xref: /linux/net/packet/af_packet.c (revision 5a558f369ef89c6fd8170ee1137274fcc08517ae)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		PACKET - implements raw packet sockets.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *
13  * Fixes:
14  *		Alan Cox	:	verify_area() now used correctly
15  *		Alan Cox	:	new skbuff lists, look ma no backlogs!
16  *		Alan Cox	:	tidied skbuff lists.
17  *		Alan Cox	:	Now uses generic datagram routines I
18  *					added. Also fixed the peek/read crash
19  *					from all old Linux datagram code.
20  *		Alan Cox	:	Uses the improved datagram code.
21  *		Alan Cox	:	Added NULL's for socket options.
22  *		Alan Cox	:	Re-commented the code.
23  *		Alan Cox	:	Use new kernel side addressing
24  *		Rob Janssen	:	Correct MTU usage.
25  *		Dave Platt	:	Counter leaks caused by incorrect
26  *					interrupt locking and some slightly
27  *					dubious gcc output. Can you read
28  *					compiler: it said _VOLATILE_
29  *	Richard Kooijman	:	Timestamp fixes.
30  *		Alan Cox	:	New buffers. Use sk->mac.raw.
31  *		Alan Cox	:	sendmsg/recvmsg support.
32  *		Alan Cox	:	Protocol setting support
33  *	Alexey Kuznetsov	:	Untied from IPv4 stack.
34  *	Cyrus Durgin		:	Fixed kerneld for kmod.
35  *	Michal Ostrowski        :       Module initialization cleanup.
36  *         Ulises Alonso        :       Frame number limit removal and
37  *                                      packet_set_ring memory leak.
38  *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
39  *					The convention is that longer addresses
40  *					will simply extend the hardware address
41  *					byte arrays at the end of sockaddr_ll
42  *					and packet_mreq.
43  *		Johann Baudy	:	Added TX RING.
44  *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
45  *					layer.
46  *					Copyright (C) 2011, <lokec@ccs.neu.edu>
47  */
48 
49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
50 
51 #include <linux/ethtool.h>
52 #include <linux/filter.h>
53 #include <linux/types.h>
54 #include <linux/mm.h>
55 #include <linux/capability.h>
56 #include <linux/fcntl.h>
57 #include <linux/socket.h>
58 #include <linux/in.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/if_packet.h>
62 #include <linux/wireless.h>
63 #include <linux/kernel.h>
64 #include <linux/kmod.h>
65 #include <linux/slab.h>
66 #include <linux/vmalloc.h>
67 #include <net/net_namespace.h>
68 #include <net/ip.h>
69 #include <net/protocol.h>
70 #include <linux/skbuff.h>
71 #include <net/sock.h>
72 #include <linux/errno.h>
73 #include <linux/timer.h>
74 #include <linux/uaccess.h>
75 #include <asm/ioctls.h>
76 #include <asm/page.h>
77 #include <asm/cacheflush.h>
78 #include <asm/io.h>
79 #include <linux/proc_fs.h>
80 #include <linux/seq_file.h>
81 #include <linux/poll.h>
82 #include <linux/module.h>
83 #include <linux/init.h>
84 #include <linux/mutex.h>
85 #include <linux/if_vlan.h>
86 #include <linux/virtio_net.h>
87 #include <linux/errqueue.h>
88 #include <linux/net_tstamp.h>
89 #include <linux/percpu.h>
90 #ifdef CONFIG_INET
91 #include <net/inet_common.h>
92 #endif
93 #include <linux/bpf.h>
94 #include <net/compat.h>
95 #include <linux/netfilter_netdev.h>
96 
97 #include "internal.h"
98 
99 /*
100    Assumptions:
101    - If the device has no dev->header_ops->create, there is no LL header
102      visible above the device. In this case, its hard_header_len should be 0.
103      The device may prepend its own header internally. In this case, its
104      needed_headroom should be set to the space needed for it to add its
105      internal header.
106      For example, a WiFi driver pretending to be an Ethernet driver should
107      set its hard_header_len to be the Ethernet header length, and set its
108      needed_headroom to be (the real WiFi header length - the fake Ethernet
109      header length).
110    - packet socket receives packets with pulled ll header,
111      so that SOCK_RAW should push it back.
112 
113 On receive:
114 -----------
115 
116 Incoming, dev_has_header(dev) == true
117    mac_header -> ll header
118    data       -> data
119 
120 Outgoing, dev_has_header(dev) == true
121    mac_header -> ll header
122    data       -> ll header
123 
124 Incoming, dev_has_header(dev) == false
125    mac_header -> data
126      However drivers often make it point to the ll header.
127      This is incorrect because the ll header should be invisible to us.
128    data       -> data
129 
130 Outgoing, dev_has_header(dev) == false
131    mac_header -> data. ll header is invisible to us.
132    data       -> data
133 
134 Resume
135   If dev_has_header(dev) == false we are unable to restore the ll header,
136     because it is invisible to us.
137 
138 
139 On transmit:
140 ------------
141 
142 dev_has_header(dev) == true
143    mac_header -> ll header
144    data       -> ll header
145 
146 dev_has_header(dev) == false (ll header is invisible to us)
147    mac_header -> data
148    data       -> data
149 
150    We should set network_header on output to the correct position,
151    packet classifier depends on it.
152  */
153 
154 /* Private packet socket structures. */
155 
156 /* identical to struct packet_mreq except it has
157  * a longer address field.
158  */
159 struct packet_mreq_max {
160 	int		mr_ifindex;
161 	unsigned short	mr_type;
162 	unsigned short	mr_alen;
163 	unsigned char	mr_address[MAX_ADDR_LEN];
164 };
165 
166 union tpacket_uhdr {
167 	struct tpacket_hdr  *h1;
168 	struct tpacket2_hdr *h2;
169 	struct tpacket3_hdr *h3;
170 	void *raw;
171 };
172 
173 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
174 		int closing, int tx_ring);
175 
176 #define V3_ALIGNMENT	(8)
177 
178 #define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
179 
180 #define BLK_PLUS_PRIV(sz_of_priv) \
181 	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
182 
183 #define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
184 #define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
185 #define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
186 #define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
187 #define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
188 #define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
189 
190 struct packet_sock;
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 		       struct packet_type *pt, struct net_device *orig_dev);
193 
194 static void *packet_previous_frame(struct packet_sock *po,
195 		struct packet_ring_buffer *rb,
196 		int status);
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 			struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 		struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 		struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
209 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
210 		struct tpacket3_hdr *);
211 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
212 		struct tpacket3_hdr *);
213 static void packet_flush_mclist(struct sock *sk);
214 static u16 packet_pick_tx_queue(struct sk_buff *skb);
215 
216 struct packet_skb_cb {
217 	union {
218 		struct sockaddr_pkt pkt;
219 		union {
220 			/* Trick: alias skb original length with
221 			 * ll.sll_family and ll.protocol in order
222 			 * to save room.
223 			 */
224 			unsigned int origlen;
225 			struct sockaddr_ll ll;
226 		};
227 	} sa;
228 };
229 
230 #define vio_le() virtio_legacy_is_little_endian()
231 
232 #define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
233 
234 #define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid)	\
236 	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
238 	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 	((x)->kactive_blk_num+1) : 0)
242 
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
245 
246 #ifdef CONFIG_NETFILTER_EGRESS
247 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
248 {
249 	struct sk_buff *next, *head = NULL, *tail;
250 	int rc;
251 
252 	rcu_read_lock();
253 	for (; skb != NULL; skb = next) {
254 		next = skb->next;
255 		skb_mark_not_on_list(skb);
256 
257 		if (!nf_hook_egress(skb, &rc, skb->dev))
258 			continue;
259 
260 		if (!head)
261 			head = skb;
262 		else
263 			tail->next = skb;
264 
265 		tail = skb;
266 	}
267 	rcu_read_unlock();
268 
269 	return head;
270 }
271 #endif
272 
273 static int packet_xmit(const struct packet_sock *po, struct sk_buff *skb)
274 {
275 	if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS))
276 		return dev_queue_xmit(skb);
277 
278 #ifdef CONFIG_NETFILTER_EGRESS
279 	if (nf_hook_egress_active()) {
280 		skb = nf_hook_direct_egress(skb);
281 		if (!skb)
282 			return NET_XMIT_DROP;
283 	}
284 #endif
285 	return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
286 }
287 
288 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
289 {
290 	struct net_device *dev;
291 
292 	rcu_read_lock();
293 	dev = rcu_dereference(po->cached_dev);
294 	dev_hold(dev);
295 	rcu_read_unlock();
296 
297 	return dev;
298 }
299 
300 static void packet_cached_dev_assign(struct packet_sock *po,
301 				     struct net_device *dev)
302 {
303 	rcu_assign_pointer(po->cached_dev, dev);
304 }
305 
306 static void packet_cached_dev_reset(struct packet_sock *po)
307 {
308 	RCU_INIT_POINTER(po->cached_dev, NULL);
309 }
310 
311 static u16 packet_pick_tx_queue(struct sk_buff *skb)
312 {
313 	struct net_device *dev = skb->dev;
314 	const struct net_device_ops *ops = dev->netdev_ops;
315 	int cpu = raw_smp_processor_id();
316 	u16 queue_index;
317 
318 #ifdef CONFIG_XPS
319 	skb->sender_cpu = cpu + 1;
320 #endif
321 	skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
322 	if (ops->ndo_select_queue) {
323 		queue_index = ops->ndo_select_queue(dev, skb, NULL);
324 		queue_index = netdev_cap_txqueue(dev, queue_index);
325 	} else {
326 		queue_index = netdev_pick_tx(dev, skb, NULL);
327 	}
328 
329 	return queue_index;
330 }
331 
332 /* __register_prot_hook must be invoked through register_prot_hook
333  * or from a context in which asynchronous accesses to the packet
334  * socket is not possible (packet_create()).
335  */
336 static void __register_prot_hook(struct sock *sk)
337 {
338 	struct packet_sock *po = pkt_sk(sk);
339 
340 	if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
341 		if (po->fanout)
342 			__fanout_link(sk, po);
343 		else
344 			dev_add_pack(&po->prot_hook);
345 
346 		sock_hold(sk);
347 		packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1);
348 	}
349 }
350 
351 static void register_prot_hook(struct sock *sk)
352 {
353 	lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
354 	__register_prot_hook(sk);
355 }
356 
357 /* If the sync parameter is true, we will temporarily drop
358  * the po->bind_lock and do a synchronize_net to make sure no
359  * asynchronous packet processing paths still refer to the elements
360  * of po->prot_hook.  If the sync parameter is false, it is the
361  * callers responsibility to take care of this.
362  */
363 static void __unregister_prot_hook(struct sock *sk, bool sync)
364 {
365 	struct packet_sock *po = pkt_sk(sk);
366 
367 	lockdep_assert_held_once(&po->bind_lock);
368 
369 	packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0);
370 
371 	if (po->fanout)
372 		__fanout_unlink(sk, po);
373 	else
374 		__dev_remove_pack(&po->prot_hook);
375 
376 	__sock_put(sk);
377 
378 	if (sync) {
379 		spin_unlock(&po->bind_lock);
380 		synchronize_net();
381 		spin_lock(&po->bind_lock);
382 	}
383 }
384 
385 static void unregister_prot_hook(struct sock *sk, bool sync)
386 {
387 	struct packet_sock *po = pkt_sk(sk);
388 
389 	if (packet_sock_flag(po, PACKET_SOCK_RUNNING))
390 		__unregister_prot_hook(sk, sync);
391 }
392 
393 static inline struct page * __pure pgv_to_page(void *addr)
394 {
395 	if (is_vmalloc_addr(addr))
396 		return vmalloc_to_page(addr);
397 	return virt_to_page(addr);
398 }
399 
400 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
401 {
402 	union tpacket_uhdr h;
403 
404 	/* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */
405 
406 	h.raw = frame;
407 	switch (po->tp_version) {
408 	case TPACKET_V1:
409 		WRITE_ONCE(h.h1->tp_status, status);
410 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
411 		break;
412 	case TPACKET_V2:
413 		WRITE_ONCE(h.h2->tp_status, status);
414 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
415 		break;
416 	case TPACKET_V3:
417 		WRITE_ONCE(h.h3->tp_status, status);
418 		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
419 		break;
420 	default:
421 		WARN(1, "TPACKET version not supported.\n");
422 		BUG();
423 	}
424 
425 	smp_wmb();
426 }
427 
428 static int __packet_get_status(const struct packet_sock *po, void *frame)
429 {
430 	union tpacket_uhdr h;
431 
432 	smp_rmb();
433 
434 	/* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */
435 
436 	h.raw = frame;
437 	switch (po->tp_version) {
438 	case TPACKET_V1:
439 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
440 		return READ_ONCE(h.h1->tp_status);
441 	case TPACKET_V2:
442 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
443 		return READ_ONCE(h.h2->tp_status);
444 	case TPACKET_V3:
445 		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
446 		return READ_ONCE(h.h3->tp_status);
447 	default:
448 		WARN(1, "TPACKET version not supported.\n");
449 		BUG();
450 		return 0;
451 	}
452 }
453 
454 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
455 				   unsigned int flags)
456 {
457 	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
458 
459 	if (shhwtstamps &&
460 	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
461 	    ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
462 		return TP_STATUS_TS_RAW_HARDWARE;
463 
464 	if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
465 	    ktime_to_timespec64_cond(skb_tstamp(skb), ts))
466 		return TP_STATUS_TS_SOFTWARE;
467 
468 	return 0;
469 }
470 
471 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
472 				    struct sk_buff *skb)
473 {
474 	union tpacket_uhdr h;
475 	struct timespec64 ts;
476 	__u32 ts_status;
477 
478 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp))))
479 		return 0;
480 
481 	h.raw = frame;
482 	/*
483 	 * versions 1 through 3 overflow the timestamps in y2106, since they
484 	 * all store the seconds in a 32-bit unsigned integer.
