xref: /linux/net/packet/af_packet.c (revision d53b8e36925256097a08d7cb749198d85cbf9b2b)
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/filter.h>
53 #include <linux/types.h>
54 #include <linux/mm.h>
55 #include <linux/capability.h>
56 #include <linux/fcntl.h>
57 #include <linux/socket.h>
58 #include <linux/in.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/if_packet.h>
62 #include <linux/wireless.h>
63 #include <linux/kernel.h>
64 #include <linux/kmod.h>
65 #include <linux/slab.h>
66 #include <linux/vmalloc.h>
67 #include <net/net_namespace.h>
68 #include <net/ip.h>
69 #include <net/protocol.h>
70 #include <linux/skbuff.h>
71 #include <net/sock.h>
72 #include <linux/errno.h>
73 #include <linux/timer.h>
74 #include <linux/uaccess.h>
75 #include <asm/ioctls.h>
76 #include <asm/page.h>
77 #include <asm/cacheflush.h>
78 #include <asm/io.h>
79 #include <linux/proc_fs.h>
80 #include <linux/seq_file.h>
81 #include <linux/poll.h>
82 #include <linux/module.h>
83 #include <linux/init.h>
84 #include <linux/mutex.h>
85 #include <linux/if_vlan.h>
86 #include <linux/virtio_net.h>
87 #include <linux/errqueue.h>
88 #include <linux/net_tstamp.h>
89 #include <linux/percpu.h>
90 #ifdef CONFIG_INET
91 #include <net/inet_common.h>
92 #endif
93 #include <linux/bpf.h>
94 #include <net/compat.h>
95 #include <linux/netfilter_netdev.h>
96 
97 #include "internal.h"
98 
99 /*
100    Assumptions:
101    - If the device has no dev->header_ops->create, there is no LL header
102      visible above the device. In this case, its hard_header_len should be 0.
103      The device may prepend its own header internally. In this case, its
104      needed_headroom should be set to the space needed for it to add its
105      internal header.
106      For example, a WiFi driver pretending to be an Ethernet driver should
107      set its hard_header_len to be the Ethernet header length, and set its
108      needed_headroom to be (the real WiFi header length - the fake Ethernet
109      header length).
110    - packet socket receives packets with pulled ll header,
111      so that SOCK_RAW should push it back.
112 
113 On receive:
114 -----------
115 
116 Incoming, dev_has_header(dev) == true
117    mac_header -> ll header
118    data       -> data
119 
120 Outgoing, dev_has_header(dev) == true
121    mac_header -> ll header
122    data       -> ll header
123 
124 Incoming, dev_has_header(dev) == false
125    mac_header -> data
126      However drivers often make it point to the ll header.
127      This is incorrect because the ll header should be invisible to us.
128    data       -> data
129 
130 Outgoing, dev_has_header(dev) == false
131    mac_header -> data. ll header is invisible to us.
132    data       -> data
133 
134 Resume
135   If dev_has_header(dev) == false we are unable to restore the ll header,
136     because it is invisible to us.
137 
138 
139 On transmit:
140 ------------
141 
142 dev_has_header(dev) == true
143    mac_header -> ll header
144    data       -> ll header
145 
146 dev_has_header(dev) == false (ll header is invisible to us)
147    mac_header -> data
148    data       -> data
149 
150    We should set network_header on output to the correct position,
151    packet classifier depends on it.
152  */
153 
154 /* Private packet socket structures. */
155 
156 /* identical to struct packet_mreq except it has
157  * a longer address field.
158  */
159 struct packet_mreq_max {
160 	int		mr_ifindex;
161 	unsigned short	mr_type;
162 	unsigned short	mr_alen;
163 	unsigned char	mr_address[MAX_ADDR_LEN];
164 };
165 
166 union tpacket_uhdr {
167 	struct tpacket_hdr  *h1;
168 	struct tpacket2_hdr *h2;
169 	struct tpacket3_hdr *h3;
170 	void *raw;
171 };
172 
173 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
174 		int closing, int tx_ring);
175 
176 #define V3_ALIGNMENT	(8)
177 
178 #define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
179 
180 #define BLK_PLUS_PRIV(sz_of_priv) \
181 	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
182 
183 #define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
184 #define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
185 #define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
186 #define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
187 #define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
188 #define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
189 
190 struct packet_sock;
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 		       struct packet_type *pt, struct net_device *orig_dev);
193 
194 static void *packet_previous_frame(struct packet_sock *po,
195 		struct packet_ring_buffer *rb,
196 		int status);
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 			struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 		struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 		struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
209 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
210 		struct tpacket3_hdr *);
211 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
212 		struct tpacket3_hdr *);
213 static void packet_flush_mclist(struct sock *sk);
214 static u16 packet_pick_tx_queue(struct sk_buff *skb);
215 
216 struct packet_skb_cb {
217 	union {
218 		struct sockaddr_pkt pkt;
219 		union {
220 			/* Trick: alias skb original length with
221 			 * ll.sll_family and ll.protocol in order
222 			 * to save room.
223 			 */
224 			unsigned int origlen;
225 			struct sockaddr_ll ll;
226 		};
227 	} sa;
228 };
229 
230 #define vio_le() virtio_legacy_is_little_endian()
231 
232 #define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
233 
234 #define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid)	\
236 	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
238 	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 	((x)->kactive_blk_num+1) : 0)
242 
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
245 
246 #ifdef CONFIG_NETFILTER_EGRESS
247 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
248 {
249 	struct sk_buff *next, *head = NULL, *tail;
250 	int rc;
251 
252 	rcu_read_lock();
253 	for (; skb != NULL; skb = next) {
254 		next = skb->next;
255 		skb_mark_not_on_list(skb);
256 
257 		if (!nf_hook_egress(skb, &rc, skb->dev))
258 			continue;
259 
260 		if (!head)
261 			head = skb;
262 		else
263 			tail->next = skb;
264 
265 		tail = skb;
266 	}
267 	rcu_read_unlock();
268 
269 	return head;
270 }
271 #endif
272 
273 static int packet_xmit(const struct packet_sock *po, struct sk_buff *skb)
274 {
275 	if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS))
276 		return dev_queue_xmit(skb);
277 
278 #ifdef CONFIG_NETFILTER_EGRESS
279 	if (nf_hook_egress_active()) {
280 		skb = nf_hook_direct_egress(skb);
281 		if (!skb)
282 			return NET_XMIT_DROP;
283 	}
284 #endif
285 	return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
286 }
287 
288 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
289 {
290 	struct net_device *dev;
291 
292 	rcu_read_lock();
293 	dev = rcu_dereference(po->cached_dev);
294 	dev_hold(dev);
295 	rcu_read_unlock();
296 
297 	return dev;
298 }
299 
300 static void packet_cached_dev_assign(struct packet_sock *po,
301 				     struct net_device *dev)
302 {
303 	rcu_assign_pointer(po->cached_dev, dev);
304 }
305 
306 static void packet_cached_dev_reset(struct packet_sock *po)
307 {
308 	RCU_INIT_POINTER(po->cached_dev, NULL);
309 }
310 
311 static u16 packet_pick_tx_queue(struct sk_buff *skb)
312 {
313 	struct net_device *dev = skb->dev;
314 	const struct net_device_ops *ops = dev->netdev_ops;
315 	int cpu = raw_smp_processor_id();
316 	u16 queue_index;
317 
318 #ifdef CONFIG_XPS
319 	skb->sender_cpu = cpu + 1;
320 #endif
321 	skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
322 	if (ops->ndo_select_queue) {
323 		queue_index = ops->ndo_select_queue(dev, skb, NULL);
324 		queue_index = netdev_cap_txqueue(dev, queue_index);
325 	} else {
326 		queue_index = netdev_pick_tx(dev, skb, NULL);
327 	}
328 
329 	return queue_index;
330 }
331 
332 /* __register_prot_hook must be invoked through register_prot_hook
333  * or from a context in which asynchronous accesses to the packet
334  * socket is not possible (packet_create()).
335  */
336 static void __register_prot_hook(struct sock *sk)
337 {
338 	struct packet_sock *po = pkt_sk(sk);
339 
340 	if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
341 		if (po->fanout)
342 			__fanout_link(sk, po);
343 		else
344 			dev_add_pack(&po->prot_hook);
345 
346 		sock_hold(sk);
347 		packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1);
348 	}
349 }
350 
351 static void register_prot_hook(struct sock *sk)
352 {
353 	lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
354 	__register_prot_hook(sk);
355 }
356 
357 /* If the sync parameter is true, we will temporarily drop
358  * the po->bind_lock and do a synchronize_net to make sure no
359  * asynchronous packet processing paths still refer to the elements
360  * of po->prot_hook.  If the sync parameter is false, it is the
361  * callers responsibility to take care of this.
362  */
363 static void __unregister_prot_hook(struct sock *sk, bool sync)
364 {
365 	struct packet_sock *po = pkt_sk(sk);
366 
367 	lockdep_assert_held_once(&po->bind_lock);
368 
369 	packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0);
370 
371 	if (po->fanout)
372 		__fanout_unlink(sk, po);
373 	else
374 		__dev_remove_pack(&po->prot_hook);
375 
376 	__sock_put(sk);
377 
378 	if (sync) {
379 		spin_unlock(&po->bind_lock);
380 		synchronize_net();
381 		spin_lock(&po->bind_lock);
382 	}
383 }
384 
385 static void unregister_prot_hook(struct sock *sk, bool sync)
386 {
387 	struct packet_sock *po = pkt_sk(sk);
388 
389 	if (packet_sock_flag(po, PACKET_SOCK_RUNNING))
390 		__unregister_prot_hook(sk, sync);
391 }
392 
393 static inline struct page * __pure pgv_to_page(void *addr)
394 {
395 	if (is_vmalloc_addr(addr))
396 		return vmalloc_to_page(addr);
397 	return virt_to_page(addr);
398 }
399 
400 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
401 {
402 	union tpacket_uhdr h;
403 
404 	/* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */
405 
406 	h.raw = frame;
407 	switch (po->tp_version) {
408 	case TPACKET_V1:
409 		WRITE_ONCE(h.h1->tp_status, status);
410 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
411 		break;
412 	case TPACKET_V2:
413 		WRITE_ONCE(h.h2->tp_status, status);
414 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
415 		break;
416 	case TPACKET_V3:
417 		WRITE_ONCE(h.h3->tp_status, status);
418 		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
419 		break;
420 	default:
421 		WARN(1, "TPACKET version not supported.\n");
422 		BUG();
423 	}
424 
425 	smp_wmb();
426 }
427 
428 static int __packet_get_status(const struct packet_sock *po, void *frame)
429 {
430 	union tpacket_uhdr h;
431 
432 	smp_rmb();
433 
434 	/* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */
435 
436 	h.raw = frame;
437 	switch (po->tp_version) {
438 	case TPACKET_V1:
439 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
440 		return READ_ONCE(h.h1->tp_status);
441 	case TPACKET_V2:
442 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
443 		return READ_ONCE(h.h2->tp_status);
444 	case TPACKET_V3:
445 		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
446 		return READ_ONCE(h.h3->tp_status);
447 	default:
448 		WARN(1, "TPACKET version not supported.\n");
449 		BUG();
450 		return 0;
451 	}
452 }
453 
454 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
455 				   unsigned int flags)
456 {
457 	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
458 
459 	if (shhwtstamps &&
460 	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
461 	    ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
462 		return TP_STATUS_TS_RAW_HARDWARE;
463 
464 	if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
465 	    ktime_to_timespec64_cond(skb_tstamp(skb), ts))
466 		return TP_STATUS_TS_SOFTWARE;
467 
468 	return 0;
469 }
470 
471 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
472 				    struct sk_buff *skb)
473 {
474 	union tpacket_uhdr h;
475 	struct timespec64 ts;
476 	__u32 ts_status;
477 
478 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp))))
479 		return 0;
480 
481 	h.raw = frame;
482 	/*
483 	 * versions 1 through 3 overflow the timestamps in y2106, since they
484 	 * all store the seconds in a 32-bit unsigned integer.
485 	 * If we create a version 4, that should have a 64-bit timestamp,
486 	 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
487 	 * nanoseconds.
488 	 */
489 	switch (po->tp_version) {
490 	case TPACKET_V1:
491 		h.h1->tp_sec = ts.tv_sec;
492 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
493 		break;
494 	case TPACKET_V2:
495 		h.h2->tp_sec = ts.tv_sec;
496 		h.h2->tp_nsec = ts.tv_nsec;
497 		break;
498 	case TPACKET_V3:
499 		h.h3->tp_sec = ts.tv_sec;
500 		h.h3->tp_nsec = ts.tv_nsec;
501 		break;
502 	default:
503 		WARN(1, "TPACKET version not supported.\n");
504 		BUG();
505 	}
506 
507 	/* one flush is safe, as both fields always lie on the same cacheline */
508 	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
509 	smp_wmb();
510 
511 	return ts_status;
512 }
513 
514 static void *packet_lookup_frame(const struct packet_sock *po,
515 				 const struct packet_ring_buffer *rb,
516 				 unsigned int position,
517 				 int status)
518 {
519 	unsigned int pg_vec_pos, frame_offset;
520 	union tpacket_uhdr h;
521 
522 	pg_vec_pos = position / rb->frames_per_block;
523 	frame_offset = position % rb->frames_per_block;
524 
525 	h.raw = rb->pg_vec[pg_vec_pos].buffer +
526 		(frame_offset * rb->frame_size);
527 
528 	if (status != __packet_get_status(po, h.raw))
529 		return NULL;
530 
531 	return h.raw;
532 }
533 
534 static void *packet_current_frame(struct packet_sock *po,
535 		struct packet_ring_buffer *rb,
536 		int status)
537 {
538 	return packet_lookup_frame(po, rb, rb->head, status);
539 }
540 
541 static u16 vlan_get_tci(struct sk_buff *skb, struct net_device *dev)
542 {
543 	u8 *skb_orig_data = skb->data;
544 	int skb_orig_len = skb->len;
545 	struct vlan_hdr vhdr, *vh;
546 	unsigned int header_len;
547 
548 	if (!dev)
549 		return 0;
550 
551 	/* In the SOCK_DGRAM scenario, skb data starts at the network
552 	 * protocol, which is after the VLAN headers. The outer VLAN
553 	 * header is at the hard_header_len offset in non-variable
554 	 * length link layer headers. If it's a VLAN device, the
555 	 * min_header_len should be used to exclude the VLAN header
556 	 * size.
557 	 */
558 	if (dev->min_header_len == dev->hard_header_len)
559 		header_len = dev->hard_header_len;
560 	else if (is_vlan_dev(dev))
561 		header_len = dev->min_header_len;
562 	else
563 		return 0;
564 
565 	skb_push(skb, skb->data - skb_mac_header(skb));
566 	vh = skb_header_pointer(skb, header_len, sizeof(vhdr), &vhdr);
567 	if (skb_orig_data != skb->data) {
568 		skb->data = skb_orig_data;
569 		skb->len = skb_orig_len;
570 	}
571 	if (unlikely(!vh))
572 		return 0;
573 
574 	return ntohs(vh->h_vlan_TCI);
575 }
576 
577 static __be16 vlan_get_protocol_dgram(struct sk_buff *skb)
578 {
579 	__be16 proto = skb->protocol;
580 
581 	if (unlikely(eth_type_vlan(proto))) {
582 		u8 *skb_orig_data = skb->data;
583 		int skb_orig_len = skb->len;
584 
585 		skb_push(skb, skb->data - skb_mac_header(skb));
586 		proto = __vlan_get_protocol(skb, proto, NULL);
587 		if (skb_orig_data != skb->data) {
588 			skb->data = skb_orig_data;
589 			skb->len = skb_orig_len;
590 		}
591 	}
592 
593 	return proto;
594 }
595 
596 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
597 {
598 	del_timer_sync(&pkc->retire_blk_timer);
599 }
600 
601 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
602 		struct sk_buff_head *rb_queue)
603 {
604 	struct tpacket_kbdq_core *pkc;
605 
606 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
607 
608 	spin_lock_bh(&rb_queue->lock);
609 	pkc->delete_blk_timer = 1;
610 	spin_unlock_bh(&rb_queue->lock);
611 
612 	prb_del_retire_blk_timer(pkc);
613 }
614 
615 static void prb_setup_retire_blk_timer(struct packet_sock *po)
616 {
617 	struct tpacket_kbdq_core *pkc;
618 
619 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
620 	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
621 		    0);
622 	pkc->retire_blk_timer.expires = jiffies;
623 }
624 
625 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
626 				int blk_size_in_bytes)
627 {
628 	struct net_device *dev;
629 	unsigned int mbits, div;
630 	struct ethtool_link_ksettings ecmd;
631 	int err;
632 
633 	rtnl_lock();
634 	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
635 	if (unlikely(!dev)) {
636 		rtnl_unlock();
637 		return DEFAULT_PRB_RETIRE_TOV;
638 	}
639 	err = __ethtool_get_link_ksettings(dev, &ecmd);
640 	rtnl_unlock();
641 	if (err)
642 		return DEFAULT_PRB_RETIRE_TOV;
643 
644 	/* If the link speed is so slow you don't really
645 	 * need to worry about perf anyways
646 	 */
647 	if (ecmd.base.speed < SPEED_1000 ||
648 	    ecmd.base.speed == SPEED_UNKNOWN)
649 		return DEFAULT_PRB_RETIRE_TOV;
650 
651 	div = ecmd.base.speed / 1000;
652 	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
653 
654 	if (div)
655 		mbits /= div;
656 
657 	if (div)
658 		return mbits + 1;
659 	return mbits;
660 }
661 
662 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
663 			union tpacket_req_u *req_u)
664 {
665 	p1->feature_req_word = req_u->req3.tp_feature_req_word;
666 }
667 
668 static void init_prb_bdqc(struct packet_sock *po,
669 			struct packet_ring_buffer *rb,
670 			struct pgv *pg_vec,
671 			union tpacket_req_u *req_u)
672 {
673 	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
674 	struct tpacket_block_desc *pbd;
675 
676 	memset(p1, 0x0, sizeof(*p1));
677 
678 	p1->knxt_seq_num = 1;
679 	p1->pkbdq = pg_vec;
680 	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
681 	p1->pkblk_start	= pg_vec[0].buffer;
682 	p1->kblk_size = req_u->req3.tp_block_size;
683 	p1->knum_blocks	= req_u->req3.tp_block_nr;
684 	p1->hdrlen = po->tp_hdrlen;
685 	p1->version = po->tp_version;
686 	p1->last_kactive_blk_num = 0;
687 	po->stats.stats3.tp_freeze_q_cnt = 0;
688 	if (req_u->req3.tp_retire_blk_tov)
689 		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
690 	else
691 		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
692 						req_u->req3.tp_block_size);
693 	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
694 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
695 	rwlock_init(&p1->blk_fill_in_prog_lock);
696 
697 	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
698 	prb_init_ft_ops(p1, req_u);
699 	prb_setup_retire_blk_timer(po);
700 	prb_open_block(p1, pbd);
701 }
702 
703 /*  Do NOT update the last_blk_num first.
