xref: /linux/net/packet/af_packet.c (revision 1cb4298810e27e037d3ca07286ecbb97e89ba58d)
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_sock_destruct(struct sock *sk)
1326 {
1327 	skb_queue_purge(&sk->sk_error_queue);
1328 
1329 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1330 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1331 
1332 	if (!sock_flag(sk, SOCK_DEAD)) {
1333 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1334 		return;
1335 	}
1336 }
1337 
1338 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1339 {
1340 	u32 *history = po->rollover->history;
1341 	u32 victim, rxhash;
1342 	int i, count = 0;
1343 
1344 	rxhash = skb_get_hash(skb);
1345 	for (i = 0; i < ROLLOVER_HLEN; i++)
1346 		if (READ_ONCE(history[i]) == rxhash)
1347 			count++;
1348 
1349 	victim = get_random_u32_below(ROLLOVER_HLEN);
1350 
1351 	/* Avoid dirtying the cache line if possible */
1352 	if (READ_ONCE(history[victim]) != rxhash)
1353 		WRITE_ONCE(history[victim], rxhash);
1354 
1355 	return count > (ROLLOVER_HLEN >> 1);
1356 }
1357 
1358 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1359 				      struct sk_buff *skb,
1360 				      unsigned int num)
1361 {
1362 	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1363 }
1364 
1365 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1366 				    struct sk_buff *skb,
1367 				    unsigned int num)
1368 {
1369 	unsigned int val = atomic_inc_return(&f->rr_cur);
1370 
1371 	return val % num;
1372 }
1373 
1374 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1375 				     struct sk_buff *skb,
1376 				     unsigned int num)
1377 {
1378 	return smp_processor_id() % num;
1379 }
1380 
1381 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1382 				     struct sk_buff *skb,
1383 				     unsigned int num)
1384 {
1385 	return get_random_u32_below(num);
1386 }
1387 
1388 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1389 					  struct sk_buff *skb,
1390 					  unsigned int idx, bool try_self,
1391 					  unsigned int num)
1392 {
1393 	struct packet_sock *po, *po_next, *po_skip = NULL;
1394 	unsigned int i, j, room = ROOM_NONE;
1395 
1396 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1397 
1398 	if (try_self) {
1399 		room = packet_rcv_has_room(po, skb);
1400 		if (room == ROOM_NORMAL ||
1401 		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1402 			return idx;
1403 		po_skip = po;
1404 	}
1405 
1406 	i = j = min_t(int, po->rollover->sock, num - 1);
1407 	do {
1408 		po_next = pkt_sk(rcu_dereference(f->arr[i]));
1409 		if (po_next != po_skip &&
1410 		    !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1411 		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1412 			if (i != j)
1413 				po->rollover->sock = i;
1414 			atomic_long_inc(&po->rollover->num);
1415 			if (room == ROOM_LOW)
1416 				atomic_long_inc(&po->rollover->num_huge);
1417 			return i;
1418 		}
1419 
1420 		if (++i == num)
1421 			i = 0;
1422 	} while (i != j);
1423 
1424 	atomic_long_inc(&po->rollover->num_failed);
1425 	return idx;
1426 }
1427 
1428 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1429 				    struct sk_buff *skb,
1430 				    unsigned int num)
1431 {
1432 	return skb_get_queue_mapping(skb) % num;
1433 }
1434 
1435 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1436 				     struct sk_buff *skb,
1437 				     unsigned int num)
1438 {
1439 	struct bpf_prog *prog;
1440 	unsigned int ret = 0;
1441 
1442 	rcu_read_lock();
1443 	prog = rcu_dereference(f->bpf_prog);
1444 	if (prog)
1445 		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1446 	rcu_read_unlock();
1447 
1448 	return ret;
1449 }
1450 
1451 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1452 {
1453 	return f->flags & (flag >> 8);
1454 }
1455 
1456 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1457 			     struct packet_type *pt, struct net_device *orig_dev)
1458 {
1459 	struct packet_fanout *f = pt->af_packet_priv;
1460 	unsigned int num = READ_ONCE(f->num_members);
1461 	struct net *net = read_pnet(&f->net);
1462 	struct packet_sock *po;
1463 	unsigned int idx;
1464 
1465 	if (!net_eq(dev_net(dev), net) || !num) {
1466 		kfree_skb(skb);
1467 		return 0;
1468 	}
1469 
1470 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1471 		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1472 		if (!skb)
1473 			return 0;
1474 	}
1475 	switch (f->type) {
1476 	case PACKET_FANOUT_HASH:
1477 	default:
1478 		idx = fanout_demux_hash(f, skb, num);
1479 		break;
1480 	case PACKET_FANOUT_LB:
1481 		idx = fanout_demux_lb(f, skb, num);
1482 		break;
1483 	case PACKET_FANOUT_CPU:
1484 		idx = fanout_demux_cpu(f, skb, num);
1485 		break;
1486 	case PACKET_FANOUT_RND:
1487 		idx = fanout_demux_rnd(f, skb, num);
1488 		break;
1489 	case PACKET_FANOUT_QM:
1490 		idx = fanout_demux_qm(f, skb, num);
1491 		break;
1492 	case PACKET_FANOUT_ROLLOVER:
1493 		idx = fanout_demux_rollover(f, skb, 0, false, num);
1494 		break;
1495 	case PACKET_FANOUT_CBPF:
1496 	case PACKET_FANOUT_EBPF:
1497 		idx = fanout_demux_bpf(f, skb, num);
1498 		break;
1499 	}
1500 
1501 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1502 		idx = fanout_demux_rollover(f, skb, idx, true, num);
1503 
1504 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1505 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1506 }
1507 
1508 DEFINE_MUTEX(fanout_mutex);
1509 EXPORT_SYMBOL_GPL(fanout_mutex);
1510 static LIST_HEAD(fanout_list);
1511 static u16 fanout_next_id;
1512 
1513 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1514 {
1515 	struct packet_fanout *f = po->fanout;
1516 
1517 	spin_lock(&f->lock);
1518 	rcu_assign_pointer(f->arr[f->num_members], sk);
1519 	smp_wmb();
1520 	f->num_members++;
1521 	if (f->num_members == 1)
1522 		dev_add_pack(&f->prot_hook);
1523 	spin_unlock(&f->lock);
1524 }
1525 
1526 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1527 {
1528 	struct packet_fanout *f = po->fanout;
1529 	int i;
1530 
1531 	spin_lock(&f->lock);
1532 	for (i = 0; i < f->num_members; i++) {
1533 		if (rcu_dereference_protected(f->arr[i],
1534 					      lockdep_is_held(&f->lock)) == sk)
1535 			break;
1536 	}
1537 	BUG_ON(i >= f->num_members);
1538 	rcu_assign_pointer(f->arr[i],
1539 			   rcu_dereference_protected(f->arr[f->num_members - 1],
1540 						     lockdep_is_held(&f->lock)));
1541 	f->num_members--;
1542 	if (f->num_members == 0)
1543 		__dev_remove_pack(&f->prot_hook);
1544 	spin_unlock(&f->lock);
1545 }
1546 
1547 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1548 {
1549 	if (sk->sk_family != PF_PACKET)
1550 		return false;
1551 
1552 	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1553 }
1554 
1555 static void fanout_init_data(struct packet_fanout *f)
1556 {
1557 	switch (f->type) {
1558 	case PACKET_FANOUT_LB:
1559 		atomic_set(&f->rr_cur, 0);
1560 		break;
1561 	case PACKET_FANOUT_CBPF:
1562 	case PACKET_FANOUT_EBPF:
1563 		RCU_INIT_POINTER(f->bpf_prog, NULL);
1564 		break;
1565 	}
1566 }
1567 
1568 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1569 {
1570 	struct bpf_prog *old;
1571 
1572 	spin_lock(&f->lock);
1573 	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1574 	rcu_assign_pointer(f->bpf_prog, new);
1575 	spin_unlock(&f->lock);
1576 
1577 	if (old) {
1578 		synchronize_net();
1579 		bpf_prog_destroy(old);
1580 	}
1581 }
1582 
1583 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1584 				unsigned int len)
1585 {
1586 	struct bpf_prog *new;
1587 	struct sock_fprog fprog;
1588 	int ret;
1589 
1590 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1591 		return -EPERM;
1592 
1593 	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1594 	if (ret)
1595 		return ret;
1596 
1597 	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1598 	if (ret)
1599 		return ret;
1600 
1601 	__fanout_set_data_bpf(po->fanout, new);
1602 	return 0;
1603 }
1604 
1605 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1606 				unsigned int len)
1607 {
1608 	struct bpf_prog *new;
1609 	u32 fd;
1610 
1611 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1612 		return -EPERM;
1613 	if (len != sizeof(fd))
1614 		return -EINVAL;
1615 	if (copy_from_sockptr(&fd, data, len))
1616 		return -EFAULT;
1617 
1618 	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1619 	if (IS_ERR(new))
1620 		return PTR_ERR(new);
1621 
1622 	__fanout_set_data_bpf(po->fanout, new);
1623 	return 0;
1624 }
1625 
1626 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1627 			   unsigned int len)
1628 {
1629 	switch (po->fanout->type) {
1630 	case PACKET_FANOUT_CBPF:
1631 		return fanout_set_data_cbpf(po, data, len);
1632 	case PACKET_FANOUT_EBPF:
1633 		return fanout_set_data_ebpf(po, data, len);
1634 	default:
1635 		return -EINVAL;
1636 	}
1637 }
1638 
1639 static void fanout_release_data(struct packet_fanout *f)
1640 {
1641 	switch (f->type) {
1642 	case PACKET_FANOUT_CBPF:
1643 	case PACKET_FANOUT_EBPF:
1644 		__fanout_set_data_bpf(f, NULL);
1645 	}
1646 }
1647 
1648 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1649 {
1650 	struct packet_fanout *f;
1651 
1652 	list_for_each_entry(f, &fanout_list, list) {
1653 		if (f->id == candidate_id &&
1654 		    read_pnet(&f->net) == sock_net(sk)) {
1655 			return false;
1656 		}
1657 	}
1658 	return true;
1659 }
1660 
1661 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1662 {
1663 	u16 id = fanout_next_id;
1664 
1665 	do {
1666 		if (__fanout_id_is_free(sk, id)) {
1667 			*new_id = id;
1668 			fanout_next_id = id + 1;
1669 			return true;
1670 		}
1671 
1672 		id++;
1673 	} while (id != fanout_next_id);
1674 
1675 	return false;
1676 }
1677 
1678 static int fanout_add(struct sock *sk, struct fanout_args *args)
1679 {
1680 	struct packet_rollover *rollover = NULL;
1681 	struct packet_sock *po = pkt_sk(sk);
1682 	u16 type_flags = args->type_flags;
1683 	struct packet_fanout *f, *match;
1684 	u8 type = type_flags & 0xff;
1685 	u8 flags = type_flags >> 8;
1686 	u16 id = args->id;
1687 	int err;
1688 
1689 	switch (type) {
1690 	case PACKET_FANOUT_ROLLOVER:
1691 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1692 			return -EINVAL;
1693 		break;
1694 	case PACKET_FANOUT_HASH:
1695 	case PACKET_FANOUT_LB:
1696 	case PACKET_FANOUT_CPU:
1697 	case PACKET_FANOUT_RND:
1698 	case PACKET_FANOUT_QM:
1699 	case PACKET_FANOUT_CBPF:
1700 	case PACKET_FANOUT_EBPF:
1701 		break;
1702 	default:
1703 		return -EINVAL;
1704 	}
1705 
1706 	mutex_lock(&fanout_mutex);
1707 
1708 	err = -EALREADY;
1709 	if (po->fanout)
1710 		goto out;
1711 
1712 	if (type == PACKET_FANOUT_ROLLOVER ||
1713 	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1714 		err = -ENOMEM;
1715 		rollover = kzalloc_obj(*rollover);
1716 		if (!rollover)
1717 			goto out;
1718 		atomic_long_set(&rollover->num, 0);
1719 		atomic_long_set(&rollover->num_huge, 0);
1720 		atomic_long_set(&rollover->num_failed, 0);
1721 	}
1722 
1723 	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1724 		if (id != 0) {
1725 			err = -EINVAL;
1726 			goto out;
1727 		}
1728 		if (!fanout_find_new_id(sk, &id)) {
1729 			err = -ENOMEM;
1730 			goto out;
1731 		}
1732 		/* ephemeral flag for the first socket in the group: drop it */
1733 		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1734 	}
1735 
1736 	match = NULL;
1737 	list_for_each_entry(f, &fanout_list, list) {
1738 		if (f->id == id &&
1739 		    read_pnet(&f->net) == sock_net(sk)) {
1740 			match = f;
1741 			break;
1742 		}
1743 	}
1744 	err = -EINVAL;
1745 	if (match) {
1746 		if (match->flags != flags)
1747 			goto out;
1748 		if (args->max_num_members &&
1749 		    args->max_num_members != match->max_num_members)
1750 			goto out;
1751 	} else {
1752 		if (args->max_num_members > PACKET_FANOUT_MAX)
1753 			goto out;
1754 		if (!args->max_num_members)
1755 			/* legacy PACKET_FANOUT_MAX */
1756 			args->max_num_members = 256;
1757 		err = -ENOMEM;
1758 		match = kvzalloc_flex(*match, arr, args->max_num_members);
1759 		if (!match)
1760 			goto out;
1761 		write_pnet(&match->net, sock_net(sk));
1762 		match->id = id;
1763 		match->type = type;
1764 		match->flags = flags;
1765 		INIT_LIST_HEAD(&match->list);
1766 		spin_lock_init(&match->lock);
1767 		refcount_set(&match->sk_ref, 0);
1768 		fanout_init_data(match);
1769 		match->prot_hook.type = po->prot_hook.type;
1770 		match->prot_hook.dev = po->prot_hook.dev;
1771 		match->prot_hook.func = packet_rcv_fanout;
1772 		match->prot_hook.af_packet_priv = match;
1773 		match->prot_hook.af_packet_net = read_pnet(&match->net);
1774 		match->prot_hook.id_match = match_fanout_group;
1775 		match->max_num_members = args->max_num_members;
1776 		match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1777 		list_add(&match->list, &fanout_list);
1778 	}
1779 	err = -EINVAL;
1780 
1781 	spin_lock(&po->bind_lock);
1782 	if (po->num &&
1783 	    match->type == type &&
1784 	    match->prot_hook.type == po->prot_hook.type &&
1785 	    match->prot_hook.dev == po->prot_hook.dev) {
1786 		err = -ENOSPC;
1787 		if (refcount_read(&match->sk_ref) < match->max_num_members) {
1788 			/* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1789 			WRITE_ONCE(po->fanout, match);
1790 
1791 			po->rollover = rollover;
1792 			rollover = NULL;
1793 			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1794 			if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
1795 				__dev_remove_pack(&po->prot_hook);
1796 				__fanout_link(sk, po);
1797 			}
1798 			err = 0;
1799 		}
1800 	}
1801 	spin_unlock(&po->bind_lock);
1802 
1803 	if (err && !refcount_read(&match->sk_ref)) {
1804 		list_del(&match->list);
1805 		kvfree(match);
1806 	}
1807 
1808 out:
1809 	kfree(rollover);
1810 	mutex_unlock(&fanout_mutex);
1811 	return err;
1812 }
1813 
1814 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1815  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1816  * It is the responsibility of the caller to call fanout_release_data() and
1817  * free the returned packet_fanout (after synchronize_net())
1818  */
1819 static struct packet_fanout *fanout_release(struct sock *sk)
1820 {
1821 	struct packet_sock *po = pkt_sk(sk);
1822 	struct packet_fanout *f;
1823 
1824 	mutex_lock(&fanout_mutex);
1825 	f = po->fanout;
1826 	if (f) {
1827 		po->fanout = NULL;
1828 
1829 		if (refcount_dec_and_test(&f->sk_ref))
1830 			list_del(&f->list);
1831 		else
1832 			f = NULL;
1833 	}
1834 	mutex_unlock(&fanout_mutex);
1835 
1836 	return f;
1837 }
1838 
1839 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1840 					  struct sk_buff *skb)
1841 {
1842 	/* Earlier code assumed this would be a VLAN pkt, double-check
1843 	 * this now that we have the actual packet in hand. We can only
1844 	 * do this check on Ethernet devices.
