xref: /linux/net/packet/af_packet.c (revision 33ff111d7ba3beb86e28938d6382bb5beabd865a)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		PACKET - implements raw packet sockets.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *
13  * Fixes:
14  *		Alan Cox	:	verify_area() now used correctly
15  *		Alan Cox	:	new skbuff lists, look ma no backlogs!
16  *		Alan Cox	:	tidied skbuff lists.
17  *		Alan Cox	:	Now uses generic datagram routines I
18  *					added. Also fixed the peek/read crash
19  *					from all old Linux datagram code.
20  *		Alan Cox	:	Uses the improved datagram code.
21  *		Alan Cox	:	Added NULL's for socket options.
22  *		Alan Cox	:	Re-commented the code.
23  *		Alan Cox	:	Use new kernel side addressing
24  *		Rob Janssen	:	Correct MTU usage.
25  *		Dave Platt	:	Counter leaks caused by incorrect
26  *					interrupt locking and some slightly
27  *					dubious gcc output. Can you read
28  *					compiler: it said _VOLATILE_
29  *	Richard Kooijman	:	Timestamp fixes.
30  *		Alan Cox	:	New buffers. Use sk->mac.raw.
31  *		Alan Cox	:	sendmsg/recvmsg support.
32  *		Alan Cox	:	Protocol setting support
33  *	Alexey Kuznetsov	:	Untied from IPv4 stack.
34  *	Cyrus Durgin		:	Fixed kerneld for kmod.
35  *	Michal Ostrowski        :       Module initialization cleanup.
36  *         Ulises Alonso        :       Frame number limit removal and
37  *                                      packet_set_ring memory leak.
38  *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
39  *					The convention is that longer addresses
40  *					will simply extend the hardware address
41  *					byte arrays at the end of sockaddr_ll
42  *					and packet_mreq.
43  *		Johann Baudy	:	Added TX RING.
44  *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
45  *					layer.
46  *					Copyright (C) 2011, <lokec@ccs.neu.edu>
47  */
48 
49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
50 
51 #include <linux/ethtool.h>
52 #include <linux/uio.h>
53 #include <linux/filter.h>
54 #include <linux/types.h>
55 #include <linux/mm.h>
56 #include <linux/capability.h>
57 #include <linux/fcntl.h>
58 #include <linux/socket.h>
59 #include <linux/in.h>
60 #include <linux/inet.h>
61 #include <linux/netdevice.h>
62 #include <linux/if_packet.h>
63 #include <linux/wireless.h>
64 #include <linux/kernel.h>
65 #include <linux/kmod.h>
66 #include <linux/slab.h>
67 #include <linux/vmalloc.h>
68 #include <net/net_namespace.h>
69 #include <net/ip.h>
70 #include <net/protocol.h>
71 #include <linux/skbuff.h>
72 #include <net/sock.h>
73 #include <linux/errno.h>
74 #include <linux/timer.h>
75 #include <linux/uaccess.h>
76 #include <asm/ioctls.h>
77 #include <asm/page.h>
78 #include <asm/cacheflush.h>
79 #include <asm/io.h>
80 #include <linux/proc_fs.h>
81 #include <linux/seq_file.h>
82 #include <linux/poll.h>
83 #include <linux/module.h>
84 #include <linux/init.h>
85 #include <linux/mutex.h>
86 #include <linux/if_vlan.h>
87 #include <linux/virtio_net.h>
88 #include <linux/errqueue.h>
89 #include <linux/net_tstamp.h>
90 #include <linux/percpu.h>
91 #include <linux/workqueue.h>
92 #ifdef CONFIG_INET
93 #include <net/inet_common.h>
94 #endif
95 #include <linux/bpf.h>
96 #include <net/compat.h>
97 #include <linux/netfilter_netdev.h>
98 
99 #include "internal.h"
100 
101 /*
102    Assumptions:
103    - If the device has no dev->header_ops->create, there is no LL header
104      visible above the device. In this case, its hard_header_len should be 0.
105      The device may prepend its own header internally. In this case, its
106      needed_headroom should be set to the space needed for it to add its
107      internal header.
108      For example, a WiFi driver pretending to be an Ethernet driver should
109      set its hard_header_len to be the Ethernet header length, and set its
110      needed_headroom to be (the real WiFi header length - the fake Ethernet
111      header length).
112    - packet socket receives packets with pulled ll header,
113      so that SOCK_RAW should push it back.
114 
115 On receive:
116 -----------
117 
118 Incoming, dev_has_header(dev) == true
119    mac_header -> ll header
120    data       -> data
121 
122 Outgoing, dev_has_header(dev) == true
123    mac_header -> ll header
124    data       -> ll header
125 
126 Incoming, dev_has_header(dev) == false
127    mac_header -> data
128      However drivers often make it point to the ll header.
129      This is incorrect because the ll header should be invisible to us.
130    data       -> data
131 
132 Outgoing, dev_has_header(dev) == false
133    mac_header -> data. ll header is invisible to us.
134    data       -> data
135 
136 Resume
137   If dev_has_header(dev) == false we are unable to restore the ll header,
138     because it is invisible to us.
139 
140 
141 On transmit:
142 ------------
143 
144 dev_has_header(dev) == true
145    mac_header -> ll header
146    data       -> ll header
147 
148 dev_has_header(dev) == false (ll header is invisible to us)
149    mac_header -> data
150    data       -> data
151 
152    We should set network_header on output to the correct position,
153    packet classifier depends on it.
154  */
155 
156 /* Private packet socket structures. */
157 
158 /* identical to struct packet_mreq except it has
159  * a longer address field.
160  */
161 struct packet_mreq_max {
162 	int		mr_ifindex;
163 	unsigned short	mr_type;
164 	unsigned short	mr_alen;
165 	unsigned char	mr_address[MAX_ADDR_LEN];
166 };
167 
168 union tpacket_uhdr {
169 	struct tpacket_hdr  *h1;
170 	struct tpacket2_hdr *h2;
171 	struct tpacket3_hdr *h3;
172 	void *raw;
173 };
174 
175 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
176 		int closing, int tx_ring);
177 
178 #define V3_ALIGNMENT	(8)
179 
180 #define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
181 
182 #define BLK_PLUS_PRIV(sz_of_priv) \
183 	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
184 
185 #define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
186 #define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
187 #define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
188 #define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
189 #define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
190 #define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
191 
192 struct packet_sock;
193 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
194 		       struct packet_type *pt, struct net_device *orig_dev);
195 
196 static void *packet_previous_frame(struct packet_sock *po,
197 		struct packet_ring_buffer *rb,
198 		int status);
199 static void packet_increment_head(struct packet_ring_buffer *buff);
200 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
202 			struct packet_sock *);
203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
204 		struct packet_sock *, unsigned int status);
205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
206 static void prb_open_block(struct tpacket_kbdq_core *,
207 		struct tpacket_block_desc *);
208 static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *);
209 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
210 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
211 		struct tpacket3_hdr *);
212 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
213 		struct tpacket3_hdr *);
214 static void packet_flush_mclist(struct sock *sk);
215 static u16 packet_pick_tx_queue(struct sk_buff *skb);
216 
217 struct packet_skb_cb {
218 	union {
219 		struct sockaddr_pkt pkt;
220 		union {
221 			/* Trick: alias skb original length with
222 			 * ll.sll_family and ll.protocol in order
223 			 * to save room.
224 			 */
225 			unsigned int origlen;
226 			struct sockaddr_ll ll;
227 		};
228 	} sa;
229 };
230 
231 #define vio_le() virtio_legacy_is_little_endian()
232 
233 #define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
234 
235 #define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
236 #define GET_PBLOCK_DESC(x, bid)	\
237 	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
239 	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
240 #define GET_NEXT_PRB_BLK_NUM(x) \
241 	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
242 	((x)->kactive_blk_num+1) : 0)
243 
244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
245 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246 
247 #ifdef CONFIG_NETFILTER_EGRESS
248 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
249 {
250 	struct sk_buff *next, *head = NULL, *tail;
251 	int rc;
252 
253 	rcu_read_lock();
254 	for (; skb != NULL; skb = next) {
255 		next = skb->next;
256 		skb_mark_not_on_list(skb);
257 
258 		if (!nf_hook_egress(skb, &rc, skb->dev))
259 			continue;
260 
261 		if (!head)
262 			head = skb;
263 		else
264 			tail->next = skb;
265 
266 		tail = skb;
267 	}
268 	rcu_read_unlock();
269 
270 	return head;
271 }
272 #endif
273 
274 static int packet_xmit(const struct packet_sock *po, struct sk_buff *skb)
275 {
276 	if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS))
277 		return dev_queue_xmit(skb);
278 
279 #ifdef CONFIG_NETFILTER_EGRESS
280 	if (nf_hook_egress_active()) {
281 		skb = nf_hook_direct_egress(skb);
282 		if (!skb)
283 			return NET_XMIT_DROP;
284 	}
285 #endif
286 	return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
287 }
288 
289 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
290 {
291 	struct net_device *dev;
292 
293 	rcu_read_lock();
294 	dev = rcu_dereference(po->cached_dev);
295 	dev_hold(dev);
296 	rcu_read_unlock();
297 
298 	return dev;
299 }
300 
301 static void packet_cached_dev_assign(struct packet_sock *po,
302 				     struct net_device *dev)
303 {
304 	rcu_assign_pointer(po->cached_dev, dev);
305 }
306 
307 static void packet_cached_dev_reset(struct packet_sock *po)
308 {
309 	RCU_INIT_POINTER(po->cached_dev, NULL);
310 }
311 
312 static u16 packet_pick_tx_queue(struct sk_buff *skb)
313 {
314 	struct net_device *dev = skb->dev;
315 	const struct net_device_ops *ops = dev->netdev_ops;
316 	int cpu = raw_smp_processor_id();
317 	u16 queue_index;
318 
319 #ifdef CONFIG_XPS
320 	skb->sender_cpu = cpu + 1;
321 #endif
322 	skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
323 	if (ops->ndo_select_queue) {
324 		queue_index = ops->ndo_select_queue(dev, skb, NULL);
325 		queue_index = netdev_cap_txqueue(dev, queue_index);
326 	} else {
327 		queue_index = netdev_pick_tx(dev, skb, NULL);
328 	}
329 
330 	return queue_index;
331 }
332 
333 /* __register_prot_hook must be invoked through register_prot_hook
334  * or from a context in which asynchronous accesses to the packet
335  * socket is not possible (packet_create()).
336  */
337 static void __register_prot_hook(struct sock *sk)
338 {
339 	struct packet_sock *po = pkt_sk(sk);
340 
341 	if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
342 		if (po->fanout)
343 			__fanout_link(sk, po);
344 		else
345 			dev_add_pack(&po->prot_hook);
346 
347 		sock_hold(sk);
348 		packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1);
349 	}
350 }
351 
352 static void register_prot_hook(struct sock *sk)
353 {
354 	lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
355 	__register_prot_hook(sk);
356 }
357 
358 /* If the sync parameter is true, we will temporarily drop
359  * the po->bind_lock and do a synchronize_net to make sure no
360  * asynchronous packet processing paths still refer to the elements
361  * of po->prot_hook.  If the sync parameter is false, it is the
362  * callers responsibility to take care of this.
363  */
364 static void __unregister_prot_hook(struct sock *sk, bool sync)
365 {
366 	struct packet_sock *po = pkt_sk(sk);
367 
368 	lockdep_assert_held_once(&po->bind_lock);
369 
370 	packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0);
371 
372 	if (po->fanout)
373 		__fanout_unlink(sk, po);
374 	else
375 		__dev_remove_pack(&po->prot_hook);
376 
377 	__sock_put(sk);
378 
379 	if (sync) {
380 		spin_unlock(&po->bind_lock);
381 		synchronize_net();
382 		spin_lock(&po->bind_lock);
383 	}
384 }
385 
386 static void unregister_prot_hook(struct sock *sk, bool sync)
387 {
388 	struct packet_sock *po = pkt_sk(sk);
389 
390 	if (packet_sock_flag(po, PACKET_SOCK_RUNNING))
391 		__unregister_prot_hook(sk, sync);
392 }
393 
394 static inline struct page * __pure pgv_to_page(void *addr)
395 {
396 	if (is_vmalloc_addr(addr))
397 		return vmalloc_to_page(addr);
398 	return virt_to_page(addr);
399 }
400 
401 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
402 {
403 	union tpacket_uhdr h;
404 
405 	/* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */
406 
407 	h.raw = frame;
408 	switch (po->tp_version) {
409 	case TPACKET_V1:
410 		WRITE_ONCE(h.h1->tp_status, status);
411 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
412 		break;
413 	case TPACKET_V2:
414 		WRITE_ONCE(h.h2->tp_status, status);
415 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
416 		break;
417 	case TPACKET_V3:
418 		WRITE_ONCE(h.h3->tp_status, status);
419 		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
420 		break;
421 	default:
422 		WARN(1, "TPACKET version not supported.\n");
423 		BUG();
424 	}
425 
426 	smp_wmb();
427 }
428 
429 static int __packet_get_status(const struct packet_sock *po, void *frame)
430 {
431 	union tpacket_uhdr h;
432 
433 	smp_rmb();
434 
435 	/* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */
436 
437 	h.raw = frame;
438 	switch (po->tp_version) {
439 	case TPACKET_V1:
440 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
441 		return READ_ONCE(h.h1->tp_status);
442 	case TPACKET_V2:
443 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
444 		return READ_ONCE(h.h2->tp_status);
445 	case TPACKET_V3:
446 		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
447 		return READ_ONCE(h.h3->tp_status);
448 	default:
449 		WARN(1, "TPACKET version not supported.\n");
450 		BUG();
451 		return 0;
452 	}
453 }
454 
455 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
456 				   unsigned int flags)
457 {
458 	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
459 
460 	if (shhwtstamps &&
461 	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
462 	    ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
463 		return TP_STATUS_TS_RAW_HARDWARE;
464 
465 	if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
466 	    ktime_to_timespec64_cond(skb_tstamp(skb), ts))
467 		return TP_STATUS_TS_SOFTWARE;
468 
469 	return 0;
470 }
471 
472 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
473 				    struct sk_buff *skb)
474 {
475 	union tpacket_uhdr h;
476 	struct timespec64 ts;
477 	__u32 ts_status;
478 
479 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp))))
480 		return 0;
481 
482 	h.raw = frame;
483 	/*
484 	 * versions 1 through 3 overflow the timestamps in y2106, since they
485 	 * all store the seconds in a 32-bit unsigned integer.
486 	 * If we create a version 4, that should have a 64-bit timestamp,
487 	 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
488 	 * nanoseconds.
489 	 */
490 	switch (po->tp_version) {
491 	case TPACKET_V1:
492 		h.h1->tp_sec = ts.tv_sec;
493 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
494 		break;
495 	case TPACKET_V2:
496 		h.h2->tp_sec = ts.tv_sec;
497 		h.h2->tp_nsec = ts.tv_nsec;
498 		break;
499 	case TPACKET_V3:
500 		h.h3->tp_sec = ts.tv_sec;
501 		h.h3->tp_nsec = ts.tv_nsec;
502 		break;
503 	default:
504 		WARN(1, "TPACKET version not supported.\n");
505 		BUG();
506 	}
507 
508 	/* one flush is safe, as both fields always lie on the same cacheline */
509 	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
510 	smp_wmb();
511 
512 	return ts_status;
513 }
514 
515 static void *packet_lookup_frame(const struct packet_sock *po,
516 				 const struct packet_ring_buffer *rb,
517 				 unsigned int position,
518 				 int status)
519 {
520 	unsigned int pg_vec_pos, frame_offset;
521 	union tpacket_uhdr h;
522 
523 	pg_vec_pos = position / rb->frames_per_block;
524 	frame_offset = position % rb->frames_per_block;
525 
526 	h.raw = rb->pg_vec[pg_vec_pos].buffer +
527 		(frame_offset * rb->frame_size);
528 
529 	if (status != __packet_get_status(po, h.raw))
530 		return NULL;
531 
532 	return h.raw;
533 }
534 
535 static void *packet_current_frame(struct packet_sock *po,
536 		struct packet_ring_buffer *rb,
537 		int status)
538 {
539 	return packet_lookup_frame(po, rb, rb->head, status);
540 }
541 
542 static u16 vlan_get_tci(const struct sk_buff *skb, struct net_device *dev)
543 {
544 	struct vlan_hdr vhdr, *vh;
545 	unsigned int header_len;
546 
547 	if (!dev)
548 		return 0;
549 
550 	/* In the SOCK_DGRAM scenario, skb data starts at the network
551 	 * protocol, which is after the VLAN headers. The outer VLAN
552 	 * header is at the hard_header_len offset in non-variable
553 	 * length link layer headers. If it's a VLAN device, the
554 	 * min_header_len should be used to exclude the VLAN header
555 	 * size.
556 	 */
557 	if (dev->min_header_len == dev->hard_header_len)
558 		header_len = dev->hard_header_len;
559 	else if (is_vlan_dev(dev))
560 		header_len = dev->min_header_len;
561 	else
562 		return 0;
563 
564 	vh = skb_header_pointer(skb, skb_mac_offset(skb) + header_len,
565 				sizeof(vhdr), &vhdr);
566 	if (unlikely(!vh))
567 		return 0;
568 
569 	return ntohs(vh->h_vlan_TCI);
570 }
571 
572 static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
573 {
574 	__be16 proto = skb->protocol;
575 
576 	if (unlikely(eth_type_vlan(proto)))
577 		proto = vlan_get_protocol_offset_inline(skb, proto,
578 							skb_mac_offset(skb),
579 							NULL);
580 
581 	return proto;
582 }
583 
584 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
585 		struct sk_buff_head *rb_queue)
586 {
587 	struct tpacket_kbdq_core *pkc;
588 
589 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
590 	hrtimer_cancel(&pkc->retire_blk_timer);
591 }
592 
593 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
594 				int blk_size_in_bytes)
595 {
596 	struct net_device *dev;
597 	unsigned int mbits, div;
598 	struct ethtool_link_ksettings ecmd;
599 	int err;
600 
601 	rtnl_lock();
602 	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
603 	if (unlikely(!dev)) {
604 		rtnl_unlock();
605 		return DEFAULT_PRB_RETIRE_TOV;
606 	}
607 	err = __ethtool_get_link_ksettings(dev, &ecmd);
608 	rtnl_unlock();
609 	if (err)
610 		return DEFAULT_PRB_RETIRE_TOV;
611 
612 	/* If the link speed is so slow you don't really
613 	 * need to worry about perf anyways
614 	 */
615 	if (ecmd.base.speed < SPEED_1000 ||
616 	    ecmd.base.speed == SPEED_UNKNOWN)
617 		return DEFAULT_PRB_RETIRE_TOV;
618 
619 	div = ecmd.base.speed / 1000;
620 	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
621 
622 	if (div)
623 		mbits /= div;
624 
625 	if (div)
626 		return mbits + 1;
627 	return mbits;
628 }
629 
630 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
631 			union tpacket_req_u *req_u)
632 {
633 	p1->feature_req_word = req_u->req3.tp_feature_req_word;
634 }
635 
636 static void init_prb_bdqc(struct packet_sock *po,
637 			struct packet_ring_buffer *rb,
638 			struct pgv *pg_vec,
639 			union tpacket_req_u *req_u)
640 {
641 	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
642 	struct tpacket_block_desc *pbd;
643 
644 	memset(p1, 0x0, sizeof(*p1));
645 
646 	p1->knxt_seq_num = 1;
647 	p1->pkbdq = pg_vec;
648 	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
649 	p1->pkblk_start	= pg_vec[0].buffer;
650 	p1->kblk_size = req_u->req3.tp_block_size;
651 	p1->knum_blocks	= req_u->req3.tp_block_nr;
652 	p1->hdrlen = po->tp_hdrlen;
653 	p1->version = po->tp_version;
654 	po->stats.stats3.tp_freeze_q_cnt = 0;
655 	if (req_u->req3.tp_retire_blk_tov)
656 		p1->interval_ktime = ms_to_ktime(req_u->req3.tp_retire_blk_tov);
657 	else
658 		p1->interval_ktime = ms_to_ktime(prb_calc_retire_blk_tmo(po,
659 						 req_u->req3.tp_block_size));
660 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
661 	rwlock_init(&p1->blk_fill_in_prog_lock);
662 
663 	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
664 	prb_init_ft_ops(p1, req_u);
665 	hrtimer_setup(&p1->retire_blk_timer, prb_retire_rx_blk_timer_expired,
666 		      CLOCK_MONOTONIC, HRTIMER_MODE_REL_SOFT);
667 	hrtimer_start(&p1->retire_blk_timer, p1->interval_ktime,
668 		      HRTIMER_MODE_REL_SOFT);
669 	prb_open_block(p1, pbd);
670 }
671 
672 /*
673  * With a 1MB block-size, on a 1Gbps line, it will take
674  * i) ~8 ms to fill a block + ii) memcpy etc.