704  *  Assumes sk_buff_head lock is held.
705  */
706 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
707 {
708 	mod_timer(&pkc->retire_blk_timer,
709 			jiffies + pkc->tov_in_jiffies);
710 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
711 }
712 
713 /*
714  * Timer logic:
715  * 1) We refresh the timer only when we open a block.
716  *    By doing this we don't waste cycles refreshing the timer
717  *	  on packet-by-packet basis.
718  *
719  * With a 1MB block-size, on a 1Gbps line, it will take
720  * i) ~8 ms to fill a block + ii) memcpy etc.
721  * In this cut we are not accounting for the memcpy time.
722  *
723  * So, if the user sets the 'tmo' to 10ms then the timer
724  * will never fire while the block is still getting filled
725  * (which is what we want). However, the user could choose
726  * to close a block early and that's fine.
727  *
728  * But when the timer does fire, we check whether or not to refresh it.
729  * Since the tmo granularity is in msecs, it is not too expensive
730  * to refresh the timer, lets say every '8' msecs.
731  * Either the user can set the 'tmo' or we can derive it based on
732  * a) line-speed and b) block-size.
733  * prb_calc_retire_blk_tmo() calculates the tmo.
734  *
735  */
736 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
737 {
738 	struct packet_sock *po =
739 		from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
740 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
741 	unsigned int frozen;
742 	struct tpacket_block_desc *pbd;
743 
744 	spin_lock(&po->sk.sk_receive_queue.lock);
745 
746 	frozen = prb_queue_frozen(pkc);
747 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
748 
749 	if (unlikely(pkc->delete_blk_timer))
750 		goto out;
751 
752 	/* We only need to plug the race when the block is partially filled.
753 	 * tpacket_rcv:
754 	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
755 	 *		copy_bits() is in progress ...
756 	 *		timer fires on other cpu:
757 	 *		we can't retire the current block because copy_bits
758 	 *		is in progress.
759 	 *
760 	 */
761 	if (BLOCK_NUM_PKTS(pbd)) {
762 		/* Waiting for skb_copy_bits to finish... */
763 		write_lock(&pkc->blk_fill_in_prog_lock);
764 		write_unlock(&pkc->blk_fill_in_prog_lock);
765 	}
766 
767 	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
768 		if (!frozen) {
769 			if (!BLOCK_NUM_PKTS(pbd)) {
770 				/* An empty block. Just refresh the timer. */
771 				goto refresh_timer;
772 			}
773 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
774 			if (!prb_dispatch_next_block(pkc, po))
775 				goto refresh_timer;
776 			else
777 				goto out;
778 		} else {
779 			/* Case 1. Queue was frozen because user-space was
780 			 *	   lagging behind.
781 			 */
782 			if (prb_curr_blk_in_use(pbd)) {
783 				/*
784 				 * Ok, user-space is still behind.
785 				 * So just refresh the timer.
786 				 */
787 				goto refresh_timer;
788 			} else {
789 			       /* Case 2. queue was frozen,user-space caught up,
790 				* now the link went idle && the timer fired.
791 				* We don't have a block to close.So we open this
792 				* block and restart the timer.
793 				* opening a block thaws the queue,restarts timer
794 				* Thawing/timer-refresh is a side effect.
795 				*/
796 				prb_open_block(pkc, pbd);
797 				goto out;
798 			}
799 		}
800 	}
801 
802 refresh_timer:
803 	_prb_refresh_rx_retire_blk_timer(pkc);
804 
805 out:
806 	spin_unlock(&po->sk.sk_receive_queue.lock);
807 }
808 
809 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
810 		struct tpacket_block_desc *pbd1, __u32 status)
811 {
812 	/* Flush everything minus the block header */
813 
814 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
815 	u8 *start, *end;
816 
817 	start = (u8 *)pbd1;
818 
819 	/* Skip the block header(we know header WILL fit in 4K) */
820 	start += PAGE_SIZE;
821 
822 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
823 	for (; start < end; start += PAGE_SIZE)
824 		flush_dcache_page(pgv_to_page(start));
825 
826 	smp_wmb();
827 #endif
828 
829 	/* Now update the block status. */
830 
831 	BLOCK_STATUS(pbd1) = status;
832 
833 	/* Flush the block header */
834 
835 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
836 	start = (u8 *)pbd1;
837 	flush_dcache_page(pgv_to_page(start));
838 
839 	smp_wmb();
840 #endif
841 }
842 
843 /*
844  * Side effect:
845  *
846  * 1) flush the block
847  * 2) Increment active_blk_num
848  *
849  * Note:We DONT refresh the timer on purpose.
850  *	Because almost always the next block will be opened.
851  */
852 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
853 		struct tpacket_block_desc *pbd1,
854 		struct packet_sock *po, unsigned int stat)
855 {
856 	__u32 status = TP_STATUS_USER | stat;
857 
858 	struct tpacket3_hdr *last_pkt;
859 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
860 	struct sock *sk = &po->sk;
861 
862 	if (atomic_read(&po->tp_drops))
863 		status |= TP_STATUS_LOSING;
864 
865 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
866 	last_pkt->tp_next_offset = 0;
867 
868 	/* Get the ts of the last pkt */
869 	if (BLOCK_NUM_PKTS(pbd1)) {
870 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
871 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
872 	} else {
873 		/* Ok, we tmo'd - so get the current time.
874 		 *
875 		 * It shouldn't really happen as we don't close empty
876 		 * blocks. See prb_retire_rx_blk_timer_expired().
877 		 */
878 		struct timespec64 ts;
879 		ktime_get_real_ts64(&ts);
880 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
881 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
882 	}
883 
884 	smp_wmb();
885 
886 	/* Flush the block */
887 	prb_flush_block(pkc1, pbd1, status);
888 
889 	sk->sk_data_ready(sk);
890 
891 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
892 }
893 
894 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
895 {
896 	pkc->reset_pending_on_curr_blk = 0;
897 }
898 
899 /*
900  * Side effect of opening a block:
901  *
902  * 1) prb_queue is thawed.
903  * 2) retire_blk_timer is refreshed.
904  *
905  */
906 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
907 	struct tpacket_block_desc *pbd1)
908 {
909 	struct timespec64 ts;
910 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
911 
912 	smp_rmb();
913 
914 	/* We could have just memset this but we will lose the
915 	 * flexibility of making the priv area sticky
916 	 */
917 
918 	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
919 	BLOCK_NUM_PKTS(pbd1) = 0;
920 	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
921 
922 	ktime_get_real_ts64(&ts);
923 
924 	h1->ts_first_pkt.ts_sec = ts.tv_sec;
925 	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
926 
927 	pkc1->pkblk_start = (char *)pbd1;
928 	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
929 
930 	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
931 	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
932 
933 	pbd1->version = pkc1->version;
934 	pkc1->prev = pkc1->nxt_offset;
935 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
936 
937 	prb_thaw_queue(pkc1);
938 	_prb_refresh_rx_retire_blk_timer(pkc1);
939 
940 	smp_wmb();
941 }
942 
943 /*
944  * Queue freeze logic:
945  * 1) Assume tp_block_nr = 8 blocks.
946  * 2) At time 't0', user opens Rx ring.
947  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
948  * 4) user-space is either sleeping or processing block '0'.
949  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
950  *    it will close block-7,loop around and try to fill block '0'.
951  *    call-flow:
952  *    __packet_lookup_frame_in_block
953  *      prb_retire_current_block()
954  *      prb_dispatch_next_block()
955  *        |->(BLOCK_STATUS == USER) evaluates to true
956  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
957  * 6) Now there are two cases:
958  *    6.1) Link goes idle right after the queue is frozen.
959  *         But remember, the last open_block() refreshed the timer.
960  *         When this timer expires,it will refresh itself so that we can
961  *         re-open block-0 in near future.
962  *    6.2) Link is busy and keeps on receiving packets. This is a simple
963  *         case and __packet_lookup_frame_in_block will check if block-0
964  *         is free and can now be re-used.
965  */
966 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
967 				  struct packet_sock *po)
968 {
969 	pkc->reset_pending_on_curr_blk = 1;
970 	po->stats.stats3.tp_freeze_q_cnt++;
971 }
972 
973 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
974 
975 /*
976  * If the next block is free then we will dispatch it
977  * and return a good offset.
978  * Else, we will freeze the queue.
979  * So, caller must check the return value.
980  */
981 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
982 		struct packet_sock *po)
983 {
984 	struct tpacket_block_desc *pbd;
985 
986 	smp_rmb();
987 
988 	/* 1. Get current block num */
989 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
990 
991 	/* 2. If this block is currently in_use then freeze the queue */
992 	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
993 		prb_freeze_queue(pkc, po);
994 		return NULL;
995 	}
996 
997 	/*
998 	 * 3.
999 	 * open this block and return the offset where the first packet
1000 	 * needs to get stored.
1001 	 */
1002 	prb_open_block(pkc, pbd);
1003 	return (void *)pkc->nxt_offset;
1004 }
1005 
1006 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
1007 		struct packet_sock *po, unsigned int status)
1008 {
1009 	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1010 
1011 	/* retire/close the current block */
1012 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
1013 		/*
1014 		 * Plug the case where copy_bits() is in progress on
1015 		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
1016 		 * have space to copy the pkt in the current block and
1017 		 * called prb_retire_current_block()
1018 		 *
1019 		 * We don't need to worry about the TMO case because
1020 		 * the timer-handler already handled this case.
1021 		 */
1022 		if (!(status & TP_STATUS_BLK_TMO)) {
1023 			/* Waiting for skb_copy_bits to finish... */
1024 			write_lock(&pkc->blk_fill_in_prog_lock);
1025 			write_unlock(&pkc->blk_fill_in_prog_lock);
1026 		}
1027 		prb_close_block(pkc, pbd, po, status);
1028 		return;
1029 	}
1030 }
1031 
1032 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
1033 {
1034 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
1035 }
1036 
1037 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
1038 {
1039 	return pkc->reset_pending_on_curr_blk;
1040 }
1041 
1042 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
1043 	__releases(&pkc->blk_fill_in_prog_lock)
1044 {
1045 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1046 
1047 	read_unlock(&pkc->blk_fill_in_prog_lock);
1048 }
1049 
1050 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
1051 			struct tpacket3_hdr *ppd)
1052 {
1053 	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
1054 }
1055 
1056 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1057 			struct tpacket3_hdr *ppd)
1058 {
1059 	ppd->hv1.tp_rxhash = 0;
1060 }
1061 
1062 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1063 			struct tpacket3_hdr *ppd)
1064 {
1065 	struct packet_sock *po = container_of(pkc, struct packet_sock, rx_ring.prb_bdqc);
1066 
1067 	if (skb_vlan_tag_present(pkc->skb)) {
1068 		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1069 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1070 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1071 	} else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) {
1072 		ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev);
1073 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol);
1074 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1075 	} else {
1076 		ppd->hv1.tp_vlan_tci = 0;
1077 		ppd->hv1.tp_vlan_tpid = 0;
1078 		ppd->tp_status = TP_STATUS_AVAILABLE;
1079 	}
1080 }
1081 
1082 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1083 			struct tpacket3_hdr *ppd)
1084 {
1085 	ppd->hv1.tp_padding = 0;
1086 	prb_fill_vlan_info(pkc, ppd);
1087 
1088 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1089 		prb_fill_rxhash(pkc, ppd);
1090 	else
1091 		prb_clear_rxhash(pkc, ppd);
1092 }
1093 
1094 static void prb_fill_curr_block(char *curr,
1095 				struct tpacket_kbdq_core *pkc,
1096 				struct tpacket_block_desc *pbd,
1097 				unsigned int len)
1098 	__acquires(&pkc->blk_fill_in_prog_lock)
1099 {
1100 	struct tpacket3_hdr *ppd;
1101 
1102 	ppd  = (struct tpacket3_hdr *)curr;
1103 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1104 	pkc->prev = curr;
1105 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1106 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1107 	BLOCK_NUM_PKTS(pbd) += 1;
1108 	read_lock(&pkc->blk_fill_in_prog_lock);
1109 	prb_run_all_ft_ops(pkc, ppd);
1110 }
1111 
1112 /* Assumes caller has the sk->rx_queue.lock */
1113 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1114 					    struct sk_buff *skb,
1115 					    unsigned int len
1116 					    )
1117 {
1118 	struct tpacket_kbdq_core *pkc;
1119 	struct tpacket_block_desc *pbd;
1120 	char *curr, *end;
1121 
1122 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1123 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1124 
1125 	/* Queue is frozen when user space is lagging behind */
1126 	if (prb_queue_frozen(pkc)) {
1127 		/*
1128 		 * Check if that last block which caused the queue to freeze,
1129 		 * is still in_use by user-space.
1130 		 */
1131 		if (prb_curr_blk_in_use(pbd)) {
1132 			/* Can't record this packet */
1133 			return NULL;
1134 		} else {
1135 			/*
1136 			 * Ok, the block was released by user-space.
1137 			 * Now let's open that block.
1138 			 * opening a block also thaws the queue.
1139 			 * Thawing is a side effect.
1140 			 */
1141 			prb_open_block(pkc, pbd);
1142 		}
1143 	}
1144 
1145 	smp_mb();
1146 	curr = pkc->nxt_offset;
1147 	pkc->skb = skb;
1148 	end = (char *)pbd + pkc->kblk_size;
1149 
1150 	/* first try the current block */
1151 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1152 		prb_fill_curr_block(curr, pkc, pbd, len);
1153 		return (void *)curr;
1154 	}
1155 
1156 	/* Ok, close the current block */
1157 	prb_retire_current_block(pkc, po, 0);
1158 
1159 	/* Now, try to dispatch the next block */
1160 	curr = (char *)prb_dispatch_next_block(pkc, po);
1161 	if (curr) {
1162 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1163 		prb_fill_curr_block(curr, pkc, pbd, len);
1164 		return (void *)curr;
1165 	}
1166 
1167 	/*
1168 	 * No free blocks are available.user_space hasn't caught up yet.
1169 	 * Queue was just frozen and now this packet will get dropped.
