xref: /linux/net/packet/af_packet.c (revision 26ba30221c03364d6ed9910be8da4c1fd871b07b)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		PACKET - implements raw packet sockets.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *
13  * Fixes:
14  *		Alan Cox	:	verify_area() now used correctly
15  *		Alan Cox	:	new skbuff lists, look ma no backlogs!
16  *		Alan Cox	:	tidied skbuff lists.
17  *		Alan Cox	:	Now uses generic datagram routines I
18  *					added. Also fixed the peek/read crash
19  *					from all old Linux datagram code.
20  *		Alan Cox	:	Uses the improved datagram code.
21  *		Alan Cox	:	Added NULL's for socket options.
22  *		Alan Cox	:	Re-commented the code.
23  *		Alan Cox	:	Use new kernel side addressing
24  *		Rob Janssen	:	Correct MTU usage.
25  *		Dave Platt	:	Counter leaks caused by incorrect
26  *					interrupt locking and some slightly
27  *					dubious gcc output. Can you read
28  *					compiler: it said _VOLATILE_
29  *	Richard Kooijman	:	Timestamp fixes.
30  *		Alan Cox	:	New buffers. Use sk->mac.raw.
31  *		Alan Cox	:	sendmsg/recvmsg support.
32  *		Alan Cox	:	Protocol setting support
33  *	Alexey Kuznetsov	:	Untied from IPv4 stack.
34  *	Cyrus Durgin		:	Fixed kerneld for kmod.
35  *	Michal Ostrowski        :       Module initialization cleanup.
36  *         Ulises Alonso        :       Frame number limit removal and
37  *                                      packet_set_ring memory leak.
38  *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
39  *					The convention is that longer addresses
40  *					will simply extend the hardware address
41  *					byte arrays at the end of sockaddr_ll
42  *					and packet_mreq.
43  *		Johann Baudy	:	Added TX RING.
44  *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
45  *					layer.
46  *					Copyright (C) 2011, <lokec@ccs.neu.edu>
47  */
48 
49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
50 
51 #include <linux/ethtool.h>
52 #include <linux/uio.h>
53 #include <linux/filter.h>
54 #include <linux/types.h>
55 #include <linux/mm.h>
56 #include <linux/capability.h>
57 #include <linux/fcntl.h>
58 #include <linux/socket.h>
59 #include <linux/in.h>
60 #include <linux/inet.h>
61 #include <linux/netdevice.h>
62 #include <linux/if_packet.h>
63 #include <linux/wireless.h>
64 #include <linux/kernel.h>
65 #include <linux/kmod.h>
66 #include <linux/slab.h>
67 #include <linux/vmalloc.h>
68 #include <net/net_namespace.h>
69 #include <net/ip.h>
70 #include <net/protocol.h>
71 #include <linux/skbuff.h>
72 #include <net/sock.h>
73 #include <linux/errno.h>
74 #include <linux/timer.h>
75 #include <linux/uaccess.h>
76 #include <asm/ioctls.h>
77 #include <asm/page.h>
78 #include <asm/cacheflush.h>
79 #include <asm/io.h>
80 #include <linux/proc_fs.h>
81 #include <linux/seq_file.h>
82 #include <linux/poll.h>
83 #include <linux/module.h>
84 #include <linux/init.h>
85 #include <linux/mutex.h>
86 #include <linux/if_vlan.h>
87 #include <linux/virtio_net.h>
88 #include <linux/errqueue.h>
89 #include <linux/net_tstamp.h>
90 #include <linux/percpu.h>
91 #ifdef CONFIG_INET
92 #include <net/inet_common.h>
93 #endif
94 #include <linux/bpf.h>
95 #include <net/compat.h>
96 #include <linux/netfilter_netdev.h>
97 
98 #include "internal.h"
99 
100 /*
101    Assumptions:
102    - If the device has no dev->header_ops->create, there is no LL header
103      visible above the device. In this case, its hard_header_len should be 0.
104      The device may prepend its own header internally. In this case, its
105      needed_headroom should be set to the space needed for it to add its
106      internal header.
107      For example, a WiFi driver pretending to be an Ethernet driver should
108      set its hard_header_len to be the Ethernet header length, and set its
109      needed_headroom to be (the real WiFi header length - the fake Ethernet
110      header length).
111    - packet socket receives packets with pulled ll header,
112      so that SOCK_RAW should push it back.
113 
114 On receive:
115 -----------
116 
117 Incoming, dev_has_header(dev) == true
118    mac_header -> ll header
119    data       -> data
120 
121 Outgoing, dev_has_header(dev) == true
122    mac_header -> ll header
123    data       -> ll header
124 
125 Incoming, dev_has_header(dev) == false
126    mac_header -> data
127      However drivers often make it point to the ll header.
128      This is incorrect because the ll header should be invisible to us.
129    data       -> data
130 
131 Outgoing, dev_has_header(dev) == false
132    mac_header -> data. ll header is invisible to us.
133    data       -> data
134 
135 Resume
136   If dev_has_header(dev) == false we are unable to restore the ll header,
137     because it is invisible to us.
138 
139 
140 On transmit:
141 ------------
142 
143 dev_has_header(dev) == true
144    mac_header -> ll header
145    data       -> ll header
146 
147 dev_has_header(dev) == false (ll header is invisible to us)
148    mac_header -> data
149    data       -> data
150 
151    We should set network_header on output to the correct position,
152    packet classifier depends on it.
153  */
154 
155 /* Private packet socket structures. */
156 
157 /* identical to struct packet_mreq except it has
158  * a longer address field.
159  */
160 struct packet_mreq_max {
161 	int		mr_ifindex;
162 	unsigned short	mr_type;
163 	unsigned short	mr_alen;
164 	unsigned char	mr_address[MAX_ADDR_LEN];
165 };
166 
167 union tpacket_uhdr {
168 	struct tpacket_hdr  *h1;
169 	struct tpacket2_hdr *h2;
170 	struct tpacket3_hdr *h3;
171 	void *raw;
172 };
173 
174 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
175 		int closing, int tx_ring);
176 
177 #define V3_ALIGNMENT	(8)
178 
179 #define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
180 
181 #define BLK_PLUS_PRIV(sz_of_priv) \
182 	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
183 
184 #define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
185 #define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
186 #define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
187 #define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
188 #define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
189 #define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
190 
191 struct packet_sock;
192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
193 		       struct packet_type *pt, struct net_device *orig_dev);
194 
195 static void *packet_previous_frame(struct packet_sock *po,
196 		struct packet_ring_buffer *rb,
197 		int status);
198 static void packet_increment_head(struct packet_ring_buffer *buff);
199 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
201 			struct packet_sock *);
202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
203 		struct packet_sock *, unsigned int status);
204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
205 static void prb_open_block(struct tpacket_kbdq_core *,
206 		struct tpacket_block_desc *);
207 static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *);
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(const struct sk_buff *skb, struct net_device *dev)
542 {
543 	struct vlan_hdr vhdr, *vh;
544 	unsigned int header_len;
545 
546 	if (!dev)
547 		return 0;
548 
549 	/* In the SOCK_DGRAM scenario, skb data starts at the network
550 	 * protocol, which is after the VLAN headers. The outer VLAN
551 	 * header is at the hard_header_len offset in non-variable
552 	 * length link layer headers. If it's a VLAN device, the
553 	 * min_header_len should be used to exclude the VLAN header
554 	 * size.
555 	 */
556 	if (dev->min_header_len == dev->hard_header_len)
557 		header_len = dev->hard_header_len;
558 	else if (is_vlan_dev(dev))
559 		header_len = dev->min_header_len;
560 	else
561 		return 0;
562 
563 	vh = skb_header_pointer(skb, skb_mac_offset(skb) + header_len,
564 				sizeof(vhdr), &vhdr);
565 	if (unlikely(!vh))
566 		return 0;
567 
568 	return ntohs(vh->h_vlan_TCI);
569 }
570 
571 static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
572 {
573 	__be16 proto = skb->protocol;
574 
575 	if (unlikely(eth_type_vlan(proto)))
576 		proto = vlan_get_protocol_offset_inline(skb, proto,
577 							skb_mac_offset(skb),
578 							NULL);
579 
580 	return proto;
581 }
582 
583 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
584 		struct sk_buff_head *rb_queue)
585 {
586 	struct tpacket_kbdq_core *pkc;
587 
588 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
589 	hrtimer_cancel(&pkc->retire_blk_timer);
590 }
591 
592 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
593 				int blk_size_in_bytes)
594 {
595 	struct net_device *dev;
596 	unsigned int mbits, div;
597 	struct ethtool_link_ksettings ecmd;
598 	int err;
599 
600 	rtnl_lock();
601 	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
602 	if (unlikely(!dev)) {
603 		rtnl_unlock();
604 		return DEFAULT_PRB_RETIRE_TOV;
605 	}
606 	err = __ethtool_get_link_ksettings(dev, &ecmd);
607 	rtnl_unlock();
608 	if (err)
609 		return DEFAULT_PRB_RETIRE_TOV;
610 
611 	/* If the link speed is so slow you don't really
612 	 * need to worry about perf anyways
613 	 */
614 	if (ecmd.base.speed < SPEED_1000 ||
615 	    ecmd.base.speed == SPEED_UNKNOWN)
616 		return DEFAULT_PRB_RETIRE_TOV;
617 
618 	div = ecmd.base.speed / 1000;
619 	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
620 
621 	if (div)
622 		mbits /= div;
623 
624 	if (div)
625 		return mbits + 1;
626 	return mbits;
627 }
628 
629 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
630 			union tpacket_req_u *req_u)
631 {
632 	p1->feature_req_word = req_u->req3.tp_feature_req_word;
633 }
634 
635 static void init_prb_bdqc(struct packet_sock *po,
636 			struct packet_ring_buffer *rb,
637 			struct pgv *pg_vec,
638 			union tpacket_req_u *req_u)
639 {
640 	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
641 	struct tpacket_block_desc *pbd;
642 
643 	memset(p1, 0x0, sizeof(*p1));
644 
645 	p1->knxt_seq_num = 1;
646 	p1->pkbdq = pg_vec;
647 	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
648 	p1->pkblk_start	= pg_vec[0].buffer;
649 	p1->kblk_size = req_u->req3.tp_block_size;
650 	p1->knum_blocks	= req_u->req3.tp_block_nr;
651 	p1->hdrlen = po->tp_hdrlen;
652 	p1->version = po->tp_version;
653 	po->stats.stats3.tp_freeze_q_cnt = 0;
654 	if (req_u->req3.tp_retire_blk_tov)
655 		p1->interval_ktime = ms_to_ktime(req_u->req3.tp_retire_blk_tov);
656 	else
657 		p1->interval_ktime = ms_to_ktime(prb_calc_retire_blk_tmo(po,
658 						 req_u->req3.tp_block_size));
659 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
660 	rwlock_init(&p1->blk_fill_in_prog_lock);
661 
662 	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
663 	prb_init_ft_ops(p1, req_u);
664 	hrtimer_setup(&p1->retire_blk_timer, prb_retire_rx_blk_timer_expired,
665 		      CLOCK_MONOTONIC, HRTIMER_MODE_REL_SOFT);
666 	hrtimer_start(&p1->retire_blk_timer, p1->interval_ktime,
667 		      HRTIMER_MODE_REL_SOFT);
668 	prb_open_block(p1, pbd);
669 }
670 
671 /*
672  * With a 1MB block-size, on a 1Gbps line, it will take
673  * i) ~8 ms to fill a block + ii) memcpy etc.
674  * In this cut we are not accounting for the memcpy time.
675  *
676  * Since the tmo granularity is in msecs, it is not too expensive
677  * to refresh the timer, lets say every '8' msecs.
678  * Either the user can set the 'tmo' or we can derive it based on
679  * a) line-speed and b) block-size.
680  * prb_calc_retire_blk_tmo() calculates the tmo.
681  */
682 static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *t)
683 {
684 	struct packet_sock *po =
685 		timer_container_of(po, t, rx_ring.prb_bdqc.retire_blk_timer);
686 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
687 	unsigned int frozen;
688 	struct tpacket_block_desc *pbd;
689 
690 	spin_lock(&po->sk.sk_receive_queue.lock);
691 
692 	frozen = prb_queue_frozen(pkc);
693 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
694 
695 	/* We only need to plug the race when the block is partially filled.
696 	 * tpacket_rcv:
697 	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
698 	 *		copy_bits() is in progress ...
699 	 *		timer fires on other cpu:
700 	 *		we can't retire the current block because copy_bits
701 	 *		is in progress.
702 	 *
703 	 */
704 	if (BLOCK_NUM_PKTS(pbd)) {
705 		/* Waiting for skb_copy_bits to finish... */
706 		write_lock(&pkc->blk_fill_in_prog_lock);
707 		write_unlock(&pkc->blk_fill_in_prog_lock);
708 	}
709 
710 	if (!frozen) {
711 		if (BLOCK_NUM_PKTS(pbd)) {
712 			/* Not an empty block. Need retire the block. */
713 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
714 			prb_dispatch_next_block(pkc, po);
715 		}
716 	} else {
717 		/* Case 1. Queue was frozen because user-space was
718 		 * lagging behind.
719 		 */
720 		if (!prb_curr_blk_in_use(pbd)) {
721 			/* Case 2. queue was frozen,user-space caught up,
722 			 * now the link went idle && the timer fired.
723 			 * We don't have a block to close.So we open this
724 			 * block and restart the timer.
725 			 * opening a block thaws the queue,restarts timer
726 			 * Thawing/timer-refresh is a side effect.
727 			 */
728 			prb_open_block(pkc, pbd);
729 		}
730 	}
731 
732 	hrtimer_forward_now(&pkc->retire_blk_timer, pkc->interval_ktime);
733 	spin_unlock(&po->sk.sk_receive_queue.lock);
734 	return HRTIMER_RESTART;
735 }
736 
737 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
738 		struct tpacket_block_desc *pbd1, __u32 status)
739 {
740 	/* Flush everything minus the block header */
741 
742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
743 	u8 *start, *end;
744 
745 	start = (u8 *)pbd1;
746 
747 	/* Skip the block header(we know header WILL fit in 4K) */
748 	start += PAGE_SIZE;
749 
750 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
751 	for (; start < end; start += PAGE_SIZE)
752 		flush_dcache_page(pgv_to_page(start));
753 
754 	smp_wmb();
755 #endif
756 
757 	/* Now update the block status. */
758 
759 	BLOCK_STATUS(pbd1) = status;
760 
761 	/* Flush the block header */
762 
763 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
764 	start = (u8 *)pbd1;
765 	flush_dcache_page(pgv_to_page(start));
766 
767 	smp_wmb();
768 #endif
769 }
770 
771 /*
772  * Side effect:
773  *
774  * 1) flush the block
775  * 2) Increment active_blk_num
776  *
777  * Note:We DONT refresh the timer on purpose.
778  *	Because almost always the next block will be opened.
779  */
780 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
781 		struct tpacket_block_desc *pbd1,
782 		struct packet_sock *po, unsigned int stat)
783 {
784 	__u32 status = TP_STATUS_USER | stat;
785 
786 	struct tpacket3_hdr *last_pkt;
787 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
788 	struct sock *sk = &po->sk;
789 
790 	if (atomic_read(&po->tp_drops))
791 		status |= TP_STATUS_LOSING;
792 
793 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
794 	last_pkt->tp_next_offset = 0;
795 
796 	/* Get the ts of the last pkt */
797 	if (BLOCK_NUM_PKTS(pbd1)) {
798 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
799 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
800 	} else {
801 		/* Ok, we tmo'd - so get the current time.
802 		 *
803 		 * It shouldn't really happen as we don't close empty
804 		 * blocks. See prb_retire_rx_blk_timer_expired().
805 		 */
806 		struct timespec64 ts;
807 		ktime_get_real_ts64(&ts);
808 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
809 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
810 	}
811 
812 	smp_wmb();
813 
814 	/* Flush the block */
815 	prb_flush_block(pkc1, pbd1, status);
816 
817 	sk->sk_data_ready(sk);
818 
819 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
820 }
821 
822 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
823 {
824 	pkc->reset_pending_on_curr_blk = 0;
825 }
826 
827 /*
828  * prb_open_block is called by tpacket_rcv or timer callback.
829  *
830  * Reasons why NOT update hrtimer in prb_open_block:
831  * 1) It will increase complexity to distinguish the two caller scenario.
832  * 2) hrtimer_cancel and hrtimer_start need to be called if you want to update
833  * TMO of an already enqueued hrtimer, leading to complex shutdown logic.
834  *
835  * One side effect of NOT update hrtimer when called by tpacket_rcv is that
836  * a newly opened block triggered by tpacket_rcv may be retired earlier than
837  * expected. On the other hand, if timeout is updated in prb_open_block, the
838  * frequent reception of network packets that leads to prb_open_block being
839  * called may cause hrtimer to be removed and enqueued repeatedly.
840  */
841 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
842 	struct tpacket_block_desc *pbd1)
843 {
844 	struct timespec64 ts;
845 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
846 
847 	smp_rmb();
848 
849 	/* We could have just memset this but we will lose the
850 	 * flexibility of making the priv area sticky
851 	 */
852 
853 	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
854 	BLOCK_NUM_PKTS(pbd1) = 0;
855 	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
856 
857 	ktime_get_real_ts64(&ts);
858 
859 	h1->ts_first_pkt.ts_sec = ts.tv_sec;
860 	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
861 
862 	pkc1->pkblk_start = (char *)pbd1;
863 	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864 
865 	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
867 
868 	pbd1->version = pkc1->version;
869 	pkc1->prev = pkc1->nxt_offset;
870 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
871 
872 	prb_thaw_queue(pkc1);
873 
874 	smp_wmb();
875 }
876 
877 /*
878  * Queue freeze logic:
879  * 1) Assume tp_block_nr = 8 blocks.
880  * 2) At time 't0', user opens Rx ring.
881  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882  * 4) user-space is either sleeping or processing block '0'.
883  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884  *    it will close block-7,loop around and try to fill block '0'.
