb->priority = sk->sk_priority;
1445 	skb->mark = sk->sk_mark;
1446 	err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1447 	if (err < 0)
1448 		goto out_unlock;
1449 
1450 	if (unlikely(extra_len == 4))
1451 		skb->no_fcs = 1;
1452 
1453 	dev_queue_xmit(skb);
1454 	rcu_read_unlock();
1455 	return len;
1456 
1457 out_unlock:
1458 	rcu_read_unlock();
1459 out_free:
1460 	kfree_skb(skb);
1461 	return err;
1462 }
1463 
1464 static unsigned int run_filter(const struct sk_buff *skb,
1465 				      const struct sock *sk,
1466 				      unsigned int res)
1467 {
1468 	struct sk_filter *filter;
1469 
1470 	rcu_read_lock();
1471 	filter = rcu_dereference(sk->sk_filter);
1472 	if (filter != NULL)
1473 		res = SK_RUN_FILTER(filter, skb);
1474 	rcu_read_unlock();
1475 
1476 	return res;
1477 }
1478 
1479 /*
1480  * This function makes lazy skb cloning in hope that most of packets
1481  * are discarded by BPF.
1482  *
1483  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1484  * and skb->cb are mangled. It works because (and until) packets
1485  * falling here are owned by current CPU. Output packets are cloned
1486  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1487  * sequencially, so that if we return skb to original state on exit,
1488  * we will not harm anyone.
1489  */
1490 
1491 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1492 		      struct packet_type *pt, struct net_device *orig_dev)
1493 {
1494 	struct sock *sk;
1495 	struct sockaddr_ll *sll;
1496 	struct packet_sock *po;
1497 	u8 *skb_head = skb->data;
1498 	int skb_len = skb->len;
1499 	unsigned int snaplen, res;
1500 
1501 	if (skb->pkt_type == PACKET_LOOPBACK)
1502 		goto drop;
1503 
1504 	sk = pt->af_packet_priv;
1505 	po = pkt_sk(sk);
1506 
1507 	if (!net_eq(dev_net(dev), sock_net(sk)))
1508 		goto drop;
1509 
1510 	skb->dev = dev;
1511 
1512 	if (dev->header_ops) {
1513 		/* The device has an explicit notion of ll header,
1514 		 * exported to higher levels.
1515 		 *
1516 		 * Otherwise, the device hides details of its frame
1517 		 * structure, so that corresponding packet head is
1518 		 * never delivered to user.
1519 		 */
1520 		if (sk->sk_type != SOCK_DGRAM)
1521 			skb_push(skb, skb->data - skb_mac_header(skb));
1522 		else if (skb->pkt_type == PACKET_OUTGOING) {
1523 			/* Special case: outgoing packets have ll header at head */
1524 			skb_pull(skb, skb_network_offset(skb));
1525 		}
1526 	}
1527 
1528 	snaplen = skb->len;
1529 
1530 	res = run_filter(skb, sk, snaplen);
1531 	if (!res)
1532 		goto drop_n_restore;
1533 	if (snaplen > res)
1534 		snaplen = res;
1535 
1536 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1537 		goto drop_n_acct;
1538 
1539 	if (skb_shared(skb)) {
1540 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1541 		if (nskb == NULL)
1542 			goto drop_n_acct;
1543 
1544 		if (skb_head != skb->data) {
1545 			skb->data = skb_head;
1546 			skb->len = skb_len;
1547 		}
1548 		consume_skb(skb);
1549 		skb = nskb;
1550 	}
1551 
1552 	BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1553 		     sizeof(skb->cb));
1554 
1555 	sll = &PACKET_SKB_CB(skb)->sa.ll;
1556 	sll->sll_family = AF_PACKET;
1557 	sll->sll_hatype = dev->type;
1558 	sll->sll_protocol = skb->protocol;
1559 	sll->sll_pkttype = skb->pkt_type;
1560 	if (unlikely(po->origdev))
1561 		sll->sll_ifindex = orig_dev->ifindex;
1562 	else
1563 		sll->sll_ifindex = dev->ifindex;
1564 
1565 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1566 
1567 	PACKET_SKB_CB(skb)->origlen = skb->len;
1568 
1569 	if (pskb_trim(skb, snaplen))
1570 		goto drop_n_acct;
1571 
1572 	skb_set_owner_r(skb, sk);
1573 	skb->dev = NULL;
1574 	skb_dst_drop(skb);
1575 
1576 	/* drop conntrack reference */
1577 	nf_reset(skb);
1578 
1579 	spin_lock(&sk->sk_receive_queue.lock);
1580 	po->stats.tp_packets++;
1581 	skb->dropcount = atomic_read(&sk->sk_drops);
1582 	__skb_queue_tail(&sk->sk_receive_queue, skb);
1583 	spin_unlock(&sk->sk_receive_queue.lock);
1584 	sk->sk_data_ready(sk, skb->len);
1585 	return 0;
1586 
1587 drop_n_acct:
1588 	spin_lock(&sk->sk_receive_queue.lock);
1589 	po->stats.tp_drops++;
1590 	atomic_inc(&sk->sk_drops);
1591 	spin_unlock(&sk->sk_receive_queue.lock);
1592 
1593 drop_n_restore:
1594 	if (skb_head != skb->data && skb_shared(skb)) {
1595 		skb->data = skb_head;
1596 		skb->len = skb_len;
1597 	}
1598 drop:
1599 	consume_skb(skb);
1600 	return 0;
1601 }
1602 
1603 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1604 		       struct packet_type *pt, struct net_device *orig_dev)
1605 {
1606 	struct sock *sk;
1607 	struct packet_sock *po;
1608 	struct sockaddr_ll *sll;
1609 	union {
1610 		struct tpacket_hdr *h1;
1611 		struct tpacket2_hdr *h2;
1612 		struct tpacket3_hdr *h3;
1613 		void *raw;
1614 	} h;
1615 	u8 *skb_head = skb->data;
1616 	int skb_len = skb->len;
1617 	unsigned int snaplen, res;
1618 	unsigned long status = TP_STATUS_USER;
1619 	unsigned short macoff, netoff, hdrlen;
1620 	struct sk_buff *copy_skb = NULL;
1621 	struct timeval tv;
1622 	struct timespec ts;
1623 	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1624 
1625 	if (skb->pkt_type == PACKET_LOOPBACK)
1626 		goto drop;
1627 
1628 	sk = pt->af_packet_priv;
1629 	po = pkt_sk(sk);
1630 
1631 	if (!net_eq(dev_net(dev), sock_net(sk)))
1632 		goto drop;
1633 
1634 	if (dev->header_ops) {
1635 		if (sk->sk_type != SOCK_DGRAM)
1636 			skb_push(skb, skb->data - skb_mac_header(skb));
1637 		else if (skb->pkt_type == PACKET_OUTGOING) {
1638 			/* Special case: outgoing packets have ll header at head */
1639 			skb_pull(skb, skb_network_offset(skb));
1640 		}
1641 	}
1642 
1643 	if (skb->ip_summed == CHECKSUM_PARTIAL)
1644 		status |= TP_STATUS_CSUMNOTREADY;
1645 
1646 	snaplen = skb->len;
1647 
1648 	res = run_filter(skb, sk, snaplen);
1649 	if (!res)
1650 		goto drop_n_restore;
1651 	if (snaplen > res)
1652 		snaplen = res;
1653 
1654 	if (sk->sk_type == SOCK_DGRAM) {
1655 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1656 				  po->tp_reserve;
1657 	} else {
1658 		unsigned int maclen = skb_network_offset(skb);
1659 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
1660 				       (maclen < 16 ? 16 : maclen)) +
1661 			po->tp_reserve;
1662 		macoff = netoff - maclen;
1663 	}
1664 	if (po->tp_version <= TPACKET_V2) {
1665 		if (macoff + snaplen > po->rx_ring.frame_size) {
1666 			if (po->copy_thresh &&
1667 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1668 				if (skb_shared(skb)) {
1669 					copy_skb = skb_clone(skb, GFP_ATOMIC);
1670 				} else {
1671 					copy_skb = skb_get(skb);
1672 					skb_head = skb->data;
1673 				}
1674 				if (copy_skb)
1675 					skb_set_owner_r(copy_skb, sk);
1676 			}
1677 			snaplen = po->rx_ring.frame_size - macoff;
1678 			if ((int)snaplen < 0)
1679 				snaplen = 0;
1680 		}
1681 	}
1682 	spin_lock(&sk->sk_receive_queue.lock);
1683 	h.raw = packet_current_rx_frame(po, skb,
1684 					TP_STATUS_KERNEL, (macoff+snaplen));
1685 	if (!h.raw)
1686 		goto ring_is_full;
1687 	if (po->tp_version <= TPACKET_V2) {
1688 		packet_increment_rx_head(po, &po->rx_ring);
1689 	/*
1690 	 * LOSING will be reported till you read the stats,
1691 	 * because it's COR - Clear On Read.
1692 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1693 	 * at packet level.