1170 	 */
1171 	return NULL;
1172 }
1173 
1174 static void *packet_current_rx_frame(struct packet_sock *po,
1175 					    struct sk_buff *skb,
1176 					    int status, unsigned int len)
1177 {
1178 	char *curr = NULL;
1179 	switch (po->tp_version) {
1180 	case TPACKET_V1:
1181 	case TPACKET_V2:
1182 		curr = packet_lookup_frame(po, &po->rx_ring,
1183 					po->rx_ring.head, status);
1184 		return curr;
1185 	case TPACKET_V3:
1186 		return __packet_lookup_frame_in_block(po, skb, len);
1187 	default:
1188 		WARN(1, "TPACKET version not supported\n");
1189 		BUG();
1190 		return NULL;
1191 	}
1192 }
1193 
1194 static void *prb_lookup_block(const struct packet_sock *po,
1195 			      const struct packet_ring_buffer *rb,
1196 			      unsigned int idx,
1197 			      int status)
1198 {
1199 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1200 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1201 
1202 	if (status != BLOCK_STATUS(pbd))
1203 		return NULL;
1204 	return pbd;
1205 }
1206 
1207 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1208 {
1209 	unsigned int prev;
1210 	if (rb->prb_bdqc.kactive_blk_num)
1211 		prev = rb->prb_bdqc.kactive_blk_num-1;
1212 	else
1213 		prev = rb->prb_bdqc.knum_blocks-1;
1214 	return prev;
1215 }
1216 
1217 /* Assumes caller has held the rx_queue.lock */
1218 static void *__prb_previous_block(struct packet_sock *po,
1219 					 struct packet_ring_buffer *rb,
1220 					 int status)
1221 {
1222 	unsigned int previous = prb_previous_blk_num(rb);
1223 	return prb_lookup_block(po, rb, previous, status);
1224 }
1225 
1226 static void *packet_previous_rx_frame(struct packet_sock *po,
1227 					     struct packet_ring_buffer *rb,
1228 					     int status)
1229 {
1230 	if (po->tp_version <= TPACKET_V2)
1231 		return packet_previous_frame(po, rb, status);
1232 
1233 	return __prb_previous_block(po, rb, status);
1234 }
1235 
1236 static void packet_increment_rx_head(struct packet_sock *po,
1237 					    struct packet_ring_buffer *rb)
1238 {
1239 	switch (po->tp_version) {
1240 	case TPACKET_V1:
1241 	case TPACKET_V2:
1242 		return packet_increment_head(rb);
1243 	case TPACKET_V3:
1244 	default:
1245 		WARN(1, "TPACKET version not supported.\n");
1246 		BUG();
1247 		return;
1248 	}
1249 }
1250 
1251 static void *packet_previous_frame(struct packet_sock *po,
1252 		struct packet_ring_buffer *rb,
1253 		int status)
1254 {
1255 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1256 	return packet_lookup_frame(po, rb, previous, status);
1257 }
1258 
1259 static void packet_increment_head(struct packet_ring_buffer *buff)
1260 {
1261 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1262 }
1263 
1264 static void packet_inc_pending(struct packet_ring_buffer *rb)
1265 {
1266 	this_cpu_inc(*rb->pending_refcnt);
1267 }
1268 
1269 static void packet_dec_pending(struct packet_ring_buffer *rb)
1270 {
1271 	this_cpu_dec(*rb->pending_refcnt);
1272 }
1273 
1274 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1275 {
1276 	unsigned int refcnt = 0;
1277 	int cpu;
1278 
1279 	/* We don't use pending refcount in rx_ring. */
1280 	if (rb->pending_refcnt == NULL)
1281 		return 0;
1282 
1283 	for_each_possible_cpu(cpu)
1284 		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1285 
1286 	return refcnt;
1287 }
1288 
1289 static int packet_alloc_pending(struct packet_sock *po)
1290 {
1291 	po->rx_ring.pending_refcnt = NULL;
1292 
1293 	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1294 	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1295 		return -ENOBUFS;
1296 
1297 	return 0;
1298 }
1299 
1300 static void packet_free_pending(struct packet_sock *po)
1301 {
1302 	free_percpu(po->tx_ring.pending_refcnt);
1303 }
1304 
1305 #define ROOM_POW_OFF	2
1306 #define ROOM_NONE	0x0
1307 #define ROOM_LOW	0x1
1308 #define ROOM_NORMAL	0x2
1309 
1310 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1311 {
1312 	int idx, len;
1313 
1314 	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1315 	idx = READ_ONCE(po->rx_ring.head);
1316 	if (pow_off)
1317 		idx += len >> pow_off;
1318 	if (idx >= len)
1319 		idx -= len;
1320 	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1321 }
1322 
1323 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1324 {
1325 	int idx, len;
1326 
1327 	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1328 	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1329 	if (pow_off)
1330 		idx += len >> pow_off;
1331 	if (idx >= len)
1332 		idx -= len;
1333 	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1334 }
1335 
1336 static int __packet_rcv_has_room(const struct packet_sock *po,
1337 				 const struct sk_buff *skb)
1338 {
1339 	const struct sock *sk = &po->sk;
1340 	int ret = ROOM_NONE;
1341 
1342 	if (po->prot_hook.func != tpacket_rcv) {
1343 		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1344 		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1345 				   - (skb ? skb->truesize : 0);
1346 
1347 		if (avail > (rcvbuf >> ROOM_POW_OFF))
1348 			return ROOM_NORMAL;
1349 		else if (avail > 0)
1350 			return ROOM_LOW;
1351 		else
1352 			return ROOM_NONE;
1353 	}
1354 
1355 	if (po->tp_version == TPACKET_V3) {
1356 		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1357 			ret = ROOM_NORMAL;
1358 		else if (__tpacket_v3_has_room(po, 0))
1359 			ret = ROOM_LOW;
1360 	} else {
1361 		if (__tpacket_has_room(po, ROOM_POW_OFF))
1362 			ret = ROOM_NORMAL;
1363 		else if (__tpacket_has_room(po, 0))
1364 			ret = ROOM_LOW;
1365 	}
1366 
1367 	return ret;
1368 }
1369 
1370 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1371 {
1372 	bool pressure;
1373 	int ret;
1374 
1375 	ret = __packet_rcv_has_room(po, skb);
1376 	pressure = ret != ROOM_NORMAL;
1377 
1378 	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1379 		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1380 
1381 	return ret;
1382 }
1383 
1384 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1385 {
1386 	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1387 	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1388 		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1389 }
1390 
1391 static void packet_sock_destruct(struct sock *sk)
1392 {
1393 	skb_queue_purge(&sk->sk_error_queue);
1394 
1395 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1396 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1397 
1398 	if (!sock_flag(sk, SOCK_DEAD)) {
1399 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1400 		return;
1401 	}
1402 }
1403 
1404 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1405 {
1406 	u32 *history = po->rollover->history;
1407 	u32 victim, rxhash;
1408 	int i, count = 0;
1409 
1410 	rxhash = skb_get_hash(skb);
1411 	for (i = 0; i < ROLLOVER_HLEN; i++)
1412 		if (READ_ONCE(history[i]) == rxhash)
1413 			count++;
1414 
1415 	victim = get_random_u32_below(ROLLOVER_HLEN);
1416 
1417 	/* Avoid dirtying the cache line if possible */
1418 	if (READ_ONCE(history[victim]) != rxhash)
1419 		WRITE_ONCE(history[victim], rxhash);
1420 
1421 	return count > (ROLLOVER_HLEN >> 1);
1422 }
1423 
1424 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1425 				      struct sk_buff *skb,
1426 				      unsigned int num)
1427 {
1428 	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1429 }
1430 
1431 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1432 				    struct sk_buff *skb,
1433 				    unsigned int num)
1434 {
1435 	unsigned int val = atomic_inc_return(&f->rr_cur);
1436 
1437 	return val % num;
1438 }
1439 
1440 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1441 				     struct sk_buff *skb,
1442 				     unsigned int num)
1443 {
1444 	return smp_processor_id() % num;
1445 }
1446 
1447 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1448 				     struct sk_buff *skb,
1449 				     unsigned int num)
1450 {
1451 	return get_random_u32_below(num);
1452 }
1453 
1454 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1455 					  struct sk_buff *skb,
1456 					  unsigned int idx, bool try_self,
1457 					  unsigned int num)
1458 {
1459 	struct packet_sock *po, *po_next, *po_skip = NULL;
1460 	unsigned int i, j, room = ROOM_NONE;
1461 
1462 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1463 
1464 	if (try_self) {
1465 		room = packet_rcv_has_room(po, skb);
1466 		if (room == ROOM_NORMAL ||
1467 		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1468 			return idx;
1469 		po_skip = po;
1470 	}
1471 
1472 	i = j = min_t(int, po->rollover->sock, num - 1);
1473 	do {
1474 		po_next = pkt_sk(rcu_dereference(f->arr[i]));
1475 		if (po_next != po_skip &&
1476 		    !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1477 		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1478 			if (i != j)
1479 				po->rollover->sock = i;
1480 			atomic_long_inc(&po->rollover->num);
1481 			if (room == ROOM_LOW)
1482 				atomic_long_inc(&po->rollover->num_huge);
1483 			return i;
1484 		}
1485 
1486 		if (++i == num)
1487 			i = 0;
1488 	} while (i != j);
1489 
1490 	atomic_long_inc(&po->rollover->num_failed);
1491 	return idx;
1492 }
1493 
1494 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1495 				    struct sk_buff *skb,
1496 				    unsigned int num)
1497 {
1498 	return skb_get_queue_mapping(skb) % num;
1499 }
1500 
1501 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1502 				     struct sk_buff *skb,
1503 				     unsigned int num)
1504 {
1505 	struct bpf_prog *prog;
1506 	unsigned int ret = 0;
1507 
1508 	rcu_read_lock();
1509 	prog = rcu_dereference(f->bpf_prog);
1510 	if (prog)
1511 		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1512 	rcu_read_unlock();
1513 
1514 	return ret;
1515 }
1516 
1517 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1518 {
1519 	return f->flags & (flag >> 8);
1520 }
1521 
1522 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1523 			     struct packet_type *pt, struct net_device *orig_dev)
1524 {
1525 	struct packet_fanout *f = pt->af_packet_priv;
1526 	unsigned int num = READ_ONCE(f->num_members);
1527 	struct net *net = read_pnet(&f->net);
1528 	struct packet_sock *po;
1529 	unsigned int idx;
1530 
1531 	if (!net_eq(dev_net(dev), net) || !num) {
1532 		kfree_skb(skb);
1533 		return 0;
1534 	}
1535 
1536 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1537 		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1538 		if (!skb)
1539 			return 0;
1540 	}
1541 	switch (f->type) {
1542 	case PACKET_FANOUT_HASH:
1543 	default:
1544 		idx = fanout_demux_hash(f, skb, num);
1545 		break;
1546 	case PACKET_FANOUT_LB:
1547 		idx = fanout_demux_lb(f, skb, num);
1548 		break;
1549 	case PACKET_FANOUT_CPU:
1550 		idx = fanout_demux_cpu(f, skb, num);
1551 		break;
1552 	case PACKET_FANOUT_RND:
1553 		idx = fanout_demux_rnd(f, skb, num);
1554 		break;
1555 	case PACKET_FANOUT_QM:
1556 		idx = fanout_demux_qm(f, skb, num);
1557 		break;
1558 	case PACKET_FANOUT_ROLLOVER:
1559 		idx = fanout_demux_rollover(f, skb, 0, false, num);
1560 		break;
1561 	case PACKET_FANOUT_CBPF:
1562 	case PACKET_FANOUT_EBPF:
1563 		idx = fanout_demux_bpf(f, skb, num);
1564 		break;
1565 	}
1566 
1567 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1568 		idx = fanout_demux_rollover(f, skb, idx, true, num);
1569 
1570 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1571 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1572 }
1573 
1574 DEFINE_MUTEX(fanout_mutex);
1575 EXPORT_SYMBOL_GPL(fanout_mutex);
1576 static LIST_HEAD(fanout_list);
1577 static u16 fanout_next_id;
1578 
1579 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1580 {
1581 	struct packet_fanout *f = po->fanout;
1582 
1583 	spin_lock(&f->lock);
1584 	rcu_assign_pointer(f->arr[f->num_members], sk);
1585 	smp_wmb();
1586 	f->num_members++;
1587 	if (f->num_members == 1)
1588 		dev_add_pack(&f->prot_hook);
1589 	spin_unlock(&f->lock);
1590 }
1591 
1592 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1593 {
1594 	struct packet_fanout *f = po->fanout;
1595 	int i;
1596 
1597 	spin_lock(&f->lock);
1598 	for (i = 0; i < f->num_members; i++) {
1599 		if (rcu_dereference_protected(f->arr[i],
1600 					      lockdep_is_held(&f->lock)) == sk)
1601 			break;
1602 	}
1603 	BUG_ON(i >= f->num_members);
1604 	rcu_assign_pointer(f->arr[i],
1605 			   rcu_dereference_protected(f->arr[f->num_members - 1],
1606 						     lockdep_is_held(&f->lock)));
1607 	f->num_members--;
1608 	if (f->num_members == 0)
1609 		__dev_remove_pack(&f->prot_hook);
1610 	spin_unlock(&f->lock);
1611 }
1612 
1613 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1614 {
1615 	if (sk->sk_family != PF_PACKET)
1616 		return false;
1617 
1618 	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1619 }
1620 
1621 static void fanout_init_data(struct packet_fanout *f)
1622 {
1623 	switch (f->type) {
1624 	case PACKET_FANOUT_LB:
1625 		atomic_set(&f->rr_cur, 0);
1626 		break;
1627 	case PACKET_FANOUT_CBPF:
1628 	case PACKET_FANOUT_EBPF:
1629 		RCU_INIT_POINTER(f->bpf_prog, NULL);
1630 		break;
1631 	}
1632 }
1633 
1634 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1635 {
1636 	struct bpf_prog *old;
1637 
1638 	spin_lock(&f->lock);
1639 	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1640 	rcu_assign_pointer(f->bpf_prog, new);
1641 	spin_unlock(&f->lock);
1642 
1643 	if (old) {
1644 		synchronize_net();
1645 		bpf_prog_destroy(old);
1646 	}
1647 }
1648 
1649 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1650 				unsigned int len)
1651 {
1652 	struct bpf_prog *new;
1653 	struct sock_fprog fprog;
1654 	int ret;
1655 
1656 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1657 		return -EPERM;
1658 
1659 	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1660 	if (ret)
1661 		return ret;
1662 
1663 	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1664 	if (ret)
1665 		return ret;
1666 
1667 	__fanout_set_data_bpf(po->fanout, new);
1668 	return 0;
1669 }
1670 
1671 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1672 				unsigned int len)
1673 {
1674 	struct bpf_prog *new;
1675 	u32 fd;
1676 
1677 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1678 		return -EPERM;
1679 	if (len != sizeof(fd))
1680 		return -EINVAL;
1681 	if (copy_from_sockptr(&fd, data, len))
1682 		return -EFAULT;
1683 
1684 	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1685 	if (IS_ERR(new))
1686 		return PTR_ERR(new);
1687 
1688 	__fanout_set_data_bpf(po->fanout, new);
1689 	return 0;
1690 }
1691 
1692 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1693 			   unsigned int len)
1694 {
1695 	switch (po->fanout->type) {
1696 	case PACKET_FANOUT_CBPF:
1697 		return fanout_set_data_cbpf(po, data, len);
1698 	case PACKET_FANOUT_EBPF:
1699 		return fanout_set_data_ebpf(po, data, len);
1700 	default:
1701 		return -EINVAL;
1702 	}
1703 }
1704 
1705 static void fanout_release_data(struct packet_fanout *f)
1706 {
1707 	switch (f->type) {
1708 	case PACKET_FANOUT_CBPF:
1709 	case PACKET_FANOUT_EBPF:
1710 		__fanout_set_data_bpf(f, NULL);
1711 	}
1712 }
1713 
1714 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1715 {
1716 	struct packet_fanout *f;
1717 
1718 	list_for_each_entry(f, &fanout_list, list) {
1719 		if (f->id == candidate_id &&
1720 		    read_pnet(&f->net) == sock_net(sk)) {
1721 			return false;
1722 		}
1723 	}
1724 	return true;
1725 }
1726 
1727 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1728 {
1729 	u16 id = fanout_next_id;
1730 
1731 	do {
1732 		if (__fanout_id_is_free(sk, id)) {
1733 			*new_id = id;
1734 			fanout_next_id = id + 1;
1735 			return true;
1736 		}
1737 
1738 		id++;
1739 	} while (id != fanout_next_id);
1740 
1741 	return false;
1742 }
1743 
1744 static int fanout_add(struct sock *sk, struct fanout_args *args)
1745 {
1746 	struct packet_rollover *rollover = NULL;
1747 	struct packet_sock *po = pkt_sk(sk);
1748 	u16 type_flags = args->type_flags;
1749 	struct packet_fanout *f, *match;
1750 	u8 type = type_flags & 0xff;
1751 	u8 flags = type_flags >> 8;
1752 	u16 id = args->id;
1753 	int err;
1754 
1755 	switch (type) {
1756 	case PACKET_FANOUT_ROLLOVER:
1757 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1758 			return -EINVAL;
1759 		break;
1760 	case PACKET_FANOUT_HASH:
1761 	case PACKET_FANOUT_LB:
1762 	case PACKET_FANOUT_CPU:
1763 	case PACKET_FANOUT_RND:
1764 	case PACKET_FANOUT_QM:
1765 	case PACKET_FANOUT_CBPF:
1766 	case PACKET_FANOUT_EBPF:
1767 		break;
1768 	default:
1769 		return -EINVAL;
1770 	}
1771 
1772 	mutex_lock(&fanout_mutex);
1773 
1774 	err = -EALREADY;
1775 	if (po->fanout)
1776 		goto out;
1777 
1778 	if (type == PACKET_FANOUT_ROLLOVER ||
1779 	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1780 		err = -ENOMEM;
1781 		rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1782 		if (!rollover)
1783 			goto out;
1784 		atomic_long_set(&rollover->num, 0);
1785 		atomic_long_set(&rollover->num_huge, 0);
1786 		atomic_long_set(&rollover->num_failed, 0);
1787 	}
1788 
1789 	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1790 		if (id != 0) {
1791 			err = -EINVAL;
1792 			goto out;
1793 		}
1794 		if (!fanout_find_new_id(sk, &id)) {
1795 			err = -ENOMEM;
1796 			goto out;
1797 		}
1798 		/* ephemeral flag for the first socket in the group: drop it */
1799 		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1800 	}
1801 
1802 	match = NULL;
1803 	list_for_each_entry(f, &fanout_list, list) {
1804 		if (f->id == id &&
1805 		    read_pnet(&f->net) == sock_net(sk)) {
1806 			match = f;
1807 			break;
1808 		}
1809 	}
1810 	err = -EINVAL;
1811 	if (match) {
1812 		if (match->flags != flags)
1813 			goto out;
1814 		if (args->max_num_members &&
1815 		    args->max_num_members != match->max_num_members)
1816 			goto out;
1817 	} else {
1818 		if (args->max_num_members > PACKET_FANOUT_MAX)
1819 			goto out;
1820 		if (!args->max_num_members)
1821 			/* legacy PACKET_FANOUT_MAX */
1822 			args->max_num_members = 256;
1823 		err = -ENOMEM;
1824 		match = kvzalloc(struct_size(match, arr, args->max_num_members),
1825 				 GFP_KERNEL);
1826 		if (!match)
1827 			goto out;
1828 		write_pnet(&match->net, sock_net(sk));
1829 		match->id = id;
1830 		match->type = type;
1831 		match->flags = flags;
1832 		INIT_LIST_HEAD(&match->list);
1833 		spin_lock_init(&match->lock);
1834 		refcount_set(&match->sk_ref, 0);
1835 		fanout_init_data(match);
1836 		match->prot_hook.type = po->prot_hook.type;
1837 		match->prot_hook.dev = po->prot_hook.dev;
1838 		match->prot_hook.func = packet_rcv_fanout;
1839 		match->prot_hook.af_packet_priv = match;
1840 		match->prot_hook.af_packet_net = read_pnet(&match->net);
1841 		match->prot_hook.id_match = match_fanout_group;
1842 		match->max_num_members = args->max_num_members;
1843 		match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1844 		list_add(&match->list, &fanout_list);
1845 	}
1846 	err = -EINVAL;
1847 
1848 	spin_lock(&po->bind_lock);
1849 	if (packet_sock_flag(po, PACKET_SOCK_RUNNING) &&
1850 	    match->type == type &&
1851 	    match->prot_hook.type == po->prot_hook.type &&
1852 	    match->prot_hook.dev == po->prot_hook.dev) {
1853 		err = -ENOSPC;
1854 		if (refcount_read(&match->sk_ref) < match->max_num_members) {
1855 			__dev_remove_pack(&po->prot_hook);
1856 
1857 			/* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1858 			WRITE_ONCE(po->fanout, match);
1859 
1860 			po->rollover = rollover;
1861 			rollover = NULL;
1862 			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1863 			__fanout_link(sk, po);
1864 			err = 0;
1865 		}
1866 	}
1867 	spin_unlock(&po->bind_lock);
1868 
1869 	if (err && !refcount_read(&match->sk_ref)) {
1870 		list_del(&match->list);
1871 		kvfree(match);
1872 	}
1873 
1874 out:
1875 	kfree(rollover);
1876 	mutex_unlock(&fanout_mutex);
1877 	return err;
1878 }
1879 
1880 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1881  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1882  * It is the responsibility of the caller to call fanout_release_data() and
1883  * free the returned packet_fanout (after synchronize_net())
1884  */
1885 static struct packet_fanout *fanout_release(struct sock *sk)
1886 {
1887 	struct packet_sock *po = pkt_sk(sk);
1888 	struct packet_fanout *f;
1889 
1890 	mutex_lock(&fanout_mutex);
1891 	f = po->fanout;
1892 	if (f) {
1893 		po->fanout = NULL;
1894 
1895 		if (refcount_dec_and_test(&f->sk_ref))
1896 			list_del(&f->list);
1897 		else
1898 			f = NULL;
1899 	}
1900 	mutex_unlock(&fanout_mutex);
1901 
1902 	return f;
1903 }
1904 
1905 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1906 					  struct sk_buff *skb)
1907 {
1908 	/* Earlier code assumed this would be a VLAN pkt, double-check
1909 	 * this now that we have the actual packet in hand. We can only
1910 	 * do this check on Ethernet devices.
1911 	 */
1912 	if (unlikely(dev->type != ARPHRD_ETHER))
1913 		return false;
1914 
1915 	skb_reset_mac_header(skb);
1916 	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1917 }
1918 
1919 static const struct proto_ops packet_ops;
1920 
1921 static const struct proto_ops packet_ops_spkt;
1922 
1923 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1924 			   struct packet_type *pt, struct net_device *orig_dev)
1925 {
1926 	struct sock *sk;
1927 	struct sockaddr_pkt *spkt;
1928 
1929 	/*
1930 	 *	When we registered the protocol we saved the socket in the data
1931 	 *	field for just this event.