1845 	 */
1846 	if (unlikely(dev->type != ARPHRD_ETHER))
1847 		return false;
1848 
1849 	skb_reset_mac_header(skb);
1850 	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1851 }
1852 
1853 static const struct proto_ops packet_ops;
1854 
1855 static const struct proto_ops packet_ops_spkt;
1856 
1857 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1858 			   struct packet_type *pt, struct net_device *orig_dev)
1859 {
1860 	struct sock *sk;
1861 	struct sockaddr_pkt *spkt;
1862 
1863 	/*
1864 	 *	When we registered the protocol we saved the socket in the data
1865 	 *	field for just this event.
1866 	 */
1867 
1868 	sk = pt->af_packet_priv;
1869 
1870 	/*
1871 	 *	Yank back the headers [hope the device set this
1872 	 *	right or kerboom...]
1873 	 *
1874 	 *	Incoming packets have ll header pulled,
1875 	 *	push it back.
1876 	 *
1877 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1878 	 *	so that this procedure is noop.
1879 	 */
1880 
1881 	if (skb->pkt_type == PACKET_LOOPBACK)
1882 		goto out;
1883 
1884 	if (!net_eq(dev_net(dev), sock_net(sk)))
1885 		goto out;
1886 
1887 	skb = skb_share_check(skb, GFP_ATOMIC);
1888 	if (skb == NULL)
1889 		goto oom;
1890 
1891 	/* drop any routing info */
1892 	skb_dst_drop(skb);
1893 
1894 	/* drop conntrack reference */
1895 	nf_reset_ct(skb);
1896 
1897 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1898 
1899 	skb_push(skb, skb->data - skb_mac_header(skb));
1900 
1901 	/*
1902 	 *	The SOCK_PACKET socket receives _all_ frames.
1903 	 */
1904 
1905 	spkt->spkt_family = dev->type;
1906 	strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1907 	spkt->spkt_protocol = skb->protocol;
1908 
1909 	/*
1910 	 *	Charge the memory to the socket. This is done specifically
1911 	 *	to prevent sockets using all the memory up.
1912 	 */
1913 
1914 	if (sock_queue_rcv_skb(sk, skb) == 0)
1915 		return 0;
1916 
1917 out:
1918 	kfree_skb(skb);
1919 oom:
1920 	return 0;
1921 }
1922 
1923 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1924 {
1925 	int depth;
1926 
1927 	/* On TX skb->data is the L2 header; anchor it for all socket types. */
1928 	skb_reset_mac_header(skb);
1929 
1930 	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1931 	    sock->type == SOCK_RAW)
1932 		skb->protocol = dev_parse_header_protocol(skb);
1933 
1934 	/* Move network header to the right position for VLAN tagged packets */
1935 	if (likely(skb->dev->type == ARPHRD_ETHER) &&
1936 	    eth_type_vlan(skb->protocol) &&
1937 	    vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1938 		skb_set_network_header(skb, depth);
1939 
1940 	skb_probe_transport_header(skb);
1941 }
1942 
1943 /*
1944  *	Output a raw packet to a device layer. This bypasses all the other
1945  *	protocol layers and you must therefore supply it with a complete frame
1946  */
1947 
1948 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1949 			       size_t len)
1950 {
1951 	struct sock *sk = sock->sk;
1952 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1953 	struct sk_buff *skb = NULL;
1954 	struct net_device *dev;
1955 	struct sockcm_cookie sockc;
1956 	__be16 proto = 0;
1957 	int err;
1958 	int extra_len = 0;
1959 
1960 	/*
1961 	 *	Get and verify the address.
1962 	 */
1963 
1964 	if (saddr) {
1965 		if (msg->msg_namelen < sizeof(struct sockaddr))
1966 			return -EINVAL;
1967 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1968 			proto = saddr->spkt_protocol;
1969 	} else
1970 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1971 
1972 	/*
1973 	 *	Find the device first to size check it
1974 	 */
1975 
1976 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1977 retry:
1978 	rcu_read_lock();
1979 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1980 	err = -ENODEV;
1981 	if (dev == NULL)
1982 		goto out_unlock;
1983 
1984 	err = -ENETDOWN;
1985 	if (!(dev->flags & IFF_UP))
1986 		goto out_unlock;
1987 
1988 	/*
1989 	 * You may not queue a frame bigger than the mtu. This is the lowest level
1990 	 * raw protocol and you must do your own fragmentation at this level.
1991 	 */
1992 
1993 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1994 		if (!netif_supports_nofcs(dev)) {
1995 			err = -EPROTONOSUPPORT;
1996 			goto out_unlock;
1997 		}
1998 		extra_len = 4; /* We're doing our own CRC */
1999 	}
2000 
2001 	err = -EMSGSIZE;
2002 	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2003 		goto out_unlock;
2004 
2005 	if (!skb) {
2006 		size_t reserved = LL_RESERVED_SPACE(dev);
2007 		int tlen = dev->needed_tailroom;
2008 		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2009 
2010 		rcu_read_unlock();
2011 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2012 		if (skb == NULL)
2013 			return -ENOBUFS;
2014 		/* FIXME: Save some space for broken drivers that write a hard
2015 		 * header at transmission time by themselves. PPP is the notable
2016 		 * one here. This should really be fixed at the driver level.
2017 		 */
2018 		skb_reserve(skb, reserved);
2019 		skb_reset_network_header(skb);
2020 
2021 		/* Try to align data part correctly */
2022 		if (hhlen) {
2023 			skb->data -= hhlen;
2024 			skb->tail -= hhlen;
2025 			if (len < hhlen)
2026 				skb_reset_network_header(skb);
2027 		}
2028 		err = memcpy_from_msg(skb_put(skb, len), msg, len);
2029 		if (err)
2030 			goto out_free;
2031 		goto retry;
2032 	}
2033 
2034 	if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2035 		err = -EINVAL;
2036 		goto out_unlock;
2037 	}
2038 	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2039 	    !packet_extra_vlan_len_allowed(dev, skb)) {
2040 		err = -EMSGSIZE;
2041 		goto out_unlock;
2042 	}
2043 
2044 	sockcm_init(&sockc, sk);
2045 	if (msg->msg_controllen) {
2046 		err = sock_cmsg_send(sk, msg, &sockc);
2047 		if (unlikely(err))
2048 			goto out_unlock;
2049 	}
2050 
2051 	skb->protocol = proto;
2052 	skb->dev = dev;
2053 	skb->priority = sockc.priority;
2054 	skb->mark = sockc.mark;
2055 	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
2056 	skb_setup_tx_timestamp(skb, &sockc);
2057 
2058 	if (unlikely(extra_len == 4))
2059 		skb->no_fcs = 1;
2060 
2061 	packet_parse_headers(skb, sock);
2062 
2063 	dev_queue_xmit(skb);
2064 	rcu_read_unlock();
2065 	return len;
2066 
2067 out_unlock:
2068 	rcu_read_unlock();
2069 out_free:
2070 	kfree_skb(skb);
2071 	return err;
2072 }
2073 
2074 static unsigned int run_filter(struct sk_buff *skb,
2075 			       const struct sock *sk,
2076 			       unsigned int res)
2077 {
2078 	struct sk_filter *filter;
2079 
2080 	rcu_read_lock();
2081 	filter = rcu_dereference(sk->sk_filter);
2082 	if (filter != NULL)
2083 		res = bpf_prog_run_clear_cb(filter->prog, skb);
2084 	rcu_read_unlock();
2085 
2086 	return res;
2087 }
2088 
2089 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2090 			   size_t *len, int vnet_hdr_sz)
2091 {
2092 	struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2093 
2094 	if (*len < vnet_hdr_sz)
2095 		return -EINVAL;
2096 	*len -= vnet_hdr_sz;
2097 
2098 	if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2099 		return -EINVAL;
2100 
2101 	return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2102 }
2103 
2104 /*
2105  * This function makes lazy skb cloning in hope that most of packets
2106  * are discarded by BPF.
2107  *
2108  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2109  * and skb->cb are mangled. It works because (and until) packets
2110  * falling here are owned by current CPU. Output packets are cloned
2111  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2112  * sequentially, so that if we return skb to original state on exit,
2113  * we will not harm anyone.
2114  */
2115 
2116 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2117 		      struct packet_type *pt, struct net_device *orig_dev)
2118 {
2119 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2120 	struct sock *sk = NULL;
2121 	struct sockaddr_ll *sll;
2122 	struct packet_sock *po;
2123 	u8 *skb_head = skb->data;
2124 	int skb_len = skb->len;
2125 	unsigned int snaplen, res;
2126 
2127 	if (skb->pkt_type == PACKET_LOOPBACK)
2128 		goto drop;
2129 
2130 	sk = pt->af_packet_priv;
2131 	po = pkt_sk(sk);
2132 
2133 	if (!net_eq(dev_net(dev), sock_net(sk)))
2134 		goto drop;
2135 
2136 	skb->dev = dev;
2137 
2138 	if (dev_has_header(dev)) {
2139 		/* The device has an explicit notion of ll header,
2140 		 * exported to higher levels.
2141 		 *
2142 		 * Otherwise, the device hides details of its frame
2143 		 * structure, so that corresponding packet head is
2144 		 * never delivered to user.
2145 		 */
2146 		if (sk->sk_type != SOCK_DGRAM)
2147 			skb_push(skb, skb->data - skb_mac_header(skb));
2148 		else if (skb->pkt_type == PACKET_OUTGOING) {
2149 			/* Special case: outgoing packets have ll header at head */
2150 			skb_pull(skb, skb_network_offset(skb));
2151 		}
2152 	}
2153 
2154 	snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2155 
2156 	res = run_filter(skb, sk, snaplen);
2157 	if (!res)
2158 		goto drop_n_restore;
2159 	if (snaplen > res)
2160 		snaplen = res;
2161 
2162 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2163 		goto drop_n_acct;
2164 
2165 	if (skb_shared(skb)) {
2166 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2167 		if (nskb == NULL)
2168 			goto drop_n_acct;
2169 
2170 		if (skb_head != skb->data) {
2171 			skb->data = skb_head;
2172 			skb->len = skb_len;
2173 		}
2174 		consume_skb(skb);
2175 		skb = nskb;
2176 	}
2177 
2178 	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2179 
2180 	sll = &PACKET_SKB_CB(skb)->sa.ll;
2181 	sll->sll_hatype = dev->type;
2182 	sll->sll_pkttype = skb->pkt_type;
2183 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2184 		sll->sll_ifindex = orig_dev->ifindex;
2185 	else
2186 		sll->sll_ifindex = dev->ifindex;
2187 
2188 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2189 
2190 	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2191 	 * Use their space for storing the original skb length.