485 	 * If we create a version 4, that should have a 64-bit timestamp,
486 	 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
487 	 * nanoseconds.
488 	 */
489 	switch (po->tp_version) {
490 	case TPACKET_V1:
491 		h.h1->tp_sec = ts.tv_sec;
492 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
493 		break;
494 	case TPACKET_V2:
495 		h.h2->tp_sec = ts.tv_sec;
496 		h.h2->tp_nsec = ts.tv_nsec;
497 		break;
498 	case TPACKET_V3:
499 		h.h3->tp_sec = ts.tv_sec;
500 		h.h3->tp_nsec = ts.tv_nsec;
501 		break;
502 	default:
503 		WARN(1, "TPACKET version not supported.\n");
504 		BUG();
505 	}
506 
507 	/* one flush is safe, as both fields always lie on the same cacheline */
508 	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
509 	smp_wmb();
510 
511 	return ts_status;
512 }
513 
514 static void *packet_lookup_frame(const struct packet_sock *po,
515 				 const struct packet_ring_buffer *rb,
516 				 unsigned int position,
517 				 int status)
518 {
519 	unsigned int pg_vec_pos, frame_offset;
520 	union tpacket_uhdr h;
521 
522 	pg_vec_pos = position / rb->frames_per_block;
523 	frame_offset = position % rb->frames_per_block;
524 
525 	h.raw = rb->pg_vec[pg_vec_pos].buffer +
526 		(frame_offset * rb->frame_size);
527 
528 	if (status != __packet_get_status(po, h.raw))
529 		return NULL;
530 
531 	return h.raw;
532 }
533 
534 static void *packet_current_frame(struct packet_sock *po,
535 		struct packet_ring_buffer *rb,
536 		int status)
537 {
538 	return packet_lookup_frame(po, rb, rb->head, status);
539 }
540 
541 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
542 {
543 	del_timer_sync(&pkc->retire_blk_timer);
544 }
545 
546 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
547 		struct sk_buff_head *rb_queue)
548 {
549 	struct tpacket_kbdq_core *pkc;
550 
551 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552 
553 	spin_lock_bh(&rb_queue->lock);
554 	pkc->delete_blk_timer = 1;
555 	spin_unlock_bh(&rb_queue->lock);
556 
557 	prb_del_retire_blk_timer(pkc);
558 }
559 
560 static void prb_setup_retire_blk_timer(struct packet_sock *po)
561 {
562 	struct tpacket_kbdq_core *pkc;
563 
564 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
565 	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
566 		    0);
567 	pkc->retire_blk_timer.expires = jiffies;
568 }
569 
570 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
571 				int blk_size_in_bytes)
572 {
573 	struct net_device *dev;
574 	unsigned int mbits, div;
575 	struct ethtool_link_ksettings ecmd;
576 	int err;
577 
578 	rtnl_lock();
579 	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
580 	if (unlikely(!dev)) {
581 		rtnl_unlock();
582 		return DEFAULT_PRB_RETIRE_TOV;
583 	}
584 	err = __ethtool_get_link_ksettings(dev, &ecmd);
585 	rtnl_unlock();
586 	if (err)
587 		return DEFAULT_PRB_RETIRE_TOV;
588 
589 	/* If the link speed is so slow you don't really
590 	 * need to worry about perf anyways
591 	 */
592 	if (ecmd.base.speed < SPEED_1000 ||
593 	    ecmd.base.speed == SPEED_UNKNOWN)
594 		return DEFAULT_PRB_RETIRE_TOV;
595 
596 	div = ecmd.base.speed / 1000;
597 	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
598 
599 	if (div)
600 		mbits /= div;
601 
602 	if (div)
603 		return mbits + 1;
604 	return mbits;
605 }
606 
607 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
608 			union tpacket_req_u *req_u)
609 {
610 	p1->feature_req_word = req_u->req3.tp_feature_req_word;
611 }
612 
613 static void init_prb_bdqc(struct packet_sock *po,
614 			struct packet_ring_buffer *rb,
615 			struct pgv *pg_vec,
616 			union tpacket_req_u *req_u)
617 {
618 	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
619 	struct tpacket_block_desc *pbd;
620 
621 	memset(p1, 0x0, sizeof(*p1));
622 
623 	p1->knxt_seq_num = 1;
624 	p1->pkbdq = pg_vec;
625 	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
626 	p1->pkblk_start	= pg_vec[0].buffer;
627 	p1->kblk_size = req_u->req3.tp_block_size;
628 	p1->knum_blocks	= req_u->req3.tp_block_nr;
629 	p1->hdrlen = po->tp_hdrlen;
630 	p1->version = po->tp_version;
631 	p1->last_kactive_blk_num = 0;
632 	po->stats.stats3.tp_freeze_q_cnt = 0;
633 	if (req_u->req3.tp_retire_blk_tov)
634 		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
635 	else
636 		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
637 						req_u->req3.tp_block_size);
638 	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
639 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
640 	rwlock_init(&p1->blk_fill_in_prog_lock);
641 
642 	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
643 	prb_init_ft_ops(p1, req_u);
644 	prb_setup_retire_blk_timer(po);
645 	prb_open_block(p1, pbd);
646 }
647 
648 /*  Do NOT update the last_blk_num first.
649  *  Assumes sk_buff_head lock is held.
650  */
651 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
652 {
653 	mod_timer(&pkc->retire_blk_timer,
654 			jiffies + pkc->tov_in_jiffies);
655 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
656 }
657 
658 /*
659  * Timer logic:
660  * 1) We refresh the timer only when we open a block.
661  *    By doing this we don't waste cycles refreshing the timer
662  *	  on packet-by-packet basis.
663  *
664  * With a 1MB block-size, on a 1Gbps line, it will take
665  * i) ~8 ms to fill a block + ii) memcpy etc.
666  * In this cut we are not accounting for the memcpy time.
667  *
668  * So, if the user sets the 'tmo' to 10ms then the timer
669  * will never fire while the block is still getting filled
670  * (which is what we want). However, the user could choose
671  * to close a block early and that's fine.
672  *
673  * But when the timer does fire, we check whether or not to refresh it.
674  * Since the tmo granularity is in msecs, it is not too expensive
675  * to refresh the timer, lets say every '8' msecs.
676  * Either the user can set the 'tmo' or we can derive it based on
677  * a) line-speed and b) block-size.
678  * prb_calc_retire_blk_tmo() calculates the tmo.
679  *
680  */
681 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
682 {
683 	struct packet_sock *po =
684 		from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
685 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
686 	unsigned int frozen;
687 	struct tpacket_block_desc *pbd;
688 
689 	spin_lock(&po->sk.sk_receive_queue.lock);
690 
691 	frozen = prb_queue_frozen(pkc);
692 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
693 
694 	if (unlikely(pkc->delete_blk_timer))
695 		goto out;
696 
697 	/* We only need to plug the race when the block is partially filled.
698 	 * tpacket_rcv:
699 	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
700 	 *		copy_bits() is in progress ...
701 	 *		timer fires on other cpu:
702 	 *		we can't retire the current block because copy_bits
703 	 *		is in progress.
704 	 *
705 	 */
706 	if (BLOCK_NUM_PKTS(pbd)) {
707 		/* Waiting for skb_copy_bits to finish... */
708 		write_lock(&pkc->blk_fill_in_prog_lock);
709 		write_unlock(&pkc->blk_fill_in_prog_lock);
710 	}
711 
712 	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
713 		if (!frozen) {
714 			if (!BLOCK_NUM_PKTS(pbd)) {
715 				/* An empty block. Just refresh the timer. */
716 				goto refresh_timer;
717 			}
718 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
719 			if (!prb_dispatch_next_block(pkc, po))
720 				goto refresh_timer;
721 			else
722 				goto out;
723 		} else {
724 			/* Case 1. Queue was frozen because user-space was
725 			 *	   lagging behind.
726 			 */
727 			if (prb_curr_blk_in_use(pbd)) {
728 				/*
729 				 * Ok, user-space is still behind.
730 				 * So just refresh the timer.
731 				 */
732 				goto refresh_timer;
733 			} else {
734 			       /* Case 2. queue was frozen,user-space caught up,
735 				* now the link went idle && the timer fired.
736 				* We don't have a block to close.So we open this
737 				* block and restart the timer.
738 				* opening a block thaws the queue,restarts timer
739 				* Thawing/timer-refresh is a side effect.
740 				*/
741 				prb_open_block(pkc, pbd);
742 				goto out;
743 			}
744 		}
745 	}
746 
747 refresh_timer:
748 	_prb_refresh_rx_retire_blk_timer(pkc);
749 
750 out:
751 	spin_unlock(&po->sk.sk_receive_queue.lock);
752 }
753 
754 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
755 		struct tpacket_block_desc *pbd1, __u32 status)
756 {
757 	/* Flush everything minus the block header */
758 
759 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
760 	u8 *start, *end;
761 
762 	start = (u8 *)pbd1;
763 
764 	/* Skip the block header(we know header WILL fit in 4K) */
765 	start += PAGE_SIZE;
766 
767 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
768 	for (; start < end; start += PAGE_SIZE)
769 		flush_dcache_page(pgv_to_page(start));
770 
771 	smp_wmb();
772 #endif
773 
774 	/* Now update the block status. */
775 
776 	BLOCK_STATUS(pbd1) = status;
777 
778 	/* Flush the block header */
779 
780 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
781 	start = (u8 *)pbd1;
782 	flush_dcache_page(pgv_to_page(start));
783 
784 	smp_wmb();
785 #endif
786 }
787 
788 /*
789  * Side effect:
790  *
791  * 1) flush the block
792  * 2) Increment active_blk_num
793  *
794  * Note:We DONT refresh the timer on purpose.
795  *	Because almost always the next block will be opened.
796  */
797 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
798 		struct tpacket_block_desc *pbd1,
799 		struct packet_sock *po, unsigned int stat)
800 {
801 	__u32 status = TP_STATUS_USER | stat;
802 
803 	struct tpacket3_hdr *last_pkt;
804 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
805 	struct sock *sk = &po->sk;
806 
807 	if (atomic_read(&po->tp_drops))
808 		status |= TP_STATUS_LOSING;
809 
810 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
811 	last_pkt->tp_next_offset = 0;
812 
813 	/* Get the ts of the last pkt */
814 	if (BLOCK_NUM_PKTS(pbd1)) {
815 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
816 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
817 	} else {
818 		/* Ok, we tmo'd - so get the current time.
819 		 *
820 		 * It shouldn't really happen as we don't close empty
821 		 * blocks. See prb_retire_rx_blk_timer_expired().
822 		 */
823 		struct timespec64 ts;
824 		ktime_get_real_ts64(&ts);
825 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
826 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
827 	}
828 
829 	smp_wmb();
830 
831 	/* Flush the block */
832 	prb_flush_block(pkc1, pbd1, status);
833 
834 	sk->sk_data_ready(sk);
835 
836 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
837 }
838 
839 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
840 {
841 	pkc->reset_pending_on_curr_blk = 0;
842 }
843 
844 /*
845  * Side effect of opening a block:
846  *
847  * 1) prb_queue is thawed.
848  * 2) retire_blk_timer is refreshed.
849  *
850  */
851 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
852 	struct tpacket_block_desc *pbd1)
853 {
854 	struct timespec64 ts;
855 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
856 
857 	smp_rmb();
858 
859 	/* We could have just memset this but we will lose the
860 	 * flexibility of making the priv area sticky
861 	 */
862 
863 	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
864 	BLOCK_NUM_PKTS(pbd1) = 0;
865 	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 
867 	ktime_get_real_ts64(&ts);
868 
869 	h1->ts_first_pkt.ts_sec = ts.tv_sec;
870 	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
871 
872 	pkc1->pkblk_start = (char *)pbd1;
873 	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
874 
875 	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
876 	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
877 
878 	pbd1->version = pkc1->version;
879 	pkc1->prev = pkc1->nxt_offset;
880 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
881 
882 	prb_thaw_queue(pkc1);
883 	_prb_refresh_rx_retire_blk_timer(pkc1);
884 
885 	smp_wmb();
886 }
887 
888 /*
889  * Queue freeze logic:
890  * 1) Assume tp_block_nr = 8 blocks.
891  * 2) At time 't0', user opens Rx ring.
892  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
893  * 4) user-space is either sleeping or processing block '0'.