1932 	 */
1933 
1934 	sk = pt->af_packet_priv;
1935 
1936 	/*
1937 	 *	Yank back the headers [hope the device set this
1938 	 *	right or kerboom...]
1939 	 *
1940 	 *	Incoming packets have ll header pulled,
1941 	 *	push it back.
1942 	 *
1943 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1944 	 *	so that this procedure is noop.
1945 	 */
1946 
1947 	if (skb->pkt_type == PACKET_LOOPBACK)
1948 		goto out;
1949 
1950 	if (!net_eq(dev_net(dev), sock_net(sk)))
1951 		goto out;
1952 
1953 	skb = skb_share_check(skb, GFP_ATOMIC);
1954 	if (skb == NULL)
1955 		goto oom;
1956 
1957 	/* drop any routing info */
1958 	skb_dst_drop(skb);
1959 
1960 	/* drop conntrack reference */
1961 	nf_reset_ct(skb);
1962 
1963 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1964 
1965 	skb_push(skb, skb->data - skb_mac_header(skb));
1966 
1967 	/*
1968 	 *	The SOCK_PACKET socket receives _all_ frames.
1969 	 */
1970 
1971 	spkt->spkt_family = dev->type;
1972 	strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1973 	spkt->spkt_protocol = skb->protocol;
1974 
1975 	/*
1976 	 *	Charge the memory to the socket. This is done specifically
1977 	 *	to prevent sockets using all the memory up.
1978 	 */
1979 
1980 	if (sock_queue_rcv_skb(sk, skb) == 0)
1981 		return 0;
1982 
1983 out:
1984 	kfree_skb(skb);
1985 oom:
1986 	return 0;
1987 }
1988 
1989 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1990 {
1991 	int depth;
1992 
1993 	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1994 	    sock->type == SOCK_RAW) {
1995 		skb_reset_mac_header(skb);
1996 		skb->protocol = dev_parse_header_protocol(skb);
1997 	}
1998 
1999 	/* Move network header to the right position for VLAN tagged packets */
2000 	if (likely(skb->dev->type == ARPHRD_ETHER) &&
2001 	    eth_type_vlan(skb->protocol) &&
2002 	    vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
2003 		skb_set_network_header(skb, depth);
2004 
2005 	skb_probe_transport_header(skb);
2006 }
2007 
2008 /*
2009  *	Output a raw packet to a device layer. This bypasses all the other
2010  *	protocol layers and you must therefore supply it with a complete frame
2011  */
2012 
2013 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
2014 			       size_t len)
2015 {
2016 	struct sock *sk = sock->sk;
2017 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
2018 	struct sk_buff *skb = NULL;
2019 	struct net_device *dev;
2020 	struct sockcm_cookie sockc;
2021 	__be16 proto = 0;
2022 	int err;
2023 	int extra_len = 0;
2024 
2025 	/*
2026 	 *	Get and verify the address.
2027 	 */
2028 
2029 	if (saddr) {
2030 		if (msg->msg_namelen < sizeof(struct sockaddr))
2031 			return -EINVAL;
2032 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
2033 			proto = saddr->spkt_protocol;
2034 	} else
2035 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
2036 
2037 	/*
2038 	 *	Find the device first to size check it
2039 	 */
2040 
2041 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
2042 retry:
2043 	rcu_read_lock();
2044 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
2045 	err = -ENODEV;
2046 	if (dev == NULL)
2047 		goto out_unlock;
2048 
2049 	err = -ENETDOWN;
2050 	if (!(dev->flags & IFF_UP))
2051 		goto out_unlock;
2052 
2053 	/*
2054 	 * You may not queue a frame bigger than the mtu. This is the lowest level
2055 	 * raw protocol and you must do your own fragmentation at this level.
2056 	 */
2057 
2058 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2059 		if (!netif_supports_nofcs(dev)) {
2060 			err = -EPROTONOSUPPORT;
2061 			goto out_unlock;
2062 		}
2063 		extra_len = 4; /* We're doing our own CRC */
2064 	}
2065 
2066 	err = -EMSGSIZE;
2067 	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2068 		goto out_unlock;
2069 
2070 	if (!skb) {
2071 		size_t reserved = LL_RESERVED_SPACE(dev);
2072 		int tlen = dev->needed_tailroom;
2073 		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2074 
2075 		rcu_read_unlock();
2076 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2077 		if (skb == NULL)
2078 			return -ENOBUFS;
2079 		/* FIXME: Save some space for broken drivers that write a hard
2080 		 * header at transmission time by themselves. PPP is the notable
2081 		 * one here. This should really be fixed at the driver level.
2082 		 */
2083 		skb_reserve(skb, reserved);
2084 		skb_reset_network_header(skb);
2085 
2086 		/* Try to align data part correctly */
2087 		if (hhlen) {
2088 			skb->data -= hhlen;
2089 			skb->tail -= hhlen;
2090 			if (len < hhlen)
2091 				skb_reset_network_header(skb);
2092 		}
2093 		err = memcpy_from_msg(skb_put(skb, len), msg, len);
2094 		if (err)
2095 			goto out_free;
2096 		goto retry;
2097 	}
2098 
2099 	if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2100 		err = -EINVAL;
2101 		goto out_unlock;
2102 	}
2103 	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2104 	    !packet_extra_vlan_len_allowed(dev, skb)) {
2105 		err = -EMSGSIZE;
2106 		goto out_unlock;
2107 	}
2108 
2109 	sockcm_init(&sockc, sk);
2110 	if (msg->msg_controllen) {
2111 		err = sock_cmsg_send(sk, msg, &sockc);
2112 		if (unlikely(err))
2113 			goto out_unlock;
2114 	}
2115 
2116 	skb->protocol = proto;
2117 	skb->dev = dev;
2118 	skb->priority = READ_ONCE(sk->sk_priority);
2119 	skb->mark = READ_ONCE(sk->sk_mark);
2120 	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
2121 	skb_setup_tx_timestamp(skb, sockc.tsflags);
2122 
2123 	if (unlikely(extra_len == 4))
2124 		skb->no_fcs = 1;
2125 
2126 	packet_parse_headers(skb, sock);
2127 
2128 	dev_queue_xmit(skb);
2129 	rcu_read_unlock();
2130 	return len;
2131 
2132 out_unlock:
2133 	rcu_read_unlock();
2134 out_free:
2135 	kfree_skb(skb);
2136 	return err;
2137 }
2138 
2139 static unsigned int run_filter(struct sk_buff *skb,
2140 			       const struct sock *sk,
2141 			       unsigned int res)
2142 {
2143 	struct sk_filter *filter;
2144 
2145 	rcu_read_lock();
2146 	filter = rcu_dereference(sk->sk_filter);
2147 	if (filter != NULL)
2148 		res = bpf_prog_run_clear_cb(filter->prog, skb);
2149 	rcu_read_unlock();
2150 
2151 	return res;
2152 }
2153 
2154 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2155 			   size_t *len, int vnet_hdr_sz)
2156 {
2157 	struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2158 
2159 	if (*len < vnet_hdr_sz)
2160 		return -EINVAL;
2161 	*len -= vnet_hdr_sz;
2162 
2163 	if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2164 		return -EINVAL;
2165 
2166 	return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2167 }
2168 
2169 /*
2170  * This function makes lazy skb cloning in hope that most of packets
2171  * are discarded by BPF.
2172  *
2173  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2174  * and skb->cb are mangled. It works because (and until) packets
2175  * falling here are owned by current CPU. Output packets are cloned
2176  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2177  * sequentially, so that if we return skb to original state on exit,
2178  * we will not harm anyone.
2179  */
2180 
2181 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2182 		      struct packet_type *pt, struct net_device *orig_dev)
2183 {
2184 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2185 	struct sock *sk = NULL;
2186 	struct sockaddr_ll *sll;
2187 	struct packet_sock *po;
2188 	u8 *skb_head = skb->data;
2189 	int skb_len = skb->len;
2190 	unsigned int snaplen, res;
2191 
2192 	if (skb->pkt_type == PACKET_LOOPBACK)
2193 		goto drop;
2194 
2195 	sk = pt->af_packet_priv;
2196 	po = pkt_sk(sk);
2197 
2198 	if (!net_eq(dev_net(dev), sock_net(sk)))
2199 		goto drop;
2200 
2201 	skb->dev = dev;
2202 
2203 	if (dev_has_header(dev)) {
2204 		/* The device has an explicit notion of ll header,
2205 		 * exported to higher levels.
2206 		 *
2207 		 * Otherwise, the device hides details of its frame
2208 		 * structure, so that corresponding packet head is
2209 		 * never delivered to user.
2210 		 */
2211 		if (sk->sk_type != SOCK_DGRAM)
2212 			skb_push(skb, skb->data - skb_mac_header(skb));
2213 		else if (skb->pkt_type == PACKET_OUTGOING) {
2214 			/* Special case: outgoing packets have ll header at head */
2215 			skb_pull(skb, skb_network_offset(skb));
2216 		}
2217 	}
2218 
2219 	snaplen = skb->len;
2220 
2221 	res = run_filter(skb, sk, snaplen);
2222 	if (!res)
2223 		goto drop_n_restore;
2224 	if (snaplen > res)
2225 		snaplen = res;
2226 
2227 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2228 		goto drop_n_acct;
2229 
2230 	if (skb_shared(skb)) {
2231 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2232 		if (nskb == NULL)
2233 			goto drop_n_acct;
2234 
2235 		if (skb_head != skb->data) {
2236 			skb->data = skb_head;
2237 			skb->len = skb_len;
2238 		}
2239 		consume_skb(skb);
2240 		skb = nskb;
2241 	}
2242 
2243 	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2244 
2245 	sll = &PACKET_SKB_CB(skb)->sa.ll;
2246 	sll->sll_hatype = dev->type;
2247 	sll->sll_pkttype = skb->pkt_type;
2248 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2249 		sll->sll_ifindex = orig_dev->ifindex;
2250 	else
2251 		sll->sll_ifindex = dev->ifindex;
2252 
2253 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2254 
2255 	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2256 	 * Use their space for storing the original skb length.
2257 	 */
2258 	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2259 
2260 	if (pskb_trim(skb, snaplen))
2261 		goto drop_n_acct;
2262 
2263 	skb_set_owner_r(skb, sk);
2264 	skb->dev = NULL;
2265 	skb_dst_drop(skb);
2266 
2267 	/* drop conntrack reference */
2268 	nf_reset_ct(skb);
2269 
2270 	spin_lock(&sk->sk_receive_queue.lock);
2271 	po->stats.stats1.tp_packets++;
2272 	sock_skb_set_dropcount(sk, skb);
2273 	skb_clear_delivery_time(skb);
2274 	__skb_queue_tail(&sk->sk_receive_queue, skb);
2275 	spin_unlock(&sk->sk_receive_queue.lock);
2276 	sk->sk_data_ready(sk);
2277 	return 0;
2278 
2279 drop_n_acct:
2280 	atomic_inc(&po->tp_drops);
2281 	atomic_inc(&sk->sk_drops);
2282 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2283 
2284 drop_n_restore:
2285 	if (skb_head != skb->data && skb_shared(skb)) {
2286 		skb->data = skb_head;
2287 		skb->len = skb_len;
2288 	}
2289 drop:
2290 	sk_skb_reason_drop(sk, skb, drop_reason);
2291 	return 0;
2292 }
2293 
2294 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2295 		       struct packet_type *pt, struct net_device *orig_dev)
2296 {
2297 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2298 	struct sock *sk = NULL;
2299 	struct packet_sock *po;
2300 	struct sockaddr_ll *sll;
2301 	union tpacket_uhdr h;
2302 	u8 *skb_head = skb->data;
2303 	int skb_len = skb->len;
2304 	unsigned int snaplen, res;
2305 	unsigned long status = TP_STATUS_USER;
2306 	unsigned short macoff, hdrlen;
2307 	unsigned int netoff;
2308 	struct sk_buff *copy_skb = NULL;
2309 	struct timespec64 ts;
2310 	__u32 ts_status;
2311 	unsigned int slot_id = 0;
2312 	int vnet_hdr_sz = 0;
2313 
2314 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2315 	 * We may add members to them until current aligned size without forcing
2316 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2317 	 */
2318 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2319 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2320 
2321 	if (skb->pkt_type == PACKET_LOOPBACK)
2322 		goto drop;
2323 
2324 	sk = pt->af_packet_priv;
2325 	po = pkt_sk(sk);
2326 
2327 	if (!net_eq(dev_net(dev), sock_net(sk)))
2328 		goto drop;
2329 
2330 	if (dev_has_header(dev)) {
2331 		if (sk->sk_type != SOCK_DGRAM)
2332 			skb_push(skb, skb->data - skb_mac_header(skb));
2333 		else if (skb->pkt_type == PACKET_OUTGOING) {
2334 			/* Special case: outgoing packets have ll header at head */
2335 			skb_pull(skb, skb_network_offset(skb));
2336 		}
2337 	}
2338 
2339 	snaplen = skb->len;
2340 
2341 	res = run_filter(skb, sk, snaplen);
2342 	if (!res)
2343 		goto drop_n_restore;
2344 
2345 	/* If we are flooded, just give up */
2346 	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2347 		atomic_inc(&po->tp_drops);
2348 		goto drop_n_restore;
2349 	}
2350 
2351 	if (skb->ip_summed == CHECKSUM_PARTIAL)
2352 		status |= TP_STATUS_CSUMNOTREADY;
2353 	else if (skb->pkt_type != PACKET_OUTGOING &&
2354 		 skb_csum_unnecessary(skb))
2355 		status |= TP_STATUS_CSUM_VALID;
2356 	if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2357 		status |= TP_STATUS_GSO_TCP;
2358 
2359 	if (snaplen > res)
2360 		snaplen = res;
2361 
2362 	if (sk->sk_type == SOCK_DGRAM) {
2363 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2364 				  po->tp_reserve;
2365 	} else {
2366 		unsigned int maclen = skb_network_offset(skb);
2367 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2368 				       (maclen < 16 ? 16 : maclen)) +
2369 				       po->tp_reserve;
2370 		vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2371 		if (vnet_hdr_sz)
2372 			netoff += vnet_hdr_sz;
2373 		macoff = netoff - maclen;
2374 	}
2375 	if (netoff > USHRT_MAX) {
2376 		atomic_inc(&po->tp_drops);
2377 		goto drop_n_restore;
2378 	}
2379 	if (po->tp_version <= TPACKET_V2) {
2380 		if (macoff + snaplen > po->rx_ring.frame_size) {
2381 			if (READ_ONCE(po->copy_thresh) &&
2382 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2383 				if (skb_shared(skb)) {
2384 					copy_skb = skb_clone(skb, GFP_ATOMIC);
2385 				} else {
2386 					copy_skb = skb_get(skb);
2387 					skb_head = skb->data;
2388 				}
2389 				if (copy_skb) {
2390 					memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2391 					       sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2392 					skb_set_owner_r(copy_skb, sk);
2393 				}
2394 			}
2395 			snaplen = po->rx_ring.frame_size - macoff;
2396 			if ((int)snaplen < 0) {
2397 				snaplen = 0;
2398 				vnet_hdr_sz = 0;
2399 			}
2400 		}
2401 	} else if (unlikely(macoff + snaplen >
2402 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2403 		u32 nval;
2404 
2405 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2406 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2407 			    snaplen, nval, macoff);
2408 		snaplen = nval;
2409 		if (unlikely((int)snaplen < 0)) {
2410 			snaplen = 0;
2411 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2412 			vnet_hdr_sz = 0;
2413 		}
2414 	}
2415 	spin_lock(&sk->sk_receive_queue.lock);
2416 	h.raw = packet_current_rx_frame(po, skb,
2417 					TP_STATUS_KERNEL, (macoff+snaplen));
2418 	if (!h.raw)
2419 		goto drop_n_account;
2420 
2421 	if (po->tp_version <= TPACKET_V2) {
2422 		slot_id = po->rx_ring.head;
2423 		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2424 			goto drop_n_account;
2425 		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2426 	}
2427 
2428 	if (vnet_hdr_sz &&
2429 	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2430 				    sizeof(struct virtio_net_hdr),
2431 				    vio_le(), true, 0)) {
2432 		if (po->tp_version == TPACKET_V3)
2433 			prb_clear_blk_fill_status(&po->rx_ring);
2434 		goto drop_n_account;
2435 	}
2436 
2437 	if (po->tp_version <= TPACKET_V2) {
2438 		packet_increment_rx_head(po, &po->rx_ring);
2439 	/*
2440 	 * LOSING will be reported till you read the stats,
2441 	 * because it's COR - Clear On Read.
2442 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2443 	 * at packet level.
2444 	 */
2445 		if (atomic_read(&po->tp_drops))
2446 			status |= TP_STATUS_LOSING;
2447 	}
2448 
2449 	po->stats.stats1.tp_packets++;
2450 	if (copy_skb) {
2451 		status |= TP_STATUS_COPY;
2452 		skb_clear_delivery_time(copy_skb);
2453 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2454 	}
2455 	spin_unlock(&sk->sk_receive_queue.lock);
2456 
2457 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2458 
2459 	/* Always timestamp; prefer an existing software timestamp taken
2460 	 * closer to the time of capture.
2461 	 */
2462 	ts_status = tpacket_get_timestamp(skb, &ts,
2463 					  READ_ONCE(po->tp_tstamp) |
2464 					  SOF_TIMESTAMPING_SOFTWARE);
2465 	if (!ts_status)
2466 		ktime_get_real_ts64(&ts);
2467 
2468 	status |= ts_status;
2469 
2470 	switch (po->tp_version) {
2471 	case TPACKET_V1:
2472 		h.h1->tp_len = skb->len;
2473 		h.h1->tp_snaplen = snaplen;
2474 		h.h1->tp_mac = macoff;
2475 		h.h1->tp_net = netoff;
2476 		h.h1->tp_sec = ts.tv_sec;
2477 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2478 		hdrlen = sizeof(*h.h1);
2479 		break;
2480 	case TPACKET_V2:
2481 		h.h2->tp_len = skb->len;
2482 		h.h2->tp_snaplen = snaplen;
2483 		h.h2->tp_mac = macoff;
2484 		h.h2->tp_net = netoff;
2485 		h.h2->tp_sec = ts.tv_sec;
2486 		h.h2->tp_nsec = ts.tv_nsec;
2487 		if (skb_vlan_tag_present(skb)) {
2488 			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2489 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2490 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2491 		} else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
2492 			h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev);
2493 			h.h2->tp_vlan_tpid = ntohs(skb->protocol);
2494 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2495 		} else {
2496 			h.h2->tp_vlan_tci = 0;
2497 			h.h2->tp_vlan_tpid = 0;
2498 		}
2499 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2500 		hdrlen = sizeof(*h.h2);
2501 		break;
2502 	case TPACKET_V3:
2503 		/* tp_nxt_offset,vlan are already populated above.