2192 	 */
2193 	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2194 
2195 	if (pskb_trim(skb, snaplen))
2196 		goto drop_n_acct;
2197 
2198 	skb_set_owner_r(skb, sk);
2199 	skb->dev = NULL;
2200 	skb_dst_drop(skb);
2201 
2202 	/* drop conntrack reference */
2203 	nf_reset_ct(skb);
2204 
2205 	spin_lock(&sk->sk_receive_queue.lock);
2206 	po->stats.stats1.tp_packets++;
2207 	sock_skb_set_dropcount(sk, skb);
2208 	skb_clear_delivery_time(skb);
2209 	__skb_queue_tail(&sk->sk_receive_queue, skb);
2210 	spin_unlock(&sk->sk_receive_queue.lock);
2211 	sk->sk_data_ready(sk);
2212 	return 0;
2213 
2214 drop_n_acct:
2215 	atomic_inc(&po->tp_drops);
2216 	sk_drops_inc(sk);
2217 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2218 
2219 drop_n_restore:
2220 	if (skb_head != skb->data && skb_shared(skb)) {
2221 		skb->data = skb_head;
2222 		skb->len = skb_len;
2223 	}
2224 drop:
2225 	sk_skb_reason_drop(sk, skb, drop_reason);
2226 	return 0;
2227 }
2228 
2229 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2230 		       struct packet_type *pt, struct net_device *orig_dev)
2231 {
2232 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2233 	struct sock *sk = NULL;
2234 	struct packet_sock *po;
2235 	struct sockaddr_ll *sll;
2236 	union tpacket_uhdr h;
2237 	u8 *skb_head = skb->data;
2238 	int skb_len = skb->len;
2239 	unsigned int snaplen, res;
2240 	unsigned long status = TP_STATUS_USER;
2241 	unsigned short macoff, hdrlen;
2242 	unsigned int netoff;
2243 	struct sk_buff *copy_skb = NULL;
2244 	struct timespec64 ts;
2245 	__u32 ts_status;
2246 	unsigned int slot_id = 0;
2247 	int vnet_hdr_sz = 0;
2248 
2249 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2250 	 * We may add members to them until current aligned size without forcing
2251 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2252 	 */
2253 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2254 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2255 
2256 	if (skb->pkt_type == PACKET_LOOPBACK)
2257 		goto drop;
2258 
2259 	sk = pt->af_packet_priv;
2260 	po = pkt_sk(sk);
2261 
2262 	if (!net_eq(dev_net(dev), sock_net(sk)))
2263 		goto drop;
2264 
2265 	if (dev_has_header(dev)) {
2266 		if (sk->sk_type != SOCK_DGRAM)
2267 			skb_push(skb, skb->data - skb_mac_header(skb));
2268 		else if (skb->pkt_type == PACKET_OUTGOING) {
2269 			/* Special case: outgoing packets have ll header at head */
2270 			skb_pull(skb, skb_network_offset(skb));
2271 		}
2272 	}
2273 
2274 	snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2275 
2276 	res = run_filter(skb, sk, snaplen);
2277 	if (!res)
2278 		goto drop_n_restore;
2279 
2280 	/* If we are flooded, just give up */
2281 	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2282 		atomic_inc(&po->tp_drops);
2283 		goto drop_n_restore;
2284 	}
2285 
2286 	if (skb->ip_summed == CHECKSUM_PARTIAL)
2287 		status |= TP_STATUS_CSUMNOTREADY;
2288 	else if (skb->pkt_type != PACKET_OUTGOING &&
2289 		 skb_csum_unnecessary(skb))
2290 		status |= TP_STATUS_CSUM_VALID;
2291 	if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2292 		status |= TP_STATUS_GSO_TCP;
2293 
2294 	if (snaplen > res)
2295 		snaplen = res;
2296 
2297 	if (sk->sk_type == SOCK_DGRAM) {
2298 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2299 				  po->tp_reserve;
2300 	} else {
2301 		unsigned int maclen = skb_network_offset(skb);
2302 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2303 				       (maclen < 16 ? 16 : maclen)) +
2304 				       po->tp_reserve;
2305 		vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2306 		if (vnet_hdr_sz)
2307 			netoff += vnet_hdr_sz;
2308 		macoff = netoff - maclen;
2309 	}
2310 	if (netoff > USHRT_MAX) {
2311 		atomic_inc(&po->tp_drops);
2312 		goto drop_n_restore;
2313 	}
2314 	if (po->tp_version <= TPACKET_V2) {
2315 		if (macoff + snaplen > po->rx_ring.frame_size) {
2316 			if (READ_ONCE(po->copy_thresh) &&
2317 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2318 				if (skb_shared(skb)) {
2319 					copy_skb = skb_clone(skb, GFP_ATOMIC);
2320 				} else {
2321 					copy_skb = skb_get(skb);
2322 					skb_head = skb->data;
2323 				}
2324 				if (copy_skb) {
2325 					memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2326 					       sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2327 					skb_set_owner_r(copy_skb, sk);
2328 				}
2329 			}
2330 			snaplen = po->rx_ring.frame_size - macoff;
2331 			if ((int)snaplen < 0) {
2332 				snaplen = 0;
2333 				vnet_hdr_sz = 0;
2334 			}
2335 		}
2336 	} else if (unlikely(macoff + snaplen >
2337 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2338 		u32 nval;
2339 
2340 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2341 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2342 			    snaplen, nval, macoff);
2343 		snaplen = nval;
2344 		if (unlikely((int)snaplen < 0)) {
2345 			snaplen = 0;
2346 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2347 			vnet_hdr_sz = 0;
2348 		}
2349 	}
2350 	spin_lock(&sk->sk_receive_queue.lock);
2351 	h.raw = packet_current_rx_frame(po, skb,
2352 					TP_STATUS_KERNEL, (macoff+snaplen));
2353 	if (!h.raw)
2354 		goto drop_n_account;
2355 
2356 	if (po->tp_version <= TPACKET_V2) {
2357 		slot_id = po->rx_ring.head;
2358 		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2359 			goto drop_n_account;
2360 		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2361 	}
2362 
2363 	if (vnet_hdr_sz &&
2364 	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2365 				    sizeof(struct virtio_net_hdr),
2366 				    vio_le(), true, 0)) {
2367 		if (po->tp_version == TPACKET_V3)
2368 			prb_clear_blk_fill_status(&po->rx_ring);
2369 		goto drop_n_account;
2370 	}
2371 
2372 	if (po->tp_version <= TPACKET_V2) {
2373 		packet_increment_rx_head(po, &po->rx_ring);
2374 	/*
2375 	 * LOSING will be reported till you read the stats,
2376 	 * because it's COR - Clear On Read.
2377 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2378 	 * at packet level.
2379 	 */
2380 		if (atomic_read(&po->tp_drops))
2381 			status |= TP_STATUS_LOSING;
2382 	}
2383 
2384 	po->stats.stats1.tp_packets++;
2385 	if (copy_skb) {
2386 		status |= TP_STATUS_COPY;
2387 		skb_clear_delivery_time(copy_skb);
2388 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2389 	}
2390 	spin_unlock(&sk->sk_receive_queue.lock);
2391 
2392 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2393 
2394 	/* Always timestamp; prefer an existing software timestamp taken
2395 	 * closer to the time of capture.
2396 	 */
2397 	ts_status = tpacket_get_timestamp(skb, &ts,
2398 					  READ_ONCE(po->tp_tstamp) |
2399 					  SOF_TIMESTAMPING_SOFTWARE);
2400 	if (!ts_status)
2401 		ktime_get_real_ts64(&ts);
2402 
2403 	status |= ts_status;
2404 
2405 	switch (po->tp_version) {
2406 	case TPACKET_V1:
2407 		h.h1->tp_len = skb->len;
2408 		h.h1->tp_snaplen = snaplen;
2409 		h.h1->tp_mac = macoff;
2410 		h.h1->tp_net = netoff;
2411 		h.h1->tp_sec = ts.tv_sec;
2412 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2413 		hdrlen = sizeof(*h.h1);
2414 		break;
2415 	case TPACKET_V2:
2416 		h.h2->tp_len = skb->len;
2417 		h.h2->tp_snaplen = snaplen;
2418 		h.h2->tp_mac = macoff;
2419 		h.h2->tp_net = netoff;
2420 		h.h2->tp_sec = ts.tv_sec;
2421 		h.h2->tp_nsec = ts.tv_nsec;
2422 		if (skb_vlan_tag_present(skb)) {
2423 			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2424 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2425 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2426 		} else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
2427 			h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev);
2428 			h.h2->tp_vlan_tpid = ntohs(skb->protocol);
2429 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2430 		} else {
2431 			h.h2->tp_vlan_tci = 0;
2432 			h.h2->tp_vlan_tpid = 0;
2433 		}
2434 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2435 		hdrlen = sizeof(*h.h2);
2436 		break;
2437 	case TPACKET_V3:
2438 		/* tp_nxt_offset,vlan are already populated above.
2439 		 * So DONT clear those fields here
2440 		 */
2441 		h.h3->tp_status |= status;
2442 		h.h3->tp_len = skb->len;
2443 		h.h3->tp_snaplen = snaplen;
2444 		h.h3->tp_mac = macoff;
2445 		h.h3->tp_net = netoff;
2446 		h.h3->tp_sec  = ts.tv_sec;
2447 		h.h3->tp_nsec = ts.tv_nsec;
2448 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2449 		hdrlen = sizeof(*h.h3);
2450 		break;
2451 	default:
2452 		BUG();
2453 	}
2454 
2455 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2456 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2457 	sll->sll_family = AF_PACKET;
2458 	sll->sll_hatype = dev->type;
2459 	sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ?
2460 		vlan_get_protocol_dgram(skb) : skb->protocol;
2461 	sll->sll_pkttype = skb->pkt_type;
2462 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2463 		sll->sll_ifindex = orig_dev->ifindex;
2464 	else
2465 		sll->sll_ifindex = dev->ifindex;
2466 
2467 	smp_mb();
2468 
2469 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2470 	if (po->tp_version <= TPACKET_V2) {
2471 		u8 *start, *end;
2472 
2473 		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2474 					macoff + snaplen);
2475 
2476 		for (start = h.raw; start < end; start += PAGE_SIZE)
2477 			flush_dcache_page(pgv_to_page(start));
2478 	}
2479 	smp_wmb();
2480 #endif
2481 
2482 	if (po->tp_version <= TPACKET_V2) {
2483 		spin_lock(&sk->sk_receive_queue.lock);
2484 		__packet_set_status(po, h.raw, status);
2485 		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2486 		spin_unlock(&sk->sk_receive_queue.lock);
2487 		sk->sk_data_ready(sk);
2488 	} else if (po->tp_version == TPACKET_V3) {
2489 		prb_clear_blk_fill_status(&po->rx_ring);
2490 	}
2491 
2492 drop_n_restore:
2493 	if (skb_head != skb->data && skb_shared(skb)) {
2494 		skb->data = skb_head;
2495 		skb->len = skb_len;
2496 	}
2497 drop:
2498 	sk_skb_reason_drop(sk, skb, drop_reason);
2499 	return 0;
2500 
2501 drop_n_account:
2502 	spin_unlock(&sk->sk_receive_queue.lock);
2503 	atomic_inc(&po->tp_drops);
2504 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2505 
2506 	sk->sk_data_ready(sk);
2507 	sk_skb_reason_drop(sk, copy_skb, drop_reason);
2508 	goto drop_n_restore;
2509 }
2510 
2511 static void tpacket_destruct_skb(struct sk_buff *skb)
2512 {
2513 	struct packet_sock *po = pkt_sk(skb->sk);
2514 
2515 	if (likely(po->tx_ring.pg_vec)) {
2516 		void *ph;
2517 		__u32 ts;
2518 
2519 		ph = skb_zcopy_get_nouarg(skb);
2520 		packet_dec_pending(&po->tx_ring);
2521 
2522 		ts = __packet_set_timestamp(po, ph, skb);
2523 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2524 
2525 		complete(&po->skb_completion);
2526 	}
2527 
2528 	sock_wfree(skb);
2529 }
2530 
2531 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2532 {
2533 	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2534 	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2535 	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2536 	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2537 		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2538 			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2539 			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2540 
2541 	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2542 		return -EINVAL;
2543 
2544 	return 0;
2545 }
2546 
2547 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2548 				 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2549 {
2550 	int ret;
2551 
2552 	if (*len < vnet_hdr_sz)
2553 		return -EINVAL;
2554 	*len -= vnet_hdr_sz;
2555 
2556 	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2557 		return -EFAULT;
2558 
2559 	ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2560 	if (ret)
2561 		return ret;
2562 
2563 	/* move iter to point to the start of mac header */
2564 	if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2565 		iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2566 
2567 	return 0;
2568 }
2569 
2570 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2571 		void *frame, struct net_device *dev, void *data, int tp_len,
2572 		__be16 proto, unsigned char *addr, int hlen, int copylen,
2573 		const struct sockcm_cookie *sockc)
2574 {
2575 	union tpacket_uhdr ph;
2576 	int to_write, offset, len, nr_frags, len_max;
2577 	struct socket *sock = po->sk.sk_socket;
2578 	struct page *page;
2579 	int err;
2580 
2581 	ph.raw = frame;
2582 
2583 	skb->protocol = proto;
2584 	skb->dev = dev;
2585 	skb->priority = sockc->priority;
2586 	skb->mark = sockc->mark;
2587 	skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid);
2588 	skb_setup_tx_timestamp(skb, sockc);
2589 	skb_zcopy_set_nouarg(skb, ph.raw);
2590 
2591 	skb_reserve(skb, hlen);
2592 	skb_reset_network_header(skb);
2593 
2594 	to_write = tp_len;
2595 
2596 	if (sock->type == SOCK_DGRAM) {
2597 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2598 				NULL, tp_len);
2599 		if (unlikely(err < 0))
2600 			return -EINVAL;
2601 	} else if (copylen) {
2602 		int hdrlen = min_t(int, copylen, tp_len);
2603 
2604 		skb_push(skb, dev->hard_header_len);
2605 		skb_put(skb, copylen - dev->hard_header_len);
2606 		err = skb_store_bits(skb, 0, data, hdrlen);
2607 		if (unlikely(err))
2608 			return err;
2609 		if (!dev_validate_header(dev, skb->data, hdrlen))
2610 			return -EINVAL;
2611 
2612 		data += hdrlen;
2613 		to_write -= hdrlen;
2614 	}
2615 
2616 	offset = offset_in_page(data);
2617 	len_max = PAGE_SIZE - offset;
2618 	len = ((to_write > len_max) ? len_max : to_write);
2619 
2620 	skb->data_len = to_write;
2621 	skb->len += to_write;
2622 	skb->truesize += to_write;
2623 	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2624 
2625 	while (likely(to_write)) {
2626 		nr_frags = skb_shinfo(skb)->nr_frags;
2627 
2628 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2629 			pr_err("Packet exceed the number of skb frags(%u)\n",
2630 			       (unsigned int)MAX_SKB_FRAGS);
2631 			return -EFAULT;
2632 		}
2633 
2634 		page = pgv_to_page(data);
2635 		data += len;
2636 		flush_dcache_page(page);
2637 		get_page(page);
2638 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2639 		to_write -= len;
2640 		offset = 0;
2641 		len_max = PAGE_SIZE;
2642 		len = ((to_write > len_max) ? len_max : to_write);
2643 	}
2644 
2645 	packet_parse_headers(skb, sock);
2646 
2647 	return tp_len;
2648 }
2649 
2650 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2651 				int size_max, void **data)
2652 {
2653 	union tpacket_uhdr ph;
2654 	int tp_len, off;
2655 
2656 	ph.raw = frame;
2657 
2658 	switch (po->tp_version) {
2659 	case TPACKET_V3:
2660 		if (ph.h3->tp_next_offset != 0) {
2661 			pr_warn_once("variable sized slot not supported");
2662 			return -EINVAL;
2663 		}
2664 		tp_len = ph.h3->tp_len;
2665 		break;
2666 	case TPACKET_V2:
2667 		tp_len = ph.h2->tp_len;
2668 		break;
2669 	default:
2670 		tp_len = ph.h1->tp_len;
2671 		break;
2672 	}
2673 	if (unlikely(tp_len > size_max)) {
2674 		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2675 		return -EMSGSIZE;
2676 	}
2677 
2678 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2679 		int off_min, off_max;
2680 
2681 		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2682 		off_max = po->tx_ring.frame_size - tp_len;
2683 		if (po->sk.sk_type == SOCK_DGRAM) {
2684 			switch (po->tp_version) {
2685 			case TPACKET_V3:
2686 				off = ph.h3->tp_net;
2687 				break;
2688 			case TPACKET_V2:
2689 				off = ph.h2->tp_net;
2690 				break;
2691 			default:
2692 				off = ph.h1->tp_net;
2693 				break;
2694 			}
2695 		} else {
2696 			switch (po->tp_version) {
2697 			case TPACKET_V3:
2698 				off = ph.h3->tp_mac;
2699 				break;
2700 			case TPACKET_V2:
2701 				off = ph.h2->tp_mac;
2702 				break;
2703 			default:
2704 				off = ph.h1->tp_mac;
2705 				break;
2706 			}
2707 		}
2708 		if (unlikely((off < off_min) || (off_max < off)))
2709 			return -EINVAL;
2710 	} else {
2711 		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2712 	}
2713 
2714 	*data = frame + off;
2715 	return tp_len;
2716 }
2717 
2718 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2719 {
2720 	struct sk_buff *skb = NULL;
2721 	struct net_device *dev;
2722 	struct virtio_net_hdr vnet_hdr;
2723 	bool has_vnet_hdr = false;
2724 	struct sockcm_cookie sockc;
2725 	__be16 proto;
2726 	int err, reserve = 0;
2727 	void *ph;
2728 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2729 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2730 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2731 	unsigned char *addr = NULL;
2732 	int tp_len, size_max;
2733 	void *data;
2734 	int len_sum = 0;
2735 	int status = TP_STATUS_AVAILABLE;
2736 	int hlen, tlen, copylen = 0;
2737 	long timeo;
2738 
2739 	mutex_lock(&po->pg_vec_lock);
2740 
2741 	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2742 	 * we need to confirm it under protection of pg_vec_lock.