675  * In this cut we are not accounting for the memcpy time.
676  *
677  * Since the tmo granularity is in msecs, it is not too expensive
678  * to refresh the timer, lets say every '8' msecs.
679  * Either the user can set the 'tmo' or we can derive it based on
680  * a) line-speed and b) block-size.
681  * prb_calc_retire_blk_tmo() calculates the tmo.
682  */
683 static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *t)
684 {
685 	struct packet_sock *po =
686 		timer_container_of(po, t, rx_ring.prb_bdqc.retire_blk_timer);
687 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
688 	unsigned int frozen;
689 	struct tpacket_block_desc *pbd;
690 
691 	spin_lock(&po->sk.sk_receive_queue.lock);
692 
693 	frozen = prb_queue_frozen(pkc);
694 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
695 
696 	/* We only need to plug the race when the block is partially filled.
697 	 * tpacket_rcv:
698 	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
699 	 *		copy_bits() is in progress ...
700 	 *		timer fires on other cpu:
701 	 *		we can't retire the current block because copy_bits
702 	 *		is in progress.
703 	 *
704 	 */
705 	if (BLOCK_NUM_PKTS(pbd)) {
706 		/* Waiting for skb_copy_bits to finish... */
707 		write_lock(&pkc->blk_fill_in_prog_lock);
708 		write_unlock(&pkc->blk_fill_in_prog_lock);
709 	}
710 
711 	if (!frozen) {
712 		if (BLOCK_NUM_PKTS(pbd)) {
713 			/* Not an empty block. Need retire the block. */
714 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
715 			prb_dispatch_next_block(pkc, po);
716 		}
717 	} else {
718 		/* Case 1. Queue was frozen because user-space was
719 		 * lagging behind.
720 		 */
721 		if (!prb_curr_blk_in_use(pbd)) {
722 			/* Case 2. queue was frozen,user-space caught up,
723 			 * now the link went idle && the timer fired.
724 			 * We don't have a block to close.So we open this
725 			 * block and restart the timer.
726 			 * opening a block thaws the queue,restarts timer
727 			 * Thawing/timer-refresh is a side effect.
728 			 */
729 			prb_open_block(pkc, pbd);
730 		}
731 	}
732 
733 	hrtimer_forward_now(&pkc->retire_blk_timer, pkc->interval_ktime);
734 	spin_unlock(&po->sk.sk_receive_queue.lock);
735 	return HRTIMER_RESTART;
736 }
737 
738 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
739 		struct tpacket_block_desc *pbd1, __u32 status)
740 {
741 	/* Flush everything minus the block header */
742 
743 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
744 	u8 *start, *end;
745 
746 	start = (u8 *)pbd1;
747 
748 	/* Skip the block header(we know header WILL fit in 4K) */
749 	start += PAGE_SIZE;
750 
751 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
752 	for (; start < end; start += PAGE_SIZE)
753 		flush_dcache_page(pgv_to_page(start));
754 
755 	smp_wmb();
756 #endif
757 
758 	/* Now update the block status. */
759 
760 	BLOCK_STATUS(pbd1) = status;
761 
762 	/* Flush the block header */
763 
764 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
765 	start = (u8 *)pbd1;
766 	flush_dcache_page(pgv_to_page(start));
767 
768 	smp_wmb();
769 #endif
770 }
771 
772 /*
773  * Side effect:
774  *
775  * 1) flush the block
776  * 2) Increment active_blk_num
777  *
778  * Note:We DONT refresh the timer on purpose.
779  *	Because almost always the next block will be opened.
780  */
781 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
782 		struct tpacket_block_desc *pbd1,
783 		struct packet_sock *po, unsigned int stat)
784 {
785 	__u32 status = TP_STATUS_USER | stat;
786 
787 	struct tpacket3_hdr *last_pkt;
788 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
789 	struct sock *sk = &po->sk;
790 
791 	if (atomic_read(&po->tp_drops))
792 		status |= TP_STATUS_LOSING;
793 
794 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
795 	last_pkt->tp_next_offset = 0;
796 
797 	/* Get the ts of the last pkt */
798 	if (BLOCK_NUM_PKTS(pbd1)) {
799 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
800 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
801 	} else {
802 		/* Ok, we tmo'd - so get the current time.
803 		 *
804 		 * It shouldn't really happen as we don't close empty
805 		 * blocks. See prb_retire_rx_blk_timer_expired().
806 		 */
807 		struct timespec64 ts;
808 		ktime_get_real_ts64(&ts);
809 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
810 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
811 	}
812 
813 	smp_wmb();
814 
815 	/* Flush the block */
816 	prb_flush_block(pkc1, pbd1, status);
817 
818 	sk->sk_data_ready(sk);
819 
820 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
821 }
822 
823 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
824 {
825 	pkc->reset_pending_on_curr_blk = 0;
826 }
827 
828 /*
829  * prb_open_block is called by tpacket_rcv or timer callback.
830  *
831  * Reasons why NOT update hrtimer in prb_open_block:
832  * 1) It will increase complexity to distinguish the two caller scenario.
833  * 2) hrtimer_cancel and hrtimer_start need to be called if you want to update
834  * TMO of an already enqueued hrtimer, leading to complex shutdown logic.
835  *
836  * One side effect of NOT update hrtimer when called by tpacket_rcv is that
837  * a newly opened block triggered by tpacket_rcv may be retired earlier than
838  * expected. On the other hand, if timeout is updated in prb_open_block, the
839  * frequent reception of network packets that leads to prb_open_block being
840  * called may cause hrtimer to be removed and enqueued repeatedly.
841  */
842 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
843 	struct tpacket_block_desc *pbd1)
844 {
845 	struct timespec64 ts;
846 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
847 
848 	smp_rmb();
849 
850 	/* We could have just memset this but we will lose the
851 	 * flexibility of making the priv area sticky
852 	 */
853 
854 	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
855 	BLOCK_NUM_PKTS(pbd1) = 0;
856 	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
857 
858 	ktime_get_real_ts64(&ts);
859 
860 	h1->ts_first_pkt.ts_sec = ts.tv_sec;
861 	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
862 
863 	pkc1->pkblk_start = (char *)pbd1;
864 	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 
866 	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
867 	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
868 
869 	pbd1->version = pkc1->version;
870 	pkc1->prev = pkc1->nxt_offset;
871 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
872 
873 	prb_thaw_queue(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 			/* Waiting for skb_copy_bits to finish... */
959 			write_lock(&pkc->blk_fill_in_prog_lock);
960 			write_unlock(&pkc->blk_fill_in_prog_lock);
961 		}
962 		prb_close_block(pkc, pbd, po, status);
963 		return;
964 	}
965 }
966 
967 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
968 {
969 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
970 }
971 
972 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
973 {
974 	return pkc->reset_pending_on_curr_blk;
975 }
976 
977 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
978 	__releases(&pkc->blk_fill_in_prog_lock)
979 {
980 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
981 
982 	read_unlock(&pkc->blk_fill_in_prog_lock);
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 	struct packet_sock *po = container_of(pkc, struct packet_sock, rx_ring.prb_bdqc);
1001 
1002 	if (skb_vlan_tag_present(pkc->skb)) {
1003 		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1004 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1005 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1006 	} else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) {
1007 		ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev);
1008 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol);
1009 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1010 	} else {
1011 		ppd->hv1.tp_vlan_tci = 0;
1012 		ppd->hv1.tp_vlan_tpid = 0;
1013 		ppd->tp_status = TP_STATUS_AVAILABLE;
1014 	}
1015 }
1016 
1017 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1018 			struct tpacket3_hdr *ppd)
1019 {
1020 	ppd->hv1.tp_padding = 0;
1021 	prb_fill_vlan_info(pkc, ppd);
1022 
1023 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1024 		prb_fill_rxhash(pkc, ppd);
1025 	else
1026 		prb_clear_rxhash(pkc, ppd);
1027 }
1028 
1029 static void prb_fill_curr_block(char *curr,
1030 				struct tpacket_kbdq_core *pkc,
1031 				struct tpacket_block_desc *pbd,
1032 				unsigned int len)
1033 	__acquires(&pkc->blk_fill_in_prog_lock)
1034 {
1035 	struct tpacket3_hdr *ppd;
1036 
1037 	ppd  = (struct tpacket3_hdr *)curr;
1038 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1039 	pkc->prev = curr;
1040 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1042 	BLOCK_NUM_PKTS(pbd) += 1;
1043 	read_lock(&pkc->blk_fill_in_prog_lock);
1044 	prb_run_all_ft_ops(pkc, ppd);
1045 }
1046 
1047 /* Assumes caller has the sk->rx_queue.lock */
1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1049 					    struct sk_buff *skb,
1050 					    unsigned int len
1051 					    )
1052 {
1053 	struct tpacket_kbdq_core *pkc;
1054 	struct tpacket_block_desc *pbd;
1055 	char *curr, *end;
1056 
1057 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1058 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1059 
1060 	/* Queue is frozen when user space is lagging behind */
1061 	if (prb_queue_frozen(pkc)) {
1062 		/*
1063 		 * Check if that last block which caused the queue to freeze,
1064 		 * is still in_use by user-space.
1065 		 */
1066 		if (prb_curr_blk_in_use(pbd)) {
1067 			/* Can't record this packet */
1068 			return NULL;
1069 		} else {
1070 			/*
1071 			 * Ok, the block was released by user-space.
1072 			 * Now let's open that block.
1073 			 * opening a block also thaws the queue.
1074 			 * Thawing is a side effect.
1075 			 */
1076 			prb_open_block(pkc, pbd);
1077 		}
1078 	}
1079 
1080 	smp_mb();
1081 	curr = pkc->nxt_offset;
1082 	pkc->skb = skb;
1083 	end = (char *)pbd + pkc->kblk_size;
1084 
1085 	/* first try the current block */
1086 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1087 		prb_fill_curr_block(curr, pkc, pbd, len);
1088 		return (void *)curr;
1089 	}
1090 
1091 	/* Ok, close the current block */
1092 	prb_retire_current_block(pkc, po, 0);
1093 
1094 	/* Now, try to dispatch the next block */
1095 	curr = (char *)prb_dispatch_next_block(pkc, po);
1096 	if (curr) {
1097 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1098 		prb_fill_curr_block(curr, pkc, pbd, len);
1099 		return (void *)curr;
1100 	}
1101 
1102 	/*
1103 	 * No free blocks are available.user_space hasn't caught up yet.
1104 	 * Queue was just frozen and now this packet will get dropped.
1105 	 */
1106 	return NULL;
1107 }
1108 
1109 static void *packet_current_rx_frame(struct packet_sock *po,
1110 					    struct sk_buff *skb,
1111 					    int status, unsigned int len)
1112 {
1113 	char *curr = NULL;
1114 	switch (po->tp_version) {
1115 	case TPACKET_V1:
1116 	case TPACKET_V2:
1117 		curr = packet_lookup_frame(po, &po->rx_ring,
1118 					po->rx_ring.head, status);
1119 		return curr;
1120 	case TPACKET_V3:
1121 		return __packet_lookup_frame_in_block(po, skb, len);
1122 	default:
1123 		WARN(1, "TPACKET version not supported\n");
1124 		BUG();
1125 		return NULL;
1126 	}
1127 }
1128 
1129 static void *prb_lookup_block(const struct packet_sock *po,
1130 			      const struct packet_ring_buffer *rb,
1131 			      unsigned int idx,
1132 			      int status)
1133 {
1134 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1135 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1136 
1137 	if (status != BLOCK_STATUS(pbd))
1138 		return NULL;
1139 	return pbd;
1140 }
1141 
1142 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1143 {
1144 	unsigned int prev;
1145 	if (rb->prb_bdqc.kactive_blk_num)
1146 		prev = rb->prb_bdqc.kactive_blk_num-1;
1147 	else
1148 		prev = rb->prb_bdqc.knum_blocks-1;
1149 	return prev;
1150 }
1151 
1152 /* Assumes caller has held the rx_queue.lock */
1153 static void *__prb_previous_block(struct packet_sock *po,
1154 					 struct packet_ring_buffer *rb,
1155 					 int status)
1156 {
1157 	unsigned int previous = prb_previous_blk_num(rb);
1158 	return prb_lookup_block(po, rb, previous, status);
1159 }
1160 
1161 static void *packet_previous_rx_frame(struct packet_sock *po,
1162 					     struct packet_ring_buffer *rb,
1163 					     int status)
1164 {
1165 	if (po->tp_version <= TPACKET_V2)
1166 		return packet_previous_frame(po, rb, status);
1167 
1168 	return __prb_previous_block(po, rb, status);
1169 }
1170 
1171 static void packet_increment_rx_head(struct packet_sock *po,
1172 					    struct packet_ring_buffer *rb)
1173 {
1174 	switch (po->tp_version) {
1175 	case TPACKET_V1:
1176 	case TPACKET_V2:
1177 		return packet_increment_head(rb);
1178 	case TPACKET_V3:
1179 	default:
1180 		WARN(1, "TPACKET version not supported.\n");
1181 		BUG();
1182 		return;
1183 	}
1184 }
1185 
1186 static void *packet_previous_frame(struct packet_sock *po,
1187 		struct packet_ring_buffer *rb,
1188 		int status)
1189 {
1190 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1191 	return packet_lookup_frame(po, rb, previous, status);
1192 }
1193 
1194 static void packet_increment_head(struct packet_ring_buffer *buff)
1195 {
1196 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1197 }
1198 
1199 static void packet_inc_pending(struct packet_ring_buffer *rb)
1200 {
1201 	this_cpu_inc(*rb->pending_refcnt);
1202 }
1203 
1204 static void packet_dec_pending(struct packet_ring_buffer *rb)
1205 {
1206 	this_cpu_dec(*rb->pending_refcnt);
1207 }
1208 
1209 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1210 {
1211 	unsigned int refcnt = 0;
1212 	int cpu;
1213 
1214 	/* We don't use pending refcount in rx_ring. */
1215 	if (rb->pending_refcnt == NULL)
1216 		return 0;
1217 
1218 	for_each_possible_cpu(cpu)
1219 		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1220 
1221 	return refcnt;
1222 }
1223 
1224 static int packet_alloc_pending(struct packet_sock *po)
1225 {
1226 	po->rx_ring.pending_refcnt = NULL;
1227 
1228 	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1229 	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1230 		return -ENOBUFS;
1231 
1232 	return 0;
1233 }
1234 
1235 static void packet_free_pending(struct packet_sock *po)
1236 {
1237 	free_percpu(po->tx_ring.pending_refcnt);
1238 }
1239 
1240 #define ROOM_POW_OFF	2
1241 #define ROOM_NONE	0x0
1242 #define ROOM_LOW	0x1
1243 #define ROOM_NORMAL	0x2
1244 
1245 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1246 {
1247 	int idx, len;
1248 
1249 	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1250 	idx = READ_ONCE(po->rx_ring.head);
1251 	if (pow_off)
1252 		idx += len >> pow_off;
1253 	if (idx >= len)
1254 		idx -= len;
1255 	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1256 }
1257 
1258 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1259 {
1260 	int idx, len;
1261 
1262 	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1263 	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1264 	if (pow_off)
1265 		idx += len >> pow_off;
1266 	if (idx >= len)
1267 		idx -= len;
1268 	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1269 }
1270 
1271 static int __packet_rcv_has_room(const struct packet_sock *po,
1272 				 const struct sk_buff *skb)
1273 {
1274 	const struct sock *sk = &po->sk;
1275 	int ret = ROOM_NONE;
1276 
1277 	if (po->prot_hook.func != tpacket_rcv) {
1278 		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1279 		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1280 				   - (skb ? skb->truesize : 0);
1281 
1282 		if (avail > (rcvbuf >> ROOM_POW_OFF))
1283 			return ROOM_NORMAL;
1284 		else if (avail > 0)
1285 			return ROOM_LOW;
1286 		else
1287 			return ROOM_NONE;
1288 	}
1289 
1290 	if (po->tp_version == TPACKET_V3) {
1291 		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1292 			ret = ROOM_NORMAL;
1293 		else if (__tpacket_v3_has_room(po, 0))
1294 			ret = ROOM_LOW;
1295 	} else {
1296 		if (__tpacket_has_room(po, ROOM_POW_OFF))
1297 			ret = ROOM_NORMAL;
1298 		else if (__tpacket_has_room(po, 0))
1299 			ret = ROOM_LOW;
1300 	}
1301 
1302 	return ret;
1303 }
1304 
1305 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1306 {
1307 	bool pressure;
1308 	int ret;
1309 
1310 	ret = __packet_rcv_has_room(po, skb);
1311 	pressure = ret != ROOM_NORMAL;
1312 
1313 	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1314 		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1315 
1316 	return ret;
1317 }
1318 
1319 static void __packet_rcv_try_clear_pressure(struct packet_sock *po)
1320 {
1321 	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1322 	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1323 		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1324 }
1325 
1326 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1327 {
1328 	struct sock *sk = &po->sk;
1329 
1330 	if (!packet_sock_flag(po, PACKET_SOCK_PRESSURE))
1331 		return;
1332 
1333 	spin_lock_bh(&sk->sk_receive_queue.lock);
1334 	__packet_rcv_try_clear_pressure(po);
1335 	spin_unlock_bh(&sk->sk_receive_queue.lock);
1336 }
1337 
1338 static void packet_sock_destruct(struct sock *sk)
1339 {
1340 	skb_queue_purge(&sk->sk_error_queue);
1341 
1342 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1343 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1344 
1345 	packet_free_pending(pkt_sk(sk));
1346 
1347 	if (!sock_flag(sk, SOCK_DEAD)) {