2504 		 * So DONT clear those fields here
2505 		 */
2506 		h.h3->tp_status |= status;
2507 		h.h3->tp_len = skb->len;
2508 		h.h3->tp_snaplen = snaplen;
2509 		h.h3->tp_mac = macoff;
2510 		h.h3->tp_net = netoff;
2511 		h.h3->tp_sec  = ts.tv_sec;
2512 		h.h3->tp_nsec = ts.tv_nsec;
2513 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2514 		hdrlen = sizeof(*h.h3);
2515 		break;
2516 	default:
2517 		BUG();
2518 	}
2519 
2520 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2521 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2522 	sll->sll_family = AF_PACKET;
2523 	sll->sll_hatype = dev->type;
2524 	sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ?
2525 		vlan_get_protocol_dgram(skb) : skb->protocol;
2526 	sll->sll_pkttype = skb->pkt_type;
2527 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2528 		sll->sll_ifindex = orig_dev->ifindex;
2529 	else
2530 		sll->sll_ifindex = dev->ifindex;
2531 
2532 	smp_mb();
2533 
2534 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2535 	if (po->tp_version <= TPACKET_V2) {
2536 		u8 *start, *end;
2537 
2538 		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2539 					macoff + snaplen);
2540 
2541 		for (start = h.raw; start < end; start += PAGE_SIZE)
2542 			flush_dcache_page(pgv_to_page(start));
2543 	}
2544 	smp_wmb();
2545 #endif
2546 
2547 	if (po->tp_version <= TPACKET_V2) {
2548 		spin_lock(&sk->sk_receive_queue.lock);
2549 		__packet_set_status(po, h.raw, status);
2550 		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2551 		spin_unlock(&sk->sk_receive_queue.lock);
2552 		sk->sk_data_ready(sk);
2553 	} else if (po->tp_version == TPACKET_V3) {
2554 		prb_clear_blk_fill_status(&po->rx_ring);
2555 	}
2556 
2557 drop_n_restore:
2558 	if (skb_head != skb->data && skb_shared(skb)) {
2559 		skb->data = skb_head;
2560 		skb->len = skb_len;
2561 	}
2562 drop:
2563 	sk_skb_reason_drop(sk, skb, drop_reason);
2564 	return 0;
2565 
2566 drop_n_account:
2567 	spin_unlock(&sk->sk_receive_queue.lock);
2568 	atomic_inc(&po->tp_drops);
2569 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2570 
2571 	sk->sk_data_ready(sk);
2572 	sk_skb_reason_drop(sk, copy_skb, drop_reason);
2573 	goto drop_n_restore;
2574 }
2575 
2576 static void tpacket_destruct_skb(struct sk_buff *skb)
2577 {
2578 	struct packet_sock *po = pkt_sk(skb->sk);
2579 
2580 	if (likely(po->tx_ring.pg_vec)) {
2581 		void *ph;
2582 		__u32 ts;
2583 
2584 		ph = skb_zcopy_get_nouarg(skb);
2585 		packet_dec_pending(&po->tx_ring);
2586 
2587 		ts = __packet_set_timestamp(po, ph, skb);
2588 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2589 
2590 		complete(&po->skb_completion);
2591 	}
2592 
2593 	sock_wfree(skb);
2594 }
2595 
2596 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2597 {
2598 	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2599 	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2600 	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2601 	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2602 		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2603 			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2604 			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2605 
2606 	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2607 		return -EINVAL;
2608 
2609 	return 0;
2610 }
2611 
2612 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2613 				 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2614 {
2615 	int ret;
2616 
2617 	if (*len < vnet_hdr_sz)
2618 		return -EINVAL;
2619 	*len -= vnet_hdr_sz;
2620 
2621 	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2622 		return -EFAULT;
2623 
2624 	ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2625 	if (ret)
2626 		return ret;
2627 
2628 	/* move iter to point to the start of mac header */
2629 	if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2630 		iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2631 
2632 	return 0;
2633 }
2634 
2635 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2636 		void *frame, struct net_device *dev, void *data, int tp_len,
2637 		__be16 proto, unsigned char *addr, int hlen, int copylen,
2638 		const struct sockcm_cookie *sockc)
2639 {
2640 	union tpacket_uhdr ph;
2641 	int to_write, offset, len, nr_frags, len_max;
2642 	struct socket *sock = po->sk.sk_socket;
2643 	struct page *page;
2644 	int err;
2645 
2646 	ph.raw = frame;
2647 
2648 	skb->protocol = proto;
2649 	skb->dev = dev;
2650 	skb->priority = READ_ONCE(po->sk.sk_priority);
2651 	skb->mark = READ_ONCE(po->sk.sk_mark);
2652 	skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid);
2653 	skb_setup_tx_timestamp(skb, sockc->tsflags);
2654 	skb_zcopy_set_nouarg(skb, ph.raw);
2655 
2656 	skb_reserve(skb, hlen);
2657 	skb_reset_network_header(skb);
2658 
2659 	to_write = tp_len;
2660 
2661 	if (sock->type == SOCK_DGRAM) {
2662 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2663 				NULL, tp_len);
2664 		if (unlikely(err < 0))
2665 			return -EINVAL;
2666 	} else if (copylen) {
2667 		int hdrlen = min_t(int, copylen, tp_len);
2668 
2669 		skb_push(skb, dev->hard_header_len);
2670 		skb_put(skb, copylen - dev->hard_header_len);
2671 		err = skb_store_bits(skb, 0, data, hdrlen);
2672 		if (unlikely(err))
2673 			return err;
2674 		if (!dev_validate_header(dev, skb->data, hdrlen))
2675 			return -EINVAL;
2676 
2677 		data += hdrlen;
2678 		to_write -= hdrlen;
2679 	}
2680 
2681 	offset = offset_in_page(data);
2682 	len_max = PAGE_SIZE - offset;
2683 	len = ((to_write > len_max) ? len_max : to_write);
2684 
2685 	skb->data_len = to_write;
2686 	skb->len += to_write;
2687 	skb->truesize += to_write;
2688 	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2689 
2690 	while (likely(to_write)) {
2691 		nr_frags = skb_shinfo(skb)->nr_frags;
2692 
2693 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2694 			pr_err("Packet exceed the number of skb frags(%u)\n",
2695 			       (unsigned int)MAX_SKB_FRAGS);
2696 			return -EFAULT;
2697 		}
2698 
2699 		page = pgv_to_page(data);
2700 		data += len;
2701 		flush_dcache_page(page);
2702 		get_page(page);
2703 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2704 		to_write -= len;
2705 		offset = 0;
2706 		len_max = PAGE_SIZE;
2707 		len = ((to_write > len_max) ? len_max : to_write);
2708 	}
2709 
2710 	packet_parse_headers(skb, sock);
2711 
2712 	return tp_len;
2713 }
2714 
2715 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2716 				int size_max, void **data)
2717 {
2718 	union tpacket_uhdr ph;
2719 	int tp_len, off;
2720 
2721 	ph.raw = frame;
2722 
2723 	switch (po->tp_version) {
2724 	case TPACKET_V3:
2725 		if (ph.h3->tp_next_offset != 0) {
2726 			pr_warn_once("variable sized slot not supported");
2727 			return -EINVAL;
2728 		}
2729 		tp_len = ph.h3->tp_len;
2730 		break;
2731 	case TPACKET_V2:
2732 		tp_len = ph.h2->tp_len;
2733 		break;
2734 	default:
2735 		tp_len = ph.h1->tp_len;
2736 		break;
2737 	}
2738 	if (unlikely(tp_len > size_max)) {
2739 		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2740 		return -EMSGSIZE;
2741 	}
2742 
2743 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2744 		int off_min, off_max;
2745 
2746 		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2747 		off_max = po->tx_ring.frame_size - tp_len;
2748 		if (po->sk.sk_type == SOCK_DGRAM) {
2749 			switch (po->tp_version) {
2750 			case TPACKET_V3:
2751 				off = ph.h3->tp_net;
2752 				break;
2753 			case TPACKET_V2:
2754 				off = ph.h2->tp_net;
2755 				break;
2756 			default:
2757 				off = ph.h1->tp_net;
2758 				break;
2759 			}
2760 		} else {
2761 			switch (po->tp_version) {
2762 			case TPACKET_V3:
2763 				off = ph.h3->tp_mac;
2764 				break;
2765 			case TPACKET_V2:
2766 				off = ph.h2->tp_mac;
2767 				break;
2768 			default:
2769 				off = ph.h1->tp_mac;
2770 				break;
2771 			}
2772 		}
2773 		if (unlikely((off < off_min) || (off_max < off)))
2774 			return -EINVAL;
2775 	} else {
2776 		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2777 	}
2778 
2779 	*data = frame + off;
2780 	return tp_len;
2781 }
2782 
2783 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2784 {
2785 	struct sk_buff *skb = NULL;
2786 	struct net_device *dev;
2787 	struct virtio_net_hdr *vnet_hdr = NULL;
2788 	struct sockcm_cookie sockc;
2789 	__be16 proto;
2790 	int err, reserve = 0;
2791 	void *ph;
2792 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2793 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2794 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2795 	unsigned char *addr = NULL;
2796 	int tp_len, size_max;
2797 	void *data;
2798 	int len_sum = 0;
2799 	int status = TP_STATUS_AVAILABLE;
2800 	int hlen, tlen, copylen = 0;
2801 	long timeo = 0;
2802 
2803 	mutex_lock(&po->pg_vec_lock);
2804 
2805 	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2806 	 * we need to confirm it under protection of pg_vec_lock.
2807 	 */
2808 	if (unlikely(!po->tx_ring.pg_vec)) {
2809 		err = -EBUSY;
2810 		goto out;
2811 	}
2812 	if (likely(saddr == NULL)) {
2813 		dev	= packet_cached_dev_get(po);
2814 		proto	= READ_ONCE(po->num);
2815 	} else {
2816 		err = -EINVAL;
2817 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2818 			goto out;
2819 		if (msg->msg_namelen < (saddr->sll_halen
2820 					+ offsetof(struct sockaddr_ll,
2821 						sll_addr)))
2822 			goto out;
2823 		proto	= saddr->sll_protocol;
2824 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2825 		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2826 			if (dev && msg->msg_namelen < dev->addr_len +
2827 				   offsetof(struct sockaddr_ll, sll_addr))
2828 				goto out_put;
2829 			addr = saddr->sll_addr;
2830 		}
2831 	}
2832 
2833 	err = -ENXIO;
2834 	if (unlikely(dev == NULL))
2835 		goto out;
2836 	err = -ENETDOWN;
2837 	if (unlikely(!(dev->flags & IFF_UP)))
2838 		goto out_put;
2839 
2840 	sockcm_init(&sockc, &po->sk);
2841 	if (msg->msg_controllen) {
2842 		err = sock_cmsg_send(&po->sk, msg, &sockc);
2843 		if (unlikely(err))
2844 			goto out_put;
2845 	}
2846 
2847 	if (po->sk.sk_socket->type == SOCK_RAW)
2848 		reserve = dev->hard_header_len;
2849 	size_max = po->tx_ring.frame_size
2850 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2851 
2852 	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2853 		size_max = dev->mtu + reserve + VLAN_HLEN;
2854 
2855 	reinit_completion(&po->skb_completion);
2856 
2857 	do {
2858 		ph = packet_current_frame(po, &po->tx_ring,
2859 					  TP_STATUS_SEND_REQUEST);
2860 		if (unlikely(ph == NULL)) {
2861 			if (need_wait && skb) {
2862 				timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2863 				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2864 				if (timeo <= 0) {
2865 					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2866 					goto out_put;
2867 				}
2868 			}
2869 			/* check for additional frames */
2870 			continue;
2871 		}
2872 
2873 		skb = NULL;
2874 		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2875 		if (tp_len < 0)
2876 			goto tpacket_error;
2877 
2878 		status = TP_STATUS_SEND_REQUEST;
2879 		hlen = LL_RESERVED_SPACE(dev);
2880 		tlen = dev->needed_tailroom;
2881 		if (vnet_hdr_sz) {
2882 			vnet_hdr = data;
2883 			data += vnet_hdr_sz;
2884 			tp_len -= vnet_hdr_sz;
2885 			if (tp_len < 0 ||
2886 			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2887 				tp_len = -EINVAL;
2888 				goto tpacket_error;
2889 			}
2890 			copylen = __virtio16_to_cpu(vio_le(),
2891 						    vnet_hdr->hdr_len);
2892 		}
2893 		copylen = max_t(int, copylen, dev->hard_header_len);
2894 		skb = sock_alloc_send_skb(&po->sk,
2895 				hlen + tlen + sizeof(struct sockaddr_ll) +
2896 				(copylen - dev->hard_header_len),
2897 				!need_wait, &err);
2898 
2899 		if (unlikely(skb == NULL)) {
2900 			/* we assume the socket was initially writeable ... */
2901 			if (likely(len_sum > 0))
2902 				err = len_sum;
2903 			goto out_status;
2904 		}
2905 		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2906 					  addr, hlen, copylen, &sockc);
2907 		if (likely(tp_len >= 0) &&
2908 		    tp_len > dev->mtu + reserve &&
2909 		    !vnet_hdr_sz &&
2910 		    !packet_extra_vlan_len_allowed(dev, skb))
2911 			tp_len = -EMSGSIZE;
2912 
2913 		if (unlikely(tp_len < 0)) {
2914 tpacket_error:
2915 			if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2916 				__packet_set_status(po, ph,
2917 						TP_STATUS_AVAILABLE);
2918 				packet_increment_head(&po->tx_ring);
2919 				kfree_skb(skb);
2920 				continue;
2921 			} else {
2922 				status = TP_STATUS_WRONG_FORMAT;
2923 				err = tp_len;
2924 				goto out_status;
2925 			}
2926 		}
2927 
2928 		if (vnet_hdr_sz) {
2929 			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2930 				tp_len = -EINVAL;
2931 				goto tpacket_error;
2932 			}
2933 			virtio_net_hdr_set_proto(skb, vnet_hdr);
2934 		}
2935 
2936 		skb->destructor = tpacket_destruct_skb;
2937 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2938 		packet_inc_pending(&po->tx_ring);
2939 
2940 		status = TP_STATUS_SEND_REQUEST;
2941 		err = packet_xmit(po, skb);
2942 		if (unlikely(err != 0)) {
2943 			if (err > 0)
2944 				err = net_xmit_errno(err);
2945 			if (err && __packet_get_status(po, ph) ==
2946 				   TP_STATUS_AVAILABLE) {
2947 				/* skb was destructed already */
2948 				skb = NULL;
2949 				goto out_status;
2950 			}
2951 			/*
2952 			 * skb was dropped but not destructed yet;
2953 			 * let's treat it like congestion or err < 0
2954 			 */
2955 			err = 0;
2956 		}
2957 		packet_increment_head(&po->tx_ring);
2958 		len_sum += tp_len;
2959 	} while (likely((ph != NULL) ||
2960 		/* Note: packet_read_pending() might be slow if we have
2961 		 * to call it as it's per_cpu variable, but in fast-path
2962 		 * we already short-circuit the loop with the first
2963 		 * condition, and luckily don't have to go that path
2964 		 * anyway.
2965 		 */
2966 		 (need_wait && packet_read_pending(&po->tx_ring))));
2967 
2968 	err = len_sum;
2969 	goto out_put;
2970 
2971 out_status:
2972 	__packet_set_status(po, ph, status);
2973 	kfree_skb(skb);
2974 out_put:
2975 	dev_put(dev);
2976 out:
2977 	mutex_unlock(&po->pg_vec_lock);
2978 	return err;
2979 }
2980 
2981 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2982 				        size_t reserve, size_t len,
2983 				        size_t linear, int noblock,
2984 				        int *err)
2985 {
2986 	struct sk_buff *skb;
2987 
2988 	/* Under a page?  Don't bother with paged skb. */
2989 	if (prepad + len < PAGE_SIZE || !linear)
2990 		linear = len;
2991 
2992 	if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2993 		linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2994 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2995 				   err, PAGE_ALLOC_COSTLY_ORDER);
2996 	if (!skb)
2997 		return NULL;
2998 
2999 	skb_reserve(skb, reserve);
3000 	skb_put(skb, linear);
3001 	skb->data_len = len - linear;
3002 	skb->len += len - linear;
3003 
3004 	return skb;
3005 }
3006 
3007 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
3008 {
3009 	struct sock *sk = sock->sk;
3010 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
3011 	struct sk_buff *skb;
3012 	struct net_device *dev;
3013 	__be16 proto;
3014 	unsigned char *addr = NULL;
3015 	int err, reserve = 0;
3016 	struct sockcm_cookie sockc;
3017 	struct virtio_net_hdr vnet_hdr = { 0 };
3018 	int offset = 0;
3019 	struct packet_sock *po = pkt_sk(sk);
3020 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
3021 	int hlen, tlen, linear;
3022 	int extra_len = 0;
3023 
3024 	/*
3025 	 *	Get and verify the address.