894  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
895  *    it will close block-7,loop around and try to fill block '0'.
896  *    call-flow:
897  *    __packet_lookup_frame_in_block
898  *      prb_retire_current_block()
899  *      prb_dispatch_next_block()
900  *        |->(BLOCK_STATUS == USER) evaluates to true
901  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
902  * 6) Now there are two cases:
903  *    6.1) Link goes idle right after the queue is frozen.
904  *         But remember, the last open_block() refreshed the timer.
905  *         When this timer expires,it will refresh itself so that we can
906  *         re-open block-0 in near future.
907  *    6.2) Link is busy and keeps on receiving packets. This is a simple
908  *         case and __packet_lookup_frame_in_block will check if block-0
909  *         is free and can now be re-used.
910  */
911 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
912 				  struct packet_sock *po)
913 {
914 	pkc->reset_pending_on_curr_blk = 1;
915 	po->stats.stats3.tp_freeze_q_cnt++;
916 }
917 
918 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
919 
920 /*
921  * If the next block is free then we will dispatch it
922  * and return a good offset.
923  * Else, we will freeze the queue.
924  * So, caller must check the return value.
925  */
926 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
927 		struct packet_sock *po)
928 {
929 	struct tpacket_block_desc *pbd;
930 
931 	smp_rmb();
932 
933 	/* 1. Get current block num */
934 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
935 
936 	/* 2. If this block is currently in_use then freeze the queue */
937 	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
938 		prb_freeze_queue(pkc, po);
939 		return NULL;
940 	}
941 
942 	/*
943 	 * 3.
944 	 * open this block and return the offset where the first packet
945 	 * needs to get stored.
946 	 */
947 	prb_open_block(pkc, pbd);
948 	return (void *)pkc->nxt_offset;
949 }
950 
951 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
952 		struct packet_sock *po, unsigned int status)
953 {
954 	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
955 
956 	/* retire/close the current block */
957 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
958 		/*
959 		 * Plug the case where copy_bits() is in progress on
960 		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
961 		 * have space to copy the pkt in the current block and
962 		 * called prb_retire_current_block()
963 		 *
964 		 * We don't need to worry about the TMO case because
965 		 * the timer-handler already handled this case.
966 		 */
967 		if (!(status & TP_STATUS_BLK_TMO)) {
968 			/* Waiting for skb_copy_bits to finish... */
969 			write_lock(&pkc->blk_fill_in_prog_lock);
970 			write_unlock(&pkc->blk_fill_in_prog_lock);
971 		}
972 		prb_close_block(pkc, pbd, po, status);
973 		return;
974 	}
975 }
976 
977 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
978 {
979 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
980 }
981 
982 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
983 {
984 	return pkc->reset_pending_on_curr_blk;
985 }
986 
987 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
988 	__releases(&pkc->blk_fill_in_prog_lock)
989 {
990 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
991 
992 	read_unlock(&pkc->blk_fill_in_prog_lock);
993 }
994 
995 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
996 			struct tpacket3_hdr *ppd)
997 {
998 	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
999 }
1000 
1001 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1002 			struct tpacket3_hdr *ppd)
1003 {
1004 	ppd->hv1.tp_rxhash = 0;
1005 }
1006 
1007 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1008 			struct tpacket3_hdr *ppd)
1009 {
1010 	if (skb_vlan_tag_present(pkc->skb)) {
1011 		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1012 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1013 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1014 	} else {
1015 		ppd->hv1.tp_vlan_tci = 0;
1016 		ppd->hv1.tp_vlan_tpid = 0;
1017 		ppd->tp_status = TP_STATUS_AVAILABLE;
1018 	}
1019 }
1020 
1021 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1022 			struct tpacket3_hdr *ppd)
1023 {
1024 	ppd->hv1.tp_padding = 0;
1025 	prb_fill_vlan_info(pkc, ppd);
1026 
1027 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1028 		prb_fill_rxhash(pkc, ppd);
1029 	else
1030 		prb_clear_rxhash(pkc, ppd);
1031 }
1032 
1033 static void prb_fill_curr_block(char *curr,
1034 				struct tpacket_kbdq_core *pkc,
1035 				struct tpacket_block_desc *pbd,
1036 				unsigned int len)
1037 	__acquires(&pkc->blk_fill_in_prog_lock)
1038 {
1039 	struct tpacket3_hdr *ppd;
1040 
1041 	ppd  = (struct tpacket3_hdr *)curr;
1042 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1043 	pkc->prev = curr;
1044 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1045 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1046 	BLOCK_NUM_PKTS(pbd) += 1;
1047 	read_lock(&pkc->blk_fill_in_prog_lock);
1048 	prb_run_all_ft_ops(pkc, ppd);
1049 }
1050 
1051 /* Assumes caller has the sk->rx_queue.lock */
1052 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1053 					    struct sk_buff *skb,
1054 					    unsigned int len
1055 					    )
1056 {
1057 	struct tpacket_kbdq_core *pkc;
1058 	struct tpacket_block_desc *pbd;
1059 	char *curr, *end;
1060 
1061 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1062 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 
1064 	/* Queue is frozen when user space is lagging behind */
1065 	if (prb_queue_frozen(pkc)) {
1066 		/*
1067 		 * Check if that last block which caused the queue to freeze,
1068 		 * is still in_use by user-space.
1069 		 */
1070 		if (prb_curr_blk_in_use(pbd)) {
1071 			/* Can't record this packet */
1072 			return NULL;
1073 		} else {
1074 			/*
1075 			 * Ok, the block was released by user-space.
1076 			 * Now let's open that block.
1077 			 * opening a block also thaws the queue.
1078 			 * Thawing is a side effect.
1079 			 */
1080 			prb_open_block(pkc, pbd);
1081 		}
1082 	}
1083 
1084 	smp_mb();
1085 	curr = pkc->nxt_offset;
1086 	pkc->skb = skb;
1087 	end = (char *)pbd + pkc->kblk_size;
1088 
1089 	/* first try the current block */
1090 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1091 		prb_fill_curr_block(curr, pkc, pbd, len);
1092 		return (void *)curr;
1093 	}
1094 
1095 	/* Ok, close the current block */
1096 	prb_retire_current_block(pkc, po, 0);
1097 
1098 	/* Now, try to dispatch the next block */
1099 	curr = (char *)prb_dispatch_next_block(pkc, po);
1100 	if (curr) {
1101 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1102 		prb_fill_curr_block(curr, pkc, pbd, len);
1103 		return (void *)curr;
1104 	}
1105 
1106 	/*
1107 	 * No free blocks are available.user_space hasn't caught up yet.
1108 	 * Queue was just frozen and now this packet will get dropped.
1109 	 */
1110 	return NULL;
1111 }
1112 
1113 static void *packet_current_rx_frame(struct packet_sock *po,
1114 					    struct sk_buff *skb,
1115 					    int status, unsigned int len)
1116 {
1117 	char *curr = NULL;
1118 	switch (po->tp_version) {
1119 	case TPACKET_V1:
1120 	case TPACKET_V2:
1121 		curr = packet_lookup_frame(po, &po->rx_ring,
1122 					po->rx_ring.head, status);
1123 		return curr;
1124 	case TPACKET_V3:
1125 		return __packet_lookup_frame_in_block(po, skb, len);
1126 	default:
1127 		WARN(1, "TPACKET version not supported\n");
1128 		BUG();
1129 		return NULL;
1130 	}
1131 }
1132 
1133 static void *prb_lookup_block(const struct packet_sock *po,
1134 			      const struct packet_ring_buffer *rb,
1135 			      unsigned int idx,
1136 			      int status)
1137 {
1138 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1139 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1140 
1141 	if (status != BLOCK_STATUS(pbd))
1142 		return NULL;
1143 	return pbd;
1144 }
1145 
1146 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1147 {
1148 	unsigned int prev;
1149 	if (rb->prb_bdqc.kactive_blk_num)
1150 		prev = rb->prb_bdqc.kactive_blk_num-1;
1151 	else
1152 		prev = rb->prb_bdqc.knum_blocks-1;
1153 	return prev;
1154 }
1155 
1156 /* Assumes caller has held the rx_queue.lock */
1157 static void *__prb_previous_block(struct packet_sock *po,
1158 					 struct packet_ring_buffer *rb,
1159 					 int status)
1160 {
1161 	unsigned int previous = prb_previous_blk_num(rb);
1162 	return prb_lookup_block(po, rb, previous, status);
1163 }
1164 
1165 static void *packet_previous_rx_frame(struct packet_sock *po,
1166 					     struct packet_ring_buffer *rb,
1167 					     int status)
1168 {
1169 	if (po->tp_version <= TPACKET_V2)
1170 		return packet_previous_frame(po, rb, status);
1171 
1172 	return __prb_previous_block(po, rb, status);
1173 }
1174 
1175 static void packet_increment_rx_head(struct packet_sock *po,
1176 					    struct packet_ring_buffer *rb)
1177 {
1178 	switch (po->tp_version) {
1179 	case TPACKET_V1:
1180 	case TPACKET_V2:
1181 		return packet_increment_head(rb);
1182 	case TPACKET_V3:
1183 	default:
1184 		WARN(1, "TPACKET version not supported.\n");
1185 		BUG();
1186 		return;
1187 	}
1188 }
1189 
1190 static void *packet_previous_frame(struct packet_sock *po,
1191 		struct packet_ring_buffer *rb,
1192 		int status)
1193 {
1194 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1195 	return packet_lookup_frame(po, rb, previous, status);
1196 }
1197 
1198 static void packet_increment_head(struct packet_ring_buffer *buff)
1199 {
1200 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1201 }
1202 
1203 static void packet_inc_pending(struct packet_ring_buffer *rb)
1204 {
1205 	this_cpu_inc(*rb->pending_refcnt);
1206 }
1207 
1208 static void packet_dec_pending(struct packet_ring_buffer *rb)
1209 {
1210 	this_cpu_dec(*rb->pending_refcnt);
1211 }
1212 
1213 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1214 {
1215 	unsigned int refcnt = 0;
1216 	int cpu;
1217 
1218 	/* We don't use pending refcount in rx_ring. */
1219 	if (rb->pending_refcnt == NULL)
1220 		return 0;
1221 
1222 	for_each_possible_cpu(cpu)
1223 		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1224 
1225 	return refcnt;
1226 }
1227 
1228 static int packet_alloc_pending(struct packet_sock *po)
1229 {
1230 	po->rx_ring.pending_refcnt = NULL;
1231 
1232 	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1233 	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1234 		return -ENOBUFS;
1235 
1236 	return 0;
1237 }
1238 
1239 static void packet_free_pending(struct packet_sock *po)
1240 {
1241 	free_percpu(po->tx_ring.pending_refcnt);
1242 }
1243 
1244 #define ROOM_POW_OFF	2
1245 #define ROOM_NONE	0x0
1246 #define ROOM_LOW	0x1
1247 #define ROOM_NORMAL	0x2
1248 
1249 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1250 {
1251 	int idx, len;
1252 
1253 	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1254 	idx = READ_ONCE(po->rx_ring.head);
1255 	if (pow_off)
1256 		idx += len >> pow_off;
1257 	if (idx >= len)
1258 		idx -= len;
1259 	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1260 }
1261 
1262 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1263 {
1264 	int idx, len;
1265 
1266 	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1267 	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1268 	if (pow_off)
1269 		idx += len >> pow_off;
1270 	if (idx >= len)
1271 		idx -= len;
1272 	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1273 }
1274 
1275 static int __packet_rcv_has_room(const struct packet_sock *po,
1276 				 const struct sk_buff *skb)
1277 {
1278 	const struct sock *sk = &po->sk;
1279 	int ret = ROOM_NONE;
1280 
1281 	if (po->prot_hook.func != tpacket_rcv) {
1282 		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1283 		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1284 				   - (skb ? skb->truesize : 0);
1285 
1286 		if (avail > (rcvbuf >> ROOM_POW_OFF))
1287 			return ROOM_NORMAL;
1288 		else if (avail > 0)
1289 			return ROOM_LOW;
1290 		else
1291 			return ROOM_NONE;
1292 	}
1293 
1294 	if (po->tp_version == TPACKET_V3) {
1295 		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1296 			ret = ROOM_NORMAL;
1297 		else if (__tpacket_v3_has_room(po, 0))
1298 			ret = ROOM_LOW;
1299 	} else {
1300 		if (__tpacket_has_room(po, ROOM_POW_OFF))
1301 			ret = ROOM_NORMAL;
1302 		else if (__tpacket_has_room(po, 0))
1303 			ret = ROOM_LOW;
1304 	}
1305 
1306 	return ret;
1307 }
1308 
1309 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1310 {
1311 	bool pressure;
1312 	int ret;
1313 
1314 	ret = __packet_rcv_has_room(po, skb);
1315 	pressure = ret != ROOM_NORMAL;
1316 
1317 	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1318 		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1319 
1320 	return ret;
1321 }
1322 
1323 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1324 {
1325 	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1326 	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1327 		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1328 }
1329 
1330 static void packet_sock_destruct(struct sock *sk)
1331 {
1332 	skb_queue_purge(&sk->sk_error_queue);
1333 
1334 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1335 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1336 
1337 	if (!sock_flag(sk, SOCK_DEAD)) {
1338 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1339 		return;
1340 	}
1341 }
1342 
1343 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1344 {
1345 	u32 *history = po->rollover->history;
1346 	u32 victim, rxhash;
1347 	int i, count = 0;
1348 
1349 	rxhash = skb_get_hash(skb);
1350 	for (i = 0; i < ROLLOVER_HLEN; i++)
1351 		if (READ_ONCE(history[i]) == rxhash)
1352 			count++;
1353 
1354 	victim = get_random_u32_below(ROLLOVER_HLEN);
1355 