1694 	 */
1695 		if (po->stats.tp_drops)
1696 			status |= TP_STATUS_LOSING;
1697 	}
1698 	po->stats.tp_packets++;
1699 	if (copy_skb) {
1700 		status |= TP_STATUS_COPY;
1701 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1702 	}
1703 	spin_unlock(&sk->sk_receive_queue.lock);
1704 
1705 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1706 
1707 	switch (po->tp_version) {
1708 	case TPACKET_V1:
1709 		h.h1->tp_len = skb->len;
1710 		h.h1->tp_snaplen = snaplen;
1711 		h.h1->tp_mac = macoff;
1712 		h.h1->tp_net = netoff;
1713 		if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1714 				&& shhwtstamps->syststamp.tv64)
1715 			tv = ktime_to_timeval(shhwtstamps->syststamp);
1716 		else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1717 				&& shhwtstamps->hwtstamp.tv64)
1718 			tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1719 		else if (skb->tstamp.tv64)
1720 			tv = ktime_to_timeval(skb->tstamp);
1721 		else
1722 			do_gettimeofday(&tv);
1723 		h.h1->tp_sec = tv.tv_sec;
1724 		h.h1->tp_usec = tv.tv_usec;
1725 		hdrlen = sizeof(*h.h1);
1726 		break;
1727 	case TPACKET_V2:
1728 		h.h2->tp_len = skb->len;
1729 		h.h2->tp_snaplen = snaplen;
1730 		h.h2->tp_mac = macoff;
1731 		h.h2->tp_net = netoff;
1732 		if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1733 				&& shhwtstamps->syststamp.tv64)
1734 			ts = ktime_to_timespec(shhwtstamps->syststamp);
1735 		else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1736 				&& shhwtstamps->hwtstamp.tv64)
1737 			ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1738 		else if (skb->tstamp.tv64)
1739 			ts = ktime_to_timespec(skb->tstamp);
1740 		else
1741 			getnstimeofday(&ts);
1742 		h.h2->tp_sec = ts.tv_sec;
1743 		h.h2->tp_nsec = ts.tv_nsec;
1744 		if (vlan_tx_tag_present(skb)) {
1745 			h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1746 			status |= TP_STATUS_VLAN_VALID;
1747 		} else {
1748 			h.h2->tp_vlan_tci = 0;
1749 		}
1750 		h.h2->tp_padding = 0;
1751 		hdrlen = sizeof(*h.h2);
1752 		break;
1753 	case TPACKET_V3:
1754 		/* tp_nxt_offset,vlan are already populated above.
1755 		 * So DONT clear those fields here
1756 		 */
1757 		h.h3->tp_status |= status;
1758 		h.h3->tp_len = skb->len;
1759 		h.h3->tp_snaplen = snaplen;
1760 		h.h3->tp_mac = macoff;
1761 		h.h3->tp_net = netoff;
1762 		if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1763 				&& shhwtstamps->syststamp.tv64)
1764 			ts = ktime_to_timespec(shhwtstamps->syststamp);
1765 		else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1766 				&& shhwtstamps->hwtstamp.tv64)
1767 			ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1768 		else if (skb->tstamp.tv64)
1769 			ts = ktime_to_timespec(skb->tstamp);
1770 		else
1771 			getnstimeofday(&ts);
1772 		h.h3->tp_sec  = ts.tv_sec;
1773 		h.h3->tp_nsec = ts.tv_nsec;
1774 		hdrlen = sizeof(*h.h3);
1775 		break;
1776 	default:
1777 		BUG();
1778 	}
1779 
1780 	sll = h.raw + TPACKET_ALIGN(hdrlen);
1781 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1782 	sll->sll_family = AF_PACKET;
1783 	sll->sll_hatype = dev->type;
1784 	sll->sll_protocol = skb->protocol;
1785 	sll->sll_pkttype = skb->pkt_type;
1786 	if (unlikely(po->origdev))
1787 		sll->sll_ifindex = orig_dev->ifindex;
1788 	else
1789 		sll->sll_ifindex = dev->ifindex;
1790 
1791 	smp_mb();
1792 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1793 	{
1794 		u8 *start, *end;
1795 
1796 		if (po->tp_version <= TPACKET_V2) {
1797 			end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1798 				+ macoff + snaplen);
1799 			for (start = h.raw; start < end; start += PAGE_SIZE)
1800 				flush_dcache_page(pgv_to_page(start));
1801 		}
1802 		smp_wmb();
1803 	}
1804 #endif
1805 	if (po->tp_version <= TPACKET_V2)
1806 		__packet_set_status(po, h.raw, status);
1807 	else
1808 		prb_clear_blk_fill_status(&po->rx_ring);
1809 
1810 	sk->sk_data_ready(sk, 0);
1811 
1812 drop_n_restore:
1813 	if (skb_head != skb->data && skb_shared(skb)) {
1814 		skb->data = skb_head;
1815 		skb->len = skb_len;
1816 	}
1817 drop:
1818 	kfree_skb(skb);
1819 	return 0;
1820 
1821 ring_is_full:
1822 	po->stats.tp_drops++;
1823 	spin_unlock(&sk->sk_receive_queue.lock);
1824 
1825 	sk->sk_data_ready(sk, 0);
1826 	kfree_skb(copy_skb);
1827 	goto drop_n_restore;
1828 }
1829 
1830 static void tpacket_destruct_skb(struct sk_buff *skb)
1831 {
1832 	struct packet_sock *po = pkt_sk(skb->sk);
1833 	void *ph;
1834 
1835 	if (likely(po->tx_ring.pg_vec)) {
1836 		ph = skb_shinfo(skb)->destructor_arg;
1837 		BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1838 		atomic_dec(&po->tx_ring.pending);
1839 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1840 	}
1841 
1842 	sock_wfree(skb);
1843 }
1844 
1845 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1846 		void *frame, struct net_device *dev, int size_max,
1847 		__be16 proto, unsigned char *addr, int hlen)
1848 {
1849 	union {
1850 		struct tpacket_hdr *h1;
1851 		struct tpacket2_hdr *h2;
1852 		void *raw;
1853 	} ph;
1854 	int to_write, offset, len, tp_len, nr_frags, len_max;
1855 	struct socket *sock = po->sk.sk_socket;
1856 	struct page *page;
1857 	void *data;
1858 	int err;
1859 
1860 	ph.raw = frame;
1861 
1862 	skb->protocol = proto;
1863 	skb->dev = dev;
1864 	skb->priority = po->sk.sk_priority;
1865 	skb->mark = po->sk.sk_mark;
1866 	skb_shinfo(skb)->destructor_arg = ph.raw;
1867 
1868 	switch (po->tp_version) {
1869 	case TPACKET_V2:
1870 		tp_len = ph.h2->tp_len;
1871 		break;
1872 	default:
1873 		tp_len = ph.h1->tp_len;
1874 		break;
1875 	}
1876 	if (unlikely(tp_len > size_max)) {
1877 		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1878 		return -EMSGSIZE;
1879 	}
1880 
1881 	skb_reserve(skb, hlen);
1882 	skb_reset_network_header(skb);
1883 
1884 	data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1885 	to_write = tp_len;
1886 
1887 	if (sock->type == SOCK_DGRAM) {
1888 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
1889 				NULL, tp_len);
1890 		if (unlikely(err < 0))
1891 			return -EINVAL;
1892 	} else if (dev->hard_header_len) {
1893 		/* net device doesn't like empty head */
1894 		if (unlikely(tp_len <= dev->hard_header_len)) {
1895 			pr_err("packet size is too short (%d < %d)\n",
1896 			       tp_len, dev->hard_header_len);
1897 			return -EINVAL;
1898 		}
1899 
1900 		skb_push(skb, dev->hard_header_len);
1901 		err = skb_store_bits(skb, 0, data,
1902 				dev->hard_header_len);
1903 		if (unlikely(err))
1904 			return err;
1905 
1906 		data += dev->hard_header_len;
1907 		to_write -= dev->hard_header_len;
1908 	}
1909 
1910 	err = -EFAULT;
1911 	offset = offset_in_page(data);
1912 	len_max = PAGE_SIZE - offset;
1913 	len = ((to_write > len_max) ? len_max : to_write);
1914 
1915 	skb->data_len = to_write;
1916 	skb->len += to_write;
1917 	skb->truesize += to_write;
1918 	atomic_add(to_write, &po->sk.sk_wmem_alloc);
1919 
1920 	while (likely(to_write)) {
1921 		nr_frags = skb_shinfo(skb)->nr_frags;
1922 
1923 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
1924 			pr_err("Packet exceed the number of skb frags(%lu)\n",
1925 			       MAX_SKB_FRAGS);
1926 			return -EFAULT;
1927 		}
1928 
1929 		page = pgv_to_page(data);
1930 		data += len;
1931 		flush_dcache_page(page);
1932 		get_page(page);
1933 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
1934 		to_write -= len;
1935 		offset = 0;
1936 		len_max = PAGE_SIZE;
1937 		len = ((to_write > len_max) ? len_max : to_write);
1938 	}
1939 
1940 	return tp_len;
1941 }
1942 
1943 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
1944 {
1945 	struct sk_buff *skb;
1946 	struct net_device *dev;
1947 	__be16 proto;
1948 	bool need_rls_dev = false;
1949 	int err, reserve = 0;
1950 	void *ph;
1951 	struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1952 	int tp_len, size_max;
1953 	unsigned char *addr;
1954 	int len_sum = 0;
1955 	int status = TP_STATUS_AVAILABLE;
1956 	int hlen, tlen;
1957 
1958 	mutex_lock(&po->pg_vec_lock);
1959 
1960 	err = -EBUSY;
1961 	if (saddr == NULL) {
1962 		dev = po->prot_hook.dev;
1963 		proto	= po->num;
1964 		addr	= NULL;
1965 	} else {
1966 		err = -EINVAL;
1967 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1968 			goto out;
1969 		if (msg->msg_namelen < (saddr->sll_halen
1970 					+ offsetof(struct sockaddr_ll,
1971 						sll_addr)))
1972 			goto out;
1973 		proto	= saddr->sll_protocol;
1974 		addr	= saddr->sll_addr;
1975 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
1976 		need_rls_dev = true;
1977 	}
1978 
1979 	err = -ENXIO;
1980 	if (unlikely(dev == NULL))
1981 		goto out;
1982 
1983 	reserve = dev->hard_header_len;
1984 
1985 	err = -ENETDOWN;
1986 	if (unlikely(!(dev->flags & IFF_UP)))