3026 	 */
3027 
3028 	if (likely(saddr == NULL)) {
3029 		dev	= packet_cached_dev_get(po);
3030 		proto	= READ_ONCE(po->num);
3031 	} else {
3032 		err = -EINVAL;
3033 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
3034 			goto out;
3035 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
3036 			goto out;
3037 		proto	= saddr->sll_protocol;
3038 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
3039 		if (sock->type == SOCK_DGRAM) {
3040 			if (dev && msg->msg_namelen < dev->addr_len +
3041 				   offsetof(struct sockaddr_ll, sll_addr))
3042 				goto out_unlock;
3043 			addr = saddr->sll_addr;
3044 		}
3045 	}
3046 
3047 	err = -ENXIO;
3048 	if (unlikely(dev == NULL))
3049 		goto out_unlock;
3050 	err = -ENETDOWN;
3051 	if (unlikely(!(dev->flags & IFF_UP)))
3052 		goto out_unlock;
3053 
3054 	sockcm_init(&sockc, sk);
3055 	sockc.mark = READ_ONCE(sk->sk_mark);
3056 	if (msg->msg_controllen) {
3057 		err = sock_cmsg_send(sk, msg, &sockc);
3058 		if (unlikely(err))
3059 			goto out_unlock;
3060 	}
3061 
3062 	if (sock->type == SOCK_RAW)
3063 		reserve = dev->hard_header_len;
3064 	if (vnet_hdr_sz) {
3065 		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3066 		if (err)
3067 			goto out_unlock;
3068 	}
3069 
3070 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3071 		if (!netif_supports_nofcs(dev)) {
3072 			err = -EPROTONOSUPPORT;
3073 			goto out_unlock;
3074 		}
3075 		extra_len = 4; /* We're doing our own CRC */
3076 	}
3077 
3078 	err = -EMSGSIZE;
3079 	if (!vnet_hdr.gso_type &&
3080 	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3081 		goto out_unlock;
3082 
3083 	err = -ENOBUFS;
3084 	hlen = LL_RESERVED_SPACE(dev);
3085 	tlen = dev->needed_tailroom;
3086 	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3087 	linear = max(linear, min_t(int, len, dev->hard_header_len));
3088 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3089 			       msg->msg_flags & MSG_DONTWAIT, &err);
3090 	if (skb == NULL)
3091 		goto out_unlock;
3092 
3093 	skb_reset_network_header(skb);
3094 
3095 	err = -EINVAL;
3096 	if (sock->type == SOCK_DGRAM) {
3097 		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3098 		if (unlikely(offset < 0))
3099 			goto out_free;
3100 	} else if (reserve) {
3101 		skb_reserve(skb, -reserve);
3102 		if (len < reserve + sizeof(struct ipv6hdr) &&
3103 		    dev->min_header_len != dev->hard_header_len)
3104 			skb_reset_network_header(skb);
3105 	}
3106 
3107 	/* Returns -EFAULT on error */
3108 	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3109 	if (err)
3110 		goto out_free;
3111 
3112 	if ((sock->type == SOCK_RAW &&
3113 	     !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3114 		err = -EINVAL;
3115 		goto out_free;
3116 	}
3117 
3118 	skb_setup_tx_timestamp(skb, sockc.tsflags);
3119 
3120 	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3121 	    !packet_extra_vlan_len_allowed(dev, skb)) {
3122 		err = -EMSGSIZE;
3123 		goto out_free;
3124 	}
3125 
3126 	skb->protocol = proto;
3127 	skb->dev = dev;
3128 	skb->priority = READ_ONCE(sk->sk_priority);
3129 	skb->mark = sockc.mark;
3130 	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
3131 
3132 	if (unlikely(extra_len == 4))
3133 		skb->no_fcs = 1;
3134 
3135 	packet_parse_headers(skb, sock);
3136 
3137 	if (vnet_hdr_sz) {
3138 		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3139 		if (err)
3140 			goto out_free;
3141 		len += vnet_hdr_sz;
3142 		virtio_net_hdr_set_proto(skb, &vnet_hdr);
3143 	}
3144 
3145 	err = packet_xmit(po, skb);
3146 
3147 	if (unlikely(err != 0)) {
3148 		if (err > 0)
3149 			err = net_xmit_errno(err);
3150 		if (err)
3151 			goto out_unlock;
3152 	}
3153 
3154 	dev_put(dev);
3155 
3156 	return len;
3157 
3158 out_free:
3159 	kfree_skb(skb);
3160 out_unlock:
3161 	dev_put(dev);
3162 out:
3163 	return err;
3164 }
3165 
3166 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3167 {
3168 	struct sock *sk = sock->sk;
3169 	struct packet_sock *po = pkt_sk(sk);
3170 
3171 	/* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3172 	 * tpacket_snd() will redo the check safely.
3173 	 */
3174 	if (data_race(po->tx_ring.pg_vec))
3175 		return tpacket_snd(po, msg);
3176 
3177 	return packet_snd(sock, msg, len);
3178 }
3179 
3180 /*
3181  *	Close a PACKET socket. This is fairly simple. We immediately go
3182  *	to 'closed' state and remove our protocol entry in the device list.
3183  */
3184 
3185 static int packet_release(struct socket *sock)
3186 {
3187 	struct sock *sk = sock->sk;
3188 	struct packet_sock *po;
3189 	struct packet_fanout *f;
3190 	struct net *net;
3191 	union tpacket_req_u req_u;
3192 
3193 	if (!sk)
3194 		return 0;
3195 
3196 	net = sock_net(sk);
3197 	po = pkt_sk(sk);
3198 
3199 	mutex_lock(&net->packet.sklist_lock);
3200 	sk_del_node_init_rcu(sk);
3201 	mutex_unlock(&net->packet.sklist_lock);
3202 
3203 	sock_prot_inuse_add(net, sk->sk_prot, -1);
3204 
3205 	spin_lock(&po->bind_lock);
3206 	unregister_prot_hook(sk, false);
3207 	packet_cached_dev_reset(po);
3208 
3209 	if (po->prot_hook.dev) {
3210 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3211 		po->prot_hook.dev = NULL;
3212 	}
3213 	spin_unlock(&po->bind_lock);
3214 
3215 	packet_flush_mclist(sk);
3216 
3217 	lock_sock(sk);
3218 	if (po->rx_ring.pg_vec) {
3219 		memset(&req_u, 0, sizeof(req_u));
3220 		packet_set_ring(sk, &req_u, 1, 0);
3221 	}
3222 
3223 	if (po->tx_ring.pg_vec) {
3224 		memset(&req_u, 0, sizeof(req_u));
3225 		packet_set_ring(sk, &req_u, 1, 1);
3226 	}
3227 	release_sock(sk);
3228 
3229 	f = fanout_release(sk);
3230 
3231 	synchronize_net();
3232 
3233 	kfree(po->rollover);
3234 	if (f) {
3235 		fanout_release_data(f);
3236 		kvfree(f);
3237 	}
3238 	/*
3239 	 *	Now the socket is dead. No more input will appear.
3240 	 */
3241 	sock_orphan(sk);
3242 	sock->sk = NULL;
3243 
3244 	/* Purge queues */
3245 
3246 	skb_queue_purge(&sk->sk_receive_queue);
3247 	packet_free_pending(po);
3248 
3249 	sock_put(sk);
3250 	return 0;
3251 }
3252 
3253 /*
3254  *	Attach a packet hook.
3255  */
3256 
3257 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3258 			  __be16 proto)
3259 {
3260 	struct packet_sock *po = pkt_sk(sk);
3261 	struct net_device *dev = NULL;
3262 	bool unlisted = false;
3263 	bool need_rehook;
3264 	int ret = 0;
3265 
3266 	lock_sock(sk);
3267 	spin_lock(&po->bind_lock);
3268 	if (!proto)
3269 		proto = po->num;
3270 
3271 	rcu_read_lock();
3272 
3273 	if (po->fanout) {
3274 		ret = -EINVAL;
3275 		goto out_unlock;
3276 	}
3277 
3278 	if (name) {
3279 		dev = dev_get_by_name_rcu(sock_net(sk), name);
3280 		if (!dev) {
3281 			ret = -ENODEV;
3282 			goto out_unlock;
3283 		}
3284 	} else if (ifindex) {
3285 		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3286 		if (!dev) {
3287 			ret = -ENODEV;
3288 			goto out_unlock;
3289 		}
3290 	}
3291 
3292 	need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3293 
3294 	if (need_rehook) {
3295 		dev_hold(dev);
3296 		if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3297 			rcu_read_unlock();
3298 			/* prevents packet_notifier() from calling
3299 			 * register_prot_hook()
3300 			 */
3301 			WRITE_ONCE(po->num, 0);
3302 			__unregister_prot_hook(sk, true);
3303 			rcu_read_lock();
3304 			if (dev)
3305 				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3306 								 dev->ifindex);
3307 		}
3308 
3309 		BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3310 		WRITE_ONCE(po->num, proto);
3311 		po->prot_hook.type = proto;
3312 
3313 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3314 
3315 		if (unlikely(unlisted)) {
3316 			po->prot_hook.dev = NULL;
3317 			WRITE_ONCE(po->ifindex, -1);
3318 			packet_cached_dev_reset(po);
3319 		} else {
3320 			netdev_hold(dev, &po->prot_hook.dev_tracker,
3321 				    GFP_ATOMIC);
3322 			po->prot_hook.dev = dev;
3323 			WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3324 			packet_cached_dev_assign(po, dev);
3325 		}
3326 		dev_put(dev);
3327 	}
3328 
3329 	if (proto == 0 || !need_rehook)
3330 		goto out_unlock;
3331 
3332 	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3333 		register_prot_hook(sk);
3334 	} else {
3335 		sk->sk_err = ENETDOWN;
3336 		if (!sock_flag(sk, SOCK_DEAD))
3337 			sk_error_report(sk);
3338 	}
3339 
3340 out_unlock:
3341 	rcu_read_unlock();
3342 	spin_unlock(&po->bind_lock);
3343 	release_sock(sk);
3344 	return ret;
3345 }
3346 
3347 /*
3348  *	Bind a packet socket to a device
3349  */
3350 
3351 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3352 			    int addr_len)
3353 {
3354 	struct sock *sk = sock->sk;
3355 	char name[sizeof(uaddr->sa_data_min) + 1];
3356 
3357 	/*
3358 	 *	Check legality
3359 	 */
3360 
3361 	if (addr_len != sizeof(struct sockaddr))
3362 		return -EINVAL;
3363 	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3364 	 * zero-terminated.
3365 	 */
3366 	memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3367 	name[sizeof(uaddr->sa_data_min)] = 0;
3368 
3369 	return packet_do_bind(sk, name, 0, 0);
3370 }
3371 
3372 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3373 {
3374 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3375 	struct sock *sk = sock->sk;
3376 
3377 	/*
3378 	 *	Check legality
3379 	 */
3380 
3381 	if (addr_len < sizeof(struct sockaddr_ll))
3382 		return -EINVAL;
3383 	if (sll->sll_family != AF_PACKET)
3384 		return -EINVAL;
3385 
3386 	return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3387 }
3388 
3389 static struct proto packet_proto = {
3390 	.name	  = "PACKET",
3391 	.owner	  = THIS_MODULE,
3392 	.obj_size = sizeof(struct packet_sock),
3393 };
3394 
3395 /*
3396  *	Create a packet of type SOCK_PACKET.
3397  */
3398 
3399 static int packet_create(struct net *net, struct socket *sock, int protocol,
3400 			 int kern)
3401 {
3402 	struct sock *sk;
3403 	struct packet_sock *po;
3404 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3405 	int err;
3406 
3407 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3408 		return -EPERM;
3409 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3410 	    sock->type != SOCK_PACKET)
3411 		return -ESOCKTNOSUPPORT;
3412 
3413 	sock->state = SS_UNCONNECTED;
3414 
3415 	err = -ENOBUFS;
3416 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3417 	if (sk == NULL)
3418 		goto out;
3419 
3420 	sock->ops = &packet_ops;
3421 	if (sock->type == SOCK_PACKET)
3422 		sock->ops = &packet_ops_spkt;
3423 
3424 	sock_init_data(sock, sk);
3425 
3426 	po = pkt_sk(sk);
3427 	init_completion(&po->skb_completion);
3428 	sk->sk_family = PF_PACKET;
3429 	po->num = proto;
3430 
3431 	err = packet_alloc_pending(po);
3432 	if (err)
3433 		goto out2;
3434 
3435 	packet_cached_dev_reset(po);
3436 
3437 	sk->sk_destruct = packet_sock_destruct;
3438 
3439 	/*
3440 	 *	Attach a protocol block
3441 	 */
3442 
3443 	spin_lock_init(&po->bind_lock);
3444 	mutex_init(&po->pg_vec_lock);
3445 	po->rollover = NULL;
3446 	po->prot_hook.func = packet_rcv;
3447 
3448 	if (sock->type == SOCK_PACKET)
3449 		po->prot_hook.func = packet_rcv_spkt;
3450 
3451 	po->prot_hook.af_packet_priv = sk;
3452 	po->prot_hook.af_packet_net = sock_net(sk);
3453 
3454 	if (proto) {
3455 		po->prot_hook.type = proto;
3456 		__register_prot_hook(sk);
3457 	}
3458 
3459 	mutex_lock(&net->packet.sklist_lock);
3460 	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3461 	mutex_unlock(&net->packet.sklist_lock);
3462 
3463 	sock_prot_inuse_add(net, &packet_proto, 1);
3464 
3465 	return 0;
3466 out2:
3467 	sk_free(sk);
3468 out:
3469 	return err;
3470 }
3471 
3472 /*
3473  *	Pull a packet from our receive queue and hand it to the user.
3474  *	If necessary we block.
3475  */
3476 
3477 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3478 			  int flags)
3479 {
3480 	struct sock *sk = sock->sk;
3481 	struct sk_buff *skb;
3482 	int copied, err;
3483 	int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3484 	unsigned int origlen = 0;
3485 
3486 	err = -EINVAL;
3487 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3488 		goto out;
3489 
3490 #if 0
3491 	/* What error should we return now? EUNATTACH? */
3492 	if (pkt_sk(sk)->ifindex < 0)
3493 		return -ENODEV;
3494 #endif
3495 
3496 	if (flags & MSG_ERRQUEUE) {
3497 		err = sock_recv_errqueue(sk, msg, len,
3498 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3499 		goto out;
3500 	}
3501 
3502 	/*
3503 	 *	Call the generic datagram receiver. This handles all sorts
3504 	 *	of horrible races and re-entrancy so we can forget about it
3505 	 *	in the protocol layers.
3506 	 *
3507 	 *	Now it will return ENETDOWN, if device have just gone down,
3508 	 *	but then it will block.
3509 	 */
3510 
3511 	skb = skb_recv_datagram(sk, flags, &err);
3512 
3513 	/*
3514 	 *	An error occurred so return it. Because skb_recv_datagram()
3515 	 *	handles the blocking we don't see and worry about blocking
3516 	 *	retries.
3517 	 */
3518 
3519 	if (skb == NULL)
3520 		goto out;
3521 
3522 	packet_rcv_try_clear_pressure(pkt_sk(sk));
3523 
3524 	if (vnet_hdr_len) {
3525 		err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3526 		if (err)
3527 			goto out_free;
3528 	}
3529 
3530 	/* You lose any data beyond the buffer you gave. If it worries
3531 	 * a user program they can ask the device for its MTU
3532 	 * anyway.
3533 	 */
3534 	copied = skb->len;
3535 	if (copied > len) {
3536 		copied = len;
3537 		msg->msg_flags |= MSG_TRUNC;
3538 	}
3539 
3540 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3541 	if (err)
3542 		goto out_free;
3543 
3544 	if (sock->type != SOCK_PACKET) {
3545 		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3546 
3547 		/* Original length was stored in sockaddr_ll fields */
3548 		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3549 		sll->sll_family = AF_PACKET;
3550 		sll->sll_protocol = (sock->type == SOCK_DGRAM) ?
3551 			vlan_get_protocol_dgram(skb) : skb->protocol;
3552 	}
3553 
3554 	sock_recv_cmsgs(msg, sk, skb);
3555 
3556 	if (msg->msg_name) {
3557 		const size_t max_len = min(sizeof(skb->cb),
3558 					   sizeof(struct sockaddr_storage));
3559 		int copy_len;
3560 
3561 		/* If the address length field is there to be filled
3562 		 * in, we fill it in now.
3563 		 */
3564 		if (sock->type == SOCK_PACKET) {
3565 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3566 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3567 			copy_len = msg->msg_namelen;
3568 		} else {
3569 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3570 
3571 			msg->msg_namelen = sll->sll_halen +
3572 				offsetof(struct sockaddr_ll, sll_addr);
3573 			copy_len = msg->msg_namelen;
3574 			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3575 				memset(msg->msg_name +
3576 				       offsetof(struct sockaddr_ll, sll_addr),
3577 				       0, sizeof(sll->sll_addr));
3578 				msg->msg_namelen = sizeof(struct sockaddr_ll);
3579 			}
3580 		}
3581 		if (WARN_ON_ONCE(copy_len > max_len)) {
3582 			copy_len = max_len;
3583 			msg->msg_namelen = copy_len;
3584 		}
3585 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3586 	}
3587 
3588 	if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3589 		struct tpacket_auxdata aux;
3590 
3591 		aux.tp_status = TP_STATUS_USER;
3592 		if (skb->ip_summed == CHECKSUM_PARTIAL)
3593 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3594 		else if (skb->pkt_type != PACKET_OUTGOING &&
3595 			 skb_csum_unnecessary(skb))
3596 			aux.tp_status |= TP_STATUS_CSUM_VALID;
3597 		if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3598 			aux.tp_status |= TP_STATUS_GSO_TCP;
3599 
3600 		aux.tp_len = origlen;
3601 		aux.tp_snaplen = skb->len;
3602 		aux.tp_mac = 0;
3603 		aux.tp_net = skb_network_offset(skb);
3604 		if (skb_vlan_tag_present(skb)) {
3605 			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3606 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3607 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3608 		} else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
3609 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3610 			struct net_device *dev;
3611 
3612 			rcu_read_lock();
3613 			dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex);
3614 			if (dev) {
3615 				aux.tp_vlan_tci = vlan_get_tci(skb, dev);
3616 				aux.tp_vlan_tpid = ntohs(skb->protocol);
3617 				aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3618 			} else {
3619 				aux.tp_vlan_tci = 0;
3620 				aux.tp_vlan_tpid = 0;
3621 			}
3622 			rcu_read_unlock();
3623 		} else {
3624 			aux.tp_vlan_tci = 0;
3625 			aux.tp_vlan_tpid = 0;
3626 		}
3627 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3628 	}
3629 
3630 	/*
3631 	 *	Free or return the buffer as appropriate. Again this
3632 	 *	hides all the races and re-entrancy issues from us.
