xref: /openbmc/linux/net/packet/af_packet.c (revision 0f4630f3)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		PACKET - implements raw packet sockets.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
11  *
12  * Fixes:
13  *		Alan Cox	:	verify_area() now used correctly
14  *		Alan Cox	:	new skbuff lists, look ma no backlogs!
15  *		Alan Cox	:	tidied skbuff lists.
16  *		Alan Cox	:	Now uses generic datagram routines I
17  *					added. Also fixed the peek/read crash
18  *					from all old Linux datagram code.
19  *		Alan Cox	:	Uses the improved datagram code.
20  *		Alan Cox	:	Added NULL's for socket options.
21  *		Alan Cox	:	Re-commented the code.
22  *		Alan Cox	:	Use new kernel side addressing
23  *		Rob Janssen	:	Correct MTU usage.
24  *		Dave Platt	:	Counter leaks caused by incorrect
25  *					interrupt locking and some slightly
26  *					dubious gcc output. Can you read
27  *					compiler: it said _VOLATILE_
28  *	Richard Kooijman	:	Timestamp fixes.
29  *		Alan Cox	:	New buffers. Use sk->mac.raw.
30  *		Alan Cox	:	sendmsg/recvmsg support.
31  *		Alan Cox	:	Protocol setting support
32  *	Alexey Kuznetsov	:	Untied from IPv4 stack.
33  *	Cyrus Durgin		:	Fixed kerneld for kmod.
34  *	Michal Ostrowski        :       Module initialization cleanup.
35  *         Ulises Alonso        :       Frame number limit removal and
36  *                                      packet_set_ring memory leak.
37  *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
38  *					The convention is that longer addresses
39  *					will simply extend the hardware address
40  *					byte arrays at the end of sockaddr_ll
41  *					and packet_mreq.
42  *		Johann Baudy	:	Added TX RING.
43  *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
44  *					layer.
45  *					Copyright (C) 2011, <lokec@ccs.neu.edu>
46  *
47  *
48  *		This program is free software; you can redistribute it and/or
49  *		modify it under the terms of the GNU General Public License
50  *		as published by the Free Software Foundation; either version
51  *		2 of the License, or (at your option) any later version.
52  *
53  */
54 
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
78 #include <asm/page.h>
79 #include <asm/cacheflush.h>
80 #include <asm/io.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
92 #ifdef CONFIG_INET
93 #include <net/inet_common.h>
94 #endif
95 #include <linux/bpf.h>
96 
97 #include "internal.h"
98 
99 /*
100    Assumptions:
101    - if device has no dev->hard_header routine, it adds and removes ll header
102      inside itself. In this case ll header is invisible outside of device,
103      but higher levels still should reserve dev->hard_header_len.
104      Some devices are enough clever to reallocate skb, when header
105      will not fit to reserved space (tunnel), another ones are silly
106      (PPP).
107    - packet socket receives packets with pulled ll header,
108      so that SOCK_RAW should push it back.
109 
110 On receive:
111 -----------
112 
113 Incoming, dev->hard_header!=NULL
114    mac_header -> ll header
115    data       -> data
116 
117 Outgoing, dev->hard_header!=NULL
118    mac_header -> ll header
119    data       -> ll header
120 
121 Incoming, dev->hard_header==NULL
122    mac_header -> UNKNOWN position. It is very likely, that it points to ll
123 		 header.  PPP makes it, that is wrong, because introduce
124 		 assymetry between rx and tx paths.
125    data       -> data
126 
127 Outgoing, dev->hard_header==NULL
128    mac_header -> data. ll header is still not built!
129    data       -> data
130 
131 Resume
132   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
133 
134 
135 On transmit:
136 ------------
137 
138 dev->hard_header != NULL
139    mac_header -> ll header
140    data       -> ll header
141 
142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
143    mac_header -> data
144    data       -> data
145 
146    We should set nh.raw on output to correct posistion,
147    packet classifier depends on it.
148  */
149 
150 /* Private packet socket structures. */
151 
152 /* identical to struct packet_mreq except it has
153  * a longer address field.
154  */
155 struct packet_mreq_max {
156 	int		mr_ifindex;
157 	unsigned short	mr_type;
158 	unsigned short	mr_alen;
159 	unsigned char	mr_address[MAX_ADDR_LEN];
160 };
161 
162 union tpacket_uhdr {
163 	struct tpacket_hdr  *h1;
164 	struct tpacket2_hdr *h2;
165 	struct tpacket3_hdr *h3;
166 	void *raw;
167 };
168 
169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170 		int closing, int tx_ring);
171 
172 #define V3_ALIGNMENT	(8)
173 
174 #define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 
179 #define PGV_FROM_VMALLOC 1
180 
181 #define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
182 #define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
183 #define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
184 #define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
185 #define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
186 #define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
187 #define BLOCK_PRIV(x)		((void *)((char *)(x) + BLOCK_O2PRIV(x)))
188 
189 struct packet_sock;
190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 		       struct packet_type *pt, struct net_device *orig_dev);
193 
194 static void *packet_previous_frame(struct packet_sock *po,
195 		struct packet_ring_buffer *rb,
196 		int status);
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
199 			struct tpacket_block_desc *);
200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
201 			struct packet_sock *);
202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
203 		struct packet_sock *, unsigned int status);
204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
205 static void prb_open_block(struct tpacket_kbdq_core *,
206 		struct tpacket_block_desc *);
207 static void prb_retire_rx_blk_timer_expired(unsigned long);
208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
209 static void prb_init_blk_timer(struct packet_sock *,
210 		struct tpacket_kbdq_core *,
211 		void (*func) (unsigned long));
212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
214 		struct tpacket3_hdr *);
215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
216 		struct tpacket3_hdr *);
217 static void packet_flush_mclist(struct sock *sk);
218 
219 struct packet_skb_cb {
220 	union {
221 		struct sockaddr_pkt pkt;
222 		union {
223 			/* Trick: alias skb original length with
224 			 * ll.sll_family and ll.protocol in order
225 			 * to save room.
226 			 */
227 			unsigned int origlen;
228 			struct sockaddr_ll ll;
229 		};
230 	} sa;
231 };
232 
233 #define vio_le() virtio_legacy_is_little_endian()
234 
235 #define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
236 
237 #define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
238 #define GET_PBLOCK_DESC(x, bid)	\
239 	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
241 	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
242 #define GET_NEXT_PRB_BLK_NUM(x) \
243 	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244 	((x)->kactive_blk_num+1) : 0)
245 
246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
248 
249 static int packet_direct_xmit(struct sk_buff *skb)
250 {
251 	struct net_device *dev = skb->dev;
252 	netdev_features_t features;
253 	struct netdev_queue *txq;
254 	int ret = NETDEV_TX_BUSY;
255 
256 	if (unlikely(!netif_running(dev) ||
257 		     !netif_carrier_ok(dev)))
258 		goto drop;
259 
260 	features = netif_skb_features(skb);
261 	if (skb_needs_linearize(skb, features) &&
262 	    __skb_linearize(skb))
263 		goto drop;
264 
265 	txq = skb_get_tx_queue(dev, skb);
266 
267 	local_bh_disable();
268 
269 	HARD_TX_LOCK(dev, txq, smp_processor_id());
270 	if (!netif_xmit_frozen_or_drv_stopped(txq))
271 		ret = netdev_start_xmit(skb, dev, txq, false);
272 	HARD_TX_UNLOCK(dev, txq);
273 
274 	local_bh_enable();
275 
276 	if (!dev_xmit_complete(ret))
277 		kfree_skb(skb);
278 
279 	return ret;
280 drop:
281 	atomic_long_inc(&dev->tx_dropped);
282 	kfree_skb(skb);
283 	return NET_XMIT_DROP;
284 }
285 
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287 {
288 	struct net_device *dev;
289 
290 	rcu_read_lock();
291 	dev = rcu_dereference(po->cached_dev);
292 	if (likely(dev))
293 		dev_hold(dev);
294 	rcu_read_unlock();
295 
296 	return dev;
297 }
298 
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 				     struct net_device *dev)
301 {
302 	rcu_assign_pointer(po->cached_dev, dev);
303 }
304 
305 static void packet_cached_dev_reset(struct packet_sock *po)
306 {
307 	RCU_INIT_POINTER(po->cached_dev, NULL);
308 }
309 
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
311 {
312 	return po->xmit == packet_direct_xmit;
313 }
314 
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
316 {
317 	return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
318 }
319 
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
321 {
322 	const struct net_device_ops *ops = dev->netdev_ops;
323 	u16 queue_index;
324 
325 	if (ops->ndo_select_queue) {
326 		queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 						    __packet_pick_tx_queue);
328 		queue_index = netdev_cap_txqueue(dev, queue_index);
329 	} else {
330 		queue_index = __packet_pick_tx_queue(dev, skb);
331 	}
332 
333 	skb_set_queue_mapping(skb, queue_index);
334 }
335 
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337  * or from a context in which asynchronous accesses to the packet
338  * socket is not possible (packet_create()).
339  */
340 static void register_prot_hook(struct sock *sk)
341 {
342 	struct packet_sock *po = pkt_sk(sk);
343 
344 	if (!po->running) {
345 		if (po->fanout)
346 			__fanout_link(sk, po);
347 		else
348 			dev_add_pack(&po->prot_hook);
349 
350 		sock_hold(sk);
351 		po->running = 1;
352 	}
353 }
354 
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356  * held.   If the sync parameter is true, we will temporarily drop
357  * the po->bind_lock and do a synchronize_net to make sure no
358  * asynchronous packet processing paths still refer to the elements
359  * of po->prot_hook.  If the sync parameter is false, it is the
360  * callers responsibility to take care of this.
361  */
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
363 {
364 	struct packet_sock *po = pkt_sk(sk);
365 
366 	po->running = 0;
367 
368 	if (po->fanout)
369 		__fanout_unlink(sk, po);
370 	else
371 		__dev_remove_pack(&po->prot_hook);
372 
373 	__sock_put(sk);
374 
375 	if (sync) {
376 		spin_unlock(&po->bind_lock);
377 		synchronize_net();
378 		spin_lock(&po->bind_lock);
379 	}
380 }
381 
382 static void unregister_prot_hook(struct sock *sk, bool sync)
383 {
384 	struct packet_sock *po = pkt_sk(sk);
385 
386 	if (po->running)
387 		__unregister_prot_hook(sk, sync);
388 }
389 
390 static inline struct page * __pure pgv_to_page(void *addr)
391 {
392 	if (is_vmalloc_addr(addr))
393 		return vmalloc_to_page(addr);
394 	return virt_to_page(addr);
395 }
396 
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
398 {
399 	union tpacket_uhdr h;
400 
401 	h.raw = frame;
402 	switch (po->tp_version) {
403 	case TPACKET_V1:
404 		h.h1->tp_status = status;
405 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
406 		break;
407 	case TPACKET_V2:
408 		h.h2->tp_status = status;
409 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
410 		break;
411 	case TPACKET_V3:
412 	default:
413 		WARN(1, "TPACKET version not supported.\n");
414 		BUG();
415 	}
416 
417 	smp_wmb();
418 }
419 
420 static int __packet_get_status(struct packet_sock *po, void *frame)
421 {
422 	union tpacket_uhdr h;
423 
424 	smp_rmb();
425 
426 	h.raw = frame;
427 	switch (po->tp_version) {
428 	case TPACKET_V1:
429 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
430 		return h.h1->tp_status;
431 	case TPACKET_V2:
432 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
433 		return h.h2->tp_status;
434 	case TPACKET_V3:
435 	default:
436 		WARN(1, "TPACKET version not supported.\n");
437 		BUG();
438 		return 0;
439 	}
440 }
441 
442 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
443 				   unsigned int flags)
444 {
445 	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
446 
447 	if (shhwtstamps &&
448 	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 	    ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450 		return TP_STATUS_TS_RAW_HARDWARE;
451 
452 	if (ktime_to_timespec_cond(skb->tstamp, ts))
453 		return TP_STATUS_TS_SOFTWARE;
454 
455 	return 0;
456 }
457 
458 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
459 				    struct sk_buff *skb)
460 {
461 	union tpacket_uhdr h;
462 	struct timespec ts;
463 	__u32 ts_status;
464 
465 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
466 		return 0;
467 
468 	h.raw = frame;
469 	switch (po->tp_version) {
470 	case TPACKET_V1:
471 		h.h1->tp_sec = ts.tv_sec;
472 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
473 		break;
474 	case TPACKET_V2:
475 		h.h2->tp_sec = ts.tv_sec;
476 		h.h2->tp_nsec = ts.tv_nsec;
477 		break;
478 	case TPACKET_V3:
479 	default:
480 		WARN(1, "TPACKET version not supported.\n");
481 		BUG();
482 	}
483 
484 	/* one flush is safe, as both fields always lie on the same cacheline */
485 	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
486 	smp_wmb();
487 
488 	return ts_status;
489 }
490 
491 static void *packet_lookup_frame(struct packet_sock *po,
492 		struct packet_ring_buffer *rb,
493 		unsigned int position,
494 		int status)
495 {
496 	unsigned int pg_vec_pos, frame_offset;
497 	union tpacket_uhdr h;
498 
499 	pg_vec_pos = position / rb->frames_per_block;
500 	frame_offset = position % rb->frames_per_block;
501 
502 	h.raw = rb->pg_vec[pg_vec_pos].buffer +
503 		(frame_offset * rb->frame_size);
504 
505 	if (status != __packet_get_status(po, h.raw))
506 		return NULL;
507 
508 	return h.raw;
509 }
510 
511 static void *packet_current_frame(struct packet_sock *po,
512 		struct packet_ring_buffer *rb,
513 		int status)
514 {
515 	return packet_lookup_frame(po, rb, rb->head, status);
516 }
517 
518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
519 {
520 	del_timer_sync(&pkc->retire_blk_timer);
521 }
522 
523 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
524 		struct sk_buff_head *rb_queue)
525 {
526 	struct tpacket_kbdq_core *pkc;
527 
528 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
529 
530 	spin_lock_bh(&rb_queue->lock);
531 	pkc->delete_blk_timer = 1;
532 	spin_unlock_bh(&rb_queue->lock);
533 
534 	prb_del_retire_blk_timer(pkc);
535 }
536 
537 static void prb_init_blk_timer(struct packet_sock *po,
538 		struct tpacket_kbdq_core *pkc,
539 		void (*func) (unsigned long))
540 {
541 	init_timer(&pkc->retire_blk_timer);
542 	pkc->retire_blk_timer.data = (long)po;
543 	pkc->retire_blk_timer.function = func;
544 	pkc->retire_blk_timer.expires = jiffies;
545 }
546 
547 static void prb_setup_retire_blk_timer(struct packet_sock *po)
548 {
549 	struct tpacket_kbdq_core *pkc;
550 
551 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552 	prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
553 }
554 
555 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
556 				int blk_size_in_bytes)
557 {
558 	struct net_device *dev;
559 	unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
560 	struct ethtool_cmd ecmd;
561 	int err;
562 	u32 speed;
563 
564 	rtnl_lock();
565 	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
566 	if (unlikely(!dev)) {
567 		rtnl_unlock();
568 		return DEFAULT_PRB_RETIRE_TOV;
569 	}
570 	err = __ethtool_get_settings(dev, &ecmd);
571 	speed = ethtool_cmd_speed(&ecmd);
572 	rtnl_unlock();
573 	if (!err) {
574 		/*
575 		 * If the link speed is so slow you don't really
576 		 * need to worry about perf anyways
577 		 */
578 		if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
579 			return DEFAULT_PRB_RETIRE_TOV;
580 		} else {
581 			msec = 1;
582 			div = speed / 1000;
583 		}
584 	}
585 
586 	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
587 
588 	if (div)
589 		mbits /= div;
590 
591 	tmo = mbits * msec;
592 
593 	if (div)
594 		return tmo+1;
595 	return tmo;
596 }
597 
598 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
599 			union tpacket_req_u *req_u)
600 {
601 	p1->feature_req_word = req_u->req3.tp_feature_req_word;
602 }
603 
604 static void init_prb_bdqc(struct packet_sock *po,
605 			struct packet_ring_buffer *rb,
606 			struct pgv *pg_vec,
607 			union tpacket_req_u *req_u)
608 {
609 	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
610 	struct tpacket_block_desc *pbd;
611 
612 	memset(p1, 0x0, sizeof(*p1));
613 
614 	p1->knxt_seq_num = 1;
615 	p1->pkbdq = pg_vec;
616 	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
617 	p1->pkblk_start	= pg_vec[0].buffer;
618 	p1->kblk_size = req_u->req3.tp_block_size;
619 	p1->knum_blocks	= req_u->req3.tp_block_nr;
620 	p1->hdrlen = po->tp_hdrlen;
621 	p1->version = po->tp_version;
622 	p1->last_kactive_blk_num = 0;
623 	po->stats.stats3.tp_freeze_q_cnt = 0;
624 	if (req_u->req3.tp_retire_blk_tov)
625 		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
626 	else
627 		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
628 						req_u->req3.tp_block_size);
629 	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
630 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
631 
632 	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
633 	prb_init_ft_ops(p1, req_u);
634 	prb_setup_retire_blk_timer(po);
635 	prb_open_block(p1, pbd);
636 }
637 
638 /*  Do NOT update the last_blk_num first.