885  *    call-flow:
886  *    __packet_lookup_frame_in_block
887  *      prb_retire_current_block()
888  *      prb_dispatch_next_block()
889  *        |->(BLOCK_STATUS == USER) evaluates to true
890  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
891  * 6) Now there are two cases:
892  *    6.1) Link goes idle right after the queue is frozen.
893  *         But remember, the last open_block() refreshed the timer.
894  *         When this timer expires,it will refresh itself so that we can
895  *         re-open block-0 in near future.
896  *    6.2) Link is busy and keeps on receiving packets. This is a simple
897  *         case and __packet_lookup_frame_in_block will check if block-0
898  *         is free and can now be re-used.
899  */
900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
901 				  struct packet_sock *po)
902 {
903 	pkc->reset_pending_on_curr_blk = 1;
904 	po->stats.stats3.tp_freeze_q_cnt++;
905 }
906 
907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
908 
909 /*
910  * If the next block is free then we will dispatch it
911  * and return a good offset.
912  * Else, we will freeze the queue.
913  * So, caller must check the return value.
914  */
915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
916 		struct packet_sock *po)
917 {
918 	struct tpacket_block_desc *pbd;
919 
920 	smp_rmb();
921 
922 	/* 1. Get current block num */
923 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
924 
925 	/* 2. If this block is currently in_use then freeze the queue */
926 	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
927 		prb_freeze_queue(pkc, po);
928 		return NULL;
929 	}
930 
931 	/*
932 	 * 3.
933 	 * open this block and return the offset where the first packet
934 	 * needs to get stored.
935 	 */
936 	prb_open_block(pkc, pbd);
937 	return (void *)pkc->nxt_offset;
938 }
939 
940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
941 		struct packet_sock *po, unsigned int status)
942 {
943 	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
944 
945 	/* retire/close the current block */
946 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
947 		/*
948 		 * Plug the case where copy_bits() is in progress on
949 		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950 		 * have space to copy the pkt in the current block and
951 		 * called prb_retire_current_block()
952 		 *
953 		 * We don't need to worry about the TMO case because
954 		 * the timer-handler already handled this case.
955 		 */
956 		if (!(status & TP_STATUS_BLK_TMO)) {
957 			/* Waiting for skb_copy_bits to finish... */
958 			write_lock(&pkc->blk_fill_in_prog_lock);
959 			write_unlock(&pkc->blk_fill_in_prog_lock);
960 		}
961 		prb_close_block(pkc, pbd, po, status);
962 		return;
963 	}
964 }
965 
966 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
967 {
968 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
969 }
970 
971 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
972 {
973 	return pkc->reset_pending_on_curr_blk;
974 }
975 
976 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
977 	__releases(&pkc->blk_fill_in_prog_lock)
978 {
979 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
980 
981 	read_unlock(&pkc->blk_fill_in_prog_lock);
982 }
983 
984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
985 			struct tpacket3_hdr *ppd)
986 {
987 	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
988 }
989 
990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
991 			struct tpacket3_hdr *ppd)
992 {
993 	ppd->hv1.tp_rxhash = 0;
994 }
995 
996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
997 			struct tpacket3_hdr *ppd)
998 {
999 	struct packet_sock *po = container_of(pkc, struct packet_sock, rx_ring.prb_bdqc);
1000 
1001 	if (skb_vlan_tag_present(pkc->skb)) {
1002 		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1003 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1004 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1005 	} else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) {
1006 		ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev);
1007 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol);
1008 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1009 	} else {
1010 		ppd->hv1.tp_vlan_tci = 0;
1011 		ppd->hv1.tp_vlan_tpid = 0;
1012 		ppd->tp_status = TP_STATUS_AVAILABLE;
1013 	}
1014 }
1015 
1016 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1017 			struct tpacket3_hdr *ppd)
1018 {
1019 	ppd->hv1.tp_padding = 0;
1020 	prb_fill_vlan_info(pkc, ppd);
1021 
1022 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1023 		prb_fill_rxhash(pkc, ppd);
1024 	else
1025 		prb_clear_rxhash(pkc, ppd);
1026 }
1027 
1028 static void prb_fill_curr_block(char *curr,
1029 				struct tpacket_kbdq_core *pkc,
1030 				struct tpacket_block_desc *pbd,
1031 				unsigned int len)
1032 	__acquires(&pkc->blk_fill_in_prog_lock)
1033 {
1034 	struct tpacket3_hdr *ppd;
1035 
1036 	ppd  = (struct tpacket3_hdr *)curr;
1037 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1038 	pkc->prev = curr;
1039 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1040 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 	BLOCK_NUM_PKTS(pbd) += 1;
1042 	read_lock(&pkc->blk_fill_in_prog_lock);
1043 	prb_run_all_ft_ops(pkc, ppd);
1044 }
1045 
1046 /* Assumes caller has the sk->rx_queue.lock */
1047 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1048 					    struct sk_buff *skb,
1049 					    unsigned int len
1050 					    )
1051 {
1052 	struct tpacket_kbdq_core *pkc;
1053 	struct tpacket_block_desc *pbd;
1054 	char *curr, *end;
1055 
1056 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1057 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1058 
1059 	/* Queue is frozen when user space is lagging behind */
1060 	if (prb_queue_frozen(pkc)) {
1061 		/*
1062 		 * Check if that last block which caused the queue to freeze,
1063 		 * is still in_use by user-space.
1064 		 */
1065 		if (prb_curr_blk_in_use(pbd)) {
1066 			/* Can't record this packet */
1067 			return NULL;
1068 		} else {
1069 			/*
1070 			 * Ok, the block was released by user-space.
1071 			 * Now let's open that block.
1072 			 * opening a block also thaws the queue.
1073 			 * Thawing is a side effect.
1074 			 */
1075 			prb_open_block(pkc, pbd);
1076 		}
1077 	}
1078 
1079 	smp_mb();
1080 	curr = pkc->nxt_offset;
1081 	pkc->skb = skb;
1082 	end = (char *)pbd + pkc->kblk_size;
1083 
1084 	/* first try the current block */
1085 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1086 		prb_fill_curr_block(curr, pkc, pbd, len);
1087 		return (void *)curr;
1088 	}
1089 
1090 	/* Ok, close the current block */
1091 	prb_retire_current_block(pkc, po, 0);
1092 
1093 	/* Now, try to dispatch the next block */
1094 	curr = (char *)prb_dispatch_next_block(pkc, po);
1095 	if (curr) {
1096 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1097 		prb_fill_curr_block(curr, pkc, pbd, len);
1098 		return (void *)curr;
1099 	}
1100 
1101 	/*
1102 	 * No free blocks are available.user_space hasn't caught up yet.
1103 	 * Queue was just frozen and now this packet will get dropped.
1104 	 */
1105 	return NULL;
1106 }
1107 
1108 static void *packet_current_rx_frame(struct packet_sock *po,
1109 					    struct sk_buff *skb,
1110 					    int status, unsigned int len)
1111 {
1112 	char *curr = NULL;
1113 	switch (po->tp_version) {
1114 	case TPACKET_V1:
1115 	case TPACKET_V2:
1116 		curr = packet_lookup_frame(po, &po->rx_ring,
1117 					po->rx_ring.head, status);
1118 		return curr;
1119 	case TPACKET_V3:
1120 		return __packet_lookup_frame_in_block(po, skb, len);
1121 	default:
1122 		WARN(1, "TPACKET version not supported\n");
1123 		BUG();
1124 		return NULL;
1125 	}
1126 }
1127 
1128 static void *prb_lookup_block(const struct packet_sock *po,
1129 			      const struct packet_ring_buffer *rb,
1130 			      unsigned int idx,
1131 			      int status)
1132 {
1133 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1134 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1135 
1136 	if (status != BLOCK_STATUS(pbd))
1137 		return NULL;
1138 	return pbd;
1139 }
1140 
1141 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1142 {
1143 	unsigned int prev;
1144 	if (rb->prb_bdqc.kactive_blk_num)
1145 		prev = rb->prb_bdqc.kactive_blk_num-1;
1146 	else
1147 		prev = rb->prb_bdqc.knum_blocks-1;
1148 	return prev;
1149 }
1150 
1151 /* Assumes caller has held the rx_queue.lock */
1152 static void *__prb_previous_block(struct packet_sock *po,
1153 					 struct packet_ring_buffer *rb,
1154 					 int status)
1155 {
1156 	unsigned int previous = prb_previous_blk_num(rb);
1157 	return prb_lookup_block(po, rb, previous, status);
1158 }
1159 
1160 static void *packet_previous_rx_frame(struct packet_sock *po,
1161 					     struct packet_ring_buffer *rb,
1162 					     int status)
1163 {
1164 	if (po->tp_version <= TPACKET_V2)
1165 		return packet_previous_frame(po, rb, status);
1166 
1167 	return __prb_previous_block(po, rb, status);
1168 }
1169 
1170 static void packet_increment_rx_head(struct packet_sock *po,
1171 					    struct packet_ring_buffer *rb)
1172 {
1173 	switch (po->tp_version) {
1174 	case TPACKET_V1:
1175 	case TPACKET_V2:
1176 		return packet_increment_head(rb);
1177 	case TPACKET_V3:
1178 	default:
1179 		WARN(1, "TPACKET version not supported.\n");
1180 		BUG();
1181 		return;
1182 	}
1183 }
1184 
1185 static void *packet_previous_frame(struct packet_sock *po,
1186 		struct packet_ring_buffer *rb,
1187 		int status)
1188 {
1189 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1190 	return packet_lookup_frame(po, rb, previous, status);
1191 }
1192 
1193 static void packet_increment_head(struct packet_ring_buffer *buff)
1194 {
1195 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1196 }
1197 
1198 static void packet_inc_pending(struct packet_ring_buffer *rb)
1199 {
1200 	this_cpu_inc(*rb->pending_refcnt);
1201 }
1202 
1203 static void packet_dec_pending(struct packet_ring_buffer *rb)
1204 {
1205 	this_cpu_dec(*rb->pending_refcnt);
1206 }
1207 
1208 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1209 {
1210 	unsigned int refcnt = 0;
1211 	int cpu;
1212 
1213 	/* We don't use pending refcount in rx_ring. */
1214 	if (rb->pending_refcnt == NULL)
1215 		return 0;
1216 
1217 	for_each_possible_cpu(cpu)
1218 		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1219 
1220 	return refcnt;
1221 }
1222 
1223 static int packet_alloc_pending(struct packet_sock *po)
1224 {
1225 	po->rx_ring.pending_refcnt = NULL;
1226 
1227 	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1228 	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1229 		return -ENOBUFS;
1230 
1231 	return 0;
1232 }
1233 
1234 static void packet_free_pending(struct packet_sock *po)
1235 {
1236 	free_percpu(po->tx_ring.pending_refcnt);
1237 }
1238 
1239 #define ROOM_POW_OFF	2
1240 #define ROOM_NONE	0x0
1241 #define ROOM_LOW	0x1
1242 #define ROOM_NORMAL	0x2
1243 
1244 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1245 {
1246 	int idx, len;
1247 
1248 	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1249 	idx = READ_ONCE(po->rx_ring.head);
1250 	if (pow_off)
1251 		idx += len >> pow_off;
1252 	if (idx >= len)
1253 		idx -= len;
1254 	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1255 }
1256 
1257 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1258 {
1259 	int idx, len;
1260 
1261 	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1262 	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1263 	if (pow_off)
1264 		idx += len >> pow_off;
1265 	if (idx >= len)
1266 		idx -= len;
1267 	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1268 }
1269 
1270 static int __packet_rcv_has_room(const struct packet_sock *po,
1271 				 const struct sk_buff *skb)
1272 {
1273 	const struct sock *sk = &po->sk;
1274 	int ret = ROOM_NONE;
1275 
1276 	if (po->prot_hook.func != tpacket_rcv) {
1277 		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1278 		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1279 				   - (skb ? skb->truesize : 0);
1280 
1281 		if (avail > (rcvbuf >> ROOM_POW_OFF))
1282 			return ROOM_NORMAL;
1283 		else if (avail > 0)
1284 			return ROOM_LOW;
1285 		else
1286 			return ROOM_NONE;
1287 	}
1288 
1289 	if (po->tp_version == TPACKET_V3) {
1290 		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1291 			ret = ROOM_NORMAL;
1292 		else if (__tpacket_v3_has_room(po, 0))
1293 			ret = ROOM_LOW;
1294 	} else {
1295 		if (__tpacket_has_room(po, ROOM_POW_OFF))
1296 			ret = ROOM_NORMAL;
1297 		else if (__tpacket_has_room(po, 0))
1298 			ret = ROOM_LOW;
1299 	}
1300 
1301 	return ret;
1302 }
1303 
1304 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1305 {
1306 	bool pressure;
1307 	int ret;
1308 
1309 	ret = __packet_rcv_has_room(po, skb);
1310 	pressure = ret != ROOM_NORMAL;
1311 
1312 	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1313 		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1314 
1315 	return ret;
1316 }
1317 
1318 static void __packet_rcv_try_clear_pressure(struct packet_sock *po)
1319 {
1320 	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1321 	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1322 		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1323 }
1324 
1325 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1326 {
1327 	struct sock *sk = &po->sk;
1328 
1329 	if (!packet_sock_flag(po, PACKET_SOCK_PRESSURE))
1330 		return;
1331 
1332 	spin_lock_bh(&sk->sk_receive_queue.lock);
1333 	__packet_rcv_try_clear_pressure(po);
1334 	spin_unlock_bh(&sk->sk_receive_queue.lock);
1335 }
1336 
1337 static void packet_sock_destruct(struct sock *sk)
1338 {
1339 	skb_queue_purge(&sk->sk_error_queue);
1340 
1341 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1342 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1343 
1344 	if (!sock_flag(sk, SOCK_DEAD)) {
1345 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1346 		return;
1347 	}
1348 }
1349 
1350 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1351 {
1352 	u32 *history = po->rollover->history;
1353 	u32 victim, rxhash;
1354 	int i, count = 0;
1355 
1356 	rxhash = skb_get_hash(skb);
1357 	for (i = 0; i < ROLLOVER_HLEN; i++)
1358 		if (READ_ONCE(history[i]) == rxhash)
1359 			count++;
1360 
1361 	victim = get_random_u32_below(ROLLOVER_HLEN);
1362 
1363 	/* Avoid dirtying the cache line if possible */
1364 	if (READ_ONCE(history[victim]) != rxhash)
1365 		WRITE_ONCE(history[victim], rxhash);
1366 
1367 	return count > (ROLLOVER_HLEN >> 1);
1368 }
1369 
1370 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1371 				      struct sk_buff *skb,
1372 				      unsigned int num)
1373 {
1374 	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1375 }
1376 
1377 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1378 				    struct sk_buff *skb,
1379 				    unsigned int num)
1380 {
1381 	unsigned int val = atomic_inc_return(&f->rr_cur);
1382 
1383 	return val % num;
1384 }
1385 
1386 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1387 				     struct sk_buff *skb,
1388 				     unsigned int num)
1389 {
1390 	return smp_processor_id() % num;
1391 }
1392 
1393 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1394 				     struct sk_buff *skb,
1395 				     unsigned int num)
1396 {
1397 	return get_random_u32_below(num);
1398 }
1399 
1400 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1401 					  struct sk_buff *skb,
1402 					  unsigned int idx, bool try_self,
1403 					  unsigned int num)
1404 {
1405 	struct packet_sock *po, *po_next, *po_skip = NULL;
1406 	unsigned int i, j, room = ROOM_NONE;
1407 
1408 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1409 
1410 	if (try_self) {
1411 		room = packet_rcv_has_room(po, skb);
1412 		if (room == ROOM_NORMAL ||
1413 		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1414 			return idx;
1415 		po_skip = po;
1416 	}
1417 
1418 	i = j = min_t(int, po->rollover->sock, num - 1);
1419 	do {
1420 		po_next = pkt_sk(rcu_dereference(f->arr[i]));
1421 		if (po_next != po_skip &&
1422 		    !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1423 		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1424 			if (i != j)
1425 				po->rollover->sock = i;
1426 			atomic_long_inc(&po->rollover->num);
1427 			if (room == ROOM_LOW)
1428 				atomic_long_inc(&po->rollover->num_huge);
1429 			return i;
1430 		}
1431 
1432 		if (++i == num)
1433 			i = 0;
1434 	} while (i != j);
1435 
1436 	atomic_long_inc(&po->rollover->num_failed);
1437 	return idx;
1438 }
1439 
1440 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1441 				    struct sk_buff *skb,
1442 				    unsigned int num)