1348 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1349 		return;
1350 	}
1351 }
1352 
1353 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1354 {
1355 	u32 *history = po->rollover->history;
1356 	u32 victim, rxhash;
1357 	int i, count = 0;
1358 
1359 	rxhash = skb_get_hash(skb);
1360 	for (i = 0; i < ROLLOVER_HLEN; i++)
1361 		if (READ_ONCE(history[i]) == rxhash)
1362 			count++;
1363 
1364 	victim = get_random_u32_below(ROLLOVER_HLEN);
1365 
1366 	/* Avoid dirtying the cache line if possible */
1367 	if (READ_ONCE(history[victim]) != rxhash)
1368 		WRITE_ONCE(history[victim], rxhash);
1369 
1370 	return count > (ROLLOVER_HLEN >> 1);
1371 }
1372 
1373 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1374 				      struct sk_buff *skb,
1375 				      unsigned int num)
1376 {
1377 	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1378 }
1379 
1380 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1381 				    struct sk_buff *skb,
1382 				    unsigned int num)
1383 {
1384 	unsigned int val = atomic_inc_return(&f->rr_cur);
1385 
1386 	return val % num;
1387 }
1388 
1389 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1390 				     struct sk_buff *skb,
1391 				     unsigned int num)
1392 {
1393 	return smp_processor_id() % num;
1394 }
1395 
1396 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1397 				     struct sk_buff *skb,
1398 				     unsigned int num)
1399 {
1400 	return get_random_u32_below(num);
1401 }
1402 
1403 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1404 					  struct sk_buff *skb,
1405 					  unsigned int idx, bool try_self,
1406 					  unsigned int num)
1407 {
1408 	struct packet_sock *po, *po_next, *po_skip = NULL;
1409 	unsigned int i, j, room = ROOM_NONE;
1410 
1411 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1412 
1413 	if (try_self) {
1414 		room = packet_rcv_has_room(po, skb);
1415 		if (room == ROOM_NORMAL ||
1416 		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1417 			return idx;
1418 		po_skip = po;
1419 	}
1420 
1421 	i = j = min_t(int, po->rollover->sock, num - 1);
1422 	do {
1423 		po_next = pkt_sk(rcu_dereference(f->arr[i]));
1424 		if (po_next != po_skip &&
1425 		    !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1426 		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1427 			if (i != j)
1428 				po->rollover->sock = i;
1429 			atomic_long_inc(&po->rollover->num);
1430 			if (room == ROOM_LOW)
1431 				atomic_long_inc(&po->rollover->num_huge);
1432 			return i;
1433 		}
1434 
1435 		if (++i == num)
1436 			i = 0;
1437 	} while (i != j);
1438 
1439 	atomic_long_inc(&po->rollover->num_failed);
1440 	return idx;
1441 }
1442 
1443 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1444 				    struct sk_buff *skb,
1445 				    unsigned int num)
1446 {
1447 	return skb_get_queue_mapping(skb) % num;
1448 }
1449 
1450 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1451 				     struct sk_buff *skb,
1452 				     unsigned int num)
1453 {
1454 	struct bpf_prog *prog;
1455 	unsigned int ret = 0;
1456 
1457 	rcu_read_lock();
1458 	prog = rcu_dereference(f->bpf_prog);
1459 	if (prog)
1460 		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1461 	rcu_read_unlock();
1462 
1463 	return ret;
1464 }
1465 
1466 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1467 {
1468 	return f->flags & (flag >> 8);
1469 }
1470 
1471 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1472 			     struct packet_type *pt, struct net_device *orig_dev)
1473 {
1474 	struct packet_fanout *f = pt->af_packet_priv;
1475 	unsigned int num = READ_ONCE(f->num_members);
1476 	struct net *net = read_pnet(&f->net);
1477 	struct packet_sock *po;
1478 	unsigned int idx;
1479 
1480 	if (!net_eq(dev_net(dev), net) || !num) {
1481 		kfree_skb(skb);
1482 		return 0;
1483 	}
1484 
1485 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1486 		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1487 		if (!skb)
1488 			return 0;
1489 	}
1490 	switch (f->type) {
1491 	case PACKET_FANOUT_HASH:
1492 	default:
1493 		idx = fanout_demux_hash(f, skb, num);
1494 		break;
1495 	case PACKET_FANOUT_LB:
1496 		idx = fanout_demux_lb(f, skb, num);
1497 		break;
1498 	case PACKET_FANOUT_CPU:
1499 		idx = fanout_demux_cpu(f, skb, num);
1500 		break;
1501 	case PACKET_FANOUT_RND:
1502 		idx = fanout_demux_rnd(f, skb, num);
1503 		break;
1504 	case PACKET_FANOUT_QM:
1505 		idx = fanout_demux_qm(f, skb, num);
1506 		break;
1507 	case PACKET_FANOUT_ROLLOVER:
1508 		idx = fanout_demux_rollover(f, skb, 0, false, num);
1509 		break;
1510 	case PACKET_FANOUT_CBPF:
1511 	case PACKET_FANOUT_EBPF:
1512 		idx = fanout_demux_bpf(f, skb, num);
1513 		break;
1514 	}
1515 
1516 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1517 		idx = fanout_demux_rollover(f, skb, idx, true, num);
1518 
1519 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1520 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1521 }
1522 
1523 DEFINE_MUTEX(fanout_mutex);
1524 EXPORT_SYMBOL_GPL(fanout_mutex);
1525 static LIST_HEAD(fanout_list);
1526 static u16 fanout_next_id;
1527 
1528 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1529 {
1530 	struct packet_fanout *f = po->fanout;
1531 
1532 	spin_lock(&f->lock);
1533 	rcu_assign_pointer(f->arr[f->num_members], sk);
1534 	smp_wmb();
1535 	f->num_members++;
1536 	if (f->num_members == 1)
1537 		dev_add_pack(&f->prot_hook);
1538 	spin_unlock(&f->lock);
1539 }
1540 
1541 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1542 {
1543 	struct packet_fanout *f = po->fanout;
1544 	int i;
1545 
1546 	spin_lock(&f->lock);
1547 	for (i = 0; i < f->num_members; i++) {
1548 		if (rcu_dereference_protected(f->arr[i],
1549 					      lockdep_is_held(&f->lock)) == sk)
1550 			break;
1551 	}
1552 	BUG_ON(i >= f->num_members);
1553 	rcu_assign_pointer(f->arr[i],
1554 			   rcu_dereference_protected(f->arr[f->num_members - 1],
1555 						     lockdep_is_held(&f->lock)));
1556 	f->num_members--;
1557 	if (f->num_members == 0)
1558 		__dev_remove_pack(&f->prot_hook);
1559 	spin_unlock(&f->lock);
1560 }
1561 
1562 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1563 {
1564 	if (sk->sk_family != PF_PACKET)
1565 		return false;
1566 
1567 	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1568 }
1569 
1570 static void fanout_init_data(struct packet_fanout *f)
1571 {
1572 	switch (f->type) {
1573 	case PACKET_FANOUT_LB:
1574 		atomic_set(&f->rr_cur, 0);
1575 		break;
1576 	case PACKET_FANOUT_CBPF:
1577 	case PACKET_FANOUT_EBPF:
1578 		RCU_INIT_POINTER(f->bpf_prog, NULL);
1579 		break;
1580 	}
1581 }
1582 
1583 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1584 {
1585 	struct bpf_prog *old;
1586 
1587 	spin_lock(&f->lock);
1588 	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1589 	rcu_assign_pointer(f->bpf_prog, new);
1590 	spin_unlock(&f->lock);
1591 
1592 	if (old) {
1593 		synchronize_net();
1594 		bpf_prog_destroy(old);
1595 	}
1596 }
1597 
1598 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1599 				unsigned int len)
1600 {
1601 	struct bpf_prog *new;
1602 	struct sock_fprog fprog;
1603 	int ret;
1604 
1605 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1606 		return -EPERM;
1607 
1608 	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1609 	if (ret)
1610 		return ret;
1611 
1612 	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1613 	if (ret)
1614 		return ret;
1615 
1616 	__fanout_set_data_bpf(po->fanout, new);
1617 	return 0;
1618 }
1619 
1620 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1621 				unsigned int len)
1622 {
1623 	struct bpf_prog *new;
1624 	u32 fd;
1625 
1626 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1627 		return -EPERM;
1628 	if (len != sizeof(fd))
1629 		return -EINVAL;
1630 	if (copy_from_sockptr(&fd, data, len))
1631 		return -EFAULT;
1632 
1633 	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1634 	if (IS_ERR(new))
1635 		return PTR_ERR(new);
1636 
1637 	__fanout_set_data_bpf(po->fanout, new);
1638 	return 0;
1639 }
1640 
1641 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1642 			   unsigned int len)
1643 {
1644 	switch (po->fanout->type) {
1645 	case PACKET_FANOUT_CBPF:
1646 		return fanout_set_data_cbpf(po, data, len);
1647 	case PACKET_FANOUT_EBPF:
1648 		return fanout_set_data_ebpf(po, data, len);
1649 	default:
1650 		return -EINVAL;
1651 	}
1652 }
1653 
1654 static void fanout_release_data(struct packet_fanout *f)
1655 {
1656 	switch (f->type) {
1657 	case PACKET_FANOUT_CBPF:
1658 	case PACKET_FANOUT_EBPF:
1659 		__fanout_set_data_bpf(f, NULL);
1660 	}
1661 }
1662 
1663 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1664 {
1665 	struct packet_fanout *f;
1666 
1667 	list_for_each_entry(f, &fanout_list, list) {
1668 		if (f->id == candidate_id &&
1669 		    read_pnet(&f->net) == sock_net(sk)) {
1670 			return false;
1671 		}
1672 	}
1673 	return true;
1674 }
1675 
1676 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1677 {
1678 	u16 id = fanout_next_id;
1679 
1680 	do {
1681 		if (__fanout_id_is_free(sk, id)) {
1682 			*new_id = id;
1683 			fanout_next_id = id + 1;
1684 			return true;
1685 		}
1686 
1687 		id++;
1688 	} while (id != fanout_next_id);
1689 
1690 	return false;
1691 }
1692 
1693 static int fanout_add(struct sock *sk, struct fanout_args *args)
1694 {
1695 	struct packet_rollover *rollover = NULL;
1696 	struct packet_sock *po = pkt_sk(sk);
1697 	u16 type_flags = args->type_flags;
1698 	struct packet_fanout *f, *match;
1699 	u8 type = type_flags & 0xff;
1700 	u8 flags = type_flags >> 8;
1701 	u16 id = args->id;
1702 	int err;
1703 
1704 	switch (type) {
1705 	case PACKET_FANOUT_ROLLOVER:
1706 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1707 			return -EINVAL;
1708 		break;
1709 	case PACKET_FANOUT_HASH:
1710 	case PACKET_FANOUT_LB:
1711 	case PACKET_FANOUT_CPU:
1712 	case PACKET_FANOUT_RND:
1713 	case PACKET_FANOUT_QM:
1714 	case PACKET_FANOUT_CBPF:
1715 	case PACKET_FANOUT_EBPF:
1716 		break;
1717 	default:
1718 		return -EINVAL;
1719 	}
1720 
1721 	mutex_lock(&fanout_mutex);
1722 
1723 	err = -EALREADY;
1724 	if (po->fanout)
1725 		goto out;
1726 
1727 	if (type == PACKET_FANOUT_ROLLOVER ||
1728 	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1729 		err = -ENOMEM;
1730 		rollover = kzalloc_obj(*rollover);
1731 		if (!rollover)
1732 			goto out;
1733 		atomic_long_set(&rollover->num, 0);
1734 		atomic_long_set(&rollover->num_huge, 0);
1735 		atomic_long_set(&rollover->num_failed, 0);
1736 	}
1737 
1738 	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1739 		if (id != 0) {
1740 			err = -EINVAL;
1741 			goto out;
1742 		}
1743 		if (!fanout_find_new_id(sk, &id)) {
1744 			err = -ENOMEM;
1745 			goto out;
1746 		}
1747 		/* ephemeral flag for the first socket in the group: drop it */
1748 		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1749 	}
1750 
1751 	match = NULL;
1752 	list_for_each_entry(f, &fanout_list, list) {
1753 		if (f->id == id &&
1754 		    read_pnet(&f->net) == sock_net(sk)) {
1755 			match = f;
1756 			break;
1757 		}
1758 	}
1759 	err = -EINVAL;
1760 	if (match) {
1761 		if (match->flags != flags)
1762 			goto out;
1763 		if (args->max_num_members &&
1764 		    args->max_num_members != match->max_num_members)
1765 			goto out;
1766 	} else {
1767 		if (args->max_num_members > PACKET_FANOUT_MAX)
1768 			goto out;
1769 		if (!args->max_num_members)
1770 			/* legacy PACKET_FANOUT_MAX */
1771 			args->max_num_members = 256;
1772 		err = -ENOMEM;
1773 		match = kvzalloc_flex(*match, arr, args->max_num_members);
1774 		if (!match)
1775 			goto out;
1776 		write_pnet(&match->net, sock_net(sk));
1777 		match->id = id;
1778 		match->type = type;
1779 		match->flags = flags;
1780 		INIT_LIST_HEAD(&match->list);
1781 		spin_lock_init(&match->lock);
1782 		refcount_set(&match->sk_ref, 0);
1783 		fanout_init_data(match);
1784 		match->prot_hook.type = po->prot_hook.type;
1785 		match->prot_hook.dev = po->prot_hook.dev;
1786 		match->prot_hook.func = packet_rcv_fanout;
1787 		match->prot_hook.af_packet_priv = match;
1788 		match->prot_hook.af_packet_net = read_pnet(&match->net);
1789 		match->prot_hook.id_match = match_fanout_group;
1790 		match->max_num_members = args->max_num_members;
1791 		match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1792 		list_add(&match->list, &fanout_list);
1793 	}
1794 	err = -EINVAL;
1795 
1796 	spin_lock(&po->bind_lock);
1797 	if (po->num &&
1798 	    match->type == type &&
1799 	    match->prot_hook.type == po->prot_hook.type &&
1800 	    match->prot_hook.dev == po->prot_hook.dev) {
1801 		err = -ENOSPC;
1802 		if (refcount_read(&match->sk_ref) < match->max_num_members) {
1803 			/* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1804 			WRITE_ONCE(po->fanout, match);
1805 
1806 			po->rollover = rollover;
1807 			rollover = NULL;
1808 			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1809 			if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
1810 				__dev_remove_pack(&po->prot_hook);
1811 				__fanout_link(sk, po);
1812 			}
1813 			err = 0;
1814 		}
1815 	}
1816 	spin_unlock(&po->bind_lock);
1817 
1818 	if (err && !refcount_read(&match->sk_ref)) {
1819 		list_del(&match->list);
1820 		kvfree(match);
1821 	}
1822 
1823 out:
1824 	kfree(rollover);
1825 	mutex_unlock(&fanout_mutex);
1826 	return err;
1827 }
1828 
1829 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1830  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1831  * It is the responsibility of the caller to call fanout_release_data() and
1832  * free the returned packet_fanout (after synchronize_net())
1833  */
1834 static struct packet_fanout *fanout_release(struct sock *sk)
1835 {
1836 	struct packet_sock *po = pkt_sk(sk);
1837 	struct packet_fanout *f;
1838 
1839 	mutex_lock(&fanout_mutex);
1840 	f = po->fanout;
1841 	if (f) {
1842 		po->fanout = NULL;
1843 
1844 		if (refcount_dec_and_test(&f->sk_ref))
1845 			list_del(&f->list);
1846 		else
1847 			f = NULL;
1848 	}
1849 	mutex_unlock(&fanout_mutex);
1850 
1851 	return f;
1852 }
1853 
1854 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1855 					  struct sk_buff *skb)
1856 {
1857 	/* Earlier code assumed this would be a VLAN pkt, double-check
1858 	 * this now that we have the actual packet in hand. We can only
1859 	 * do this check on Ethernet devices.
1860 	 */
1861 	if (unlikely(dev->type != ARPHRD_ETHER))
1862 		return false;
1863 
1864 	skb_reset_mac_header(skb);
1865 	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1866 }
1867 
1868 static const struct proto_ops packet_ops;
1869 
1870 static const struct proto_ops packet_ops_spkt;
1871 
1872 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1873 			   struct packet_type *pt, struct net_device *orig_dev)
1874 {
1875 	struct sock *sk;
1876 	struct sockaddr_pkt *spkt;
1877 
1878 	/*
1879 	 *	When we registered the protocol we saved the socket in the data
1880 	 *	field for just this event.
1881 	 */
1882 
1883 	sk = pt->af_packet_priv;
1884 
1885 	/*
1886 	 *	Yank back the headers [hope the device set this
1887 	 *	right or kerboom...]
1888 	 *
1889 	 *	Incoming packets have ll header pulled,
1890 	 *	push it back.
1891 	 *
1892 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1893 	 *	so that this procedure is noop.
1894 	 */
1895 
1896 	if (skb->pkt_type == PACKET_LOOPBACK)
1897 		goto out;
1898 
1899 	if (!net_eq(dev_net(dev), sock_net(sk)))
1900 		goto out;
1901 
1902 	skb = skb_share_check(skb, GFP_ATOMIC);
1903 	if (skb == NULL)
1904 		goto oom;
1905 
1906 	/* drop any routing info */
1907 	skb_dst_drop(skb);
1908 
1909 	/* drop conntrack reference */
1910 	nf_reset_ct(skb);
1911 
1912 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1913 
1914 	skb_push(skb, skb->data - skb_mac_header(skb));
1915 
1916 	/*
1917 	 *	The SOCK_PACKET socket receives _all_ frames.
1918 	 */
1919 
1920 	spkt->spkt_family = dev->type;
1921 	strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1922 	spkt->spkt_protocol = skb->protocol;
1923 
1924 	/*
1925 	 *	Charge the memory to the socket. This is done specifically
1926 	 *	to prevent sockets using all the memory up.
1927 	 */
1928 
1929 	if (sock_queue_rcv_skb(sk, skb) == 0)
1930 		return 0;
1931 
1932 out:
1933 	kfree_skb(skb);
1934 oom:
1935 	return 0;
1936 }
1937 
1938 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1939 {
1940 	int depth;
1941 
1942 	/* On TX skb->data is the L2 header; anchor it for all socket types. */
1943 	skb_reset_mac_header(skb);
1944 
1945 	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1946 	    sock->type == SOCK_RAW)
1947 		skb->protocol = dev_parse_header_protocol(skb);
1948 
1949 	skb_probe_transport_header(skb);
1950 
1951 	/* Move network header to the right position for VLAN tagged packets */
1952 	if (likely(skb->dev->type == ARPHRD_ETHER) &&
1953 	    eth_type_vlan(skb->protocol) &&
1954 	    vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1955 		skb_set_network_header(skb, depth);
1956 }
1957 
1958 /*
1959  *	Output a raw packet to a device layer. This bypasses all the other
1960  *	protocol layers and you must therefore supply it with a complete frame
1961  */
1962 
1963 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1964 			       size_t len)
1965 {
1966 	struct sock *sk = sock->sk;
1967 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1968 	struct sk_buff *skb = NULL;
1969 	struct net_device *dev;
1970 	struct sockcm_cookie sockc;
1971 	__be16 proto = 0;
1972 	int hard_header_len;
1973 	int extra_len = 0;
1974 	int err;
1975 
1976 	/*
1977 	 *	Get and verify the address.