639  *  Assumes sk_buff_head lock is held.
640  */
641 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
642 {
643 	mod_timer(&pkc->retire_blk_timer,
644 			jiffies + pkc->tov_in_jiffies);
645 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
646 }
647 
648 /*
649  * Timer logic:
650  * 1) We refresh the timer only when we open a block.
651  *    By doing this we don't waste cycles refreshing the timer
652  *	  on packet-by-packet basis.
653  *
654  * With a 1MB block-size, on a 1Gbps line, it will take
655  * i) ~8 ms to fill a block + ii) memcpy etc.
656  * In this cut we are not accounting for the memcpy time.
657  *
658  * So, if the user sets the 'tmo' to 10ms then the timer
659  * will never fire while the block is still getting filled
660  * (which is what we want). However, the user could choose
661  * to close a block early and that's fine.
662  *
663  * But when the timer does fire, we check whether or not to refresh it.
664  * Since the tmo granularity is in msecs, it is not too expensive
665  * to refresh the timer, lets say every '8' msecs.
666  * Either the user can set the 'tmo' or we can derive it based on
667  * a) line-speed and b) block-size.
668  * prb_calc_retire_blk_tmo() calculates the tmo.
669  *
670  */
671 static void prb_retire_rx_blk_timer_expired(unsigned long data)
672 {
673 	struct packet_sock *po = (struct packet_sock *)data;
674 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
675 	unsigned int frozen;
676 	struct tpacket_block_desc *pbd;
677 
678 	spin_lock(&po->sk.sk_receive_queue.lock);
679 
680 	frozen = prb_queue_frozen(pkc);
681 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
682 
683 	if (unlikely(pkc->delete_blk_timer))
684 		goto out;
685 
686 	/* We only need to plug the race when the block is partially filled.
687 	 * tpacket_rcv:
688 	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
689 	 *		copy_bits() is in progress ...
690 	 *		timer fires on other cpu:
691 	 *		we can't retire the current block because copy_bits
692 	 *		is in progress.
693 	 *
694 	 */
695 	if (BLOCK_NUM_PKTS(pbd)) {
696 		while (atomic_read(&pkc->blk_fill_in_prog)) {
697 			/* Waiting for skb_copy_bits to finish... */
698 			cpu_relax();
699 		}
700 	}
701 
702 	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
703 		if (!frozen) {
704 			if (!BLOCK_NUM_PKTS(pbd)) {
705 				/* An empty block. Just refresh the timer. */
706 				goto refresh_timer;
707 			}
708 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
709 			if (!prb_dispatch_next_block(pkc, po))
710 				goto refresh_timer;
711 			else
712 				goto out;
713 		} else {
714 			/* Case 1. Queue was frozen because user-space was
715 			 *	   lagging behind.
716 			 */
717 			if (prb_curr_blk_in_use(pkc, pbd)) {
718 				/*
719 				 * Ok, user-space is still behind.
720 				 * So just refresh the timer.
721 				 */
722 				goto refresh_timer;
723 			} else {
724 			       /* Case 2. queue was frozen,user-space caught up,
725 				* now the link went idle && the timer fired.
726 				* We don't have a block to close.So we open this
727 				* block and restart the timer.
728 				* opening a block thaws the queue,restarts timer
729 				* Thawing/timer-refresh is a side effect.
730 				*/
731 				prb_open_block(pkc, pbd);
732 				goto out;
733 			}
734 		}
735 	}
736 
737 refresh_timer:
738 	_prb_refresh_rx_retire_blk_timer(pkc);
739 
740 out:
741 	spin_unlock(&po->sk.sk_receive_queue.lock);
742 }
743 
744 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
745 		struct tpacket_block_desc *pbd1, __u32 status)
746 {
747 	/* Flush everything minus the block header */
748 
749 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
750 	u8 *start, *end;
751 
752 	start = (u8 *)pbd1;
753 
754 	/* Skip the block header(we know header WILL fit in 4K) */
755 	start += PAGE_SIZE;
756 
757 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
758 	for (; start < end; start += PAGE_SIZE)
759 		flush_dcache_page(pgv_to_page(start));
760 
761 	smp_wmb();
762 #endif
763 
764 	/* Now update the block status. */
765 
766 	BLOCK_STATUS(pbd1) = status;
767 
768 	/* Flush the block header */
769 
770 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
771 	start = (u8 *)pbd1;
772 	flush_dcache_page(pgv_to_page(start));
773 
774 	smp_wmb();
775 #endif
776 }
777 
778 /*
779  * Side effect:
780  *
781  * 1) flush the block
782  * 2) Increment active_blk_num
783  *
784  * Note:We DONT refresh the timer on purpose.
785  *	Because almost always the next block will be opened.
786  */
787 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
788 		struct tpacket_block_desc *pbd1,
789 		struct packet_sock *po, unsigned int stat)
790 {
791 	__u32 status = TP_STATUS_USER | stat;
792 
793 	struct tpacket3_hdr *last_pkt;
794 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
795 	struct sock *sk = &po->sk;
796 
797 	if (po->stats.stats3.tp_drops)
798 		status |= TP_STATUS_LOSING;
799 
800 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
801 	last_pkt->tp_next_offset = 0;
802 
803 	/* Get the ts of the last pkt */
804 	if (BLOCK_NUM_PKTS(pbd1)) {
805 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
806 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
807 	} else {
808 		/* Ok, we tmo'd - so get the current time.
809 		 *
810 		 * It shouldn't really happen as we don't close empty
811 		 * blocks. See prb_retire_rx_blk_timer_expired().
812 		 */
813 		struct timespec ts;
814 		getnstimeofday(&ts);
815 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
816 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
817 	}
818 
819 	smp_wmb();
820 
821 	/* Flush the block */
822 	prb_flush_block(pkc1, pbd1, status);
823 
824 	sk->sk_data_ready(sk);
825 
826 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
827 }
828 
829 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
830 {
831 	pkc->reset_pending_on_curr_blk = 0;
832 }
833 
834 /*
835  * Side effect of opening a block:
836  *
837  * 1) prb_queue is thawed.
838  * 2) retire_blk_timer is refreshed.
839  *
840  */
841 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
842 	struct tpacket_block_desc *pbd1)
843 {
844 	struct timespec ts;
845 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
846 
847 	smp_rmb();
848 
849 	/* We could have just memset this but we will lose the
850 	 * flexibility of making the priv area sticky
851 	 */
852 
853 	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
854 	BLOCK_NUM_PKTS(pbd1) = 0;
855 	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
856 
857 	getnstimeofday(&ts);
858 
859 	h1->ts_first_pkt.ts_sec = ts.tv_sec;
860 	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
861 
862 	pkc1->pkblk_start = (char *)pbd1;
863 	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864 
865 	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
867 
868 	pbd1->version = pkc1->version;
869 	pkc1->prev = pkc1->nxt_offset;
870 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
871 
872 	prb_thaw_queue(pkc1);
873 	_prb_refresh_rx_retire_blk_timer(pkc1);
874 
875 	smp_wmb();
876 }
877 
878 /*
879  * Queue freeze logic:
880  * 1) Assume tp_block_nr = 8 blocks.
881  * 2) At time 't0', user opens Rx ring.
882  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
883  * 4) user-space is either sleeping or processing block '0'.
884  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
885  *    it will close block-7,loop around and try to fill block '0'.
886  *    call-flow:
887  *    __packet_lookup_frame_in_block
888  *      prb_retire_current_block()
889  *      prb_dispatch_next_block()
890  *        |->(BLOCK_STATUS == USER) evaluates to true
891  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
892  * 6) Now there are two cases:
893  *    6.1) Link goes idle right after the queue is frozen.
894  *         But remember, the last open_block() refreshed the timer.
895  *         When this timer expires,it will refresh itself so that we can
896  *         re-open block-0 in near future.
897  *    6.2) Link is busy and keeps on receiving packets. This is a simple
898  *         case and __packet_lookup_frame_in_block will check if block-0
899  *         is free and can now be re-used.
900  */
901 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
902 				  struct packet_sock *po)
903 {
904 	pkc->reset_pending_on_curr_blk = 1;
905 	po->stats.stats3.tp_freeze_q_cnt++;
906 }
907 
908 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
909 
910 /*
911  * If the next block is free then we will dispatch it
912  * and return a good offset.
913  * Else, we will freeze the queue.
914  * So, caller must check the return value.
915  */
916 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
917 		struct packet_sock *po)
918 {
919 	struct tpacket_block_desc *pbd;
920 
921 	smp_rmb();
922 
923 	/* 1. Get current block num */
924 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
925 
926 	/* 2. If this block is currently in_use then freeze the queue */
927 	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
928 		prb_freeze_queue(pkc, po);
929 		return NULL;
930 	}
931 
932 	/*
933 	 * 3.
934 	 * open this block and return the offset where the first packet
935 	 * needs to get stored.
936 	 */
937 	prb_open_block(pkc, pbd);
938 	return (void *)pkc->nxt_offset;
939 }
940 
941 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
942 		struct packet_sock *po, unsigned int status)
943 {
944 	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
945 
946 	/* retire/close the current block */
947 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
948 		/*
949 		 * Plug the case where copy_bits() is in progress on
950 		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
951 		 * have space to copy the pkt in the current block and
952 		 * called prb_retire_current_block()
953 		 *
954 		 * We don't need to worry about the TMO case because
955 		 * the timer-handler already handled this case.
956 		 */
957 		if (!(status & TP_STATUS_BLK_TMO)) {
958 			while (atomic_read(&pkc->blk_fill_in_prog)) {
959 				/* Waiting for skb_copy_bits to finish... */
960 				cpu_relax();
961 			}
962 		}
963 		prb_close_block(pkc, pbd, po, status);
964 		return;
965 	}
966 }
967 
968 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
969 				      struct tpacket_block_desc *pbd)
970 {
971 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
972 }
973 
974 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
975 {
976 	return pkc->reset_pending_on_curr_blk;
977 }
978 
979 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
980 {
981 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
982 	atomic_dec(&pkc->blk_fill_in_prog);
983 }
984 
985 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
986 			struct tpacket3_hdr *ppd)
987 {
988 	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
989 }
990 
991 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
992 			struct tpacket3_hdr *ppd)
993 {
994 	ppd->hv1.tp_rxhash = 0;
995 }
996 
997 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
998 			struct tpacket3_hdr *ppd)
999 {
1000 	if (skb_vlan_tag_present(pkc->skb)) {
1001 		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1002 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1003 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1004 	} else {
1005 		ppd->hv1.tp_vlan_tci = 0;
1006 		ppd->hv1.tp_vlan_tpid = 0;
1007 		ppd->tp_status = TP_STATUS_AVAILABLE;
1008 	}
1009 }
1010 
1011 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1012 			struct tpacket3_hdr *ppd)
1013 {
1014 	ppd->hv1.tp_padding = 0;
1015 	prb_fill_vlan_info(pkc, ppd);
1016 
1017 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1018 		prb_fill_rxhash(pkc, ppd);
1019 	else
1020 		prb_clear_rxhash(pkc, ppd);
1021 }
1022 
1023 static void prb_fill_curr_block(char *curr,
1024 				struct tpacket_kbdq_core *pkc,
1025 				struct tpacket_block_desc *pbd,
1026 				unsigned int len)
1027 {
1028 	struct tpacket3_hdr *ppd;
1029 
1030 	ppd  = (struct tpacket3_hdr *)curr;
1031 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1032 	pkc->prev = curr;
1033 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1035 	BLOCK_NUM_PKTS(pbd) += 1;
1036 	atomic_inc(&pkc->blk_fill_in_prog);
1037 	prb_run_all_ft_ops(pkc, ppd);
1038 }
1039 
1040 /* Assumes caller has the sk->rx_queue.lock */
1041 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1042 					    struct sk_buff *skb,
1043 						int status,
1044 					    unsigned int len
1045 					    )
1046 {
1047 	struct tpacket_kbdq_core *pkc;
1048 	struct tpacket_block_desc *pbd;
1049 	char *curr, *end;
1050 
1051 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1052 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1053 
1054 	/* Queue is frozen when user space is lagging behind */
1055 	if (prb_queue_frozen(pkc)) {
1056 		/*
1057 		 * Check if that last block which caused the queue to freeze,
1058 		 * is still in_use by user-space.
1059 		 */
1060 		if (prb_curr_blk_in_use(pkc, pbd)) {
1061 			/* Can't record this packet */
1062 			return NULL;
1063 		} else {
1064 			/*
1065 			 * Ok, the block was released by user-space.
1066 			 * Now let's open that block.
1067 			 * opening a block also thaws the queue.
1068 			 * Thawing is a side effect.
1069 			 */
1070 			prb_open_block(pkc, pbd);
1071 		}
1072 	}
1073 
1074 	smp_mb();
1075 	curr = pkc->nxt_offset;
1076 	pkc->skb = skb;
1077 	end = (char *)pbd + pkc->kblk_size;
1078 
1079 	/* first try the current block */
1080 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1081 		prb_fill_curr_block(curr, pkc, pbd, len);
1082 		return (void *)curr;
1083 	}
1084 
1085 	/* Ok, close the current block */
1086 	prb_retire_current_block(pkc, po, 0);
1087 
1088 	/* Now, try to dispatch the next block */
1089 	curr = (char *)prb_dispatch_next_block(pkc, po);
1090 	if (curr) {
1091 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1092 		prb_fill_curr_block(curr, pkc, pbd, len);
1093 		return (void *)curr;
1094 	}
1095 
1096 	/*
1097 	 * No free blocks are available.user_space hasn't caught up yet.
1098 	 * Queue was just frozen and now this packet will get dropped.