1987 		goto out_put;
1988 
1989 	size_max = po->tx_ring.frame_size
1990 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1991 
1992 	if (size_max > dev->mtu + reserve)
1993 		size_max = dev->mtu + reserve;
1994 
1995 	do {
1996 		ph = packet_current_frame(po, &po->tx_ring,
1997 				TP_STATUS_SEND_REQUEST);
1998 
1999 		if (unlikely(ph == NULL)) {
2000 			schedule();
2001 			continue;
2002 		}
2003 
2004 		status = TP_STATUS_SEND_REQUEST;
2005 		hlen = LL_RESERVED_SPACE(dev);
2006 		tlen = dev->needed_tailroom;
2007 		skb = sock_alloc_send_skb(&po->sk,
2008 				hlen + tlen + sizeof(struct sockaddr_ll),
2009 				0, &err);
2010 
2011 		if (unlikely(skb == NULL))
2012 			goto out_status;
2013 
2014 		tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2015 				addr, hlen);
2016 
2017 		if (unlikely(tp_len < 0)) {
2018 			if (po->tp_loss) {
2019 				__packet_set_status(po, ph,
2020 						TP_STATUS_AVAILABLE);
2021 				packet_increment_head(&po->tx_ring);
2022 				kfree_skb(skb);
2023 				continue;
2024 			} else {
2025 				status = TP_STATUS_WRONG_FORMAT;
2026 				err = tp_len;
2027 				goto out_status;
2028 			}
2029 		}
2030 
2031 		skb->destructor = tpacket_destruct_skb;
2032 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2033 		atomic_inc(&po->tx_ring.pending);
2034 
2035 		status = TP_STATUS_SEND_REQUEST;
2036 		err = dev_queue_xmit(skb);
2037 		if (unlikely(err > 0)) {
2038 			err = net_xmit_errno(err);
2039 			if (err && __packet_get_status(po, ph) ==
2040 				   TP_STATUS_AVAILABLE) {
2041 				/* skb was destructed already */
2042 				skb = NULL;
2043 				goto out_status;
2044 			}
2045 			/*
2046 			 * skb was dropped but not destructed yet;
2047 			 * let's treat it like congestion or err < 0
2048 			 */
2049 			err = 0;
2050 		}
2051 		packet_increment_head(&po->tx_ring);
2052 		len_sum += tp_len;
2053 	} while (likely((ph != NULL) ||
2054 			((!(msg->msg_flags & MSG_DONTWAIT)) &&
2055 			 (atomic_read(&po->tx_ring.pending))))
2056 		);
2057 
2058 	err = len_sum;
2059 	goto out_put;
2060 
2061 out_status:
2062 	__packet_set_status(po, ph, status);
2063 	kfree_skb(skb);
2064 out_put:
2065 	if (need_rls_dev)
2066 		dev_put(dev);
2067 out:
2068 	mutex_unlock(&po->pg_vec_lock);
2069 	return err;
2070 }
2071 
2072 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2073 				        size_t reserve, size_t len,
2074 				        size_t linear, int noblock,
2075 				        int *err)
2076 {
2077 	struct sk_buff *skb;
2078 
2079 	/* Under a page?  Don't bother with paged skb. */
2080 	if (prepad + len < PAGE_SIZE || !linear)
2081 		linear = len;
2082 
2083 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2084 				   err);
2085 	if (!skb)
2086 		return NULL;
2087 
2088 	skb_reserve(skb, reserve);
2089 	skb_put(skb, linear);
2090 	skb->data_len = len - linear;
2091 	skb->len += len - linear;
2092 
2093 	return skb;
2094 }
2095 
2096 static int packet_snd(struct socket *sock,
2097 			  struct msghdr *msg, size_t len)
2098 {
2099 	struct sock *sk = sock->sk;
2100 	struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2101 	struct sk_buff *skb;
2102 	struct net_device *dev;
2103 	__be16 proto;
2104 	bool need_rls_dev = false;
2105 	unsigned char *addr;
2106 	int err, reserve = 0;
2107 	struct virtio_net_hdr vnet_hdr = { 0 };
2108 	int offset = 0;
2109 	int vnet_hdr_len;
2110 	struct packet_sock *po = pkt_sk(sk);
2111 	unsigned short gso_type = 0;
2112 	int hlen, tlen;
2113 	int extra_len = 0;
2114 
2115 	/*
2116 	 *	Get and verify the address.
2117 	 */
2118 
2119 	if (saddr == NULL) {
2120 		dev = po->prot_hook.dev;
2121 		proto	= po->num;
2122 		addr	= NULL;
2123 	} else {
2124 		err = -EINVAL;
2125 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2126 			goto out;
2127 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2128 			goto out;
2129 		proto	= saddr->sll_protocol;
2130 		addr	= saddr->sll_addr;
2131 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2132 		need_rls_dev = true;
2133 	}
2134 
2135 	err = -ENXIO;
2136 	if (dev == NULL)
2137 		goto out_unlock;
2138 	if (sock->type == SOCK_RAW)
2139 		reserve = dev->hard_header_len;
2140 
2141 	err = -ENETDOWN;
2142 	if (!(dev->flags & IFF_UP))
2143 		goto out_unlock;
2144 
2145 	if (po->has_vnet_hdr) {
2146 		vnet_hdr_len = sizeof(vnet_hdr);
2147 
2148 		err = -EINVAL;
2149 		if (len < vnet_hdr_len)
2150 			goto out_unlock;
2151 
2152 		len -= vnet_hdr_len;
2153 
2154 		err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2155 				       vnet_hdr_len);
2156 		if (err < 0)
2157 			goto out_unlock;
2158 
2159 		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2160 		    (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2161 		      vnet_hdr.hdr_len))
2162 			vnet_hdr.hdr_len = vnet_hdr.csum_start +
2163 						 vnet_hdr.csum_offset + 2;
2164 
2165 		err = -EINVAL;
2166 		if (vnet_hdr.hdr_len > len)
2167 			goto out_unlock;
2168 
2169 		if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2170 			switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2171 			case VIRTIO_NET_HDR_GSO_TCPV4:
2172 				gso_type = SKB_GSO_TCPV4;
2173 				break;
2174 			case VIRTIO_NET_HDR_GSO_TCPV6:
2175 				gso_type = SKB_GSO_TCPV6;
2176 				break;
2177 			case VIRTIO_NET_HDR_GSO_UDP:
2178 				gso_type = SKB_GSO_UDP;
2179 				break;
2180 			default:
2181 				goto out_unlock;
2182 			}
2183 
2184 			if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2185 				gso_type |= SKB_GSO_TCP_ECN;
2186 
2187 			if (vnet_hdr.gso_size == 0)
2188 				goto out_unlock;
2189 
2190 		}
2191 	}
2192 
2193 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2194 		if (!netif_supports_nofcs(dev)) {
2195 			err = -EPROTONOSUPPORT;
2196 			goto out_unlock;
2197 		}
2198 		extra_len = 4; /* We're doing our own CRC */
2199 	}
2200 
2201 	err = -EMSGSIZE;
2202 	if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2203 		goto out_unlock;
2204 
2205 	err = -ENOBUFS;
2206 	hlen = LL_RESERVED_SPACE(dev);
2207 	tlen = dev->needed_tailroom;
2208 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2209 			       msg->msg_flags & MSG_DONTWAIT, &err);
2210 	if (skb == NULL)
2211 		goto out_unlock;
2212 
2213 	skb_set_network_header(skb, reserve);
2214 
2215 	err = -EINVAL;
2216 	if (sock->type == SOCK_DGRAM &&
2217 	    (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2218 		goto out_free;
2219 
2220 	/* Returns -EFAULT on error */
2221 	err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2222 	if (err)
2223 		goto out_free;
2224 	err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2225 	if (err < 0)
2226 		goto out_free;
2227 
2228 	if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2229 		/* Earlier code assumed this would be a VLAN pkt,
2230 		 * double-check this now that we have the actual
2231 		 * packet in hand.
2232 		 */
2233 		struct ethhdr *ehdr;
2234 		skb_reset_mac_header(skb);
2235 		ehdr = eth_hdr(skb);
2236 		if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2237 			err = -EMSGSIZE;
2238 			goto out_free;
2239 		}
2240 	}
2241 
2242 	skb->protocol = proto;
2243 	skb->dev = dev;
2244 	skb->priority = sk->sk_priority;
2245 	skb->mark = sk->sk_mark;
2246 
2247 	if (po->has_vnet_hdr) {
2248 		if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2249 			if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2250 						  vnet_hdr.csum_offset)) {
2251 				err = -EINVAL;
2252 				goto out_free;
2253 			}
2254 		}
2255 
2256 		skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2257 		skb_shinfo(skb)->gso_type = gso_type;
2258 
2259 		/* Header must be checked, and gso_segs computed. */
2260 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2261 		skb_shinfo(skb)->gso_segs = 0;
2262 
2263 		len += vnet_hdr_len;
2264 	}
2265 
2266 	if (unlikely(extra_len == 4))
2267 		skb->no_fcs = 1;
2268 
2269 	/*
2270 	 *	Now send it
2271 	 */
2272 
2273 	err = dev_queue_xmit(skb);
2274 	if (err > 0 && (err = net_xmit_errno(err)) != 0)
2275 		goto out_unlock;
2276 
2277 	if (need_rls_dev)
2278 		dev_put(dev);
2279 
2280 	return len;
2281 
2282 out_free:
2283 	kfree_skb(skb);
2284 out_unlock:
2285 	if (dev && need_rls_dev)
2286 		dev_put(dev);
2287 out:
2288 	return err;
2289 }
2290 
2291 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2292 		struct msghdr *msg, size_t len)
2293 {
2294 	struct sock *sk = sock->sk;
2295 	struct packet_sock *po = pkt_sk(sk);
2296 	if (po->tx_ring.pg_vec)
2297 		return tpacket_snd(po, msg);
2298 	else
2299 		return packet_snd(sock, msg, len);
2300 }
2301 
2302 /*
2303  *	Close a PACKET socket. This is fairly simple. We immediately go
2304  *	to 'closed' state and remove our protocol entry in the device list.