2743 	 */
2744 	if (unlikely(!po->tx_ring.pg_vec)) {
2745 		err = -EBUSY;
2746 		goto out;
2747 	}
2748 	if (likely(saddr == NULL)) {
2749 		dev	= packet_cached_dev_get(po);
2750 		proto	= READ_ONCE(po->num);
2751 	} else {
2752 		err = -EINVAL;
2753 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2754 			goto out;
2755 		if (msg->msg_namelen < (saddr->sll_halen
2756 					+ offsetof(struct sockaddr_ll,
2757 						sll_addr)))
2758 			goto out;
2759 		proto	= saddr->sll_protocol;
2760 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2761 		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2762 			if (dev && msg->msg_namelen < dev->addr_len +
2763 				   offsetof(struct sockaddr_ll, sll_addr))
2764 				goto out_put;
2765 			addr = saddr->sll_addr;
2766 		}
2767 	}
2768 
2769 	err = -ENXIO;
2770 	if (unlikely(dev == NULL))
2771 		goto out;
2772 	err = -ENETDOWN;
2773 	if (unlikely(!(dev->flags & IFF_UP)))
2774 		goto out_put;
2775 
2776 	sockcm_init(&sockc, &po->sk);
2777 	if (msg->msg_controllen) {
2778 		err = sock_cmsg_send(&po->sk, msg, &sockc);
2779 		if (unlikely(err))
2780 			goto out_put;
2781 	}
2782 
2783 	if (po->sk.sk_socket->type == SOCK_RAW)
2784 		reserve = dev->hard_header_len;
2785 	size_max = po->tx_ring.frame_size
2786 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2787 
2788 	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2789 		size_max = dev->mtu + reserve + VLAN_HLEN;
2790 
2791 	timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2792 	reinit_completion(&po->skb_completion);
2793 
2794 	do {
2795 		ph = packet_current_frame(po, &po->tx_ring,
2796 					  TP_STATUS_SEND_REQUEST);
2797 		if (unlikely(ph == NULL)) {
2798 			/* Note: packet_read_pending() might be slow if we
2799 			 * have to call it as it's per_cpu variable, but in
2800 			 * fast-path we don't have to call it, only when ph
2801 			 * is NULL, we need to check the pending_refcnt.
2802 			 */
2803 			if (need_wait && packet_read_pending(&po->tx_ring)) {
2804 				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2805 				if (timeo <= 0) {
2806 					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2807 					goto out_put;
2808 				}
2809 				/* check for additional frames */
2810 				continue;
2811 			} else
2812 				break;
2813 		}
2814 
2815 		skb = NULL;
2816 		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2817 		if (tp_len < 0)
2818 			goto tpacket_error;
2819 
2820 		status = TP_STATUS_SEND_REQUEST;
2821 		hlen = LL_RESERVED_SPACE(dev);
2822 		tlen = dev->needed_tailroom;
2823 		if (vnet_hdr_sz) {
2824 			data += vnet_hdr_sz;
2825 			tp_len -= vnet_hdr_sz;
2826 			if (tp_len < 0) {
2827 				tp_len = -EINVAL;
2828 				goto tpacket_error;
2829 			}
2830 			memcpy(&vnet_hdr, data - vnet_hdr_sz, sizeof(vnet_hdr));
2831 			if (__packet_snd_vnet_parse(&vnet_hdr, tp_len)) {
2832 				tp_len = -EINVAL;
2833 				goto tpacket_error;
2834 			}
2835 			copylen = __virtio16_to_cpu(vio_le(),
2836 						    vnet_hdr.hdr_len);
2837 			has_vnet_hdr = true;
2838 		}
2839 		copylen = max_t(int, copylen, dev->hard_header_len);
2840 		skb = sock_alloc_send_skb(&po->sk,
2841 				hlen + tlen + sizeof(struct sockaddr_ll) +
2842 				(copylen - dev->hard_header_len),
2843 				!need_wait, &err);
2844 
2845 		if (unlikely(skb == NULL)) {
2846 			/* we assume the socket was initially writeable ... */
2847 			if (likely(len_sum > 0))
2848 				err = len_sum;
2849 			goto out_status;
2850 		}
2851 		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2852 					  addr, hlen, copylen, &sockc);
2853 		if (likely(tp_len >= 0) &&
2854 		    tp_len > dev->mtu + reserve &&
2855 		    !vnet_hdr_sz &&
2856 		    !packet_extra_vlan_len_allowed(dev, skb))
2857 			tp_len = -EMSGSIZE;
2858 
2859 		if (unlikely(tp_len < 0)) {
2860 tpacket_error:
2861 			if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2862 				__packet_set_status(po, ph,
2863 						TP_STATUS_AVAILABLE);
2864 				packet_increment_head(&po->tx_ring);
2865 				kfree_skb(skb);
2866 				continue;
2867 			} else {
2868 				status = TP_STATUS_WRONG_FORMAT;
2869 				err = tp_len;
2870 				goto out_status;
2871 			}
2872 		}
2873 
2874 		if (has_vnet_hdr) {
2875 			if (virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le())) {
2876 				tp_len = -EINVAL;
2877 				goto tpacket_error;
2878 			}
2879 			virtio_net_hdr_set_proto(skb, &vnet_hdr);
2880 		}
2881 
2882 		skb->destructor = tpacket_destruct_skb;
2883 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2884 		packet_inc_pending(&po->tx_ring);
2885 
2886 		status = TP_STATUS_SEND_REQUEST;
2887 		err = packet_xmit(po, skb);
2888 		if (unlikely(err != 0)) {
2889 			if (err > 0)
2890 				err = net_xmit_errno(err);
2891 			if (err && __packet_get_status(po, ph) ==
2892 				   TP_STATUS_AVAILABLE) {
2893 				/* skb was destructed already */
2894 				skb = NULL;
2895 				goto out_status;
2896 			}
2897 			/*
2898 			 * skb was dropped but not destructed yet;
2899 			 * let's treat it like congestion or err < 0
2900 			 */
2901 			err = 0;
2902 		}
2903 		packet_increment_head(&po->tx_ring);
2904 		len_sum += tp_len;
2905 	} while (1);
2906 
2907 	err = len_sum;
2908 	goto out_put;
2909 
2910 out_status:
2911 	__packet_set_status(po, ph, status);
2912 	kfree_skb(skb);
2913 out_put:
2914 	dev_put(dev);
2915 out:
2916 	mutex_unlock(&po->pg_vec_lock);
2917 	return err;
2918 }
2919 
2920 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2921 				        size_t reserve, size_t len,
2922 				        size_t linear, int noblock,
2923 				        int *err)
2924 {
2925 	struct sk_buff *skb;
2926 
2927 	/* Under a page?  Don't bother with paged skb. */
2928 	if (prepad + len < PAGE_SIZE || !linear)
2929 		linear = len;
2930 
2931 	if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2932 		linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2933 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2934 				   err, PAGE_ALLOC_COSTLY_ORDER);
2935 	if (!skb)
2936 		return NULL;
2937 
2938 	skb_reserve(skb, reserve);
2939 	skb_put(skb, linear);
2940 	skb->data_len = len - linear;
2941 	skb->len += len - linear;
2942 
2943 	return skb;
2944 }
2945 
2946 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2947 {
2948 	struct sock *sk = sock->sk;
2949 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2950 	struct sk_buff *skb;
2951 	struct net_device *dev;
2952 	__be16 proto;
2953 	unsigned char *addr = NULL;
2954 	int err, reserve = 0;
2955 	struct sockcm_cookie sockc;
2956 	struct virtio_net_hdr vnet_hdr = { 0 };
2957 	int offset = 0;
2958 	struct packet_sock *po = pkt_sk(sk);
2959 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2960 	int hlen, tlen, linear;
2961 	int extra_len = 0;
2962 
2963 	/*
2964 	 *	Get and verify the address.
2965 	 */
2966 
2967 	if (likely(saddr == NULL)) {
2968 		dev	= packet_cached_dev_get(po);
2969 		proto	= READ_ONCE(po->num);
2970 	} else {
2971 		err = -EINVAL;
2972 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2973 			goto out;
2974 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2975 			goto out;
2976 		proto	= saddr->sll_protocol;
2977 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2978 		if (sock->type == SOCK_DGRAM) {
2979 			if (dev && msg->msg_namelen < dev->addr_len +
2980 				   offsetof(struct sockaddr_ll, sll_addr))
2981 				goto out_unlock;
2982 			addr = saddr->sll_addr;
2983 		}
2984 	}
2985 
2986 	err = -ENXIO;
2987 	if (unlikely(dev == NULL))
2988 		goto out_unlock;
2989 	err = -ENETDOWN;
2990 	if (unlikely(!(dev->flags & IFF_UP)))
2991 		goto out_unlock;
2992 
2993 	sockcm_init(&sockc, sk);
2994 	if (msg->msg_controllen) {
2995 		err = sock_cmsg_send(sk, msg, &sockc);
2996 		if (unlikely(err))
2997 			goto out_unlock;
2998 	}
2999 
3000 	if (sock->type == SOCK_RAW)
3001 		reserve = dev->hard_header_len;
3002 	if (vnet_hdr_sz) {
3003 		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3004 		if (err)
3005 			goto out_unlock;
3006 	}
3007 
3008 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3009 		if (!netif_supports_nofcs(dev)) {
3010 			err = -EPROTONOSUPPORT;
3011 			goto out_unlock;
3012 		}
3013 		extra_len = 4; /* We're doing our own CRC */
3014 	}
3015 
3016 	err = -EMSGSIZE;
3017 	if (!vnet_hdr.gso_type &&
3018 	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3019 		goto out_unlock;
3020 
3021 	err = -ENOBUFS;
3022 	hlen = LL_RESERVED_SPACE(dev);
3023 	tlen = dev->needed_tailroom;
3024 	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3025 	linear = max(linear, min_t(int, len, dev->hard_header_len));
3026 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3027 			       msg->msg_flags & MSG_DONTWAIT, &err);
3028 	if (skb == NULL)
3029 		goto out_unlock;
3030 
3031 	skb_reset_network_header(skb);
3032 
3033 	err = -EINVAL;
3034 	if (sock->type == SOCK_DGRAM) {
3035 		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3036 		if (unlikely(offset < 0))
3037 			goto out_free;
3038 	} else if (reserve) {
3039 		skb_reserve(skb, -reserve);
3040 		if (len < reserve + sizeof(struct ipv6hdr) &&
3041 		    dev->min_header_len != dev->hard_header_len)
3042 			skb_reset_network_header(skb);
3043 	}
3044 
3045 	/* Returns -EFAULT on error */
3046 	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3047 	if (err)
3048 		goto out_free;
3049 
3050 	if ((sock->type == SOCK_RAW &&
3051 	     !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3052 		err = -EINVAL;
3053 		goto out_free;
3054 	}
3055 
3056 	skb_setup_tx_timestamp(skb, &sockc);
3057 
3058 	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3059 	    !packet_extra_vlan_len_allowed(dev, skb)) {
3060 		err = -EMSGSIZE;
3061 		goto out_free;
3062 	}
3063 
3064 	skb->protocol = proto;
3065 	skb->dev = dev;
3066 	skb->priority = sockc.priority;
3067 	skb->mark = sockc.mark;
3068 	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
3069 
3070 	if (unlikely(extra_len == 4))
3071 		skb->no_fcs = 1;
3072 
3073 	packet_parse_headers(skb, sock);
3074 
3075 	if (vnet_hdr_sz) {
3076 		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3077 		if (err)
3078 			goto out_free;
3079 		len += vnet_hdr_sz;
3080 		virtio_net_hdr_set_proto(skb, &vnet_hdr);
3081 	}
3082 
3083 	err = packet_xmit(po, skb);
3084 
3085 	if (unlikely(err != 0)) {
3086 		if (err > 0)
3087 			err = net_xmit_errno(err);
3088 		if (err)
3089 			goto out_unlock;
3090 	}
3091 
3092 	dev_put(dev);
3093 
3094 	return len;
3095 
3096 out_free:
3097 	kfree_skb(skb);
3098 out_unlock:
3099 	dev_put(dev);
3100 out:
3101 	return err;
3102 }
3103 
3104 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3105 {
3106 	struct sock *sk = sock->sk;
3107 	struct packet_sock *po = pkt_sk(sk);
3108 
3109 	/* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3110 	 * tpacket_snd() will redo the check safely.