1099 	 */
1100 	return NULL;
1101 }
1102 
1103 static void *packet_current_rx_frame(struct packet_sock *po,
1104 					    struct sk_buff *skb,
1105 					    int status, unsigned int len)
1106 {
1107 	char *curr = NULL;
1108 	switch (po->tp_version) {
1109 	case TPACKET_V1:
1110 	case TPACKET_V2:
1111 		curr = packet_lookup_frame(po, &po->rx_ring,
1112 					po->rx_ring.head, status);
1113 		return curr;
1114 	case TPACKET_V3:
1115 		return __packet_lookup_frame_in_block(po, skb, status, len);
1116 	default:
1117 		WARN(1, "TPACKET version not supported\n");
1118 		BUG();
1119 		return NULL;
1120 	}
1121 }
1122 
1123 static void *prb_lookup_block(struct packet_sock *po,
1124 				     struct packet_ring_buffer *rb,
1125 				     unsigned int idx,
1126 				     int status)
1127 {
1128 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1129 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1130 
1131 	if (status != BLOCK_STATUS(pbd))
1132 		return NULL;
1133 	return pbd;
1134 }
1135 
1136 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1137 {
1138 	unsigned int prev;
1139 	if (rb->prb_bdqc.kactive_blk_num)
1140 		prev = rb->prb_bdqc.kactive_blk_num-1;
1141 	else
1142 		prev = rb->prb_bdqc.knum_blocks-1;
1143 	return prev;
1144 }
1145 
1146 /* Assumes caller has held the rx_queue.lock */
1147 static void *__prb_previous_block(struct packet_sock *po,
1148 					 struct packet_ring_buffer *rb,
1149 					 int status)
1150 {
1151 	unsigned int previous = prb_previous_blk_num(rb);
1152 	return prb_lookup_block(po, rb, previous, status);
1153 }
1154 
1155 static void *packet_previous_rx_frame(struct packet_sock *po,
1156 					     struct packet_ring_buffer *rb,
1157 					     int status)
1158 {
1159 	if (po->tp_version <= TPACKET_V2)
1160 		return packet_previous_frame(po, rb, status);
1161 
1162 	return __prb_previous_block(po, rb, status);
1163 }
1164 
1165 static void packet_increment_rx_head(struct packet_sock *po,
1166 					    struct packet_ring_buffer *rb)
1167 {
1168 	switch (po->tp_version) {
1169 	case TPACKET_V1:
1170 	case TPACKET_V2:
1171 		return packet_increment_head(rb);
1172 	case TPACKET_V3:
1173 	default:
1174 		WARN(1, "TPACKET version not supported.\n");
1175 		BUG();
1176 		return;
1177 	}
1178 }
1179 
1180 static void *packet_previous_frame(struct packet_sock *po,
1181 		struct packet_ring_buffer *rb,
1182 		int status)
1183 {
1184 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1185 	return packet_lookup_frame(po, rb, previous, status);
1186 }
1187 
1188 static void packet_increment_head(struct packet_ring_buffer *buff)
1189 {
1190 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1191 }
1192 
1193 static void packet_inc_pending(struct packet_ring_buffer *rb)
1194 {
1195 	this_cpu_inc(*rb->pending_refcnt);
1196 }
1197 
1198 static void packet_dec_pending(struct packet_ring_buffer *rb)
1199 {
1200 	this_cpu_dec(*rb->pending_refcnt);
1201 }
1202 
1203 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1204 {
1205 	unsigned int refcnt = 0;
1206 	int cpu;
1207 
1208 	/* We don't use pending refcount in rx_ring. */
1209 	if (rb->pending_refcnt == NULL)
1210 		return 0;
1211 
1212 	for_each_possible_cpu(cpu)
1213 		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1214 
1215 	return refcnt;
1216 }
1217 
1218 static int packet_alloc_pending(struct packet_sock *po)
1219 {
1220 	po->rx_ring.pending_refcnt = NULL;
1221 
1222 	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1223 	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1224 		return -ENOBUFS;
1225 
1226 	return 0;
1227 }
1228 
1229 static void packet_free_pending(struct packet_sock *po)
1230 {
1231 	free_percpu(po->tx_ring.pending_refcnt);
1232 }
1233 
1234 #define ROOM_POW_OFF	2
1235 #define ROOM_NONE	0x0
1236 #define ROOM_LOW	0x1
1237 #define ROOM_NORMAL	0x2
1238 
1239 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1240 {
1241 	int idx, len;
1242 
1243 	len = po->rx_ring.frame_max + 1;
1244 	idx = po->rx_ring.head;
1245 	if (pow_off)
1246 		idx += len >> pow_off;
1247 	if (idx >= len)
1248 		idx -= len;
1249 	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1250 }
1251 
1252 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1253 {
1254 	int idx, len;
1255 
1256 	len = po->rx_ring.prb_bdqc.knum_blocks;
1257 	idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1258 	if (pow_off)
1259 		idx += len >> pow_off;
1260 	if (idx >= len)
1261 		idx -= len;
1262 	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1263 }
1264 
1265 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266 {
1267 	struct sock *sk = &po->sk;
1268 	int ret = ROOM_NONE;
1269 
1270 	if (po->prot_hook.func != tpacket_rcv) {
1271 		int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1272 					  - (skb ? skb->truesize : 0);
1273 		if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1274 			return ROOM_NORMAL;
1275 		else if (avail > 0)
1276 			return ROOM_LOW;
1277 		else
1278 			return ROOM_NONE;
1279 	}
1280 
1281 	if (po->tp_version == TPACKET_V3) {
1282 		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1283 			ret = ROOM_NORMAL;
1284 		else if (__tpacket_v3_has_room(po, 0))
1285 			ret = ROOM_LOW;
1286 	} else {
1287 		if (__tpacket_has_room(po, ROOM_POW_OFF))
1288 			ret = ROOM_NORMAL;
1289 		else if (__tpacket_has_room(po, 0))
1290 			ret = ROOM_LOW;
1291 	}
1292 
1293 	return ret;
1294 }
1295 
1296 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1297 {
1298 	int ret;
1299 	bool has_room;
1300 
1301 	spin_lock_bh(&po->sk.sk_receive_queue.lock);
1302 	ret = __packet_rcv_has_room(po, skb);
1303 	has_room = ret == ROOM_NORMAL;
1304 	if (po->pressure == has_room)
1305 		po->pressure = !has_room;
1306 	spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1307 
1308 	return ret;
1309 }
1310 
1311 static void packet_sock_destruct(struct sock *sk)
1312 {
1313 	skb_queue_purge(&sk->sk_error_queue);
1314 
1315 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1316 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1317 
1318 	if (!sock_flag(sk, SOCK_DEAD)) {
1319 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1320 		return;
1321 	}
1322 
1323 	sk_refcnt_debug_dec(sk);
1324 }
1325 
1326 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1327 {
1328 	u32 rxhash;
1329 	int i, count = 0;
1330 
1331 	rxhash = skb_get_hash(skb);
1332 	for (i = 0; i < ROLLOVER_HLEN; i++)
1333 		if (po->rollover->history[i] == rxhash)
1334 			count++;
1335 
1336 	po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1337 	return count > (ROLLOVER_HLEN >> 1);
1338 }
1339 
1340 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1341 				      struct sk_buff *skb,
1342 				      unsigned int num)
1343 {
1344 	return reciprocal_scale(skb_get_hash(skb), num);
1345 }
1346 
1347 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1348 				    struct sk_buff *skb,
1349 				    unsigned int num)
1350 {
1351 	unsigned int val = atomic_inc_return(&f->rr_cur);
1352 
1353 	return val % num;
1354 }
1355 
1356 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1357 				     struct sk_buff *skb,
1358 				     unsigned int num)
1359 {
1360 	return smp_processor_id() % num;
1361 }
1362 
1363 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1364 				     struct sk_buff *skb,
1365 				     unsigned int num)
1366 {
1367 	return prandom_u32_max(num);
1368 }
1369 
1370 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1371 					  struct sk_buff *skb,
1372 					  unsigned int idx, bool try_self,
1373 					  unsigned int num)
1374 {
1375 	struct packet_sock *po, *po_next, *po_skip = NULL;
1376 	unsigned int i, j, room = ROOM_NONE;
1377 
1378 	po = pkt_sk(f->arr[idx]);
1379 
1380 	if (try_self) {
1381 		room = packet_rcv_has_room(po, skb);
1382 		if (room == ROOM_NORMAL ||
1383 		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1384 			return idx;
1385 		po_skip = po;
1386 	}
1387 
1388 	i = j = min_t(int, po->rollover->sock, num - 1);
1389 	do {
1390 		po_next = pkt_sk(f->arr[i]);
1391 		if (po_next != po_skip && !po_next->pressure &&
1392 		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1393 			if (i != j)
1394 				po->rollover->sock = i;
1395 			atomic_long_inc(&po->rollover->num);
1396 			if (room == ROOM_LOW)
1397 				atomic_long_inc(&po->rollover->num_huge);
1398 			return i;
1399 		}
1400 
1401 		if (++i == num)
1402 			i = 0;
1403 	} while (i != j);
1404 
1405 	atomic_long_inc(&po->rollover->num_failed);
1406 	return idx;
1407 }
1408 
1409 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1410 				    struct sk_buff *skb,
1411 				    unsigned int num)
1412 {
1413 	return skb_get_queue_mapping(skb) % num;
1414 }
1415 
1416 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1417 				     struct sk_buff *skb,
1418 				     unsigned int num)
1419 {
1420 	struct bpf_prog *prog;
1421 	unsigned int ret = 0;
1422 
1423 	rcu_read_lock();
1424 	prog = rcu_dereference(f->bpf_prog);
1425 	if (prog)
1426 		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1427 	rcu_read_unlock();
1428 
1429 	return ret;
1430 }
1431 
1432 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1433 {
1434 	return f->flags & (flag >> 8);
1435 }
1436 
1437 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1438 			     struct packet_type *pt, struct net_device *orig_dev)
1439 {
1440 	struct packet_fanout *f = pt->af_packet_priv;
1441 	unsigned int num = READ_ONCE(f->num_members);
1442 	struct net *net = read_pnet(&f->net);
1443 	struct packet_sock *po;
1444 	unsigned int idx;
1445 
1446 	if (!net_eq(dev_net(dev), net) || !num) {
1447 		kfree_skb(skb);
1448 		return 0;
1449 	}
1450 
1451 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1452 		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1453 		if (!skb)
1454 			return 0;
1455 	}
1456 	switch (f->type) {
1457 	case PACKET_FANOUT_HASH:
1458 	default:
1459 		idx = fanout_demux_hash(f, skb, num);
1460 		break;
1461 	case PACKET_FANOUT_LB:
1462 		idx = fanout_demux_lb(f, skb, num);
1463 		break;
1464 	case PACKET_FANOUT_CPU:
1465 		idx = fanout_demux_cpu(f, skb, num);
1466 		break;
1467 	case PACKET_FANOUT_RND:
1468 		idx = fanout_demux_rnd(f, skb, num);
1469 		break;
1470 	case PACKET_FANOUT_QM:
1471 		idx = fanout_demux_qm(f, skb, num);
1472 		break;
1473 	case PACKET_FANOUT_ROLLOVER:
1474 		idx = fanout_demux_rollover(f, skb, 0, false, num);
1475 		break;
1476 	case PACKET_FANOUT_CBPF:
1477 	case PACKET_FANOUT_EBPF:
1478 		idx = fanout_demux_bpf(f, skb, num);
1479 		break;
1480 	}
1481 
1482 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1483 		idx = fanout_demux_rollover(f, skb, idx, true, num);
1484 
1485 	po = pkt_sk(f->arr[idx]);
1486 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1487 }
1488 
1489 DEFINE_MUTEX(fanout_mutex);
1490 EXPORT_SYMBOL_GPL(fanout_mutex);
1491 static LIST_HEAD(fanout_list);
1492 
1493 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1494 {
1495 	struct packet_fanout *f = po->fanout;
1496 
1497 	spin_lock(&f->lock);
1498 	f->arr[f->num_members] = sk;
1499 	smp_wmb();
1500 	f->num_members++;
1501 	spin_unlock(&f->lock);
1502 }
1503 
1504 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1505 {
1506 	struct packet_fanout *f = po->fanout;
1507 	int i;
1508 
1509 	spin_lock(&f->lock);
1510 	for (i = 0; i < f->num_members; i++) {
1511 		if (f->arr[i] == sk)
1512 			break;
1513 	}
1514 	BUG_ON(i >= f->num_members);
1515 	f->arr[i] = f->arr[f->num_members - 1];
1516 	f->num_members--;
1517 	spin_unlock(&f->lock);
1518 }
1519 
1520 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1521 {
1522 	if (sk->sk_family != PF_PACKET)
1523 		return false;
1524 
1525 	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1526 }
1527 
1528 static void fanout_init_data(struct packet_fanout *f)
1529 {
1530 	switch (f->type) {
1531 	case PACKET_FANOUT_LB:
1532 		atomic_set(&f->rr_cur, 0);
1533 		break;
1534 	case PACKET_FANOUT_CBPF:
1535 	case PACKET_FANOUT_EBPF:
1536 		RCU_INIT_POINTER(f->bpf_prog, NULL);
1537 		break;
1538 	}
1539 }
1540 
1541 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1542 {
1543 	struct bpf_prog *old;
1544 
1545 	spin_lock(&f->lock);
1546 	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1547 	rcu_assign_pointer(f->bpf_prog, new);
1548 	spin_unlock(&f->lock);
1549 
1550 	if (old) {
1551 		synchronize_net();
1552 		bpf_prog_destroy(old);
1553 	}
1554 }
1555 
1556 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1557 				unsigned int len)
1558 {
1559 	struct bpf_prog *new;
1560 	struct sock_fprog fprog;
1561 	int ret;
1562 
1563 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1564 		return -EPERM;
1565 	if (len != sizeof(fprog))
1566 		return -EINVAL;
1567 	if (copy_from_user(&fprog, data, len))
1568 		return -EFAULT;
1569 
1570 	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1571 	if (ret)
1572 		return ret;
1573 
1574 	__fanout_set_data_bpf(po->fanout, new);
1575 	return 0;
1576 }
1577 
1578 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1579 				unsigned int len)
1580 {
1581 	struct bpf_prog *new;
1582 	u32 fd;
1583 
1584 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1585 		return -EPERM;
1586 	if (len != sizeof(fd))
1587 		return -EINVAL;
1588 	if (copy_from_user(&fd, data, len))
1589 		return -EFAULT;
1590 
1591 	new = bpf_prog_get(fd);
1592 	if (IS_ERR(new))
1593 		return PTR_ERR(new);
1594 	if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1595 		bpf_prog_put(new);
1596 		return -EINVAL;
1597 	}
1598 
1599 	__fanout_set_data_bpf(po->fanout, new);
1600 	return 0;
1601 }
1602 
1603 static int fanout_set_data(struct packet_sock *po, char __user *data,
1604 			   unsigned int len)
1605 {
1606 	switch (po->fanout->type) {
1607 	case PACKET_FANOUT_CBPF:
1608 		return fanout_set_data_cbpf(po, data, len);
1609 	case PACKET_FANOUT_EBPF:
1610 		return fanout_set_data_ebpf(po, data, len);
1611 	default:
1612 		return -EINVAL;
1613 	};
1614 }
1615 
1616 static void fanout_release_data(struct packet_fanout *f)
1617 {
1618 	switch (f->type) {
1619 	case PACKET_FANOUT_CBPF:
1620 	case PACKET_FANOUT_EBPF:
1621 		__fanout_set_data_bpf(f, NULL);
1622 	};
1623 }
1624 
1625 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1626 {
1627 	struct packet_sock *po = pkt_sk(sk);
1628 	struct packet_fanout *f, *match;
1629 	u8 type = type_flags & 0xff;
1630 	u8 flags = type_flags >> 8;
1631 	int err;
1632 
1633 	switch (type) {
1634 	case PACKET_FANOUT_ROLLOVER:
1635 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1636 			return -EINVAL;
1637 	case PACKET_FANOUT_HASH:
1638 	case PACKET_FANOUT_LB:
1639 	case PACKET_FANOUT_CPU:
1640 	case PACKET_FANOUT_RND:
1641 	case PACKET_FANOUT_QM:
1642 	case PACKET_FANOUT_CBPF:
1643 	case PACKET_FANOUT_EBPF:
1644 		break;
1645 	default:
1646 		return -EINVAL;
1647 	}
1648 
1649 	if (!po->running)
1650 		return -EINVAL;
1651 
1652 	if (po->fanout)
1653 		return -EALREADY;
1654 
1655 	if (type == PACKET_FANOUT_ROLLOVER ||
1656 	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1657 		po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1658 		if (!po->rollover)
1659 			return -ENOMEM;
1660 		atomic_long_set(&po->rollover->num, 0);
1661 		atomic_long_set(&po->rollover->num_huge, 0);
1662 		atomic_long_set(&po->rollover->num_failed, 0);
1663 	}
1664 
1665 	mutex_lock(&fanout_mutex);
1666 	match = NULL;
1667 	list_for_each_entry(f, &fanout_list, list) {
1668 		if (f->id == id &&
1669 		    read_pnet(&f->net) == sock_net(sk)) {
1670 			match = f;
1671 			break;
1672 		}
1673 	}
1674 	err = -EINVAL;
1675 	if (match && match->flags != flags)
1676 		goto out;
1677 	if (!match) {
1678 		err = -ENOMEM;
1679 		match = kzalloc(sizeof(*match), GFP_KERNEL);
1680 		if (!match)
1681 			goto out;
1682 		write_pnet(&match->net, sock_net(sk));
1683 		match->id = id;
1684 		match->type = type;
1685 		match->flags = flags;
1686 		INIT_LIST_HEAD(&match->list);
1687 		spin_lock_init(&match->lock);
1688 		atomic_set(&match->sk_ref, 0);
1689 		fanout_init_data(match);
1690 		match->prot_hook.type = po->prot_hook.type;
1691 		match->prot_hook.dev = po->prot_hook.dev;
1692 		match->prot_hook.func = packet_rcv_fanout;
1693 		match->prot_hook.af_packet_priv = match;
1694 		match->prot_hook.id_match = match_fanout_group;
1695 		dev_add_pack(&match->prot_hook);
1696 		list_add(&match->list, &fanout_list);
1697 	}
1698 	err = -EINVAL;
1699 	if (match->type == type &&
1700 	    match->prot_hook.type == po->prot_hook.type &&
1701 	    match->prot_hook.dev == po->prot_hook.dev) {
1702 		err = -ENOSPC;
1703 		if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1704 			__dev_remove_pack(&po->prot_hook);
1705 			po->fanout = match;
1706 			atomic_inc(&match->sk_ref);
1707 			__fanout_link(sk, po);
1708 			err = 0;
1709 		}
1710 	}
1711 out:
1712 	mutex_unlock(&fanout_mutex);
1713 	if (err) {
1714 		kfree(po->rollover);
1715 		po->rollover = NULL;
1716 	}
1717 	return err;
1718 }
1719 
1720 static void fanout_release(struct sock *sk)
1721 {
1722 	struct packet_sock *po = pkt_sk(sk);
1723 	struct packet_fanout *f;
1724 
1725 	f = po->fanout;
1726 	if (!f)
1727 		return;
1728 
1729 	mutex_lock(&fanout_mutex);
1730 	po->fanout = NULL;
1731 
1732 	if (atomic_dec_and_test(&f->sk_ref)) {
1733 		list_del(&f->list);
1734 		dev_remove_pack(&f->prot_hook);
1735 		fanout_release_data(f);
1736 		kfree(f);
1737 	}
1738 	mutex_unlock(&fanout_mutex);
1739 
1740 	if (po->rollover)
1741 		kfree_rcu(po->rollover, rcu);
1742 }
1743 
1744 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1745 					  struct sk_buff *skb)
1746 {
1747 	/* Earlier code assumed this would be a VLAN pkt, double-check
1748 	 * this now that we have the actual packet in hand. We can only
1749 	 * do this check on Ethernet devices.
1750 	 */
1751 	if (unlikely(dev->type != ARPHRD_ETHER))
1752 		return false;
1753 
1754 	skb_reset_mac_header(skb);
1755 	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1756 }
1757 
1758 static const struct proto_ops packet_ops;
1759 
1760 static const struct proto_ops packet_ops_spkt;
1761 
1762 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1763 			   struct packet_type *pt, struct net_device *orig_dev)
1764 {
1765 	struct sock *sk;
1766 	struct sockaddr_pkt *spkt;
1767 
1768 	/*
1769 	 *	When we registered the protocol we saved the socket in the data
1770 	 *	field for just this event.
1771 	 */
1772 
1773 	sk = pt->af_packet_priv;
1774 
1775 	/*
1776 	 *	Yank back the headers [hope the device set this
1777 	 *	right or kerboom...]