2305  */
2306 
2307 static int packet_release(struct socket *sock)
2308 {
2309 	struct sock *sk = sock->sk;
2310 	struct packet_sock *po;
2311 	struct net *net;
2312 	union tpacket_req_u req_u;
2313 
2314 	if (!sk)
2315 		return 0;
2316 
2317 	net = sock_net(sk);
2318 	po = pkt_sk(sk);
2319 
2320 	mutex_lock(&net->packet.sklist_lock);
2321 	sk_del_node_init_rcu(sk);
2322 	mutex_unlock(&net->packet.sklist_lock);
2323 
2324 	preempt_disable();
2325 	sock_prot_inuse_add(net, sk->sk_prot, -1);
2326 	preempt_enable();
2327 
2328 	spin_lock(&po->bind_lock);
2329 	unregister_prot_hook(sk, false);
2330 	if (po->prot_hook.dev) {
2331 		dev_put(po->prot_hook.dev);
2332 		po->prot_hook.dev = NULL;
2333 	}
2334 	spin_unlock(&po->bind_lock);
2335 
2336 	packet_flush_mclist(sk);
2337 
2338 	memset(&req_u, 0, sizeof(req_u));
2339 
2340 	if (po->rx_ring.pg_vec)
2341 		packet_set_ring(sk, &req_u, 1, 0);
2342 
2343 	if (po->tx_ring.pg_vec)
2344 		packet_set_ring(sk, &req_u, 1, 1);
2345 
2346 	fanout_release(sk);
2347 
2348 	synchronize_net();
2349 	/*
2350 	 *	Now the socket is dead. No more input will appear.
2351 	 */
2352 	sock_orphan(sk);
2353 	sock->sk = NULL;
2354 
2355 	/* Purge queues */
2356 
2357 	skb_queue_purge(&sk->sk_receive_queue);
2358 	sk_refcnt_debug_release(sk);
2359 
2360 	sock_put(sk);
2361 	return 0;
2362 }
2363 
2364 /*
2365  *	Attach a packet hook.
2366  */
2367 
2368 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2369 {
2370 	struct packet_sock *po = pkt_sk(sk);
2371 
2372 	if (po->fanout) {
2373 		if (dev)
2374 			dev_put(dev);
2375 
2376 		return -EINVAL;
2377 	}
2378 
2379 	lock_sock(sk);
2380 
2381 	spin_lock(&po->bind_lock);
2382 	unregister_prot_hook(sk, true);
2383 	po->num = protocol;
2384 	po->prot_hook.type = protocol;
2385 	if (po->prot_hook.dev)
2386 		dev_put(po->prot_hook.dev);
2387 	po->prot_hook.dev = dev;
2388 
2389 	po->ifindex = dev ? dev->ifindex : 0;
2390 
2391 	if (protocol == 0)
2392 		goto out_unlock;
2393 
2394 	if (!dev || (dev->flags & IFF_UP)) {
2395 		register_prot_hook(sk);
2396 	} else {
2397 		sk->sk_err = ENETDOWN;
2398 		if (!sock_flag(sk, SOCK_DEAD))
2399 			sk->sk_error_report(sk);
2400 	}
2401 
2402 out_unlock:
2403 	spin_unlock(&po->bind_lock);
2404 	release_sock(sk);
2405 	return 0;
2406 }
2407 
2408 /*
2409  *	Bind a packet socket to a device
2410  */
2411 
2412 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2413 			    int addr_len)
2414 {
2415 	struct sock *sk = sock->sk;
2416 	char name[15];
2417 	struct net_device *dev;
2418 	int err = -ENODEV;
2419 
2420 	/*
2421 	 *	Check legality
2422 	 */
2423 
2424 	if (addr_len != sizeof(struct sockaddr))
2425 		return -EINVAL;
2426 	strlcpy(name, uaddr->sa_data, sizeof(name));
2427 
2428 	dev = dev_get_by_name(sock_net(sk), name);
2429 	if (dev)
2430 		err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2431 	return err;
2432 }
2433 
2434 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2435 {
2436 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2437 	struct sock *sk = sock->sk;
2438 	struct net_device *dev = NULL;
2439 	int err;
2440 
2441 
2442 	/*
2443 	 *	Check legality
2444 	 */
2445 
2446 	if (addr_len < sizeof(struct sockaddr_ll))
2447 		return -EINVAL;
2448 	if (sll->sll_family != AF_PACKET)
2449 		return -EINVAL;
2450 
2451 	if (sll->sll_ifindex) {
2452 		err = -ENODEV;
2453 		dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2454 		if (dev == NULL)
2455 			goto out;
2456 	}
2457 	err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2458 
2459 out:
2460 	return err;
2461 }
2462 
2463 static struct proto packet_proto = {
2464 	.name	  = "PACKET",
2465 	.owner	  = THIS_MODULE,
2466 	.obj_size = sizeof(struct packet_sock),
2467 };
2468 
2469 /*
2470  *	Create a packet of type SOCK_PACKET.
2471  */
2472 
2473 static int packet_create(struct net *net, struct socket *sock, int protocol,
2474 			 int kern)
2475 {
2476 	struct sock *sk;
2477 	struct packet_sock *po;
2478 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
2479 	int err;
2480 
2481 	if (!capable(CAP_NET_RAW))
2482 		return -EPERM;
2483 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2484 	    sock->type != SOCK_PACKET)
2485 		return -ESOCKTNOSUPPORT;
2486 
2487 	sock->state = SS_UNCONNECTED;
2488 
2489 	err = -ENOBUFS;
2490 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2491 	if (sk == NULL)
2492 		goto out;
2493 
2494 	sock->ops = &packet_ops;
2495 	if (sock->type == SOCK_PACKET)
2496 		sock->ops = &packet_ops_spkt;
2497 
2498 	sock_init_data(sock, sk);
2499 
2500 	po = pkt_sk(sk);
2501 	sk->sk_family = PF_PACKET;
2502 	po->num = proto;
2503 
2504 	sk->sk_destruct = packet_sock_destruct;
2505 	sk_refcnt_debug_inc(sk);
2506 
2507 	/*
2508 	 *	Attach a protocol block
2509 	 */
2510 
2511 	spin_lock_init(&po->bind_lock);
2512 	mutex_init(&po->pg_vec_lock);
2513 	po->prot_hook.func = packet_rcv;
2514 
2515 	if (sock->type == SOCK_PACKET)
2516 		po->prot_hook.func = packet_rcv_spkt;
2517 
2518 	po->prot_hook.af_packet_priv = sk;
2519 
2520 	if (proto) {
2521 		po->prot_hook.type = proto;
2522 		register_prot_hook(sk);
2523 	}
2524 
2525 	mutex_lock(&net->packet.sklist_lock);
2526 	sk_add_node_rcu(sk, &net->packet.sklist);
2527 	mutex_unlock(&net->packet.sklist_lock);
2528 
2529 	preempt_disable();
2530 	sock_prot_inuse_add(net, &packet_proto, 1);
2531 	preempt_enable();
2532 
2533 	return 0;
2534 out:
2535 	return err;
2536 }
2537 
2538 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2539 {
2540 	struct sock_exterr_skb *serr;
2541 	struct sk_buff *skb, *skb2;
2542 	int copied, err;
2543 
2544 	err = -EAGAIN;
2545 	skb = skb_dequeue(&sk->sk_error_queue);
2546 	if (skb == NULL)
2547 		goto out;
2548 
2549 	copied = skb->len;
2550 	if (copied > len) {
2551 		msg->msg_flags |= MSG_TRUNC;
2552 		copied = len;
2553 	}
2554 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2555 	if (err)
2556 		goto out_free_skb;
2557 
2558 	sock_recv_timestamp(msg, sk, skb);
2559 
2560 	serr = SKB_EXT_ERR(skb);
2561 	put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2562 		 sizeof(serr->ee), &serr->ee);
2563 
2564 	msg->msg_flags |= MSG_ERRQUEUE;
2565 	err = copied;
2566 
2567 	/* Reset and regenerate socket error */
2568 	spin_lock_bh(&sk->sk_error_queue.lock);
2569 	sk->sk_err = 0;
2570 	if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2571 		sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2572 		spin_unlock_bh(&sk->sk_error_queue.lock);
2573 		sk->sk_error_report(sk);
2574 	} else
2575 		spin_unlock_bh(&sk->sk_error_queue.lock);
2576 
2577 out_free_skb:
2578 	kfree_skb(skb);
2579 out:
2580 	return err;
2581 }
2582 
2583 /*
2584  *	Pull a packet from our receive queue and hand it to the user.
2585  *	If necessary we block.
2586  */
2587 
2588 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2589 			  struct msghdr *msg, size_t len, int flags)
2590 {
2591 	struct sock *sk = sock->sk;
2592 	struct sk_buff *skb;
2593 	int copied, err;
2594 	struct sockaddr_ll *sll;
2595 	int vnet_hdr_len = 0;
2596 
2597 	err = -EINVAL;
2598 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2599 		goto out;
2600 
2601 #if 0
2602 	/* What error should we return now? EUNATTACH? */
2603 	if (pkt_sk(sk)->ifindex < 0)
2604 		return -ENODEV;
2605 #endif
2606 
2607 	if (flags & MSG_ERRQUEUE) {
2608 		err = packet_recv_error(sk, msg, len);
2609 		goto out;
2610 	}
2611 
2612 	/*
2613 	 *	Call the generic datagram receiver. This handles all sorts
2614 	 *	of horrible races and re-entrancy so we can forget about it
2615 	 *	in the protocol layers.
2616 	 *
2617 	 *	Now it will return ENETDOWN, if device have just gone down,
2618 	 *	but then it will block.