3111 	 */
3112 	if (data_race(po->tx_ring.pg_vec))
3113 		return tpacket_snd(po, msg);
3114 
3115 	return packet_snd(sock, msg, len);
3116 }
3117 
3118 /*
3119  *	Close a PACKET socket. This is fairly simple. We immediately go
3120  *	to 'closed' state and remove our protocol entry in the device list.
3121  */
3122 
3123 static int packet_release(struct socket *sock)
3124 {
3125 	struct sock *sk = sock->sk;
3126 	struct packet_sock *po;
3127 	struct packet_fanout *f;
3128 	struct net *net;
3129 	union tpacket_req_u req_u;
3130 
3131 	if (!sk)
3132 		return 0;
3133 
3134 	net = sock_net(sk);
3135 	po = pkt_sk(sk);
3136 
3137 	mutex_lock(&net->packet.sklist_lock);
3138 	sk_del_node_init_rcu(sk);
3139 	mutex_unlock(&net->packet.sklist_lock);
3140 
3141 	sock_prot_inuse_add(net, sk->sk_prot, -1);
3142 
3143 	spin_lock(&po->bind_lock);
3144 	unregister_prot_hook(sk, false);
3145 	WRITE_ONCE(po->num, 0);
3146 	packet_cached_dev_reset(po);
3147 
3148 	if (po->prot_hook.dev) {
3149 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3150 		po->prot_hook.dev = NULL;
3151 	}
3152 	spin_unlock(&po->bind_lock);
3153 
3154 	packet_flush_mclist(sk);
3155 
3156 	lock_sock(sk);
3157 	if (po->rx_ring.pg_vec) {
3158 		memset(&req_u, 0, sizeof(req_u));
3159 		packet_set_ring(sk, &req_u, 1, 0);
3160 	}
3161 
3162 	if (po->tx_ring.pg_vec) {
3163 		memset(&req_u, 0, sizeof(req_u));
3164 		packet_set_ring(sk, &req_u, 1, 1);
3165 	}
3166 	release_sock(sk);
3167 
3168 	f = fanout_release(sk);
3169 
3170 	synchronize_net();
3171 
3172 	kfree(po->rollover);
3173 	if (f) {
3174 		fanout_release_data(f);
3175 		kvfree(f);
3176 	}
3177 	/*
3178 	 *	Now the socket is dead. No more input will appear.
3179 	 */
3180 	sock_orphan(sk);
3181 	sock->sk = NULL;
3182 
3183 	/* Purge queues */
3184 
3185 	skb_queue_purge(&sk->sk_receive_queue);
3186 	packet_free_pending(po);
3187 
3188 	sock_put(sk);
3189 	return 0;
3190 }
3191 
3192 /*
3193  *	Attach a packet hook.
3194  */
3195 
3196 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3197 			  __be16 proto)
3198 {
3199 	struct packet_sock *po = pkt_sk(sk);
3200 	struct net_device *dev = NULL;
3201 	bool unlisted = false;
3202 	bool need_rehook;
3203 	int ret = 0;
3204 
3205 	lock_sock(sk);
3206 	spin_lock(&po->bind_lock);
3207 	if (!proto)
3208 		proto = po->num;
3209 
3210 	rcu_read_lock();
3211 
3212 	if (po->fanout) {
3213 		ret = -EINVAL;
3214 		goto out_unlock;
3215 	}
3216 
3217 	if (name) {
3218 		dev = dev_get_by_name_rcu(sock_net(sk), name);
3219 		if (!dev) {
3220 			ret = -ENODEV;
3221 			goto out_unlock;
3222 		}
3223 	} else if (ifindex) {
3224 		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3225 		if (!dev) {
3226 			ret = -ENODEV;
3227 			goto out_unlock;
3228 		}
3229 	}
3230 
3231 	need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3232 
3233 	if (need_rehook) {
3234 		dev_hold(dev);
3235 		if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3236 			rcu_read_unlock();
3237 			/* prevents packet_notifier() from calling
3238 			 * register_prot_hook()
3239 			 */
3240 			WRITE_ONCE(po->num, 0);
3241 			__unregister_prot_hook(sk, true);
3242 			rcu_read_lock();
3243 			if (dev)
3244 				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3245 								 dev->ifindex);
3246 		}
3247 
3248 		BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3249 		WRITE_ONCE(po->num, proto);
3250 		po->prot_hook.type = proto;
3251 
3252 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3253 
3254 		if (unlikely(unlisted)) {
3255 			po->prot_hook.dev = NULL;
3256 			WRITE_ONCE(po->ifindex, -1);
3257 			packet_cached_dev_reset(po);
3258 		} else {
3259 			netdev_hold(dev, &po->prot_hook.dev_tracker,
3260 				    GFP_ATOMIC);
3261 			po->prot_hook.dev = dev;
3262 			WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3263 			packet_cached_dev_assign(po, dev);
3264 		}
3265 		dev_put(dev);
3266 	}
3267 
3268 	if (proto == 0 || !need_rehook)
3269 		goto out_unlock;
3270 
3271 	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3272 		register_prot_hook(sk);
3273 	} else {
3274 		sk->sk_err = ENETDOWN;
3275 		if (!sock_flag(sk, SOCK_DEAD))
3276 			sk_error_report(sk);
3277 	}
3278 
3279 out_unlock:
3280 	rcu_read_unlock();
3281 	spin_unlock(&po->bind_lock);
3282 	release_sock(sk);
3283 	return ret;
3284 }
3285 
3286 /*
3287  *	Bind a packet socket to a device
3288  */
3289 
3290 static int packet_bind_spkt(struct socket *sock, struct sockaddr_unsized *uaddr,
3291 			    int addr_len)
3292 {
3293 	struct sock *sk = sock->sk;
3294 	struct sockaddr *sa = (struct sockaddr *)uaddr;
3295 	char name[sizeof(sa->sa_data) + 1];
3296 
3297 	/*
3298 	 *	Check legality
3299 	 */
3300 
3301 	if (addr_len != sizeof(struct sockaddr))
3302 		return -EINVAL;
3303 	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3304 	 * zero-terminated.
3305 	 */
3306 	memcpy(name, sa->sa_data, sizeof(sa->sa_data));
3307 	name[sizeof(sa->sa_data)] = 0;
3308 
3309 	return packet_do_bind(sk, name, 0, 0);
3310 }
3311 
3312 static int packet_bind(struct socket *sock, struct sockaddr_unsized *uaddr, int addr_len)
3313 {
3314 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3315 	struct sock *sk = sock->sk;
3316 
3317 	/*
3318 	 *	Check legality
3319 	 */
3320 
3321 	if (addr_len < sizeof(struct sockaddr_ll))
3322 		return -EINVAL;
3323 	if (sll->sll_family != AF_PACKET)
3324 		return -EINVAL;
3325 
3326 	return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3327 }
3328 
3329 static struct proto packet_proto = {
3330 	.name	  = "PACKET",
3331 	.owner	  = THIS_MODULE,
3332 	.obj_size = sizeof(struct packet_sock),
3333 };
3334 
3335 /*
3336  *	Create a packet of type SOCK_PACKET.
3337  */
3338 
3339 static int packet_create(struct net *net, struct socket *sock, int protocol,
3340 			 int kern)
3341 {
3342 	struct sock *sk;
3343 	struct packet_sock *po;
3344 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3345 	int err;
3346 
3347 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3348 		return -EPERM;
3349 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3350 	    sock->type != SOCK_PACKET)
3351 		return -ESOCKTNOSUPPORT;
3352 
3353 	sock->state = SS_UNCONNECTED;
3354 
3355 	err = -ENOBUFS;
3356 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3357 	if (sk == NULL)
3358 		goto out;
3359 
3360 	sock->ops = &packet_ops;
3361 	if (sock->type == SOCK_PACKET)
3362 		sock->ops = &packet_ops_spkt;
3363 
3364 	po = pkt_sk(sk);
3365 	err = packet_alloc_pending(po);
3366 	if (err)
3367 		goto out_sk_free;
3368 
3369 	sock_init_data(sock, sk);
3370 
3371 	init_completion(&po->skb_completion);
3372 	sk->sk_family = PF_PACKET;
3373 	po->num = proto;
3374 
3375 	packet_cached_dev_reset(po);
3376 
3377 	sk->sk_destruct = packet_sock_destruct;
3378 
3379 	/*
3380 	 *	Attach a protocol block
3381 	 */
3382 
3383 	spin_lock_init(&po->bind_lock);
3384 	mutex_init(&po->pg_vec_lock);
3385 	po->rollover = NULL;
3386 	po->prot_hook.func = packet_rcv;
3387 
3388 	if (sock->type == SOCK_PACKET)
3389 		po->prot_hook.func = packet_rcv_spkt;
3390 
3391 	po->prot_hook.af_packet_priv = sk;
3392 	po->prot_hook.af_packet_net = sock_net(sk);
3393 
3394 	if (proto) {
3395 		po->prot_hook.type = proto;
3396 		__register_prot_hook(sk);
3397 	}
3398 
3399 	mutex_lock(&net->packet.sklist_lock);
3400 	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3401 	mutex_unlock(&net->packet.sklist_lock);
3402 
3403 	sock_prot_inuse_add(net, &packet_proto, 1);
3404 
3405 	return 0;
3406 out_sk_free:
3407 	sk_free(sk);
3408 out:
3409 	return err;
3410 }
3411 
3412 /*
3413  *	Pull a packet from our receive queue and hand it to the user.
3414  *	If necessary we block.
3415  */
3416 
3417 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3418 			  int flags)
3419 {
3420 	struct sock *sk = sock->sk;
3421 	struct sk_buff *skb;
3422 	int copied, err;
3423 	int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3424 	unsigned int origlen = 0;
3425 
3426 	err = -EINVAL;
3427 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3428 		goto out;
3429 
3430 #if 0
3431 	/* What error should we return now? EUNATTACH? */
3432 	if (pkt_sk(sk)->ifindex < 0)
3433 		return -ENODEV;
3434 #endif
3435 
3436 	if (flags & MSG_ERRQUEUE) {
3437 		err = sock_recv_errqueue(sk, msg, len,
3438 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3439 		goto out;
3440 	}
3441 
3442 	/*
3443 	 *	Call the generic datagram receiver. This handles all sorts
3444 	 *	of horrible races and re-entrancy so we can forget about it
3445 	 *	in the protocol layers.
3446 	 *
3447 	 *	Now it will return ENETDOWN, if device have just gone down,
3448 	 *	but then it will block.
3449 	 */
3450 
3451 	skb = skb_recv_datagram(sk, flags, &err);
3452 
3453 	/*
3454 	 *	An error occurred so return it. Because skb_recv_datagram()
3455 	 *	handles the blocking we don't see and worry about blocking
3456 	 *	retries.
3457 	 */
3458 
3459 	if (skb == NULL)
3460 		goto out;
3461 
3462 	packet_rcv_try_clear_pressure(pkt_sk(sk));
3463 
3464 	if (vnet_hdr_len) {
3465 		err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3466 		if (err)
3467 			goto out_free;
3468 	}
3469 
3470 	/* You lose any data beyond the buffer you gave. If it worries
3471 	 * a user program they can ask the device for its MTU
3472 	 * anyway.
3473 	 */
3474 	copied = skb->len;
3475 	if (copied > len) {
3476 		copied = len;
3477 		msg->msg_flags |= MSG_TRUNC;
3478 	}
3479 
3480 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3481 	if (err)
3482 		goto out_free;
3483 
3484 	if (sock->type != SOCK_PACKET) {
3485 		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3486 
3487 		/* Original length was stored in sockaddr_ll fields */
3488 		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3489 		sll->sll_family = AF_PACKET;
3490 		sll->sll_protocol = (sock->type == SOCK_DGRAM) ?
3491 			vlan_get_protocol_dgram(skb) : skb->protocol;
3492 	}
3493 
3494 	sock_recv_cmsgs(msg, sk, skb);
3495 
3496 	if (msg->msg_name) {
3497 		const size_t max_len = min(sizeof(skb->cb),
3498 					   sizeof(struct sockaddr_storage));
3499 		int copy_len;
3500 
3501 		/* If the address length field is there to be filled
3502 		 * in, we fill it in now.