1778 	 *
1779 	 *	Incoming packets have ll header pulled,
1780 	 *	push it back.
1781 	 *
1782 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1783 	 *	so that this procedure is noop.
1784 	 */
1785 
1786 	if (skb->pkt_type == PACKET_LOOPBACK)
1787 		goto out;
1788 
1789 	if (!net_eq(dev_net(dev), sock_net(sk)))
1790 		goto out;
1791 
1792 	skb = skb_share_check(skb, GFP_ATOMIC);
1793 	if (skb == NULL)
1794 		goto oom;
1795 
1796 	/* drop any routing info */
1797 	skb_dst_drop(skb);
1798 
1799 	/* drop conntrack reference */
1800 	nf_reset(skb);
1801 
1802 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1803 
1804 	skb_push(skb, skb->data - skb_mac_header(skb));
1805 
1806 	/*
1807 	 *	The SOCK_PACKET socket receives _all_ frames.
1808 	 */
1809 
1810 	spkt->spkt_family = dev->type;
1811 	strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1812 	spkt->spkt_protocol = skb->protocol;
1813 
1814 	/*
1815 	 *	Charge the memory to the socket. This is done specifically
1816 	 *	to prevent sockets using all the memory up.
1817 	 */
1818 
1819 	if (sock_queue_rcv_skb(sk, skb) == 0)
1820 		return 0;
1821 
1822 out:
1823 	kfree_skb(skb);
1824 oom:
1825 	return 0;
1826 }
1827 
1828 
1829 /*
1830  *	Output a raw packet to a device layer. This bypasses all the other
1831  *	protocol layers and you must therefore supply it with a complete frame
1832  */
1833 
1834 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1835 			       size_t len)
1836 {
1837 	struct sock *sk = sock->sk;
1838 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1839 	struct sk_buff *skb = NULL;
1840 	struct net_device *dev;
1841 	__be16 proto = 0;
1842 	int err;
1843 	int extra_len = 0;
1844 
1845 	/*
1846 	 *	Get and verify the address.
1847 	 */
1848 
1849 	if (saddr) {
1850 		if (msg->msg_namelen < sizeof(struct sockaddr))
1851 			return -EINVAL;
1852 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1853 			proto = saddr->spkt_protocol;
1854 	} else
1855 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1856 
1857 	/*
1858 	 *	Find the device first to size check it
1859 	 */
1860 
1861 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1862 retry:
1863 	rcu_read_lock();
1864 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1865 	err = -ENODEV;
1866 	if (dev == NULL)
1867 		goto out_unlock;
1868 
1869 	err = -ENETDOWN;
1870 	if (!(dev->flags & IFF_UP))
1871 		goto out_unlock;
1872 
1873 	/*
1874 	 * You may not queue a frame bigger than the mtu. This is the lowest level
1875 	 * raw protocol and you must do your own fragmentation at this level.
1876 	 */
1877 
1878 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1879 		if (!netif_supports_nofcs(dev)) {
1880 			err = -EPROTONOSUPPORT;
1881 			goto out_unlock;
1882 		}
1883 		extra_len = 4; /* We're doing our own CRC */
1884 	}
1885 
1886 	err = -EMSGSIZE;
1887 	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1888 		goto out_unlock;
1889 
1890 	if (!skb) {
1891 		size_t reserved = LL_RESERVED_SPACE(dev);
1892 		int tlen = dev->needed_tailroom;
1893 		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1894 
1895 		rcu_read_unlock();
1896 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1897 		if (skb == NULL)
1898 			return -ENOBUFS;
1899 		/* FIXME: Save some space for broken drivers that write a hard
1900 		 * header at transmission time by themselves. PPP is the notable
1901 		 * one here. This should really be fixed at the driver level.
1902 		 */
1903 		skb_reserve(skb, reserved);
1904 		skb_reset_network_header(skb);
1905 
1906 		/* Try to align data part correctly */
1907 		if (hhlen) {
1908 			skb->data -= hhlen;
1909 			skb->tail -= hhlen;
1910 			if (len < hhlen)
1911 				skb_reset_network_header(skb);
1912 		}
1913 		err = memcpy_from_msg(skb_put(skb, len), msg, len);
1914 		if (err)
1915 			goto out_free;
1916 		goto retry;
1917 	}
1918 
1919 	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1920 	    !packet_extra_vlan_len_allowed(dev, skb)) {
1921 		err = -EMSGSIZE;
1922 		goto out_unlock;
1923 	}
1924 
1925 	skb->protocol = proto;
1926 	skb->dev = dev;
1927 	skb->priority = sk->sk_priority;
1928 	skb->mark = sk->sk_mark;
1929 
1930 	sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1931 
1932 	if (unlikely(extra_len == 4))
1933 		skb->no_fcs = 1;
1934 
1935 	skb_probe_transport_header(skb, 0);
1936 
1937 	dev_queue_xmit(skb);
1938 	rcu_read_unlock();
1939 	return len;
1940 
1941 out_unlock:
1942 	rcu_read_unlock();
1943 out_free:
1944 	kfree_skb(skb);
1945 	return err;
1946 }
1947 
1948 static unsigned int run_filter(struct sk_buff *skb,
1949 			       const struct sock *sk,
1950 			       unsigned int res)
1951 {
1952 	struct sk_filter *filter;
1953 
1954 	rcu_read_lock();
1955 	filter = rcu_dereference(sk->sk_filter);
1956 	if (filter != NULL)
1957 		res = bpf_prog_run_clear_cb(filter->prog, skb);
1958 	rcu_read_unlock();
1959 
1960 	return res;
1961 }
1962 
1963 /*
1964  * This function makes lazy skb cloning in hope that most of packets
1965  * are discarded by BPF.
1966  *
1967  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1968  * and skb->cb are mangled. It works because (and until) packets
1969  * falling here are owned by current CPU. Output packets are cloned
1970  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1971  * sequencially, so that if we return skb to original state on exit,
1972  * we will not harm anyone.
1973  */
1974 
1975 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1976 		      struct packet_type *pt, struct net_device *orig_dev)
1977 {
1978 	struct sock *sk;
1979 	struct sockaddr_ll *sll;
1980 	struct packet_sock *po;
1981 	u8 *skb_head = skb->data;
1982 	int skb_len = skb->len;
1983 	unsigned int snaplen, res;
1984 
1985 	if (skb->pkt_type == PACKET_LOOPBACK)
1986 		goto drop;
1987 
1988 	sk = pt->af_packet_priv;
1989 	po = pkt_sk(sk);
1990 
1991 	if (!net_eq(dev_net(dev), sock_net(sk)))
1992 		goto drop;
1993 
1994 	skb->dev = dev;
1995 
1996 	if (dev->header_ops) {
1997 		/* The device has an explicit notion of ll header,
1998 		 * exported to higher levels.
1999 		 *
2000 		 * Otherwise, the device hides details of its frame
2001 		 * structure, so that corresponding packet head is
2002 		 * never delivered to user.
2003 		 */
2004 		if (sk->sk_type != SOCK_DGRAM)
2005 			skb_push(skb, skb->data - skb_mac_header(skb));
2006 		else if (skb->pkt_type == PACKET_OUTGOING) {
2007 			/* Special case: outgoing packets have ll header at head */
2008 			skb_pull(skb, skb_network_offset(skb));
2009 		}
2010 	}
2011 
2012 	snaplen = skb->len;
2013 
2014 	res = run_filter(skb, sk, snaplen);
2015 	if (!res)
2016 		goto drop_n_restore;
2017 	if (snaplen > res)
2018 		snaplen = res;
2019 
2020 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2021 		goto drop_n_acct;
2022 
2023 	if (skb_shared(skb)) {
2024 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2025 		if (nskb == NULL)
2026 			goto drop_n_acct;
2027 
2028 		if (skb_head != skb->data) {
2029 			skb->data = skb_head;
2030 			skb->len = skb_len;
2031 		}
2032 		consume_skb(skb);
2033 		skb = nskb;
2034 	}
2035 
2036 	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2037 
2038 	sll = &PACKET_SKB_CB(skb)->sa.ll;
2039 	sll->sll_hatype = dev->type;
2040 	sll->sll_pkttype = skb->pkt_type;
2041 	if (unlikely(po->origdev))
2042 		sll->sll_ifindex = orig_dev->ifindex;
2043 	else
2044 		sll->sll_ifindex = dev->ifindex;
2045 
2046 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2047 
2048 	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2049 	 * Use their space for storing the original skb length.
2050 	 */
2051 	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2052 
2053 	if (pskb_trim(skb, snaplen))
2054 		goto drop_n_acct;
2055 
2056 	skb_set_owner_r(skb, sk);
2057 	skb->dev = NULL;
2058 	skb_dst_drop(skb);
2059 
2060 	/* drop conntrack reference */
2061 	nf_reset(skb);
2062 
2063 	spin_lock(&sk->sk_receive_queue.lock);
2064 	po->stats.stats1.tp_packets++;
2065 	sock_skb_set_dropcount(sk, skb);
2066 	__skb_queue_tail(&sk->sk_receive_queue, skb);
2067 	spin_unlock(&sk->sk_receive_queue.lock);
2068 	sk->sk_data_ready(sk);
2069 	return 0;
2070 
2071 drop_n_acct:
2072 	spin_lock(&sk->sk_receive_queue.lock);
2073 	po->stats.stats1.tp_drops++;
2074 	atomic_inc(&sk->sk_drops);
2075 	spin_unlock(&sk->sk_receive_queue.lock);
2076 
2077 drop_n_restore:
2078 	if (skb_head != skb->data && skb_shared(skb)) {
2079 		skb->data = skb_head;
2080 		skb->len = skb_len;
2081 	}
2082 drop:
2083 	consume_skb(skb);
2084 	return 0;
2085 }
2086 
2087 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2088 		       struct packet_type *pt, struct net_device *orig_dev)
2089 {
2090 	struct sock *sk;
2091 	struct packet_sock *po;
2092 	struct sockaddr_ll *sll;
2093 	union tpacket_uhdr h;
2094 	u8 *skb_head = skb->data;
2095 	int skb_len = skb->len;
2096 	unsigned int snaplen, res;
2097 	unsigned long status = TP_STATUS_USER;
2098 	unsigned short macoff, netoff, hdrlen;
2099 	struct sk_buff *copy_skb = NULL;
2100 	struct timespec ts;
2101 	__u32 ts_status;
2102 
2103 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2104 	 * We may add members to them until current aligned size without forcing
2105 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2106 	 */
2107 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2108 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2109 
2110 	if (skb->pkt_type == PACKET_LOOPBACK)
2111 		goto drop;
2112 
2113 	sk = pt->af_packet_priv;
2114 	po = pkt_sk(sk);
2115 
2116 	if (!net_eq(dev_net(dev), sock_net(sk)))
2117 		goto drop;
2118 
2119 	if (dev->header_ops) {
2120 		if (sk->sk_type != SOCK_DGRAM)
2121 			skb_push(skb, skb->data - skb_mac_header(skb));
2122 		else if (skb->pkt_type == PACKET_OUTGOING) {
2123 			/* Special case: outgoing packets have ll header at head */
2124 			skb_pull(skb, skb_network_offset(skb));
2125 		}
2126 	}
2127 
2128 	snaplen = skb->len;
2129 
2130 	res = run_filter(skb, sk, snaplen);
2131 	if (!res)
2132 		goto drop_n_restore;
2133 
2134 	if (skb->ip_summed == CHECKSUM_PARTIAL)
2135 		status |= TP_STATUS_CSUMNOTREADY;
2136 	else if (skb->pkt_type != PACKET_OUTGOING &&
2137 		 (skb->ip_summed == CHECKSUM_COMPLETE ||
2138 		  skb_csum_unnecessary(skb)))
2139 		status |= TP_STATUS_CSUM_VALID;
2140 
2141 	if (snaplen > res)
2142 		snaplen = res;
2143 
2144 	if (sk->sk_type == SOCK_DGRAM) {
2145 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2146 				  po->tp_reserve;
2147 	} else {
2148 		unsigned int maclen = skb_network_offset(skb);
2149 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2150 				       (maclen < 16 ? 16 : maclen)) +
2151 			po->tp_reserve;
2152 		macoff = netoff - maclen;
2153 	}
2154 	if (po->tp_version <= TPACKET_V2) {
2155 		if (macoff + snaplen > po->rx_ring.frame_size) {
2156 			if (po->copy_thresh &&
2157 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2158 				if (skb_shared(skb)) {
2159 					copy_skb = skb_clone(skb, GFP_ATOMIC);
2160 				} else {
2161 					copy_skb = skb_get(skb);
2162 					skb_head = skb->data;
2163 				}
2164 				if (copy_skb)
2165 					skb_set_owner_r(copy_skb, sk);
2166 			}
2167 			snaplen = po->rx_ring.frame_size - macoff;
2168 			if ((int)snaplen < 0)
2169 				snaplen = 0;
2170 		}
2171 	} else if (unlikely(macoff + snaplen >
2172 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2173 		u32 nval;
2174 
2175 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2176 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2177 			    snaplen, nval, macoff);
2178 		snaplen = nval;
2179 		if (unlikely((int)snaplen < 0)) {
2180 			snaplen = 0;
2181 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2182 		}
2183 	}
2184 	spin_lock(&sk->sk_receive_queue.lock);
2185 	h.raw = packet_current_rx_frame(po, skb,
2186 					TP_STATUS_KERNEL, (macoff+snaplen));
2187 	if (!h.raw)
2188 		goto ring_is_full;
2189 	if (po->tp_version <= TPACKET_V2) {
2190 		packet_increment_rx_head(po, &po->rx_ring);
2191 	/*
2192 	 * LOSING will be reported till you read the stats,
2193 	 * because it's COR - Clear On Read.
2194 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2195 	 * at packet level.
2196 	 */
2197 		if (po->stats.stats1.tp_drops)
2198 			status |= TP_STATUS_LOSING;
2199 	}
2200 	po->stats.stats1.tp_packets++;
2201 	if (copy_skb) {
2202 		status |= TP_STATUS_COPY;
2203 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2204 	}
2205 	spin_unlock(&sk->sk_receive_queue.lock);
2206 
2207 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2208 
2209 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2210 		getnstimeofday(&ts);
2211 
2212 	status |= ts_status;
2213 
2214 	switch (po->tp_version) {
2215 	case TPACKET_V1:
2216 		h.h1->tp_len = skb->len;
2217 		h.h1->tp_snaplen = snaplen;
2218 		h.h1->tp_mac = macoff;
2219 		h.h1->tp_net = netoff;
2220 		h.h1->tp_sec = ts.tv_sec;
2221 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2222 		hdrlen = sizeof(*h.h1);
2223 		break;
2224 	case TPACKET_V2:
2225 		h.h2->tp_len = skb->len;
2226 		h.h2->tp_snaplen = snaplen;
2227 		h.h2->tp_mac = macoff;
2228 		h.h2->tp_net = netoff;
2229 		h.h2->tp_sec = ts.tv_sec;
2230 		h.h2->tp_nsec = ts.tv_nsec;
2231 		if (skb_vlan_tag_present(skb)) {
2232 			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2233 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2234 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2235 		} else {
2236 			h.h2->tp_vlan_tci = 0;
2237 			h.h2->tp_vlan_tpid = 0;
2238 		}
2239 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2240 		hdrlen = sizeof(*h.h2);
2241 		break;
2242 	case TPACKET_V3:
2243 		/* tp_nxt_offset,vlan are already populated above.