3633 	 */
3634 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3635 
3636 out_free:
3637 	skb_free_datagram(sk, skb);
3638 out:
3639 	return err;
3640 }
3641 
3642 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3643 			       int peer)
3644 {
3645 	struct net_device *dev;
3646 	struct sock *sk	= sock->sk;
3647 
3648 	if (peer)
3649 		return -EOPNOTSUPP;
3650 
3651 	uaddr->sa_family = AF_PACKET;
3652 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3653 	rcu_read_lock();
3654 	dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3655 	if (dev)
3656 		strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3657 	rcu_read_unlock();
3658 
3659 	return sizeof(*uaddr);
3660 }
3661 
3662 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3663 			  int peer)
3664 {
3665 	struct net_device *dev;
3666 	struct sock *sk = sock->sk;
3667 	struct packet_sock *po = pkt_sk(sk);
3668 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3669 	int ifindex;
3670 
3671 	if (peer)
3672 		return -EOPNOTSUPP;
3673 
3674 	ifindex = READ_ONCE(po->ifindex);
3675 	sll->sll_family = AF_PACKET;
3676 	sll->sll_ifindex = ifindex;
3677 	sll->sll_protocol = READ_ONCE(po->num);
3678 	sll->sll_pkttype = 0;
3679 	rcu_read_lock();
3680 	dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3681 	if (dev) {
3682 		sll->sll_hatype = dev->type;
3683 		sll->sll_halen = dev->addr_len;
3684 
3685 		/* Let __fortify_memcpy_chk() know the actual buffer size. */
3686 		memcpy(((struct sockaddr_storage *)sll)->__data +
3687 		       offsetof(struct sockaddr_ll, sll_addr) -
3688 		       offsetofend(struct sockaddr_ll, sll_family),
3689 		       dev->dev_addr, dev->addr_len);
3690 	} else {
3691 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3692 		sll->sll_halen = 0;
3693 	}
3694 	rcu_read_unlock();
3695 
3696 	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3697 }
3698 
3699 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3700 			 int what)
3701 {
3702 	switch (i->type) {
3703 	case PACKET_MR_MULTICAST:
3704 		if (i->alen != dev->addr_len)
3705 			return -EINVAL;
3706 		if (what > 0)
3707 			return dev_mc_add(dev, i->addr);
3708 		else
3709 			return dev_mc_del(dev, i->addr);
3710 		break;
3711 	case PACKET_MR_PROMISC:
3712 		return dev_set_promiscuity(dev, what);
3713 	case PACKET_MR_ALLMULTI:
3714 		return dev_set_allmulti(dev, what);
3715 	case PACKET_MR_UNICAST:
3716 		if (i->alen != dev->addr_len)
3717 			return -EINVAL;
3718 		if (what > 0)
3719 			return dev_uc_add(dev, i->addr);
3720 		else
3721 			return dev_uc_del(dev, i->addr);
3722 		break;
3723 	default:
3724 		break;
3725 	}
3726 	return 0;
3727 }
3728 
3729 static void packet_dev_mclist_delete(struct net_device *dev,
3730 				     struct packet_mclist **mlp)
3731 {
3732 	struct packet_mclist *ml;
3733 
3734 	while ((ml = *mlp) != NULL) {
3735 		if (ml->ifindex == dev->ifindex) {
3736 			packet_dev_mc(dev, ml, -1);
3737 			*mlp = ml->next;
3738 			kfree(ml);
3739 		} else
3740 			mlp = &ml->next;
3741 	}
3742 }
3743 
3744 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3745 {
3746 	struct packet_sock *po = pkt_sk(sk);
3747 	struct packet_mclist *ml, *i;
3748 	struct net_device *dev;
3749 	int err;
3750 
3751 	rtnl_lock();
3752 
3753 	err = -ENODEV;
3754 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3755 	if (!dev)
3756 		goto done;
3757 
3758 	err = -EINVAL;
3759 	if (mreq->mr_alen > dev->addr_len)
3760 		goto done;
3761 
3762 	err = -ENOBUFS;
3763 	i = kmalloc(sizeof(*i), GFP_KERNEL);
3764 	if (i == NULL)
3765 		goto done;
3766 
3767 	err = 0;
3768 	for (ml = po->mclist; ml; ml = ml->next) {
3769 		if (ml->ifindex == mreq->mr_ifindex &&
3770 		    ml->type == mreq->mr_type &&
3771 		    ml->alen == mreq->mr_alen &&
3772 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3773 			ml->count++;
3774 			/* Free the new element ... */
3775 			kfree(i);
3776 			goto done;
3777 		}
3778 	}
3779 
3780 	i->type = mreq->mr_type;
3781 	i->ifindex = mreq->mr_ifindex;
3782 	i->alen = mreq->mr_alen;
3783 	memcpy(i->addr, mreq->mr_address, i->alen);
3784 	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3785 	i->count = 1;
3786 	i->next = po->mclist;
3787 	po->mclist = i;
3788 	err = packet_dev_mc(dev, i, 1);
3789 	if (err) {
3790 		po->mclist = i->next;
3791 		kfree(i);
3792 	}
3793 
3794 done:
3795 	rtnl_unlock();
3796 	return err;
3797 }
3798 
3799 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3800 {
3801 	struct packet_mclist *ml, **mlp;
3802 
3803 	rtnl_lock();
3804 
3805 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3806 		if (ml->ifindex == mreq->mr_ifindex &&
3807 		    ml->type == mreq->mr_type &&
3808 		    ml->alen == mreq->mr_alen &&
3809 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3810 			if (--ml->count == 0) {
3811 				struct net_device *dev;
3812 				*mlp = ml->next;
3813 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3814 				if (dev)
3815 					packet_dev_mc(dev, ml, -1);
3816 				kfree(ml);
3817 			}
3818 			break;
3819 		}
3820 	}
3821 	rtnl_unlock();
3822 	return 0;
3823 }
3824 
3825 static void packet_flush_mclist(struct sock *sk)
3826 {
3827 	struct packet_sock *po = pkt_sk(sk);
3828 	struct packet_mclist *ml;
3829 
3830 	if (!po->mclist)
3831 		return;
3832 
3833 	rtnl_lock();
3834 	while ((ml = po->mclist) != NULL) {
3835 		struct net_device *dev;
3836 
3837 		po->mclist = ml->next;
3838 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3839 		if (dev != NULL)
3840 			packet_dev_mc(dev, ml, -1);
3841 		kfree(ml);
3842 	}
3843 	rtnl_unlock();
3844 }
3845 
3846 static int
3847 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3848 		  unsigned int optlen)
3849 {
3850 	struct sock *sk = sock->sk;
3851 	struct packet_sock *po = pkt_sk(sk);
3852 	int ret;
3853 
3854 	if (level != SOL_PACKET)
3855 		return -ENOPROTOOPT;
3856 
3857 	switch (optname) {
3858 	case PACKET_ADD_MEMBERSHIP:
3859 	case PACKET_DROP_MEMBERSHIP:
3860 	{
3861 		struct packet_mreq_max mreq;
3862 		int len = optlen;
3863 		memset(&mreq, 0, sizeof(mreq));
3864 		if (len < sizeof(struct packet_mreq))
3865 			return -EINVAL;
3866 		if (len > sizeof(mreq))
3867 			len = sizeof(mreq);
3868 		if (copy_from_sockptr(&mreq, optval, len))
3869 			return -EFAULT;
3870 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3871 			return -EINVAL;
3872 		if (optname == PACKET_ADD_MEMBERSHIP)
3873 			ret = packet_mc_add(sk, &mreq);
3874 		else
3875 			ret = packet_mc_drop(sk, &mreq);
3876 		return ret;
3877 	}
3878 
3879 	case PACKET_RX_RING:
3880 	case PACKET_TX_RING:
3881 	{
3882 		union tpacket_req_u req_u;
3883 
3884 		ret = -EINVAL;
3885 		lock_sock(sk);
3886 		switch (po->tp_version) {
3887 		case TPACKET_V1:
3888 		case TPACKET_V2:
3889 			if (optlen < sizeof(req_u.req))
3890 				break;
3891 			ret = copy_from_sockptr(&req_u.req, optval,
3892 						sizeof(req_u.req)) ?
3893 						-EINVAL : 0;
3894 			break;
3895 		case TPACKET_V3:
3896 		default:
3897 			if (optlen < sizeof(req_u.req3))
3898 				break;
3899 			ret = copy_from_sockptr(&req_u.req3, optval,
3900 						sizeof(req_u.req3)) ?
3901 						-EINVAL : 0;
3902 			break;
3903 		}
3904 		if (!ret)
3905 			ret = packet_set_ring(sk, &req_u, 0,
3906 					      optname == PACKET_TX_RING);
3907 		release_sock(sk);
3908 		return ret;
3909 	}
3910 	case PACKET_COPY_THRESH:
3911 	{
3912 		int val;
3913 
3914 		if (optlen != sizeof(val))
3915 			return -EINVAL;
3916 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3917 			return -EFAULT;
3918 
3919 		WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3920 		return 0;
3921 	}
3922 	case PACKET_VERSION:
3923 	{
3924 		int val;
3925 
3926 		if (optlen != sizeof(val))
3927 			return -EINVAL;
3928 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3929 			return -EFAULT;
3930 		switch (val) {
3931 		case TPACKET_V1:
3932 		case TPACKET_V2:
3933 		case TPACKET_V3:
3934 			break;
3935 		default:
3936 			return -EINVAL;
3937 		}
3938 		lock_sock(sk);
3939 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3940 			ret = -EBUSY;
3941 		} else {
3942 			po->tp_version = val;
3943 			ret = 0;
3944 		}
3945 		release_sock(sk);
3946 		return ret;
3947 	}
3948 	case PACKET_RESERVE:
3949 	{
3950 		unsigned int val;
3951 
3952 		if (optlen != sizeof(val))
3953 			return -EINVAL;
3954 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3955 			return -EFAULT;
3956 		if (val > INT_MAX)
3957 			return -EINVAL;
3958 		lock_sock(sk);
3959 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3960 			ret = -EBUSY;
3961 		} else {
3962 			po->tp_reserve = val;
3963 			ret = 0;
3964 		}
3965 		release_sock(sk);
3966 		return ret;
3967 	}
3968 	case PACKET_LOSS:
3969 	{
3970 		unsigned int val;
3971 
3972 		if (optlen != sizeof(val))
3973 			return -EINVAL;
3974 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3975 			return -EFAULT;
3976 
3977 		lock_sock(sk);
3978 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3979 			ret = -EBUSY;
3980 		} else {
3981 			packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3982 			ret = 0;
3983 		}
3984 		release_sock(sk);
3985 		return ret;
3986 	}
3987 	case PACKET_AUXDATA:
3988 	{
3989 		int val;
3990 
3991 		if (optlen < sizeof(val))
3992 			return -EINVAL;
3993 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3994 			return -EFAULT;
3995 
3996 		packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3997 		return 0;
3998 	}
3999 	case PACKET_ORIGDEV:
4000 	{
4001 		int val;
4002 
4003 		if (optlen < sizeof(val))
4004 			return -EINVAL;
4005 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4006 			return -EFAULT;
4007 
4008 		packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
4009 		return 0;
4010 	}
4011 	case PACKET_VNET_HDR:
4012 	case PACKET_VNET_HDR_SZ:
4013 	{
4014 		int val, hdr_len;
4015 
4016 		if (sock->type != SOCK_RAW)
4017 			return -EINVAL;
4018 		if (optlen < sizeof(val))
4019 			return -EINVAL;
4020 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4021 			return -EFAULT;
4022 
4023 		if (optname == PACKET_VNET_HDR_SZ) {
4024 			if (val && val != sizeof(struct virtio_net_hdr) &&
4025 			    val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
4026 				return -EINVAL;
4027 			hdr_len = val;
4028 		} else {
4029 			hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
4030 		}
4031 		lock_sock(sk);
4032 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
4033 			ret = -EBUSY;
4034 		} else {
4035 			WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
4036 			ret = 0;
4037 		}
4038 		release_sock(sk);
4039 		return ret;
4040 	}
4041 	case PACKET_TIMESTAMP:
4042 	{
4043 		int val;
4044 
4045 		if (optlen != sizeof(val))
4046 			return -EINVAL;
4047 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4048 			return -EFAULT;
4049 
4050 		WRITE_ONCE(po->tp_tstamp, val);
4051 		return 0;
4052 	}
4053 	case PACKET_FANOUT:
4054 	{
4055 		struct fanout_args args = { 0 };
4056 
4057 		if (optlen != sizeof(int) && optlen != sizeof(args))
4058 			return -EINVAL;
4059 		if (copy_from_sockptr(&args, optval, optlen))
4060 			return -EFAULT;
4061 
4062 		return fanout_add(sk, &args);
4063 	}
4064 	case PACKET_FANOUT_DATA:
4065 	{
4066 		/* Paired with the WRITE_ONCE() in fanout_add() */
4067 		if (!READ_ONCE(po->fanout))
4068 			return -EINVAL;
4069 
4070 		return fanout_set_data(po, optval, optlen);
4071 	}
4072 	case PACKET_IGNORE_OUTGOING:
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 		if (val < 0 || val > 1)
4081 			return -EINVAL;
4082 
4083 		WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4084 		return 0;
4085 	}
4086 	case PACKET_TX_HAS_OFF:
4087 	{
4088 		unsigned int val;
4089 
4090 		if (optlen != sizeof(val))
4091 			return -EINVAL;
4092 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4093 			return -EFAULT;
4094 
4095 		lock_sock(sk);
4096 		if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4097 			packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4098 
4099 		release_sock(sk);
4100 		return 0;
4101 	}
4102 	case PACKET_QDISC_BYPASS:
4103 	{
4104 		int val;
4105 
4106 		if (optlen != sizeof(val))
4107 			return -EINVAL;
4108 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4109 			return -EFAULT;
4110 
4111 		packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4112 		return 0;
4113 	}
4114 	default:
4115 		return -ENOPROTOOPT;
4116 	}
4117 }
4118 
4119 static int packet_getsockopt(struct socket *sock, int level, int optname,
4120 			     char __user *optval, int __user *optlen)
4121 {
4122 	int len;
4123 	int val, lv = sizeof(val);
4124 	struct sock *sk = sock->sk;
4125 	struct packet_sock *po = pkt_sk(sk);
4126 	void *data = &val;
4127 	union tpacket_stats_u st;
4128 	struct tpacket_rollover_stats rstats;
4129 	int drops;
4130 
4131 	if (level != SOL_PACKET)
4132 		return -ENOPROTOOPT;
4133 
4134 	if (get_user(len, optlen))
4135 		return -EFAULT;
4136 
4137 	if (len < 0)
4138 		return -EINVAL;
4139 
4140 	switch (optname) {
4141 	case PACKET_STATISTICS:
4142 		spin_lock_bh(&sk->sk_receive_queue.lock);
4143 		memcpy(&st, &po->stats, sizeof(st));
4144 		memset(&po->stats, 0, sizeof(po->stats));
4145 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4146 		drops = atomic_xchg(&po->tp_drops, 0);
4147 
4148 		if (po->tp_version == TPACKET_V3) {
4149 			lv = sizeof(struct tpacket_stats_v3);
4150 			st.stats3.tp_drops = drops;
4151 			st.stats3.tp_packets += drops;
4152 			data = &st.stats3;
4153 		} else {
4154 			lv = sizeof(struct tpacket_stats);
4155 			st.stats1.tp_drops = drops;
4156 			st.stats1.tp_packets += drops;
4157 			data = &st.stats1;
4158 		}
4159 
4160 		break;
4161 	case PACKET_AUXDATA:
4162 		val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4163 		break;
4164 	case PACKET_ORIGDEV:
4165 		val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4166 		break;
4167 	case PACKET_VNET_HDR:
4168 		val = !!READ_ONCE(po->vnet_hdr_sz);
4169 		break;
4170 	case PACKET_VNET_HDR_SZ:
4171 		val = READ_ONCE(po->vnet_hdr_sz);
4172 		break;
4173 	case PACKET_COPY_THRESH:
4174 		val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4175 		break;
4176 	case PACKET_VERSION:
4177 		val = po->tp_version;
4178 		break;
4179 	case PACKET_HDRLEN:
4180 		if (len > sizeof(int))
4181 			len = sizeof(int);
4182 		if (len < sizeof(int))
4183 			return -EINVAL;
4184 		if (copy_from_user(&val, optval, len))
4185 			return -EFAULT;
4186 		switch (val) {
4187 		case TPACKET_V1:
4188 			val = sizeof(struct tpacket_hdr);
4189 			break;
4190 		case TPACKET_V2:
4191 			val = sizeof(struct tpacket2_hdr);
4192 			break;
4193 		case TPACKET_V3:
4194 			val = sizeof(struct tpacket3_hdr);
4195 			break;
4196 		default:
4197 			return -EINVAL;
4198 		}
4199 		break;
4200 	case PACKET_RESERVE:
4201 		val = po->tp_reserve;
4202 		break;
4203 	case PACKET_LOSS:
4204 		val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4205 		break;
4206 	case PACKET_TIMESTAMP:
4207 		val = READ_ONCE(po->tp_tstamp);
4208 		break;
4209 	case PACKET_FANOUT:
4210 		val = (po->fanout ?
4211 		       ((u32)po->fanout->id |
4212 			((u32)po->fanout->type << 16) |
4213 			((u32)po->fanout->flags << 24)) :
4214 		       0);
4215 		break;
4216 	case PACKET_IGNORE_OUTGOING:
4217 		val = READ_ONCE(po->prot_hook.ignore_outgoing);
4218 		break;
4219 	case PACKET_ROLLOVER_STATS:
4220 		if (!po->rollover)
4221 			return -EINVAL;
4222 		rstats.tp_all = atomic_long_read(&po->rollover->num);
4223 		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4224 		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4225 		data = &rstats;
4226 		lv = sizeof(rstats);
4227 		break;
4228 	case PACKET_TX_HAS_OFF:
4229 		val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4230 		break;
4231 	case PACKET_QDISC_BYPASS:
4232 		val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4233 		break;
4234 	default:
4235 		return -ENOPROTOOPT;
4236 	}
4237 
4238 	if (len > lv)
4239 		len = lv;
4240 	if (put_user(len, optlen))
4241 		return -EFAULT;
4242 	if (copy_to_user(optval, data, len))
4243 		return -EFAULT;
4244 	return 0;
4245 }
4246 
4247 static int packet_notifier(struct notifier_block *this,
4248 			   unsigned long msg, void *ptr)
4249 {
4250 	struct sock *sk;
4251 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4252 	struct net *net = dev_net(dev);
4253 
4254 	rcu_read_lock();
4255 	sk_for_each_rcu(sk, &net->packet.sklist) {
4256 		struct packet_sock *po = pkt_sk(sk);
4257 
4258 		switch (msg) {
4259 		case NETDEV_UNREGISTER:
4260 			if (po->mclist)
4261 				packet_dev_mclist_delete(dev, &po->mclist);
4262 			fallthrough;
4263 
4264 		case NETDEV_DOWN:
4265 			if (dev->ifindex == po->ifindex) {
4266 				spin_lock(&po->bind_lock);
4267 				if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4268 					__unregister_prot_hook(sk, false);
4269 					sk->sk_err = ENETDOWN;
4270 					if (!sock_flag(sk, SOCK_DEAD))
4271 						sk_error_report(sk);
4272 				}
4273 				if (msg == NETDEV_UNREGISTER) {
4274 					packet_cached_dev_reset(po);
4275 					WRITE_ONCE(po->ifindex, -1);
4276 					netdev_put(po->prot_hook.dev,
4277 						   &po->prot_hook.dev_tracker);
4278 					po->prot_hook.dev = NULL;
4279 				}
4280 				spin_unlock(&po->bind_lock);
4281 			}
4282 			break;
4283 		case NETDEV_UP:
4284 			if (dev->ifindex == po->ifindex) {
4285 				spin_lock(&po->bind_lock);
4286 				if (po->num)
4287 					register_prot_hook(sk);
4288 				spin_unlock(&po->bind_lock);
4289 			}
4290 			break;
4291 		}
4292 	}
4293 	rcu_read_unlock();
4294 	return NOTIFY_DONE;
4295 }
4296 
4297 
4298 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4299 			unsigned long arg)
4300 {
4301 	struct sock *sk = sock->sk;
4302 
4303 	switch (cmd) {
4304 	case SIOCOUTQ:
4305 	{
4306 		int amount = sk_wmem_alloc_get(sk);
4307 
4308 		return put_user(amount, (int __user *)arg);
4309 	}
4310 	case SIOCINQ:
4311 	{
4312 		struct sk_buff *skb;
4313 		int amount = 0;
4314 
4315 		spin_lock_bh(&sk->sk_receive_queue.lock);
4316 		skb = skb_peek(&sk->sk_receive_queue);
4317 		if (skb)
4318 			amount = skb->len;
4319 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4320 		return put_user(amount, (int __user *)arg);
4321 	}
4322 #ifdef CONFIG_INET
4323 	case SIOCADDRT:
4324 	case SIOCDELRT:
4325 	case SIOCDARP:
4326 	case SIOCGARP:
4327 	case SIOCSARP:
4328 	case SIOCGIFADDR:
4329 	case SIOCSIFADDR:
4330 	case SIOCGIFBRDADDR:
4331 	case SIOCSIFBRDADDR:
4332 	case SIOCGIFNETMASK:
4333 	case SIOCSIFNETMASK:
4334 	case SIOCGIFDSTADDR:
4335 	case SIOCSIFDSTADDR:
4336 	case SIOCSIFFLAGS:
4337 		return inet_dgram_ops.ioctl(sock, cmd, arg);
4338 #endif
4339 
4340 	default:
4341 		return -ENOIOCTLCMD;
4342 	}
4343 	return 0;
4344 }
4345 
4346 static __poll_t packet_poll(struct file *file, struct socket *sock,
4347 				poll_table *wait)
4348 {
4349 	struct sock *sk = sock->sk;
4350 	struct packet_sock *po = pkt_sk(sk);
4351 	__poll_t mask = datagram_poll(file, sock, wait);
4352 
4353 	spin_lock_bh(&sk->sk_receive_queue.lock);
4354 	if (po->rx_ring.pg_vec) {
4355 		if (!packet_previous_rx_frame(po, &po->rx_ring,
4356 			TP_STATUS_KERNEL))
4357 			mask |= EPOLLIN | EPOLLRDNORM;
4358 	}
4359 	packet_rcv_try_clear_pressure(po);
4360 	spin_unlock_bh(&sk->sk_receive_queue.lock);
4361 	spin_lock_bh(&sk->sk_write_queue.lock);
4362 	if (po->tx_ring.pg_vec) {
4363 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4364 			mask |= EPOLLOUT | EPOLLWRNORM;
4365 	}
4366 	spin_unlock_bh(&sk->sk_write_queue.lock);
4367 	return mask;
4368 }
4369 
4370 
4371 /* Dirty? Well, I still did not learn better way to account
4372  * for user mmaps.