3503 		 */
3504 		if (sock->type == SOCK_PACKET) {
3505 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3506 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3507 			copy_len = msg->msg_namelen;
3508 		} else {
3509 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3510 
3511 			msg->msg_namelen = sll->sll_halen +
3512 				offsetof(struct sockaddr_ll, sll_addr);
3513 			copy_len = msg->msg_namelen;
3514 			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3515 				memset(msg->msg_name +
3516 				       offsetof(struct sockaddr_ll, sll_addr),
3517 				       0, sizeof(sll->sll_addr));
3518 				msg->msg_namelen = sizeof(struct sockaddr_ll);
3519 			}
3520 		}
3521 		if (WARN_ON_ONCE(copy_len > max_len)) {
3522 			copy_len = max_len;
3523 			msg->msg_namelen = copy_len;
3524 		}
3525 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3526 	}
3527 
3528 	if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3529 		struct tpacket_auxdata aux;
3530 
3531 		aux.tp_status = TP_STATUS_USER;
3532 		if (skb->ip_summed == CHECKSUM_PARTIAL)
3533 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3534 		else if (skb->pkt_type != PACKET_OUTGOING &&
3535 			 skb_csum_unnecessary(skb))
3536 			aux.tp_status |= TP_STATUS_CSUM_VALID;
3537 		if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3538 			aux.tp_status |= TP_STATUS_GSO_TCP;
3539 
3540 		aux.tp_len = origlen;
3541 		aux.tp_snaplen = skb->len;
3542 		aux.tp_mac = 0;
3543 		aux.tp_net = skb_network_offset(skb);
3544 		if (skb_vlan_tag_present(skb)) {
3545 			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3546 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3547 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3548 		} else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
3549 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3550 			struct net_device *dev;
3551 
3552 			rcu_read_lock();
3553 			dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex);
3554 			if (dev) {
3555 				aux.tp_vlan_tci = vlan_get_tci(skb, dev);
3556 				aux.tp_vlan_tpid = ntohs(skb->protocol);
3557 				aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3558 			} else {
3559 				aux.tp_vlan_tci = 0;
3560 				aux.tp_vlan_tpid = 0;
3561 			}
3562 			rcu_read_unlock();
3563 		} else {
3564 			aux.tp_vlan_tci = 0;
3565 			aux.tp_vlan_tpid = 0;
3566 		}
3567 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3568 	}
3569 
3570 	/*
3571 	 *	Free or return the buffer as appropriate. Again this
3572 	 *	hides all the races and re-entrancy issues from us.
3573 	 */
3574 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3575 
3576 out_free:
3577 	skb_free_datagram(sk, skb);
3578 out:
3579 	return err;
3580 }
3581 
3582 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3583 			       int peer)
3584 {
3585 	struct net_device *dev;
3586 	struct sock *sk	= sock->sk;
3587 
3588 	if (peer)
3589 		return -EOPNOTSUPP;
3590 
3591 	uaddr->sa_family = AF_PACKET;
3592 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3593 	rcu_read_lock();
3594 	dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3595 	if (dev)
3596 		strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3597 	rcu_read_unlock();
3598 
3599 	return sizeof(*uaddr);
3600 }
3601 
3602 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3603 			  int peer)
3604 {
3605 	struct net_device *dev;
3606 	struct sock *sk = sock->sk;
3607 	struct packet_sock *po = pkt_sk(sk);
3608 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3609 	int ifindex;
3610 
3611 	if (peer)
3612 		return -EOPNOTSUPP;
3613 
3614 	ifindex = READ_ONCE(po->ifindex);
3615 	sll->sll_family = AF_PACKET;
3616 	sll->sll_ifindex = ifindex;
3617 	sll->sll_protocol = READ_ONCE(po->num);
3618 	sll->sll_pkttype = 0;
3619 	rcu_read_lock();
3620 	dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3621 	if (dev) {
3622 		sll->sll_hatype = dev->type;
3623 		sll->sll_halen = dev->addr_len;
3624 
3625 		/* Let __fortify_memcpy_chk() know the actual buffer size. */
3626 		memcpy(((struct sockaddr_storage *)sll)->__data +
3627 		       offsetof(struct sockaddr_ll, sll_addr) -
3628 		       offsetofend(struct sockaddr_ll, sll_family),
3629 		       dev->dev_addr, dev->addr_len);
3630 	} else {
3631 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3632 		sll->sll_halen = 0;
3633 	}
3634 	rcu_read_unlock();
3635 
3636 	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3637 }
3638 
3639 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3640 			 int what)
3641 {
3642 	switch (i->type) {
3643 	case PACKET_MR_MULTICAST:
3644 		if (i->alen != dev->addr_len)
3645 			return -EINVAL;
3646 		if (what > 0)
3647 			return dev_mc_add(dev, i->addr);
3648 		else
3649 			return dev_mc_del(dev, i->addr);
3650 		break;
3651 	case PACKET_MR_PROMISC:
3652 		return dev_set_promiscuity(dev, what);
3653 	case PACKET_MR_ALLMULTI:
3654 		return dev_set_allmulti(dev, what);
3655 	case PACKET_MR_UNICAST:
3656 		if (i->alen != dev->addr_len)
3657 			return -EINVAL;
3658 		if (what > 0)
3659 			return dev_uc_add(dev, i->addr);
3660 		else
3661 			return dev_uc_del(dev, i->addr);
3662 		break;
3663 	default:
3664 		break;
3665 	}
3666 	return 0;
3667 }
3668 
3669 static void packet_dev_mclist_delete(struct net_device *dev,
3670 				     struct packet_mclist **mlp,
3671 				     struct list_head *list)
3672 {
3673 	struct packet_mclist *ml;
3674 
3675 	while ((ml = *mlp) != NULL) {
3676 		if (ml->ifindex == dev->ifindex) {
3677 			list_add(&ml->remove_list, list);
3678 			*mlp = ml->next;
3679 		} else
3680 			mlp = &ml->next;
3681 	}
3682 }
3683 
3684 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3685 {
3686 	struct packet_sock *po = pkt_sk(sk);
3687 	struct packet_mclist *ml, *i;
3688 	struct net_device *dev;
3689 	int err;
3690 
3691 	rtnl_lock();
3692 
3693 	err = -ENODEV;
3694 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3695 	if (!dev)
3696 		goto done;
3697 
3698 	err = -EINVAL;
3699 	if (mreq->mr_alen > dev->addr_len)
3700 		goto done;
3701 
3702 	err = -ENOBUFS;
3703 	i = kmalloc_obj(*i);
3704 	if (i == NULL)
3705 		goto done;
3706 
3707 	err = 0;
3708 	for (ml = po->mclist; ml; ml = ml->next) {
3709 		if (ml->ifindex == mreq->mr_ifindex &&
3710 		    ml->type == mreq->mr_type &&
3711 		    ml->alen == mreq->mr_alen &&
3712 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3713 			ml->count++;
3714 			/* Free the new element ... */
3715 			kfree(i);
3716 			goto done;
3717 		}
3718 	}
3719 
3720 	i->type = mreq->mr_type;
3721 	i->ifindex = mreq->mr_ifindex;
3722 	i->alen = mreq->mr_alen;
3723 	memcpy(i->addr, mreq->mr_address, i->alen);
3724 	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3725 	i->count = 1;
3726 	INIT_LIST_HEAD(&i->remove_list);
3727 	i->next = po->mclist;
3728 	po->mclist = i;
3729 	err = packet_dev_mc(dev, i, 1);
3730 	if (err) {
3731 		po->mclist = i->next;
3732 		kfree(i);
3733 	}
3734 
3735 done:
3736 	rtnl_unlock();
3737 	return err;
3738 }
3739 
3740 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3741 {
3742 	struct packet_mclist *ml, **mlp;
3743 
3744 	rtnl_lock();
3745 
3746 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3747 		if (ml->ifindex == mreq->mr_ifindex &&
3748 		    ml->type == mreq->mr_type &&
3749 		    ml->alen == mreq->mr_alen &&
3750 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3751 			if (--ml->count == 0) {
3752 				struct net_device *dev;
3753 				*mlp = ml->next;
3754 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3755 				if (dev)
3756 					packet_dev_mc(dev, ml, -1);
3757 				kfree(ml);
3758 			}
3759 			break;
3760 		}
3761 	}
3762 	rtnl_unlock();
3763 	return 0;
3764 }
3765 
3766 static void packet_flush_mclist(struct sock *sk)
3767 {
3768 	struct packet_sock *po = pkt_sk(sk);
3769 	struct packet_mclist *ml;
3770 
3771 	if (!po->mclist)
3772 		return;
3773 
3774 	rtnl_lock();
3775 	while ((ml = po->mclist) != NULL) {
3776 		struct net_device *dev;
3777 
3778 		po->mclist = ml->next;
3779 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3780 		if (dev != NULL)
3781 			packet_dev_mc(dev, ml, -1);
3782 		kfree(ml);
3783 	}
3784 	rtnl_unlock();
3785 }
3786 
3787 static int
3788 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3789 		  unsigned int optlen)
3790 {
3791 	struct sock *sk = sock->sk;
3792 	struct packet_sock *po = pkt_sk(sk);
3793 	int ret;
3794 
3795 	if (level != SOL_PACKET)
3796 		return -ENOPROTOOPT;
3797 
3798 	switch (optname) {
3799 	case PACKET_ADD_MEMBERSHIP:
3800 	case PACKET_DROP_MEMBERSHIP:
3801 	{
3802 		struct packet_mreq_max mreq;
3803 		int len = optlen;
3804 		memset(&mreq, 0, sizeof(mreq));
3805 		if (len < sizeof(struct packet_mreq))
3806 			return -EINVAL;
3807 		if (len > sizeof(mreq))
3808 			len = sizeof(mreq);
3809 		if (copy_from_sockptr(&mreq, optval, len))
3810 			return -EFAULT;
3811 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3812 			return -EINVAL;
3813 		if (optname == PACKET_ADD_MEMBERSHIP)
3814 			ret = packet_mc_add(sk, &mreq);
3815 		else
3816 			ret = packet_mc_drop(sk, &mreq);
3817 		return ret;
3818 	}
3819 
3820 	case PACKET_RX_RING:
3821 	case PACKET_TX_RING:
3822 	{
3823 		union tpacket_req_u req_u;
3824 
3825 		ret = -EINVAL;
3826 		lock_sock(sk);
3827 		switch (po->tp_version) {
3828 		case TPACKET_V1:
3829 		case TPACKET_V2:
3830 			if (optlen < sizeof(req_u.req))
3831 				break;
3832 			ret = copy_from_sockptr(&req_u.req, optval,
3833 						sizeof(req_u.req)) ?
3834 						-EINVAL : 0;
3835 			break;
3836 		case TPACKET_V3:
3837 		default:
3838 			if (optlen < sizeof(req_u.req3))
3839 				break;
3840 			ret = copy_from_sockptr(&req_u.req3, optval,
3841 						sizeof(req_u.req3)) ?