2244 		 * So DONT clear those fields here
2245 		 */
2246 		h.h3->tp_status |= status;
2247 		h.h3->tp_len = skb->len;
2248 		h.h3->tp_snaplen = snaplen;
2249 		h.h3->tp_mac = macoff;
2250 		h.h3->tp_net = netoff;
2251 		h.h3->tp_sec  = ts.tv_sec;
2252 		h.h3->tp_nsec = ts.tv_nsec;
2253 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2254 		hdrlen = sizeof(*h.h3);
2255 		break;
2256 	default:
2257 		BUG();
2258 	}
2259 
2260 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2261 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2262 	sll->sll_family = AF_PACKET;
2263 	sll->sll_hatype = dev->type;
2264 	sll->sll_protocol = skb->protocol;
2265 	sll->sll_pkttype = skb->pkt_type;
2266 	if (unlikely(po->origdev))
2267 		sll->sll_ifindex = orig_dev->ifindex;
2268 	else
2269 		sll->sll_ifindex = dev->ifindex;
2270 
2271 	smp_mb();
2272 
2273 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2274 	if (po->tp_version <= TPACKET_V2) {
2275 		u8 *start, *end;
2276 
2277 		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2278 					macoff + snaplen);
2279 
2280 		for (start = h.raw; start < end; start += PAGE_SIZE)
2281 			flush_dcache_page(pgv_to_page(start));
2282 	}
2283 	smp_wmb();
2284 #endif
2285 
2286 	if (po->tp_version <= TPACKET_V2) {
2287 		__packet_set_status(po, h.raw, status);
2288 		sk->sk_data_ready(sk);
2289 	} else {
2290 		prb_clear_blk_fill_status(&po->rx_ring);
2291 	}
2292 
2293 drop_n_restore:
2294 	if (skb_head != skb->data && skb_shared(skb)) {
2295 		skb->data = skb_head;
2296 		skb->len = skb_len;
2297 	}
2298 drop:
2299 	kfree_skb(skb);
2300 	return 0;
2301 
2302 ring_is_full:
2303 	po->stats.stats1.tp_drops++;
2304 	spin_unlock(&sk->sk_receive_queue.lock);
2305 
2306 	sk->sk_data_ready(sk);
2307 	kfree_skb(copy_skb);
2308 	goto drop_n_restore;
2309 }
2310 
2311 static void tpacket_destruct_skb(struct sk_buff *skb)
2312 {
2313 	struct packet_sock *po = pkt_sk(skb->sk);
2314 
2315 	if (likely(po->tx_ring.pg_vec)) {
2316 		void *ph;
2317 		__u32 ts;
2318 
2319 		ph = skb_shinfo(skb)->destructor_arg;
2320 		packet_dec_pending(&po->tx_ring);
2321 
2322 		ts = __packet_set_timestamp(po, ph, skb);
2323 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2324 	}
2325 
2326 	sock_wfree(skb);
2327 }
2328 
2329 static bool ll_header_truncated(const struct net_device *dev, int len)
2330 {
2331 	/* net device doesn't like empty head */
2332 	if (unlikely(len < dev->hard_header_len)) {
2333 		net_warn_ratelimited("%s: packet size is too short (%d < %d)\n",
2334 				     current->comm, len, dev->hard_header_len);
2335 		return true;
2336 	}
2337 
2338 	return false;
2339 }
2340 
2341 static void tpacket_set_protocol(const struct net_device *dev,
2342 				 struct sk_buff *skb)
2343 {
2344 	if (dev->type == ARPHRD_ETHER) {
2345 		skb_reset_mac_header(skb);
2346 		skb->protocol = eth_hdr(skb)->h_proto;
2347 	}
2348 }
2349 
2350 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2351 		void *frame, struct net_device *dev, int size_max,
2352 		__be16 proto, unsigned char *addr, int hlen)
2353 {
2354 	union tpacket_uhdr ph;
2355 	int to_write, offset, len, tp_len, nr_frags, len_max;
2356 	struct socket *sock = po->sk.sk_socket;
2357 	struct page *page;
2358 	void *data;
2359 	int err;
2360 
2361 	ph.raw = frame;
2362 
2363 	skb->protocol = proto;
2364 	skb->dev = dev;
2365 	skb->priority = po->sk.sk_priority;
2366 	skb->mark = po->sk.sk_mark;
2367 	sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2368 	skb_shinfo(skb)->destructor_arg = ph.raw;
2369 
2370 	switch (po->tp_version) {
2371 	case TPACKET_V2:
2372 		tp_len = ph.h2->tp_len;
2373 		break;
2374 	default:
2375 		tp_len = ph.h1->tp_len;
2376 		break;
2377 	}
2378 	if (unlikely(tp_len > size_max)) {
2379 		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2380 		return -EMSGSIZE;
2381 	}
2382 
2383 	skb_reserve(skb, hlen);
2384 	skb_reset_network_header(skb);
2385 
2386 	if (unlikely(po->tp_tx_has_off)) {
2387 		int off_min, off_max, off;
2388 		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2389 		off_max = po->tx_ring.frame_size - tp_len;
2390 		if (sock->type == SOCK_DGRAM) {
2391 			switch (po->tp_version) {
2392 			case TPACKET_V2:
2393 				off = ph.h2->tp_net;
2394 				break;
2395 			default:
2396 				off = ph.h1->tp_net;
2397 				break;
2398 			}
2399 		} else {
2400 			switch (po->tp_version) {
2401 			case TPACKET_V2:
2402 				off = ph.h2->tp_mac;
2403 				break;
2404 			default:
2405 				off = ph.h1->tp_mac;
2406 				break;
2407 			}
2408 		}
2409 		if (unlikely((off < off_min) || (off_max < off)))
2410 			return -EINVAL;
2411 		data = ph.raw + off;
2412 	} else {
2413 		data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2414 	}
2415 	to_write = tp_len;
2416 
2417 	if (sock->type == SOCK_DGRAM) {
2418 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2419 				NULL, tp_len);
2420 		if (unlikely(err < 0))
2421 			return -EINVAL;
2422 	} else if (dev->hard_header_len) {
2423 		if (ll_header_truncated(dev, tp_len))
2424 			return -EINVAL;
2425 
2426 		skb_push(skb, dev->hard_header_len);
2427 		err = skb_store_bits(skb, 0, data,
2428 				dev->hard_header_len);
2429 		if (unlikely(err))
2430 			return err;
2431 		if (!skb->protocol)
2432 			tpacket_set_protocol(dev, skb);
2433 
2434 		data += dev->hard_header_len;
2435 		to_write -= dev->hard_header_len;
2436 	}
2437 
2438 	offset = offset_in_page(data);
2439 	len_max = PAGE_SIZE - offset;
2440 	len = ((to_write > len_max) ? len_max : to_write);
2441 
2442 	skb->data_len = to_write;
2443 	skb->len += to_write;
2444 	skb->truesize += to_write;
2445 	atomic_add(to_write, &po->sk.sk_wmem_alloc);
2446 
2447 	while (likely(to_write)) {
2448 		nr_frags = skb_shinfo(skb)->nr_frags;
2449 
2450 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2451 			pr_err("Packet exceed the number of skb frags(%lu)\n",
2452 			       MAX_SKB_FRAGS);
2453 			return -EFAULT;
2454 		}
2455 
2456 		page = pgv_to_page(data);
2457 		data += len;
2458 		flush_dcache_page(page);
2459 		get_page(page);
2460 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2461 		to_write -= len;
2462 		offset = 0;
2463 		len_max = PAGE_SIZE;
2464 		len = ((to_write > len_max) ? len_max : to_write);
2465 	}
2466 
2467 	skb_probe_transport_header(skb, 0);
2468 
2469 	return tp_len;
2470 }
2471 
2472 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2473 {
2474 	struct sk_buff *skb;
2475 	struct net_device *dev;
2476 	__be16 proto;
2477 	int err, reserve = 0;
2478 	void *ph;
2479 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2480 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2481 	int tp_len, size_max;
2482 	unsigned char *addr;
2483 	int len_sum = 0;
2484 	int status = TP_STATUS_AVAILABLE;
2485 	int hlen, tlen;
2486 
2487 	mutex_lock(&po->pg_vec_lock);
2488 
2489 	if (likely(saddr == NULL)) {
2490 		dev	= packet_cached_dev_get(po);
2491 		proto	= po->num;
2492 		addr	= NULL;
2493 	} else {
2494 		err = -EINVAL;
2495 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2496 			goto out;
2497 		if (msg->msg_namelen < (saddr->sll_halen
2498 					+ offsetof(struct sockaddr_ll,
2499 						sll_addr)))
2500 			goto out;
2501 		proto	= saddr->sll_protocol;
2502 		addr	= saddr->sll_addr;
2503 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2504 	}
2505 
2506 	err = -ENXIO;
2507 	if (unlikely(dev == NULL))
2508 		goto out;
2509 	err = -ENETDOWN;
2510 	if (unlikely(!(dev->flags & IFF_UP)))
2511 		goto out_put;
2512 
2513 	if (po->sk.sk_socket->type == SOCK_RAW)
2514 		reserve = dev->hard_header_len;
2515 	size_max = po->tx_ring.frame_size
2516 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2517 
2518 	if (size_max > dev->mtu + reserve + VLAN_HLEN)
2519 		size_max = dev->mtu + reserve + VLAN_HLEN;
2520 
2521 	do {
2522 		ph = packet_current_frame(po, &po->tx_ring,
2523 					  TP_STATUS_SEND_REQUEST);
2524 		if (unlikely(ph == NULL)) {
2525 			if (need_wait && need_resched())
2526 				schedule();
2527 			continue;
2528 		}
2529 
2530 		status = TP_STATUS_SEND_REQUEST;
2531 		hlen = LL_RESERVED_SPACE(dev);
2532 		tlen = dev->needed_tailroom;
2533 		skb = sock_alloc_send_skb(&po->sk,
2534 				hlen + tlen + sizeof(struct sockaddr_ll),
2535 				!need_wait, &err);
2536 
2537 		if (unlikely(skb == NULL)) {
2538 			/* we assume the socket was initially writeable ... */
2539 			if (likely(len_sum > 0))
2540 				err = len_sum;
2541 			goto out_status;
2542 		}
2543 		tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2544 					  addr, hlen);
2545 		if (likely(tp_len >= 0) &&
2546 		    tp_len > dev->mtu + reserve &&
2547 		    !packet_extra_vlan_len_allowed(dev, skb))
2548 			tp_len = -EMSGSIZE;
2549 
2550 		if (unlikely(tp_len < 0)) {
2551 			if (po->tp_loss) {
2552 				__packet_set_status(po, ph,
2553 						TP_STATUS_AVAILABLE);
2554 				packet_increment_head(&po->tx_ring);
2555 				kfree_skb(skb);
2556 				continue;
2557 			} else {
2558 				status = TP_STATUS_WRONG_FORMAT;
2559 				err = tp_len;
2560 				goto out_status;
2561 			}
2562 		}
2563 
2564 		packet_pick_tx_queue(dev, skb);
2565 
2566 		skb->destructor = tpacket_destruct_skb;
2567 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2568 		packet_inc_pending(&po->tx_ring);
2569 
2570 		status = TP_STATUS_SEND_REQUEST;
2571 		err = po->xmit(skb);
2572 		if (unlikely(err > 0)) {
2573 			err = net_xmit_errno(err);
2574 			if (err && __packet_get_status(po, ph) ==
2575 				   TP_STATUS_AVAILABLE) {
2576 				/* skb was destructed already */
2577 				skb = NULL;
2578 				goto out_status;
2579 			}
2580 			/*
2581 			 * skb was dropped but not destructed yet;
2582 			 * let's treat it like congestion or err < 0
2583 			 */
2584 			err = 0;
2585 		}
2586 		packet_increment_head(&po->tx_ring);
2587 		len_sum += tp_len;
2588 	} while (likely((ph != NULL) ||
2589 		/* Note: packet_read_pending() might be slow if we have
2590 		 * to call it as it's per_cpu variable, but in fast-path
2591 		 * we already short-circuit the loop with the first
2592 		 * condition, and luckily don't have to go that path
2593 		 * anyway.
2594 		 */
2595 		 (need_wait && packet_read_pending(&po->tx_ring))));
2596 
2597 	err = len_sum;
2598 	goto out_put;
2599 
2600 out_status:
2601 	__packet_set_status(po, ph, status);
2602 	kfree_skb(skb);
2603 out_put:
2604 	dev_put(dev);
2605 out:
2606 	mutex_unlock(&po->pg_vec_lock);
2607 	return err;
2608 }
2609 
2610 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2611 				        size_t reserve, size_t len,
2612 				        size_t linear, int noblock,
2613 				        int *err)
2614 {
2615 	struct sk_buff *skb;
2616 
2617 	/* Under a page?  Don't bother with paged skb. */
2618 	if (prepad + len < PAGE_SIZE || !linear)
2619 		linear = len;
2620 
2621 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2622 				   err, 0);
2623 	if (!skb)
2624 		return NULL;
2625 
2626 	skb_reserve(skb, reserve);
2627 	skb_put(skb, linear);
2628 	skb->data_len = len - linear;
2629 	skb->len += len - linear;
2630 
2631 	return skb;
2632 }
2633 
2634 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2635 {
2636 	struct sock *sk = sock->sk;
2637 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2638 	struct sk_buff *skb;
2639 	struct net_device *dev;
2640 	__be16 proto;
2641 	unsigned char *addr;
2642 	int err, reserve = 0;
2643 	struct sockcm_cookie sockc;
2644 	struct virtio_net_hdr vnet_hdr = { 0 };
2645 	int offset = 0;
2646 	int vnet_hdr_len;
2647 	struct packet_sock *po = pkt_sk(sk);
2648 	unsigned short gso_type = 0;
2649 	int hlen, tlen;
2650 	int extra_len = 0;
2651 	ssize_t n;
2652 
2653 	/*
2654 	 *	Get and verify the address.
2655 	 */
2656 
2657 	if (likely(saddr == NULL)) {
2658 		dev	= packet_cached_dev_get(po);
2659 		proto	= po->num;
2660 		addr	= NULL;
2661 	} else {
2662 		err = -EINVAL;
2663 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2664 			goto out;
2665 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2666 			goto out;
2667 		proto	= saddr->sll_protocol;
2668 		addr	= saddr->sll_addr;
2669 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2670 	}
2671 
2672 	err = -ENXIO;
2673 	if (unlikely(dev == NULL))
2674 		goto out_unlock;
2675 	err = -ENETDOWN;
2676 	if (unlikely(!(dev->flags & IFF_UP)))
2677 		goto out_unlock;
2678 
2679 	sockc.mark = sk->sk_mark;
2680 	if (msg->msg_controllen) {
2681 		err = sock_cmsg_send(sk, msg, &sockc);
2682 		if (unlikely(err))
2683 			goto out_unlock;
2684 	}
2685 
2686 	if (sock->type == SOCK_RAW)
2687 		reserve = dev->hard_header_len;
2688 	if (po->has_vnet_hdr) {
2689 		vnet_hdr_len = sizeof(vnet_hdr);
2690 
2691 		err = -EINVAL;
2692 		if (len < vnet_hdr_len)
2693 			goto out_unlock;
2694 
2695 		len -= vnet_hdr_len;
2696 
2697 		err = -EFAULT;
2698 		n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2699 		if (n != vnet_hdr_len)
2700 			goto out_unlock;
2701 
2702 		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2703 		    (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2704 		     __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2705 		      __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2706 			vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2707 				 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2708 				__virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2709 
2710 		err = -EINVAL;
2711 		if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2712 			goto out_unlock;
2713 
2714 		if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2715 			switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2716 			case VIRTIO_NET_HDR_GSO_TCPV4:
2717 				gso_type = SKB_GSO_TCPV4;
2718 				break;
2719 			case VIRTIO_NET_HDR_GSO_TCPV6:
2720 				gso_type = SKB_GSO_TCPV6;
2721 				break;
2722 			case VIRTIO_NET_HDR_GSO_UDP:
2723 				gso_type = SKB_GSO_UDP;
2724 				break;
2725 			default:
2726 				goto out_unlock;
2727 			}
2728 
2729 			if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2730 				gso_type |= SKB_GSO_TCP_ECN;
2731 
2732 			if (vnet_hdr.gso_size == 0)
2733 				goto out_unlock;
2734 
2735 		}
2736 	}
2737 
2738 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2739 		if (!netif_supports_nofcs(dev)) {
2740 			err = -EPROTONOSUPPORT;
2741 			goto out_unlock;
2742 		}
2743 		extra_len = 4; /* We're doing our own CRC */
2744 	}
2745 
2746 	err = -EMSGSIZE;
2747 	if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2748 		goto out_unlock;
2749 
2750 	err = -ENOBUFS;
2751 	hlen = LL_RESERVED_SPACE(dev);
2752 	tlen = dev->needed_tailroom;
2753 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2754 			       __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2755 			       msg->msg_flags & MSG_DONTWAIT, &err);
2756 	if (skb == NULL)
2757 		goto out_unlock;
2758 
2759 	skb_set_network_header(skb, reserve);
2760 
2761 	err = -EINVAL;
2762 	if (sock->type == SOCK_DGRAM) {
2763 		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2764 		if (unlikely(offset < 0))
2765 			goto out_free;
2766 	} else {
2767 		if (ll_header_truncated(dev, len))
2768 			goto out_free;
2769 	}
2770 
2771 	/* Returns -EFAULT on error */
2772 	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2773 	if (err)
2774 		goto out_free;
2775 
2776 	sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2777 
2778 	if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2779 	    !packet_extra_vlan_len_allowed(dev, skb)) {
2780 		err = -EMSGSIZE;
2781 		goto out_free;
2782 	}
2783 
2784 	skb->protocol = proto;
2785 	skb->dev = dev;
2786 	skb->priority = sk->sk_priority;
2787 	skb->mark = sockc.mark;
2788 
2789 	packet_pick_tx_queue(dev, skb);
2790 
2791 	if (po->has_vnet_hdr) {
2792 		if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2793 			u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2794 			u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2795 			if (!skb_partial_csum_set(skb, s, o)) {
2796 				err = -EINVAL;
2797 				goto out_free;
2798 			}
2799 		}
2800 
2801 		skb_shinfo(skb)->gso_size =
2802 			__virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2803 		skb_shinfo(skb)->gso_type = gso_type;
2804 
2805 		/* Header must be checked, and gso_segs computed. */
2806 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2807 		skb_shinfo(skb)->gso_segs = 0;
2808 
2809 		len += vnet_hdr_len;
2810 	}
2811 
2812 	skb_probe_transport_header(skb, reserve);
2813 
2814 	if (unlikely(extra_len == 4))
2815 		skb->no_fcs = 1;
2816 
2817 	err = po->xmit(skb);
2818 	if (err > 0 && (err = net_xmit_errno(err)) != 0)
2819 		goto out_unlock;
2820 
2821 	dev_put(dev);
2822 
2823 	return len;
2824 
2825 out_free:
2826 	kfree_skb(skb);
2827 out_unlock:
2828 	if (dev)
2829 		dev_put(dev);
2830 out:
2831 	return err;
2832 }
2833 
2834 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2835 {
2836 	struct sock *sk = sock->sk;
2837 	struct packet_sock *po = pkt_sk(sk);
2838 
2839 	if (po->tx_ring.pg_vec)
2840 		return tpacket_snd(po, msg);
2841 	else
2842 		return packet_snd(sock, msg, len);
2843 }
2844 
2845 /*
2846  *	Close a PACKET socket. This is fairly simple. We immediately go
2847  *	to 'closed' state and remove our protocol entry in the device list.