1443 {
1444 	return skb_get_queue_mapping(skb) % num;
1445 }
1446 
1447 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1448 				     struct sk_buff *skb,
1449 				     unsigned int num)
1450 {
1451 	struct bpf_prog *prog;
1452 	unsigned int ret = 0;
1453 
1454 	rcu_read_lock();
1455 	prog = rcu_dereference(f->bpf_prog);
1456 	if (prog)
1457 		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1458 	rcu_read_unlock();
1459 
1460 	return ret;
1461 }
1462 
1463 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1464 {
1465 	return f->flags & (flag >> 8);
1466 }
1467 
1468 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1469 			     struct packet_type *pt, struct net_device *orig_dev)
1470 {
1471 	struct packet_fanout *f = pt->af_packet_priv;
1472 	unsigned int num = READ_ONCE(f->num_members);
1473 	struct net *net = read_pnet(&f->net);
1474 	struct packet_sock *po;
1475 	unsigned int idx;
1476 
1477 	if (!net_eq(dev_net(dev), net) || !num) {
1478 		kfree_skb(skb);
1479 		return 0;
1480 	}
1481 
1482 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1483 		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1484 		if (!skb)
1485 			return 0;
1486 	}
1487 	switch (f->type) {
1488 	case PACKET_FANOUT_HASH:
1489 	default:
1490 		idx = fanout_demux_hash(f, skb, num);
1491 		break;
1492 	case PACKET_FANOUT_LB:
1493 		idx = fanout_demux_lb(f, skb, num);
1494 		break;
1495 	case PACKET_FANOUT_CPU:
1496 		idx = fanout_demux_cpu(f, skb, num);
1497 		break;
1498 	case PACKET_FANOUT_RND:
1499 		idx = fanout_demux_rnd(f, skb, num);
1500 		break;
1501 	case PACKET_FANOUT_QM:
1502 		idx = fanout_demux_qm(f, skb, num);
1503 		break;
1504 	case PACKET_FANOUT_ROLLOVER:
1505 		idx = fanout_demux_rollover(f, skb, 0, false, num);
1506 		break;
1507 	case PACKET_FANOUT_CBPF:
1508 	case PACKET_FANOUT_EBPF:
1509 		idx = fanout_demux_bpf(f, skb, num);
1510 		break;
1511 	}
1512 
1513 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1514 		idx = fanout_demux_rollover(f, skb, idx, true, num);
1515 
1516 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1517 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1518 }
1519 
1520 DEFINE_MUTEX(fanout_mutex);
1521 EXPORT_SYMBOL_GPL(fanout_mutex);
1522 static LIST_HEAD(fanout_list);
1523 static u16 fanout_next_id;
1524 
1525 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1526 {
1527 	struct packet_fanout *f = po->fanout;
1528 
1529 	spin_lock(&f->lock);
1530 	rcu_assign_pointer(f->arr[f->num_members], sk);
1531 	smp_wmb();
1532 	f->num_members++;
1533 	if (f->num_members == 1)
1534 		dev_add_pack(&f->prot_hook);
1535 	spin_unlock(&f->lock);
1536 }
1537 
1538 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1539 {
1540 	struct packet_fanout *f = po->fanout;
1541 	int i;
1542 
1543 	spin_lock(&f->lock);
1544 	for (i = 0; i < f->num_members; i++) {
1545 		if (rcu_dereference_protected(f->arr[i],
1546 					      lockdep_is_held(&f->lock)) == sk)
1547 			break;
1548 	}
1549 	BUG_ON(i >= f->num_members);
1550 	rcu_assign_pointer(f->arr[i],
1551 			   rcu_dereference_protected(f->arr[f->num_members - 1],
1552 						     lockdep_is_held(&f->lock)));
1553 	f->num_members--;
1554 	if (f->num_members == 0)
1555 		__dev_remove_pack(&f->prot_hook);
1556 	spin_unlock(&f->lock);
1557 }
1558 
1559 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1560 {
1561 	if (sk->sk_family != PF_PACKET)
1562 		return false;
1563 
1564 	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1565 }
1566 
1567 static void fanout_init_data(struct packet_fanout *f)
1568 {
1569 	switch (f->type) {
1570 	case PACKET_FANOUT_LB:
1571 		atomic_set(&f->rr_cur, 0);
1572 		break;
1573 	case PACKET_FANOUT_CBPF:
1574 	case PACKET_FANOUT_EBPF:
1575 		RCU_INIT_POINTER(f->bpf_prog, NULL);
1576 		break;
1577 	}
1578 }
1579 
1580 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1581 {
1582 	struct bpf_prog *old;
1583 
1584 	spin_lock(&f->lock);
1585 	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1586 	rcu_assign_pointer(f->bpf_prog, new);
1587 	spin_unlock(&f->lock);
1588 
1589 	if (old) {
1590 		synchronize_net();
1591 		bpf_prog_destroy(old);
1592 	}
1593 }
1594 
1595 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1596 				unsigned int len)
1597 {
1598 	struct bpf_prog *new;
1599 	struct sock_fprog fprog;
1600 	int ret;
1601 
1602 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1603 		return -EPERM;
1604 
1605 	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1606 	if (ret)
1607 		return ret;
1608 
1609 	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1610 	if (ret)
1611 		return ret;
1612 
1613 	__fanout_set_data_bpf(po->fanout, new);
1614 	return 0;
1615 }
1616 
1617 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1618 				unsigned int len)
1619 {
1620 	struct bpf_prog *new;
1621 	u32 fd;
1622 
1623 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1624 		return -EPERM;
1625 	if (len != sizeof(fd))
1626 		return -EINVAL;
1627 	if (copy_from_sockptr(&fd, data, len))
1628 		return -EFAULT;
1629 
1630 	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1631 	if (IS_ERR(new))
1632 		return PTR_ERR(new);
1633 
1634 	__fanout_set_data_bpf(po->fanout, new);
1635 	return 0;
1636 }
1637 
1638 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1639 			   unsigned int len)
1640 {
1641 	switch (po->fanout->type) {
1642 	case PACKET_FANOUT_CBPF:
1643 		return fanout_set_data_cbpf(po, data, len);
1644 	case PACKET_FANOUT_EBPF:
1645 		return fanout_set_data_ebpf(po, data, len);
1646 	default:
1647 		return -EINVAL;
1648 	}
1649 }
1650 
1651 static void fanout_release_data(struct packet_fanout *f)
1652 {
1653 	switch (f->type) {
1654 	case PACKET_FANOUT_CBPF:
1655 	case PACKET_FANOUT_EBPF:
1656 		__fanout_set_data_bpf(f, NULL);
1657 	}
1658 }
1659 
1660 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1661 {
1662 	struct packet_fanout *f;
1663 
1664 	list_for_each_entry(f, &fanout_list, list) {
1665 		if (f->id == candidate_id &&
1666 		    read_pnet(&f->net) == sock_net(sk)) {
1667 			return false;
1668 		}
1669 	}
1670 	return true;
1671 }
1672 
1673 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1674 {
1675 	u16 id = fanout_next_id;
1676 
1677 	do {
1678 		if (__fanout_id_is_free(sk, id)) {
1679 			*new_id = id;
1680 			fanout_next_id = id + 1;
1681 			return true;
1682 		}
1683 
1684 		id++;
1685 	} while (id != fanout_next_id);
1686 
1687 	return false;
1688 }
1689 
1690 static int fanout_add(struct sock *sk, struct fanout_args *args)
1691 {
1692 	struct packet_rollover *rollover = NULL;
1693 	struct packet_sock *po = pkt_sk(sk);
1694 	u16 type_flags = args->type_flags;
1695 	struct packet_fanout *f, *match;
1696 	u8 type = type_flags & 0xff;
1697 	u8 flags = type_flags >> 8;
1698 	u16 id = args->id;
1699 	int err;
1700 
1701 	switch (type) {
1702 	case PACKET_FANOUT_ROLLOVER:
1703 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1704 			return -EINVAL;
1705 		break;
1706 	case PACKET_FANOUT_HASH:
1707 	case PACKET_FANOUT_LB:
1708 	case PACKET_FANOUT_CPU:
1709 	case PACKET_FANOUT_RND:
1710 	case PACKET_FANOUT_QM:
1711 	case PACKET_FANOUT_CBPF:
1712 	case PACKET_FANOUT_EBPF:
1713 		break;
1714 	default:
1715 		return -EINVAL;
1716 	}
1717 
1718 	mutex_lock(&fanout_mutex);
1719 
1720 	err = -EALREADY;
1721 	if (po->fanout)
1722 		goto out;
1723 
1724 	if (type == PACKET_FANOUT_ROLLOVER ||
1725 	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1726 		err = -ENOMEM;
1727 		rollover = kzalloc_obj(*rollover);
1728 		if (!rollover)
1729 			goto out;
1730 		atomic_long_set(&rollover->num, 0);
1731 		atomic_long_set(&rollover->num_huge, 0);
1732 		atomic_long_set(&rollover->num_failed, 0);
1733 	}
1734 
1735 	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1736 		if (id != 0) {
1737 			err = -EINVAL;
1738 			goto out;
1739 		}
1740 		if (!fanout_find_new_id(sk, &id)) {
1741 			err = -ENOMEM;
1742 			goto out;
1743 		}
1744 		/* ephemeral flag for the first socket in the group: drop it */
1745 		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1746 	}
1747 
1748 	match = NULL;
1749 	list_for_each_entry(f, &fanout_list, list) {
1750 		if (f->id == id &&
1751 		    read_pnet(&f->net) == sock_net(sk)) {
1752 			match = f;
1753 			break;
1754 		}
1755 	}
1756 	err = -EINVAL;
1757 	if (match) {
1758 		if (match->flags != flags)
1759 			goto out;
1760 		if (args->max_num_members &&
1761 		    args->max_num_members != match->max_num_members)
1762 			goto out;
1763 	} else {
1764 		if (args->max_num_members > PACKET_FANOUT_MAX)
1765 			goto out;
1766 		if (!args->max_num_members)
1767 			/* legacy PACKET_FANOUT_MAX */
1768 			args->max_num_members = 256;
1769 		err = -ENOMEM;
1770 		match = kvzalloc_flex(*match, arr, args->max_num_members);
1771 		if (!match)
1772 			goto out;
1773 		write_pnet(&match->net, sock_net(sk));
1774 		match->id = id;
1775 		match->type = type;
1776 		match->flags = flags;
1777 		INIT_LIST_HEAD(&match->list);
1778 		spin_lock_init(&match->lock);
1779 		refcount_set(&match->sk_ref, 0);
1780 		fanout_init_data(match);
1781 		match->prot_hook.type = po->prot_hook.type;
1782 		match->prot_hook.dev = po->prot_hook.dev;
1783 		match->prot_hook.func = packet_rcv_fanout;
1784 		match->prot_hook.af_packet_priv = match;
1785 		match->prot_hook.af_packet_net = read_pnet(&match->net);
1786 		match->prot_hook.id_match = match_fanout_group;
1787 		match->max_num_members = args->max_num_members;
1788 		match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1789 		list_add(&match->list, &fanout_list);
1790 	}
1791 	err = -EINVAL;
1792 
1793 	spin_lock(&po->bind_lock);
1794 	if (po->num &&
1795 	    match->type == type &&
1796 	    match->prot_hook.type == po->prot_hook.type &&
1797 	    match->prot_hook.dev == po->prot_hook.dev) {
1798 		err = -ENOSPC;
1799 		if (refcount_read(&match->sk_ref) < match->max_num_members) {
1800 			/* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1801 			WRITE_ONCE(po->fanout, match);
1802 
1803 			po->rollover = rollover;
1804 			rollover = NULL;
1805 			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1806 			if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
1807 				__dev_remove_pack(&po->prot_hook);
1808 				__fanout_link(sk, po);
1809 			}
1810 			err = 0;
1811 		}
1812 	}
1813 	spin_unlock(&po->bind_lock);
1814 
1815 	if (err && !refcount_read(&match->sk_ref)) {
1816 		list_del(&match->list);
1817 		kvfree(match);
1818 	}
1819 
1820 out:
1821 	kfree(rollover);
1822 	mutex_unlock(&fanout_mutex);
1823 	return err;
1824 }
1825 
1826 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1827  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1828  * It is the responsibility of the caller to call fanout_release_data() and
1829  * free the returned packet_fanout (after synchronize_net())
1830  */
1831 static struct packet_fanout *fanout_release(struct sock *sk)
1832 {
1833 	struct packet_sock *po = pkt_sk(sk);
1834 	struct packet_fanout *f;
1835 
1836 	mutex_lock(&fanout_mutex);
1837 	f = po->fanout;
1838 	if (f) {
1839 		po->fanout = NULL;
1840 
1841 		if (refcount_dec_and_test(&f->sk_ref))
1842 			list_del(&f->list);
1843 		else
1844 			f = NULL;
1845 	}
1846 	mutex_unlock(&fanout_mutex);
1847 
1848 	return f;
1849 }
1850 
1851 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1852 					  struct sk_buff *skb)
1853 {
1854 	/* Earlier code assumed this would be a VLAN pkt, double-check
1855 	 * this now that we have the actual packet in hand. We can only
1856 	 * do this check on Ethernet devices.
1857 	 */
1858 	if (unlikely(dev->type != ARPHRD_ETHER))
1859 		return false;
1860 
1861 	skb_reset_mac_header(skb);
1862 	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1863 }
1864 
1865 static const struct proto_ops packet_ops;
1866 
1867 static const struct proto_ops packet_ops_spkt;
1868 
1869 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1870 			   struct packet_type *pt, struct net_device *orig_dev)
1871 {
1872 	struct sock *sk;
1873 	struct sockaddr_pkt *spkt;
1874 
1875 	/*
1876 	 *	When we registered the protocol we saved the socket in the data
1877 	 *	field for just this event.
1878 	 */
1879 
1880 	sk = pt->af_packet_priv;
1881 
1882 	/*
1883 	 *	Yank back the headers [hope the device set this
1884 	 *	right or kerboom...]
1885 	 *
1886 	 *	Incoming packets have ll header pulled,
1887 	 *	push it back.
1888 	 *
1889 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1890 	 *	so that this procedure is noop.
1891 	 */
1892 
1893 	if (skb->pkt_type == PACKET_LOOPBACK)
1894 		goto out;
1895 
1896 	if (!net_eq(dev_net(dev), sock_net(sk)))
1897 		goto out;
1898 
1899 	skb = skb_share_check(skb, GFP_ATOMIC);
1900 	if (skb == NULL)
1901 		goto oom;
1902 
1903 	/* drop any routing info */
1904 	skb_dst_drop(skb);
1905 
1906 	/* drop conntrack reference */
1907 	nf_reset_ct(skb);
1908 
1909 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1910 
1911 	skb_push(skb, skb->data - skb_mac_header(skb));
1912 
1913 	/*
1914 	 *	The SOCK_PACKET socket receives _all_ frames.
1915 	 */
1916 
1917 	spkt->spkt_family = dev->type;
1918 	strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1919 	spkt->spkt_protocol = skb->protocol;
1920 
1921 	/*
1922 	 *	Charge the memory to the socket. This is done specifically
1923 	 *	to prevent sockets using all the memory up.
1924 	 */
1925 
1926 	if (sock_queue_rcv_skb(sk, skb) == 0)
1927 		return 0;
1928 
1929 out:
1930 	kfree_skb(skb);
1931 oom:
1932 	return 0;
1933 }
1934 
1935 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1936 {
1937 	int depth;
1938 
1939 	/* On TX skb->data is the L2 header; anchor it for all socket types. */
1940 	skb_reset_mac_header(skb);
1941 
1942 	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1943 	    sock->type == SOCK_RAW)
1944 		skb->protocol = dev_parse_header_protocol(skb);
1945 
1946 	/* Move network header to the right position for VLAN tagged packets */
1947 	if (likely(skb->dev->type == ARPHRD_ETHER) &&
1948 	    eth_type_vlan(skb->protocol) &&
1949 	    vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1950 		skb_set_network_header(skb, depth);
1951 
1952 	skb_probe_transport_header(skb);
1953 }
1954 
1955 /*
1956  *	Output a raw packet to a device layer. This bypasses all the other
1957  *	protocol layers and you must therefore supply it with a complete frame
1958  */
1959 
1960 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1961 			       size_t len)
1962 {
1963 	struct sock *sk = sock->sk;
1964 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1965 	struct sk_buff *skb = NULL;
1966 	struct net_device *dev;
1967 	struct sockcm_cookie sockc;
1968 	__be16 proto = 0;
1969 	int hard_header_len;
1970 	int extra_len = 0;
1971 	int err;
1972 
1973 	/*
1974 	 *	Get and verify the address.
1975 	 */
1976 
1977 	if (saddr) {
1978 		if (msg->msg_namelen < sizeof(struct sockaddr))
1979 			return -EINVAL;
1980 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1981 			proto = saddr->spkt_protocol;
1982 	} else
1983 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1984 
1985 	/*
1986 	 *	Find the device first to size check it
1987 	 */
1988 
1989 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1990 retry:
1991 	rcu_read_lock();
1992 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1993 	err = -ENODEV;
1994 	if (dev == NULL)
1995 		goto out_unlock;
1996 
1997 	err = -ENETDOWN;
1998 	if (!(dev->flags & IFF_UP))
1999 		goto out_unlock;
2000 
2001 	/*
2002 	 * You may not queue a frame bigger than the mtu. This is the lowest level
2003 	 * raw protocol and you must do your own fragmentation at this level.
2004 	 */
2005 
2006 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2007 		if (!netif_supports_nofcs(dev)) {
2008 			err = -EPROTONOSUPPORT;
2009 			goto out_unlock;
2010 		}
2011 		extra_len = 4; /* We're doing our own CRC */
2012 	}
2013 
2014 	/* Keep the allocation-time header length across retry. */
2015 	if (!skb)
2016 		hard_header_len = READ_ONCE(dev->hard_header_len);
2017 
2018 	err = -EMSGSIZE;
2019 	if (len > dev->mtu + hard_header_len + VLAN_HLEN + extra_len)
2020 		goto out_unlock;
2021 
2022 	if (!skb) {
2023 		size_t reserved = LL_RESERVED_SPACE_EX(dev, hard_header_len);
2024 		int tlen = dev->needed_tailroom;
2025 		unsigned int hhlen = dev->header_ops ? hard_header_len : 0;
2026 
2027 		rcu_read_unlock();
2028 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2029 		if (skb == NULL)
2030 			return -ENOBUFS;
2031 		/* FIXME: Save some space for broken drivers that write a hard
2032 		 * header at transmission time by themselves. PPP is the notable
2033 		 * one here. This should really be fixed at the driver level.