1978 	 */
1979 
1980 	if (saddr) {
1981 		if (msg->msg_namelen < sizeof(struct sockaddr))
1982 			return -EINVAL;
1983 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1984 			proto = saddr->spkt_protocol;
1985 	} else
1986 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1987 
1988 	/*
1989 	 *	Find the device first to size check it
1990 	 */
1991 
1992 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1993 retry:
1994 	rcu_read_lock();
1995 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1996 	err = -ENODEV;
1997 	if (dev == NULL)
1998 		goto out_unlock;
1999 
2000 	err = -ENETDOWN;
2001 	if (!(dev->flags & IFF_UP))
2002 		goto out_unlock;
2003 
2004 	/*
2005 	 * You may not queue a frame bigger than the mtu. This is the lowest level
2006 	 * raw protocol and you must do your own fragmentation at this level.
2007 	 */
2008 
2009 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2010 		if (!netif_supports_nofcs(dev)) {
2011 			err = -EPROTONOSUPPORT;
2012 			goto out_unlock;
2013 		}
2014 		extra_len = 4; /* We're doing our own CRC */
2015 	}
2016 
2017 	/* Keep the allocation-time header length across retry. */
2018 	if (!skb)
2019 		hard_header_len = READ_ONCE(dev->hard_header_len);
2020 
2021 	err = -EMSGSIZE;
2022 	if (len > dev->mtu + hard_header_len + VLAN_HLEN + extra_len)
2023 		goto out_unlock;
2024 
2025 	if (!skb) {
2026 		size_t reserved = LL_RESERVED_SPACE_EX(dev, hard_header_len);
2027 		int tlen = dev->needed_tailroom;
2028 		unsigned int hhlen = dev->header_ops ? hard_header_len : 0;
2029 
2030 		rcu_read_unlock();
2031 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2032 		if (skb == NULL)
2033 			return -ENOBUFS;
2034 		/* FIXME: Save some space for broken drivers that write a hard
2035 		 * header at transmission time by themselves. PPP is the notable
2036 		 * one here. This should really be fixed at the driver level.
2037 		 */
2038 		skb_reserve(skb, reserved);
2039 		skb_reset_network_header(skb);
2040 
2041 		/* Try to align data part correctly */
2042 		if (hhlen) {
2043 			skb->data -= hhlen;
2044 			skb->tail -= hhlen;
2045 			if (len < hhlen)
2046 				skb_reset_network_header(skb);
2047 		}
2048 		err = memcpy_from_msg(skb_put(skb, len), msg, len);
2049 		if (err)
2050 			goto out_free;
2051 		goto retry;
2052 	}
2053 
2054 	if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2055 		err = -EINVAL;
2056 		goto out_unlock;
2057 	}
2058 	if (len > (dev->mtu + hard_header_len + extra_len) &&
2059 	    !packet_extra_vlan_len_allowed(dev, skb)) {
2060 		err = -EMSGSIZE;
2061 		goto out_unlock;
2062 	}
2063 
2064 	sockcm_init(&sockc, sk);
2065 	if (msg->msg_controllen) {
2066 		err = sock_cmsg_send(sk, msg, &sockc);
2067 		if (unlikely(err))
2068 			goto out_unlock;
2069 	}
2070 
2071 	skb->protocol = proto;
2072 	skb->dev = dev;
2073 	skb->priority = sockc.priority;
2074 	skb->mark = sockc.mark;
2075 	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
2076 	skb_setup_tx_timestamp(skb, &sockc);
2077 
2078 	if (unlikely(extra_len == 4))
2079 		skb->no_fcs = 1;
2080 
2081 	packet_parse_headers(skb, sock);
2082 
2083 	dev_queue_xmit(skb);
2084 	rcu_read_unlock();
2085 	return len;
2086 
2087 out_unlock:
2088 	rcu_read_unlock();
2089 out_free:
2090 	kfree_skb(skb);
2091 	return err;
2092 }
2093 
2094 static unsigned int run_filter(struct sk_buff *skb,
2095 			       const struct sock *sk,
2096 			       unsigned int res)
2097 {
2098 	struct sk_filter *filter;
2099 
2100 	rcu_read_lock();
2101 	filter = rcu_dereference(sk->sk_filter);
2102 	if (filter != NULL)
2103 		res = bpf_prog_run_clear_cb(filter->prog, skb);
2104 	rcu_read_unlock();
2105 
2106 	return res;
2107 }
2108 
2109 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2110 			   size_t *len, int vnet_hdr_sz)
2111 {
2112 	struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2113 
2114 	if (*len < vnet_hdr_sz)
2115 		return -EINVAL;
2116 	*len -= vnet_hdr_sz;
2117 
2118 	if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2119 		return -EINVAL;
2120 
2121 	return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2122 }
2123 
2124 /*
2125  * This function makes lazy skb cloning in hope that most of packets
2126  * are discarded by BPF.
2127  *
2128  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2129  * and skb->cb are mangled. It works because (and until) packets
2130  * falling here are owned by current CPU. Output packets are cloned
2131  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2132  * sequentially, so that if we return skb to original state on exit,
2133  * we will not harm anyone.
2134  */
2135 
2136 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2137 		      struct packet_type *pt, struct net_device *orig_dev)
2138 {
2139 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2140 	struct sock *sk = NULL;
2141 	struct sockaddr_ll *sll;
2142 	struct packet_sock *po;
2143 	u8 *skb_head = skb->data;
2144 	int skb_len = skb->len;
2145 	unsigned int snaplen, res;
2146 
2147 	if (skb->pkt_type == PACKET_LOOPBACK)
2148 		goto drop;
2149 
2150 	sk = pt->af_packet_priv;
2151 	po = pkt_sk(sk);
2152 
2153 	if (!net_eq(dev_net(dev), sock_net(sk)))
2154 		goto drop;
2155 
2156 	skb->dev = dev;
2157 
2158 	if (dev_has_header(dev)) {
2159 		/* The device has an explicit notion of ll header,
2160 		 * exported to higher levels.
2161 		 *
2162 		 * Otherwise, the device hides details of its frame
2163 		 * structure, so that corresponding packet head is
2164 		 * never delivered to user.
2165 		 */
2166 		if (sk->sk_type != SOCK_DGRAM)
2167 			skb_push(skb, skb->data - skb_mac_header(skb));
2168 		else if (skb->pkt_type == PACKET_OUTGOING) {
2169 			/* Special case: outgoing packets have ll header at head */
2170 			skb_pull(skb, skb_network_offset(skb));
2171 		}
2172 	}
2173 
2174 	snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2175 
2176 	res = run_filter(skb, sk, snaplen);
2177 	if (!res)
2178 		goto drop_n_restore;
2179 	if (snaplen > res)
2180 		snaplen = res;
2181 
2182 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2183 		goto drop_n_acct;
2184 
2185 	if (skb_shared(skb)) {
2186 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2187 		if (nskb == NULL)
2188 			goto drop_n_acct;
2189 
2190 		if (skb_head != skb->data) {
2191 			skb->data = skb_head;
2192 			skb->len = skb_len;
2193 		}
2194 		consume_skb(skb);
2195 		skb = nskb;
2196 	}
2197 
2198 	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2199 
2200 	sll = &PACKET_SKB_CB(skb)->sa.ll;
2201 	sll->sll_hatype = dev->type;
2202 	sll->sll_pkttype = skb->pkt_type;
2203 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2204 		sll->sll_ifindex = orig_dev->ifindex;
2205 	else
2206 		sll->sll_ifindex = dev->ifindex;
2207 
2208 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2209 
2210 	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2211 	 * Use their space for storing the original skb length.
2212 	 */
2213 	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2214 
2215 	if (pskb_trim(skb, snaplen))
2216 		goto drop_n_acct;
2217 
2218 	skb_set_owner_r(skb, sk);
2219 	skb->dev = NULL;
2220 	skb_dst_drop(skb);
2221 
2222 	/* drop conntrack reference */
2223 	nf_reset_ct(skb);
2224 
2225 	spin_lock(&sk->sk_receive_queue.lock);
2226 	po->stats.stats1.tp_packets++;
2227 	sock_skb_set_dropcount(sk, skb);
2228 	skb_clear_delivery_time(skb);
2229 	__skb_queue_tail(&sk->sk_receive_queue, skb);
2230 	spin_unlock(&sk->sk_receive_queue.lock);
2231 	sk->sk_data_ready(sk);
2232 	return 0;
2233 
2234 drop_n_acct:
2235 	atomic_inc(&po->tp_drops);
2236 	sk_drops_inc(sk);
2237 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2238 
2239 drop_n_restore:
2240 	if (skb_head != skb->data && skb_shared(skb)) {
2241 		skb->data = skb_head;
2242 		skb->len = skb_len;
2243 	}
2244 drop:
2245 	sk_skb_reason_drop(sk, skb, drop_reason);
2246 	return 0;
2247 }
2248 
2249 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2250 		       struct packet_type *pt, struct net_device *orig_dev)
2251 {
2252 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2253 	struct sock *sk = NULL;
2254 	struct packet_sock *po;
2255 	struct sockaddr_ll *sll;
2256 	union tpacket_uhdr h;
2257 	u8 *skb_head = skb->data;
2258 	int skb_len = skb->len;
2259 	unsigned int snaplen, res;
2260 	unsigned long status = TP_STATUS_USER;
2261 	unsigned short macoff, hdrlen;
2262 	unsigned int netoff;
2263 	struct sk_buff *copy_skb = NULL;
2264 	struct timespec64 ts;
2265 	__u32 ts_status;
2266 	unsigned int slot_id = 0;
2267 	int vnet_hdr_sz = 0;
2268 
2269 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2270 	 * We may add members to them until current aligned size without forcing
2271 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2272 	 */
2273 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2274 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2275 
2276 	if (skb->pkt_type == PACKET_LOOPBACK)
2277 		goto drop;
2278 
2279 	sk = pt->af_packet_priv;
2280 	po = pkt_sk(sk);
2281 
2282 	if (!net_eq(dev_net(dev), sock_net(sk)))
2283 		goto drop;
2284 
2285 	if (dev_has_header(dev)) {
2286 		if (sk->sk_type != SOCK_DGRAM)
2287 			skb_push(skb, skb->data - skb_mac_header(skb));
2288 		else if (skb->pkt_type == PACKET_OUTGOING) {
2289 			/* Special case: outgoing packets have ll header at head */
2290 			skb_pull(skb, skb_network_offset(skb));
2291 		}
2292 	}
2293 
2294 	snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2295 
2296 	res = run_filter(skb, sk, snaplen);
2297 	if (!res)
2298 		goto drop_n_restore;
2299 
2300 	/* If we are flooded, just give up */
2301 	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2302 		atomic_inc(&po->tp_drops);
2303 		goto drop_n_restore;
2304 	}
2305 
2306 	if (skb->ip_summed == CHECKSUM_PARTIAL)
2307 		status |= TP_STATUS_CSUMNOTREADY;
2308 	else if (skb->pkt_type != PACKET_OUTGOING &&
2309 		 skb_csum_unnecessary(skb))
2310 		status |= TP_STATUS_CSUM_VALID;
2311 	if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2312 		status |= TP_STATUS_GSO_TCP;
2313 
2314 	if (snaplen > res)
2315 		snaplen = res;
2316 
2317 	if (sk->sk_type == SOCK_DGRAM) {
2318 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2319 				  po->tp_reserve;
2320 	} else {
2321 		unsigned int maclen = skb_network_offset(skb);
2322 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2323 				       (maclen < 16 ? 16 : maclen)) +
2324 				       po->tp_reserve;
2325 		vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2326 		if (vnet_hdr_sz)
2327 			netoff += vnet_hdr_sz;
2328 		macoff = netoff - maclen;
2329 	}
2330 	if (netoff > USHRT_MAX) {
2331 		atomic_inc(&po->tp_drops);
2332 		goto drop_n_restore;
2333 	}
2334 	if (po->tp_version <= TPACKET_V2) {
2335 		if (macoff + snaplen > po->rx_ring.frame_size) {
2336 			if (READ_ONCE(po->copy_thresh) &&
2337 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2338 				if (skb_shared(skb)) {
2339 					copy_skb = skb_clone(skb, GFP_ATOMIC);
2340 				} else {
2341 					copy_skb = skb_get(skb);
2342 					skb_head = skb->data;
2343 				}
2344 				if (copy_skb) {
2345 					memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2346 					       sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2347 					skb_set_owner_r(copy_skb, sk);
2348 				}
2349 			}
2350 			snaplen = po->rx_ring.frame_size - macoff;
2351 			if ((int)snaplen < 0) {
2352 				snaplen = 0;
2353 				vnet_hdr_sz = 0;
2354 			}
2355 		}
2356 	} else if (unlikely(macoff + snaplen >
2357 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2358 		u32 nval;
2359 
2360 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2361 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2362 			    snaplen, nval, macoff);
2363 		snaplen = nval;
2364 		if (unlikely((int)snaplen < 0)) {
2365 			snaplen = 0;
2366 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2367 			vnet_hdr_sz = 0;
2368 		}
2369 	}
2370 	spin_lock(&sk->sk_receive_queue.lock);
2371 	h.raw = packet_current_rx_frame(po, skb,
2372 					TP_STATUS_KERNEL, (macoff+snaplen));
2373 	if (!h.raw)
2374 		goto drop_n_account;
2375 
2376 	if (po->tp_version <= TPACKET_V2) {
2377 		slot_id = po->rx_ring.head;
2378 		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2379 			goto drop_n_account;
2380 		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2381 	}
2382 
2383 	if (vnet_hdr_sz &&
2384 	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2385 				    sizeof(struct virtio_net_hdr),
2386 				    vio_le(), true, 0)) {
2387 		if (po->tp_version <= TPACKET_V2)
2388 			__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2389 		else
2390 			prb_clear_blk_fill_status(&po->rx_ring);
2391 		goto drop_n_account;
2392 	}
2393 
2394 	if (po->tp_version <= TPACKET_V2) {
2395 		packet_increment_rx_head(po, &po->rx_ring);
2396 	/*
2397 	 * LOSING will be reported till you read the stats,
2398 	 * because it's COR - Clear On Read.
2399 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2400 	 * at packet level.
2401 	 */
2402 		if (atomic_read(&po->tp_drops))
2403 			status |= TP_STATUS_LOSING;
2404 	}
2405 
2406 	po->stats.stats1.tp_packets++;
2407 	if (copy_skb) {
2408 		status |= TP_STATUS_COPY;
2409 		skb_clear_delivery_time(copy_skb);
2410 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2411 	}
2412 	spin_unlock(&sk->sk_receive_queue.lock);
2413 
2414 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2415 
2416 	/* Always timestamp; prefer an existing software timestamp taken
2417 	 * closer to the time of capture.
2418 	 */
2419 	ts_status = tpacket_get_timestamp(skb, &ts,
2420 					  READ_ONCE(po->tp_tstamp) |
2421 					  SOF_TIMESTAMPING_SOFTWARE);
2422 	if (!ts_status)
2423 		ktime_get_real_ts64(&ts);
2424 
2425 	status |= ts_status;
2426 
2427 	switch (po->tp_version) {
2428 	case TPACKET_V1:
2429 		h.h1->tp_len = skb->len;
2430 		h.h1->tp_snaplen = snaplen;
2431 		h.h1->tp_mac = macoff;
2432 		h.h1->tp_net = netoff;
2433 		h.h1->tp_sec = ts.tv_sec;
2434 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2435 		hdrlen = sizeof(*h.h1);
2436 		break;
2437 	case TPACKET_V2:
2438 		h.h2->tp_len = skb->len;
2439 		h.h2->tp_snaplen = snaplen;
2440 		h.h2->tp_mac = macoff;
2441 		h.h2->tp_net = netoff;
2442 		h.h2->tp_sec = ts.tv_sec;
2443 		h.h2->tp_nsec = ts.tv_nsec;
2444 		if (skb_vlan_tag_present(skb)) {
2445 			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2446 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2447 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2448 		} else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
2449 			h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev);
2450 			h.h2->tp_vlan_tpid = ntohs(skb->protocol);
2451 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2452 		} else {
2453 			h.h2->tp_vlan_tci = 0;
2454 			h.h2->tp_vlan_tpid = 0;
2455 		}
2456 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2457 		hdrlen = sizeof(*h.h2);
2458 		break;
2459 	case TPACKET_V3:
2460 		/* tp_nxt_offset,vlan are already populated above.
2461 		 * So DONT clear those fields here
2462 		 */
2463 		h.h3->tp_status |= status;
2464 		h.h3->tp_len = skb->len;
2465 		h.h3->tp_snaplen = snaplen;
2466 		h.h3->tp_mac = macoff;
2467 		h.h3->tp_net = netoff;
2468 		h.h3->tp_sec  = ts.tv_sec;
2469 		h.h3->tp_nsec = ts.tv_nsec;
2470 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2471 		hdrlen = sizeof(*h.h3);
2472 		break;
2473 	default:
2474 		BUG();
2475 	}
2476 
2477 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2478 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2479 	sll->sll_family = AF_PACKET;
2480 	sll->sll_hatype = dev->type;
2481 	sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ?