2848  */
2849 
2850 static int packet_release(struct socket *sock)
2851 {
2852 	struct sock *sk = sock->sk;
2853 	struct packet_sock *po;
2854 	struct net *net;
2855 	union tpacket_req_u req_u;
2856 
2857 	if (!sk)
2858 		return 0;
2859 
2860 	net = sock_net(sk);
2861 	po = pkt_sk(sk);
2862 
2863 	mutex_lock(&net->packet.sklist_lock);
2864 	sk_del_node_init_rcu(sk);
2865 	mutex_unlock(&net->packet.sklist_lock);
2866 
2867 	preempt_disable();
2868 	sock_prot_inuse_add(net, sk->sk_prot, -1);
2869 	preempt_enable();
2870 
2871 	spin_lock(&po->bind_lock);
2872 	unregister_prot_hook(sk, false);
2873 	packet_cached_dev_reset(po);
2874 
2875 	if (po->prot_hook.dev) {
2876 		dev_put(po->prot_hook.dev);
2877 		po->prot_hook.dev = NULL;
2878 	}
2879 	spin_unlock(&po->bind_lock);
2880 
2881 	packet_flush_mclist(sk);
2882 
2883 	if (po->rx_ring.pg_vec) {
2884 		memset(&req_u, 0, sizeof(req_u));
2885 		packet_set_ring(sk, &req_u, 1, 0);
2886 	}
2887 
2888 	if (po->tx_ring.pg_vec) {
2889 		memset(&req_u, 0, sizeof(req_u));
2890 		packet_set_ring(sk, &req_u, 1, 1);
2891 	}
2892 
2893 	fanout_release(sk);
2894 
2895 	synchronize_net();
2896 	/*
2897 	 *	Now the socket is dead. No more input will appear.
2898 	 */
2899 	sock_orphan(sk);
2900 	sock->sk = NULL;
2901 
2902 	/* Purge queues */
2903 
2904 	skb_queue_purge(&sk->sk_receive_queue);
2905 	packet_free_pending(po);
2906 	sk_refcnt_debug_release(sk);
2907 
2908 	sock_put(sk);
2909 	return 0;
2910 }
2911 
2912 /*
2913  *	Attach a packet hook.
2914  */
2915 
2916 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2917 			  __be16 proto)
2918 {
2919 	struct packet_sock *po = pkt_sk(sk);
2920 	struct net_device *dev_curr;
2921 	__be16 proto_curr;
2922 	bool need_rehook;
2923 	struct net_device *dev = NULL;
2924 	int ret = 0;
2925 	bool unlisted = false;
2926 
2927 	if (po->fanout)
2928 		return -EINVAL;
2929 
2930 	lock_sock(sk);
2931 	spin_lock(&po->bind_lock);
2932 	rcu_read_lock();
2933 
2934 	if (name) {
2935 		dev = dev_get_by_name_rcu(sock_net(sk), name);
2936 		if (!dev) {
2937 			ret = -ENODEV;
2938 			goto out_unlock;
2939 		}
2940 	} else if (ifindex) {
2941 		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2942 		if (!dev) {
2943 			ret = -ENODEV;
2944 			goto out_unlock;
2945 		}
2946 	}
2947 
2948 	if (dev)
2949 		dev_hold(dev);
2950 
2951 	proto_curr = po->prot_hook.type;
2952 	dev_curr = po->prot_hook.dev;
2953 
2954 	need_rehook = proto_curr != proto || dev_curr != dev;
2955 
2956 	if (need_rehook) {
2957 		if (po->running) {
2958 			rcu_read_unlock();
2959 			__unregister_prot_hook(sk, true);
2960 			rcu_read_lock();
2961 			dev_curr = po->prot_hook.dev;
2962 			if (dev)
2963 				unlisted = !dev_get_by_index_rcu(sock_net(sk),
2964 								 dev->ifindex);
2965 		}
2966 
2967 		po->num = proto;
2968 		po->prot_hook.type = proto;
2969 
2970 		if (unlikely(unlisted)) {
2971 			dev_put(dev);
2972 			po->prot_hook.dev = NULL;
2973 			po->ifindex = -1;
2974 			packet_cached_dev_reset(po);
2975 		} else {
2976 			po->prot_hook.dev = dev;
2977 			po->ifindex = dev ? dev->ifindex : 0;
2978 			packet_cached_dev_assign(po, dev);
2979 		}
2980 	}
2981 	if (dev_curr)
2982 		dev_put(dev_curr);
2983 
2984 	if (proto == 0 || !need_rehook)
2985 		goto out_unlock;
2986 
2987 	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
2988 		register_prot_hook(sk);
2989 	} else {
2990 		sk->sk_err = ENETDOWN;
2991 		if (!sock_flag(sk, SOCK_DEAD))
2992 			sk->sk_error_report(sk);
2993 	}
2994 
2995 out_unlock:
2996 	rcu_read_unlock();
2997 	spin_unlock(&po->bind_lock);
2998 	release_sock(sk);
2999 	return ret;
3000 }
3001 
3002 /*
3003  *	Bind a packet socket to a device
3004  */
3005 
3006 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3007 			    int addr_len)
3008 {
3009 	struct sock *sk = sock->sk;
3010 	char name[15];
3011 
3012 	/*
3013 	 *	Check legality
3014 	 */
3015 
3016 	if (addr_len != sizeof(struct sockaddr))
3017 		return -EINVAL;
3018 	strlcpy(name, uaddr->sa_data, sizeof(name));
3019 
3020 	return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3021 }
3022 
3023 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3024 {
3025 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3026 	struct sock *sk = sock->sk;
3027 
3028 	/*
3029 	 *	Check legality
3030 	 */
3031 
3032 	if (addr_len < sizeof(struct sockaddr_ll))
3033 		return -EINVAL;
3034 	if (sll->sll_family != AF_PACKET)
3035 		return -EINVAL;
3036 
3037 	return packet_do_bind(sk, NULL, sll->sll_ifindex,
3038 			      sll->sll_protocol ? : pkt_sk(sk)->num);
3039 }
3040 
3041 static struct proto packet_proto = {
3042 	.name	  = "PACKET",
3043 	.owner	  = THIS_MODULE,
3044 	.obj_size = sizeof(struct packet_sock),
3045 };
3046 
3047 /*
3048  *	Create a packet of type SOCK_PACKET.
3049  */
3050 
3051 static int packet_create(struct net *net, struct socket *sock, int protocol,
3052 			 int kern)
3053 {
3054 	struct sock *sk;
3055 	struct packet_sock *po;
3056 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3057 	int err;
3058 
3059 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3060 		return -EPERM;
3061 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3062 	    sock->type != SOCK_PACKET)
3063 		return -ESOCKTNOSUPPORT;
3064 
3065 	sock->state = SS_UNCONNECTED;
3066 
3067 	err = -ENOBUFS;
3068 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3069 	if (sk == NULL)
3070 		goto out;
3071 
3072 	sock->ops = &packet_ops;
3073 	if (sock->type == SOCK_PACKET)
3074 		sock->ops = &packet_ops_spkt;
3075 
3076 	sock_init_data(sock, sk);
3077 
3078 	po = pkt_sk(sk);
3079 	sk->sk_family = PF_PACKET;
3080 	po->num = proto;
3081 	po->xmit = dev_queue_xmit;
3082 
3083 	err = packet_alloc_pending(po);
3084 	if (err)
3085 		goto out2;
3086 
3087 	packet_cached_dev_reset(po);
3088 
3089 	sk->sk_destruct = packet_sock_destruct;
3090 	sk_refcnt_debug_inc(sk);
3091 
3092 	/*
3093 	 *	Attach a protocol block
3094 	 */
3095 
3096 	spin_lock_init(&po->bind_lock);
3097 	mutex_init(&po->pg_vec_lock);
3098 	po->rollover = NULL;
3099 	po->prot_hook.func = packet_rcv;
3100 
3101 	if (sock->type == SOCK_PACKET)
3102 		po->prot_hook.func = packet_rcv_spkt;
3103 
3104 	po->prot_hook.af_packet_priv = sk;
3105 
3106 	if (proto) {
3107 		po->prot_hook.type = proto;
3108 		register_prot_hook(sk);
3109 	}
3110 
3111 	mutex_lock(&net->packet.sklist_lock);
3112 	sk_add_node_rcu(sk, &net->packet.sklist);
3113 	mutex_unlock(&net->packet.sklist_lock);
3114 
3115 	preempt_disable();
3116 	sock_prot_inuse_add(net, &packet_proto, 1);
3117 	preempt_enable();
3118 
3119 	return 0;
3120 out2:
3121 	sk_free(sk);
3122 out:
3123 	return err;
3124 }
3125 
3126 /*
3127  *	Pull a packet from our receive queue and hand it to the user.
3128  *	If necessary we block.
3129  */
3130 
3131 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3132 			  int flags)
3133 {
3134 	struct sock *sk = sock->sk;
3135 	struct sk_buff *skb;
3136 	int copied, err;
3137 	int vnet_hdr_len = 0;
3138 	unsigned int origlen = 0;
3139 
3140 	err = -EINVAL;
3141 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3142 		goto out;
3143 
3144 #if 0
3145 	/* What error should we return now? EUNATTACH? */
3146 	if (pkt_sk(sk)->ifindex < 0)
3147 		return -ENODEV;
3148 #endif
3149 
3150 	if (flags & MSG_ERRQUEUE) {
3151 		err = sock_recv_errqueue(sk, msg, len,
3152 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3153 		goto out;
3154 	}
3155 
3156 	/*
3157 	 *	Call the generic datagram receiver. This handles all sorts
3158 	 *	of horrible races and re-entrancy so we can forget about it
3159 	 *	in the protocol layers.
3160 	 *
3161 	 *	Now it will return ENETDOWN, if device have just gone down,
3162 	 *	but then it will block.
3163 	 */
3164 
3165 	skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3166 
3167 	/*
3168 	 *	An error occurred so return it. Because skb_recv_datagram()
3169 	 *	handles the blocking we don't see and worry about blocking
3170 	 *	retries.
3171 	 */
3172 
3173 	if (skb == NULL)
3174 		goto out;
3175 
3176 	if (pkt_sk(sk)->pressure)
3177 		packet_rcv_has_room(pkt_sk(sk), NULL);
3178 
3179 	if (pkt_sk(sk)->has_vnet_hdr) {
3180 		struct virtio_net_hdr vnet_hdr = { 0 };
3181 
3182 		err = -EINVAL;
3183 		vnet_hdr_len = sizeof(vnet_hdr);
3184 		if (len < vnet_hdr_len)
3185 			goto out_free;
3186 
3187 		len -= vnet_hdr_len;
3188 
3189 		if (skb_is_gso(skb)) {
3190 			struct skb_shared_info *sinfo = skb_shinfo(skb);
3191 
3192 			/* This is a hint as to how much should be linear. */
3193 			vnet_hdr.hdr_len =
3194 				__cpu_to_virtio16(vio_le(), skb_headlen(skb));
3195 			vnet_hdr.gso_size =
3196 				__cpu_to_virtio16(vio_le(), sinfo->gso_size);
3197 			if (sinfo->gso_type & SKB_GSO_TCPV4)
3198 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3199 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
3200 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3201 			else if (sinfo->gso_type & SKB_GSO_UDP)
3202 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3203 			else if (sinfo->gso_type & SKB_GSO_FCOE)
3204 				goto out_free;
3205 			else
3206 				BUG();
3207 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3208 				vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3209 		} else
3210 			vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3211 
3212 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
3213 			vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3214 			vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3215 					  skb_checksum_start_offset(skb));
3216 			vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3217 							 skb->csum_offset);
3218 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3219 			vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3220 		} /* else everything is zero */
3221 
3222 		err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3223 		if (err < 0)
3224 			goto out_free;
3225 	}
3226 
3227 	/* You lose any data beyond the buffer you gave. If it worries
3228 	 * a user program they can ask the device for its MTU
3229 	 * anyway.
3230 	 */
3231 	copied = skb->len;
3232 	if (copied > len) {
3233 		copied = len;
3234 		msg->msg_flags |= MSG_TRUNC;
3235 	}
3236 
3237 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3238 	if (err)
3239 		goto out_free;
3240 
3241 	if (sock->type != SOCK_PACKET) {
3242 		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3243 
3244 		/* Original length was stored in sockaddr_ll fields */
3245 		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3246 		sll->sll_family = AF_PACKET;
3247 		sll->sll_protocol = skb->protocol;
3248 	}
3249 
3250 	sock_recv_ts_and_drops(msg, sk, skb);
3251 
3252 	if (msg->msg_name) {
3253 		/* If the address length field is there to be filled
3254 		 * in, we fill it in now.
3255 		 */
3256 		if (sock->type == SOCK_PACKET) {
3257 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3258 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3259 		} else {
3260 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3261 
3262 			msg->msg_namelen = sll->sll_halen +
3263 				offsetof(struct sockaddr_ll, sll_addr);
3264 		}
3265 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3266 		       msg->msg_namelen);
3267 	}
3268 
3269 	if (pkt_sk(sk)->auxdata) {
3270 		struct tpacket_auxdata aux;
3271 
3272 		aux.tp_status = TP_STATUS_USER;
3273 		if (skb->ip_summed == CHECKSUM_PARTIAL)
3274 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3275 		else if (skb->pkt_type != PACKET_OUTGOING &&
3276 			 (skb->ip_summed == CHECKSUM_COMPLETE ||
3277 			  skb_csum_unnecessary(skb)))
3278 			aux.tp_status |= TP_STATUS_CSUM_VALID;
3279 
3280 		aux.tp_len = origlen;
3281 		aux.tp_snaplen = skb->len;
3282 		aux.tp_mac = 0;
3283 		aux.tp_net = skb_network_offset(skb);
3284 		if (skb_vlan_tag_present(skb)) {
3285 			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3286 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3287 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3288 		} else {
3289 			aux.tp_vlan_tci = 0;
3290 			aux.tp_vlan_tpid = 0;
3291 		}
3292 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3293 	}
3294 
3295 	/*
3296 	 *	Free or return the buffer as appropriate. Again this
3297 	 *	hides all the races and re-entrancy issues from us.