3842 						-EINVAL : 0;
3843 			break;
3844 		}
3845 		if (!ret)
3846 			ret = packet_set_ring(sk, &req_u, 0,
3847 					      optname == PACKET_TX_RING);
3848 		release_sock(sk);
3849 		return ret;
3850 	}
3851 	case PACKET_COPY_THRESH:
3852 	{
3853 		int val;
3854 
3855 		if (optlen != sizeof(val))
3856 			return -EINVAL;
3857 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3858 			return -EFAULT;
3859 
3860 		WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3861 		return 0;
3862 	}
3863 	case PACKET_VERSION:
3864 	{
3865 		int val;
3866 
3867 		if (optlen != sizeof(val))
3868 			return -EINVAL;
3869 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3870 			return -EFAULT;
3871 		switch (val) {
3872 		case TPACKET_V1:
3873 		case TPACKET_V2:
3874 		case TPACKET_V3:
3875 			break;
3876 		default:
3877 			return -EINVAL;
3878 		}
3879 		lock_sock(sk);
3880 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3881 			ret = -EBUSY;
3882 		} else {
3883 			po->tp_version = val;
3884 			ret = 0;
3885 		}
3886 		release_sock(sk);
3887 		return ret;
3888 	}
3889 	case PACKET_RESERVE:
3890 	{
3891 		unsigned int val;
3892 
3893 		if (optlen != sizeof(val))
3894 			return -EINVAL;
3895 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3896 			return -EFAULT;
3897 		if (val > INT_MAX)
3898 			return -EINVAL;
3899 		lock_sock(sk);
3900 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3901 			ret = -EBUSY;
3902 		} else {
3903 			po->tp_reserve = val;
3904 			ret = 0;
3905 		}
3906 		release_sock(sk);
3907 		return ret;
3908 	}
3909 	case PACKET_LOSS:
3910 	{
3911 		unsigned int val;
3912 
3913 		if (optlen != sizeof(val))
3914 			return -EINVAL;
3915 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3916 			return -EFAULT;
3917 
3918 		lock_sock(sk);
3919 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3920 			ret = -EBUSY;
3921 		} else {
3922 			packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3923 			ret = 0;
3924 		}
3925 		release_sock(sk);
3926 		return ret;
3927 	}
3928 	case PACKET_AUXDATA:
3929 	{
3930 		int val;
3931 
3932 		if (optlen < sizeof(val))
3933 			return -EINVAL;
3934 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3935 			return -EFAULT;
3936 
3937 		packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3938 		return 0;
3939 	}
3940 	case PACKET_ORIGDEV:
3941 	{
3942 		int val;
3943 
3944 		if (optlen < sizeof(val))
3945 			return -EINVAL;
3946 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3947 			return -EFAULT;
3948 
3949 		packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3950 		return 0;
3951 	}
3952 	case PACKET_VNET_HDR:
3953 	case PACKET_VNET_HDR_SZ:
3954 	{
3955 		int val, hdr_len;
3956 
3957 		if (sock->type != SOCK_RAW)
3958 			return -EINVAL;
3959 		if (optlen < sizeof(val))
3960 			return -EINVAL;
3961 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3962 			return -EFAULT;
3963 
3964 		if (optname == PACKET_VNET_HDR_SZ) {
3965 			if (val && val != sizeof(struct virtio_net_hdr) &&
3966 			    val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
3967 				return -EINVAL;
3968 			hdr_len = val;
3969 		} else {
3970 			hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
3971 		}
3972 		lock_sock(sk);
3973 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3974 			ret = -EBUSY;
3975 		} else {
3976 			WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
3977 			ret = 0;
3978 		}
3979 		release_sock(sk);
3980 		return ret;
3981 	}
3982 	case PACKET_TIMESTAMP:
3983 	{
3984 		int val;
3985 
3986 		if (optlen != sizeof(val))
3987 			return -EINVAL;
3988 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3989 			return -EFAULT;
3990 
3991 		WRITE_ONCE(po->tp_tstamp, val);
3992 		return 0;
3993 	}
3994 	case PACKET_FANOUT:
3995 	{
3996 		struct fanout_args args = { 0 };
3997 
3998 		if (optlen != sizeof(int) && optlen != sizeof(args))
3999 			return -EINVAL;
4000 		if (copy_from_sockptr(&args, optval, optlen))
4001 			return -EFAULT;
4002 
4003 		return fanout_add(sk, &args);
4004 	}
4005 	case PACKET_FANOUT_DATA:
4006 	{
4007 		/* Paired with the WRITE_ONCE() in fanout_add() */
4008 		if (!READ_ONCE(po->fanout))
4009 			return -EINVAL;
4010 
4011 		return fanout_set_data(po, optval, optlen);
4012 	}
4013 	case PACKET_IGNORE_OUTGOING:
4014 	{
4015 		int val;
4016 
4017 		if (optlen != sizeof(val))
4018 			return -EINVAL;
4019 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4020 			return -EFAULT;
4021 		if (val < 0 || val > 1)
4022 			return -EINVAL;
4023 
4024 		WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4025 		return 0;
4026 	}
4027 	case PACKET_TX_HAS_OFF:
4028 	{
4029 		unsigned int val;
4030 
4031 		if (optlen != sizeof(val))
4032 			return -EINVAL;
4033 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4034 			return -EFAULT;
4035 
4036 		lock_sock(sk);
4037 		if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4038 			packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4039 
4040 		release_sock(sk);
4041 		return 0;
4042 	}
4043 	case PACKET_QDISC_BYPASS:
4044 	{
4045 		int val;
4046 
4047 		if (optlen != sizeof(val))
4048 			return -EINVAL;
4049 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4050 			return -EFAULT;
4051 
4052 		packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4053 		return 0;
4054 	}
4055 	default:
4056 		return -ENOPROTOOPT;
4057 	}
4058 }
4059 
4060 static int packet_getsockopt(struct socket *sock, int level, int optname,
4061 			     sockopt_t *opt)
4062 {
4063 	int len;
4064 	int val, lv = sizeof(val);
4065 	struct sock *sk = sock->sk;
4066 	struct packet_sock *po = pkt_sk(sk);
4067 	void *data = &val;
4068 	union tpacket_stats_u st;
4069 	struct tpacket_rollover_stats rstats;
4070 	int drops;
4071 
4072 	if (level != SOL_PACKET)
4073 		return -ENOPROTOOPT;
4074 
4075 	len = opt->optlen;
4076 
4077 	if (len < 0)
4078 		return -EINVAL;
4079 
4080 	switch (optname) {
4081 	case PACKET_STATISTICS:
4082 		spin_lock_bh(&sk->sk_receive_queue.lock);
4083 		memcpy(&st, &po->stats, sizeof(st));
4084 		memset(&po->stats, 0, sizeof(po->stats));
4085 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4086 		drops = atomic_xchg(&po->tp_drops, 0);
4087 
4088 		if (po->tp_version == TPACKET_V3) {
4089 			lv = sizeof(struct tpacket_stats_v3);
4090 			st.stats3.tp_drops = drops;
4091 			st.stats3.tp_packets += drops;
4092 			data = &st.stats3;
4093 		} else {
4094 			lv = sizeof(struct tpacket_stats);
4095 			st.stats1.tp_drops = drops;
4096 			st.stats1.tp_packets += drops;
4097 			data = &st.stats1;
4098 		}
4099 
4100 		break;
4101 	case PACKET_AUXDATA:
4102 		val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4103 		break;
4104 	case PACKET_ORIGDEV:
4105 		val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4106 		break;
4107 	case PACKET_VNET_HDR:
4108 		val = !!READ_ONCE(po->vnet_hdr_sz);
4109 		break;
4110 	case PACKET_VNET_HDR_SZ:
4111 		val = READ_ONCE(po->vnet_hdr_sz);
4112 		break;
4113 	case PACKET_COPY_THRESH:
4114 		val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4115 		break;
4116 	case PACKET_VERSION:
4117 		val = po->tp_version;
4118 		break;
4119 	case PACKET_HDRLEN:
4120 		if (len > sizeof(int))
4121 			len = sizeof(int);
4122 		if (len < sizeof(int))
4123 			return -EINVAL;
4124 		if (copy_from_iter(&val, len, &opt->iter_in) != len)
4125 			return -EFAULT;
4126 		switch (val) {
4127 		case TPACKET_V1:
4128 			val = sizeof(struct tpacket_hdr);
4129 			break;
4130 		case TPACKET_V2:
4131 			val = sizeof(struct tpacket2_hdr);
4132 			break;
4133 		case TPACKET_V3:
4134 			val = sizeof(struct tpacket3_hdr);
4135 			break;
4136 		default:
4137 			return -EINVAL;
4138 		}
4139 		break;
4140 	case PACKET_RESERVE:
4141 		val = po->tp_reserve;
4142 		break;
4143 	case PACKET_LOSS:
4144 		val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4145 		break;
4146 	case PACKET_TIMESTAMP:
4147 		val = READ_ONCE(po->tp_tstamp);
4148 		break;
4149 	case PACKET_FANOUT:
4150 		val = (po->fanout ?
4151 		       ((u32)po->fanout->id |
4152 			((u32)po->fanout->type << 16) |
4153 			((u32)po->fanout->flags << 24)) :
4154 		       0);
4155 		break;
4156 	case PACKET_IGNORE_OUTGOING:
4157 		val = READ_ONCE(po->prot_hook.ignore_outgoing);
4158 		break;
4159 	case PACKET_ROLLOVER_STATS:
4160 		if (!po->rollover)
4161 			return -EINVAL;
4162 		rstats.tp_all = atomic_long_read(&po->rollover->num);
4163 		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4164 		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4165 		data = &rstats;
4166 		lv = sizeof(rstats);
4167 		break;
4168 	case PACKET_TX_HAS_OFF:
4169 		val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4170 		break;
4171 	case PACKET_QDISC_BYPASS:
4172 		val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4173 		break;
4174 	default:
4175 		return -ENOPROTOOPT;
4176 	}
4177 
4178 	if (len > lv)
4179 		len = lv;
4180 	opt->optlen = len;
4181 	if (copy_to_iter(data, len, &opt->iter_out) != len)
4182 		return -EFAULT;
4183 	return 0;
4184 }
4185 
4186 static int packet_notifier(struct notifier_block *this,
4187 			   unsigned long msg, void *ptr)
4188 {
4189 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4190 	struct net *net = dev_net(dev);
4191 	struct packet_mclist *ml, *tmp;
4192 	LIST_HEAD(mclist);
4193 	struct sock *sk;
4194 
4195 	rcu_read_lock();
4196 	sk_for_each_rcu(sk, &net->packet.sklist) {
4197 		struct packet_sock *po = pkt_sk(sk);
4198 
4199 		switch (msg) {
4200 		case NETDEV_UNREGISTER:
4201 			if (po->mclist)
4202 				packet_dev_mclist_delete(dev, &po->mclist,
4203 							 &mclist);
4204 			fallthrough;
4205 
4206 		case NETDEV_DOWN:
4207 			if (dev->ifindex == po->ifindex) {
4208 				spin_lock(&po->bind_lock);
4209 				if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4210 					__unregister_prot_hook(sk, false);
4211 					sk->sk_err = ENETDOWN;
4212 					if (!sock_flag(sk, SOCK_DEAD))
4213 						sk_error_report(sk);
4214 				}
4215 				if (msg == NETDEV_UNREGISTER) {
4216 					packet_cached_dev_reset(po);
4217 					WRITE_ONCE(po->ifindex, -1);
4218 					netdev_put(po->prot_hook.dev,
4219 						   &po->prot_hook.dev_tracker);
4220 					po->prot_hook.dev = NULL;
4221 				}
4222 				spin_unlock(&po->bind_lock);
4223 			}
4224 			break;
4225 		case NETDEV_UP:
4226 			if (dev->ifindex == po->ifindex) {
4227 				spin_lock(&po->bind_lock);
4228 				if (po->num)
4229 					register_prot_hook(sk);
4230 				spin_unlock(&po->bind_lock);
4231 			}
4232 			break;
4233 		}
4234 	}
4235 	rcu_read_unlock();
4236 
4237 	/* packet_dev_mc might grab instance locks so can't run under rcu */
4238 	list_for_each_entry_safe(ml, tmp, &mclist, remove_list) {
4239 		packet_dev_mc(dev, ml, -1);
4240 		kfree(ml);
4241 	}
4242 
4243 	return NOTIFY_DONE;
4244 }
4245 
4246 
4247 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4248 			unsigned long arg)
4249 {
4250 	struct sock *sk = sock->sk;
4251 
4252 	switch (cmd) {
4253 	case SIOCOUTQ:
4254 	{
4255 		int amount = sk_wmem_alloc_get(sk);
4256 
4257 		return put_user(amount, (int __user *)arg);
4258 	}
4259 	case SIOCINQ:
4260 	{
4261 		struct sk_buff *skb;
4262 		int amount = 0;
4263 
4264 		spin_lock_bh(&sk->sk_receive_queue.lock);
4265 		skb = skb_peek(&sk->sk_receive_queue);
4266 		if (skb)
4267 			amount = skb->len;
4268 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4269 		return put_user(amount, (int __user *)arg);
4270 	}
4271 #ifdef CONFIG_INET
4272 	case SIOCADDRT:
4273 	case SIOCDELRT:
4274 	case SIOCDARP:
4275 	case SIOCGARP:
4276 	case SIOCSARP:
4277 	case SIOCGIFADDR:
4278 	case SIOCSIFADDR:
4279 	case SIOCGIFBRDADDR:
4280 	case SIOCSIFBRDADDR:
4281 	case SIOCGIFNETMASK:
4282 	case SIOCSIFNETMASK:
4283 	case SIOCGIFDSTADDR:
4284 	case SIOCSIFDSTADDR:
4285 	case SIOCSIFFLAGS:
4286 		return inet_dgram_ops.ioctl(sock, cmd, arg);
4287 #endif
4288 
4289 	default:
4290 		return -ENOIOCTLCMD;
4291 	}
4292 	return 0;
4293 }
4294 
4295 static __poll_t packet_poll(struct file *file, struct socket *sock,
4296 				poll_table *wait)
4297 {
4298 	struct sock *sk = sock->sk;
4299 	struct packet_sock *po = pkt_sk(sk);
4300 	__poll_t mask = datagram_poll(file, sock, wait);
4301 
4302 	spin_lock_bh(&sk->sk_receive_queue.lock);
4303 	if (po->rx_ring.pg_vec) {
4304 		if (!packet_previous_rx_frame(po, &po->rx_ring,
4305 			TP_STATUS_KERNEL))
4306 			mask |= EPOLLIN | EPOLLRDNORM;
4307 	}
4308 	packet_rcv_try_clear_pressure(po);
4309 	spin_unlock_bh(&sk->sk_receive_queue.lock);
4310 	spin_lock_bh(&sk->sk_write_queue.lock);
4311 	if (po->tx_ring.pg_vec) {
4312 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4313 			mask |= EPOLLOUT | EPOLLWRNORM;
4314 	}
4315 	spin_unlock_bh(&sk->sk_write_queue.lock);
4316 	return mask;
4317 }
4318 
4319 
4320 /* Dirty? Well, I still did not learn better way to account
4321  * for user mmaps.