2034 		 */
2035 		skb_reserve(skb, reserved);
2036 		skb_reset_network_header(skb);
2037 
2038 		/* Try to align data part correctly */
2039 		if (hhlen) {
2040 			skb->data -= hhlen;
2041 			skb->tail -= hhlen;
2042 			if (len < hhlen)
2043 				skb_reset_network_header(skb);
2044 		}
2045 		err = memcpy_from_msg(skb_put(skb, len), msg, len);
2046 		if (err)
2047 			goto out_free;
2048 		goto retry;
2049 	}
2050 
2051 	if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2052 		err = -EINVAL;
2053 		goto out_unlock;
2054 	}
2055 	if (len > (dev->mtu + hard_header_len + extra_len) &&
2056 	    !packet_extra_vlan_len_allowed(dev, skb)) {
2057 		err = -EMSGSIZE;
2058 		goto out_unlock;
2059 	}
2060 
2061 	sockcm_init(&sockc, sk);
2062 	if (msg->msg_controllen) {
2063 		err = sock_cmsg_send(sk, msg, &sockc);
2064 		if (unlikely(err))
2065 			goto out_unlock;
2066 	}
2067 
2068 	skb->protocol = proto;
2069 	skb->dev = dev;
2070 	skb->priority = sockc.priority;
2071 	skb->mark = sockc.mark;
2072 	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
2073 	skb_setup_tx_timestamp(skb, &sockc);
2074 
2075 	if (unlikely(extra_len == 4))
2076 		skb->no_fcs = 1;
2077 
2078 	packet_parse_headers(skb, sock);
2079 
2080 	dev_queue_xmit(skb);
2081 	rcu_read_unlock();
2082 	return len;
2083 
2084 out_unlock:
2085 	rcu_read_unlock();
2086 out_free:
2087 	kfree_skb(skb);
2088 	return err;
2089 }
2090 
2091 static unsigned int run_filter(struct sk_buff *skb,
2092 			       const struct sock *sk,
2093 			       unsigned int res)
2094 {
2095 	struct sk_filter *filter;
2096 
2097 	rcu_read_lock();
2098 	filter = rcu_dereference(sk->sk_filter);
2099 	if (filter != NULL)
2100 		res = bpf_prog_run_clear_cb(filter->prog, skb);
2101 	rcu_read_unlock();
2102 
2103 	return res;
2104 }
2105 
2106 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2107 			   size_t *len, int vnet_hdr_sz)
2108 {
2109 	struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2110 
2111 	if (*len < vnet_hdr_sz)
2112 		return -EINVAL;
2113 	*len -= vnet_hdr_sz;
2114 
2115 	if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2116 		return -EINVAL;
2117 
2118 	return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2119 }
2120 
2121 /*
2122  * This function makes lazy skb cloning in hope that most of packets
2123  * are discarded by BPF.
2124  *
2125  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2126  * and skb->cb are mangled. It works because (and until) packets
2127  * falling here are owned by current CPU. Output packets are cloned
2128  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2129  * sequentially, so that if we return skb to original state on exit,
2130  * we will not harm anyone.
2131  */
2132 
2133 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2134 		      struct packet_type *pt, struct net_device *orig_dev)
2135 {
2136 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2137 	struct sock *sk = NULL;
2138 	struct sockaddr_ll *sll;
2139 	struct packet_sock *po;
2140 	u8 *skb_head = skb->data;
2141 	int skb_len = skb->len;
2142 	unsigned int snaplen, res;
2143 
2144 	if (skb->pkt_type == PACKET_LOOPBACK)
2145 		goto drop;
2146 
2147 	sk = pt->af_packet_priv;
2148 	po = pkt_sk(sk);
2149 
2150 	if (!net_eq(dev_net(dev), sock_net(sk)))
2151 		goto drop;
2152 
2153 	skb->dev = dev;
2154 
2155 	if (dev_has_header(dev)) {
2156 		/* The device has an explicit notion of ll header,
2157 		 * exported to higher levels.
2158 		 *
2159 		 * Otherwise, the device hides details of its frame
2160 		 * structure, so that corresponding packet head is
2161 		 * never delivered to user.
2162 		 */
2163 		if (sk->sk_type != SOCK_DGRAM)
2164 			skb_push(skb, skb->data - skb_mac_header(skb));
2165 		else if (skb->pkt_type == PACKET_OUTGOING) {
2166 			/* Special case: outgoing packets have ll header at head */
2167 			skb_pull(skb, skb_network_offset(skb));
2168 		}
2169 	}
2170 
2171 	snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2172 
2173 	res = run_filter(skb, sk, snaplen);
2174 	if (!res)
2175 		goto drop_n_restore;
2176 	if (snaplen > res)
2177 		snaplen = res;
2178 
2179 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2180 		goto drop_n_acct;
2181 
2182 	if (skb_shared(skb)) {
2183 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2184 		if (nskb == NULL)
2185 			goto drop_n_acct;
2186 
2187 		if (skb_head != skb->data) {
2188 			skb->data = skb_head;
2189 			skb->len = skb_len;
2190 		}
2191 		consume_skb(skb);
2192 		skb = nskb;
2193 	}
2194 
2195 	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2196 
2197 	sll = &PACKET_SKB_CB(skb)->sa.ll;
2198 	sll->sll_hatype = dev->type;
2199 	sll->sll_pkttype = skb->pkt_type;
2200 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2201 		sll->sll_ifindex = orig_dev->ifindex;
2202 	else
2203 		sll->sll_ifindex = dev->ifindex;
2204 
2205 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2206 
2207 	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2208 	 * Use their space for storing the original skb length.
2209 	 */
2210 	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2211 
2212 	if (pskb_trim(skb, snaplen))
2213 		goto drop_n_acct;
2214 
2215 	skb_set_owner_r(skb, sk);
2216 	skb->dev = NULL;
2217 	skb_dst_drop(skb);
2218 
2219 	/* drop conntrack reference */
2220 	nf_reset_ct(skb);
2221 
2222 	spin_lock(&sk->sk_receive_queue.lock);
2223 	po->stats.stats1.tp_packets++;
2224 	sock_skb_set_dropcount(sk, skb);
2225 	skb_clear_delivery_time(skb);
2226 	__skb_queue_tail(&sk->sk_receive_queue, skb);
2227 	spin_unlock(&sk->sk_receive_queue.lock);
2228 	sk->sk_data_ready(sk);
2229 	return 0;
2230 
2231 drop_n_acct:
2232 	atomic_inc(&po->tp_drops);
2233 	sk_drops_inc(sk);
2234 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2235 
2236 drop_n_restore:
2237 	if (skb_head != skb->data && skb_shared(skb)) {
2238 		skb->data = skb_head;
2239 		skb->len = skb_len;
2240 	}
2241 drop:
2242 	sk_skb_reason_drop(sk, skb, drop_reason);
2243 	return 0;
2244 }
2245 
2246 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2247 		       struct packet_type *pt, struct net_device *orig_dev)
2248 {
2249 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2250 	struct sock *sk = NULL;
2251 	struct packet_sock *po;
2252 	struct sockaddr_ll *sll;
2253 	union tpacket_uhdr h;
2254 	u8 *skb_head = skb->data;
2255 	int skb_len = skb->len;
2256 	unsigned int snaplen, res;
2257 	unsigned long status = TP_STATUS_USER;
2258 	unsigned short macoff, hdrlen;
2259 	unsigned int netoff;
2260 	struct sk_buff *copy_skb = NULL;
2261 	struct timespec64 ts;
2262 	__u32 ts_status;
2263 	unsigned int slot_id = 0;
2264 	int vnet_hdr_sz = 0;
2265 
2266 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2267 	 * We may add members to them until current aligned size without forcing
2268 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2269 	 */
2270 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2271 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2272 
2273 	if (skb->pkt_type == PACKET_LOOPBACK)
2274 		goto drop;
2275 
2276 	sk = pt->af_packet_priv;
2277 	po = pkt_sk(sk);
2278 
2279 	if (!net_eq(dev_net(dev), sock_net(sk)))
2280 		goto drop;
2281 
2282 	if (dev_has_header(dev)) {
2283 		if (sk->sk_type != SOCK_DGRAM)
2284 			skb_push(skb, skb->data - skb_mac_header(skb));
2285 		else if (skb->pkt_type == PACKET_OUTGOING) {
2286 			/* Special case: outgoing packets have ll header at head */
2287 			skb_pull(skb, skb_network_offset(skb));
2288 		}
2289 	}
2290 
2291 	snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2292 
2293 	res = run_filter(skb, sk, snaplen);
2294 	if (!res)
2295 		goto drop_n_restore;
2296 
2297 	/* If we are flooded, just give up */
2298 	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2299 		atomic_inc(&po->tp_drops);
2300 		goto drop_n_restore;
2301 	}
2302 
2303 	if (skb->ip_summed == CHECKSUM_PARTIAL)
2304 		status |= TP_STATUS_CSUMNOTREADY;
2305 	else if (skb->pkt_type != PACKET_OUTGOING &&
2306 		 skb_csum_unnecessary(skb))
2307 		status |= TP_STATUS_CSUM_VALID;
2308 	if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2309 		status |= TP_STATUS_GSO_TCP;
2310 
2311 	if (snaplen > res)
2312 		snaplen = res;
2313 
2314 	if (sk->sk_type == SOCK_DGRAM) {
2315 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2316 				  po->tp_reserve;
2317 	} else {
2318 		unsigned int maclen = skb_network_offset(skb);
2319 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2320 				       (maclen < 16 ? 16 : maclen)) +
2321 				       po->tp_reserve;
2322 		vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2323 		if (vnet_hdr_sz)
2324 			netoff += vnet_hdr_sz;
2325 		macoff = netoff - maclen;
2326 	}
2327 	if (netoff > USHRT_MAX) {
2328 		atomic_inc(&po->tp_drops);
2329 		goto drop_n_restore;
2330 	}
2331 	if (po->tp_version <= TPACKET_V2) {
2332 		if (macoff + snaplen > po->rx_ring.frame_size) {
2333 			if (READ_ONCE(po->copy_thresh) &&
2334 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2335 				if (skb_shared(skb)) {
2336 					copy_skb = skb_clone(skb, GFP_ATOMIC);
2337 				} else {
2338 					copy_skb = skb_get(skb);
2339 					skb_head = skb->data;
2340 				}
2341 				if (copy_skb) {
2342 					memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2343 					       sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2344 					skb_set_owner_r(copy_skb, sk);
2345 				}
2346 			}
2347 			snaplen = po->rx_ring.frame_size - macoff;
2348 			if ((int)snaplen < 0) {
2349 				snaplen = 0;
2350 				vnet_hdr_sz = 0;
2351 			}
2352 		}
2353 	} else if (unlikely(macoff + snaplen >
2354 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2355 		u32 nval;
2356 
2357 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2358 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2359 			    snaplen, nval, macoff);
2360 		snaplen = nval;
2361 		if (unlikely((int)snaplen < 0)) {
2362 			snaplen = 0;
2363 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2364 			vnet_hdr_sz = 0;
2365 		}
2366 	}
2367 	spin_lock(&sk->sk_receive_queue.lock);
2368 	h.raw = packet_current_rx_frame(po, skb,
2369 					TP_STATUS_KERNEL, (macoff+snaplen));
2370 	if (!h.raw)
2371 		goto drop_n_account;
2372 
2373 	if (po->tp_version <= TPACKET_V2) {
2374 		slot_id = po->rx_ring.head;
2375 		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2376 			goto drop_n_account;
2377 		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2378 	}
2379 
2380 	if (vnet_hdr_sz &&
2381 	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2382 				    sizeof(struct virtio_net_hdr),
2383 				    vio_le(), true, 0)) {
2384 		if (po->tp_version == TPACKET_V3)
2385 			prb_clear_blk_fill_status(&po->rx_ring);
2386 		goto drop_n_account;
2387 	}
2388 
2389 	if (po->tp_version <= TPACKET_V2) {
2390 		packet_increment_rx_head(po, &po->rx_ring);
2391 	/*
2392 	 * LOSING will be reported till you read the stats,
2393 	 * because it's COR - Clear On Read.
2394 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2395 	 * at packet level.
2396 	 */
2397 		if (atomic_read(&po->tp_drops))
2398 			status |= TP_STATUS_LOSING;
2399 	}
2400 
2401 	po->stats.stats1.tp_packets++;
2402 	if (copy_skb) {
2403 		status |= TP_STATUS_COPY;
2404 		skb_clear_delivery_time(copy_skb);
2405 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2406 	}
2407 	spin_unlock(&sk->sk_receive_queue.lock);
2408 
2409 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2410 
2411 	/* Always timestamp; prefer an existing software timestamp taken
2412 	 * closer to the time of capture.
2413 	 */
2414 	ts_status = tpacket_get_timestamp(skb, &ts,
2415 					  READ_ONCE(po->tp_tstamp) |
2416 					  SOF_TIMESTAMPING_SOFTWARE);
2417 	if (!ts_status)
2418 		ktime_get_real_ts64(&ts);
2419 
2420 	status |= ts_status;
2421 
2422 	switch (po->tp_version) {
2423 	case TPACKET_V1:
2424 		h.h1->tp_len = skb->len;
2425 		h.h1->tp_snaplen = snaplen;
2426 		h.h1->tp_mac = macoff;
2427 		h.h1->tp_net = netoff;
2428 		h.h1->tp_sec = ts.tv_sec;
2429 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2430 		hdrlen = sizeof(*h.h1);
2431 		break;
2432 	case TPACKET_V2:
2433 		h.h2->tp_len = skb->len;
2434 		h.h2->tp_snaplen = snaplen;
2435 		h.h2->tp_mac = macoff;
2436 		h.h2->tp_net = netoff;
2437 		h.h2->tp_sec = ts.tv_sec;
2438 		h.h2->tp_nsec = ts.tv_nsec;
2439 		if (skb_vlan_tag_present(skb)) {
2440 			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2441 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2442 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2443 		} else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
2444 			h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev);
2445 			h.h2->tp_vlan_tpid = ntohs(skb->protocol);
2446 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2447 		} else {
2448 			h.h2->tp_vlan_tci = 0;
2449 			h.h2->tp_vlan_tpid = 0;
2450 		}
2451 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2452 		hdrlen = sizeof(*h.h2);
2453 		break;
2454 	case TPACKET_V3:
2455 		/* tp_nxt_offset,vlan are already populated above.
2456 		 * So DONT clear those fields here
2457 		 */
2458 		h.h3->tp_status |= status;
2459 		h.h3->tp_len = skb->len;
2460 		h.h3->tp_snaplen = snaplen;
2461 		h.h3->tp_mac = macoff;
2462 		h.h3->tp_net = netoff;
2463 		h.h3->tp_sec  = ts.tv_sec;
2464 		h.h3->tp_nsec = ts.tv_nsec;
2465 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2466 		hdrlen = sizeof(*h.h3);
2467 		break;
2468 	default:
2469 		BUG();
2470 	}
2471 
2472 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2473 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2474 	sll->sll_family = AF_PACKET;
2475 	sll->sll_hatype = dev->type;
2476 	sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ?