2482 		vlan_get_protocol_dgram(skb) : skb->protocol;
2483 	sll->sll_pkttype = skb->pkt_type;
2484 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2485 		sll->sll_ifindex = orig_dev->ifindex;
2486 	else
2487 		sll->sll_ifindex = dev->ifindex;
2488 
2489 	smp_mb();
2490 
2491 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2492 	if (po->tp_version <= TPACKET_V2) {
2493 		u8 *start, *end;
2494 
2495 		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2496 					macoff + snaplen);
2497 
2498 		for (start = h.raw; start < end; start += PAGE_SIZE)
2499 			flush_dcache_page(pgv_to_page(start));
2500 	}
2501 	smp_wmb();
2502 #endif
2503 
2504 	if (po->tp_version <= TPACKET_V2) {
2505 		spin_lock(&sk->sk_receive_queue.lock);
2506 		__packet_set_status(po, h.raw, status);
2507 		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2508 		spin_unlock(&sk->sk_receive_queue.lock);
2509 		sk->sk_data_ready(sk);
2510 	} else if (po->tp_version == TPACKET_V3) {
2511 		prb_clear_blk_fill_status(&po->rx_ring);
2512 	}
2513 
2514 drop_n_restore:
2515 	if (skb_head != skb->data && skb_shared(skb)) {
2516 		skb->data = skb_head;
2517 		skb->len = skb_len;
2518 	}
2519 drop:
2520 	sk_skb_reason_drop(sk, skb, drop_reason);
2521 	return 0;
2522 
2523 drop_n_account:
2524 	spin_unlock(&sk->sk_receive_queue.lock);
2525 	atomic_inc(&po->tp_drops);
2526 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2527 
2528 	sk->sk_data_ready(sk);
2529 	sk_skb_reason_drop(sk, copy_skb, drop_reason);
2530 	goto drop_n_restore;
2531 }
2532 
2533 static void tpacket_destruct_skb(struct sk_buff *skb)
2534 {
2535 	struct packet_sock *po = pkt_sk(skb->sk);
2536 
2537 	if (likely(po->tx_ring.pg_vec)) {
2538 		void *ph;
2539 		__u32 ts;
2540 
2541 		ph = skb_zcopy_get_nouarg(skb);
2542 
2543 		ts = __packet_set_timestamp(po, ph, skb);
2544 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2545 
2546 		packet_dec_pending(&po->tx_ring);
2547 		complete(&po->skb_completion);
2548 	}
2549 
2550 	sock_wfree(skb);
2551 }
2552 
2553 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2554 {
2555 	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2556 	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2557 	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2558 	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2559 		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2560 			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2561 			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2562 
2563 	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2564 		return -EINVAL;
2565 
2566 	return 0;
2567 }
2568 
2569 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2570 				 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2571 {
2572 	int ret;
2573 
2574 	if (*len < vnet_hdr_sz)
2575 		return -EINVAL;
2576 	*len -= vnet_hdr_sz;
2577 
2578 	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2579 		return -EFAULT;
2580 
2581 	ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2582 	if (ret)
2583 		return ret;
2584 
2585 	/* move iter to point to the start of mac header */
2586 	if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2587 		iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2588 
2589 	return 0;
2590 }
2591 
2592 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2593 		void *frame, struct net_device *dev, void *data, int tp_len,
2594 		__be16 proto, unsigned char *addr, int hlen, int copylen,
2595 		int hard_header_len,
2596 		const struct sockcm_cookie *sockc)
2597 {
2598 	union tpacket_uhdr ph;
2599 	int to_write, offset, len, nr_frags, len_max;
2600 	struct socket *sock = po->sk.sk_socket;
2601 	struct page *page;
2602 	int err;
2603 
2604 	ph.raw = frame;
2605 
2606 	skb->protocol = proto;
2607 	skb->dev = dev;
2608 	skb->priority = sockc->priority;
2609 	skb->mark = sockc->mark;
2610 	skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid);
2611 	skb_setup_tx_timestamp(skb, sockc);
2612 	skb_zcopy_set_nouarg(skb, ph.raw);
2613 
2614 	skb_reserve(skb, hlen);
2615 	skb_reset_network_header(skb);
2616 
2617 	to_write = tp_len;
2618 
2619 	if (sock->type == SOCK_DGRAM) {
2620 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2621 				NULL, tp_len);
2622 		if (unlikely(err < 0))
2623 			return -EINVAL;
2624 	} else if (copylen) {
2625 		int hdrlen = min_t(int, copylen, tp_len);
2626 
2627 		skb_push(skb, hard_header_len);
2628 		skb_put(skb, copylen - hard_header_len);
2629 		err = skb_store_bits(skb, 0, data, hdrlen);
2630 		if (unlikely(err))
2631 			return err;
2632 		if (!dev_validate_header(dev, skb->data, hdrlen))
2633 			return -EINVAL;
2634 
2635 		data += hdrlen;
2636 		to_write -= hdrlen;
2637 	}
2638 
2639 	offset = offset_in_page(data);
2640 	len_max = PAGE_SIZE - offset;
2641 	len = ((to_write > len_max) ? len_max : to_write);
2642 
2643 	skb->data_len = to_write;
2644 	skb->len += to_write;
2645 	skb->truesize += to_write;
2646 	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2647 
2648 	while (likely(to_write)) {
2649 		nr_frags = skb_shinfo(skb)->nr_frags;
2650 
2651 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2652 			pr_err("Packet exceed the number of skb frags(%u)\n",
2653 			       (unsigned int)MAX_SKB_FRAGS);
2654 			return -EFAULT;
2655 		}
2656 
2657 		page = pgv_to_page(data);
2658 		data += len;
2659 		flush_dcache_page(page);
2660 		get_page(page);
2661 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2662 		to_write -= len;
2663 		offset = 0;
2664 		len_max = PAGE_SIZE;
2665 		len = ((to_write > len_max) ? len_max : to_write);
2666 	}
2667 
2668 	if (unlikely(!skb->len))
2669 		return -EINVAL;
2670 
2671 	packet_parse_headers(skb, sock);
2672 
2673 	return tp_len;
2674 }
2675 
2676 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2677 				int size_max, void **data)
2678 {
2679 	union tpacket_uhdr ph;
2680 	u32 tp_len;
2681 	int off;
2682 
2683 	ph.raw = frame;
2684 
2685 	switch (po->tp_version) {
2686 	case TPACKET_V3:
2687 		if (ph.h3->tp_next_offset != 0) {
2688 			pr_warn_once("variable sized slot not supported");
2689 			return -EINVAL;
2690 		}
2691 		tp_len = ph.h3->tp_len;
2692 		break;
2693 	case TPACKET_V2:
2694 		tp_len = ph.h2->tp_len;
2695 		break;
2696 	default:
2697 		tp_len = ph.h1->tp_len;
2698 		break;
2699 	}
2700 	if (unlikely(tp_len > size_max)) {
2701 		pr_err("packet size is too long (%u > %d)\n", tp_len, size_max);
2702 		return -EMSGSIZE;
2703 	}
2704 
2705 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2706 		int off_min, off_max;
2707 
2708 		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2709 		off_max = po->tx_ring.frame_size - tp_len;
2710 		if (po->sk.sk_type == SOCK_DGRAM) {
2711 			switch (po->tp_version) {
2712 			case TPACKET_V3:
2713 				off = ph.h3->tp_net;
2714 				break;
2715 			case TPACKET_V2:
2716 				off = ph.h2->tp_net;
2717 				break;
2718 			default:
2719 				off = ph.h1->tp_net;
2720 				break;
2721 			}
2722 		} else {
2723 			switch (po->tp_version) {
2724 			case TPACKET_V3:
2725 				off = ph.h3->tp_mac;
2726 				break;
2727 			case TPACKET_V2:
2728 				off = ph.h2->tp_mac;
2729 				break;
2730 			default:
2731 				off = ph.h1->tp_mac;
2732 				break;
2733 			}
2734 		}
2735 		if (unlikely((off < off_min) || (off_max < off)))
2736 			return -EINVAL;
2737 	} else {
2738 		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2739 	}
2740 
2741 	*data = frame + off;
2742 	return tp_len;
2743 }
2744 
2745 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2746 {
2747 	struct sk_buff *skb = NULL;
2748 	struct net_device *dev;
2749 	struct virtio_net_hdr vnet_hdr;
2750 	bool has_vnet_hdr = false;
2751 	struct sockcm_cookie sockc;
2752 	__be16 proto;
2753 	int err, reserve = 0;
2754 	void *ph;
2755 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2756 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2757 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2758 	unsigned char *addr = NULL;
2759 	int tp_len, size_max;
2760 	void *data;
2761 	int len_sum = 0;
2762 	int status = TP_STATUS_AVAILABLE;
2763 	int hard_header_len, hlen, tlen, copylen = 0;
2764 	long timeo;
2765 
2766 	mutex_lock(&po->pg_vec_lock);
2767 
2768 	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2769 	 * we need to confirm it under protection of pg_vec_lock.
2770 	 */
2771 	if (unlikely(!po->tx_ring.pg_vec)) {
2772 		err = -EBUSY;
2773 		goto out;
2774 	}
2775 	if (likely(saddr == NULL)) {
2776 		dev	= packet_cached_dev_get(po);
2777 		proto	= READ_ONCE(po->num);
2778 	} else {
2779 		err = -EINVAL;
2780 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2781 			goto out;
2782 		if (msg->msg_namelen < (saddr->sll_halen
2783 					+ offsetof(struct sockaddr_ll,
2784 						sll_addr)))
2785 			goto out;
2786 		proto	= saddr->sll_protocol;
2787 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2788 		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2789 			if (dev && msg->msg_namelen < dev->addr_len +
2790 				   offsetof(struct sockaddr_ll, sll_addr))
2791 				goto out_put;
2792 			addr = saddr->sll_addr;
2793 		}
2794 	}
2795 
2796 	err = -ENXIO;
2797 	if (unlikely(dev == NULL))
2798 		goto out;
2799 	err = -ENETDOWN;
2800 	if (unlikely(!(dev->flags & IFF_UP)))
2801 		goto out_put;
2802 
2803 	sockcm_init(&sockc, &po->sk);
2804 	if (msg->msg_controllen) {
2805 		err = sock_cmsg_send(&po->sk, msg, &sockc);
2806 		if (unlikely(err))
2807 			goto out_put;
2808 	}
2809 
2810 	hard_header_len = READ_ONCE(dev->hard_header_len);
2811 	if (po->sk.sk_socket->type == SOCK_RAW)
2812 		reserve = hard_header_len;
2813 	size_max = po->tx_ring.frame_size
2814 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2815 
2816 	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2817 		size_max = dev->mtu + reserve + VLAN_HLEN;
2818 
2819 	timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2820 	reinit_completion(&po->skb_completion);
2821 
2822 	do {
2823 		ph = packet_current_frame(po, &po->tx_ring,
2824 					  TP_STATUS_SEND_REQUEST);
2825 		if (unlikely(ph == NULL)) {
2826 			/* Note: packet_read_pending() might be slow if we
2827 			 * have to call it as it's per_cpu variable, but in
2828 			 * fast-path we don't have to call it, only when ph
2829 			 * is NULL, we need to check the pending_refcnt.
2830 			 */
2831 			if (need_wait && packet_read_pending(&po->tx_ring)) {
2832 				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2833 				if (timeo <= 0) {
2834 					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2835 					goto out_put;
2836 				}
2837 				/* check for additional frames */
2838 				continue;
2839 			} else
2840 				break;
2841 		}
2842 
2843 		skb = NULL;
2844 		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2845 		if (tp_len < 0)
2846 			goto tpacket_error;
2847 
2848 		status = TP_STATUS_SEND_REQUEST;
2849 		hlen = LL_RESERVED_SPACE_EX(dev, hard_header_len);
2850 		tlen = dev->needed_tailroom;
2851 		if (vnet_hdr_sz) {
2852 			data += vnet_hdr_sz;
2853 			tp_len -= vnet_hdr_sz;
2854 			if (tp_len < 0) {
2855 				tp_len = -EINVAL;
2856 				goto tpacket_error;
2857 			}
2858 			memcpy(&vnet_hdr, data - vnet_hdr_sz, sizeof(vnet_hdr));
2859 			if (__packet_snd_vnet_parse(&vnet_hdr, tp_len)) {
2860 				tp_len = -EINVAL;
2861 				goto tpacket_error;
2862 			}
2863 			copylen = __virtio16_to_cpu(vio_le(),
2864 						    vnet_hdr.hdr_len);
2865 			has_vnet_hdr = true;
2866 		}
2867 		copylen = max_t(int, copylen, hard_header_len);
2868 		skb = sock_alloc_send_skb(&po->sk,
2869 				hlen + tlen + sizeof(struct sockaddr_ll) +
2870 				(copylen - hard_header_len),
2871 				!need_wait, &err);
2872 
2873 		if (unlikely(skb == NULL)) {
2874 			/* we assume the socket was initially writeable ... */
2875 			if (likely(len_sum > 0))
2876 				err = len_sum;
2877 			goto out_status;
2878 		}
2879 		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2880 					  addr, hlen, copylen, hard_header_len,
2881 					  &sockc);
2882 		if (likely(tp_len >= 0) &&
2883 		    tp_len > dev->mtu + reserve &&
2884 		    !vnet_hdr_sz &&
2885 		    !packet_extra_vlan_len_allowed(dev, skb))
2886 			tp_len = -EMSGSIZE;
2887 
2888 		if (unlikely(tp_len < 0)) {
2889 tpacket_error:
2890 			if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2891 				__packet_set_status(po, ph,
2892 						TP_STATUS_AVAILABLE);
2893 				packet_increment_head(&po->tx_ring);
2894 				kfree_skb(skb);
2895 				continue;
2896 			} else {
2897 				status = TP_STATUS_WRONG_FORMAT;
2898 				err = tp_len;
2899 				goto out_status;
2900 			}
2901 		}
2902 
2903 		if (has_vnet_hdr) {
2904 			if (virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le())) {
2905 				tp_len = -EINVAL;
2906 				goto tpacket_error;
2907 			}
2908 			virtio_net_hdr_set_proto(skb, &vnet_hdr);
2909 		}
2910 
2911 		skb->destructor = tpacket_destruct_skb;
2912 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2913 		packet_inc_pending(&po->tx_ring);
2914 
2915 		status = TP_STATUS_SEND_REQUEST;
2916 		err = packet_xmit(po, skb);
2917 		if (unlikely(err != 0)) {
2918 			if (err > 0)
2919 				err = net_xmit_errno(err);
2920 			if (err && __packet_get_status(po, ph) ==
2921 				   TP_STATUS_AVAILABLE) {
2922 				/* skb was destructed already */
2923 				skb = NULL;
2924 				goto out_status;
2925 			}
2926 			/*
2927 			 * skb was dropped but not destructed yet;
2928 			 * let's treat it like congestion or err < 0
2929 			 */
2930 			err = 0;
2931 		}
2932 		packet_increment_head(&po->tx_ring);
2933 		len_sum += tp_len;
2934 	} while (1);
2935 
2936 	err = len_sum;
2937 	goto out_put;
2938 
2939 out_status:
2940 	__packet_set_status(po, ph, status);
2941 	kfree_skb(skb);
2942 out_put:
2943 	dev_put(dev);
2944 out:
2945 	mutex_unlock(&po->pg_vec_lock);
2946 	return err;
2947 }
2948 
2949 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2950 				        size_t reserve, size_t len,
2951 				        size_t linear, int noblock,
2952 				        int *err)
2953 {
2954 	struct sk_buff *skb;
2955 
2956 	/* Under a page?  Don't bother with paged skb. */
2957 	if (prepad + len < PAGE_SIZE || !linear)
2958 		linear = len;
2959 
2960 	if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2961 		linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2962 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2963 				   err, PAGE_ALLOC_COSTLY_ORDER);
2964 	if (!skb)
2965 		return NULL;
2966 
2967 	skb_reserve(skb, reserve);
2968 	skb_put(skb, linear);
2969 	skb->data_len = len - linear;
2970 	skb->len += len - linear;
2971 
2972 	return skb;
2973 }
2974 
2975 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2976 {
2977 	struct sock *sk = sock->sk;
2978 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2979 	struct sk_buff *skb;
2980 	struct net_device *dev;
2981 	__be16 proto;
2982 	unsigned char *addr = NULL;
2983 	int err, reserve = 0;
2984 	struct sockcm_cookie sockc;
2985 	struct virtio_net_hdr vnet_hdr = { 0 };
2986 	int offset = 0;
2987 	struct packet_sock *po = pkt_sk(sk);
2988 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2989 	int hard_header_len, hlen, tlen, linear;
2990 	int extra_len = 0;
2991 
2992 	/*
2993 	 *	Get and verify the address.
2994 	 */
2995 
2996 	if (likely(saddr == NULL)) {
2997 		dev	= packet_cached_dev_get(po);
2998 		proto	= READ_ONCE(po->num);
2999 	} else {
3000 		err = -EINVAL;
3001 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
3002 			goto out;
3003 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
3004 			goto out;
3005 		proto	= saddr->sll_protocol;
3006 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
3007 		if (sock->type == SOCK_DGRAM) {
3008 			if (dev && msg->msg_namelen < dev->addr_len +
3009 				   offsetof(struct sockaddr_ll, sll_addr))
3010 				goto out_unlock;
3011 			addr = saddr->sll_addr;
3012 		}
3013 	}
3014 
3015 	err = -ENXIO;
3016 	if (unlikely(dev == NULL))
3017 		goto out_unlock;
3018 	err = -ENETDOWN;
3019 	if (unlikely(!(dev->flags & IFF_UP)))
3020 		goto out_unlock;
3021 
3022 	sockcm_init(&sockc, sk);
3023 	if (msg->msg_controllen) {
3024 		err = sock_cmsg_send(sk, msg, &sockc);
3025 		if (unlikely(err))
3026 			goto out_unlock;
3027 	}
3028 
3029 	hard_header_len = READ_ONCE(dev->hard_header_len);
3030 	if (sock->type == SOCK_RAW)
3031 		reserve = hard_header_len;
3032 	if (vnet_hdr_sz) {
3033 		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3034 		if (err)
3035 			goto out_unlock;
3036 	}
3037 
3038 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3039 		if (!netif_supports_nofcs(dev)) {
3040 			err = -EPROTONOSUPPORT;
3041 			goto out_unlock;
3042 		}
3043 		extra_len = 4; /* We're doing our own CRC */
3044 	}
3045 
3046 	err = -EMSGSIZE;
3047 	if (!vnet_hdr.gso_type &&
3048 	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3049 		goto out_unlock;
3050 
3051 	err = -ENOBUFS;
3052 	hlen = LL_RESERVED_SPACE_EX(dev, hard_header_len);
3053 	tlen = dev->needed_tailroom;
3054 	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3055 	linear = max(linear, min_t(int, len, hard_header_len));
3056 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3057 			       msg->msg_flags & MSG_DONTWAIT, &err);
3058 	if (skb == NULL)
3059 		goto out_unlock;
3060 
3061 	skb_reset_network_header(skb);
3062 
3063 	err = -EINVAL;
3064 	if (sock->type == SOCK_DGRAM) {
3065 		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3066 		if (unlikely(offset < 0))
3067 			goto out_free;
3068 	} else if (reserve) {
3069 		skb_reserve(skb, -reserve);
3070 		if (len < reserve + sizeof(struct ipv6hdr) &&
3071 		    dev->min_header_len != hard_header_len)
3072 			skb_reset_network_header(skb);
3073 	}
3074 
3075 	/* Returns -EFAULT on error */
3076 	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3077 	if (err)
3078 		goto out_free;
3079 
3080 	if ((sock->type == SOCK_RAW &&
3081 	     !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3082 		err = -EINVAL;
3083 		goto out_free;
3084 	}
3085 
3086 	skb_setup_tx_timestamp(skb, &sockc);
3087 
3088 	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3089 	    !packet_extra_vlan_len_allowed(dev, skb)) {
3090 		err = -EMSGSIZE;
3091 		goto out_free;
3092 	}
3093 
3094 	skb->protocol = proto;
3095 	skb->dev = dev;
3096 	skb->priority = sockc.priority;
3097 	skb->mark = sockc.mark;
3098 	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
3099 
3100 	if (unlikely(extra_len == 4))
3101 		skb->no_fcs = 1;
3102 
3103 	packet_parse_headers(skb, sock);
3104 
3105 	if (vnet_hdr_sz) {
3106 		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3107 		if (err)
3108 			goto out_free;
3109 		len += vnet_hdr_sz;
3110 		virtio_net_hdr_set_proto(skb, &vnet_hdr);
3111 	}
3112 
3113 	err = packet_xmit(po, skb);
3114 
3115 	if (unlikely(err != 0)) {
3116 		if (err > 0)
3117 			err = net_xmit_errno(err);
3118 		if (err)
3119 			goto out_unlock;
3120 	}
3121 
3122 	dev_put(dev);
3123 
3124 	return len;
3125 
3126 out_free:
3127 	kfree_skb(skb);
3128 out_unlock:
3129 	dev_put(dev);
3130 out:
3131 	return err;
3132 }
3133 
3134 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3135 {
3136 	struct sock *sk = sock->sk;
3137 	struct packet_sock *po = pkt_sk(sk);
3138 
3139 	/* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3140 	 * tpacket_snd() will redo the check safely.