2619 	 */
2620 
2621 	skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2622 
2623 	/*
2624 	 *	An error occurred so return it. Because skb_recv_datagram()
2625 	 *	handles the blocking we don't see and worry about blocking
2626 	 *	retries.
2627 	 */
2628 
2629 	if (skb == NULL)
2630 		goto out;
2631 
2632 	if (pkt_sk(sk)->has_vnet_hdr) {
2633 		struct virtio_net_hdr vnet_hdr = { 0 };
2634 
2635 		err = -EINVAL;
2636 		vnet_hdr_len = sizeof(vnet_hdr);
2637 		if (len < vnet_hdr_len)
2638 			goto out_free;
2639 
2640 		len -= vnet_hdr_len;
2641 
2642 		if (skb_is_gso(skb)) {
2643 			struct skb_shared_info *sinfo = skb_shinfo(skb);
2644 
2645 			/* This is a hint as to how much should be linear. */
2646 			vnet_hdr.hdr_len = skb_headlen(skb);
2647 			vnet_hdr.gso_size = sinfo->gso_size;
2648 			if (sinfo->gso_type & SKB_GSO_TCPV4)
2649 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2650 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
2651 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2652 			else if (sinfo->gso_type & SKB_GSO_UDP)
2653 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2654 			else if (sinfo->gso_type & SKB_GSO_FCOE)
2655 				goto out_free;
2656 			else
2657 				BUG();
2658 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2659 				vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2660 		} else
2661 			vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2662 
2663 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
2664 			vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2665 			vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2666 			vnet_hdr.csum_offset = skb->csum_offset;
2667 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2668 			vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2669 		} /* else everything is zero */
2670 
2671 		err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2672 				     vnet_hdr_len);
2673 		if (err < 0)
2674 			goto out_free;
2675 	}
2676 
2677 	/*
2678 	 *	If the address length field is there to be filled in, we fill
2679 	 *	it in now.
2680 	 */
2681 
2682 	sll = &PACKET_SKB_CB(skb)->sa.ll;
2683 	if (sock->type == SOCK_PACKET)
2684 		msg->msg_namelen = sizeof(struct sockaddr_pkt);
2685 	else
2686 		msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2687 
2688 	/*
2689 	 *	You lose any data beyond the buffer you gave. If it worries a
2690 	 *	user program they can ask the device for its MTU anyway.
2691 	 */
2692 
2693 	copied = skb->len;
2694 	if (copied > len) {
2695 		copied = len;
2696 		msg->msg_flags |= MSG_TRUNC;
2697 	}
2698 
2699 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2700 	if (err)
2701 		goto out_free;
2702 
2703 	sock_recv_ts_and_drops(msg, sk, skb);
2704 
2705 	if (msg->msg_name)
2706 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2707 		       msg->msg_namelen);
2708 
2709 	if (pkt_sk(sk)->auxdata) {
2710 		struct tpacket_auxdata aux;
2711 
2712 		aux.tp_status = TP_STATUS_USER;
2713 		if (skb->ip_summed == CHECKSUM_PARTIAL)
2714 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2715 		aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2716 		aux.tp_snaplen = skb->len;
2717 		aux.tp_mac = 0;
2718 		aux.tp_net = skb_network_offset(skb);
2719 		if (vlan_tx_tag_present(skb)) {
2720 			aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2721 			aux.tp_status |= TP_STATUS_VLAN_VALID;
2722 		} else {
2723 			aux.tp_vlan_tci = 0;
2724 		}
2725 		aux.tp_padding = 0;
2726 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2727 	}
2728 
2729 	/*
2730 	 *	Free or return the buffer as appropriate. Again this
2731 	 *	hides all the races and re-entrancy issues from us.
2732 	 */
2733 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2734 
2735 out_free:
2736 	skb_free_datagram(sk, skb);
2737 out:
2738 	return err;
2739 }
2740 
2741 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2742 			       int *uaddr_len, int peer)
2743 {
2744 	struct net_device *dev;
2745 	struct sock *sk	= sock->sk;
2746 
2747 	if (peer)
2748 		return -EOPNOTSUPP;
2749 
2750 	uaddr->sa_family = AF_PACKET;
2751 	rcu_read_lock();
2752 	dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2753 	if (dev)
2754 		strncpy(uaddr->sa_data, dev->name, 14);
2755 	else
2756 		memset(uaddr->sa_data, 0, 14);
2757 	rcu_read_unlock();
2758 	*uaddr_len = sizeof(*uaddr);
2759 
2760 	return 0;
2761 }
2762 
2763 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2764 			  int *uaddr_len, int peer)
2765 {
2766 	struct net_device *dev;
2767 	struct sock *sk = sock->sk;
2768 	struct packet_sock *po = pkt_sk(sk);
2769 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2770 
2771 	if (peer)
2772 		return -EOPNOTSUPP;
2773 
2774 	sll->sll_family = AF_PACKET;
2775 	sll->sll_ifindex = po->ifindex;
2776 	sll->sll_protocol = po->num;
2777 	sll->sll_pkttype = 0;
2778 	rcu_read_lock();
2779 	dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2780 	if (dev) {
2781 		sll->sll_hatype = dev->type;
2782 		sll->sll_halen = dev->addr_len;
2783 		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2784 	} else {
2785 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
2786 		sll->sll_halen = 0;
2787 	}
2788 	rcu_read_unlock();
2789 	*uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2790 
2791 	return 0;
2792 }
2793 
2794 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2795 			 int what)
2796 {
2797 	switch (i->type) {
2798 	case PACKET_MR_MULTICAST:
2799 		if (i->alen != dev->addr_len)
2800 			return -EINVAL;
2801 		if (what > 0)
2802 			return dev_mc_add(dev, i->addr);
2803 		else
2804 			return dev_mc_del(dev, i->addr);
2805 		break;
2806 	case PACKET_MR_PROMISC:
2807 		return dev_set_promiscuity(dev, what);
2808 		break;
2809 	case PACKET_MR_ALLMULTI:
2810 		return dev_set_allmulti(dev, what);
2811 		break;
2812 	case PACKET_MR_UNICAST:
2813 		if (i->alen != dev->addr_len)
2814 			return -EINVAL;
2815 		if (what > 0)
2816 			return dev_uc_add(dev, i->addr);
2817 		else
2818 			return dev_uc_del(dev, i->addr);
2819 		break;
2820 	default:
2821 		break;
2822 	}
2823 	return 0;
2824 }
2825 
2826 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2827 {
2828 	for ( ; i; i = i->next) {
2829 		if (i->ifindex == dev->ifindex)
2830 			packet_dev_mc(dev, i, what);
2831 	}
2832 }
2833 
2834 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2835 {
2836 	struct packet_sock *po = pkt_sk(sk);
2837 	struct packet_mclist *ml, *i;
2838 	struct net_device *dev;
2839 	int err;
2840 
2841 	rtnl_lock();
2842 
2843 	err = -ENODEV;
2844 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2845 	if (!dev)
2846 		goto done;
2847 
2848 	err = -EINVAL;
2849 	if (mreq->mr_alen > dev->addr_len)
2850 		goto done;
2851 
2852 	err = -ENOBUFS;
2853 	i = kmalloc(sizeof(*i), GFP_KERNEL);
2854 	if (i == NULL)
2855 		goto done;
2856 
2857 	err = 0;
2858 	for (ml = po->mclist; ml; ml = ml->next) {
2859 		if (ml->ifindex == mreq->mr_ifindex &&
2860 		    ml->type == mreq->mr_type &&
2861 		    ml->alen == mreq->mr_alen &&
2862 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2863 			ml->count++;
2864 			/* Free the new element ... */
2865 			kfree(i);
2866 			goto done;
2867 		}
2868 	}
2869 
2870 	i->type = mreq->mr_type;
2871 	i->ifindex = mreq->mr_ifindex;
2872 	i->alen = mreq->mr_alen;
2873 	memcpy(i->addr, mreq->mr_address, i->alen);
2874 	i->count = 1;
2875 	i->next = po->mclist;
2876 	po->mclist = i;
2877 	err = packet_dev_mc(dev, i, 1);
2878 	if (err) {
2879 		po->mclist = i->next;
2880 		kfree(i);
2881 	}
2882 
2883 done:
2884 	rtnl_unlock();
2885 	return err;
2886 }
2887 
2888 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2889 {
2890 	struct packet_mclist *ml, **mlp;
2891 
2892 	rtnl_lock();
2893 