2477 		vlan_get_protocol_dgram(skb) : skb->protocol;
2478 	sll->sll_pkttype = skb->pkt_type;
2479 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2480 		sll->sll_ifindex = orig_dev->ifindex;
2481 	else
2482 		sll->sll_ifindex = dev->ifindex;
2483 
2484 	smp_mb();
2485 
2486 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2487 	if (po->tp_version <= TPACKET_V2) {
2488 		u8 *start, *end;
2489 
2490 		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2491 					macoff + snaplen);
2492 
2493 		for (start = h.raw; start < end; start += PAGE_SIZE)
2494 			flush_dcache_page(pgv_to_page(start));
2495 	}
2496 	smp_wmb();
2497 #endif
2498 
2499 	if (po->tp_version <= TPACKET_V2) {
2500 		spin_lock(&sk->sk_receive_queue.lock);
2501 		__packet_set_status(po, h.raw, status);
2502 		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2503 		spin_unlock(&sk->sk_receive_queue.lock);
2504 		sk->sk_data_ready(sk);
2505 	} else if (po->tp_version == TPACKET_V3) {
2506 		prb_clear_blk_fill_status(&po->rx_ring);
2507 	}
2508 
2509 drop_n_restore:
2510 	if (skb_head != skb->data && skb_shared(skb)) {
2511 		skb->data = skb_head;
2512 		skb->len = skb_len;
2513 	}
2514 drop:
2515 	sk_skb_reason_drop(sk, skb, drop_reason);
2516 	return 0;
2517 
2518 drop_n_account:
2519 	spin_unlock(&sk->sk_receive_queue.lock);
2520 	atomic_inc(&po->tp_drops);
2521 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2522 
2523 	sk->sk_data_ready(sk);
2524 	sk_skb_reason_drop(sk, copy_skb, drop_reason);
2525 	goto drop_n_restore;
2526 }
2527 
2528 static void tpacket_destruct_skb(struct sk_buff *skb)
2529 {
2530 	struct packet_sock *po = pkt_sk(skb->sk);
2531 
2532 	if (likely(po->tx_ring.pg_vec)) {
2533 		void *ph;
2534 		__u32 ts;
2535 
2536 		ph = skb_zcopy_get_nouarg(skb);
2537 		packet_dec_pending(&po->tx_ring);
2538 
2539 		ts = __packet_set_timestamp(po, ph, skb);
2540 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2541 
2542 		complete(&po->skb_completion);
2543 	}
2544 
2545 	sock_wfree(skb);
2546 }
2547 
2548 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2549 {
2550 	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2551 	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2552 	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2553 	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2554 		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2555 			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2556 			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2557 
2558 	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2559 		return -EINVAL;
2560 
2561 	return 0;
2562 }
2563 
2564 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2565 				 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2566 {
2567 	int ret;
2568 
2569 	if (*len < vnet_hdr_sz)
2570 		return -EINVAL;
2571 	*len -= vnet_hdr_sz;
2572 
2573 	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2574 		return -EFAULT;
2575 
2576 	ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2577 	if (ret)
2578 		return ret;
2579 
2580 	/* move iter to point to the start of mac header */
2581 	if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2582 		iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2583 
2584 	return 0;
2585 }
2586 
2587 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2588 		void *frame, struct net_device *dev, void *data, int tp_len,
2589 		__be16 proto, unsigned char *addr, int hlen, int copylen,
2590 		int hard_header_len,
2591 		const struct sockcm_cookie *sockc)
2592 {
2593 	union tpacket_uhdr ph;
2594 	int to_write, offset, len, nr_frags, len_max;
2595 	struct socket *sock = po->sk.sk_socket;
2596 	struct page *page;
2597 	int err;
2598 
2599 	ph.raw = frame;
2600 
2601 	skb->protocol = proto;
2602 	skb->dev = dev;
2603 	skb->priority = sockc->priority;
2604 	skb->mark = sockc->mark;
2605 	skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid);
2606 	skb_setup_tx_timestamp(skb, sockc);
2607 	skb_zcopy_set_nouarg(skb, ph.raw);
2608 
2609 	skb_reserve(skb, hlen);
2610 	skb_reset_network_header(skb);
2611 
2612 	to_write = tp_len;
2613 
2614 	if (sock->type == SOCK_DGRAM) {
2615 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2616 				NULL, tp_len);
2617 		if (unlikely(err < 0))
2618 			return -EINVAL;
2619 	} else if (copylen) {
2620 		int hdrlen = min_t(int, copylen, tp_len);
2621 
2622 		skb_push(skb, hard_header_len);
2623 		skb_put(skb, copylen - hard_header_len);
2624 		err = skb_store_bits(skb, 0, data, hdrlen);
2625 		if (unlikely(err))
2626 			return err;
2627 		if (!dev_validate_header(dev, skb->data, hdrlen))
2628 			return -EINVAL;
2629 
2630 		data += hdrlen;
2631 		to_write -= hdrlen;
2632 	}
2633 
2634 	offset = offset_in_page(data);
2635 	len_max = PAGE_SIZE - offset;
2636 	len = ((to_write > len_max) ? len_max : to_write);
2637 
2638 	skb->data_len = to_write;
2639 	skb->len += to_write;
2640 	skb->truesize += to_write;
2641 	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2642 
2643 	while (likely(to_write)) {
2644 		nr_frags = skb_shinfo(skb)->nr_frags;
2645 
2646 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2647 			pr_err("Packet exceed the number of skb frags(%u)\n",
2648 			       (unsigned int)MAX_SKB_FRAGS);
2649 			return -EFAULT;
2650 		}
2651 
2652 		page = pgv_to_page(data);
2653 		data += len;
2654 		flush_dcache_page(page);
2655 		get_page(page);
2656 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2657 		to_write -= len;
2658 		offset = 0;
2659 		len_max = PAGE_SIZE;
2660 		len = ((to_write > len_max) ? len_max : to_write);
2661 	}
2662 
2663 	packet_parse_headers(skb, sock);
2664 
2665 	return tp_len;
2666 }
2667 
2668 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2669 				int size_max, void **data)
2670 {
2671 	union tpacket_uhdr ph;
2672 	int tp_len, off;
2673 
2674 	ph.raw = frame;
2675 
2676 	switch (po->tp_version) {
2677 	case TPACKET_V3:
2678 		if (ph.h3->tp_next_offset != 0) {
2679 			pr_warn_once("variable sized slot not supported");
2680 			return -EINVAL;
2681 		}
2682 		tp_len = ph.h3->tp_len;
2683 		break;
2684 	case TPACKET_V2:
2685 		tp_len = ph.h2->tp_len;
2686 		break;
2687 	default:
2688 		tp_len = ph.h1->tp_len;
2689 		break;
2690 	}
2691 	if (unlikely(tp_len > size_max)) {
2692 		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2693 		return -EMSGSIZE;
2694 	}
2695 
2696 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2697 		int off_min, off_max;
2698 
2699 		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2700 		off_max = po->tx_ring.frame_size - tp_len;
2701 		if (po->sk.sk_type == SOCK_DGRAM) {
2702 			switch (po->tp_version) {
2703 			case TPACKET_V3:
2704 				off = ph.h3->tp_net;
2705 				break;
2706 			case TPACKET_V2:
2707 				off = ph.h2->tp_net;
2708 				break;
2709 			default:
2710 				off = ph.h1->tp_net;
2711 				break;
2712 			}
2713 		} else {
2714 			switch (po->tp_version) {
2715 			case TPACKET_V3:
2716 				off = ph.h3->tp_mac;
2717 				break;
2718 			case TPACKET_V2:
2719 				off = ph.h2->tp_mac;
2720 				break;
2721 			default:
2722 				off = ph.h1->tp_mac;
2723 				break;
2724 			}
2725 		}
2726 		if (unlikely((off < off_min) || (off_max < off)))
2727 			return -EINVAL;
2728 	} else {
2729 		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2730 	}
2731 
2732 	*data = frame + off;
2733 	return tp_len;
2734 }
2735 
2736 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2737 {
2738 	struct sk_buff *skb = NULL;
2739 	struct net_device *dev;
2740 	struct virtio_net_hdr vnet_hdr;
2741 	bool has_vnet_hdr = false;
2742 	struct sockcm_cookie sockc;
2743 	__be16 proto;
2744 	int err, reserve = 0;
2745 	void *ph;
2746 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2747 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2748 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2749 	unsigned char *addr = NULL;
2750 	int tp_len, size_max;
2751 	void *data;
2752 	int len_sum = 0;
2753 	int status = TP_STATUS_AVAILABLE;
2754 	int hard_header_len, hlen, tlen, copylen = 0;
2755 	long timeo;
2756 
2757 	mutex_lock(&po->pg_vec_lock);
2758 
2759 	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2760 	 * we need to confirm it under protection of pg_vec_lock.
2761 	 */
2762 	if (unlikely(!po->tx_ring.pg_vec)) {
2763 		err = -EBUSY;
2764 		goto out;
2765 	}
2766 	if (likely(saddr == NULL)) {
2767 		dev	= packet_cached_dev_get(po);
2768 		proto	= READ_ONCE(po->num);
2769 	} else {
2770 		err = -EINVAL;
2771 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2772 			goto out;
2773 		if (msg->msg_namelen < (saddr->sll_halen
2774 					+ offsetof(struct sockaddr_ll,
2775 						sll_addr)))
2776 			goto out;
2777 		proto	= saddr->sll_protocol;
2778 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2779 		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2780 			if (dev && msg->msg_namelen < dev->addr_len +
2781 				   offsetof(struct sockaddr_ll, sll_addr))
2782 				goto out_put;
2783 			addr = saddr->sll_addr;
2784 		}
2785 	}
2786 
2787 	err = -ENXIO;
2788 	if (unlikely(dev == NULL))
2789 		goto out;
2790 	err = -ENETDOWN;
2791 	if (unlikely(!(dev->flags & IFF_UP)))
2792 		goto out_put;
2793 
2794 	sockcm_init(&sockc, &po->sk);
2795 	if (msg->msg_controllen) {
2796 		err = sock_cmsg_send(&po->sk, msg, &sockc);
2797 		if (unlikely(err))
2798 			goto out_put;
2799 	}
2800 
2801 	hard_header_len = READ_ONCE(dev->hard_header_len);
2802 	if (po->sk.sk_socket->type == SOCK_RAW)
2803 		reserve = hard_header_len;
2804 	size_max = po->tx_ring.frame_size
2805 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2806 
2807 	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2808 		size_max = dev->mtu + reserve + VLAN_HLEN;
2809 
2810 	timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2811 	reinit_completion(&po->skb_completion);
2812 
2813 	do {
2814 		ph = packet_current_frame(po, &po->tx_ring,
2815 					  TP_STATUS_SEND_REQUEST);
2816 		if (unlikely(ph == NULL)) {
2817 			/* Note: packet_read_pending() might be slow if we
2818 			 * have to call it as it's per_cpu variable, but in
2819 			 * fast-path we don't have to call it, only when ph
2820 			 * is NULL, we need to check the pending_refcnt.
2821 			 */
2822 			if (need_wait && packet_read_pending(&po->tx_ring)) {
2823 				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2824 				if (timeo <= 0) {
2825 					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2826 					goto out_put;
2827 				}
2828 				/* check for additional frames */
2829 				continue;
2830 			} else
2831 				break;
2832 		}
2833 
2834 		skb = NULL;
2835 		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2836 		if (tp_len < 0)
2837 			goto tpacket_error;
2838 
2839 		status = TP_STATUS_SEND_REQUEST;
2840 		hlen = LL_RESERVED_SPACE_EX(dev, hard_header_len);
2841 		tlen = dev->needed_tailroom;
2842 		if (vnet_hdr_sz) {
2843 			data += vnet_hdr_sz;
2844 			tp_len -= vnet_hdr_sz;
2845 			if (tp_len < 0) {
2846 				tp_len = -EINVAL;
2847 				goto tpacket_error;
2848 			}
2849 			memcpy(&vnet_hdr, data - vnet_hdr_sz, sizeof(vnet_hdr));
2850 			if (__packet_snd_vnet_parse(&vnet_hdr, tp_len)) {
2851 				tp_len = -EINVAL;
2852 				goto tpacket_error;
2853 			}
2854 			copylen = __virtio16_to_cpu(vio_le(),
2855 						    vnet_hdr.hdr_len);
2856 			has_vnet_hdr = true;
2857 		}
2858 		copylen = max_t(int, copylen, hard_header_len);
2859 		skb = sock_alloc_send_skb(&po->sk,
2860 				hlen + tlen + sizeof(struct sockaddr_ll) +
2861 				(copylen - hard_header_len),
2862 				!need_wait, &err);
2863 
2864 		if (unlikely(skb == NULL)) {
2865 			/* we assume the socket was initially writeable ... */
2866 			if (likely(len_sum > 0))
2867 				err = len_sum;
2868 			goto out_status;
2869 		}
2870 		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2871 					  addr, hlen, copylen, hard_header_len,
2872 					  &sockc);
2873 		if (likely(tp_len >= 0) &&
2874 		    tp_len > dev->mtu + reserve &&
2875 		    !vnet_hdr_sz &&
2876 		    !packet_extra_vlan_len_allowed(dev, skb))
2877 			tp_len = -EMSGSIZE;
2878 
2879 		if (unlikely(tp_len < 0)) {
2880 tpacket_error:
2881 			if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2882 				__packet_set_status(po, ph,
2883 						TP_STATUS_AVAILABLE);
2884 				packet_increment_head(&po->tx_ring);
2885 				kfree_skb(skb);
2886 				continue;
2887 			} else {
2888 				status = TP_STATUS_WRONG_FORMAT;
2889 				err = tp_len;
2890 				goto out_status;
2891 			}
2892 		}
2893 
2894 		if (has_vnet_hdr) {
2895 			if (virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le())) {
2896 				tp_len = -EINVAL;
2897 				goto tpacket_error;
2898 			}
2899 			virtio_net_hdr_set_proto(skb, &vnet_hdr);
2900 		}
2901 
2902 		skb->destructor = tpacket_destruct_skb;
2903 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2904 		packet_inc_pending(&po->tx_ring);
2905 
2906 		status = TP_STATUS_SEND_REQUEST;
2907 		err = packet_xmit(po, skb);
2908 		if (unlikely(err != 0)) {
2909 			if (err > 0)
2910 				err = net_xmit_errno(err);
2911 			if (err && __packet_get_status(po, ph) ==
2912 				   TP_STATUS_AVAILABLE) {
2913 				/* skb was destructed already */
2914 				skb = NULL;
2915 				goto out_status;
2916 			}
2917 			/*
2918 			 * skb was dropped but not destructed yet;
2919 			 * let's treat it like congestion or err < 0
2920 			 */
2921 			err = 0;
2922 		}
2923 		packet_increment_head(&po->tx_ring);
2924 		len_sum += tp_len;
2925 	} while (1);
2926 
2927 	err = len_sum;
2928 	goto out_put;
2929 
2930 out_status:
2931 	__packet_set_status(po, ph, status);
2932 	kfree_skb(skb);
2933 out_put:
2934 	dev_put(dev);
2935 out:
2936 	mutex_unlock(&po->pg_vec_lock);
2937 	return err;
2938 }
2939 
2940 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2941 				        size_t reserve, size_t len,
2942 				        size_t linear, int noblock,
2943 				        int *err)
2944 {
2945 	struct sk_buff *skb;
2946 
2947 	/* Under a page?  Don't bother with paged skb. */
2948 	if (prepad + len < PAGE_SIZE || !linear)
2949 		linear = len;
2950 
2951 	if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2952 		linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2953 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2954 				   err, PAGE_ALLOC_COSTLY_ORDER);
2955 	if (!skb)
2956 		return NULL;
2957 
2958 	skb_reserve(skb, reserve);
2959 	skb_put(skb, linear);
2960 	skb->data_len = len - linear;
2961 	skb->len += len - linear;
2962 
2963 	return skb;
2964 }
2965 
2966 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2967 {
2968 	struct sock *sk = sock->sk;
2969 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2970 	struct sk_buff *skb;
2971 	struct net_device *dev;
2972 	__be16 proto;
2973 	unsigned char *addr = NULL;
2974 	int err, reserve = 0;
2975 	struct sockcm_cookie sockc;
2976 	struct virtio_net_hdr vnet_hdr = { 0 };
2977 	int offset = 0;
2978 	struct packet_sock *po = pkt_sk(sk);
2979 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2980 	int hard_header_len, hlen, tlen, linear;
2981 	int extra_len = 0;
2982 
2983 	/*
2984 	 *	Get and verify the address.
2985 	 */
2986 
2987 	if (likely(saddr == NULL)) {
2988 		dev	= packet_cached_dev_get(po);
2989 		proto	= READ_ONCE(po->num);
2990 	} else {
2991 		err = -EINVAL;
2992 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2993 			goto out;
2994 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2995 			goto out;
2996 		proto	= saddr->sll_protocol;
2997 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2998 		if (sock->type == SOCK_DGRAM) {
2999 			if (dev && msg->msg_namelen < dev->addr_len +
3000 				   offsetof(struct sockaddr_ll, sll_addr))
3001 				goto out_unlock;
3002 			addr = saddr->sll_addr;
3003 		}
3004 	}
3005 
3006 	err = -ENXIO;
3007 	if (unlikely(dev == NULL))
3008 		goto out_unlock;
3009 	err = -ENETDOWN;
3010 	if (unlikely(!(dev->flags & IFF_UP)))
3011 		goto out_unlock;
3012 
3013 	sockcm_init(&sockc, sk);
3014 	if (msg->msg_controllen) {
3015 		err = sock_cmsg_send(sk, msg, &sockc);
3016 		if (unlikely(err))
3017 			goto out_unlock;
3018 	}
3019 
3020 	hard_header_len = READ_ONCE(dev->hard_header_len);
3021 	if (sock->type == SOCK_RAW)
3022 		reserve = hard_header_len;
3023 	if (vnet_hdr_sz) {
3024 		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3025 		if (err)
3026 			goto out_unlock;
3027 	}
3028 
3029 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3030 		if (!netif_supports_nofcs(dev)) {
3031 			err = -EPROTONOSUPPORT;
3032 			goto out_unlock;
3033 		}
3034 		extra_len = 4; /* We're doing our own CRC */
3035 	}
3036 
3037 	err = -EMSGSIZE;
3038 	if (!vnet_hdr.gso_type &&
3039 	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3040 		goto out_unlock;
3041 
3042 	err = -ENOBUFS;
3043 	hlen = LL_RESERVED_SPACE_EX(dev, hard_header_len);
3044 	tlen = dev->needed_tailroom;
3045 	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3046 	linear = max(linear, min_t(int, len, hard_header_len));
3047 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3048 			       msg->msg_flags & MSG_DONTWAIT, &err);
3049 	if (skb == NULL)
3050 		goto out_unlock;
3051 
3052 	skb_reset_network_header(skb);
3053 
3054 	err = -EINVAL;
3055 	if (sock->type == SOCK_DGRAM) {
3056 		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3057 		if (unlikely(offset < 0))
3058 			goto out_free;
3059 	} else if (reserve) {
3060 		skb_reserve(skb, -reserve);
3061 		if (len < reserve + sizeof(struct ipv6hdr) &&
3062 		    dev->min_header_len != hard_header_len)
3063 			skb_reset_network_header(skb);
3064 	}
3065 
3066 	/* Returns -EFAULT on error */
3067 	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3068 	if (err)
3069 		goto out_free;
3070 
3071 	if ((sock->type == SOCK_RAW &&
3072 	     !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3073 		err = -EINVAL;
3074 		goto out_free;
3075 	}
3076 
3077 	skb_setup_tx_timestamp(skb, &sockc);
3078 
3079 	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3080 	    !packet_extra_vlan_len_allowed(dev, skb)) {
3081 		err = -EMSGSIZE;
3082 		goto out_free;
3083 	}
3084 
3085 	skb->protocol = proto;
3086 	skb->dev = dev;
3087 	skb->priority = sockc.priority;
3088 	skb->mark = sockc.mark;
3089 	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
3090 
3091 	if (unlikely(extra_len == 4))
3092 		skb->no_fcs = 1;
3093 
3094 	packet_parse_headers(skb, sock);
3095 
3096 	if (vnet_hdr_sz) {
3097 		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3098 		if (err)
3099 			goto out_free;
3100 		len += vnet_hdr_sz;
3101 		virtio_net_hdr_set_proto(skb, &vnet_hdr);
3102 	}
3103 
3104 	err = packet_xmit(po, skb);
3105 
3106 	if (unlikely(err != 0)) {
3107 		if (err > 0)
3108 			err = net_xmit_errno(err);
3109 		if (err)
3110 			goto out_unlock;
3111 	}
3112 
3113 	dev_put(dev);
3114 
3115 	return len;
3116 
3117 out_free:
3118 	kfree_skb(skb);
3119 out_unlock:
3120 	dev_put(dev);
3121 out:
3122 	return err;
3123 }
3124 
3125 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3126 {
3127 	struct sock *sk = sock->sk;
3128 	struct packet_sock *po = pkt_sk(sk);
3129 
3130 	/* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3131 	 * tpacket_snd() will redo the check safely.
3132 	 */
3133 	if (data_race(po->tx_ring.pg_vec))
3134 		return tpacket_snd(po, msg);
3135 
3136 	return packet_snd(sock, msg, len);
3137 }
3138 
3139 /*
3140  *	Close a PACKET socket. This is fairly simple. We immediately go
3141  *	to 'closed' state and remove our protocol entry in the device list.