3141 	 */
3142 	if (data_race(po->tx_ring.pg_vec))
3143 		return tpacket_snd(po, msg);
3144 
3145 	return packet_snd(sock, msg, len);
3146 }
3147 
3148 /*
3149  *	Close a PACKET socket. This is fairly simple. We immediately go
3150  *	to 'closed' state and remove our protocol entry in the device list.
3151  */
3152 
3153 static int packet_release(struct socket *sock)
3154 {
3155 	struct sock *sk = sock->sk;
3156 	struct packet_sock *po;
3157 	struct packet_fanout *f;
3158 	struct net *net;
3159 	union tpacket_req_u req_u;
3160 
3161 	if (!sk)
3162 		return 0;
3163 
3164 	net = sock_net(sk);
3165 	po = pkt_sk(sk);
3166 
3167 	mutex_lock(&net->packet.sklist_lock);
3168 	sk_del_node_init_rcu(sk);
3169 	mutex_unlock(&net->packet.sklist_lock);
3170 
3171 	sock_prot_inuse_add(net, sk->sk_prot, -1);
3172 
3173 	spin_lock(&po->bind_lock);
3174 	unregister_prot_hook(sk, false);
3175 	WRITE_ONCE(po->num, 0);
3176 	packet_cached_dev_reset(po);
3177 
3178 	if (po->prot_hook.dev) {
3179 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3180 		po->prot_hook.dev = NULL;
3181 	}
3182 	spin_unlock(&po->bind_lock);
3183 
3184 	packet_flush_mclist(sk);
3185 
3186 	lock_sock(sk);
3187 	if (po->rx_ring.pg_vec) {
3188 		memset(&req_u, 0, sizeof(req_u));
3189 		packet_set_ring(sk, &req_u, 1, 0);
3190 	}
3191 
3192 	if (po->tx_ring.pg_vec) {
3193 		memset(&req_u, 0, sizeof(req_u));
3194 		packet_set_ring(sk, &req_u, 1, 1);
3195 	}
3196 	release_sock(sk);
3197 
3198 	f = fanout_release(sk);
3199 
3200 	synchronize_net();
3201 
3202 	kfree(po->rollover);
3203 	if (f) {
3204 		fanout_release_data(f);
3205 		kvfree(f);
3206 	}
3207 	/*
3208 	 *	Now the socket is dead. No more input will appear.
3209 	 */
3210 	sock_orphan(sk);
3211 	sock->sk = NULL;
3212 
3213 	/* Purge queues */
3214 
3215 	skb_queue_purge(&sk->sk_receive_queue);
3216 
3217 	sock_put(sk);
3218 	return 0;
3219 }
3220 
3221 /*
3222  *	Attach a packet hook.
3223  */
3224 
3225 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3226 			  __be16 proto)
3227 {
3228 	struct packet_sock *po = pkt_sk(sk);
3229 	struct net_device *dev = NULL;
3230 	bool unlisted = false;
3231 	bool need_rehook;
3232 	int ret = 0;
3233 
3234 	lock_sock(sk);
3235 	spin_lock(&po->bind_lock);
3236 	if (!proto)
3237 		proto = po->num;
3238 
3239 	rcu_read_lock();
3240 
3241 	if (po->fanout) {
3242 		ret = -EINVAL;
3243 		goto out_unlock;
3244 	}
3245 
3246 	if (name) {
3247 		dev = dev_get_by_name_rcu(sock_net(sk), name);
3248 		if (!dev) {
3249 			ret = -ENODEV;
3250 			goto out_unlock;
3251 		}
3252 	} else if (ifindex) {
3253 		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3254 		if (!dev) {
3255 			ret = -ENODEV;
3256 			goto out_unlock;
3257 		}
3258 	}
3259 
3260 	need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3261 
3262 	if (need_rehook) {
3263 		dev_hold(dev);
3264 		if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3265 			rcu_read_unlock();
3266 			/* prevents packet_notifier() from calling
3267 			 * register_prot_hook()
3268 			 */
3269 			WRITE_ONCE(po->num, 0);
3270 			__unregister_prot_hook(sk, true);
3271 			rcu_read_lock();
3272 			if (dev)
3273 				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3274 								 dev->ifindex);
3275 		}
3276 
3277 		BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3278 		WRITE_ONCE(po->num, proto);
3279 		po->prot_hook.type = proto;
3280 
3281 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3282 
3283 		if (unlikely(unlisted)) {
3284 			po->prot_hook.dev = NULL;
3285 			WRITE_ONCE(po->ifindex, -1);
3286 			packet_cached_dev_reset(po);
3287 		} else {
3288 			netdev_hold(dev, &po->prot_hook.dev_tracker,
3289 				    GFP_ATOMIC);
3290 			po->prot_hook.dev = dev;
3291 			WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3292 			packet_cached_dev_assign(po, dev);
3293 		}
3294 		dev_put(dev);
3295 	}
3296 
3297 	if (proto == 0 || !need_rehook)
3298 		goto out_unlock;
3299 
3300 	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3301 		register_prot_hook(sk);
3302 	} else {
3303 		sk->sk_err = ENETDOWN;
3304 		if (!sock_flag(sk, SOCK_DEAD))
3305 			sk_error_report(sk);
3306 	}
3307 
3308 out_unlock:
3309 	rcu_read_unlock();
3310 	spin_unlock(&po->bind_lock);
3311 	release_sock(sk);
3312 	return ret;
3313 }
3314 
3315 /*
3316  *	Bind a packet socket to a device
3317  */
3318 
3319 static int packet_bind_spkt(struct socket *sock, struct sockaddr_unsized *uaddr,
3320 			    int addr_len)
3321 {
3322 	struct sock *sk = sock->sk;
3323 	struct sockaddr *sa = (struct sockaddr *)uaddr;
3324 	char name[sizeof(sa->sa_data) + 1];
3325 
3326 	/*
3327 	 *	Check legality
3328 	 */
3329 
3330 	if (addr_len != sizeof(struct sockaddr))
3331 		return -EINVAL;
3332 	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3333 	 * zero-terminated.
3334 	 */
3335 	memcpy(name, sa->sa_data, sizeof(sa->sa_data));
3336 	name[sizeof(sa->sa_data)] = 0;
3337 
3338 	return packet_do_bind(sk, name, 0, 0);
3339 }
3340 
3341 static int packet_bind(struct socket *sock, struct sockaddr_unsized *uaddr, int addr_len)
3342 {
3343 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3344 	struct sock *sk = sock->sk;
3345 
3346 	/*
3347 	 *	Check legality
3348 	 */
3349 
3350 	if (addr_len < sizeof(struct sockaddr_ll))
3351 		return -EINVAL;
3352 	if (sll->sll_family != AF_PACKET)
3353 		return -EINVAL;
3354 
3355 	return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3356 }
3357 
3358 static struct proto packet_proto = {
3359 	.name	  = "PACKET",
3360 	.owner	  = THIS_MODULE,
3361 	.obj_size = sizeof(struct packet_sock),
3362 };
3363 
3364 /*
3365  *	Create a packet of type SOCK_PACKET.
3366  */
3367 
3368 static int packet_create(struct net *net, struct socket *sock, int protocol,
3369 			 int kern)
3370 {
3371 	struct sock *sk;
3372 	struct packet_sock *po;
3373 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3374 	int err;
3375 
3376 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3377 		return -EPERM;
3378 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3379 	    sock->type != SOCK_PACKET)
3380 		return -ESOCKTNOSUPPORT;
3381 
3382 	sock->state = SS_UNCONNECTED;
3383 
3384 	err = -ENOBUFS;
3385 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3386 	if (sk == NULL)
3387 		goto out;
3388 
3389 	sock->ops = &packet_ops;
3390 	if (sock->type == SOCK_PACKET)
3391 		sock->ops = &packet_ops_spkt;
3392 
3393 	po = pkt_sk(sk);
3394 	err = packet_alloc_pending(po);
3395 	if (err)
3396 		goto out_sk_free;
3397 
3398 	sock_init_data(sock, sk);
3399 
3400 	init_completion(&po->skb_completion);
3401 	sk->sk_family = PF_PACKET;
3402 	po->num = proto;
3403 
3404 	packet_cached_dev_reset(po);
3405 
3406 	sk->sk_destruct = packet_sock_destruct;
3407 
3408 	/*
3409 	 *	Attach a protocol block
3410 	 */
3411 
3412 	spin_lock_init(&po->bind_lock);
3413 	mutex_init(&po->pg_vec_lock);
3414 	po->rollover = NULL;
3415 	po->prot_hook.func = packet_rcv;
3416 
3417 	if (sock->type == SOCK_PACKET)
3418 		po->prot_hook.func = packet_rcv_spkt;
3419 
3420 	po->prot_hook.af_packet_priv = sk;
3421 	po->prot_hook.af_packet_net = sock_net(sk);
3422 
3423 	if (proto) {
3424 		po->prot_hook.type = proto;
3425 		__register_prot_hook(sk);
3426 	}
3427 
3428 	mutex_lock(&net->packet.sklist_lock);
3429 	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3430 	mutex_unlock(&net->packet.sklist_lock);
3431 
3432 	sock_prot_inuse_add(net, &packet_proto, 1);
3433 
3434 	return 0;
3435 out_sk_free:
3436 	sk_free(sk);
3437 out:
3438 	return err;
3439 }
3440 
3441 /*
3442  *	Pull a packet from our receive queue and hand it to the user.
3443  *	If necessary we block.
3444  */
3445 
3446 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3447 			  int flags)
3448 {
3449 	struct sock *sk = sock->sk;
3450 	struct sk_buff *skb;
3451 	int copied, err;
3452 	int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3453 	unsigned int origlen = 0;
3454 
3455 	err = -EINVAL;
3456 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3457 		goto out;
3458 
3459 #if 0
3460 	/* What error should we return now? EUNATTACH? */
3461 	if (pkt_sk(sk)->ifindex < 0)
3462 		return -ENODEV;
3463 #endif
3464 
3465 	if (flags & MSG_ERRQUEUE) {
3466 		err = sock_recv_errqueue(sk, msg, len,
3467 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3468 		goto out;
3469 	}
3470 
3471 	/*
3472 	 *	Call the generic datagram receiver. This handles all sorts
3473 	 *	of horrible races and re-entrancy so we can forget about it
3474 	 *	in the protocol layers.
3475 	 *
3476 	 *	Now it will return ENETDOWN, if device have just gone down,
3477 	 *	but then it will block.
3478 	 */
3479 
3480 	skb = skb_recv_datagram(sk, flags, &err);
3481 
3482 	/*
3483 	 *	An error occurred so return it. Because skb_recv_datagram()
3484 	 *	handles the blocking we don't see and worry about blocking
3485 	 *	retries.
3486 	 */
3487 
3488 	if (skb == NULL)
3489 		goto out;
3490 
3491 	packet_rcv_try_clear_pressure(pkt_sk(sk));
3492 
3493 	if (vnet_hdr_len) {
3494 		err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3495 		if (err)
3496 			goto out_free;
3497 	}
3498 
3499 	/* You lose any data beyond the buffer you gave. If it worries
3500 	 * a user program they can ask the device for its MTU
3501 	 * anyway.
3502 	 */
3503 	copied = skb->len;
3504 	if (copied > len) {
3505 		copied = len;
3506 		msg->msg_flags |= MSG_TRUNC;
3507 	}
3508 
3509 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3510 	if (err)
3511 		goto out_free;
3512 
3513 	if (sock->type != SOCK_PACKET) {
3514 		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3515 
3516 		/* Original length was stored in sockaddr_ll fields */
3517 		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3518 		sll->sll_family = AF_PACKET;
3519 		sll->sll_protocol = (sock->type == SOCK_DGRAM) ?
3520 			vlan_get_protocol_dgram(skb) : skb->protocol;
3521 	}
3522 
3523 	sock_recv_cmsgs(msg, sk, skb);
3524 
3525 	if (msg->msg_name) {
3526 		const size_t max_len = min(sizeof(skb->cb),
3527 					   sizeof(struct sockaddr_storage));
3528 		int copy_len;
3529 
3530 		/* If the address length field is there to be filled
3531 		 * in, we fill it in now.
3532 		 */
3533 		if (sock->type == SOCK_PACKET) {
3534 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3535 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3536 			copy_len = msg->msg_namelen;
3537 		} else {
3538 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3539 
3540 			msg->msg_namelen = sll->sll_halen +
3541 				offsetof(struct sockaddr_ll, sll_addr);
3542 			copy_len = msg->msg_namelen;
3543 			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3544 				memset(msg->msg_name +
3545 				       offsetof(struct sockaddr_ll, sll_addr),
3546 				       0, sizeof(sll->sll_addr));
3547 				msg->msg_namelen = sizeof(struct sockaddr_ll);
3548 			}
3549 		}
3550 		if (WARN_ON_ONCE(copy_len > max_len)) {
3551 			copy_len = max_len;
3552 			msg->msg_namelen = copy_len;
3553 		}
3554 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3555 	}
3556 
3557 	if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3558 		struct tpacket_auxdata aux;
3559 
3560 		aux.tp_status = TP_STATUS_USER;
3561 		if (skb->ip_summed == CHECKSUM_PARTIAL)
3562 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3563 		else if (skb->pkt_type != PACKET_OUTGOING &&
3564 			 skb_csum_unnecessary(skb))
3565 			aux.tp_status |= TP_STATUS_CSUM_VALID;
3566 		if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3567 			aux.tp_status |= TP_STATUS_GSO_TCP;
3568 
3569 		aux.tp_len = origlen;
3570 		aux.tp_snaplen = skb->len;
3571 		aux.tp_mac = 0;
3572 		aux.tp_net = skb_network_offset(skb);
3573 		if (skb_vlan_tag_present(skb)) {
3574 			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3575 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3576 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3577 		} else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
3578 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3579 			struct net_device *dev;
3580 
3581 			rcu_read_lock();
3582 			dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex);
3583 			if (dev) {
3584 				aux.tp_vlan_tci = vlan_get_tci(skb, dev);
3585 				aux.tp_vlan_tpid = ntohs(skb->protocol);
3586 				aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3587 			} else {
3588 				aux.tp_vlan_tci = 0;
3589 				aux.tp_vlan_tpid = 0;
3590 			}
3591 			rcu_read_unlock();
3592 		} else {
3593 			aux.tp_vlan_tci = 0;
3594 			aux.tp_vlan_tpid = 0;
3595 		}
3596 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3597 	}
3598 
3599 	/*
3600 	 *	Free or return the buffer as appropriate. Again this
3601 	 *	hides all the races and re-entrancy issues from us.
3602 	 */
3603 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3604 
3605 out_free:
3606 	skb_free_datagram(sk, skb);
3607 out:
3608 	return err;
3609 }
3610 
3611 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3612 			       int peer)
3613 {
3614 	struct net_device *dev;
3615 	struct sock *sk	= sock->sk;
3616 
3617 	if (peer)
3618 		return -EOPNOTSUPP;
3619 
3620 	uaddr->sa_family = AF_PACKET;
3621 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3622 	rcu_read_lock();
3623 	dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3624 	if (dev)
3625 		strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3626 	rcu_read_unlock();
3627 
3628 	return sizeof(*uaddr);
3629 }
3630 
3631 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3632 			  int peer)
3633 {
3634 	struct net_device *dev;
3635 	struct sock *sk = sock->sk;
3636 	struct packet_sock *po = pkt_sk(sk);
3637 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3638 	int ifindex;
3639 
3640 	if (peer)
3641 		return -EOPNOTSUPP;
3642 
3643 	ifindex = READ_ONCE(po->ifindex);
3644 	sll->sll_family = AF_PACKET;
3645 	sll->sll_ifindex = ifindex;
3646 	sll->sll_protocol = READ_ONCE(po->num);
3647 	sll->sll_pkttype = 0;
3648 	rcu_read_lock();
3649 	dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3650 	if (dev) {
3651 		sll->sll_hatype = dev->type;
3652 		sll->sll_halen = dev->addr_len;
3653 
3654 		/* Let __fortify_memcpy_chk() know the actual buffer size. */
3655 		memcpy(((struct sockaddr_storage *)sll)->__data +
3656 		       offsetof(struct sockaddr_ll, sll_addr) -
3657 		       offsetofend(struct sockaddr_ll, sll_family),
3658 		       dev->dev_addr, dev->addr_len);
3659 	} else {
3660 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3661 		sll->sll_halen = 0;
3662 	}
3663 	rcu_read_unlock();
3664 
3665 	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3666 }
3667 
3668 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3669 			 int what)
3670 {
3671 	switch (i->type) {
3672 	case PACKET_MR_MULTICAST:
3673 		if (i->alen != dev->addr_len)
3674 			return -EINVAL;
3675 		if (what > 0)
3676 			return dev_mc_add(dev, i->addr);
3677 		else
3678 			return dev_mc_del(dev, i->addr);
3679 		break;
3680 	case PACKET_MR_PROMISC:
3681 		return dev_set_promiscuity(dev, what);
3682 	case PACKET_MR_ALLMULTI:
3683 		return dev_set_allmulti(dev, what);
3684 	case PACKET_MR_UNICAST:
3685 		if (i->alen != dev->addr_len)
3686 			return -EINVAL;
3687 		if (what > 0)
3688 			return dev_uc_add(dev, i->addr);
3689 		else
3690 			return dev_uc_del(dev, i->addr);
3691 		break;
3692 	default:
3693 		break;
3694 	}
3695 	return 0;
3696 }
3697 
3698 static void packet_dev_mclist_delete(struct net_device *dev,
3699 				     struct packet_mclist **mlp,
3700 				     struct list_head *list)
3701 {
3702 	struct packet_mclist *ml;
3703 
3704 	while ((ml = *mlp) != NULL) {
3705 		if (ml->ifindex == dev->ifindex) {
3706 			list_add(&ml->remove_list, list);
3707 			*mlp = ml->next;
3708 		} else
3709 			mlp = &ml->next;
3710 	}
3711 }
3712 
3713 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3714 {
3715 	struct packet_sock *po = pkt_sk(sk);
3716 	struct packet_mclist *ml, *i;
3717 	struct net_device *dev;
3718 	int err;
3719 
3720 	rtnl_lock();
3721 
3722 	err = -ENODEV;
3723 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3724 	if (!dev)
3725 		goto done;
3726 
3727 	err = -EINVAL;
3728 	if (mreq->mr_alen > dev->addr_len)
3729 		goto done;
3730 
3731 	err = -ENOBUFS;
3732 	i = kmalloc_obj(*i);
3733 	if (i == NULL)
3734 		goto done;
3735 
3736 	err = 0;
3737 	for (ml = po->mclist; ml; ml = ml->next) {
3738 		if (ml->ifindex == mreq->mr_ifindex &&
3739 		    ml->type == mreq->mr_type &&
3740 		    ml->alen == mreq->mr_alen &&
3741 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3742 			ml->count++;
3743 			/* Free the new element ... */
3744 			kfree(i);
3745 			goto done;
3746 		}
3747 	}
3748 
3749 	i->type = mreq->mr_type;
3750 	i->ifindex = mreq->mr_ifindex;
3751 	i->alen = mreq->mr_alen;
3752 	memcpy(i->addr, mreq->mr_address, i->alen);
3753 	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3754 	i->count = 1;
3755 	INIT_LIST_HEAD(&i->remove_list);
3756 	i->next = po->mclist;
3757 	po->mclist = i;
3758 	err = packet_dev_mc(dev, i, 1);
3759 	if (err) {
3760 		po->mclist = i->next;
3761 		kfree(i);
3762 	}
3763 
3764 done:
3765 	rtnl_unlock();
3766 	return err;
3767 }
3768 
3769 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3770 {
3771 	struct packet_mclist *ml, **mlp;
3772 
3773 	rtnl_lock();
3774 
3775 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3776 		if (ml->ifindex == mreq->mr_ifindex &&
3777 		    ml->type == mreq->mr_type &&
3778 		    ml->alen == mreq->mr_alen &&
3779 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3780 			if (--ml->count == 0) {
3781 				struct net_device *dev;
3782 				*mlp = ml->next;
3783 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3784 				if (dev)
3785 					packet_dev_mc(dev, ml, -1);
3786 				kfree(ml);
3787 			}
3788 			break;
3789 		}
3790 	}
3791 	rtnl_unlock();
3792 	return 0;
3793 }
3794 
3795 static void packet_flush_mclist(struct sock *sk)
3796 {
3797 	struct packet_sock *po = pkt_sk(sk);
3798 	struct packet_mclist *ml;
3799 
3800 	if (!po->mclist)
3801 		return;
3802 
3803 	rtnl_lock();
3804 	while ((ml = po->mclist) != NULL) {
3805 		struct net_device *dev;
3806 
3807 		po->mclist = ml->next;
3808 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3809 		if (dev != NULL)
3810 			packet_dev_mc(dev, ml, -1);
3811 		kfree(ml);
3812 	}
3813 	rtnl_unlock();
3814 }
3815 
3816 static int
3817 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3818 		  unsigned int optlen)
3819 {
3820 	struct sock *sk = sock->sk;
3821 	struct packet_sock *po = pkt_sk(sk);
3822 	int ret;
3823 
3824 	if (level != SOL_PACKET)
3825 		return -ENOPROTOOPT;
3826 
3827 	switch (optname) {
3828 	case PACKET_ADD_MEMBERSHIP:
3829 	case PACKET_DROP_MEMBERSHIP:
3830 	{
3831 		struct packet_mreq_max mreq;
3832 		int len = optlen;
3833 		memset(&mreq, 0, sizeof(mreq));
3834 		if (len < sizeof(struct packet_mreq))
3835 			return -EINVAL;
3836 		if (len > sizeof(mreq))
3837 			len = sizeof(mreq);
3838 		if (copy_from_sockptr(&mreq, optval, len))
3839 			return -EFAULT;
3840 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3841 			return -EINVAL;
3842 		if (optname == PACKET_ADD_MEMBERSHIP)
3843 			ret = packet_mc_add(sk, &mreq);
3844 		else
3845 			ret = packet_mc_drop(sk, &mreq);
3846 		return ret;
3847 	}
3848 
3849 	case PACKET_RX_RING:
3850 	case PACKET_TX_RING:
3851 	{
3852 		union tpacket_req_u req_u;
3853 
3854 		ret = -EINVAL;
3855 		lock_sock(sk);
3856 		switch (po->tp_version) {
3857 		case TPACKET_V1:
3858 		case TPACKET_V2:
3859 			if (optlen < sizeof(req_u.req))
3860 				break;
3861 			ret = copy_from_sockptr(&req_u.req, optval,
3862 						sizeof(req_u.req)) ?