1356 	/* Avoid dirtying the cache line if possible */
1357 	if (READ_ONCE(history[victim]) != rxhash)
1358 		WRITE_ONCE(history[victim], rxhash);
1359 
1360 	return count > (ROLLOVER_HLEN >> 1);
1361 }
1362 
1363 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1364 				      struct sk_buff *skb,
1365 				      unsigned int num)
1366 {
1367 	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1368 }
1369 
1370 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1371 				    struct sk_buff *skb,
1372 				    unsigned int num)
1373 {
1374 	unsigned int val = atomic_inc_return(&f->rr_cur);
1375 
1376 	return val % num;
1377 }
1378 
1379 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1380 				     struct sk_buff *skb,
1381 				     unsigned int num)
1382 {
1383 	return smp_processor_id() % num;
1384 }
1385 
1386 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1387 				     struct sk_buff *skb,
1388 				     unsigned int num)
1389 {
1390 	return get_random_u32_below(num);
1391 }
1392 
1393 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1394 					  struct sk_buff *skb,
1395 					  unsigned int idx, bool try_self,
1396 					  unsigned int num)
1397 {
1398 	struct packet_sock *po, *po_next, *po_skip = NULL;
1399 	unsigned int i, j, room = ROOM_NONE;
1400 
1401 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1402 
1403 	if (try_self) {
1404 		room = packet_rcv_has_room(po, skb);
1405 		if (room == ROOM_NORMAL ||
1406 		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1407 			return idx;
1408 		po_skip = po;
1409 	}
1410 
1411 	i = j = min_t(int, po->rollover->sock, num - 1);
1412 	do {
1413 		po_next = pkt_sk(rcu_dereference(f->arr[i]));
1414 		if (po_next != po_skip &&
1415 		    !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1416 		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1417 			if (i != j)
1418 				po->rollover->sock = i;
1419 			atomic_long_inc(&po->rollover->num);
1420 			if (room == ROOM_LOW)
1421 				atomic_long_inc(&po->rollover->num_huge);
1422 			return i;
1423 		}
1424 
1425 		if (++i == num)
1426 			i = 0;
1427 	} while (i != j);
1428 
1429 	atomic_long_inc(&po->rollover->num_failed);
1430 	return idx;
1431 }
1432 
1433 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1434 				    struct sk_buff *skb,
1435 				    unsigned int num)
1436 {
1437 	return skb_get_queue_mapping(skb) % num;
1438 }
1439 
1440 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1441 				     struct sk_buff *skb,
1442 				     unsigned int num)
1443 {
1444 	struct bpf_prog *prog;
1445 	unsigned int ret = 0;
1446 
1447 	rcu_read_lock();
1448 	prog = rcu_dereference(f->bpf_prog);
1449 	if (prog)
1450 		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1451 	rcu_read_unlock();
1452 
1453 	return ret;
1454 }
1455 
1456 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1457 {
1458 	return f->flags & (flag >> 8);
1459 }
1460 
1461 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1462 			     struct packet_type *pt, struct net_device *orig_dev)
1463 {
1464 	struct packet_fanout *f = pt->af_packet_priv;
1465 	unsigned int num = READ_ONCE(f->num_members);
1466 	struct net *net = read_pnet(&f->net);
1467 	struct packet_sock *po;
1468 	unsigned int idx;
1469 
1470 	if (!net_eq(dev_net(dev), net) || !num) {
1471 		kfree_skb(skb);
1472 		return 0;
1473 	}
1474 
1475 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1476 		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1477 		if (!skb)
1478 			return 0;
1479 	}
1480 	switch (f->type) {
1481 	case PACKET_FANOUT_HASH:
1482 	default:
1483 		idx = fanout_demux_hash(f, skb, num);
1484 		break;
1485 	case PACKET_FANOUT_LB:
1486 		idx = fanout_demux_lb(f, skb, num);
1487 		break;
1488 	case PACKET_FANOUT_CPU:
1489 		idx = fanout_demux_cpu(f, skb, num);
1490 		break;
1491 	case PACKET_FANOUT_RND:
1492 		idx = fanout_demux_rnd(f, skb, num);
1493 		break;
1494 	case PACKET_FANOUT_QM:
1495 		idx = fanout_demux_qm(f, skb, num);
1496 		break;
1497 	case PACKET_FANOUT_ROLLOVER:
1498 		idx = fanout_demux_rollover(f, skb, 0, false, num);
1499 		break;
1500 	case PACKET_FANOUT_CBPF:
1501 	case PACKET_FANOUT_EBPF:
1502 		idx = fanout_demux_bpf(f, skb, num);
1503 		break;
1504 	}
1505 
1506 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1507 		idx = fanout_demux_rollover(f, skb, idx, true, num);
1508 
1509 	po = pkt_sk(rcu_dereference(f->arr[idx]));
1510 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1511 }
1512 
1513 DEFINE_MUTEX(fanout_mutex);
1514 EXPORT_SYMBOL_GPL(fanout_mutex);
1515 static LIST_HEAD(fanout_list);
1516 static u16 fanout_next_id;
1517 
1518 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1519 {
1520 	struct packet_fanout *f = po->fanout;
1521 
1522 	spin_lock(&f->lock);
1523 	rcu_assign_pointer(f->arr[f->num_members], sk);
1524 	smp_wmb();
1525 	f->num_members++;
1526 	if (f->num_members == 1)
1527 		dev_add_pack(&f->prot_hook);
1528 	spin_unlock(&f->lock);
1529 }
1530 
1531 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1532 {
1533 	struct packet_fanout *f = po->fanout;
1534 	int i;
1535 
1536 	spin_lock(&f->lock);
1537 	for (i = 0; i < f->num_members; i++) {
1538 		if (rcu_dereference_protected(f->arr[i],
1539 					      lockdep_is_held(&f->lock)) == sk)
1540 			break;
1541 	}
1542 	BUG_ON(i >= f->num_members);
1543 	rcu_assign_pointer(f->arr[i],
1544 			   rcu_dereference_protected(f->arr[f->num_members - 1],
1545 						     lockdep_is_held(&f->lock)));
1546 	f->num_members--;
1547 	if (f->num_members == 0)
1548 		__dev_remove_pack(&f->prot_hook);
1549 	spin_unlock(&f->lock);
1550 }
1551 
1552 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1553 {
1554 	if (sk->sk_family != PF_PACKET)
1555 		return false;
1556 
1557 	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1558 }
1559 
1560 static void fanout_init_data(struct packet_fanout *f)
1561 {
1562 	switch (f->type) {
1563 	case PACKET_FANOUT_LB:
1564 		atomic_set(&f->rr_cur, 0);
1565 		break;
1566 	case PACKET_FANOUT_CBPF:
1567 	case PACKET_FANOUT_EBPF:
1568 		RCU_INIT_POINTER(f->bpf_prog, NULL);
1569 		break;
1570 	}
1571 }
1572 
1573 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1574 {
1575 	struct bpf_prog *old;
1576 
1577 	spin_lock(&f->lock);
1578 	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1579 	rcu_assign_pointer(f->bpf_prog, new);
1580 	spin_unlock(&f->lock);
1581 
1582 	if (old) {
1583 		synchronize_net();
1584 		bpf_prog_destroy(old);
1585 	}
1586 }
1587 
1588 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1589 				unsigned int len)
1590 {
1591 	struct bpf_prog *new;
1592 	struct sock_fprog fprog;
1593 	int ret;
1594 
1595 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1596 		return -EPERM;
1597 
1598 	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1599 	if (ret)
1600 		return ret;
1601 
1602 	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1603 	if (ret)
1604 		return ret;
1605 
1606 	__fanout_set_data_bpf(po->fanout, new);
1607 	return 0;
1608 }
1609 
1610 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1611 				unsigned int len)
1612 {
1613 	struct bpf_prog *new;
1614 	u32 fd;
1615 
1616 	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1617 		return -EPERM;
1618 	if (len != sizeof(fd))
1619 		return -EINVAL;
1620 	if (copy_from_sockptr(&fd, data, len))
1621 		return -EFAULT;
1622 
1623 	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1624 	if (IS_ERR(new))
1625 		return PTR_ERR(new);
1626 
1627 	__fanout_set_data_bpf(po->fanout, new);
1628 	return 0;
1629 }
1630 
1631 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1632 			   unsigned int len)
1633 {
1634 	switch (po->fanout->type) {
1635 	case PACKET_FANOUT_CBPF:
1636 		return fanout_set_data_cbpf(po, data, len);
1637 	case PACKET_FANOUT_EBPF:
1638 		return fanout_set_data_ebpf(po, data, len);
1639 	default:
1640 		return -EINVAL;
1641 	}
1642 }
1643 
1644 static void fanout_release_data(struct packet_fanout *f)
1645 {
1646 	switch (f->type) {
1647 	case PACKET_FANOUT_CBPF:
1648 	case PACKET_FANOUT_EBPF:
1649 		__fanout_set_data_bpf(f, NULL);
1650 	}
1651 }
1652 
1653 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1654 {
1655 	struct packet_fanout *f;
1656 
1657 	list_for_each_entry(f, &fanout_list, list) {
1658 		if (f->id == candidate_id &&
1659 		    read_pnet(&f->net) == sock_net(sk)) {
1660 			return false;
1661 		}
1662 	}
1663 	return true;
1664 }
1665 
1666 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1667 {
1668 	u16 id = fanout_next_id;
1669 
1670 	do {
1671 		if (__fanout_id_is_free(sk, id)) {
1672 			*new_id = id;
1673 			fanout_next_id = id + 1;
1674 			return true;
1675 		}
1676 
1677 		id++;
1678 	} while (id != fanout_next_id);
1679 
1680 	return false;
1681 }
1682 
1683 static int fanout_add(struct sock *sk, struct fanout_args *args)
1684 {
1685 	struct packet_rollover *rollover = NULL;
1686 	struct packet_sock *po = pkt_sk(sk);
1687 	u16 type_flags = args->type_flags;
1688 	struct packet_fanout *f, *match;
1689 	u8 type = type_flags & 0xff;
1690 	u8 flags = type_flags >> 8;
1691 	u16 id = args->id;
1692 	int err;
1693 
1694 	switch (type) {
1695 	case PACKET_FANOUT_ROLLOVER:
1696 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1697 			return -EINVAL;
1698 		break;
1699 	case PACKET_FANOUT_HASH:
1700 	case PACKET_FANOUT_LB:
1701 	case PACKET_FANOUT_CPU:
1702 	case PACKET_FANOUT_RND:
1703 	case PACKET_FANOUT_QM:
1704 	case PACKET_FANOUT_CBPF:
1705 	case PACKET_FANOUT_EBPF:
1706 		break;
1707 	default:
1708 		return -EINVAL;
1709 	}
1710 
1711 	mutex_lock(&fanout_mutex);
1712 
1713 	err = -EALREADY;
1714 	if (po->fanout)
1715 		goto out;
1716 
1717 	if (type == PACKET_FANOUT_ROLLOVER ||
1718 	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1719 		err = -ENOMEM;
1720 		rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1721 		if (!rollover)
1722 			goto out;
1723 		atomic_long_set(&rollover->num, 0);
1724 		atomic_long_set(&rollover->num_huge, 0);
1725 		atomic_long_set(&rollover->num_failed, 0);
1726 	}
1727 
1728 	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1729 		if (id != 0) {
1730 			err = -EINVAL;
1731 			goto out;
1732 		}
1733 		if (!fanout_find_new_id(sk, &id)) {
1734 			err = -ENOMEM;
1735 			goto out;
1736 		}
1737 		/* ephemeral flag for the first socket in the group: drop it */
1738 		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1739 	}
1740 
1741 	match = NULL;
1742 	list_for_each_entry(f, &fanout_list, list) {
1743 		if (f->id == id &&
1744 		    read_pnet(&f->net) == sock_net(sk)) {
1745 			match = f;
1746 			break;
1747 		}
1748 	}
1749 	err = -EINVAL;
1750 	if (match) {
1751 		if (match->flags != flags)
1752 			goto out;
1753 		if (args->max_num_members &&
1754 		    args->max_num_members != match->max_num_members)
1755 			goto out;
1756 	} else {
1757 		if (args->max_num_members > PACKET_FANOUT_MAX)
1758 			goto out;
1759 		if (!args->max_num_members)
1760 			/* legacy PACKET_FANOUT_MAX */
1761 			args->max_num_members = 256;
1762 		err = -ENOMEM;
1763 		match = kvzalloc(struct_size(match, arr, args->max_num_members),
1764 				 GFP_KERNEL);
1765 		if (!match)
1766 			goto out;
1767 		write_pnet(&match->net, sock_net(sk));
1768 		match->id = id;
1769 		match->type = type;
1770 		match->flags = flags;
1771 		INIT_LIST_HEAD(&match->list);
1772 		spin_lock_init(&match->lock);
1773 		refcount_set(&match->sk_ref, 0);
1774 		fanout_init_data(match);
1775 		match->prot_hook.type = po->prot_hook.type;
1776 		match->prot_hook.dev = po->prot_hook.dev;
1777 		match->prot_hook.func = packet_rcv_fanout;
1778 		match->prot_hook.af_packet_priv = match;
1779 		match->prot_hook.af_packet_net = read_pnet(&match->net);
1780 		match->prot_hook.id_match = match_fanout_group;
1781 		match->max_num_members = args->max_num_members;
1782 		match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1783 		list_add(&match->list, &fanout_list);
1784 	}
1785 	err = -EINVAL;
1786 
1787 	spin_lock(&po->bind_lock);
1788 	if (packet_sock_flag(po, PACKET_SOCK_RUNNING) &&
1789 	    match->type == type &&
1790 	    match->prot_hook.type == po->prot_hook.type &&
1791 	    match->prot_hook.dev == po->prot_hook.dev) {
1792 		err = -ENOSPC;
1793 		if (refcount_read(&match->sk_ref) < match->max_num_members) {
1794 			__dev_remove_pack(&po->prot_hook);
1795 
1796 			/* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1797 			WRITE_ONCE(po->fanout, match);
1798 
1799 			po->rollover = rollover;
1800 			rollover = NULL;
1801 			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1802 			__fanout_link(sk, po);
1803 			err = 0;
1804 		}
1805 	}
1806 	spin_unlock(&po->bind_lock);
1807 
1808 	if (err && !refcount_read(&match->sk_ref)) {
1809 		list_del(&match->list);
1810 		kvfree(match);
1811 	}
1812 
1813 out:
1814 	kfree(rollover);
1815 	mutex_unlock(&fanout_mutex);
1816 	return err;
1817 }
1818 
1819 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1820  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1821  * It is the responsibility of the caller to call fanout_release_data() and
1822  * free the returned packet_fanout (after synchronize_net())
1823  */
1824 static struct packet_fanout *fanout_release(struct sock *sk)
1825 {
1826 	struct packet_sock *po = pkt_sk(sk);
1827 	struct packet_fanout *f;
1828 
1829 	mutex_lock(&fanout_mutex);
1830 	f = po->fanout;
1831 	if (f) {
1832 		po->fanout = NULL;
1833 
1834 		if (refcount_dec_and_test(&f->sk_ref))
1835 			list_del(&f->list);
1836 		else
1837 			f = NULL;
1838 	}
1839 	mutex_unlock(&fanout_mutex);
1840 
1841 	return f;
1842 }
1843 
1844 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1845 					  struct sk_buff *skb)
1846 {
1847 	/* Earlier code assumed this would be a VLAN pkt, double-check
1848 	 * this now that we have the actual packet in hand. We can only
1849 	 * do this check on Ethernet devices.