3142  */
3143 
3144 static int packet_release(struct socket *sock)
3145 {
3146 	struct sock *sk = sock->sk;
3147 	struct packet_sock *po;
3148 	struct packet_fanout *f;
3149 	struct net *net;
3150 	union tpacket_req_u req_u;
3151 
3152 	if (!sk)
3153 		return 0;
3154 
3155 	net = sock_net(sk);
3156 	po = pkt_sk(sk);
3157 
3158 	mutex_lock(&net->packet.sklist_lock);
3159 	sk_del_node_init_rcu(sk);
3160 	mutex_unlock(&net->packet.sklist_lock);
3161 
3162 	sock_prot_inuse_add(net, sk->sk_prot, -1);
3163 
3164 	spin_lock(&po->bind_lock);
3165 	unregister_prot_hook(sk, false);
3166 	WRITE_ONCE(po->num, 0);
3167 	packet_cached_dev_reset(po);
3168 
3169 	if (po->prot_hook.dev) {
3170 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3171 		po->prot_hook.dev = NULL;
3172 	}
3173 	spin_unlock(&po->bind_lock);
3174 
3175 	packet_flush_mclist(sk);
3176 
3177 	lock_sock(sk);
3178 	if (po->rx_ring.pg_vec) {
3179 		memset(&req_u, 0, sizeof(req_u));
3180 		packet_set_ring(sk, &req_u, 1, 0);
3181 	}
3182 
3183 	if (po->tx_ring.pg_vec) {
3184 		memset(&req_u, 0, sizeof(req_u));
3185 		packet_set_ring(sk, &req_u, 1, 1);
3186 	}
3187 	release_sock(sk);
3188 
3189 	f = fanout_release(sk);
3190 
3191 	synchronize_net();
3192 
3193 	kfree(po->rollover);
3194 	if (f) {
3195 		fanout_release_data(f);
3196 		kvfree(f);
3197 	}
3198 	/*
3199 	 *	Now the socket is dead. No more input will appear.
3200 	 */
3201 	sock_orphan(sk);
3202 	sock->sk = NULL;
3203 
3204 	/* Purge queues */
3205 
3206 	skb_queue_purge(&sk->sk_receive_queue);
3207 	packet_free_pending(po);
3208 
3209 	sock_put(sk);
3210 	return 0;
3211 }
3212 
3213 /*
3214  *	Attach a packet hook.
3215  */
3216 
3217 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3218 			  __be16 proto)
3219 {
3220 	struct packet_sock *po = pkt_sk(sk);
3221 	struct net_device *dev = NULL;
3222 	bool unlisted = false;
3223 	bool need_rehook;
3224 	int ret = 0;
3225 
3226 	lock_sock(sk);
3227 	spin_lock(&po->bind_lock);
3228 	if (!proto)
3229 		proto = po->num;
3230 
3231 	rcu_read_lock();
3232 
3233 	if (po->fanout) {
3234 		ret = -EINVAL;
3235 		goto out_unlock;
3236 	}
3237 
3238 	if (name) {
3239 		dev = dev_get_by_name_rcu(sock_net(sk), name);
3240 		if (!dev) {
3241 			ret = -ENODEV;
3242 			goto out_unlock;
3243 		}
3244 	} else if (ifindex) {
3245 		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3246 		if (!dev) {
3247 			ret = -ENODEV;
3248 			goto out_unlock;
3249 		}
3250 	}
3251 
3252 	need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3253 
3254 	if (need_rehook) {
3255 		dev_hold(dev);
3256 		if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3257 			rcu_read_unlock();
3258 			/* prevents packet_notifier() from calling
3259 			 * register_prot_hook()
3260 			 */
3261 			WRITE_ONCE(po->num, 0);
3262 			__unregister_prot_hook(sk, true);
3263 			rcu_read_lock();
3264 			if (dev)
3265 				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3266 								 dev->ifindex);
3267 		}
3268 
3269 		BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3270 		WRITE_ONCE(po->num, proto);
3271 		po->prot_hook.type = proto;
3272 
3273 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3274 
3275 		if (unlikely(unlisted)) {
3276 			po->prot_hook.dev = NULL;
3277 			WRITE_ONCE(po->ifindex, -1);
3278 			packet_cached_dev_reset(po);
3279 		} else {
3280 			netdev_hold(dev, &po->prot_hook.dev_tracker,
3281 				    GFP_ATOMIC);
3282 			po->prot_hook.dev = dev;
3283 			WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3284 			packet_cached_dev_assign(po, dev);
3285 		}
3286 		dev_put(dev);
3287 	}
3288 
3289 	if (proto == 0 || !need_rehook)
3290 		goto out_unlock;
3291 
3292 	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3293 		register_prot_hook(sk);
3294 	} else {
3295 		sk->sk_err = ENETDOWN;
3296 		if (!sock_flag(sk, SOCK_DEAD))
3297 			sk_error_report(sk);
3298 	}
3299 
3300 out_unlock:
3301 	rcu_read_unlock();
3302 	spin_unlock(&po->bind_lock);
3303 	release_sock(sk);
3304 	return ret;
3305 }
3306 
3307 /*
3308  *	Bind a packet socket to a device
3309  */
3310 
3311 static int packet_bind_spkt(struct socket *sock, struct sockaddr_unsized *uaddr,
3312 			    int addr_len)
3313 {
3314 	struct sock *sk = sock->sk;
3315 	struct sockaddr *sa = (struct sockaddr *)uaddr;
3316 	char name[sizeof(sa->sa_data) + 1];
3317 
3318 	/*
3319 	 *	Check legality
3320 	 */
3321 
3322 	if (addr_len != sizeof(struct sockaddr))
3323 		return -EINVAL;
3324 	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3325 	 * zero-terminated.
3326 	 */
3327 	memcpy(name, sa->sa_data, sizeof(sa->sa_data));
3328 	name[sizeof(sa->sa_data)] = 0;
3329 
3330 	return packet_do_bind(sk, name, 0, 0);
3331 }
3332 
3333 static int packet_bind(struct socket *sock, struct sockaddr_unsized *uaddr, int addr_len)
3334 {
3335 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3336 	struct sock *sk = sock->sk;
3337 
3338 	/*
3339 	 *	Check legality
3340 	 */
3341 
3342 	if (addr_len < sizeof(struct sockaddr_ll))
3343 		return -EINVAL;
3344 	if (sll->sll_family != AF_PACKET)
3345 		return -EINVAL;
3346 
3347 	return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3348 }
3349 
3350 static struct proto packet_proto = {
3351 	.name	  = "PACKET",
3352 	.owner	  = THIS_MODULE,
3353 	.obj_size = sizeof(struct packet_sock),
3354 };
3355 
3356 /*
3357  *	Create a packet of type SOCK_PACKET.
3358  */
3359 
3360 static int packet_create(struct net *net, struct socket *sock, int protocol,
3361 			 int kern)
3362 {
3363 	struct sock *sk;
3364 	struct packet_sock *po;
3365 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3366 	int err;
3367 
3368 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3369 		return -EPERM;
3370 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3371 	    sock->type != SOCK_PACKET)
3372 		return -ESOCKTNOSUPPORT;
3373 
3374 	sock->state = SS_UNCONNECTED;
3375 
3376 	err = -ENOBUFS;
3377 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3378 	if (sk == NULL)
3379 		goto out;
3380 
3381 	sock->ops = &packet_ops;
3382 	if (sock->type == SOCK_PACKET)
3383 		sock->ops = &packet_ops_spkt;
3384 
3385 	po = pkt_sk(sk);
3386 	err = packet_alloc_pending(po);
3387 	if (err)
3388 		goto out_sk_free;
3389 
3390 	sock_init_data(sock, sk);
3391 
3392 	init_completion(&po->skb_completion);
3393 	sk->sk_family = PF_PACKET;
3394 	po->num = proto;
3395 
3396 	packet_cached_dev_reset(po);
3397 
3398 	sk->sk_destruct = packet_sock_destruct;
3399 
3400 	/*
3401 	 *	Attach a protocol block
3402 	 */
3403 
3404 	spin_lock_init(&po->bind_lock);
3405 	mutex_init(&po->pg_vec_lock);
3406 	po->rollover = NULL;
3407 	po->prot_hook.func = packet_rcv;
3408 
3409 	if (sock->type == SOCK_PACKET)
3410 		po->prot_hook.func = packet_rcv_spkt;
3411 
3412 	po->prot_hook.af_packet_priv = sk;
3413 	po->prot_hook.af_packet_net = sock_net(sk);
3414 
3415 	if (proto) {
3416 		po->prot_hook.type = proto;
3417 		__register_prot_hook(sk);
3418 	}
3419 
3420 	mutex_lock(&net->packet.sklist_lock);
3421 	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3422 	mutex_unlock(&net->packet.sklist_lock);
3423 
3424 	sock_prot_inuse_add(net, &packet_proto, 1);
3425 
3426 	return 0;
3427 out_sk_free:
3428 	sk_free(sk);
3429 out:
3430 	return err;
3431 }
3432 
3433 /*
3434  *	Pull a packet from our receive queue and hand it to the user.
3435  *	If necessary we block.
3436  */
3437 
3438 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3439 			  int flags)
3440 {
3441 	struct sock *sk = sock->sk;
3442 	struct sk_buff *skb;
3443 	int copied, err;
3444 	int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3445 	unsigned int origlen = 0;
3446 
3447 	err = -EINVAL;
3448 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3449 		goto out;
3450 
3451 #if 0
3452 	/* What error should we return now? EUNATTACH? */
3453 	if (pkt_sk(sk)->ifindex < 0)
3454 		return -ENODEV;
3455 #endif
3456 
3457 	if (flags & MSG_ERRQUEUE) {
3458 		err = sock_recv_errqueue(sk, msg, len,
3459 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3460 		goto out;
3461 	}
3462 
3463 	/*
3464 	 *	Call the generic datagram receiver. This handles all sorts
3465 	 *	of horrible races and re-entrancy so we can forget about it
3466 	 *	in the protocol layers.
3467 	 *
3468 	 *	Now it will return ENETDOWN, if device have just gone down,
3469 	 *	but then it will block.
3470 	 */
3471 
3472 	skb = skb_recv_datagram(sk, flags, &err);
3473 
3474 	/*
3475 	 *	An error occurred so return it. Because skb_recv_datagram()
3476 	 *	handles the blocking we don't see and worry about blocking
3477 	 *	retries.
3478 	 */
3479 
3480 	if (skb == NULL)
3481 		goto out;
3482 
3483 	packet_rcv_try_clear_pressure(pkt_sk(sk));
3484 
3485 	if (vnet_hdr_len) {
3486 		err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3487 		if (err)
3488 			goto out_free;
3489 	}
3490 
3491 	/* You lose any data beyond the buffer you gave. If it worries
3492 	 * a user program they can ask the device for its MTU
3493 	 * anyway.
3494 	 */
3495 	copied = skb->len;
3496 	if (copied > len) {
3497 		copied = len;
3498 		msg->msg_flags |= MSG_TRUNC;
3499 	}
3500 
3501 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3502 	if (err)
3503 		goto out_free;
3504 
3505 	if (sock->type != SOCK_PACKET) {
3506 		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3507 
3508 		/* Original length was stored in sockaddr_ll fields */
3509 		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3510 		sll->sll_family = AF_PACKET;
3511 		sll->sll_protocol = (sock->type == SOCK_DGRAM) ?
3512 			vlan_get_protocol_dgram(skb) : skb->protocol;
3513 	}
3514 
3515 	sock_recv_cmsgs(msg, sk, skb);
3516 
3517 	if (msg->msg_name) {
3518 		const size_t max_len = min(sizeof(skb->cb),
3519 					   sizeof(struct sockaddr_storage));
3520 		int copy_len;
3521 
3522 		/* If the address length field is there to be filled
3523 		 * in, we fill it in now.
3524 		 */
3525 		if (sock->type == SOCK_PACKET) {
3526 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3527 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3528 			copy_len = msg->msg_namelen;
3529 		} else {
3530 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3531 
3532 			msg->msg_namelen = sll->sll_halen +
3533 				offsetof(struct sockaddr_ll, sll_addr);
3534 			copy_len = msg->msg_namelen;
3535 			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3536 				memset(msg->msg_name +
3537 				       offsetof(struct sockaddr_ll, sll_addr),
3538 				       0, sizeof(sll->sll_addr));
3539 				msg->msg_namelen = sizeof(struct sockaddr_ll);
3540 			}
3541 		}
3542 		if (WARN_ON_ONCE(copy_len > max_len)) {
3543 			copy_len = max_len;
3544 			msg->msg_namelen = copy_len;
3545 		}
3546 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3547 	}
3548 
3549 	if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3550 		struct tpacket_auxdata aux;
3551 
3552 		aux.tp_status = TP_STATUS_USER;
3553 		if (skb->ip_summed == CHECKSUM_PARTIAL)
3554 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3555 		else if (skb->pkt_type != PACKET_OUTGOING &&
3556 			 skb_csum_unnecessary(skb))
3557 			aux.tp_status |= TP_STATUS_CSUM_VALID;
3558 		if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3559 			aux.tp_status |= TP_STATUS_GSO_TCP;
3560 
3561 		aux.tp_len = origlen;
3562 		aux.tp_snaplen = skb->len;
3563 		aux.tp_mac = 0;
3564 		aux.tp_net = skb_network_offset(skb);
3565 		if (skb_vlan_tag_present(skb)) {
3566 			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3567 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3568 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3569 		} else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
3570 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3571 			struct net_device *dev;
3572 
3573 			rcu_read_lock();
3574 			dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex);
3575 			if (dev) {
3576 				aux.tp_vlan_tci = vlan_get_tci(skb, dev);
3577 				aux.tp_vlan_tpid = ntohs(skb->protocol);
3578 				aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3579 			} else {
3580 				aux.tp_vlan_tci = 0;
3581 				aux.tp_vlan_tpid = 0;
3582 			}
3583 			rcu_read_unlock();
3584 		} else {
3585 			aux.tp_vlan_tci = 0;
3586 			aux.tp_vlan_tpid = 0;
3587 		}
3588 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3589 	}
3590 
3591 	/*
3592 	 *	Free or return the buffer as appropriate. Again this
3593 	 *	hides all the races and re-entrancy issues from us.
3594 	 */
3595 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3596 
3597 out_free:
3598 	skb_free_datagram(sk, skb);
3599 out:
3600 	return err;
3601 }
3602 
3603 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3604 			       int peer)
3605 {
3606 	struct net_device *dev;
3607 	struct sock *sk	= sock->sk;
3608 
3609 	if (peer)
3610 		return -EOPNOTSUPP;
3611 
3612 	uaddr->sa_family = AF_PACKET;
3613 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3614 	rcu_read_lock();
3615 	dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3616 	if (dev)
3617 		strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3618 	rcu_read_unlock();
3619 
3620 	return sizeof(*uaddr);
3621 }
3622 
3623 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3624 			  int peer)
3625 {
3626 	struct net_device *dev;
3627 	struct sock *sk = sock->sk;
3628 	struct packet_sock *po = pkt_sk(sk);
3629 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3630 	int ifindex;
3631 
3632 	if (peer)
3633 		return -EOPNOTSUPP;
3634 
3635 	ifindex = READ_ONCE(po->ifindex);
3636 	sll->sll_family = AF_PACKET;
3637 	sll->sll_ifindex = ifindex;
3638 	sll->sll_protocol = READ_ONCE(po->num);
3639 	sll->sll_pkttype = 0;
3640 	rcu_read_lock();
3641 	dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3642 	if (dev) {
3643 		sll->sll_hatype = dev->type;
3644 		sll->sll_halen = dev->addr_len;
3645 
3646 		/* Let __fortify_memcpy_chk() know the actual buffer size. */
3647 		memcpy(((struct sockaddr_storage *)sll)->__data +
3648 		       offsetof(struct sockaddr_ll, sll_addr) -
3649 		       offsetofend(struct sockaddr_ll, sll_family),
3650 		       dev->dev_addr, dev->addr_len);
3651 	} else {
3652 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3653 		sll->sll_halen = 0;
3654 	}
3655 	rcu_read_unlock();
3656 
3657 	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3658 }
3659 
3660 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3661 			 int what)
3662 {
3663 	switch (i->type) {
3664 	case PACKET_MR_MULTICAST:
3665 		if (i->alen != dev->addr_len)
3666 			return -EINVAL;
3667 		if (what > 0)
3668 			return dev_mc_add(dev, i->addr);
3669 		else
3670 			return dev_mc_del(dev, i->addr);
3671 		break;
3672 	case PACKET_MR_PROMISC:
3673 		return dev_set_promiscuity(dev, what);
3674 	case PACKET_MR_ALLMULTI:
3675 		return dev_set_allmulti(dev, what);
3676 	case PACKET_MR_UNICAST:
3677 		if (i->alen != dev->addr_len)
3678 			return -EINVAL;
3679 		if (what > 0)
3680 			return dev_uc_add(dev, i->addr);
3681 		else
3682 			return dev_uc_del(dev, i->addr);
3683 		break;
3684 	default:
3685 		break;
3686 	}
3687 	return 0;
3688 }
3689 
3690 static void packet_dev_mclist_delete(struct net_device *dev,
3691 				     struct packet_mclist **mlp,
3692 				     struct list_head *list)
3693 {
3694 	struct packet_mclist *ml;
3695 
3696 	while ((ml = *mlp) != NULL) {
3697 		if (ml->ifindex == dev->ifindex) {
3698 			list_add(&ml->remove_list, list);
3699 			*mlp = ml->next;
3700 		} else
3701 			mlp = &ml->next;
3702 	}
3703 }
3704 
3705 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3706 {
3707 	struct packet_sock *po = pkt_sk(sk);
3708 	struct packet_mclist *ml, *i;
3709 	struct net_device *dev;
3710 	int err;
3711 
3712 	rtnl_lock();
3713 
3714 	err = -ENODEV;
3715 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3716 	if (!dev)
3717 		goto done;
3718 
3719 	err = -EINVAL;
3720 	if (mreq->mr_alen > dev->addr_len)
3721 		goto done;
3722 
3723 	err = -ENOBUFS;
3724 	i = kmalloc_obj(*i);
3725 	if (i == NULL)
3726 		goto done;
3727 
3728 	err = 0;
3729 	for (ml = po->mclist; ml; ml = ml->next) {
3730 		if (ml->ifindex == mreq->mr_ifindex &&
3731 		    ml->type == mreq->mr_type &&
3732 		    ml->alen == mreq->mr_alen &&
3733 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3734 			ml->count++;
3735 			/* Free the new element ... */
3736 			kfree(i);
3737 			goto done;
3738 		}
3739 	}
3740 
3741 	i->type = mreq->mr_type;
3742 	i->ifindex = mreq->mr_ifindex;
3743 	i->alen = mreq->mr_alen;
3744 	memcpy(i->addr, mreq->mr_address, i->alen);
3745 	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3746 	i->count = 1;
3747 	INIT_LIST_HEAD(&i->remove_list);
3748 	i->next = po->mclist;
3749 	po->mclist = i;
3750 	err = packet_dev_mc(dev, i, 1);
3751 	if (err) {
3752 		po->mclist = i->next;
3753 		kfree(i);
3754 	}
3755 
3756 done:
3757 	rtnl_unlock();
3758 	return err;
3759 }
3760 
3761 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3762 {
3763 	struct packet_mclist *ml, **mlp;
3764 
3765 	rtnl_lock();
3766 
3767 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3768 		if (ml->ifindex == mreq->mr_ifindex &&
3769 		    ml->type == mreq->mr_type &&
3770 		    ml->alen == mreq->mr_alen &&
3771 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3772 			if (--ml->count == 0) {
3773 				struct net_device *dev;
3774 				*mlp = ml->next;
3775 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3776 				if (dev)
3777 					packet_dev_mc(dev, ml, -1);
3778 				kfree(ml);
3779 			}
3780 			break;
3781 		}
3782 	}
3783 	rtnl_unlock();
3784 	return 0;
3785 }
3786 
3787 static void packet_flush_mclist(struct sock *sk)
3788 {
3789 	struct packet_sock *po = pkt_sk(sk);
3790 	struct packet_mclist *ml;
3791 
3792 	if (!po->mclist)
3793 		return;
3794 
3795 	rtnl_lock();
3796 	while ((ml = po->mclist) != NULL) {
3797 		struct net_device *dev;
3798 
3799 		po->mclist = ml->next;
3800 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3801 		if (dev != NULL)
3802 			packet_dev_mc(dev, ml, -1);
3803 		kfree(ml);
3804 	}
3805 	rtnl_unlock();
3806 }
3807 
3808 static int
3809 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3810 		  unsigned int optlen)
3811 {
3812 	struct sock *sk = sock->sk;
3813 	struct packet_sock *po = pkt_sk(sk);
3814 	int ret;
3815 
3816 	if (level != SOL_PACKET)
3817 		return -ENOPROTOOPT;
3818 
3819 	switch (optname) {
3820 	case PACKET_ADD_MEMBERSHIP:
3821 	case PACKET_DROP_MEMBERSHIP:
3822 	{
3823 		struct packet_mreq_max mreq;
3824 		int len = optlen;
3825 		memset(&mreq, 0, sizeof(mreq));
3826 		if (len < sizeof(struct packet_mreq))
3827 			return -EINVAL;
3828 		if (len > sizeof(mreq))
3829 			len = sizeof(mreq);
3830 		if (copy_from_sockptr(&mreq, optval, len))
3831 			return -EFAULT;
3832 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3833 			return -EINVAL;
3834 		if (optname == PACKET_ADD_MEMBERSHIP)
3835 			ret = packet_mc_add(sk, &mreq);
3836 		else
3837 			ret = packet_mc_drop(sk, &mreq);
3838 		return ret;
3839 	}
3840 
3841 	case PACKET_RX_RING:
3842 	case PACKET_TX_RING:
3843 	{
3844 		union tpacket_req_u req_u;
3845 
3846 		ret = -EINVAL;
3847 		lock_sock(sk);
3848 		switch (po->tp_version) {
3849 		case TPACKET_V1:
3850 		case TPACKET_V2:
3851 			if (optlen < sizeof(req_u.req))
3852 				break;
3853 			ret = copy_from_sockptr(&req_u.req, optval,
3854 						sizeof(req_u.req)) ?