4373  */
4374 
4375 static void packet_mm_open(struct vm_area_struct *vma)
4376 {
4377 	struct file *file = vma->vm_file;
4378 	struct socket *sock = file->private_data;
4379 	struct sock *sk = sock->sk;
4380 
4381 	if (sk)
4382 		atomic_long_inc(&pkt_sk(sk)->mapped);
4383 }
4384 
4385 static void packet_mm_close(struct vm_area_struct *vma)
4386 {
4387 	struct file *file = vma->vm_file;
4388 	struct socket *sock = file->private_data;
4389 	struct sock *sk = sock->sk;
4390 
4391 	if (sk)
4392 		atomic_long_dec(&pkt_sk(sk)->mapped);
4393 }
4394 
4395 static const struct vm_operations_struct packet_mmap_ops = {
4396 	.open	=	packet_mm_open,
4397 	.close	=	packet_mm_close,
4398 };
4399 
4400 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4401 			unsigned int len)
4402 {
4403 	int i;
4404 
4405 	for (i = 0; i < len; i++) {
4406 		if (likely(pg_vec[i].buffer)) {
4407 			if (is_vmalloc_addr(pg_vec[i].buffer))
4408 				vfree(pg_vec[i].buffer);
4409 			else
4410 				free_pages((unsigned long)pg_vec[i].buffer,
4411 					   order);
4412 			pg_vec[i].buffer = NULL;
4413 		}
4414 	}
4415 	kfree(pg_vec);
4416 }
4417 
4418 static char *alloc_one_pg_vec_page(unsigned long order)
4419 {
4420 	char *buffer;
4421 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4422 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4423 
4424 	buffer = (char *) __get_free_pages(gfp_flags, order);
4425 	if (buffer)
4426 		return buffer;
4427 
4428 	/* __get_free_pages failed, fall back to vmalloc */
4429 	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4430 	if (buffer)
4431 		return buffer;
4432 
4433 	/* vmalloc failed, lets dig into swap here */
4434 	gfp_flags &= ~__GFP_NORETRY;
4435 	buffer = (char *) __get_free_pages(gfp_flags, order);
4436 	if (buffer)
4437 		return buffer;
4438 
4439 	/* complete and utter failure */
4440 	return NULL;
4441 }
4442 
4443 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4444 {
4445 	unsigned int block_nr = req->tp_block_nr;
4446 	struct pgv *pg_vec;
4447 	int i;
4448 
4449 	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4450 	if (unlikely(!pg_vec))
4451 		goto out;
4452 
4453 	for (i = 0; i < block_nr; i++) {
4454 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4455 		if (unlikely(!pg_vec[i].buffer))
4456 			goto out_free_pgvec;
4457 	}
4458 
4459 out:
4460 	return pg_vec;
4461 
4462 out_free_pgvec:
4463 	free_pg_vec(pg_vec, order, block_nr);
4464 	pg_vec = NULL;
4465 	goto out;
4466 }
4467 
4468 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4469 		int closing, int tx_ring)
4470 {
4471 	struct pgv *pg_vec = NULL;
4472 	struct packet_sock *po = pkt_sk(sk);
4473 	unsigned long *rx_owner_map = NULL;
4474 	int was_running, order = 0;
4475 	struct packet_ring_buffer *rb;
4476 	struct sk_buff_head *rb_queue;
4477 	__be16 num;
4478 	int err;
4479 	/* Added to avoid minimal code churn */
4480 	struct tpacket_req *req = &req_u->req;
4481 
4482 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4483 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4484 
4485 	err = -EBUSY;
4486 	if (!closing) {
4487 		if (atomic_long_read(&po->mapped))
4488 			goto out;
4489 		if (packet_read_pending(rb))
4490 			goto out;
4491 	}
4492 
4493 	if (req->tp_block_nr) {
4494 		unsigned int min_frame_size;
4495 
4496 		/* Sanity tests and some calculations */
4497 		err = -EBUSY;
4498 		if (unlikely(rb->pg_vec))
4499 			goto out;
4500 
4501 		switch (po->tp_version) {
4502 		case TPACKET_V1:
4503 			po->tp_hdrlen = TPACKET_HDRLEN;
4504 			break;
4505 		case TPACKET_V2:
4506 			po->tp_hdrlen = TPACKET2_HDRLEN;
4507 			break;
4508 		case TPACKET_V3:
4509 			po->tp_hdrlen = TPACKET3_HDRLEN;
4510 			break;
4511 		}
4512 
4513 		err = -EINVAL;
4514 		if (unlikely((int)req->tp_block_size <= 0))
4515 			goto out;
4516 		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4517 			goto out;
4518 		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4519 		if (po->tp_version >= TPACKET_V3 &&
4520 		    req->tp_block_size <
4521 		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4522 			goto out;
4523 		if (unlikely(req->tp_frame_size < min_frame_size))
4524 			goto out;
4525 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4526 			goto out;
4527 
4528 		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4529 		if (unlikely(rb->frames_per_block == 0))
4530 			goto out;
4531 		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4532 			goto out;
4533 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4534 					req->tp_frame_nr))
4535 			goto out;
4536 
4537 		err = -ENOMEM;
4538 		order = get_order(req->tp_block_size);
4539 		pg_vec = alloc_pg_vec(req, order);
4540 		if (unlikely(!pg_vec))
4541 			goto out;
4542 		switch (po->tp_version) {
4543 		case TPACKET_V3:
4544 			/* Block transmit is not supported yet */
4545 			if (!tx_ring) {
4546 				init_prb_bdqc(po, rb, pg_vec, req_u);
4547 			} else {
4548 				struct tpacket_req3 *req3 = &req_u->req3;
4549 
4550 				if (req3->tp_retire_blk_tov ||
4551 				    req3->tp_sizeof_priv ||
4552 				    req3->tp_feature_req_word) {
4553 					err = -EINVAL;
4554 					goto out_free_pg_vec;
4555 				}
4556 			}
4557 			break;
4558 		default:
4559 			if (!tx_ring) {
4560 				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4561 					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4562 				if (!rx_owner_map)
4563 					goto out_free_pg_vec;
4564 			}
4565 			break;
4566 		}
4567 	}
4568 	/* Done */
4569 	else {
4570 		err = -EINVAL;
4571 		if (unlikely(req->tp_frame_nr))
4572 			goto out;
4573 	}
4574 
4575 
4576 	/* Detach socket from network */
4577 	spin_lock(&po->bind_lock);
4578 	was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4579 	num = po->num;
4580 	if (was_running) {
4581 		WRITE_ONCE(po->num, 0);
4582 		__unregister_prot_hook(sk, false);
4583 	}
4584 	spin_unlock(&po->bind_lock);
4585 
4586 	synchronize_net();
4587 
4588 	err = -EBUSY;
4589 	mutex_lock(&po->pg_vec_lock);
4590 	if (closing || atomic_long_read(&po->mapped) == 0) {
4591 		err = 0;
4592 		spin_lock_bh(&rb_queue->lock);
4593 		swap(rb->pg_vec, pg_vec);
4594 		if (po->tp_version <= TPACKET_V2)
4595 			swap(rb->rx_owner_map, rx_owner_map);
4596 		rb->frame_max = (req->tp_frame_nr - 1);
4597 		rb->head = 0;
4598 		rb->frame_size = req->tp_frame_size;
4599 		spin_unlock_bh(&rb_queue->lock);
4600 
4601 		swap(rb->pg_vec_order, order);
4602 		swap(rb->pg_vec_len, req->tp_block_nr);
4603 
4604 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4605 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4606 						tpacket_rcv : packet_rcv;
4607 		skb_queue_purge(rb_queue);
4608 		if (atomic_long_read(&po->mapped))
4609 			pr_err("packet_mmap: vma is busy: %ld\n",
4610 			       atomic_long_read(&po->mapped));
4611 	}
4612 	mutex_unlock(&po->pg_vec_lock);
4613 
4614 	spin_lock(&po->bind_lock);
4615 	if (was_running) {
4616 		WRITE_ONCE(po->num, num);
4617 		register_prot_hook(sk);
4618 	}
4619 	spin_unlock(&po->bind_lock);
4620 	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4621 		/* Because we don't support block-based V3 on tx-ring */
4622 		if (!tx_ring)
4623 			prb_shutdown_retire_blk_timer(po, rb_queue);
4624 	}
4625 
4626 out_free_pg_vec:
4627 	if (pg_vec) {
4628 		bitmap_free(rx_owner_map);
4629 		free_pg_vec(pg_vec, order, req->tp_block_nr);
4630 	}
4631 out:
4632 	return err;
4633 }
4634 
4635 static int packet_mmap(struct file *file, struct socket *sock,
4636 		struct vm_area_struct *vma)
4637 {
4638 	struct sock *sk = sock->sk;
4639 	struct packet_sock *po = pkt_sk(sk);
4640 	unsigned long size, expected_size;
4641 	struct packet_ring_buffer *rb;
4642 	unsigned long start;
4643 	int err = -EINVAL;
4644 	int i;
4645 
4646 	if (vma->vm_pgoff)
4647 		return -EINVAL;
4648 
4649 	mutex_lock(&po->pg_vec_lock);
4650 
4651 	expected_size = 0;
4652 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4653 		if (rb->pg_vec) {
4654 			expected_size += rb->pg_vec_len
4655 						* rb->pg_vec_pages
4656 						* PAGE_SIZE;
4657 		}
4658 	}
4659 
4660 	if (expected_size == 0)
4661 		goto out;
4662 
4663 	size = vma->vm_end - vma->vm_start;
4664 	if (size != expected_size)
4665 		goto out;
4666 
4667 	start = vma->vm_start;
4668 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4669 		if (rb->pg_vec == NULL)
4670 			continue;
4671 
4672 		for (i = 0; i < rb->pg_vec_len; i++) {
4673 			struct page *page;
4674 			void *kaddr = rb->pg_vec[i].buffer;
4675 			int pg_num;
4676 
4677 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4678 				page = pgv_to_page(kaddr);
4679 				err = vm_insert_page(vma, start, page);
4680 				if (unlikely(err))
4681 					goto out;
4682 				start += PAGE_SIZE;
4683 				kaddr += PAGE_SIZE;
4684 			}
4685 		}
4686 	}
4687 
4688 	atomic_long_inc(&po->mapped);
4689 	vma->vm_ops = &packet_mmap_ops;
4690 	err = 0;
4691 
4692 out:
4693 	mutex_unlock(&po->pg_vec_lock);
4694 	return err;
4695 }
4696 
4697 static const struct proto_ops packet_ops_spkt = {
4698 	.family =	PF_PACKET,
4699 	.owner =	THIS_MODULE,
4700 	.release =	packet_release,
4701 	.bind =		packet_bind_spkt,
4702 	.connect =	sock_no_connect,
4703 	.socketpair =	sock_no_socketpair,
4704 	.accept =	sock_no_accept,
4705 	.getname =	packet_getname_spkt,
4706 	.poll =		datagram_poll,
4707 	.ioctl =	packet_ioctl,
4708 	.gettstamp =	sock_gettstamp,
4709 	.listen =	sock_no_listen,
4710 	.shutdown =	sock_no_shutdown,
4711 	.sendmsg =	packet_sendmsg_spkt,
4712 	.recvmsg =	packet_recvmsg,
4713 	.mmap =		sock_no_mmap,
4714 };
4715 
4716 static const struct proto_ops packet_ops = {
4717 	.family =	PF_PACKET,
4718 	.owner =	THIS_MODULE,
4719 	.release =	packet_release,
4720 	.bind =		packet_bind,
4721 	.connect =	sock_no_connect,
4722 	.socketpair =	sock_no_socketpair,
4723 	.accept =	sock_no_accept,
4724 	.getname =	packet_getname,
4725 	.poll =		packet_poll,
4726 	.ioctl =	packet_ioctl,
4727 	.gettstamp =	sock_gettstamp,
4728 	.listen =	sock_no_listen,
4729 	.shutdown =	sock_no_shutdown,
4730 	.setsockopt =	packet_setsockopt,
4731 	.getsockopt =	packet_getsockopt,
4732 	.sendmsg =	packet_sendmsg,
4733 	.recvmsg =	packet_recvmsg,
4734 	.mmap =		packet_mmap,
4735 };
4736 
4737 static const struct net_proto_family packet_family_ops = {
4738 	.family =	PF_PACKET,
4739 	.create =	packet_create,
4740 	.owner	=	THIS_MODULE,
4741 };
4742 
4743 static struct notifier_block packet_netdev_notifier = {
4744 	.notifier_call =	packet_notifier,
4745 };
4746 
4747 #ifdef CONFIG_PROC_FS
4748 
4749 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4750 	__acquires(RCU)
4751 {
4752 	struct net *net = seq_file_net(seq);
4753 
4754 	rcu_read_lock();
4755 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4756 }
4757 
4758 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4759 {
4760 	struct net *net = seq_file_net(seq);
4761 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4762 }
4763 
4764 static void packet_seq_stop(struct seq_file *seq, void *v)
4765 	__releases(RCU)
4766 {
4767 	rcu_read_unlock();
4768 }
4769 
4770 static int packet_seq_show(struct seq_file *seq, void *v)
4771 {
4772 	if (v == SEQ_START_TOKEN)
4773 		seq_printf(seq,
4774 			   "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4775 			   IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4776 	else {
4777 		struct sock *s = sk_entry(v);
4778 		const struct packet_sock *po = pkt_sk(s);
4779 
4780 		seq_printf(seq,
4781 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4782 			   s,
4783 			   refcount_read(&s->sk_refcnt),
4784 			   s->sk_type,
4785 			   ntohs(READ_ONCE(po->num)),
4786 			   READ_ONCE(po->ifindex),
4787 			   packet_sock_flag(po, PACKET_SOCK_RUNNING),
4788 			   atomic_read(&s->sk_rmem_alloc),
4789 			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4790 			   sock_i_ino(s));
4791 	}
4792 
4793 	return 0;
4794 }
4795 
4796 static const struct seq_operations packet_seq_ops = {
4797 	.start	= packet_seq_start,
4798 	.next	= packet_seq_next,
4799 	.stop	= packet_seq_stop,
4800 	.show	= packet_seq_show,
4801 };
4802 #endif
4803 
4804 static int __net_init packet_net_init(struct net *net)
4805 {
4806 	mutex_init(&net->packet.sklist_lock);
4807 	INIT_HLIST_HEAD(&net->packet.sklist);
4808 
4809 #ifdef CONFIG_PROC_FS
4810 	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4811 			sizeof(struct seq_net_private)))
4812 		return -ENOMEM;
4813 #endif /* CONFIG_PROC_FS */
4814 
4815 	return 0;
4816 }
4817 
4818 static void __net_exit packet_net_exit(struct net *net)
4819 {
4820 	remove_proc_entry("packet", net->proc_net);
4821 	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4822 }
4823 
4824 static struct pernet_operations packet_net_ops = {
4825 	.init = packet_net_init,
4826 	.exit = packet_net_exit,
4827 };
4828 
4829 
4830 static void __exit packet_exit(void)
4831 {
4832 	sock_unregister(PF_PACKET);
4833 	proto_unregister(&packet_proto);
4834 	unregister_netdevice_notifier(&packet_netdev_notifier);
4835 	unregister_pernet_subsys(&packet_net_ops);
4836 }
4837 
4838 static int __init packet_init(void)
4839 {
4840 	int rc;
4841 
4842 	rc = register_pernet_subsys(&packet_net_ops);
4843 	if (rc)
4844 		goto out;
4845 	rc = register_netdevice_notifier(&packet_netdev_notifier);
4846 	if (rc)
4847 		goto out_pernet;
4848 	rc = proto_register(&packet_proto, 0);
4849 	if (rc)
4850 		goto out_notifier;
4851 	rc = sock_register(&packet_family_ops);
4852 	if (rc)
4853 		goto out_proto;
4854 
4855 	return 0;
4856 
4857 out_proto:
4858 	proto_unregister(&packet_proto);
4859 out_notifier:
4860 	unregister_netdevice_notifier(&packet_netdev_notifier);
4861 out_pernet:
4862 	unregister_pernet_subsys(&packet_net_ops);
4863 out:
4864 	return rc;
4865 }
4866 
4867 module_init(packet_init);
4868 module_exit(packet_exit);
4869 MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4870 MODULE_LICENSE("GPL");
4871 MODULE_ALIAS_NETPROTO(PF_PACKET);
4872