3298 	 */
3299 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3300 
3301 out_free:
3302 	skb_free_datagram(sk, skb);
3303 out:
3304 	return err;
3305 }
3306 
3307 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3308 			       int *uaddr_len, int peer)
3309 {
3310 	struct net_device *dev;
3311 	struct sock *sk	= sock->sk;
3312 
3313 	if (peer)
3314 		return -EOPNOTSUPP;
3315 
3316 	uaddr->sa_family = AF_PACKET;
3317 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3318 	rcu_read_lock();
3319 	dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3320 	if (dev)
3321 		strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3322 	rcu_read_unlock();
3323 	*uaddr_len = sizeof(*uaddr);
3324 
3325 	return 0;
3326 }
3327 
3328 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3329 			  int *uaddr_len, int peer)
3330 {
3331 	struct net_device *dev;
3332 	struct sock *sk = sock->sk;
3333 	struct packet_sock *po = pkt_sk(sk);
3334 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3335 
3336 	if (peer)
3337 		return -EOPNOTSUPP;
3338 
3339 	sll->sll_family = AF_PACKET;
3340 	sll->sll_ifindex = po->ifindex;
3341 	sll->sll_protocol = po->num;
3342 	sll->sll_pkttype = 0;
3343 	rcu_read_lock();
3344 	dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3345 	if (dev) {
3346 		sll->sll_hatype = dev->type;
3347 		sll->sll_halen = dev->addr_len;
3348 		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3349 	} else {
3350 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3351 		sll->sll_halen = 0;
3352 	}
3353 	rcu_read_unlock();
3354 	*uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3355 
3356 	return 0;
3357 }
3358 
3359 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3360 			 int what)
3361 {
3362 	switch (i->type) {
3363 	case PACKET_MR_MULTICAST:
3364 		if (i->alen != dev->addr_len)
3365 			return -EINVAL;
3366 		if (what > 0)
3367 			return dev_mc_add(dev, i->addr);
3368 		else
3369 			return dev_mc_del(dev, i->addr);
3370 		break;
3371 	case PACKET_MR_PROMISC:
3372 		return dev_set_promiscuity(dev, what);
3373 	case PACKET_MR_ALLMULTI:
3374 		return dev_set_allmulti(dev, what);
3375 	case PACKET_MR_UNICAST:
3376 		if (i->alen != dev->addr_len)
3377 			return -EINVAL;
3378 		if (what > 0)
3379 			return dev_uc_add(dev, i->addr);
3380 		else
3381 			return dev_uc_del(dev, i->addr);
3382 		break;
3383 	default:
3384 		break;
3385 	}
3386 	return 0;
3387 }
3388 
3389 static void packet_dev_mclist_delete(struct net_device *dev,
3390 				     struct packet_mclist **mlp)
3391 {
3392 	struct packet_mclist *ml;
3393 
3394 	while ((ml = *mlp) != NULL) {
3395 		if (ml->ifindex == dev->ifindex) {
3396 			packet_dev_mc(dev, ml, -1);
3397 			*mlp = ml->next;
3398 			kfree(ml);
3399 		} else
3400 			mlp = &ml->next;
3401 	}
3402 }
3403 
3404 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3405 {
3406 	struct packet_sock *po = pkt_sk(sk);
3407 	struct packet_mclist *ml, *i;
3408 	struct net_device *dev;
3409 	int err;
3410 
3411 	rtnl_lock();
3412 
3413 	err = -ENODEV;
3414 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3415 	if (!dev)
3416 		goto done;
3417 
3418 	err = -EINVAL;
3419 	if (mreq->mr_alen > dev->addr_len)
3420 		goto done;
3421 
3422 	err = -ENOBUFS;
3423 	i = kmalloc(sizeof(*i), GFP_KERNEL);
3424 	if (i == NULL)
3425 		goto done;
3426 
3427 	err = 0;
3428 	for (ml = po->mclist; ml; ml = ml->next) {
3429 		if (ml->ifindex == mreq->mr_ifindex &&
3430 		    ml->type == mreq->mr_type &&
3431 		    ml->alen == mreq->mr_alen &&
3432 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3433 			ml->count++;
3434 			/* Free the new element ... */
3435 			kfree(i);
3436 			goto done;
3437 		}
3438 	}
3439 
3440 	i->type = mreq->mr_type;
3441 	i->ifindex = mreq->mr_ifindex;
3442 	i->alen = mreq->mr_alen;
3443 	memcpy(i->addr, mreq->mr_address, i->alen);
3444 	i->count = 1;
3445 	i->next = po->mclist;
3446 	po->mclist = i;
3447 	err = packet_dev_mc(dev, i, 1);
3448 	if (err) {
3449 		po->mclist = i->next;
3450 		kfree(i);
3451 	}
3452 
3453 done:
3454 	rtnl_unlock();
3455 	return err;
3456 }
3457 
3458 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3459 {
3460 	struct packet_mclist *ml, **mlp;
3461 
3462 	rtnl_lock();
3463 
3464 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3465 		if (ml->ifindex == mreq->mr_ifindex &&
3466 		    ml->type == mreq->mr_type &&
3467 		    ml->alen == mreq->mr_alen &&
3468 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3469 			if (--ml->count == 0) {
3470 				struct net_device *dev;
3471 				*mlp = ml->next;
3472 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3473 				if (dev)
3474 					packet_dev_mc(dev, ml, -1);
3475 				kfree(ml);
3476 			}
3477 			break;
3478 		}
3479 	}
3480 	rtnl_unlock();
3481 	return 0;
3482 }
3483 
3484 static void packet_flush_mclist(struct sock *sk)
3485 {
3486 	struct packet_sock *po = pkt_sk(sk);
3487 	struct packet_mclist *ml;
3488 
3489 	if (!po->mclist)
3490 		return;
3491 
3492 	rtnl_lock();
3493 	while ((ml = po->mclist) != NULL) {
3494 		struct net_device *dev;
3495 
3496 		po->mclist = ml->next;
3497 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3498 		if (dev != NULL)
3499 			packet_dev_mc(dev, ml, -1);
3500 		kfree(ml);
3501 	}
3502 	rtnl_unlock();
3503 }
3504 
3505 static int
3506 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3507 {
3508 	struct sock *sk = sock->sk;
3509 	struct packet_sock *po = pkt_sk(sk);
3510 	int ret;
3511 
3512 	if (level != SOL_PACKET)
3513 		return -ENOPROTOOPT;
3514 
3515 	switch (optname) {
3516 	case PACKET_ADD_MEMBERSHIP:
3517 	case PACKET_DROP_MEMBERSHIP:
3518 	{
3519 		struct packet_mreq_max mreq;
3520 		int len = optlen;
3521 		memset(&mreq, 0, sizeof(mreq));
3522 		if (len < sizeof(struct packet_mreq))
3523 			return -EINVAL;
3524 		if (len > sizeof(mreq))
3525 			len = sizeof(mreq);
3526 		if (copy_from_user(&mreq, optval, len))
3527 			return -EFAULT;
3528 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3529 			return -EINVAL;
3530 		if (optname == PACKET_ADD_MEMBERSHIP)
3531 			ret = packet_mc_add(sk, &mreq);
3532 		else
3533 			ret = packet_mc_drop(sk, &mreq);
3534 		return ret;
3535 	}
3536 
3537 	case PACKET_RX_RING:
3538 	case PACKET_TX_RING:
3539 	{
3540 		union tpacket_req_u req_u;
3541 		int len;
3542 
3543 		switch (po->tp_version) {
3544 		case TPACKET_V1:
3545 		case TPACKET_V2:
3546 			len = sizeof(req_u.req);
3547 			break;
3548 		case TPACKET_V3:
3549 		default:
3550 			len = sizeof(req_u.req3);
3551 			break;
3552 		}
3553 		if (optlen < len)
3554 			return -EINVAL;
3555 		if (pkt_sk(sk)->has_vnet_hdr)
3556 			return -EINVAL;
3557 		if (copy_from_user(&req_u.req, optval, len))
3558 			return -EFAULT;
3559 		return packet_set_ring(sk, &req_u, 0,
3560 			optname == PACKET_TX_RING);
3561 	}
3562 	case PACKET_COPY_THRESH:
3563 	{
3564 		int val;
3565 
3566 		if (optlen != sizeof(val))
3567 			return -EINVAL;
3568 		if (copy_from_user(&val, optval, sizeof(val)))
3569 			return -EFAULT;
3570 
3571 		pkt_sk(sk)->copy_thresh = val;
3572 		return 0;
3573 	}
3574 	case PACKET_VERSION:
3575 	{
3576 		int val;
3577 
3578 		if (optlen != sizeof(val))
3579 			return -EINVAL;
3580 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3581 			return -EBUSY;
3582 		if (copy_from_user(&val, optval, sizeof(val)))
3583 			return -EFAULT;
3584 		switch (val) {
3585 		case TPACKET_V1:
3586 		case TPACKET_V2:
3587 		case TPACKET_V3:
3588 			po->tp_version = val;
3589 			return 0;
3590 		default:
3591 			return -EINVAL;
3592 		}
3593 	}
3594 	case PACKET_RESERVE:
3595 	{
3596 		unsigned int val;
3597 
3598 		if (optlen != sizeof(val))
3599 			return -EINVAL;
3600 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3601 			return -EBUSY;
3602 		if (copy_from_user(&val, optval, sizeof(val)))
3603 			return -EFAULT;
3604 		po->tp_reserve = val;
3605 		return 0;
3606 	}
3607 	case PACKET_LOSS:
3608 	{
3609 		unsigned int val;
3610 
3611 		if (optlen != sizeof(val))
3612 			return -EINVAL;
3613 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3614 			return -EBUSY;
3615 		if (copy_from_user(&val, optval, sizeof(val)))
3616 			return -EFAULT;
3617 		po->tp_loss = !!val;
3618 		return 0;
3619 	}
3620 	case PACKET_AUXDATA:
3621 	{
3622 		int val;
3623 
3624 		if (optlen < sizeof(val))
3625 			return -EINVAL;
3626 		if (copy_from_user(&val, optval, sizeof(val)))
3627 			return -EFAULT;
3628 
3629 		po->auxdata = !!val;
3630 		return 0;
3631 	}
3632 	case PACKET_ORIGDEV:
3633 	{
3634 		int val;
3635 
3636 		if (optlen < sizeof(val))
3637 			return -EINVAL;
3638 		if (copy_from_user(&val, optval, sizeof(val)))
3639 			return -EFAULT;
3640 
3641 		po->origdev = !!val;
3642 		return 0;
3643 	}
3644 	case PACKET_VNET_HDR:
3645 	{
3646 		int val;
3647 
3648 		if (sock->type != SOCK_RAW)
3649 			return -EINVAL;
3650 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3651 			return -EBUSY;
3652 		if (optlen < sizeof(val))
3653 			return -EINVAL;
3654 		if (copy_from_user(&val, optval, sizeof(val)))
3655 			return -EFAULT;
3656 
3657 		po->has_vnet_hdr = !!val;
3658 		return 0;
3659 	}
3660 	case PACKET_TIMESTAMP:
3661 	{
3662 		int val;
3663 
3664 		if (optlen != sizeof(val))
3665 			return -EINVAL;
3666 		if (copy_from_user(&val, optval, sizeof(val)))
3667 			return -EFAULT;
3668 
3669 		po->tp_tstamp = val;
3670 		return 0;
3671 	}
3672 	case PACKET_FANOUT:
3673 	{
3674 		int val;
3675 
3676 		if (optlen != sizeof(val))
3677 			return -EINVAL;
3678 		if (copy_from_user(&val, optval, sizeof(val)))
3679 			return -EFAULT;
3680 
3681 		return fanout_add(sk, val & 0xffff, val >> 16);
3682 	}
3683 	case PACKET_FANOUT_DATA:
3684 	{
3685 		if (!po->fanout)
3686 			return -EINVAL;
3687 
3688 		return fanout_set_data(po, optval, optlen);
3689 	}
3690 	case PACKET_TX_HAS_OFF:
3691 	{
3692 		unsigned int val;
3693 
3694 		if (optlen != sizeof(val))
3695 			return -EINVAL;
3696 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3697 			return -EBUSY;
3698 		if (copy_from_user(&val, optval, sizeof(val)))
3699 			return -EFAULT;
3700 		po->tp_tx_has_off = !!val;
3701 		return 0;
3702 	}
3703 	case PACKET_QDISC_BYPASS:
3704 	{
3705 		int val;
3706 
3707 		if (optlen != sizeof(val))
3708 			return -EINVAL;
3709 		if (copy_from_user(&val, optval, sizeof(val)))
3710 			return -EFAULT;
3711 
3712 		po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3713 		return 0;
3714 	}
3715 	default:
3716 		return -ENOPROTOOPT;
3717 	}
3718 }
3719 
3720 static int packet_getsockopt(struct socket *sock, int level, int optname,
3721 			     char __user *optval, int __user *optlen)
3722 {
3723 	int len;
3724 	int val, lv = sizeof(val);
3725 	struct sock *sk = sock->sk;
3726 	struct packet_sock *po = pkt_sk(sk);
3727 	void *data = &val;
3728 	union tpacket_stats_u st;
3729 	struct tpacket_rollover_stats rstats;
3730 
3731 	if (level != SOL_PACKET)
3732 		return -ENOPROTOOPT;
3733 
3734 	if (get_user(len, optlen))
3735 		return -EFAULT;
3736 
3737 	if (len < 0)
3738 		return -EINVAL;
3739 
3740 	switch (optname) {
3741 	case PACKET_STATISTICS:
3742 		spin_lock_bh(&sk->sk_receive_queue.lock);
3743 		memcpy(&st, &po->stats, sizeof(st));
3744 		memset(&po->stats, 0, sizeof(po->stats));
3745 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3746 
3747 		if (po->tp_version == TPACKET_V3) {
3748 			lv = sizeof(struct tpacket_stats_v3);
3749 			st.stats3.tp_packets += st.stats3.tp_drops;
3750 			data = &st.stats3;
3751 		} else {
3752 			lv = sizeof(struct tpacket_stats);
3753 			st.stats1.tp_packets += st.stats1.tp_drops;
3754 			data = &st.stats1;
3755 		}
3756 
3757 		break;
3758 	case PACKET_AUXDATA:
3759 		val = po->auxdata;
3760 		break;
3761 	case PACKET_ORIGDEV:
3762 		val = po->origdev;
3763 		break;
3764 	case PACKET_VNET_HDR:
3765 		val = po->has_vnet_hdr;
3766 		break;
3767 	case PACKET_VERSION:
3768 		val = po->tp_version;
3769 		break;
3770 	case PACKET_HDRLEN:
3771 		if (len > sizeof(int))
3772 			len = sizeof(int);
3773 		if (copy_from_user(&val, optval, len))
3774 			return -EFAULT;
3775 		switch (val) {
3776 		case TPACKET_V1:
3777 			val = sizeof(struct tpacket_hdr);
3778 			break;
3779 		case TPACKET_V2:
3780 			val = sizeof(struct tpacket2_hdr);
3781 			break;
3782 		case TPACKET_V3:
3783 			val = sizeof(struct tpacket3_hdr);
3784 			break;
3785 		default:
3786 			return -EINVAL;
3787 		}
3788 		break;
3789 	case PACKET_RESERVE:
3790 		val = po->tp_reserve;
3791 		break;
3792 	case PACKET_LOSS:
3793 		val = po->tp_loss;
3794 		break;
3795 	case PACKET_TIMESTAMP:
3796 		val = po->tp_tstamp;
3797 		break;
3798 	case PACKET_FANOUT:
3799 		val = (po->fanout ?
3800 		       ((u32)po->fanout->id |
3801 			((u32)po->fanout->type << 16) |
3802 			((u32)po->fanout->flags << 24)) :
3803 		       0);
3804 		break;
3805 	case PACKET_ROLLOVER_STATS:
3806 		if (!po->rollover)
3807 			return -EINVAL;
3808 		rstats.tp_all = atomic_long_read(&po->rollover->num);
3809 		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3810 		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3811 		data = &rstats;
3812 		lv = sizeof(rstats);
3813 		break;
3814 	case PACKET_TX_HAS_OFF:
3815 		val = po->tp_tx_has_off;
3816 		break;
3817 	case PACKET_QDISC_BYPASS:
3818 		val = packet_use_direct_xmit(po);
3819 		break;
3820 	default:
3821 		return -ENOPROTOOPT;
3822 	}
3823 
3824 	if (len > lv)
3825 		len = lv;
3826 	if (put_user(len, optlen))
3827 		return -EFAULT;
3828 	if (copy_to_user(optval, data, len))
3829 		return -EFAULT;
3830 	return 0;
3831 }
3832 
3833 
3834 static int packet_notifier(struct notifier_block *this,
3835 			   unsigned long msg, void *ptr)
3836 {
3837 	struct sock *sk;
3838 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3839 	struct net *net = dev_net(dev);
3840 
3841 	rcu_read_lock();
3842 	sk_for_each_rcu(sk, &net->packet.sklist) {
3843 		struct packet_sock *po = pkt_sk(sk);
3844 
3845 		switch (msg) {
3846 		case NETDEV_UNREGISTER:
3847 			if (po->mclist)
3848 				packet_dev_mclist_delete(dev, &po->mclist);
3849 			/* fallthrough */
3850 
3851 		case NETDEV_DOWN:
3852 			if (dev->ifindex == po->ifindex) {
3853 				spin_lock(&po->bind_lock);
3854 				if (po->running) {
3855 					__unregister_prot_hook(sk, false);
3856 					sk->sk_err = ENETDOWN;
3857 					if (!sock_flag(sk, SOCK_DEAD))
3858 						sk->sk_error_report(sk);
3859 				}
3860 				if (msg == NETDEV_UNREGISTER) {
3861 					packet_cached_dev_reset(po);
3862 					po->ifindex = -1;
3863 					if (po->prot_hook.dev)
3864 						dev_put(po->prot_hook.dev);
3865 					po->prot_hook.dev = NULL;
3866 				}
3867 				spin_unlock(&po->bind_lock);
3868 			}
3869 			break;
3870 		case NETDEV_UP:
3871 			if (dev->ifindex == po->ifindex) {
3872 				spin_lock(&po->bind_lock);
3873 				if (po->num)
3874 					register_prot_hook(sk);
3875 				spin_unlock(&po->bind_lock);
3876 			}
3877 			break;
3878 		}
3879 	}
3880 	rcu_read_unlock();
3881 	return NOTIFY_DONE;
3882 }
3883 
3884 
3885 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3886 			unsigned long arg)
3887 {
3888 	struct sock *sk = sock->sk;
3889 
3890 	switch (cmd) {
3891 	case SIOCOUTQ:
3892 	{
3893 		int amount = sk_wmem_alloc_get(sk);
3894 
3895 		return put_user(amount, (int __user *)arg);
3896 	}
3897 	case SIOCINQ:
3898 	{
3899 		struct sk_buff *skb;
3900 		int amount = 0;
3901 
3902 		spin_lock_bh(&sk->sk_receive_queue.lock);
3903 		skb = skb_peek(&sk->sk_receive_queue);
3904 		if (skb)
3905 			amount = skb->len;
3906 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3907 		return put_user(amount, (int __user *)arg);
3908 	}
3909 	case SIOCGSTAMP:
3910 		return sock_get_timestamp(sk, (struct timeval __user *)arg);
3911 	case SIOCGSTAMPNS:
3912 		return sock_get_timestampns(sk, (struct timespec __user *)arg);
3913 
3914 #ifdef CONFIG_INET
3915 	case SIOCADDRT:
3916 	case SIOCDELRT:
3917 	case SIOCDARP:
3918 	case SIOCGARP:
3919 	case SIOCSARP:
3920 	case SIOCGIFADDR:
3921 	case SIOCSIFADDR:
3922 	case SIOCGIFBRDADDR:
3923 	case SIOCSIFBRDADDR:
3924 	case SIOCGIFNETMASK:
3925 	case SIOCSIFNETMASK:
3926 	case SIOCGIFDSTADDR:
3927 	case SIOCSIFDSTADDR:
3928 	case SIOCSIFFLAGS:
3929 		return inet_dgram_ops.ioctl(sock, cmd, arg);
3930 #endif
3931 
3932 	default:
3933 		return -ENOIOCTLCMD;
3934 	}
3935 	return 0;
3936 }
3937 
3938 static unsigned int packet_poll(struct file *file, struct socket *sock,
3939 				poll_table *wait)
3940 {
3941 	struct sock *sk = sock->sk;
3942 	struct packet_sock *po = pkt_sk(sk);
3943 	unsigned int mask = datagram_poll(file, sock, wait);
3944 
3945 	spin_lock_bh(&sk->sk_receive_queue.lock);
3946 	if (po->rx_ring.pg_vec) {
3947 		if (!packet_previous_rx_frame(po, &po->rx_ring,
3948 			TP_STATUS_KERNEL))
3949 			mask |= POLLIN | POLLRDNORM;
3950 	}
3951 	if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3952 		po->pressure = 0;
3953 	spin_unlock_bh(&sk->sk_receive_queue.lock);
3954 	spin_lock_bh(&sk->sk_write_queue.lock);
3955 	if (po->tx_ring.pg_vec) {
3956 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3957 			mask |= POLLOUT | POLLWRNORM;
3958 	}
3959 	spin_unlock_bh(&sk->sk_write_queue.lock);
3960 	return mask;
3961 }
3962 
3963 
3964 /* Dirty? Well, I still did not learn better way to account
3965  * for user mmaps.