2894 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2895 		if (ml->ifindex == mreq->mr_ifindex &&
2896 		    ml->type == mreq->mr_type &&
2897 		    ml->alen == mreq->mr_alen &&
2898 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2899 			if (--ml->count == 0) {
2900 				struct net_device *dev;
2901 				*mlp = ml->next;
2902 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2903 				if (dev)
2904 					packet_dev_mc(dev, ml, -1);
2905 				kfree(ml);
2906 			}
2907 			rtnl_unlock();
2908 			return 0;
2909 		}
2910 	}
2911 	rtnl_unlock();
2912 	return -EADDRNOTAVAIL;
2913 }
2914 
2915 static void packet_flush_mclist(struct sock *sk)
2916 {
2917 	struct packet_sock *po = pkt_sk(sk);
2918 	struct packet_mclist *ml;
2919 
2920 	if (!po->mclist)
2921 		return;
2922 
2923 	rtnl_lock();
2924 	while ((ml = po->mclist) != NULL) {
2925 		struct net_device *dev;
2926 
2927 		po->mclist = ml->next;
2928 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2929 		if (dev != NULL)
2930 			packet_dev_mc(dev, ml, -1);
2931 		kfree(ml);
2932 	}
2933 	rtnl_unlock();
2934 }
2935 
2936 static int
2937 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
2938 {
2939 	struct sock *sk = sock->sk;
2940 	struct packet_sock *po = pkt_sk(sk);
2941 	int ret;
2942 
2943 	if (level != SOL_PACKET)
2944 		return -ENOPROTOOPT;
2945 
2946 	switch (optname) {
2947 	case PACKET_ADD_MEMBERSHIP:
2948 	case PACKET_DROP_MEMBERSHIP:
2949 	{
2950 		struct packet_mreq_max mreq;
2951 		int len = optlen;
2952 		memset(&mreq, 0, sizeof(mreq));
2953 		if (len < sizeof(struct packet_mreq))
2954 			return -EINVAL;
2955 		if (len > sizeof(mreq))
2956 			len = sizeof(mreq);
2957 		if (copy_from_user(&mreq, optval, len))
2958 			return -EFAULT;
2959 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
2960 			return -EINVAL;
2961 		if (optname == PACKET_ADD_MEMBERSHIP)
2962 			ret = packet_mc_add(sk, &mreq);
2963 		else
2964 			ret = packet_mc_drop(sk, &mreq);
2965 		return ret;
2966 	}
2967 
2968 	case PACKET_RX_RING:
2969 	case PACKET_TX_RING:
2970 	{
2971 		union tpacket_req_u req_u;
2972 		int len;
2973 
2974 		switch (po->tp_version) {
2975 		case TPACKET_V1:
2976 		case TPACKET_V2:
2977 			len = sizeof(req_u.req);
2978 			break;
2979 		case TPACKET_V3:
2980 		default:
2981 			len = sizeof(req_u.req3);
2982 			break;
2983 		}
2984 		if (optlen < len)
2985 			return -EINVAL;
2986 		if (pkt_sk(sk)->has_vnet_hdr)
2987 			return -EINVAL;
2988 		if (copy_from_user(&req_u.req, optval, len))
2989 			return -EFAULT;
2990 		return packet_set_ring(sk, &req_u, 0,
2991 			optname == PACKET_TX_RING);
2992 	}
2993 	case PACKET_COPY_THRESH:
2994 	{
2995 		int val;
2996 
2997 		if (optlen != sizeof(val))
2998 			return -EINVAL;
2999 		if (copy_from_user(&val, optval, sizeof(val)))
3000 			return -EFAULT;
3001 
3002 		pkt_sk(sk)->copy_thresh = val;
3003 		return 0;
3004 	}
3005 	case PACKET_VERSION:
3006 	{
3007 		int val;
3008 
3009 		if (optlen != sizeof(val))
3010 			return -EINVAL;
3011 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3012 			return -EBUSY;
3013 		if (copy_from_user(&val, optval, sizeof(val)))
3014 			return -EFAULT;
3015 		switch (val) {
3016 		case TPACKET_V1:
3017 		case TPACKET_V2:
3018 		case TPACKET_V3:
3019 			po->tp_version = val;
3020 			return 0;
3021 		default:
3022 			return -EINVAL;
3023 		}
3024 	}
3025 	case PACKET_RESERVE:
3026 	{
3027 		unsigned int val;
3028 
3029 		if (optlen != sizeof(val))
3030 			return -EINVAL;
3031 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3032 			return -EBUSY;
3033 		if (copy_from_user(&val, optval, sizeof(val)))
3034 			return -EFAULT;
3035 		po->tp_reserve = val;
3036 		return 0;
3037 	}
3038 	case PACKET_LOSS:
3039 	{
3040 		unsigned int val;
3041 
3042 		if (optlen != sizeof(val))
3043 			return -EINVAL;
3044 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3045 			return -EBUSY;
3046 		if (copy_from_user(&val, optval, sizeof(val)))
3047 			return -EFAULT;
3048 		po->tp_loss = !!val;
3049 		return 0;
3050 	}
3051 	case PACKET_AUXDATA:
3052 	{
3053 		int val;
3054 
3055 		if (optlen < sizeof(val))
3056 			return -EINVAL;
3057 		if (copy_from_user(&val, optval, sizeof(val)))
3058 			return -EFAULT;
3059 
3060 		po->auxdata = !!val;
3061 		return 0;
3062 	}
3063 	case PACKET_ORIGDEV:
3064 	{
3065 		int val;
3066 
3067 		if (optlen < sizeof(val))
3068 			return -EINVAL;
3069 		if (copy_from_user(&val, optval, sizeof(val)))
3070 			return -EFAULT;
3071 
3072 		po->origdev = !!val;
3073 		return 0;
3074 	}
3075 	case PACKET_VNET_HDR:
3076 	{
3077 		int val;
3078 
3079 		if (sock->type != SOCK_RAW)
3080 			return -EINVAL;
3081 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3082 			return -EBUSY;
3083 		if (optlen < sizeof(val))
3084 			return -EINVAL;
3085 		if (copy_from_user(&val, optval, sizeof(val)))
3086 			return -EFAULT;
3087 
3088 		po->has_vnet_hdr = !!val;
3089 		return 0;
3090 	}
3091 	case PACKET_TIMESTAMP:
3092 	{
3093 		int val;
3094 
3095 		if (optlen != sizeof(val))
3096 			return -EINVAL;
3097 		if (copy_from_user(&val, optval, sizeof(val)))
3098 			return -EFAULT;
3099 
3100 		po->tp_tstamp = val;
3101 		return 0;
3102 	}
3103 	case PACKET_FANOUT:
3104 	{
3105 		int val;
3106 
3107 		if (optlen != sizeof(val))
3108 			return -EINVAL;
3109 		if (copy_from_user(&val, optval, sizeof(val)))
3110 			return -EFAULT;
3111 
3112 		return fanout_add(sk, val & 0xffff, val >> 16);
3113 	}
3114 	default:
3115 		return -ENOPROTOOPT;
3116 	}
3117 }
3118 
3119 static int packet_getsockopt(struct socket *sock, int level, int optname,
3120 			     char __user *optval, int __user *optlen)
3121 {
3122 	int len;
3123 	int val, lv = sizeof(val);
3124 	struct sock *sk = sock->sk;
3125 	struct packet_sock *po = pkt_sk(sk);
3126 	void *data = &val;
3127 	struct tpacket_stats st;
3128 	union tpacket_stats_u st_u;
3129 
3130 	if (level != SOL_PACKET)
3131 		return -ENOPROTOOPT;
3132 
3133 	if (get_user(len, optlen))
3134 		return -EFAULT;
3135 
3136 	if (len < 0)
3137 		return -EINVAL;
3138 
3139 	switch (optname) {
3140 	case PACKET_STATISTICS:
3141 		spin_lock_bh(&sk->sk_receive_queue.lock);
3142 		if (po->tp_version == TPACKET_V3) {
3143 			lv = sizeof(struct tpacket_stats_v3);
3144 			memcpy(&st_u.stats3, &po->stats,
3145 			       sizeof(struct tpacket_stats));
3146 			st_u.stats3.tp_freeze_q_cnt =
3147 					po->stats_u.stats3.tp_freeze_q_cnt;
3148 			st_u.stats3.tp_packets += po->stats.tp_drops;
3149 			data = &st_u.stats3;
3150 		} else {
3151 			lv = sizeof(struct tpacket_stats);
3152 			st = po->stats;
3153 			st.tp_packets += st.tp_drops;
3154 			data = &st;
3155 		}
3156 		memset(&po->stats, 0, sizeof(st));
3157 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3158 		break;
3159 	case PACKET_AUXDATA:
3160 		val = po->auxdata;
3161 		break;
3162 	case PACKET_ORIGDEV:
3163 		val = po->origdev;
3164 		break;
3165 	case PACKET_VNET_HDR:
3166 		val = po->has_vnet_hdr;
3167 		break;
3168 	case PACKET_VERSION:
3169 		val = po->tp_version;
3170 		break;
3171 	case PACKET_HDRLEN:
3172 		if (len > sizeof(int))
3173 			len = sizeof(int);
3174 		if (copy_from_user(&val, optval, len))
3175 			return -EFAULT;
3176 		switch (val) {
3177 		case TPACKET_V1:
3178 			val = sizeof(struct tpacket_hdr);
3179 			break;
3180 		case TPACKET_V2:
3181 			val = sizeof(struct tpacket2_hdr);
3182 			break;
3183 		case TPACKET_V3:
3184 			val = sizeof(struct tpacket3_hdr);
3185 			break;
3186 		default:
3187 			return -EINVAL;
3188 		}
3189 		break;
3190 	case PACKET_RESERVE:
3191 		val = po->tp_reserve;
3192 		break;
3193 	case PACKET_LOSS:
3194 		val = po->tp_loss;
3195 		break;
3196 	case PACKET_TIMESTAMP:
3197 		val = po->tp_tstamp;
3198 		break;
3199 	case PACKET_FANOUT:
3200 		val = (po->fanout ?