3863 						-EINVAL : 0;
3864 			break;
3865 		case TPACKET_V3:
3866 		default:
3867 			if (optlen < sizeof(req_u.req3))
3868 				break;
3869 			ret = copy_from_sockptr(&req_u.req3, optval,
3870 						sizeof(req_u.req3)) ?
3871 						-EINVAL : 0;
3872 			break;
3873 		}
3874 		if (!ret)
3875 			ret = packet_set_ring(sk, &req_u, 0,
3876 					      optname == PACKET_TX_RING);
3877 		release_sock(sk);
3878 		return ret;
3879 	}
3880 	case PACKET_COPY_THRESH:
3881 	{
3882 		int val;
3883 
3884 		if (optlen != sizeof(val))
3885 			return -EINVAL;
3886 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3887 			return -EFAULT;
3888 
3889 		WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3890 		return 0;
3891 	}
3892 	case PACKET_VERSION:
3893 	{
3894 		int val;
3895 
3896 		if (optlen != sizeof(val))
3897 			return -EINVAL;
3898 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3899 			return -EFAULT;
3900 		switch (val) {
3901 		case TPACKET_V1:
3902 		case TPACKET_V2:
3903 		case TPACKET_V3:
3904 			break;
3905 		default:
3906 			return -EINVAL;
3907 		}
3908 		lock_sock(sk);
3909 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3910 			ret = -EBUSY;
3911 		} else {
3912 			po->tp_version = val;
3913 			ret = 0;
3914 		}
3915 		release_sock(sk);
3916 		return ret;
3917 	}
3918 	case PACKET_RESERVE:
3919 	{
3920 		unsigned int val;
3921 
3922 		if (optlen != sizeof(val))
3923 			return -EINVAL;
3924 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3925 			return -EFAULT;
3926 		if (val > INT_MAX)
3927 			return -EINVAL;
3928 		lock_sock(sk);
3929 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3930 			ret = -EBUSY;
3931 		} else {
3932 			po->tp_reserve = val;
3933 			ret = 0;
3934 		}
3935 		release_sock(sk);
3936 		return ret;
3937 	}
3938 	case PACKET_LOSS:
3939 	{
3940 		unsigned int val;
3941 
3942 		if (optlen != sizeof(val))
3943 			return -EINVAL;
3944 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3945 			return -EFAULT;
3946 
3947 		lock_sock(sk);
3948 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3949 			ret = -EBUSY;
3950 		} else {
3951 			packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3952 			ret = 0;
3953 		}
3954 		release_sock(sk);
3955 		return ret;
3956 	}
3957 	case PACKET_AUXDATA:
3958 	{
3959 		int val;
3960 
3961 		if (optlen < sizeof(val))
3962 			return -EINVAL;
3963 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3964 			return -EFAULT;
3965 
3966 		packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3967 		return 0;
3968 	}
3969 	case PACKET_ORIGDEV:
3970 	{
3971 		int val;
3972 
3973 		if (optlen < sizeof(val))
3974 			return -EINVAL;
3975 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3976 			return -EFAULT;
3977 
3978 		packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3979 		return 0;
3980 	}
3981 	case PACKET_VNET_HDR:
3982 	case PACKET_VNET_HDR_SZ:
3983 	{
3984 		int val, hdr_len;
3985 
3986 		if (sock->type != SOCK_RAW)
3987 			return -EINVAL;
3988 		if (optlen < sizeof(val))
3989 			return -EINVAL;
3990 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3991 			return -EFAULT;
3992 
3993 		if (optname == PACKET_VNET_HDR_SZ) {
3994 			if (val && val != sizeof(struct virtio_net_hdr) &&
3995 			    val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
3996 				return -EINVAL;
3997 			hdr_len = val;
3998 		} else {
3999 			hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
4000 		}
4001 		lock_sock(sk);
4002 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
4003 			ret = -EBUSY;
4004 		} else {
4005 			WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
4006 			ret = 0;
4007 		}
4008 		release_sock(sk);
4009 		return ret;
4010 	}
4011 	case PACKET_TIMESTAMP:
4012 	{
4013 		int val;
4014 
4015 		if (optlen != sizeof(val))
4016 			return -EINVAL;
4017 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4018 			return -EFAULT;
4019 
4020 		WRITE_ONCE(po->tp_tstamp, val);
4021 		return 0;
4022 	}
4023 	case PACKET_FANOUT:
4024 	{
4025 		struct fanout_args args = { 0 };
4026 
4027 		if (optlen != sizeof(int) && optlen != sizeof(args))
4028 			return -EINVAL;
4029 		if (copy_from_sockptr(&args, optval, optlen))
4030 			return -EFAULT;
4031 
4032 		return fanout_add(sk, &args);
4033 	}
4034 	case PACKET_FANOUT_DATA:
4035 	{
4036 		/* Paired with the WRITE_ONCE() in fanout_add() */
4037 		if (!READ_ONCE(po->fanout))
4038 			return -EINVAL;
4039 
4040 		return fanout_set_data(po, optval, optlen);
4041 	}
4042 	case PACKET_IGNORE_OUTGOING:
4043 	{
4044 		int val;
4045 
4046 		if (optlen != sizeof(val))
4047 			return -EINVAL;
4048 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4049 			return -EFAULT;
4050 		if (val < 0 || val > 1)
4051 			return -EINVAL;
4052 
4053 		WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4054 		return 0;
4055 	}
4056 	case PACKET_TX_HAS_OFF:
4057 	{
4058 		unsigned int val;
4059 
4060 		if (optlen != sizeof(val))
4061 			return -EINVAL;
4062 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4063 			return -EFAULT;
4064 
4065 		lock_sock(sk);
4066 		if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4067 			packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4068 
4069 		release_sock(sk);
4070 		return 0;
4071 	}
4072 	case PACKET_QDISC_BYPASS:
4073 	{
4074 		int val;
4075 
4076 		if (optlen != sizeof(val))
4077 			return -EINVAL;
4078 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4079 			return -EFAULT;
4080 
4081 		packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4082 		return 0;
4083 	}
4084 	default:
4085 		return -ENOPROTOOPT;
4086 	}
4087 }
4088 
4089 static int packet_getsockopt(struct socket *sock, int level, int optname,
4090 			     sockopt_t *opt)
4091 {
4092 	int len;
4093 	int val, lv = sizeof(val);
4094 	struct sock *sk = sock->sk;
4095 	struct packet_sock *po = pkt_sk(sk);
4096 	void *data = &val;
4097 	union tpacket_stats_u st;
4098 	struct tpacket_rollover_stats rstats;
4099 	int drops;
4100 
4101 	if (level != SOL_PACKET)
4102 		return -ENOPROTOOPT;
4103 
4104 	len = opt->optlen;
4105 
4106 	if (len < 0)
4107 		return -EINVAL;
4108 
4109 	switch (optname) {
4110 	case PACKET_STATISTICS:
4111 		spin_lock_bh(&sk->sk_receive_queue.lock);
4112 		memcpy(&st, &po->stats, sizeof(st));
4113 		memset(&po->stats, 0, sizeof(po->stats));
4114 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4115 		drops = atomic_xchg(&po->tp_drops, 0);
4116 
4117 		if (po->tp_version == TPACKET_V3) {
4118 			lv = sizeof(struct tpacket_stats_v3);
4119 			st.stats3.tp_drops = drops;
4120 			st.stats3.tp_packets += drops;
4121 			data = &st.stats3;
4122 		} else {
4123 			lv = sizeof(struct tpacket_stats);
4124 			st.stats1.tp_drops = drops;
4125 			st.stats1.tp_packets += drops;
4126 			data = &st.stats1;
4127 		}
4128 
4129 		break;
4130 	case PACKET_AUXDATA:
4131 		val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4132 		break;
4133 	case PACKET_ORIGDEV:
4134 		val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4135 		break;
4136 	case PACKET_VNET_HDR:
4137 		val = !!READ_ONCE(po->vnet_hdr_sz);
4138 		break;
4139 	case PACKET_VNET_HDR_SZ:
4140 		val = READ_ONCE(po->vnet_hdr_sz);
4141 		break;
4142 	case PACKET_COPY_THRESH:
4143 		val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4144 		break;
4145 	case PACKET_VERSION:
4146 		val = po->tp_version;
4147 		break;
4148 	case PACKET_HDRLEN:
4149 		if (len > sizeof(int))
4150 			len = sizeof(int);
4151 		if (len < sizeof(int))
4152 			return -EINVAL;
4153 		if (copy_from_iter(&val, len, &opt->iter_in) != len)
4154 			return -EFAULT;
4155 		switch (val) {
4156 		case TPACKET_V1:
4157 			val = sizeof(struct tpacket_hdr);
4158 			break;
4159 		case TPACKET_V2:
4160 			val = sizeof(struct tpacket2_hdr);
4161 			break;
4162 		case TPACKET_V3:
4163 			val = sizeof(struct tpacket3_hdr);
4164 			break;
4165 		default:
4166 			return -EINVAL;
4167 		}
4168 		break;
4169 	case PACKET_RESERVE:
4170 		val = po->tp_reserve;
4171 		break;
4172 	case PACKET_LOSS:
4173 		val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4174 		break;
4175 	case PACKET_TIMESTAMP:
4176 		val = READ_ONCE(po->tp_tstamp);
4177 		break;
4178 	case PACKET_FANOUT:
4179 		val = (po->fanout ?
4180 		       ((u32)po->fanout->id |
4181 			((u32)po->fanout->type << 16) |
4182 			((u32)po->fanout->flags << 24)) :
4183 		       0);
4184 		break;
4185 	case PACKET_IGNORE_OUTGOING:
4186 		val = READ_ONCE(po->prot_hook.ignore_outgoing);
4187 		break;
4188 	case PACKET_ROLLOVER_STATS:
4189 		if (!po->rollover)
4190 			return -EINVAL;
4191 		rstats.tp_all = atomic_long_read(&po->rollover->num);
4192 		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4193 		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4194 		data = &rstats;
4195 		lv = sizeof(rstats);
4196 		break;
4197 	case PACKET_TX_HAS_OFF:
4198 		val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4199 		break;
4200 	case PACKET_QDISC_BYPASS:
4201 		val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4202 		break;
4203 	default:
4204 		return -ENOPROTOOPT;
4205 	}
4206 
4207 	if (len > lv)
4208 		len = lv;
4209 	opt->optlen = len;
4210 	if (copy_to_iter(data, len, &opt->iter_out) != len)
4211 		return -EFAULT;
4212 	return 0;
4213 }
4214 
4215 static int packet_notifier(struct notifier_block *this,
4216 			   unsigned long msg, void *ptr)
4217 {
4218 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4219 	struct net *net = dev_net(dev);
4220 	struct packet_mclist *ml, *tmp;
4221 	LIST_HEAD(mclist);
4222 	struct sock *sk;
4223 
4224 	rcu_read_lock();
4225 	sk_for_each_rcu(sk, &net->packet.sklist) {
4226 		struct packet_sock *po = pkt_sk(sk);
4227 
4228 		switch (msg) {
4229 		case NETDEV_UNREGISTER:
4230 			if (po->mclist)
4231 				packet_dev_mclist_delete(dev, &po->mclist,
4232 							 &mclist);
4233 			fallthrough;
4234 
4235 		case NETDEV_DOWN:
4236 			if (dev->ifindex == po->ifindex) {
4237 				spin_lock(&po->bind_lock);
4238 				if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4239 					__unregister_prot_hook(sk, false);
4240 					sk->sk_err = ENETDOWN;
4241 					if (!sock_flag(sk, SOCK_DEAD))
4242 						sk_error_report(sk);
4243 				}
4244 				if (msg == NETDEV_UNREGISTER) {
4245 					packet_cached_dev_reset(po);
4246 					WRITE_ONCE(po->ifindex, -1);
4247 					netdev_put(po->prot_hook.dev,
4248 						   &po->prot_hook.dev_tracker);
4249 					po->prot_hook.dev = NULL;
4250 				}
4251 				spin_unlock(&po->bind_lock);
4252 			}
4253 			break;
4254 		case NETDEV_UP:
4255 			if (dev->ifindex == po->ifindex) {
4256 				spin_lock(&po->bind_lock);
4257 				if (po->num)
4258 					register_prot_hook(sk);
4259 				spin_unlock(&po->bind_lock);
4260 			}
4261 			break;
4262 		}
4263 	}
4264 	rcu_read_unlock();
4265 
4266 	/* packet_dev_mc might grab instance locks so can't run under rcu */
4267 	list_for_each_entry_safe(ml, tmp, &mclist, remove_list) {
4268 		packet_dev_mc(dev, ml, -1);
4269 		kfree(ml);
4270 	}
4271 
4272 	return NOTIFY_DONE;
4273 }
4274 
4275 
4276 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4277 			unsigned long arg)
4278 {
4279 	struct sock *sk = sock->sk;
4280 
4281 	switch (cmd) {
4282 	case SIOCOUTQ:
4283 	{
4284 		int amount = sk_wmem_alloc_get(sk);
4285 
4286 		return put_user(amount, (int __user *)arg);
4287 	}
4288 	case SIOCINQ:
4289 	{
4290 		struct sk_buff *skb;
4291 		int amount = 0;
4292 
4293 		spin_lock_bh(&sk->sk_receive_queue.lock);
4294 		skb = skb_peek(&sk->sk_receive_queue);
4295 		if (skb)
4296 			amount = skb->len;
4297 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4298 		return put_user(amount, (int __user *)arg);
4299 	}
4300 #ifdef CONFIG_INET
4301 	case SIOCADDRT:
4302 	case SIOCDELRT:
4303 	case SIOCDARP:
4304 	case SIOCGARP:
4305 	case SIOCSARP:
4306 	case SIOCGIFADDR:
4307 	case SIOCSIFADDR:
4308 	case SIOCGIFBRDADDR:
4309 	case SIOCSIFBRDADDR:
4310 	case SIOCGIFNETMASK:
4311 	case SIOCSIFNETMASK:
4312 	case SIOCGIFDSTADDR:
4313 	case SIOCSIFDSTADDR:
4314 	case SIOCSIFFLAGS:
4315 		return inet_dgram_ops.ioctl(sock, cmd, arg);
4316 #endif
4317 
4318 	default:
4319 		return -ENOIOCTLCMD;
4320 	}
4321 	return 0;
4322 }
4323 
4324 static __poll_t packet_poll(struct file *file, struct socket *sock,
4325 				poll_table *wait)
4326 {
4327 	struct sock *sk = sock->sk;
4328 	struct packet_sock *po = pkt_sk(sk);
4329 	__poll_t mask = datagram_poll(file, sock, wait);
4330 
4331 	spin_lock_bh(&sk->sk_receive_queue.lock);
4332 	if (po->rx_ring.pg_vec) {
4333 		if (!packet_previous_rx_frame(po, &po->rx_ring,
4334 			TP_STATUS_KERNEL))
4335 			mask |= EPOLLIN | EPOLLRDNORM;
4336 	}
4337 	__packet_rcv_try_clear_pressure(po);
4338 	spin_unlock_bh(&sk->sk_receive_queue.lock);
4339 	spin_lock_bh(&sk->sk_write_queue.lock);
4340 	if (po->tx_ring.pg_vec) {
4341 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4342 			mask |= EPOLLOUT | EPOLLWRNORM;
4343 	}
4344 	spin_unlock_bh(&sk->sk_write_queue.lock);
4345 	return mask;
4346 }
4347 
4348 
4349 /* Dirty? Well, I still did not learn better way to account
4350  * for user mmaps.