4322  */
4323 
4324 static void packet_mm_open(struct vm_area_struct *vma)
4325 {
4326 	struct file *file = vma->vm_file;
4327 	struct socket *sock = file->private_data;
4328 	struct sock *sk = sock->sk;
4329 
4330 	if (sk)
4331 		atomic_long_inc(&pkt_sk(sk)->mapped);
4332 }
4333 
4334 static void packet_mm_close(struct vm_area_struct *vma)
4335 {
4336 	struct file *file = vma->vm_file;
4337 	struct socket *sock = file->private_data;
4338 	struct sock *sk = sock->sk;
4339 
4340 	if (sk)
4341 		atomic_long_dec(&pkt_sk(sk)->mapped);
4342 }
4343 
4344 static const struct vm_operations_struct packet_mmap_ops = {
4345 	.open	=	packet_mm_open,
4346 	.close	=	packet_mm_close,
4347 };
4348 
4349 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4350 			unsigned int len)
4351 {
4352 	int i;
4353 
4354 	for (i = 0; i < len; i++) {
4355 		if (likely(pg_vec[i].buffer)) {
4356 			if (is_vmalloc_addr(pg_vec[i].buffer))
4357 				vfree(pg_vec[i].buffer);
4358 			else
4359 				free_pages((unsigned long)pg_vec[i].buffer,
4360 					   order);
4361 			pg_vec[i].buffer = NULL;
4362 		}
4363 	}
4364 	kfree(pg_vec);
4365 }
4366 
4367 static char *alloc_one_pg_vec_page(unsigned long order)
4368 {
4369 	char *buffer;
4370 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4371 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4372 
4373 	buffer = (char *) __get_free_pages(gfp_flags, order);
4374 	if (buffer)
4375 		return buffer;
4376 
4377 	/* __get_free_pages failed, fall back to vmalloc */
4378 	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4379 	if (buffer)
4380 		return buffer;
4381 
4382 	/* vmalloc failed, lets dig into swap here */
4383 	gfp_flags &= ~__GFP_NORETRY;
4384 	buffer = (char *) __get_free_pages(gfp_flags, order);
4385 	if (buffer)
4386 		return buffer;
4387 
4388 	/* complete and utter failure */
4389 	return NULL;
4390 }
4391 
4392 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4393 {
4394 	unsigned int block_nr = req->tp_block_nr;
4395 	struct pgv *pg_vec;
4396 	int i;
4397 
4398 	pg_vec = kzalloc_objs(struct pgv, block_nr, GFP_KERNEL | __GFP_NOWARN);
4399 	if (unlikely(!pg_vec))
4400 		goto out;
4401 
4402 	for (i = 0; i < block_nr; i++) {
4403 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4404 		if (unlikely(!pg_vec[i].buffer))
4405 			goto out_free_pgvec;
4406 	}
4407 
4408 out:
4409 	return pg_vec;
4410 
4411 out_free_pgvec:
4412 	free_pg_vec(pg_vec, order, block_nr);
4413 	pg_vec = NULL;
4414 	goto out;
4415 }
4416 
4417 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4418 		int closing, int tx_ring)
4419 {
4420 	struct pgv *pg_vec = NULL;
4421 	struct packet_sock *po = pkt_sk(sk);
4422 	unsigned long *rx_owner_map = NULL;
4423 	int was_running, order = 0;
4424 	struct packet_ring_buffer *rb;
4425 	struct sk_buff_head *rb_queue;
4426 	__be16 num;
4427 	int err;
4428 	/* Added to avoid minimal code churn */
4429 	struct tpacket_req *req = &req_u->req;
4430 
4431 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4432 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4433 
4434 	err = -EBUSY;
4435 	if (!closing) {
4436 		if (atomic_long_read(&po->mapped))
4437 			goto out;
4438 		if (packet_read_pending(rb))
4439 			goto out;
4440 	}
4441 
4442 	if (req->tp_block_nr) {
4443 		unsigned int min_frame_size;
4444 
4445 		/* Sanity tests and some calculations */
4446 		err = -EBUSY;
4447 		if (unlikely(rb->pg_vec))
4448 			goto out;
4449 
4450 		switch (po->tp_version) {
4451 		case TPACKET_V1:
4452 			po->tp_hdrlen = TPACKET_HDRLEN;
4453 			break;
4454 		case TPACKET_V2:
4455 			po->tp_hdrlen = TPACKET2_HDRLEN;
4456 			break;
4457 		case TPACKET_V3:
4458 			po->tp_hdrlen = TPACKET3_HDRLEN;
4459 			break;
4460 		}
4461 
4462 		err = -EINVAL;
4463 		if (unlikely((int)req->tp_block_size <= 0))
4464 			goto out;
4465 		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4466 			goto out;
4467 		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4468 		if (po->tp_version >= TPACKET_V3 &&
4469 		    req->tp_block_size <
4470 		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4471 			goto out;
4472 		if (unlikely(req->tp_frame_size < min_frame_size))
4473 			goto out;
4474 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4475 			goto out;
4476 
4477 		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4478 		if (unlikely(rb->frames_per_block == 0))
4479 			goto out;
4480 		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4481 			goto out;
4482 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4483 					req->tp_frame_nr))
4484 			goto out;
4485 
4486 		err = -ENOMEM;
4487 		order = get_order(req->tp_block_size);
4488 		pg_vec = alloc_pg_vec(req, order);
4489 		if (unlikely(!pg_vec))
4490 			goto out;
4491 		switch (po->tp_version) {
4492 		case TPACKET_V3:
4493 			/* Block transmit is not supported yet */
4494 			if (!tx_ring) {
4495 				init_prb_bdqc(po, rb, pg_vec, req_u);
4496 			} else {
4497 				struct tpacket_req3 *req3 = &req_u->req3;
4498 
4499 				if (req3->tp_retire_blk_tov ||
4500 				    req3->tp_sizeof_priv ||
4501 				    req3->tp_feature_req_word) {
4502 					err = -EINVAL;
4503 					goto out_free_pg_vec;
4504 				}
4505 			}
4506 			break;
4507 		default:
4508 			if (!tx_ring) {
4509 				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4510 					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4511 				if (!rx_owner_map)
4512 					goto out_free_pg_vec;
4513 			}
4514 			break;
4515 		}
4516 	}
4517 	/* Done */
4518 	else {
4519 		err = -EINVAL;
4520 		if (unlikely(req->tp_frame_nr))
4521 			goto out;
4522 	}
4523 
4524 
4525 	/* Detach socket from network */
4526 	spin_lock(&po->bind_lock);
4527 	was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4528 	num = po->num;
4529 	WRITE_ONCE(po->num, 0);
4530 	if (was_running)
4531 		__unregister_prot_hook(sk, false);
4532 
4533 	spin_unlock(&po->bind_lock);
4534 
4535 	synchronize_net();
4536 
4537 	err = -EBUSY;
4538 	mutex_lock(&po->pg_vec_lock);
4539 	if (closing || atomic_long_read(&po->mapped) == 0) {
4540 		err = 0;
4541 		spin_lock_bh(&rb_queue->lock);
4542 		swap(rb->pg_vec, pg_vec);
4543 		if (po->tp_version <= TPACKET_V2)
4544 			swap(rb->rx_owner_map, rx_owner_map);
4545 		rb->frame_max = (req->tp_frame_nr - 1);
4546 		rb->head = 0;
4547 		rb->frame_size = req->tp_frame_size;
4548 		spin_unlock_bh(&rb_queue->lock);
4549 
4550 		swap(rb->pg_vec_order, order);
4551 		swap(rb->pg_vec_len, req->tp_block_nr);
4552 
4553 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4554 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4555 						tpacket_rcv : packet_rcv;
4556 		skb_queue_purge(rb_queue);
4557 		if (atomic_long_read(&po->mapped))
4558 			pr_err("packet_mmap: vma is busy: %ld\n",
4559 			       atomic_long_read(&po->mapped));
4560 	}
4561 	mutex_unlock(&po->pg_vec_lock);
4562 
4563 	spin_lock(&po->bind_lock);
4564 	WRITE_ONCE(po->num, num);
4565 	/*
4566 	 * NETDEV_UNREGISTER may have invalidated the binding while bind_lock
4567 	 * was dropped above.  Do not re-add a fanout hook to a dead device.
4568 	 */
4569 	if (was_running && READ_ONCE(po->ifindex) != -1)
4570 		register_prot_hook(sk);
4571 
4572 	spin_unlock(&po->bind_lock);
4573 	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4574 		/* Because we don't support block-based V3 on tx-ring */
4575 		if (!tx_ring)
4576 			prb_shutdown_retire_blk_timer(po, rb_queue);
4577 	}
4578 
4579 out_free_pg_vec:
4580 	if (pg_vec) {
4581 		bitmap_free(rx_owner_map);
4582 		free_pg_vec(pg_vec, order, req->tp_block_nr);
4583 	}
4584 out:
4585 	return err;
4586 }
4587 
4588 static int packet_mmap(struct file *file, struct socket *sock,
4589 		struct vm_area_struct *vma)
4590 {
4591 	struct sock *sk = sock->sk;
4592 	struct packet_sock *po = pkt_sk(sk);
4593 	unsigned long size, expected_size;
4594 	struct packet_ring_buffer *rb;
4595 	unsigned long start;
4596 	int err = -EINVAL;
4597 	int i;
4598 
4599 	if (vma->vm_pgoff)
4600 		return -EINVAL;
4601 
4602 	mutex_lock(&po->pg_vec_lock);
4603 
4604 	expected_size = 0;
4605 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4606 		if (rb->pg_vec) {
4607 			expected_size += rb->pg_vec_len
4608 						* rb->pg_vec_pages
4609 						* PAGE_SIZE;
4610 		}
4611 	}
4612 
4613 	if (expected_size == 0)
4614 		goto out;
4615 
4616 	size = vma->vm_end - vma->vm_start;
4617 	if (size != expected_size)
4618 		goto out;
4619 
4620 	start = vma->vm_start;
4621 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4622 		if (rb->pg_vec == NULL)
4623 			continue;
4624 
4625 		for (i = 0; i < rb->pg_vec_len; i++) {
4626 			struct page *page;
4627 			void *kaddr = rb->pg_vec[i].buffer;
4628 			int pg_num;
4629 
4630 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4631 				page = pgv_to_page(kaddr);
4632 				err = vm_insert_page(vma, start, page);
4633 				if (unlikely(err))
4634 					goto out;
4635 				start += PAGE_SIZE;
4636 				kaddr += PAGE_SIZE;
4637 			}
4638 		}
4639 	}
4640 
4641 	atomic_long_inc(&po->mapped);
4642 	vma->vm_ops = &packet_mmap_ops;
4643 	err = 0;
4644 
4645 out:
4646 	mutex_unlock(&po->pg_vec_lock);
4647 	return err;
4648 }
4649 
4650 static const struct proto_ops packet_ops_spkt = {
4651 	.family =	PF_PACKET,
4652 	.owner =	THIS_MODULE,
4653 	.release =	packet_release,
4654 	.bind =		packet_bind_spkt,
4655 	.connect =	sock_no_connect,
4656 	.socketpair =	sock_no_socketpair,
4657 	.accept =	sock_no_accept,
4658 	.getname =	packet_getname_spkt,
4659 	.poll =		datagram_poll,
4660 	.ioctl =	packet_ioctl,
4661 	.gettstamp =	sock_gettstamp,
4662 	.listen =	sock_no_listen,
4663 	.shutdown =	sock_no_shutdown,
4664 	.sendmsg =	packet_sendmsg_spkt,
4665 	.recvmsg =	packet_recvmsg,
4666 	.mmap =		sock_no_mmap,
4667 };
4668 
4669 static const struct proto_ops packet_ops = {
4670 	.family =	PF_PACKET,
4671 	.owner =	THIS_MODULE,
4672 	.release =	packet_release,
4673 	.bind =		packet_bind,
4674 	.connect =	sock_no_connect,
4675 	.socketpair =	sock_no_socketpair,
4676 	.accept =	sock_no_accept,
4677 	.getname =	packet_getname,
4678 	.poll =		packet_poll,
4679 	.ioctl =	packet_ioctl,
4680 	.gettstamp =	sock_gettstamp,
4681 	.listen =	sock_no_listen,
4682 	.shutdown =	sock_no_shutdown,
4683 	.setsockopt =	packet_setsockopt,
4684 	.getsockopt_iter =	packet_getsockopt,
4685 	.sendmsg =	packet_sendmsg,
4686 	.recvmsg =	packet_recvmsg,
4687 	.mmap =		packet_mmap,
4688 };
4689 
4690 static const struct net_proto_family packet_family_ops = {
4691 	.family =	PF_PACKET,
4692 	.create =	packet_create,
4693 	.owner	=	THIS_MODULE,
4694 };
4695 
4696 static struct notifier_block packet_netdev_notifier = {
4697 	.notifier_call =	packet_notifier,
4698 };
4699 
4700 #ifdef CONFIG_PROC_FS
4701 
4702 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4703 	__acquires(RCU)
4704 {
4705 	struct net *net = seq_file_net(seq);
4706 
4707 	rcu_read_lock();
4708 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4709 }
4710 
4711 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4712 {
4713 	struct net *net = seq_file_net(seq);
4714 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4715 }
4716 
4717 static void packet_seq_stop(struct seq_file *seq, void *v)
4718 	__releases(RCU)
4719 {
4720 	rcu_read_unlock();
4721 }
4722 
4723 static int packet_seq_show(struct seq_file *seq, void *v)
4724 {
4725 	if (v == SEQ_START_TOKEN)
4726 		seq_printf(seq,
4727 			   "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4728 			   IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4729 	else {
4730 		struct sock *s = sk_entry(v);
4731 		const struct packet_sock *po = pkt_sk(s);
4732 
4733 		seq_printf(seq,
4734 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6llu\n",
4735 			   s,
4736 			   refcount_read(&s->sk_refcnt),
4737 			   s->sk_type,
4738 			   ntohs(READ_ONCE(po->num)),
4739 			   READ_ONCE(po->ifindex),
4740 			   packet_sock_flag(po, PACKET_SOCK_RUNNING),
4741 			   atomic_read(&s->sk_rmem_alloc),
4742 			   from_kuid_munged(seq_user_ns(seq), sk_uid(s)),
4743 			   sock_i_ino(s));
4744 	}
4745 
4746 	return 0;
4747 }
4748 
4749 static const struct seq_operations packet_seq_ops = {
4750 	.start	= packet_seq_start,
4751 	.next	= packet_seq_next,
4752 	.stop	= packet_seq_stop,
4753 	.show	= packet_seq_show,
4754 };
4755 #endif
4756 
4757 static int __net_init packet_net_init(struct net *net)
4758 {
4759 	mutex_init(&net->packet.sklist_lock);
4760 	INIT_HLIST_HEAD(&net->packet.sklist);
4761 
4762 #ifdef CONFIG_PROC_FS
4763 	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4764 			sizeof(struct seq_net_private)))
4765 		return -ENOMEM;
4766 #endif /* CONFIG_PROC_FS */
4767 
4768 	return 0;
4769 }
4770 
4771 static void __net_exit packet_net_exit(struct net *net)
4772 {
4773 	remove_proc_entry("packet", net->proc_net);
4774 	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4775 }
4776 
4777 static struct pernet_operations packet_net_ops = {
4778 	.init = packet_net_init,
4779 	.exit = packet_net_exit,
4780 };
4781 
4782 
4783 static void __exit packet_exit(void)
4784 {
4785 	sock_unregister(PF_PACKET);
4786 	proto_unregister(&packet_proto);
4787 	unregister_netdevice_notifier(&packet_netdev_notifier);
4788 	unregister_pernet_subsys(&packet_net_ops);
4789 }
4790 
4791 static int __init packet_init(void)
4792 {
4793 	int rc;
4794 
4795 	rc = register_pernet_subsys(&packet_net_ops);
4796 	if (rc)
4797 		goto out;
4798 	rc = register_netdevice_notifier(&packet_netdev_notifier);
4799 	if (rc)
4800 		goto out_pernet;
4801 	rc = proto_register(&packet_proto, 0);
4802 	if (rc)
4803 		goto out_notifier;
4804 	rc = sock_register(&packet_family_ops);
4805 	if (rc)
4806 		goto out_proto;
4807 
4808 	return 0;
4809 
4810 out_proto:
4811 	proto_unregister(&packet_proto);
4812 out_notifier:
4813 	unregister_netdevice_notifier(&packet_netdev_notifier);
4814 out_pernet:
4815 	unregister_pernet_subsys(&packet_net_ops);
4816 out:
4817 	return rc;
4818 }
4819 
4820 module_init(packet_init);
4821 module_exit(packet_exit);
4822 MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4823 MODULE_LICENSE("GPL");
4824 MODULE_ALIAS_NETPROTO(PF_PACKET);
4825