3966  */
3967 
3968 static void packet_mm_open(struct vm_area_struct *vma)
3969 {
3970 	struct file *file = vma->vm_file;
3971 	struct socket *sock = file->private_data;
3972 	struct sock *sk = sock->sk;
3973 
3974 	if (sk)
3975 		atomic_inc(&pkt_sk(sk)->mapped);
3976 }
3977 
3978 static void packet_mm_close(struct vm_area_struct *vma)
3979 {
3980 	struct file *file = vma->vm_file;
3981 	struct socket *sock = file->private_data;
3982 	struct sock *sk = sock->sk;
3983 
3984 	if (sk)
3985 		atomic_dec(&pkt_sk(sk)->mapped);
3986 }
3987 
3988 static const struct vm_operations_struct packet_mmap_ops = {
3989 	.open	=	packet_mm_open,
3990 	.close	=	packet_mm_close,
3991 };
3992 
3993 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3994 			unsigned int len)
3995 {
3996 	int i;
3997 
3998 	for (i = 0; i < len; i++) {
3999 		if (likely(pg_vec[i].buffer)) {
4000 			if (is_vmalloc_addr(pg_vec[i].buffer))
4001 				vfree(pg_vec[i].buffer);
4002 			else
4003 				free_pages((unsigned long)pg_vec[i].buffer,
4004 					   order);
4005 			pg_vec[i].buffer = NULL;
4006 		}
4007 	}
4008 	kfree(pg_vec);
4009 }
4010 
4011 static char *alloc_one_pg_vec_page(unsigned long order)
4012 {
4013 	char *buffer;
4014 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4015 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4016 
4017 	buffer = (char *) __get_free_pages(gfp_flags, order);
4018 	if (buffer)
4019 		return buffer;
4020 
4021 	/* __get_free_pages failed, fall back to vmalloc */
4022 	buffer = vzalloc((1 << order) * PAGE_SIZE);
4023 	if (buffer)
4024 		return buffer;
4025 
4026 	/* vmalloc failed, lets dig into swap here */
4027 	gfp_flags &= ~__GFP_NORETRY;
4028 	buffer = (char *) __get_free_pages(gfp_flags, order);
4029 	if (buffer)
4030 		return buffer;
4031 
4032 	/* complete and utter failure */
4033 	return NULL;
4034 }
4035 
4036 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4037 {
4038 	unsigned int block_nr = req->tp_block_nr;
4039 	struct pgv *pg_vec;
4040 	int i;
4041 
4042 	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4043 	if (unlikely(!pg_vec))
4044 		goto out;
4045 
4046 	for (i = 0; i < block_nr; i++) {
4047 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4048 		if (unlikely(!pg_vec[i].buffer))
4049 			goto out_free_pgvec;
4050 	}
4051 
4052 out:
4053 	return pg_vec;
4054 
4055 out_free_pgvec:
4056 	free_pg_vec(pg_vec, order, block_nr);
4057 	pg_vec = NULL;
4058 	goto out;
4059 }
4060 
4061 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4062 		int closing, int tx_ring)
4063 {
4064 	struct pgv *pg_vec = NULL;
4065 	struct packet_sock *po = pkt_sk(sk);
4066 	int was_running, order = 0;
4067 	struct packet_ring_buffer *rb;
4068 	struct sk_buff_head *rb_queue;
4069 	__be16 num;
4070 	int err = -EINVAL;
4071 	/* Added to avoid minimal code churn */
4072 	struct tpacket_req *req = &req_u->req;
4073 
4074 	/* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4075 	if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4076 		WARN(1, "Tx-ring is not supported.\n");
4077 		goto out;
4078 	}
4079 
4080 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4081 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4082 
4083 	err = -EBUSY;
4084 	if (!closing) {
4085 		if (atomic_read(&po->mapped))
4086 			goto out;
4087 		if (packet_read_pending(rb))
4088 			goto out;
4089 	}
4090 
4091 	if (req->tp_block_nr) {
4092 		/* Sanity tests and some calculations */
4093 		err = -EBUSY;
4094 		if (unlikely(rb->pg_vec))
4095 			goto out;
4096 
4097 		switch (po->tp_version) {
4098 		case TPACKET_V1:
4099 			po->tp_hdrlen = TPACKET_HDRLEN;
4100 			break;
4101 		case TPACKET_V2:
4102 			po->tp_hdrlen = TPACKET2_HDRLEN;
4103 			break;
4104 		case TPACKET_V3:
4105 			po->tp_hdrlen = TPACKET3_HDRLEN;
4106 			break;
4107 		}
4108 
4109 		err = -EINVAL;
4110 		if (unlikely((int)req->tp_block_size <= 0))
4111 			goto out;
4112 		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4113 			goto out;
4114 		if (po->tp_version >= TPACKET_V3 &&
4115 		    (int)(req->tp_block_size -
4116 			  BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4117 			goto out;
4118 		if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4119 					po->tp_reserve))
4120 			goto out;
4121 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4122 			goto out;
4123 
4124 		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4125 		if (unlikely(rb->frames_per_block == 0))
4126 			goto out;
4127 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4128 					req->tp_frame_nr))
4129 			goto out;
4130 
4131 		err = -ENOMEM;
4132 		order = get_order(req->tp_block_size);
4133 		pg_vec = alloc_pg_vec(req, order);
4134 		if (unlikely(!pg_vec))
4135 			goto out;
4136 		switch (po->tp_version) {
4137 		case TPACKET_V3:
4138 		/* Transmit path is not supported. We checked
4139 		 * it above but just being paranoid
4140 		 */
4141 			if (!tx_ring)
4142 				init_prb_bdqc(po, rb, pg_vec, req_u);
4143 			break;
4144 		default:
4145 			break;
4146 		}
4147 	}
4148 	/* Done */
4149 	else {
4150 		err = -EINVAL;
4151 		if (unlikely(req->tp_frame_nr))
4152 			goto out;
4153 	}
4154 
4155 	lock_sock(sk);
4156 
4157 	/* Detach socket from network */
4158 	spin_lock(&po->bind_lock);
4159 	was_running = po->running;
4160 	num = po->num;
4161 	if (was_running) {
4162 		po->num = 0;
4163 		__unregister_prot_hook(sk, false);
4164 	}
4165 	spin_unlock(&po->bind_lock);
4166 
4167 	synchronize_net();
4168 
4169 	err = -EBUSY;
4170 	mutex_lock(&po->pg_vec_lock);
4171 	if (closing || atomic_read(&po->mapped) == 0) {
4172 		err = 0;
4173 		spin_lock_bh(&rb_queue->lock);
4174 		swap(rb->pg_vec, pg_vec);
4175 		rb->frame_max = (req->tp_frame_nr - 1);
4176 		rb->head = 0;
4177 		rb->frame_size = req->tp_frame_size;
4178 		spin_unlock_bh(&rb_queue->lock);
4179 
4180 		swap(rb->pg_vec_order, order);
4181 		swap(rb->pg_vec_len, req->tp_block_nr);
4182 
4183 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4184 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4185 						tpacket_rcv : packet_rcv;
4186 		skb_queue_purge(rb_queue);
4187 		if (atomic_read(&po->mapped))
4188 			pr_err("packet_mmap: vma is busy: %d\n",
4189 			       atomic_read(&po->mapped));
4190 	}
4191 	mutex_unlock(&po->pg_vec_lock);
4192 
4193 	spin_lock(&po->bind_lock);
4194 	if (was_running) {
4195 		po->num = num;
4196 		register_prot_hook(sk);
4197 	}
4198 	spin_unlock(&po->bind_lock);
4199 	if (closing && (po->tp_version > TPACKET_V2)) {
4200 		/* Because we don't support block-based V3 on tx-ring */
4201 		if (!tx_ring)
4202 			prb_shutdown_retire_blk_timer(po, rb_queue);
4203 	}
4204 	release_sock(sk);
4205 
4206 	if (pg_vec)
4207 		free_pg_vec(pg_vec, order, req->tp_block_nr);
4208 out:
4209 	return err;
4210 }
4211 
4212 static int packet_mmap(struct file *file, struct socket *sock,
4213 		struct vm_area_struct *vma)
4214 {
4215 	struct sock *sk = sock->sk;
4216 	struct packet_sock *po = pkt_sk(sk);
4217 	unsigned long size, expected_size;
4218 	struct packet_ring_buffer *rb;
4219 	unsigned long start;
4220 	int err = -EINVAL;
4221 	int i;
4222 
4223 	if (vma->vm_pgoff)
4224 		return -EINVAL;
4225 
4226 	mutex_lock(&po->pg_vec_lock);
4227 
4228 	expected_size = 0;
4229 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4230 		if (rb->pg_vec) {
4231 			expected_size += rb->pg_vec_len
4232 						* rb->pg_vec_pages
4233 						* PAGE_SIZE;
4234 		}
4235 	}
4236 
4237 	if (expected_size == 0)
4238 		goto out;
4239 
4240 	size = vma->vm_end - vma->vm_start;
4241 	if (size != expected_size)
4242 		goto out;
4243 
4244 	start = vma->vm_start;
4245 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4246 		if (rb->pg_vec == NULL)
4247 			continue;
4248 
4249 		for (i = 0; i < rb->pg_vec_len; i++) {
4250 			struct page *page;
4251 			void *kaddr = rb->pg_vec[i].buffer;
4252 			int pg_num;
4253 
4254 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4255 				page = pgv_to_page(kaddr);
4256 				err = vm_insert_page(vma, start, page);
4257 				if (unlikely(err))
4258 					goto out;
4259 				start += PAGE_SIZE;
4260 				kaddr += PAGE_SIZE;
4261 			}
4262 		}
4263 	}
4264 
4265 	atomic_inc(&po->mapped);
4266 	vma->vm_ops = &packet_mmap_ops;
4267 	err = 0;
4268 
4269 out:
4270 	mutex_unlock(&po->pg_vec_lock);
4271 	return err;
4272 }
4273 
4274 static const struct proto_ops packet_ops_spkt = {
4275 	.family =	PF_PACKET,
4276 	.owner =	THIS_MODULE,
4277 	.release =	packet_release,
4278 	.bind =		packet_bind_spkt,
4279 	.connect =	sock_no_connect,
4280 	.socketpair =	sock_no_socketpair,
4281 	.accept =	sock_no_accept,
4282 	.getname =	packet_getname_spkt,
4283 	.poll =		datagram_poll,
4284 	.ioctl =	packet_ioctl,
4285 	.listen =	sock_no_listen,
4286 	.shutdown =	sock_no_shutdown,
4287 	.setsockopt =	sock_no_setsockopt,
4288 	.getsockopt =	sock_no_getsockopt,
4289 	.sendmsg =	packet_sendmsg_spkt,
4290 	.recvmsg =	packet_recvmsg,
4291 	.mmap =		sock_no_mmap,
4292 	.sendpage =	sock_no_sendpage,
4293 };
4294 
4295 static const struct proto_ops packet_ops = {
4296 	.family =	PF_PACKET,
4297 	.owner =	THIS_MODULE,
4298 	.release =	packet_release,
4299 	.bind =		packet_bind,
4300 	.connect =	sock_no_connect,
4301 	.socketpair =	sock_no_socketpair,
4302 	.accept =	sock_no_accept,
4303 	.getname =	packet_getname,
4304 	.poll =		packet_poll,
4305 	.ioctl =	packet_ioctl,
4306 	.listen =	sock_no_listen,
4307 	.shutdown =	sock_no_shutdown,
4308 	.setsockopt =	packet_setsockopt,
4309 	.getsockopt =	packet_getsockopt,
4310 	.sendmsg =	packet_sendmsg,
4311 	.recvmsg =	packet_recvmsg,
4312 	.mmap =		packet_mmap,
4313 	.sendpage =	sock_no_sendpage,
4314 };
4315 
4316 static const struct net_proto_family packet_family_ops = {
4317 	.family =	PF_PACKET,
4318 	.create =	packet_create,
4319 	.owner	=	THIS_MODULE,
4320 };
4321 
4322 static struct notifier_block packet_netdev_notifier = {
4323 	.notifier_call =	packet_notifier,
4324 };
4325 
4326 #ifdef CONFIG_PROC_FS
4327 
4328 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4329 	__acquires(RCU)
4330 {
4331 	struct net *net = seq_file_net(seq);
4332 
4333 	rcu_read_lock();
4334 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4335 }
4336 
4337 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4338 {
4339 	struct net *net = seq_file_net(seq);
4340 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4341 }
4342 
4343 static void packet_seq_stop(struct seq_file *seq, void *v)
4344 	__releases(RCU)
4345 {
4346 	rcu_read_unlock();
4347 }
4348 
4349 static int packet_seq_show(struct seq_file *seq, void *v)
4350 {
4351 	if (v == SEQ_START_TOKEN)
4352 		seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4353 	else {
4354 		struct sock *s = sk_entry(v);
4355 		const struct packet_sock *po = pkt_sk(s);
4356 
4357 		seq_printf(seq,
4358 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4359 			   s,
4360 			   atomic_read(&s->sk_refcnt),
4361 			   s->sk_type,
4362 			   ntohs(po->num),
4363 			   po->ifindex,
4364 			   po->running,
4365 			   atomic_read(&s->sk_rmem_alloc),
4366 			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4367 			   sock_i_ino(s));
4368 	}
4369 
4370 	return 0;
4371 }
4372 
4373 static const struct seq_operations packet_seq_ops = {
4374 	.start	= packet_seq_start,
4375 	.next	= packet_seq_next,
4376 	.stop	= packet_seq_stop,
4377 	.show	= packet_seq_show,
4378 };
4379 
4380 static int packet_seq_open(struct inode *inode, struct file *file)
4381 {
4382 	return seq_open_net(inode, file, &packet_seq_ops,
4383 			    sizeof(struct seq_net_private));
4384 }
4385 
4386 static const struct file_operations packet_seq_fops = {
4387 	.owner		= THIS_MODULE,
4388 	.open		= packet_seq_open,
4389 	.read		= seq_read,
4390 	.llseek		= seq_lseek,
4391 	.release	= seq_release_net,
4392 };
4393 
4394 #endif
4395 
4396 static int __net_init packet_net_init(struct net *net)
4397 {
4398 	mutex_init(&net->packet.sklist_lock);
4399 	INIT_HLIST_HEAD(&net->packet.sklist);
4400 
4401 	if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4402 		return -ENOMEM;
4403 
4404 	return 0;
4405 }
4406 
4407 static void __net_exit packet_net_exit(struct net *net)
4408 {
4409 	remove_proc_entry("packet", net->proc_net);
4410 }
4411 
4412 static struct pernet_operations packet_net_ops = {
4413 	.init = packet_net_init,
4414 	.exit = packet_net_exit,
4415 };
4416 
4417 
4418 static void __exit packet_exit(void)
4419 {
4420 	unregister_netdevice_notifier(&packet_netdev_notifier);
4421 	unregister_pernet_subsys(&packet_net_ops);
4422 	sock_unregister(PF_PACKET);
4423 	proto_unregister(&packet_proto);
4424 }
4425 
4426 static int __init packet_init(void)
4427 {
4428 	int rc = proto_register(&packet_proto, 0);
4429 
4430 	if (rc != 0)
4431 		goto out;
4432 
4433 	sock_register(&packet_family_ops);
4434 	register_pernet_subsys(&packet_net_ops);
4435 	register_netdevice_notifier(&packet_netdev_notifier);
4436 out:
4437 	return rc;
4438 }
4439 
4440 module_init(packet_init);
4441 module_exit(packet_exit);
4442 MODULE_LICENSE("GPL");
4443 MODULE_ALIAS_NETPROTO(PF_PACKET);
4444