4351  */
4352 
4353 static void packet_mm_open(struct vm_area_struct *vma)
4354 {
4355 	struct file *file = vma->vm_file;
4356 	struct socket *sock = file->private_data;
4357 	struct sock *sk = sock->sk;
4358 
4359 	if (sk)
4360 		atomic_long_inc(&pkt_sk(sk)->mapped);
4361 }
4362 
4363 static void packet_mm_close(struct vm_area_struct *vma)
4364 {
4365 	struct file *file = vma->vm_file;
4366 	struct socket *sock = file->private_data;
4367 	struct sock *sk = sock->sk;
4368 
4369 	if (sk)
4370 		atomic_long_dec(&pkt_sk(sk)->mapped);
4371 }
4372 
4373 static const struct vm_operations_struct packet_mmap_ops = {
4374 	.open	=	packet_mm_open,
4375 	.close	=	packet_mm_close,
4376 };
4377 
4378 struct packet_pg_vec {
4379 	struct packet_pg_vec_free *deferred;
4380 	unsigned int order;
4381 	unsigned int len;
4382 	struct pgv pg_vec[] __counted_by(len);
4383 };
4384 
4385 struct packet_pg_vec_free {
4386 	struct delayed_work work;
4387 	struct sock *sk;
4388 	struct packet_pg_vec *vec;
4389 };
4390 
4391 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4392 			unsigned int len)
4393 {
4394 	struct packet_pg_vec *vec;
4395 	int i;
4396 
4397 	vec = container_of_const(pg_vec, struct packet_pg_vec, pg_vec[0]);
4398 	for (i = 0; i < len; i++) {
4399 		if (likely(pg_vec[i].buffer)) {
4400 			if (is_vmalloc_addr(pg_vec[i].buffer))
4401 				vfree(pg_vec[i].buffer);
4402 			else
4403 				free_pages((unsigned long)pg_vec[i].buffer,
4404 					   order);
4405 			pg_vec[i].buffer = NULL;
4406 		}
4407 	}
4408 	kfree(vec->deferred);
4409 	kfree(vec);
4410 }
4411 
4412 static void packet_free_pg_vec_work(struct work_struct *work)
4413 {
4414 	struct packet_pg_vec_free *deferred;
4415 	struct packet_pg_vec *vec;
4416 	struct sock *sk;
4417 
4418 	deferred = container_of_const(to_delayed_work(work),
4419 				      struct packet_pg_vec_free, work);
4420 	vec = deferred->vec;
4421 	sk = deferred->sk;
4422 	if (sk_wmem_alloc_get(sk)) {
4423 		queue_delayed_work(system_long_wq, &deferred->work, 1);
4424 		return;
4425 	}
4426 
4427 	free_pg_vec(vec->pg_vec, vec->order, vec->len);
4428 	sock_put(sk);
4429 }
4430 
4431 static void packet_free_tx_ring(struct sock *sk, struct pgv *pg_vec,
4432 				unsigned int order, unsigned int len)
4433 {
4434 	struct packet_pg_vec_free *deferred;
4435 	struct packet_pg_vec *vec;
4436 
4437 	vec = container_of_const(pg_vec, struct packet_pg_vec, pg_vec[0]);
4438 	deferred = vec->deferred;
4439 	if (!deferred || !sk_wmem_alloc_get(sk)) {
4440 		free_pg_vec(pg_vec, order, len);
4441 		return;
4442 	}
4443 
4444 	/* A detached ring's pending count can miss late skb destructors. */
4445 	deferred->sk = sk;
4446 	sock_hold(sk);
4447 	queue_delayed_work(system_long_wq, &deferred->work, 0);
4448 }
4449 
4450 static char *alloc_one_pg_vec_page(unsigned long order)
4451 {
4452 	char *buffer;
4453 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4454 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4455 
4456 	buffer = (char *) __get_free_pages(gfp_flags, order);
4457 	if (buffer)
4458 		return buffer;
4459 
4460 	/* __get_free_pages failed, fall back to vmalloc */
4461 	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4462 	if (buffer)
4463 		return buffer;
4464 
4465 	/* vmalloc failed, lets dig into swap here */
4466 	gfp_flags &= ~__GFP_NORETRY;
4467 	buffer = (char *) __get_free_pages(gfp_flags, order);
4468 	if (buffer)
4469 		return buffer;
4470 
4471 	/* complete and utter failure */
4472 	return NULL;
4473 }
4474 
4475 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order, bool tx_ring)
4476 {
4477 	unsigned int block_nr = req->tp_block_nr;
4478 	struct packet_pg_vec *vec;
4479 	struct pgv *pg_vec;
4480 	int i;
4481 
4482 	vec = kzalloc_flex(*vec, pg_vec, block_nr, GFP_KERNEL | __GFP_NOWARN);
4483 	if (unlikely(!vec))
4484 		return NULL;
4485 	vec->order = order;
4486 	vec->len = block_nr;
4487 	pg_vec = vec->pg_vec;
4488 
4489 	for (i = 0; i < block_nr; i++) {
4490 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4491 		if (unlikely(!pg_vec[i].buffer))
4492 			goto out_free_pgvec;
4493 
4494 		if (tx_ring && !vec->deferred &&
4495 		    is_vmalloc_addr(pg_vec[i].buffer)) {
4496 			vec->deferred = kzalloc_obj(*vec->deferred,
4497 						    GFP_KERNEL | __GFP_NOWARN);
4498 			if (!vec->deferred)
4499 				goto out_free_pgvec;
4500 			vec->deferred->vec = vec;
4501 			INIT_DELAYED_WORK(&vec->deferred->work,
4502 					  packet_free_pg_vec_work);
4503 		}
4504 	}
4505 
4506 out:
4507 	return pg_vec;
4508 
4509 out_free_pgvec:
4510 	free_pg_vec(pg_vec, order, block_nr);
4511 	pg_vec = NULL;
4512 	goto out;
4513 }
4514 
4515 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4516 		int closing, int tx_ring)
4517 {
4518 	struct pgv *pg_vec = NULL;
4519 	struct packet_sock *po = pkt_sk(sk);
4520 	unsigned long *rx_owner_map = NULL;
4521 	int was_running, order = 0;
4522 	struct packet_ring_buffer *rb;
4523 	struct sk_buff_head *rb_queue;
4524 	__be16 num;
4525 	int err;
4526 	/* Added to avoid minimal code churn */
4527 	struct tpacket_req *req = &req_u->req;
4528 
4529 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4530 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4531 
4532 	err = -EBUSY;
4533 	if (!closing) {
4534 		if (atomic_long_read(&po->mapped))
4535 			goto out;
4536 		if (packet_read_pending(rb))
4537 			goto out;
4538 	}
4539 
4540 	if (req->tp_block_nr) {
4541 		unsigned int min_frame_size;
4542 
4543 		/* Sanity tests and some calculations */
4544 		err = -EBUSY;
4545 		if (unlikely(rb->pg_vec))
4546 			goto out;
4547 
4548 		switch (po->tp_version) {
4549 		case TPACKET_V1:
4550 			po->tp_hdrlen = TPACKET_HDRLEN;
4551 			break;
4552 		case TPACKET_V2:
4553 			po->tp_hdrlen = TPACKET2_HDRLEN;
4554 			break;
4555 		case TPACKET_V3:
4556 			po->tp_hdrlen = TPACKET3_HDRLEN;
4557 			break;
4558 		}
4559 
4560 		err = -EINVAL;
4561 		if (unlikely((int)req->tp_block_size <= 0))
4562 			goto out;
4563 		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4564 			goto out;
4565 		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4566 		if (po->tp_version >= TPACKET_V3 &&
4567 		    req->tp_block_size <
4568 		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4569 			goto out;
4570 		if (unlikely(req->tp_frame_size < min_frame_size))
4571 			goto out;
4572 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4573 			goto out;
4574 
4575 		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4576 		if (unlikely(rb->frames_per_block == 0))
4577 			goto out;
4578 		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4579 			goto out;
4580 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4581 					req->tp_frame_nr))
4582 			goto out;
4583 
4584 		err = -ENOMEM;
4585 		order = get_order(req->tp_block_size);
4586 		pg_vec = alloc_pg_vec(req, order, tx_ring);
4587 		if (unlikely(!pg_vec))
4588 			goto out;
4589 		switch (po->tp_version) {
4590 		case TPACKET_V3:
4591 			/* Block transmit is not supported yet */
4592 			if (!tx_ring) {
4593 				init_prb_bdqc(po, rb, pg_vec, req_u);
4594 			} else {
4595 				struct tpacket_req3 *req3 = &req_u->req3;
4596 
4597 				if (req3->tp_retire_blk_tov ||
4598 				    req3->tp_sizeof_priv ||
4599 				    req3->tp_feature_req_word) {
4600 					err = -EINVAL;
4601 					goto out_free_pg_vec;
4602 				}
4603 			}
4604 			break;
4605 		default:
4606 			if (!tx_ring) {
4607 				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4608 					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4609 				if (!rx_owner_map)
4610 					goto out_free_pg_vec;
4611 			}
4612 			break;
4613 		}
4614 	}
4615 	/* Done */
4616 	else {
4617 		err = -EINVAL;
4618 		if (unlikely(req->tp_frame_nr))
4619 			goto out;
4620 	}
4621 
4622 
4623 	/* Detach socket from network */
4624 	spin_lock(&po->bind_lock);
4625 	was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4626 	num = po->num;
4627 	WRITE_ONCE(po->num, 0);
4628 	if (was_running)
4629 		__unregister_prot_hook(sk, false);
4630 
4631 	spin_unlock(&po->bind_lock);
4632 
4633 	synchronize_net();
4634 
4635 	err = -EBUSY;
4636 	mutex_lock(&po->pg_vec_lock);
4637 	if (closing || atomic_long_read(&po->mapped) == 0) {
4638 		if (tx_ring && !closing && packet_read_pending(rb))
4639 			goto out_unlock;
4640 
4641 		err = 0;
4642 		spin_lock_bh(&rb_queue->lock);
4643 		swap(rb->pg_vec, pg_vec);
4644 		if (po->tp_version <= TPACKET_V2)
4645 			swap(rb->rx_owner_map, rx_owner_map);
4646 		rb->frame_max = (req->tp_frame_nr - 1);
4647 		rb->head = 0;
4648 		rb->frame_size = req->tp_frame_size;
4649 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4650 						tpacket_rcv : packet_rcv;
4651 		spin_unlock_bh(&rb_queue->lock);
4652 
4653 		swap(rb->pg_vec_order, order);
4654 		swap(rb->pg_vec_len, req->tp_block_nr);
4655 
4656 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4657 		skb_queue_purge(rb_queue);
4658 		if (atomic_long_read(&po->mapped))
4659 			pr_err("packet_mmap: vma is busy: %ld\n",
4660 			       atomic_long_read(&po->mapped));
4661 	}
4662 out_unlock:
4663 	mutex_unlock(&po->pg_vec_lock);
4664 
4665 	spin_lock(&po->bind_lock);
4666 	WRITE_ONCE(po->num, num);
4667 	/*
4668 	 * NETDEV_UNREGISTER may have invalidated the binding while bind_lock
4669 	 * was dropped above.  Do not re-add a fanout hook to a dead device.
4670 	 */
4671 	if (was_running && READ_ONCE(po->ifindex) != -1)
4672 		register_prot_hook(sk);
4673 
4674 	spin_unlock(&po->bind_lock);
4675 	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4676 		/* Because we don't support block-based V3 on tx-ring */
4677 		if (!tx_ring)
4678 			prb_shutdown_retire_blk_timer(po, rb_queue);
4679 	}
4680 
4681 out_free_pg_vec:
4682 	if (pg_vec) {
4683 		bitmap_free(rx_owner_map);
4684 		if (tx_ring && closing)
4685 			packet_free_tx_ring(sk, pg_vec, order, req->tp_block_nr);
4686 		else
4687 			free_pg_vec(pg_vec, order, req->tp_block_nr);
4688 	}
4689 out:
4690 	return err;
4691 }
4692 
4693 static int packet_mmap(struct file *file, struct socket *sock,
4694 		struct vm_area_struct *vma)
4695 {
4696 	struct sock *sk = sock->sk;
4697 	struct packet_sock *po = pkt_sk(sk);
4698 	unsigned long size, expected_size;
4699 	struct packet_ring_buffer *rb;
4700 	unsigned long start;
4701 	int err = -EINVAL;
4702 	int i;
4703 
4704 	if (vma->vm_pgoff)
4705 		return -EINVAL;
4706 
4707 	mutex_lock(&po->pg_vec_lock);
4708 
4709 	expected_size = 0;
4710 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4711 		if (rb->pg_vec) {
4712 			expected_size += rb->pg_vec_len
4713 						* rb->pg_vec_pages
4714 						* PAGE_SIZE;
4715 		}
4716 	}
4717 
4718 	if (expected_size == 0)
4719 		goto out;
4720 
4721 	size = vma->vm_end - vma->vm_start;
4722 	if (size != expected_size)
4723 		goto out;
4724 
4725 	start = vma->vm_start;
4726 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4727 		if (rb->pg_vec == NULL)
4728 			continue;
4729 
4730 		for (i = 0; i < rb->pg_vec_len; i++) {
4731 			struct page *page;
4732 			void *kaddr = rb->pg_vec[i].buffer;
4733 			int pg_num;
4734 
4735 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4736 				page = pgv_to_page(kaddr);
4737 				err = vm_insert_page(vma, start, page);
4738 				if (unlikely(err))
4739 					goto out;
4740 				start += PAGE_SIZE;
4741 				kaddr += PAGE_SIZE;
4742 			}
4743 		}
4744 	}
4745 
4746 	atomic_long_inc(&po->mapped);
4747 	vma->vm_ops = &packet_mmap_ops;
4748 	err = 0;
4749 
4750 out:
4751 	mutex_unlock(&po->pg_vec_lock);
4752 	return err;
4753 }
4754 
4755 static const struct proto_ops packet_ops_spkt = {
4756 	.family =	PF_PACKET,
4757 	.owner =	THIS_MODULE,
4758 	.release =	packet_release,
4759 	.bind =		packet_bind_spkt,
4760 	.connect =	sock_no_connect,
4761 	.socketpair =	sock_no_socketpair,
4762 	.accept =	sock_no_accept,
4763 	.getname =	packet_getname_spkt,
4764 	.poll =		datagram_poll,
4765 	.ioctl =	packet_ioctl,
4766 	.gettstamp =	sock_gettstamp,
4767 	.listen =	sock_no_listen,
4768 	.shutdown =	sock_no_shutdown,
4769 	.sendmsg =	packet_sendmsg_spkt,
4770 	.recvmsg =	packet_recvmsg,
4771 	.mmap =		sock_no_mmap,
4772 };
4773 
4774 static const struct proto_ops packet_ops = {
4775 	.family =	PF_PACKET,
4776 	.owner =	THIS_MODULE,
4777 	.release =	packet_release,
4778 	.bind =		packet_bind,
4779 	.connect =	sock_no_connect,
4780 	.socketpair =	sock_no_socketpair,
4781 	.accept =	sock_no_accept,
4782 	.getname =	packet_getname,
4783 	.poll =		packet_poll,
4784 	.ioctl =	packet_ioctl,
4785 	.gettstamp =	sock_gettstamp,
4786 	.listen =	sock_no_listen,
4787 	.shutdown =	sock_no_shutdown,
4788 	.setsockopt =	packet_setsockopt,
4789 	.getsockopt_iter =	packet_getsockopt,
4790 	.sendmsg =	packet_sendmsg,
4791 	.recvmsg =	packet_recvmsg,
4792 	.mmap =		packet_mmap,
4793 };
4794 
4795 static const struct net_proto_family packet_family_ops = {
4796 	.family =	PF_PACKET,
4797 	.create =	packet_create,
4798 	.owner	=	THIS_MODULE,
4799 };
4800 
4801 static struct notifier_block packet_netdev_notifier = {
4802 	.notifier_call =	packet_notifier,
4803 };
4804 
4805 #ifdef CONFIG_PROC_FS
4806 
4807 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4808 	__acquires(RCU)
4809 {
4810 	struct net *net = seq_file_net(seq);
4811 
4812 	rcu_read_lock();
4813 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4814 }
4815 
4816 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4817 {
4818 	struct net *net = seq_file_net(seq);
4819 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4820 }
4821 
4822 static void packet_seq_stop(struct seq_file *seq, void *v)
4823 	__releases(RCU)
4824 {
4825 	rcu_read_unlock();
4826 }
4827 
4828 static int packet_seq_show(struct seq_file *seq, void *v)
4829 {
4830 	if (v == SEQ_START_TOKEN)
4831 		seq_printf(seq,
4832 			   "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4833 			   IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4834 	else {
4835 		struct sock *s = sk_entry(v);
4836 		const struct packet_sock *po = pkt_sk(s);
4837 
4838 		seq_printf(seq,
4839 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6llu\n",
4840 			   s,
4841 			   refcount_read(&s->sk_refcnt),
4842 			   s->sk_type,
4843 			   ntohs(READ_ONCE(po->num)),
4844 			   READ_ONCE(po->ifindex),
4845 			   packet_sock_flag(po, PACKET_SOCK_RUNNING),
4846 			   atomic_read(&s->sk_rmem_alloc),
4847 			   from_kuid_munged(seq_user_ns(seq), sk_uid(s)),
4848 			   sock_i_ino(s));
4849 	}
4850 
4851 	return 0;
4852 }
4853 
4854 static const struct seq_operations packet_seq_ops = {
4855 	.start	= packet_seq_start,
4856 	.next	= packet_seq_next,
4857 	.stop	= packet_seq_stop,
4858 	.show	= packet_seq_show,
4859 };
4860 #endif
4861 
4862 static int __net_init packet_net_init(struct net *net)
4863 {
4864 	mutex_init(&net->packet.sklist_lock);
4865 	INIT_HLIST_HEAD(&net->packet.sklist);
4866 
4867 #ifdef CONFIG_PROC_FS
4868 	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4869 			sizeof(struct seq_net_private)))
4870 		return -ENOMEM;
4871 #endif /* CONFIG_PROC_FS */
4872 
4873 	return 0;
4874 }
4875 
4876 static void __net_exit packet_net_exit(struct net *net)
4877 {
4878 	remove_proc_entry("packet", net->proc_net);
4879 	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4880 }
4881 
4882 static struct pernet_operations packet_net_ops = {
4883 	.init = packet_net_init,
4884 	.exit = packet_net_exit,
4885 };
4886 
4887 
4888 static void __exit packet_exit(void)
4889 {
4890 	sock_unregister(PF_PACKET);
4891 	proto_unregister(&packet_proto);
4892 	unregister_netdevice_notifier(&packet_netdev_notifier);
4893 	unregister_pernet_subsys(&packet_net_ops);
4894 }
4895 
4896 static int __init packet_init(void)
4897 {
4898 	int rc;
4899 
4900 	rc = register_pernet_subsys(&packet_net_ops);
4901 	if (rc)
4902 		goto out;
4903 	rc = register_netdevice_notifier(&packet_netdev_notifier);
4904 	if (rc)
4905 		goto out_pernet;
4906 	rc = proto_register(&packet_proto, 0);
4907 	if (rc)
4908 		goto out_notifier;
4909 	rc = sock_register(&packet_family_ops);
4910 	if (rc)
4911 		goto out_proto;
4912 
4913 	return 0;
4914 
4915 out_proto:
4916 	proto_unregister(&packet_proto);
4917 out_notifier:
4918 	unregister_netdevice_notifier(&packet_netdev_notifier);
4919 out_pernet:
4920 	unregister_pernet_subsys(&packet_net_ops);
4921 out:
4922 	return rc;
4923 }
4924 
4925 module_init(packet_init);
4926 module_exit(packet_exit);
4927 MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4928 MODULE_LICENSE("GPL");
4929 MODULE_ALIAS_NETPROTO(PF_PACKET);
4930