1850 	 */
1851 	if (unlikely(dev->type != ARPHRD_ETHER))
1852 		return false;
1853 
1854 	skb_reset_mac_header(skb);
1855 	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1856 }
1857 
1858 static const struct proto_ops packet_ops;
1859 
1860 static const struct proto_ops packet_ops_spkt;
1861 
1862 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1863 			   struct packet_type *pt, struct net_device *orig_dev)
1864 {
1865 	struct sock *sk;
1866 	struct sockaddr_pkt *spkt;
1867 
1868 	/*
1869 	 *	When we registered the protocol we saved the socket in the data
1870 	 *	field for just this event.
1871 	 */
1872 
1873 	sk = pt->af_packet_priv;
1874 
1875 	/*
1876 	 *	Yank back the headers [hope the device set this
1877 	 *	right or kerboom...]
1878 	 *
1879 	 *	Incoming packets have ll header pulled,
1880 	 *	push it back.
1881 	 *
1882 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1883 	 *	so that this procedure is noop.
1884 	 */
1885 
1886 	if (skb->pkt_type == PACKET_LOOPBACK)
1887 		goto out;
1888 
1889 	if (!net_eq(dev_net(dev), sock_net(sk)))
1890 		goto out;
1891 
1892 	skb = skb_share_check(skb, GFP_ATOMIC);
1893 	if (skb == NULL)
1894 		goto oom;
1895 
1896 	/* drop any routing info */
1897 	skb_dst_drop(skb);
1898 
1899 	/* drop conntrack reference */
1900 	nf_reset_ct(skb);
1901 
1902 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1903 
1904 	skb_push(skb, skb->data - skb_mac_header(skb));
1905 
1906 	/*
1907 	 *	The SOCK_PACKET socket receives _all_ frames.
1908 	 */
1909 
1910 	spkt->spkt_family = dev->type;
1911 	strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1912 	spkt->spkt_protocol = skb->protocol;
1913 
1914 	/*
1915 	 *	Charge the memory to the socket. This is done specifically
1916 	 *	to prevent sockets using all the memory up.
1917 	 */
1918 
1919 	if (sock_queue_rcv_skb(sk, skb) == 0)
1920 		return 0;
1921 
1922 out:
1923 	kfree_skb(skb);
1924 oom:
1925 	return 0;
1926 }
1927 
1928 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1929 {
1930 	int depth;
1931 
1932 	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1933 	    sock->type == SOCK_RAW) {
1934 		skb_reset_mac_header(skb);
1935 		skb->protocol = dev_parse_header_protocol(skb);
1936 	}
1937 
1938 	/* Move network header to the right position for VLAN tagged packets */
1939 	if (likely(skb->dev->type == ARPHRD_ETHER) &&
1940 	    eth_type_vlan(skb->protocol) &&
1941 	    vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1942 		skb_set_network_header(skb, depth);
1943 
1944 	skb_probe_transport_header(skb);
1945 }
1946 
1947 /*
1948  *	Output a raw packet to a device layer. This bypasses all the other
1949  *	protocol layers and you must therefore supply it with a complete frame
1950  */
1951 
1952 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1953 			       size_t len)
1954 {
1955 	struct sock *sk = sock->sk;
1956 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1957 	struct sk_buff *skb = NULL;
1958 	struct net_device *dev;
1959 	struct sockcm_cookie sockc;
1960 	__be16 proto = 0;
1961 	int err;
1962 	int extra_len = 0;
1963 
1964 	/*
1965 	 *	Get and verify the address.
1966 	 */
1967 
1968 	if (saddr) {
1969 		if (msg->msg_namelen < sizeof(struct sockaddr))
1970 			return -EINVAL;
1971 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1972 			proto = saddr->spkt_protocol;
1973 	} else
1974 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1975 
1976 	/*
1977 	 *	Find the device first to size check it
1978 	 */
1979 
1980 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1981 retry:
1982 	rcu_read_lock();
1983 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1984 	err = -ENODEV;
1985 	if (dev == NULL)
1986 		goto out_unlock;
1987 
1988 	err = -ENETDOWN;
1989 	if (!(dev->flags & IFF_UP))
1990 		goto out_unlock;
1991 
1992 	/*
1993 	 * You may not queue a frame bigger than the mtu. This is the lowest level
1994 	 * raw protocol and you must do your own fragmentation at this level.
1995 	 */
1996 
1997 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1998 		if (!netif_supports_nofcs(dev)) {
1999 			err = -EPROTONOSUPPORT;
2000 			goto out_unlock;
2001 		}
2002 		extra_len = 4; /* We're doing our own CRC */
2003 	}
2004 
2005 	err = -EMSGSIZE;
2006 	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2007 		goto out_unlock;
2008 
2009 	if (!skb) {
2010 		size_t reserved = LL_RESERVED_SPACE(dev);
2011 		int tlen = dev->needed_tailroom;
2012 		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2013 
2014 		rcu_read_unlock();
2015 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2016 		if (skb == NULL)
2017 			return -ENOBUFS;
2018 		/* FIXME: Save some space for broken drivers that write a hard
2019 		 * header at transmission time by themselves. PPP is the notable
2020 		 * one here. This should really be fixed at the driver level.
2021 		 */
2022 		skb_reserve(skb, reserved);
2023 		skb_reset_network_header(skb);
2024 
2025 		/* Try to align data part correctly */
2026 		if (hhlen) {
2027 			skb->data -= hhlen;
2028 			skb->tail -= hhlen;
2029 			if (len < hhlen)
2030 				skb_reset_network_header(skb);
2031 		}
2032 		err = memcpy_from_msg(skb_put(skb, len), msg, len);
2033 		if (err)
2034 			goto out_free;
2035 		goto retry;
2036 	}
2037 
2038 	if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2039 		err = -EINVAL;
2040 		goto out_unlock;
2041 	}
2042 	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2043 	    !packet_extra_vlan_len_allowed(dev, skb)) {
2044 		err = -EMSGSIZE;
2045 		goto out_unlock;
2046 	}
2047 
2048 	sockcm_init(&sockc, sk);
2049 	if (msg->msg_controllen) {
2050 		err = sock_cmsg_send(sk, msg, &sockc);
2051 		if (unlikely(err))
2052 			goto out_unlock;
2053 	}
2054 
2055 	skb->protocol = proto;
2056 	skb->dev = dev;
2057 	skb->priority = READ_ONCE(sk->sk_priority);
2058 	skb->mark = READ_ONCE(sk->sk_mark);
2059 	skb->tstamp = sockc.transmit_time;
2060 
2061 	skb_setup_tx_timestamp(skb, sockc.tsflags);
2062 
2063 	if (unlikely(extra_len == 4))
2064 		skb->no_fcs = 1;
2065 
2066 	packet_parse_headers(skb, sock);
2067 
2068 	dev_queue_xmit(skb);
2069 	rcu_read_unlock();
2070 	return len;
2071 
2072 out_unlock:
2073 	rcu_read_unlock();
2074 out_free:
2075 	kfree_skb(skb);
2076 	return err;
2077 }
2078 
2079 static unsigned int run_filter(struct sk_buff *skb,
2080 			       const struct sock *sk,
2081 			       unsigned int res)
2082 {
2083 	struct sk_filter *filter;
2084 
2085 	rcu_read_lock();
2086 	filter = rcu_dereference(sk->sk_filter);
2087 	if (filter != NULL)
2088 		res = bpf_prog_run_clear_cb(filter->prog, skb);
2089 	rcu_read_unlock();
2090 
2091 	return res;
2092 }
2093 
2094 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2095 			   size_t *len, int vnet_hdr_sz)
2096 {
2097 	struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2098 
2099 	if (*len < vnet_hdr_sz)
2100 		return -EINVAL;
2101 	*len -= vnet_hdr_sz;
2102 
2103 	if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2104 		return -EINVAL;
2105 
2106 	return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2107 }
2108 
2109 /*
2110  * This function makes lazy skb cloning in hope that most of packets
2111  * are discarded by BPF.
2112  *
2113  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2114  * and skb->cb are mangled. It works because (and until) packets
2115  * falling here are owned by current CPU. Output packets are cloned
2116  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2117  * sequentially, so that if we return skb to original state on exit,
2118  * we will not harm anyone.
2119  */
2120 
2121 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2122 		      struct packet_type *pt, struct net_device *orig_dev)
2123 {
2124 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2125 	struct sock *sk;
2126 	struct sockaddr_ll *sll;
2127 	struct packet_sock *po;
2128 	u8 *skb_head = skb->data;
2129 	int skb_len = skb->len;
2130 	unsigned int snaplen, res;
2131 
2132 	if (skb->pkt_type == PACKET_LOOPBACK)
2133 		goto drop;
2134 
2135 	sk = pt->af_packet_priv;
2136 	po = pkt_sk(sk);
2137 
2138 	if (!net_eq(dev_net(dev), sock_net(sk)))
2139 		goto drop;
2140 
2141 	skb->dev = dev;
2142 
2143 	if (dev_has_header(dev)) {
2144 		/* The device has an explicit notion of ll header,
2145 		 * exported to higher levels.
2146 		 *
2147 		 * Otherwise, the device hides details of its frame
2148 		 * structure, so that corresponding packet head is
2149 		 * never delivered to user.
2150 		 */
2151 		if (sk->sk_type != SOCK_DGRAM)
2152 			skb_push(skb, skb->data - skb_mac_header(skb));
2153 		else if (skb->pkt_type == PACKET_OUTGOING) {
2154 			/* Special case: outgoing packets have ll header at head */
2155 			skb_pull(skb, skb_network_offset(skb));
2156 		}
2157 	}
2158 
2159 	snaplen = skb->len;
2160 
2161 	res = run_filter(skb, sk, snaplen);
2162 	if (!res)
2163 		goto drop_n_restore;
2164 	if (snaplen > res)
2165 		snaplen = res;
2166 
2167 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2168 		goto drop_n_acct;
2169 
2170 	if (skb_shared(skb)) {
2171 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2172 		if (nskb == NULL)
2173 			goto drop_n_acct;
2174 
2175 		if (skb_head != skb->data) {
2176 			skb->data = skb_head;
2177 			skb->len = skb_len;
2178 		}
2179 		consume_skb(skb);
2180 		skb = nskb;
2181 	}
2182 
2183 	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2184 
2185 	sll = &PACKET_SKB_CB(skb)->sa.ll;
2186 	sll->sll_hatype = dev->type;
2187 	sll->sll_pkttype = skb->pkt_type;
2188 	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2189 		sll->sll_ifindex = orig_dev->ifindex;
2190 	else
2191 		sll->sll_ifindex = dev->ifindex;
2192 
2193 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2194 
2195 	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2196 	 * Use their space for storing the original skb length.