3201 		       ((u32)po->fanout->id |
3202 			((u32)po->fanout->type << 16)) :
3203 		       0);
3204 		break;
3205 	default:
3206 		return -ENOPROTOOPT;
3207 	}
3208 
3209 	if (len > lv)
3210 		len = lv;
3211 	if (put_user(len, optlen))
3212 		return -EFAULT;
3213 	if (copy_to_user(optval, data, len))
3214 		return -EFAULT;
3215 	return 0;
3216 }
3217 
3218 
3219 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3220 {
3221 	struct sock *sk;
3222 	struct hlist_node *node;
3223 	struct net_device *dev = data;
3224 	struct net *net = dev_net(dev);
3225 
3226 	rcu_read_lock();
3227 	sk_for_each_rcu(sk, node, &net->packet.sklist) {
3228 		struct packet_sock *po = pkt_sk(sk);
3229 
3230 		switch (msg) {
3231 		case NETDEV_UNREGISTER:
3232 			if (po->mclist)
3233 				packet_dev_mclist(dev, po->mclist, -1);
3234 			/* fallthrough */
3235 
3236 		case NETDEV_DOWN:
3237 			if (dev->ifindex == po->ifindex) {
3238 				spin_lock(&po->bind_lock);
3239 				if (po->running) {
3240 					__unregister_prot_hook(sk, false);
3241 					sk->sk_err = ENETDOWN;
3242 					if (!sock_flag(sk, SOCK_DEAD))
3243 						sk->sk_error_report(sk);
3244 				}
3245 				if (msg == NETDEV_UNREGISTER) {
3246 					po->ifindex = -1;
3247 					if (po->prot_hook.dev)
3248 						dev_put(po->prot_hook.dev);
3249 					po->prot_hook.dev = NULL;
3250 				}
3251 				spin_unlock(&po->bind_lock);
3252 			}
3253 			break;
3254 		case NETDEV_UP:
3255 			if (dev->ifindex == po->ifindex) {
3256 				spin_lock(&po->bind_lock);
3257 				if (po->num)
3258 					register_prot_hook(sk);
3259 				spin_unlock(&po->bind_lock);
3260 			}
3261 			break;
3262 		}
3263 	}
3264 	rcu_read_unlock();
3265 	return NOTIFY_DONE;
3266 }
3267 
3268 
3269 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3270 			unsigned long arg)
3271 {
3272 	struct sock *sk = sock->sk;
3273 
3274 	switch (cmd) {
3275 	case SIOCOUTQ:
3276 	{
3277 		int amount = sk_wmem_alloc_get(sk);
3278 
3279 		return put_user(amount, (int __user *)arg);
3280 	}
3281 	case SIOCINQ:
3282 	{
3283 		struct sk_buff *skb;
3284 		int amount = 0;
3285 
3286 		spin_lock_bh(&sk->sk_receive_queue.lock);
3287 		skb = skb_peek(&sk->sk_receive_queue);
3288 		if (skb)
3289 			amount = skb->len;
3290 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3291 		return put_user(amount, (int __user *)arg);
3292 	}
3293 	case SIOCGSTAMP:
3294 		return sock_get_timestamp(sk, (struct timeval __user *)arg);
3295 	case SIOCGSTAMPNS:
3296 		return sock_get_timestampns(sk, (struct timespec __user *)arg);
3297 
3298 #ifdef CONFIG_INET
3299 	case SIOCADDRT:
3300 	case SIOCDELRT:
3301 	case SIOCDARP:
3302 	case SIOCGARP:
3303 	case SIOCSARP:
3304 	case SIOCGIFADDR:
3305 	case SIOCSIFADDR:
3306 	case SIOCGIFBRDADDR:
3307 	case SIOCSIFBRDADDR:
3308 	case SIOCGIFNETMASK:
3309 	case SIOCSIFNETMASK:
3310 	case SIOCGIFDSTADDR:
3311 	case SIOCSIFDSTADDR:
3312 	case SIOCSIFFLAGS:
3313 		return inet_dgram_ops.ioctl(sock, cmd, arg);
3314 #endif
3315 
3316 	default:
3317 		return -ENOIOCTLCMD;
3318 	}
3319 	return 0;
3320 }
3321 
3322 static unsigned int packet_poll(struct file *file, struct socket *sock,
3323 				poll_table *wait)
3324 {
3325 	struct sock *sk = sock->sk;
3326 	struct packet_sock *po = pkt_sk(sk);
3327 	unsigned int mask = datagram_poll(file, sock, wait);
3328 
3329 	spin_lock_bh(&sk->sk_receive_queue.lock);
3330 	if (po->rx_ring.pg_vec) {
3331 		if (!packet_previous_rx_frame(po, &po->rx_ring,
3332 			TP_STATUS_KERNEL))
3333 			mask |= POLLIN | POLLRDNORM;
3334 	}
3335 	spin_unlock_bh(&sk->sk_receive_queue.lock);
3336 	spin_lock_bh(&sk->sk_write_queue.lock);
3337 	if (po->tx_ring.pg_vec) {
3338 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3339 			mask |= POLLOUT | POLLWRNORM;
3340 	}
3341 	spin_unlock_bh(&sk->sk_write_queue.lock);
3342 	return mask;
3343 }
3344 
3345 
3346 /* Dirty? Well, I still did not learn better way to account
3347  * for user mmaps.
3348  */
3349 
3350 static void packet_mm_open(struct vm_area_struct *vma)
3351 {
3352 	struct file *file = vma->vm_file;
3353 	struct socket *sock = file->private_data;
3354 	struct sock *sk = sock->sk;
3355 
3356 	if (sk)
3357 		atomic_inc(&pkt_sk(sk)->mapped);
3358 }
3359 
3360 static void packet_mm_close(struct vm_area_struct *vma)
3361 {
3362 	struct file *file = vma->vm_file;
3363 	struct socket *sock = file->private_data;
3364 	struct sock *sk = sock->sk;
3365 
3366 	if (sk)
3367 		atomic_dec(&pkt_sk(sk)->mapped);
3368 }
3369 
3370 static const struct vm_operations_struct packet_mmap_ops = {
3371 	.open	=	packet_mm_open,
3372 	.close	=	packet_mm_close,
3373 };
3374 
3375 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3376 			unsigned int len)
3377 {
3378 	int i;
3379 
3380 	for (i = 0; i < len; i++) {
3381 		if (likely(pg_vec[i].buffer)) {
3382 			if (is_vmalloc_addr(pg_vec[i].buffer))
3383 				vfree(pg_vec[i].buffer);
3384 			else
3385 				free_pages((unsigned long)pg_vec[i].buffer,
3386 					   order);
3387 			pg_vec[i].buffer = NULL;
3388 		}
3389 	}
3390 	kfree(pg_vec);
3391 }
3392 
3393 static char *alloc_one_pg_vec_page(unsigned long order)
3394 {
3395 	char *buffer = NULL;
3396 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3397 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3398 
3399 	buffer = (char *) __get_free_pages(gfp_flags, order);
3400 
3401 	if (buffer)
3402 		return buffer;
3403 
3404 	/*
3405 	 * __get_free_pages failed, fall back to vmalloc
3406 	 */
3407 	buffer = vzalloc((1 << order) * PAGE_SIZE);
3408 
3409 	if (buffer)
3410 		return buffer;
3411 
3412 	/*
3413 	 * vmalloc failed, lets dig into swap here
3414 	 */
3415 	gfp_flags &= ~__GFP_NORETRY;
3416 	buffer = (char *)__get_free_pages(gfp_flags, order);
3417 	if (buffer)
3418 		return buffer;
3419 
3420 	/*
3421 	 * complete and utter failure
3422 	 */
3423 	return NULL;
3424 }
3425 
3426 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3427 {
3428 	unsigned int block_nr = req->tp_block_nr;
3429 	struct pgv *pg_vec;
3430 	int i;
3431 
3432 	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3433 	if (unlikely(!pg_vec))
3434 		goto out;
3435 
3436 	for (i = 0; i < block_nr; i++) {
3437 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3438 		if (unlikely(!pg_vec[i].buffer))
3439 			goto out_free_pgvec;
3440 	}
3441 
3442 out:
3443 	return pg_vec;
3444 
3445 out_free_pgvec:
3446 	free_pg_vec(pg_vec, order, block_nr);
3447 	pg_vec = NULL;
3448 	goto out;
3449 }
3450 
3451 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3452 		int closing, int tx_ring)
3453 {
3454 	struct pgv *pg_vec = NULL;
3455 	struct packet_sock *po = pkt_sk(sk);
3456 	int was_running, order = 0;
3457 	struct packet_ring_buffer *rb;
3458 	struct sk_buff_head *rb_queue;
3459 	__be16 num;
3460 	int err = -EINVAL;
3461 	/* Added to avoid minimal code churn */
3462 	struct tpacket_req *req = &req_u->req;
3463 
3464 	/* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3465 	if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3466 		WARN(1, "Tx-ring is not supported.\n");
3467 		goto out;
3468 	}
3469 
3470 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3471 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3472 
3473 	err = -EBUSY;
3474 	if (!closing) {
3475 		if (atomic_read(&po->mapped))
3476 			goto out;
3477 		if (atomic_read(&rb->pending))
3478 			goto out;
3479 	}
3480 
3481 	if (req->tp_block_nr) {
3482 		/* Sanity tests and some calculations */
3483 		err = -EBUSY;
3484 		if (unlikely(rb->pg_vec))
3485 			goto out;
3486 
3487 		switch (po->tp_version) {
3488 		case TPACKET_V1:
3489 			po->tp_hdrlen = TPACKET_HDRLEN;
3490 			break;
3491 		case TPACKET_V2:
3492 			po->tp_hdrlen = TPACKET2_HDRLEN;
3493 			break;
3494 		case TPACKET_V3:
3495 			po->tp_hdrlen = TPACKET3_HDRLEN;
3496 			break;
3497 		}
3498 
3499 		err = -EINVAL;
3500 		if (unlikely((int)req->tp_block_size <= 0))
3501 			goto out;
3502 		if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3503 			goto out;
3504 		if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3505 					po->tp_reserve))
3506 			goto out;
3507 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3508 			goto out;
3509 
3510 		rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3511 		if (unlikely(rb->frames_per_block <= 0))
3512 			goto out;
3513 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3514 					req->tp_frame_nr))
3515 			goto out;
3516 
3517 		err = -ENOMEM;
3518 		order = get_order(req->tp_block_size);