3855 						-EINVAL : 0;
3856 			break;
3857 		case TPACKET_V3:
3858 		default:
3859 			if (optlen < sizeof(req_u.req3))
3860 				break;
3861 			ret = copy_from_sockptr(&req_u.req3, optval,
3862 						sizeof(req_u.req3)) ?
3863 						-EINVAL : 0;
3864 			break;
3865 		}
3866 		if (!ret)
3867 			ret = packet_set_ring(sk, &req_u, 0,
3868 					      optname == PACKET_TX_RING);
3869 		release_sock(sk);
3870 		return ret;
3871 	}
3872 	case PACKET_COPY_THRESH:
3873 	{
3874 		int val;
3875 
3876 		if (optlen != sizeof(val))
3877 			return -EINVAL;
3878 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3879 			return -EFAULT;
3880 
3881 		WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3882 		return 0;
3883 	}
3884 	case PACKET_VERSION:
3885 	{
3886 		int val;
3887 
3888 		if (optlen != sizeof(val))
3889 			return -EINVAL;
3890 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3891 			return -EFAULT;
3892 		switch (val) {
3893 		case TPACKET_V1:
3894 		case TPACKET_V2:
3895 		case TPACKET_V3:
3896 			break;
3897 		default:
3898 			return -EINVAL;
3899 		}
3900 		lock_sock(sk);
3901 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3902 			ret = -EBUSY;
3903 		} else {
3904 			po->tp_version = val;
3905 			ret = 0;
3906 		}
3907 		release_sock(sk);
3908 		return ret;
3909 	}
3910 	case PACKET_RESERVE:
3911 	{
3912 		unsigned int val;
3913 
3914 		if (optlen != sizeof(val))
3915 			return -EINVAL;
3916 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3917 			return -EFAULT;
3918 		if (val > INT_MAX)
3919 			return -EINVAL;
3920 		lock_sock(sk);
3921 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3922 			ret = -EBUSY;
3923 		} else {
3924 			po->tp_reserve = val;
3925 			ret = 0;
3926 		}
3927 		release_sock(sk);
3928 		return ret;
3929 	}
3930 	case PACKET_LOSS:
3931 	{
3932 		unsigned int val;
3933 
3934 		if (optlen != sizeof(val))
3935 			return -EINVAL;
3936 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3937 			return -EFAULT;
3938 
3939 		lock_sock(sk);
3940 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3941 			ret = -EBUSY;
3942 		} else {
3943 			packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3944 			ret = 0;
3945 		}
3946 		release_sock(sk);
3947 		return ret;
3948 	}
3949 	case PACKET_AUXDATA:
3950 	{
3951 		int val;
3952 
3953 		if (optlen < sizeof(val))
3954 			return -EINVAL;
3955 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3956 			return -EFAULT;
3957 
3958 		packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3959 		return 0;
3960 	}
3961 	case PACKET_ORIGDEV:
3962 	{
3963 		int val;
3964 
3965 		if (optlen < sizeof(val))
3966 			return -EINVAL;
3967 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3968 			return -EFAULT;
3969 
3970 		packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3971 		return 0;
3972 	}
3973 	case PACKET_VNET_HDR:
3974 	case PACKET_VNET_HDR_SZ:
3975 	{
3976 		int val, hdr_len;
3977 
3978 		if (sock->type != SOCK_RAW)
3979 			return -EINVAL;
3980 		if (optlen < sizeof(val))
3981 			return -EINVAL;
3982 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3983 			return -EFAULT;
3984 
3985 		if (optname == PACKET_VNET_HDR_SZ) {
3986 			if (val && val != sizeof(struct virtio_net_hdr) &&
3987 			    val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
3988 				return -EINVAL;
3989 			hdr_len = val;
3990 		} else {
3991 			hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
3992 		}
3993 		lock_sock(sk);
3994 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3995 			ret = -EBUSY;
3996 		} else {
3997 			WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
3998 			ret = 0;
3999 		}
4000 		release_sock(sk);
4001 		return ret;
4002 	}
4003 	case PACKET_TIMESTAMP:
4004 	{
4005 		int val;
4006 
4007 		if (optlen != sizeof(val))
4008 			return -EINVAL;
4009 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4010 			return -EFAULT;
4011 
4012 		WRITE_ONCE(po->tp_tstamp, val);
4013 		return 0;
4014 	}
4015 	case PACKET_FANOUT:
4016 	{
4017 		struct fanout_args args = { 0 };
4018 
4019 		if (optlen != sizeof(int) && optlen != sizeof(args))
4020 			return -EINVAL;
4021 		if (copy_from_sockptr(&args, optval, optlen))
4022 			return -EFAULT;
4023 
4024 		return fanout_add(sk, &args);
4025 	}
4026 	case PACKET_FANOUT_DATA:
4027 	{
4028 		/* Paired with the WRITE_ONCE() in fanout_add() */
4029 		if (!READ_ONCE(po->fanout))
4030 			return -EINVAL;
4031 
4032 		return fanout_set_data(po, optval, optlen);
4033 	}
4034 	case PACKET_IGNORE_OUTGOING:
4035 	{
4036 		int val;
4037 
4038 		if (optlen != sizeof(val))
4039 			return -EINVAL;
4040 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4041 			return -EFAULT;
4042 		if (val < 0 || val > 1)
4043 			return -EINVAL;
4044 
4045 		WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4046 		return 0;
4047 	}
4048 	case PACKET_TX_HAS_OFF:
4049 	{
4050 		unsigned int val;
4051 
4052 		if (optlen != sizeof(val))
4053 			return -EINVAL;
4054 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4055 			return -EFAULT;
4056 
4057 		lock_sock(sk);
4058 		if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4059 			packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4060 
4061 		release_sock(sk);
4062 		return 0;
4063 	}
4064 	case PACKET_QDISC_BYPASS:
4065 	{
4066 		int val;
4067 
4068 		if (optlen != sizeof(val))
4069 			return -EINVAL;
4070 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4071 			return -EFAULT;
4072 
4073 		packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4074 		return 0;
4075 	}
4076 	default:
4077 		return -ENOPROTOOPT;
4078 	}
4079 }
4080 
4081 static int packet_getsockopt(struct socket *sock, int level, int optname,
4082 			     sockopt_t *opt)
4083 {
4084 	int len;
4085 	int val, lv = sizeof(val);
4086 	struct sock *sk = sock->sk;
4087 	struct packet_sock *po = pkt_sk(sk);
4088 	void *data = &val;
4089 	union tpacket_stats_u st;
4090 	struct tpacket_rollover_stats rstats;
4091 	int drops;
4092 
4093 	if (level != SOL_PACKET)
4094 		return -ENOPROTOOPT;
4095 
4096 	len = opt->optlen;
4097 
4098 	if (len < 0)
4099 		return -EINVAL;
4100 
4101 	switch (optname) {
4102 	case PACKET_STATISTICS:
4103 		spin_lock_bh(&sk->sk_receive_queue.lock);
4104 		memcpy(&st, &po->stats, sizeof(st));
4105 		memset(&po->stats, 0, sizeof(po->stats));
4106 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4107 		drops = atomic_xchg(&po->tp_drops, 0);
4108 
4109 		if (po->tp_version == TPACKET_V3) {
4110 			lv = sizeof(struct tpacket_stats_v3);
4111 			st.stats3.tp_drops = drops;
4112 			st.stats3.tp_packets += drops;
4113 			data = &st.stats3;
4114 		} else {
4115 			lv = sizeof(struct tpacket_stats);
4116 			st.stats1.tp_drops = drops;
4117 			st.stats1.tp_packets += drops;
4118 			data = &st.stats1;
4119 		}
4120 
4121 		break;
4122 	case PACKET_AUXDATA:
4123 		val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4124 		break;
4125 	case PACKET_ORIGDEV:
4126 		val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4127 		break;
4128 	case PACKET_VNET_HDR:
4129 		val = !!READ_ONCE(po->vnet_hdr_sz);
4130 		break;
4131 	case PACKET_VNET_HDR_SZ:
4132 		val = READ_ONCE(po->vnet_hdr_sz);
4133 		break;
4134 	case PACKET_COPY_THRESH:
4135 		val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4136 		break;
4137 	case PACKET_VERSION:
4138 		val = po->tp_version;
4139 		break;
4140 	case PACKET_HDRLEN:
4141 		if (len > sizeof(int))
4142 			len = sizeof(int);
4143 		if (len < sizeof(int))
4144 			return -EINVAL;
4145 		if (copy_from_iter(&val, len, &opt->iter_in) != len)
4146 			return -EFAULT;
4147 		switch (val) {
4148 		case TPACKET_V1:
4149 			val = sizeof(struct tpacket_hdr);
4150 			break;
4151 		case TPACKET_V2:
4152 			val = sizeof(struct tpacket2_hdr);
4153 			break;
4154 		case TPACKET_V3:
4155 			val = sizeof(struct tpacket3_hdr);
4156 			break;
4157 		default:
4158 			return -EINVAL;
4159 		}
4160 		break;
4161 	case PACKET_RESERVE:
4162 		val = po->tp_reserve;
4163 		break;
4164 	case PACKET_LOSS:
4165 		val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4166 		break;
4167 	case PACKET_TIMESTAMP:
4168 		val = READ_ONCE(po->tp_tstamp);
4169 		break;
4170 	case PACKET_FANOUT:
4171 		val = (po->fanout ?
4172 		       ((u32)po->fanout->id |
4173 			((u32)po->fanout->type << 16) |
4174 			((u32)po->fanout->flags << 24)) :
4175 		       0);
4176 		break;
4177 	case PACKET_IGNORE_OUTGOING:
4178 		val = READ_ONCE(po->prot_hook.ignore_outgoing);
4179 		break;
4180 	case PACKET_ROLLOVER_STATS:
4181 		if (!po->rollover)
4182 			return -EINVAL;
4183 		rstats.tp_all = atomic_long_read(&po->rollover->num);
4184 		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4185 		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4186 		data = &rstats;
4187 		lv = sizeof(rstats);
4188 		break;
4189 	case PACKET_TX_HAS_OFF:
4190 		val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4191 		break;
4192 	case PACKET_QDISC_BYPASS:
4193 		val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4194 		break;
4195 	default:
4196 		return -ENOPROTOOPT;
4197 	}
4198 
4199 	if (len > lv)
4200 		len = lv;
4201 	opt->optlen = len;
4202 	if (copy_to_iter(data, len, &opt->iter_out) != len)
4203 		return -EFAULT;
4204 	return 0;
4205 }
4206 
4207 static int packet_notifier(struct notifier_block *this,
4208 			   unsigned long msg, void *ptr)
4209 {
4210 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4211 	struct net *net = dev_net(dev);
4212 	struct packet_mclist *ml, *tmp;
4213 	LIST_HEAD(mclist);
4214 	struct sock *sk;
4215 
4216 	rcu_read_lock();
4217 	sk_for_each_rcu(sk, &net->packet.sklist) {
4218 		struct packet_sock *po = pkt_sk(sk);
4219 
4220 		switch (msg) {
4221 		case NETDEV_UNREGISTER:
4222 			if (po->mclist)
4223 				packet_dev_mclist_delete(dev, &po->mclist,
4224 							 &mclist);
4225 			fallthrough;
4226 
4227 		case NETDEV_DOWN:
4228 			if (dev->ifindex == po->ifindex) {
4229 				spin_lock(&po->bind_lock);
4230 				if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4231 					__unregister_prot_hook(sk, false);
4232 					sk->sk_err = ENETDOWN;
4233 					if (!sock_flag(sk, SOCK_DEAD))
4234 						sk_error_report(sk);
4235 				}
4236 				if (msg == NETDEV_UNREGISTER) {
4237 					packet_cached_dev_reset(po);
4238 					WRITE_ONCE(po->ifindex, -1);
4239 					netdev_put(po->prot_hook.dev,
4240 						   &po->prot_hook.dev_tracker);
4241 					po->prot_hook.dev = NULL;
4242 				}
4243 				spin_unlock(&po->bind_lock);
4244 			}
4245 			break;
4246 		case NETDEV_UP:
4247 			if (dev->ifindex == po->ifindex) {
4248 				spin_lock(&po->bind_lock);
4249 				if (po->num)
4250 					register_prot_hook(sk);
4251 				spin_unlock(&po->bind_lock);
4252 			}
4253 			break;
4254 		}
4255 	}
4256 	rcu_read_unlock();
4257 
4258 	/* packet_dev_mc might grab instance locks so can't run under rcu */
4259 	list_for_each_entry_safe(ml, tmp, &mclist, remove_list) {
4260 		packet_dev_mc(dev, ml, -1);
4261 		kfree(ml);
4262 	}
4263 
4264 	return NOTIFY_DONE;
4265 }
4266 
4267 
4268 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4269 			unsigned long arg)
4270 {
4271 	struct sock *sk = sock->sk;
4272 
4273 	switch (cmd) {
4274 	case SIOCOUTQ:
4275 	{
4276 		int amount = sk_wmem_alloc_get(sk);
4277 
4278 		return put_user(amount, (int __user *)arg);
4279 	}
4280 	case SIOCINQ:
4281 	{
4282 		struct sk_buff *skb;
4283 		int amount = 0;
4284 
4285 		spin_lock_bh(&sk->sk_receive_queue.lock);
4286 		skb = skb_peek(&sk->sk_receive_queue);
4287 		if (skb)
4288 			amount = skb->len;
4289 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4290 		return put_user(amount, (int __user *)arg);
4291 	}
4292 #ifdef CONFIG_INET
4293 	case SIOCADDRT:
4294 	case SIOCDELRT:
4295 	case SIOCDARP:
4296 	case SIOCGARP:
4297 	case SIOCSARP:
4298 	case SIOCGIFADDR:
4299 	case SIOCSIFADDR:
4300 	case SIOCGIFBRDADDR:
4301 	case SIOCSIFBRDADDR:
4302 	case SIOCGIFNETMASK:
4303 	case SIOCSIFNETMASK:
4304 	case SIOCGIFDSTADDR:
4305 	case SIOCSIFDSTADDR:
4306 	case SIOCSIFFLAGS:
4307 		return inet_dgram_ops.ioctl(sock, cmd, arg);
4308 #endif
4309 
4310 	default:
4311 		return -ENOIOCTLCMD;
4312 	}
4313 	return 0;
4314 }
4315 
4316 static __poll_t packet_poll(struct file *file, struct socket *sock,
4317 				poll_table *wait)
4318 {
4319 	struct sock *sk = sock->sk;
4320 	struct packet_sock *po = pkt_sk(sk);
4321 	__poll_t mask = datagram_poll(file, sock, wait);
4322 
4323 	spin_lock_bh(&sk->sk_receive_queue.lock);
4324 	if (po->rx_ring.pg_vec) {
4325 		if (!packet_previous_rx_frame(po, &po->rx_ring,
4326 			TP_STATUS_KERNEL))
4327 			mask |= EPOLLIN | EPOLLRDNORM;
4328 	}
4329 	__packet_rcv_try_clear_pressure(po);
4330 	spin_unlock_bh(&sk->sk_receive_queue.lock);
4331 	spin_lock_bh(&sk->sk_write_queue.lock);
4332 	if (po->tx_ring.pg_vec) {
4333 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4334 			mask |= EPOLLOUT | EPOLLWRNORM;
4335 	}
4336 	spin_unlock_bh(&sk->sk_write_queue.lock);
4337 	return mask;
4338 }
4339 
4340 
4341 /* Dirty? Well, I still did not learn better way to account
4342  * for user mmaps.