2197 	 */
2198 	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2199 
2200 	if (pskb_trim(skb, snaplen))
2201 		goto drop_n_acct;
2202 
2203 	skb_set_owner_r(skb, sk);
2204 	skb->dev = NULL;
2205 	skb_dst_drop(skb);
2206 
2207 	/* drop conntrack reference */
2208 	nf_reset_ct(skb);
2209 
2210 	spin_lock(&sk->sk_receive_queue.lock);
2211 	po->stats.stats1.tp_packets++;
2212 	sock_skb_set_dropcount(sk, skb);
2213 	skb_clear_delivery_time(skb);
2214 	__skb_queue_tail(&sk->sk_receive_queue, skb);
2215 	spin_unlock(&sk->sk_receive_queue.lock);
2216 	sk->sk_data_ready(sk);
2217 	return 0;
2218 
2219 drop_n_acct:
2220 	atomic_inc(&po->tp_drops);
2221 	atomic_inc(&sk->sk_drops);
2222 	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2223 
2224 drop_n_restore:
2225 	if (skb_head != skb->data && skb_shared(skb)) {
2226 		skb->data = skb_head;
2227 		skb->len = skb_len;
2228 	}
2229 drop:
2230 	kfree_skb_reason(skb, drop_reason);
2231 	return 0;
2232 }
2233 
2234 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2235 		       struct packet_type *pt, struct net_device *orig_dev)
2236 {
2237 	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2238 	struct sock *sk;
2239 	struct packet_sock *po;
2240 	struct sockaddr_ll *sll;
2241 	union tpacket_uhdr h;
2242 	u8 *skb_head = skb->data;
2243 	int skb_len = skb->len;
2244 	unsigned int snaplen, res;
2245 	unsigned long status = TP_STATUS_USER;
2246 	unsigned short macoff, hdrlen;
2247 	unsigned int netoff;
2248 	struct sk_buff *copy_skb = NULL;
2249 	struct timespec64 ts;
2250 	__u32 ts_status;
2251 	unsigned int slot_id = 0;
2252 	int vnet_hdr_sz = 0;
2253 
2254 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2255 	 * We may add members to them until current aligned size without forcing
2256 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2257 	 */
2258 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2259 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2260 
2261 	if (skb->pkt_type == PACKET_LOOPBACK)
2262 		goto drop;
2263 
2264 	sk = pt->af_packet_priv;
2265 	po = pkt_sk(sk);
2266 
2267 	if (!net_eq(dev_net(dev), sock_net(sk)))
2268 		goto drop;
2269 
2270 	if (dev_has_header(dev)) {
2271 		if (sk->sk_type != SOCK_DGRAM)
2272 			skb_push(skb, skb->data - skb_mac_header(skb));
2273 		else if (skb->pkt_type == PACKET_OUTGOING) {
2274 			/* Special case: outgoing packets have ll header at head */
2275 			skb_pull(skb, skb_network_offset(skb));
2276 		}
2277 	}
2278 
2279 	snaplen = skb->len;
2280 
2281 	res = run_filter(skb, sk, snaplen);
2282 	if (!res)
2283 		goto drop_n_restore;
2284 
2285 	/* If we are flooded, just give up */
2286 	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2287 		atomic_inc(&po->tp_drops);
2288 		goto drop_n_restore;
2289 	}
2290 
2291 	if (skb->ip_summed == CHECKSUM_PARTIAL)
2292 		status |= TP_STATUS_CSUMNOTREADY;
2293 	else if (skb->pkt_type != PACKET_OUTGOING &&
2294 		 skb_csum_unnecessary(skb))
2295 		status |= TP_STATUS_CSUM_VALID;
2296 	if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2297 		status |= TP_STATUS_GSO_TCP;
2298 
2299 	if (snaplen > res)
2300 		snaplen = res;
2301 
2302 	if (sk->sk_type == SOCK_DGRAM) {
2303 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2304 				  po->tp_reserve;
2305 	} else {
2306 		unsigned int maclen = skb_network_offset(skb);
2307 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2308 				       (maclen < 16 ? 16 : maclen)) +
2309 				       po->tp_reserve;
2310 		vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2311 		if (vnet_hdr_sz)
2312 			netoff += vnet_hdr_sz;
2313 		macoff = netoff - maclen;
2314 	}
2315 	if (netoff > USHRT_MAX) {
2316 		atomic_inc(&po->tp_drops);
2317 		goto drop_n_restore;
2318 	}
2319 	if (po->tp_version <= TPACKET_V2) {
2320 		if (macoff + snaplen > po->rx_ring.frame_size) {
2321 			if (READ_ONCE(po->copy_thresh) &&
2322 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2323 				if (skb_shared(skb)) {
2324 					copy_skb = skb_clone(skb, GFP_ATOMIC);
2325 				} else {
2326 					copy_skb = skb_get(skb);
2327 					skb_head = skb->data;
2328 				}
2329 				if (copy_skb) {
2330 					memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2331 					       sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2332 					skb_set_owner_r(copy_skb, sk);
2333 				}
2334 			}
2335 			snaplen = po->rx_ring.frame_size - macoff;
2336 			if ((int)snaplen < 0) {
2337 				snaplen = 0;
2338 				vnet_hdr_sz = 0;
2339 			}
2340 		}
2341 	} else if (unlikely(macoff + snaplen >
2342 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2343 		u32 nval;
2344 
2345 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2346 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2347 			    snaplen, nval, macoff);
2348 		snaplen = nval;
2349 		if (unlikely((int)snaplen < 0)) {
2350 			snaplen = 0;
2351 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2352 			vnet_hdr_sz = 0;
2353 		}
2354 	}
2355 	spin_lock(&sk->sk_receive_queue.lock);
2356 	h.raw = packet_current_rx_frame(po, skb,
2357 					TP_STATUS_KERNEL, (macoff+snaplen));
2358 	if (!h.raw)
2359 		goto drop_n_account;
2360 
2361 	if (po->tp_version <= TPACKET_V2) {
2362 		slot_id = po->rx_ring.head;
2363 		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2364 			goto drop_n_account;
2365 		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2366 	}
2367 
2368 	if (vnet_hdr_sz &&
2369 	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2370 				    sizeof(struct virtio_net_hdr),
2371 				    vio_le(), true, 0)) {
2372 		if (po->tp_version == TPACKET_V3)
2373 			prb_clear_blk_fill_status(&po->rx_ring);
2374 		goto drop_n_account;
2375 	}
2376 
2377 	if (po->tp_version <= TPACKET_V2) {
2378 		packet_increment_rx_head(po, &po->rx_ring);
2379 	/*
2380 	 * LOSING will be reported till you read the stats,
2381 	 * because it's COR - Clear On Read.
2382 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2383 	 * at packet level.
2384 	 */
2385 		if (atomic_read(&po->tp_drops))
2386 			status |= TP_STATUS_LOSING;
2387 	}
2388 
2389 	po->stats.stats1.tp_packets++;
2390 	if (copy_skb) {
2391 		status |= TP_STATUS_COPY;
2392 		skb_clear_delivery_time(copy_skb);
2393 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2394 	}
2395 	spin_unlock(&sk->sk_receive_queue.lock);
2396 
2397 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2398 
2399 	/* Always timestamp; prefer an existing software timestamp taken
2400 	 * closer to the time of capture.
2401 	 */
2402 	ts_status = tpacket_get_timestamp(skb, &ts,
2403 					  READ_ONCE(po->tp_tstamp) |
2404 					  SOF_TIMESTAMPING_SOFTWARE);
2405 	if (!ts_status)
2406 		ktime_get_real_ts64(&ts);
2407 
2408 	status |= ts_status;
2409 
2410 	switch (po->tp_version) {
2411 	case TPACKET_V1:
2412 		h.h1->tp_len = skb->len;
2413 		h.h1->tp_snaplen = snaplen;
2414 		h.h1->tp_mac = macoff;
2415 		h.h1->tp_net = netoff;
2416 		h.h1->tp_sec = ts.tv_sec;
2417 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2418 		hdrlen = sizeof(*h.h1);
2419 		break;
2420 	case TPACKET_V2:
2421 		h.h2->tp_len = skb->len;
2422 		h.h2->tp_snaplen = snaplen;
2423 		h.h2->tp_mac = macoff;
2424 		h.h2->tp_net = netoff;
2425 		h.h2->tp_sec = ts.tv_sec;
2426 		h.h2->tp_nsec = ts.tv_nsec;
2427 		if (skb_vlan_tag_present(skb)) {
2428 			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2429 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2430 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2431 		} else {
2432 			h.h2->tp_vlan_tci = 0;
2433 			h.h2->tp_vlan_tpid = 0;
2434 		}
2435 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2436 		hdrlen = sizeof(*h.h2);
2437 		break;
2438 	case TPACKET_V3:
2439 		/* tp_nxt_offset,vlan are already populated above.
2440 		 * So DONT clear those fields here
2441 		 */
2442 		h.h3->tp_status |= status;
2443 		h.h3->tp_len = skb->len;
2444 		h.h3->tp_snaplen = snaplen;
2445 		h.h3->tp_mac = macoff;
2446 		h.h3->tp_net = netoff;
2447 		h.h3->tp_sec  = ts.tv_sec;
2448 		h.h3->tp_nsec = ts.tv_nsec;
2449 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2450 		hdrlen = sizeof(*h.h3);
2451 		break;
2452 	default:
2453 		BUG();
2454 	}
2455 
2456 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2457 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2458 	sll->sll_family = AF_PACKET;
2459 	sll->sll_hatype = dev->type;
2460 	sll->sll_protocol = 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 	kfree_skb_reason(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 	kfree_skb_reason(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 = READ_ONCE(po->sk.sk_priority);
2586 	skb->mark = READ_ONCE(po->sk.sk_mark);
2587 	skb->tstamp = sockc->transmit_time;
2588 	skb_setup_tx_timestamp(skb, sockc->tsflags);
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 = NULL;
2723 	struct sockcm_cookie sockc;
2724 	__be16 proto;
2725 	int err, reserve = 0;
2726 	void *ph;
2727 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2728 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2729 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2730 	unsigned char *addr = NULL;
2731 	int tp_len, size_max;
2732 	void *data;
2733 	int len_sum = 0;
2734 	int status = TP_STATUS_AVAILABLE;
2735 	int hlen, tlen, copylen = 0;
2736 	long timeo = 0;
2737 
2738 	mutex_lock(&po->pg_vec_lock);
2739 
2740 	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2741 	 * we need to confirm it under protection of pg_vec_lock.