3519 		pg_vec = alloc_pg_vec(req, order);
3520 		if (unlikely(!pg_vec))
3521 			goto out;
3522 		switch (po->tp_version) {
3523 		case TPACKET_V3:
3524 		/* Transmit path is not supported. We checked
3525 		 * it above but just being paranoid
3526 		 */
3527 			if (!tx_ring)
3528 				init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3529 				break;
3530 		default:
3531 			break;
3532 		}
3533 	}
3534 	/* Done */
3535 	else {
3536 		err = -EINVAL;
3537 		if (unlikely(req->tp_frame_nr))
3538 			goto out;
3539 	}
3540 
3541 	lock_sock(sk);
3542 
3543 	/* Detach socket from network */
3544 	spin_lock(&po->bind_lock);
3545 	was_running = po->running;
3546 	num = po->num;
3547 	if (was_running) {
3548 		po->num = 0;
3549 		__unregister_prot_hook(sk, false);
3550 	}
3551 	spin_unlock(&po->bind_lock);
3552 
3553 	synchronize_net();
3554 
3555 	err = -EBUSY;
3556 	mutex_lock(&po->pg_vec_lock);
3557 	if (closing || atomic_read(&po->mapped) == 0) {
3558 		err = 0;
3559 		spin_lock_bh(&rb_queue->lock);
3560 		swap(rb->pg_vec, pg_vec);
3561 		rb->frame_max = (req->tp_frame_nr - 1);
3562 		rb->head = 0;
3563 		rb->frame_size = req->tp_frame_size;
3564 		spin_unlock_bh(&rb_queue->lock);
3565 
3566 		swap(rb->pg_vec_order, order);
3567 		swap(rb->pg_vec_len, req->tp_block_nr);
3568 
3569 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3570 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
3571 						tpacket_rcv : packet_rcv;
3572 		skb_queue_purge(rb_queue);
3573 		if (atomic_read(&po->mapped))
3574 			pr_err("packet_mmap: vma is busy: %d\n",
3575 			       atomic_read(&po->mapped));
3576 	}
3577 	mutex_unlock(&po->pg_vec_lock);
3578 
3579 	spin_lock(&po->bind_lock);
3580 	if (was_running) {
3581 		po->num = num;
3582 		register_prot_hook(sk);
3583 	}
3584 	spin_unlock(&po->bind_lock);
3585 	if (closing && (po->tp_version > TPACKET_V2)) {
3586 		/* Because we don't support block-based V3 on tx-ring */
3587 		if (!tx_ring)
3588 			prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3589 	}
3590 	release_sock(sk);
3591 
3592 	if (pg_vec)
3593 		free_pg_vec(pg_vec, order, req->tp_block_nr);
3594 out:
3595 	return err;
3596 }
3597 
3598 static int packet_mmap(struct file *file, struct socket *sock,
3599 		struct vm_area_struct *vma)
3600 {
3601 	struct sock *sk = sock->sk;
3602 	struct packet_sock *po = pkt_sk(sk);
3603 	unsigned long size, expected_size;
3604 	struct packet_ring_buffer *rb;
3605 	unsigned long start;
3606 	int err = -EINVAL;
3607 	int i;
3608 
3609 	if (vma->vm_pgoff)
3610 		return -EINVAL;
3611 
3612 	mutex_lock(&po->pg_vec_lock);
3613 
3614 	expected_size = 0;
3615 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3616 		if (rb->pg_vec) {
3617 			expected_size += rb->pg_vec_len
3618 						* rb->pg_vec_pages
3619 						* PAGE_SIZE;
3620 		}
3621 	}
3622 
3623 	if (expected_size == 0)
3624 		goto out;
3625 
3626 	size = vma->vm_end - vma->vm_start;
3627 	if (size != expected_size)
3628 		goto out;
3629 
3630 	start = vma->vm_start;
3631 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3632 		if (rb->pg_vec == NULL)
3633 			continue;
3634 
3635 		for (i = 0; i < rb->pg_vec_len; i++) {
3636 			struct page *page;
3637 			void *kaddr = rb->pg_vec[i].buffer;
3638 			int pg_num;
3639 
3640 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3641 				page = pgv_to_page(kaddr);
3642 				err = vm_insert_page(vma, start, page);
3643 				if (unlikely(err))
3644 					goto out;
3645 				start += PAGE_SIZE;
3646 				kaddr += PAGE_SIZE;
3647 			}
3648 		}
3649 	}
3650 
3651 	atomic_inc(&po->mapped);
3652 	vma->vm_ops = &packet_mmap_ops;
3653 	err = 0;
3654 
3655 out:
3656 	mutex_unlock(&po->pg_vec_lock);
3657 	return err;
3658 }
3659 
3660 static const struct proto_ops packet_ops_spkt = {
3661 	.family =	PF_PACKET,
3662 	.owner =	THIS_MODULE,
3663 	.release =	packet_release,
3664 	.bind =		packet_bind_spkt,
3665 	.connect =	sock_no_connect,
3666 	.socketpair =	sock_no_socketpair,
3667 	.accept =	sock_no_accept,
3668 	.getname =	packet_getname_spkt,
3669 	.poll =		datagram_poll,
3670 	.ioctl =	packet_ioctl,
3671 	.listen =	sock_no_listen,
3672 	.shutdown =	sock_no_shutdown,
3673 	.setsockopt =	sock_no_setsockopt,
3674 	.getsockopt =	sock_no_getsockopt,
3675 	.sendmsg =	packet_sendmsg_spkt,
3676 	.recvmsg =	packet_recvmsg,
3677 	.mmap =		sock_no_mmap,
3678 	.sendpage =	sock_no_sendpage,
3679 };
3680 
3681 static const struct proto_ops packet_ops = {
3682 	.family =	PF_PACKET,
3683 	.owner =	THIS_MODULE,
3684 	.release =	packet_release,
3685 	.bind =		packet_bind,
3686 	.connect =	sock_no_connect,
3687 	.socketpair =	sock_no_socketpair,
3688 	.accept =	sock_no_accept,
3689 	.getname =	packet_getname,
3690 	.poll =		packet_poll,
3691 	.ioctl =	packet_ioctl,
3692 	.listen =	sock_no_listen,
3693 	.shutdown =	sock_no_shutdown,
3694 	.setsockopt =	packet_setsockopt,
3695 	.getsockopt =	packet_getsockopt,
3696 	.sendmsg =	packet_sendmsg,
3697 	.recvmsg =	packet_recvmsg,
3698 	.mmap =		packet_mmap,
3699 	.sendpage =	sock_no_sendpage,
3700 };
3701 
3702 static const struct net_proto_family packet_family_ops = {
3703 	.family =	PF_PACKET,
3704 	.create =	packet_create,
3705 	.owner	=	THIS_MODULE,
3706 };
3707 
3708 static struct notifier_block packet_netdev_notifier = {
3709 	.notifier_call =	packet_notifier,
3710 };
3711 
3712 #ifdef CONFIG_PROC_FS
3713 
3714 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3715 	__acquires(RCU)
3716 {
3717 	struct net *net = seq_file_net(seq);
3718 
3719 	rcu_read_lock();
3720 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3721 }
3722 
3723 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3724 {
3725 	struct net *net = seq_file_net(seq);
3726 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3727 }
3728 
3729 static void packet_seq_stop(struct seq_file *seq, void *v)
3730 	__releases(RCU)
3731 {
3732 	rcu_read_unlock();
3733 }
3734 
3735 static int packet_seq_show(struct seq_file *seq, void *v)
3736 {
3737 	if (v == SEQ_START_TOKEN)
3738 		seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3739 	else {
3740 		struct sock *s = sk_entry(v);
3741 		const struct packet_sock *po = pkt_sk(s);
3742 
3743 		seq_printf(seq,
3744 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3745 			   s,
3746 			   atomic_read(&s->sk_refcnt),
3747 			   s->sk_type,
3748 			   ntohs(po->num),
3749 			   po->ifindex,
3750 			   po->running,
3751 			   atomic_read(&s->sk_rmem_alloc),
3752 			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3753 			   sock_i_ino(s));
3754 	}
3755 
3756 	return 0;
3757 }
3758 
3759 static const struct seq_operations packet_seq_ops = {
3760 	.start	= packet_seq_start,
3761 	.next	= packet_seq_next,
3762 	.stop	= packet_seq_stop,
3763 	.show	= packet_seq_show,
3764 };
3765 
3766 static int packet_seq_open(struct inode *inode, struct file *file)
3767 {
3768 	return seq_open_net(inode, file, &packet_seq_ops,
3769 			    sizeof(struct seq_net_private));
3770 }
3771 
3772 static const struct file_operations packet_seq_fops = {
3773 	.owner		= THIS_MODULE,
3774 	.open		= packet_seq_open,
3775 	.read		= seq_read,
3776 	.llseek		= seq_lseek,
3777 	.release	= seq_release_net,
3778 };
3779 
3780 #endif
3781 
3782 static int __net_init packet_net_init(struct net *net)
3783 {
3784 	mutex_init(&net->packet.sklist_lock);
3785 	INIT_HLIST_HEAD(&net->packet.sklist);
3786 
3787 	if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3788 		return -ENOMEM;
3789 
3790 	return 0;
3791 }
3792 
3793 static void __net_exit packet_net_exit(struct net *net)
3794 {
3795 	proc_net_remove(net, "packet");
3796 }
3797 
3798 static struct pernet_operations packet_net_ops = {
3799 	.init = packet_net_init,
3800 	.exit = packet_net_exit,
3801 };
3802 
3803 
3804 static void __exit packet_exit(void)
3805 {
3806 	unregister_netdevice_notifier(&packet_netdev_notifier);
3807 	unregister_pernet_subsys(&packet_net_ops);
3808 	sock_unregister(PF_PACKET);
3809 	proto_unregister(&packet_proto);
3810 }
3811 
3812 static int __init packet_init(void)
3813 {
3814 	int rc = proto_register(&packet_proto, 0);
3815 
3816 	if (rc != 0)
3817 		goto out;
3818 
3819 	sock_register(&packet_family_ops);
3820 	register_pernet_subsys(&packet_net_ops);
3821 	register_netdevice_notifier(&packet_netdev_notifier);
3822 out:
3823 	return rc;
3824 }
3825 
3826 module_init(packet_init);
3827 module_exit(packet_exit);
3828 MODULE_LICENSE("GPL");
3829 MODULE_ALIAS_NETPROTO(PF_PACKET);
3830