4343  */
4344 
4345 static void packet_mm_open(struct vm_area_struct *vma)
4346 {
4347 	struct file *file = vma->vm_file;
4348 	struct socket *sock = file->private_data;
4349 	struct sock *sk = sock->sk;
4350 
4351 	if (sk)
4352 		atomic_long_inc(&pkt_sk(sk)->mapped);
4353 }
4354 
4355 static void packet_mm_close(struct vm_area_struct *vma)
4356 {
4357 	struct file *file = vma->vm_file;
4358 	struct socket *sock = file->private_data;
4359 	struct sock *sk = sock->sk;
4360 
4361 	if (sk)
4362 		atomic_long_dec(&pkt_sk(sk)->mapped);
4363 }
4364 
4365 static const struct vm_operations_struct packet_mmap_ops = {
4366 	.open	=	packet_mm_open,
4367 	.close	=	packet_mm_close,
4368 };
4369 
4370 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4371 			unsigned int len)
4372 {
4373 	int i;
4374 
4375 	for (i = 0; i < len; i++) {
4376 		if (likely(pg_vec[i].buffer)) {
4377 			if (is_vmalloc_addr(pg_vec[i].buffer))
4378 				vfree(pg_vec[i].buffer);
4379 			else
4380 				free_pages((unsigned long)pg_vec[i].buffer,
4381 					   order);
4382 			pg_vec[i].buffer = NULL;
4383 		}
4384 	}
4385 	kfree(pg_vec);
4386 }
4387 
4388 static char *alloc_one_pg_vec_page(unsigned long order)
4389 {
4390 	char *buffer;
4391 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4392 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4393 
4394 	buffer = (char *) __get_free_pages(gfp_flags, order);
4395 	if (buffer)
4396 		return buffer;
4397 
4398 	/* __get_free_pages failed, fall back to vmalloc */
4399 	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4400 	if (buffer)
4401 		return buffer;
4402 
4403 	/* vmalloc failed, lets dig into swap here */
4404 	gfp_flags &= ~__GFP_NORETRY;
4405 	buffer = (char *) __get_free_pages(gfp_flags, order);
4406 	if (buffer)
4407 		return buffer;
4408 
4409 	/* complete and utter failure */
4410 	return NULL;
4411 }
4412 
4413 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4414 {
4415 	unsigned int block_nr = req->tp_block_nr;
4416 	struct pgv *pg_vec;
4417 	int i;
4418 
4419 	pg_vec = kzalloc_objs(struct pgv, block_nr, GFP_KERNEL | __GFP_NOWARN);
4420 	if (unlikely(!pg_vec))
4421 		goto out;
4422 
4423 	for (i = 0; i < block_nr; i++) {
4424 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4425 		if (unlikely(!pg_vec[i].buffer))
4426 			goto out_free_pgvec;
4427 	}
4428 
4429 out:
4430 	return pg_vec;
4431 
4432 out_free_pgvec:
4433 	free_pg_vec(pg_vec, order, block_nr);
4434 	pg_vec = NULL;
4435 	goto out;
4436 }
4437 
4438 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4439 		int closing, int tx_ring)
4440 {
4441 	struct pgv *pg_vec = NULL;
4442 	struct packet_sock *po = pkt_sk(sk);
4443 	unsigned long *rx_owner_map = NULL;
4444 	int was_running, order = 0;
4445 	struct packet_ring_buffer *rb;
4446 	struct sk_buff_head *rb_queue;
4447 	__be16 num;
4448 	int err;
4449 	/* Added to avoid minimal code churn */
4450 	struct tpacket_req *req = &req_u->req;
4451 
4452 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4453 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4454 
4455 	err = -EBUSY;
4456 	if (!closing) {
4457 		if (atomic_long_read(&po->mapped))
4458 			goto out;
4459 		if (packet_read_pending(rb))
4460 			goto out;
4461 	}
4462 
4463 	if (req->tp_block_nr) {
4464 		unsigned int min_frame_size;
4465 
4466 		/* Sanity tests and some calculations */
4467 		err = -EBUSY;
4468 		if (unlikely(rb->pg_vec))
4469 			goto out;
4470 
4471 		switch (po->tp_version) {
4472 		case TPACKET_V1:
4473 			po->tp_hdrlen = TPACKET_HDRLEN;
4474 			break;
4475 		case TPACKET_V2:
4476 			po->tp_hdrlen = TPACKET2_HDRLEN;
4477 			break;
4478 		case TPACKET_V3:
4479 			po->tp_hdrlen = TPACKET3_HDRLEN;
4480 			break;
4481 		}
4482 
4483 		err = -EINVAL;
4484 		if (unlikely((int)req->tp_block_size <= 0))
4485 			goto out;
4486 		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4487 			goto out;
4488 		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4489 		if (po->tp_version >= TPACKET_V3 &&
4490 		    req->tp_block_size <
4491 		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4492 			goto out;
4493 		if (unlikely(req->tp_frame_size < min_frame_size))
4494 			goto out;
4495 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4496 			goto out;
4497 
4498 		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4499 		if (unlikely(rb->frames_per_block == 0))
4500 			goto out;
4501 		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4502 			goto out;
4503 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4504 					req->tp_frame_nr))
4505 			goto out;
4506 
4507 		err = -ENOMEM;
4508 		order = get_order(req->tp_block_size);
4509 		pg_vec = alloc_pg_vec(req, order);
4510 		if (unlikely(!pg_vec))
4511 			goto out;
4512 		switch (po->tp_version) {
4513 		case TPACKET_V3:
4514 			/* Block transmit is not supported yet */
4515 			if (!tx_ring) {
4516 				init_prb_bdqc(po, rb, pg_vec, req_u);
4517 			} else {
4518 				struct tpacket_req3 *req3 = &req_u->req3;
4519 
4520 				if (req3->tp_retire_blk_tov ||
4521 				    req3->tp_sizeof_priv ||
4522 				    req3->tp_feature_req_word) {
4523 					err = -EINVAL;
4524 					goto out_free_pg_vec;
4525 				}
4526 			}
4527 			break;
4528 		default:
4529 			if (!tx_ring) {
4530 				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4531 					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4532 				if (!rx_owner_map)
4533 					goto out_free_pg_vec;
4534 			}
4535 			break;
4536 		}
4537 	}
4538 	/* Done */
4539 	else {
4540 		err = -EINVAL;
4541 		if (unlikely(req->tp_frame_nr))
4542 			goto out;
4543 	}
4544 
4545 
4546 	/* Detach socket from network */
4547 	spin_lock(&po->bind_lock);
4548 	was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4549 	num = po->num;
4550 	WRITE_ONCE(po->num, 0);
4551 	if (was_running)
4552 		__unregister_prot_hook(sk, false);
4553 
4554 	spin_unlock(&po->bind_lock);
4555 
4556 	synchronize_net();
4557 
4558 	err = -EBUSY;
4559 	mutex_lock(&po->pg_vec_lock);
4560 	if (closing || atomic_long_read(&po->mapped) == 0) {
4561 		err = 0;
4562 		spin_lock_bh(&rb_queue->lock);
4563 		swap(rb->pg_vec, pg_vec);
4564 		if (po->tp_version <= TPACKET_V2)
4565 			swap(rb->rx_owner_map, rx_owner_map);
4566 		rb->frame_max = (req->tp_frame_nr - 1);
4567 		rb->head = 0;
4568 		rb->frame_size = req->tp_frame_size;
4569 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4570 						tpacket_rcv : packet_rcv;
4571 		spin_unlock_bh(&rb_queue->lock);
4572 
4573 		swap(rb->pg_vec_order, order);
4574 		swap(rb->pg_vec_len, req->tp_block_nr);
4575 
4576 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4577 		skb_queue_purge(rb_queue);
4578 		if (atomic_long_read(&po->mapped))
4579 			pr_err("packet_mmap: vma is busy: %ld\n",
4580 			       atomic_long_read(&po->mapped));
4581 	}
4582 	mutex_unlock(&po->pg_vec_lock);
4583 
4584 	spin_lock(&po->bind_lock);
4585 	WRITE_ONCE(po->num, num);
4586 	/*
4587 	 * NETDEV_UNREGISTER may have invalidated the binding while bind_lock
4588 	 * was dropped above.  Do not re-add a fanout hook to a dead device.
4589 	 */
4590 	if (was_running && READ_ONCE(po->ifindex) != -1)
4591 		register_prot_hook(sk);
4592 
4593 	spin_unlock(&po->bind_lock);
4594 	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4595 		/* Because we don't support block-based V3 on tx-ring */
4596 		if (!tx_ring)
4597 			prb_shutdown_retire_blk_timer(po, rb_queue);
4598 	}
4599 
4600 out_free_pg_vec:
4601 	if (pg_vec) {
4602 		bitmap_free(rx_owner_map);
4603 		free_pg_vec(pg_vec, order, req->tp_block_nr);
4604 	}
4605 out:
4606 	return err;
4607 }
4608 
4609 static int packet_mmap(struct file *file, struct socket *sock,
4610 		struct vm_area_struct *vma)
4611 {
4612 	struct sock *sk = sock->sk;
4613 	struct packet_sock *po = pkt_sk(sk);
4614 	unsigned long size, expected_size;
4615 	struct packet_ring_buffer *rb;
4616 	unsigned long start;
4617 	int err = -EINVAL;
4618 	int i;
4619 
4620 	if (vma->vm_pgoff)
4621 		return -EINVAL;
4622 
4623 	mutex_lock(&po->pg_vec_lock);
4624 
4625 	expected_size = 0;
4626 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4627 		if (rb->pg_vec) {
4628 			expected_size += rb->pg_vec_len
4629 						* rb->pg_vec_pages
4630 						* PAGE_SIZE;
4631 		}
4632 	}
4633 
4634 	if (expected_size == 0)
4635 		goto out;
4636 
4637 	size = vma->vm_end - vma->vm_start;
4638 	if (size != expected_size)
4639 		goto out;
4640 
4641 	start = vma->vm_start;
4642 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4643 		if (rb->pg_vec == NULL)
4644 			continue;
4645 
4646 		for (i = 0; i < rb->pg_vec_len; i++) {
4647 			struct page *page;
4648 			void *kaddr = rb->pg_vec[i].buffer;
4649 			int pg_num;
4650 
4651 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4652 				page = pgv_to_page(kaddr);
4653 				err = vm_insert_page(vma, start, page);
4654 				if (unlikely(err))
4655 					goto out;
4656 				start += PAGE_SIZE;
4657 				kaddr += PAGE_SIZE;
4658 			}
4659 		}
4660 	}
4661 
4662 	atomic_long_inc(&po->mapped);
4663 	vma->vm_ops = &packet_mmap_ops;
4664 	err = 0;
4665 
4666 out:
4667 	mutex_unlock(&po->pg_vec_lock);
4668 	return err;
4669 }
4670 
4671 static const struct proto_ops packet_ops_spkt = {
4672 	.family =	PF_PACKET,
4673 	.owner =	THIS_MODULE,
4674 	.release =	packet_release,
4675 	.bind =		packet_bind_spkt,
4676 	.connect =	sock_no_connect,
4677 	.socketpair =	sock_no_socketpair,
4678 	.accept =	sock_no_accept,
4679 	.getname =	packet_getname_spkt,
4680 	.poll =		datagram_poll,
4681 	.ioctl =	packet_ioctl,
4682 	.gettstamp =	sock_gettstamp,
4683 	.listen =	sock_no_listen,
4684 	.shutdown =	sock_no_shutdown,
4685 	.sendmsg =	packet_sendmsg_spkt,
4686 	.recvmsg =	packet_recvmsg,
4687 	.mmap =		sock_no_mmap,
4688 };
4689 
4690 static const struct proto_ops packet_ops = {
4691 	.family =	PF_PACKET,
4692 	.owner =	THIS_MODULE,
4693 	.release =	packet_release,
4694 	.bind =		packet_bind,
4695 	.connect =	sock_no_connect,
4696 	.socketpair =	sock_no_socketpair,
4697 	.accept =	sock_no_accept,
4698 	.getname =	packet_getname,
4699 	.poll =		packet_poll,
4700 	.ioctl =	packet_ioctl,
4701 	.gettstamp =	sock_gettstamp,
4702 	.listen =	sock_no_listen,
4703 	.shutdown =	sock_no_shutdown,
4704 	.setsockopt =	packet_setsockopt,
4705 	.getsockopt_iter =	packet_getsockopt,
4706 	.sendmsg =	packet_sendmsg,
4707 	.recvmsg =	packet_recvmsg,
4708 	.mmap =		packet_mmap,
4709 };
4710 
4711 static const struct net_proto_family packet_family_ops = {
4712 	.family =	PF_PACKET,
4713 	.create =	packet_create,
4714 	.owner	=	THIS_MODULE,
4715 };
4716 
4717 static struct notifier_block packet_netdev_notifier = {
4718 	.notifier_call =	packet_notifier,
4719 };
4720 
4721 #ifdef CONFIG_PROC_FS
4722 
4723 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4724 	__acquires(RCU)
4725 {
4726 	struct net *net = seq_file_net(seq);
4727 
4728 	rcu_read_lock();
4729 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4730 }
4731 
4732 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4733 {
4734 	struct net *net = seq_file_net(seq);
4735 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4736 }
4737 
4738 static void packet_seq_stop(struct seq_file *seq, void *v)
4739 	__releases(RCU)
4740 {
4741 	rcu_read_unlock();
4742 }
4743 
4744 static int packet_seq_show(struct seq_file *seq, void *v)
4745 {
4746 	if (v == SEQ_START_TOKEN)
4747 		seq_printf(seq,
4748 			   "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4749 			   IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4750 	else {
4751 		struct sock *s = sk_entry(v);
4752 		const struct packet_sock *po = pkt_sk(s);
4753 
4754 		seq_printf(seq,
4755 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6llu\n",
4756 			   s,
4757 			   refcount_read(&s->sk_refcnt),
4758 			   s->sk_type,
4759 			   ntohs(READ_ONCE(po->num)),
4760 			   READ_ONCE(po->ifindex),
4761 			   packet_sock_flag(po, PACKET_SOCK_RUNNING),
4762 			   atomic_read(&s->sk_rmem_alloc),
4763 			   from_kuid_munged(seq_user_ns(seq), sk_uid(s)),
4764 			   sock_i_ino(s));
4765 	}
4766 
4767 	return 0;
4768 }
4769 
4770 static const struct seq_operations packet_seq_ops = {
4771 	.start	= packet_seq_start,
4772 	.next	= packet_seq_next,
4773 	.stop	= packet_seq_stop,
4774 	.show	= packet_seq_show,
4775 };
4776 #endif
4777 
4778 static int __net_init packet_net_init(struct net *net)
4779 {
4780 	mutex_init(&net->packet.sklist_lock);
4781 	INIT_HLIST_HEAD(&net->packet.sklist);
4782 
4783 #ifdef CONFIG_PROC_FS
4784 	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4785 			sizeof(struct seq_net_private)))
4786 		return -ENOMEM;
4787 #endif /* CONFIG_PROC_FS */
4788 
4789 	return 0;
4790 }
4791 
4792 static void __net_exit packet_net_exit(struct net *net)
4793 {
4794 	remove_proc_entry("packet", net->proc_net);
4795 	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4796 }
4797 
4798 static struct pernet_operations packet_net_ops = {
4799 	.init = packet_net_init,
4800 	.exit = packet_net_exit,
4801 };
4802 
4803 
4804 static void __exit packet_exit(void)
4805 {
4806 	sock_unregister(PF_PACKET);
4807 	proto_unregister(&packet_proto);
4808 	unregister_netdevice_notifier(&packet_netdev_notifier);
4809 	unregister_pernet_subsys(&packet_net_ops);
4810 }
4811 
4812 static int __init packet_init(void)
4813 {
4814 	int rc;
4815 
4816 	rc = register_pernet_subsys(&packet_net_ops);
4817 	if (rc)
4818 		goto out;
4819 	rc = register_netdevice_notifier(&packet_netdev_notifier);
4820 	if (rc)
4821 		goto out_pernet;
4822 	rc = proto_register(&packet_proto, 0);
4823 	if (rc)
4824 		goto out_notifier;
4825 	rc = sock_register(&packet_family_ops);
4826 	if (rc)
4827 		goto out_proto;
4828 
4829 	return 0;
4830 
4831 out_proto:
4832 	proto_unregister(&packet_proto);
4833 out_notifier:
4834 	unregister_netdevice_notifier(&packet_netdev_notifier);
4835 out_pernet:
4836 	unregister_pernet_subsys(&packet_net_ops);
4837 out:
4838 	return rc;
4839 }
4840 
4841 module_init(packet_init);
4842 module_exit(packet_exit);
4843 MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4844 MODULE_LICENSE("GPL");
4845 MODULE_ALIAS_NETPROTO(PF_PACKET);
4846