2742 	 */
2743 	if (unlikely(!po->tx_ring.pg_vec)) {
2744 		err = -EBUSY;
2745 		goto out;
2746 	}
2747 	if (likely(saddr == NULL)) {
2748 		dev	= packet_cached_dev_get(po);
2749 		proto	= READ_ONCE(po->num);
2750 	} else {
2751 		err = -EINVAL;
2752 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2753 			goto out;
2754 		if (msg->msg_namelen < (saddr->sll_halen
2755 					+ offsetof(struct sockaddr_ll,
2756 						sll_addr)))
2757 			goto out;
2758 		proto	= saddr->sll_protocol;
2759 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2760 		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2761 			if (dev && msg->msg_namelen < dev->addr_len +
2762 				   offsetof(struct sockaddr_ll, sll_addr))
2763 				goto out_put;
2764 			addr = saddr->sll_addr;
2765 		}
2766 	}
2767 
2768 	err = -ENXIO;
2769 	if (unlikely(dev == NULL))
2770 		goto out;
2771 	err = -ENETDOWN;
2772 	if (unlikely(!(dev->flags & IFF_UP)))
2773 		goto out_put;
2774 
2775 	sockcm_init(&sockc, &po->sk);
2776 	if (msg->msg_controllen) {
2777 		err = sock_cmsg_send(&po->sk, msg, &sockc);
2778 		if (unlikely(err))
2779 			goto out_put;
2780 	}
2781 
2782 	if (po->sk.sk_socket->type == SOCK_RAW)
2783 		reserve = dev->hard_header_len;
2784 	size_max = po->tx_ring.frame_size
2785 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2786 
2787 	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2788 		size_max = dev->mtu + reserve + VLAN_HLEN;
2789 
2790 	reinit_completion(&po->skb_completion);
2791 
2792 	do {
2793 		ph = packet_current_frame(po, &po->tx_ring,
2794 					  TP_STATUS_SEND_REQUEST);
2795 		if (unlikely(ph == NULL)) {
2796 			if (need_wait && skb) {
2797 				timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2798 				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2799 				if (timeo <= 0) {
2800 					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2801 					goto out_put;
2802 				}
2803 			}
2804 			/* check for additional frames */
2805 			continue;
2806 		}
2807 
2808 		skb = NULL;
2809 		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2810 		if (tp_len < 0)
2811 			goto tpacket_error;
2812 
2813 		status = TP_STATUS_SEND_REQUEST;
2814 		hlen = LL_RESERVED_SPACE(dev);
2815 		tlen = dev->needed_tailroom;
2816 		if (vnet_hdr_sz) {
2817 			vnet_hdr = data;
2818 			data += vnet_hdr_sz;
2819 			tp_len -= vnet_hdr_sz;
2820 			if (tp_len < 0 ||
2821 			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2822 				tp_len = -EINVAL;
2823 				goto tpacket_error;
2824 			}
2825 			copylen = __virtio16_to_cpu(vio_le(),
2826 						    vnet_hdr->hdr_len);
2827 		}
2828 		copylen = max_t(int, copylen, dev->hard_header_len);
2829 		skb = sock_alloc_send_skb(&po->sk,
2830 				hlen + tlen + sizeof(struct sockaddr_ll) +
2831 				(copylen - dev->hard_header_len),
2832 				!need_wait, &err);
2833 
2834 		if (unlikely(skb == NULL)) {
2835 			/* we assume the socket was initially writeable ... */
2836 			if (likely(len_sum > 0))
2837 				err = len_sum;
2838 			goto out_status;
2839 		}
2840 		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2841 					  addr, hlen, copylen, &sockc);
2842 		if (likely(tp_len >= 0) &&
2843 		    tp_len > dev->mtu + reserve &&
2844 		    !vnet_hdr_sz &&
2845 		    !packet_extra_vlan_len_allowed(dev, skb))
2846 			tp_len = -EMSGSIZE;
2847 
2848 		if (unlikely(tp_len < 0)) {
2849 tpacket_error:
2850 			if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2851 				__packet_set_status(po, ph,
2852 						TP_STATUS_AVAILABLE);
2853 				packet_increment_head(&po->tx_ring);
2854 				kfree_skb(skb);
2855 				continue;
2856 			} else {
2857 				status = TP_STATUS_WRONG_FORMAT;
2858 				err = tp_len;
2859 				goto out_status;
2860 			}
2861 		}
2862 
2863 		if (vnet_hdr_sz) {
2864 			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2865 				tp_len = -EINVAL;
2866 				goto tpacket_error;
2867 			}
2868 			virtio_net_hdr_set_proto(skb, vnet_hdr);
2869 		}
2870 
2871 		skb->destructor = tpacket_destruct_skb;
2872 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2873 		packet_inc_pending(&po->tx_ring);
2874 
2875 		status = TP_STATUS_SEND_REQUEST;
2876 		err = packet_xmit(po, skb);
2877 		if (unlikely(err != 0)) {
2878 			if (err > 0)
2879 				err = net_xmit_errno(err);
2880 			if (err && __packet_get_status(po, ph) ==
2881 				   TP_STATUS_AVAILABLE) {
2882 				/* skb was destructed already */
2883 				skb = NULL;
2884 				goto out_status;
2885 			}
2886 			/*
2887 			 * skb was dropped but not destructed yet;
2888 			 * let's treat it like congestion or err < 0
2889 			 */
2890 			err = 0;
2891 		}
2892 		packet_increment_head(&po->tx_ring);
2893 		len_sum += tp_len;
2894 	} while (likely((ph != NULL) ||
2895 		/* Note: packet_read_pending() might be slow if we have
2896 		 * to call it as it's per_cpu variable, but in fast-path
2897 		 * we already short-circuit the loop with the first
2898 		 * condition, and luckily don't have to go that path
2899 		 * anyway.
2900 		 */
2901 		 (need_wait && packet_read_pending(&po->tx_ring))));
2902 
2903 	err = len_sum;
2904 	goto out_put;
2905 
2906 out_status:
2907 	__packet_set_status(po, ph, status);
2908 	kfree_skb(skb);
2909 out_put:
2910 	dev_put(dev);
2911 out:
2912 	mutex_unlock(&po->pg_vec_lock);
2913 	return err;
2914 }
2915 
2916 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2917 				        size_t reserve, size_t len,
2918 				        size_t linear, int noblock,
2919 				        int *err)
2920 {
2921 	struct sk_buff *skb;
2922 
2923 	/* Under a page?  Don't bother with paged skb. */
2924 	if (prepad + len < PAGE_SIZE || !linear)
2925 		linear = len;
2926 
2927 	if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2928 		linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2929 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2930 				   err, PAGE_ALLOC_COSTLY_ORDER);
2931 	if (!skb)
2932 		return NULL;
2933 
2934 	skb_reserve(skb, reserve);
2935 	skb_put(skb, linear);
2936 	skb->data_len = len - linear;
2937 	skb->len += len - linear;
2938 
2939 	return skb;
2940 }
2941 
2942 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2943 {
2944 	struct sock *sk = sock->sk;
2945 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2946 	struct sk_buff *skb;
2947 	struct net_device *dev;
2948 	__be16 proto;
2949 	unsigned char *addr = NULL;
2950 	int err, reserve = 0;
2951 	struct sockcm_cookie sockc;
2952 	struct virtio_net_hdr vnet_hdr = { 0 };
2953 	int offset = 0;
2954 	struct packet_sock *po = pkt_sk(sk);
2955 	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2956 	int hlen, tlen, linear;
2957 	int extra_len = 0;
2958 
2959 	/*
2960 	 *	Get and verify the address.
2961 	 */
2962 
2963 	if (likely(saddr == NULL)) {
2964 		dev	= packet_cached_dev_get(po);
2965 		proto	= READ_ONCE(po->num);
2966 	} else {
2967 		err = -EINVAL;
2968 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2969 			goto out;
2970 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2971 			goto out;
2972 		proto	= saddr->sll_protocol;
2973 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2974 		if (sock->type == SOCK_DGRAM) {
2975 			if (dev && msg->msg_namelen < dev->addr_len +
2976 				   offsetof(struct sockaddr_ll, sll_addr))
2977 				goto out_unlock;
2978 			addr = saddr->sll_addr;
2979 		}
2980 	}
2981 
2982 	err = -ENXIO;
2983 	if (unlikely(dev == NULL))
2984 		goto out_unlock;
2985 	err = -ENETDOWN;
2986 	if (unlikely(!(dev->flags & IFF_UP)))
2987 		goto out_unlock;
2988 
2989 	sockcm_init(&sockc, sk);
2990 	sockc.mark = READ_ONCE(sk->sk_mark);
2991 	if (msg->msg_controllen) {
2992 		err = sock_cmsg_send(sk, msg, &sockc);
2993 		if (unlikely(err))
2994 			goto out_unlock;
2995 	}
2996 
2997 	if (sock->type == SOCK_RAW)
2998 		reserve = dev->hard_header_len;
2999 	if (vnet_hdr_sz) {
3000 		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3001 		if (err)
3002 			goto out_unlock;
3003 	}
3004 
3005 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3006 		if (!netif_supports_nofcs(dev)) {
3007 			err = -EPROTONOSUPPORT;
3008 			goto out_unlock;
3009 		}
3010 		extra_len = 4; /* We're doing our own CRC */
3011 	}
3012 
3013 	err = -EMSGSIZE;
3014 	if (!vnet_hdr.gso_type &&
3015 	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3016 		goto out_unlock;
3017 
3018 	err = -ENOBUFS;
3019 	hlen = LL_RESERVED_SPACE(dev);
3020 	tlen = dev->needed_tailroom;
3021 	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3022 	linear = max(linear, min_t(int, len, dev->hard_header_len));
3023 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3024 			       msg->msg_flags & MSG_DONTWAIT, &err);
3025 	if (skb == NULL)
3026 		goto out_unlock;
3027 
3028 	skb_reset_network_header(skb);
3029 
3030 	err = -EINVAL;
3031 	if (sock->type == SOCK_DGRAM) {
3032 		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3033 		if (unlikely(offset < 0))
3034 			goto out_free;
3035 	} else if (reserve) {
3036 		skb_reserve(skb, -reserve);
3037 		if (len < reserve + sizeof(struct ipv6hdr) &&
3038 		    dev->min_header_len != dev->hard_header_len)
3039 			skb_reset_network_header(skb);
3040 	}
3041 
3042 	/* Returns -EFAULT on error */
3043 	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3044 	if (err)
3045 		goto out_free;
3046 
3047 	if ((sock->type == SOCK_RAW &&
3048 	     !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3049 		err = -EINVAL;
3050 		goto out_free;
3051 	}
3052 
3053 	skb_setup_tx_timestamp(skb, sockc.tsflags);
3054 
3055 	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3056 	    !packet_extra_vlan_len_allowed(dev, skb)) {
3057 		err = -EMSGSIZE;
3058 		goto out_free;
3059 	}
3060 
3061 	skb->protocol = proto;
3062 	skb->dev = dev;
3063 	skb->priority = READ_ONCE(sk->sk_priority);
3064 	skb->mark = sockc.mark;
3065 	skb->tstamp = sockc.transmit_time;
3066 
3067 	if (unlikely(extra_len == 4))
3068 		skb->no_fcs = 1;
3069 
3070 	packet_parse_headers(skb, sock);
3071 
3072 	if (vnet_hdr_sz) {
3073 		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3074 		if (err)
3075 			goto out_free;
3076 		len += vnet_hdr_sz;
3077 		virtio_net_hdr_set_proto(skb, &vnet_hdr);
3078 	}
3079 
3080 	err = packet_xmit(po, skb);
3081 
3082 	if (unlikely(err != 0)) {
3083 		if (err > 0)
3084 			err = net_xmit_errno(err);
3085 		if (err)
3086 			goto out_unlock;
3087 	}
3088 
3089 	dev_put(dev);
3090 
3091 	return len;
3092 
3093 out_free:
3094 	kfree_skb(skb);
3095 out_unlock:
3096 	dev_put(dev);
3097 out:
3098 	return err;
3099 }
3100 
3101 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3102 {
3103 	struct sock *sk = sock->sk;
3104 	struct packet_sock *po = pkt_sk(sk);
3105 
3106 	/* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3107 	 * tpacket_snd() will redo the check safely.
3108 	 */
3109 	if (data_race(po->tx_ring.pg_vec))
3110 		return tpacket_snd(po, msg);
3111 
3112 	return packet_snd(sock, msg, len);
3113 }
3114 
3115 /*
3116  *	Close a PACKET socket. This is fairly simple. We immediately go
3117  *	to 'closed' state and remove our protocol entry in the device list.
3118  */
3119 
3120 static int packet_release(struct socket *sock)
3121 {
3122 	struct sock *sk = sock->sk;
3123 	struct packet_sock *po;
3124 	struct packet_fanout *f;
3125 	struct net *net;
3126 	union tpacket_req_u req_u;
3127 
3128 	if (!sk)
3129 		return 0;
3130 
3131 	net = sock_net(sk);
3132 	po = pkt_sk(sk);
3133 
3134 	mutex_lock(&net->packet.sklist_lock);
3135 	sk_del_node_init_rcu(sk);
3136 	mutex_unlock(&net->packet.sklist_lock);
3137 
3138 	sock_prot_inuse_add(net, sk->sk_prot, -1);
3139 
3140 	spin_lock(&po->bind_lock);
3141 	unregister_prot_hook(sk, false);
3142 	packet_cached_dev_reset(po);
3143 
3144 	if (po->prot_hook.dev) {
3145 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3146 		po->prot_hook.dev = NULL;
3147 	}
3148 	spin_unlock(&po->bind_lock);
3149 
3150 	packet_flush_mclist(sk);
3151 
3152 	lock_sock(sk);
3153 	if (po->rx_ring.pg_vec) {
3154 		memset(&req_u, 0, sizeof(req_u));
3155 		packet_set_ring(sk, &req_u, 1, 0);
3156 	}
3157 
3158 	if (po->tx_ring.pg_vec) {
3159 		memset(&req_u, 0, sizeof(req_u));
3160 		packet_set_ring(sk, &req_u, 1, 1);
3161 	}
3162 	release_sock(sk);
3163 
3164 	f = fanout_release(sk);
3165 
3166 	synchronize_net();
3167 
3168 	kfree(po->rollover);
3169 	if (f) {
3170 		fanout_release_data(f);
3171 		kvfree(f);
3172 	}
3173 	/*
3174 	 *	Now the socket is dead. No more input will appear.
3175 	 */
3176 	sock_orphan(sk);
3177 	sock->sk = NULL;
3178 
3179 	/* Purge queues */
3180 
3181 	skb_queue_purge(&sk->sk_receive_queue);
3182 	packet_free_pending(po);
3183 
3184 	sock_put(sk);
3185 	return 0;
3186 }
3187 
3188 /*
3189  *	Attach a packet hook.
3190  */
3191 
3192 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3193 			  __be16 proto)
3194 {
3195 	struct packet_sock *po = pkt_sk(sk);
3196 	struct net_device *dev = NULL;
3197 	bool unlisted = false;
3198 	bool need_rehook;
3199 	int ret = 0;
3200 
3201 	lock_sock(sk);
3202 	spin_lock(&po->bind_lock);
3203 	if (!proto)
3204 		proto = po->num;
3205 
3206 	rcu_read_lock();
3207 
3208 	if (po->fanout) {
3209 		ret = -EINVAL;
3210 		goto out_unlock;
3211 	}
3212 
3213 	if (name) {
3214 		dev = dev_get_by_name_rcu(sock_net(sk), name);
3215 		if (!dev) {
3216 			ret = -ENODEV;
3217 			goto out_unlock;
3218 		}
3219 	} else if (ifindex) {
3220 		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3221 		if (!dev) {
3222 			ret = -ENODEV;
3223 			goto out_unlock;
3224 		}
3225 	}
3226 
3227 	need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3228 
3229 	if (need_rehook) {
3230 		dev_hold(dev);
3231 		if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3232 			rcu_read_unlock();
3233 			/* prevents packet_notifier() from calling
3234 			 * register_prot_hook()
3235 			 */
3236 			WRITE_ONCE(po->num, 0);
3237 			__unregister_prot_hook(sk, true);
3238 			rcu_read_lock();
3239 			if (dev)
3240 				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3241 								 dev->ifindex);
3242 		}
3243 
3244 		BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3245 		WRITE_ONCE(po->num, proto);
3246 		po->prot_hook.type = proto;
3247 
3248 		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3249 
3250 		if (unlikely(unlisted)) {
3251 			po->prot_hook.dev = NULL;
3252 			WRITE_ONCE(po->ifindex, -1);
3253 			packet_cached_dev_reset(po);
3254 		} else {
3255 			netdev_hold(dev, &po->prot_hook.dev_tracker,
3256 				    GFP_ATOMIC);
3257 			po->prot_hook.dev = dev;
3258 			WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3259 			packet_cached_dev_assign(po, dev);
3260 		}
3261 		dev_put(dev);
3262 	}
3263 
3264 	if (proto == 0 || !need_rehook)
3265 		goto out_unlock;
3266 
3267 	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3268 		register_prot_hook(sk);
3269 	} else {
3270 		sk->sk_err = ENETDOWN;
3271 		if (!sock_flag(sk, SOCK_DEAD))
3272 			sk_error_report(sk);
3273 	}
3274 
3275 out_unlock:
3276 	rcu_read_unlock();
3277 	spin_unlock(&po->bind_lock);
3278 	release_sock(sk);
3279 	return ret;
3280 }
3281 
3282 /*
3283  *	Bind a packet socket to a device
3284  */
3285 
3286 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3287 			    int addr_len)
3288 {
3289 	struct sock *sk = sock->sk;
3290 	char name[sizeof(uaddr->sa_data_min) + 1];
3291 
3292 	/*
3293 	 *	Check legality
3294 	 */
3295 
3296 	if (addr_len != sizeof(struct sockaddr))
3297 		return -EINVAL;
3298 	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3299 	 * zero-terminated.
3300 	 */
3301 	memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3302 	name[sizeof(uaddr->sa_data_min)] = 0;
3303 
3304 	return packet_do_bind(sk, name, 0, 0);
3305 }
3306 
3307 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3308 {
3309 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3310 	struct sock *sk = sock->sk;
3311 
3312 	/*
3313 	 *	Check legality
3314 	 */
3315 
3316 	if (addr_len < sizeof(struct sockaddr_ll))
3317 		return -EINVAL;
3318 	if (sll->sll_family != AF_PACKET)
3319 		return -EINVAL;
3320 
3321 	return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3322 }
3323 
3324 static struct proto packet_proto = {
3325 	.name	  = "PACKET",
3326 	.owner	  = THIS_MODULE,
3327 	.obj_size = sizeof(struct packet_sock),
3328 };
3329 
3330 /*
3331  *	Create a packet of type SOCK_PACKET.
3332  */
3333 
3334 static int packet_create(struct net *net, struct socket *sock, int protocol,
3335 			 int kern)
3336 {
3337 	struct sock *sk;
3338 	struct packet_sock *po;
3339 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3340 	int err;
3341 
3342 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3343 		return -EPERM;
3344 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3345 	    sock->type != SOCK_PACKET)
3346 		return -ESOCKTNOSUPPORT;
3347 
3348 	sock->state = SS_UNCONNECTED;
3349 
3350 	err = -ENOBUFS;
3351 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3352 	if (sk == NULL)
3353 		goto out;
3354 
3355 	sock->ops = &packet_ops;
3356 	if (sock->type == SOCK_PACKET)
3357 		sock->ops = &packet_ops_spkt;
3358 
3359 	sock_init_data(sock, sk);
3360 
3361 	po = pkt_sk(sk);
3362 	init_completion(&po->skb_completion);
3363 	sk->sk_family = PF_PACKET;
3364 	po->num = proto;
3365 
3366 	err = packet_alloc_pending(po);
3367 	if (err)
3368 		goto out2;
3369 
3370 	packet_cached_dev_reset(po);
3371 
3372 	sk->sk_destruct = packet_sock_destruct;
3373 
3374 	/*
3375 	 *	Attach a protocol block
3376 	 */
3377 
3378 	spin_lock_init(&po->bind_lock);
3379 	mutex_init(&po->pg_vec_lock);
3380 	po->rollover = NULL;
3381 	po->prot_hook.func = packet_rcv;
3382 
3383 	if (sock->type == SOCK_PACKET)
3384 		po->prot_hook.func = packet_rcv_spkt;
3385 
3386 	po->prot_hook.af_packet_priv = sk;
3387 	po->prot_hook.af_packet_net = sock_net(sk);
3388 
3389 	if (proto) {
3390 		po->prot_hook.type = proto;
3391 		__register_prot_hook(sk);
3392 	}
3393 
3394 	mutex_lock(&net->packet.sklist_lock);
3395 	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3396 	mutex_unlock(&net->packet.sklist_lock);
3397 
3398 	sock_prot_inuse_add(net, &packet_proto, 1);
3399 
3400 	return 0;
3401 out2:
3402 	sk_free(sk);
3403 out:
3404 	return err;
3405 }
3406 
3407 /*
3408  *	Pull a packet from our receive queue and hand it to the user.
3409  *	If necessary we block.
3410  */
3411 
3412 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3413 			  int flags)
3414 {
3415 	struct sock *sk = sock->sk;
3416 	struct sk_buff *skb;
3417 	int copied, err;
3418 	int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3419 	unsigned int origlen = 0;
3420 
3421 	err = -EINVAL;
3422 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3423 		goto out;
3424 
3425 #if 0
3426 	/* What error should we return now? EUNATTACH? */
3427 	if (pkt_sk(sk)->ifindex < 0)
3428 		return -ENODEV;
3429 #endif
3430 
3431 	if (flags & MSG_ERRQUEUE) {
3432 		err = sock_recv_errqueue(sk, msg, len,
3433 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3434 		goto out;
3435 	}
3436 
3437 	/*
3438 	 *	Call the generic datagram receiver. This handles all sorts
3439 	 *	of horrible races and re-entrancy so we can forget about it
3440 	 *	in the protocol layers.
3441 	 *
3442 	 *	Now it will return ENETDOWN, if device have just gone down,
3443 	 *	but then it will block.
3444 	 */
3445 
3446 	skb = skb_recv_datagram(sk, flags, &err);
3447 
3448 	/*
3449 	 *	An error occurred so return it. Because skb_recv_datagram()
3450 	 *	handles the blocking we don't see and worry about blocking
3451 	 *	retries.
3452 	 */
3453 
3454 	if (skb == NULL)
3455 		goto out;
3456 
3457 	packet_rcv_try_clear_pressure(pkt_sk(sk));
3458 
3459 	if (vnet_hdr_len) {
3460 		err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3461 		if (err)
3462 			goto out_free;
3463 	}
3464 
3465 	/* You lose any data beyond the buffer you gave. If it worries
3466 	 * a user program they can ask the device for its MTU
3467 	 * anyway.
3468 	 */
3469 	copied = skb->len;
3470 	if (copied > len) {
3471 		copied = len;
3472 		msg->msg_flags |= MSG_TRUNC;
3473 	}
3474 
3475 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3476 	if (err)
3477 		goto out_free;
3478 
3479 	if (sock->type != SOCK_PACKET) {
3480 		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3481 
3482 		/* Original length was stored in sockaddr_ll fields */
3483 		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3484 		sll->sll_family = AF_PACKET;
3485 		sll->sll_protocol = skb->protocol;
3486 	}
3487 
3488 	sock_recv_cmsgs(msg, sk, skb);
3489 
3490 	if (msg->msg_name) {
3491 		const size_t max_len = min(sizeof(skb->cb),
3492 					   sizeof(struct sockaddr_storage));
3493 		int copy_len;
3494 
3495 		/* If the address length field is there to be filled
3496 		 * in, we fill it in now.
3497 		 */
3498 		if (sock->type == SOCK_PACKET) {
3499 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3500 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3501 			copy_len = msg->msg_namelen;
3502 		} else {
3503 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3504 
3505 			msg->msg_namelen = sll->sll_halen +
3506 				offsetof(struct sockaddr_ll, sll_addr);
3507 			copy_len = msg->msg_namelen;
3508 			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3509 				memset(msg->msg_name +
3510 				       offsetof(struct sockaddr_ll, sll_addr),
3511 				       0, sizeof(sll->sll_addr));
3512 				msg->msg_namelen = sizeof(struct sockaddr_ll);
3513 			}
3514 		}
3515 		if (WARN_ON_ONCE(copy_len > max_len)) {
3516 			copy_len = max_len;
3517 			msg->msg_namelen = copy_len;
3518 		}
3519 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3520 	}
3521 
3522 	if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3523 		struct tpacket_auxdata aux;
3524 
3525 		aux.tp_status = TP_STATUS_USER;
3526 		if (skb->ip_summed == CHECKSUM_PARTIAL)
3527 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3528 		else if (skb->pkt_type != PACKET_OUTGOING &&
3529 			 skb_csum_unnecessary(skb))
3530 			aux.tp_status |= TP_STATUS_CSUM_VALID;
3531 		if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3532 			aux.tp_status |= TP_STATUS_GSO_TCP;
3533 
3534 		aux.tp_len = origlen;
3535 		aux.tp_snaplen = skb->len;
3536 		aux.tp_mac = 0;
3537 		aux.tp_net = skb_network_offset(skb);
3538 		if (skb_vlan_tag_present(skb)) {
3539 			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3540 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3541 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3542 		} else {
3543 			aux.tp_vlan_tci = 0;
3544 			aux.tp_vlan_tpid = 0;
3545 		}
3546 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3547 	}
3548 
3549 	/*
3550 	 *	Free or return the buffer as appropriate. Again this
3551 	 *	hides all the races and re-entrancy issues from us.
3552 	 */
3553 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3554 
3555 out_free:
3556 	skb_free_datagram(sk, skb);
3557 out:
3558 	return err;
3559 }
3560 
3561 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3562 			       int peer)
3563 {
3564 	struct net_device *dev;
3565 	struct sock *sk	= sock->sk;
3566 
3567 	if (peer)
3568 		return -EOPNOTSUPP;
3569 
3570 	uaddr->sa_family = AF_PACKET;
3571 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3572 	rcu_read_lock();
3573 	dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3574 	if (dev)
3575 		strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3576 	rcu_read_unlock();
3577 
3578 	return sizeof(*uaddr);
3579 }
3580 
3581 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3582 			  int peer)
3583 {
3584 	struct net_device *dev;
3585 	struct sock *sk = sock->sk;
3586 	struct packet_sock *po = pkt_sk(sk);
3587 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3588 	int ifindex;
3589 
3590 	if (peer)
3591 		return -EOPNOTSUPP;
3592 
3593 	ifindex = READ_ONCE(po->ifindex);
3594 	sll->sll_family = AF_PACKET;
3595 	sll->sll_ifindex = ifindex;
3596 	sll->sll_protocol = READ_ONCE(po->num);
3597 	sll->sll_pkttype = 0;
3598 	rcu_read_lock();
3599 	dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3600 	if (dev) {
3601 		sll->sll_hatype = dev->type;
3602 		sll->sll_halen = dev->addr_len;
3603 
3604 		/* Let __fortify_memcpy_chk() know the actual buffer size. */
3605 		memcpy(((struct sockaddr_storage *)sll)->__data +
3606 		       offsetof(struct sockaddr_ll, sll_addr) -
3607 		       offsetofend(struct sockaddr_ll, sll_family),
3608 		       dev->dev_addr, dev->addr_len);
3609 	} else {
3610 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3611 		sll->sll_halen = 0;
3612 	}
3613 	rcu_read_unlock();
3614 
3615 	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3616 }
3617 
3618 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3619 			 int what)
3620 {
3621 	switch (i->type) {
3622 	case PACKET_MR_MULTICAST:
3623 		if (i->alen != dev->addr_len)
3624 			return -EINVAL;
3625 		if (what > 0)
3626 			return dev_mc_add(dev, i->addr);
3627 		else
3628 			return dev_mc_del(dev, i->addr);
3629 		break;
3630 	case PACKET_MR_PROMISC:
3631 		return dev_set_promiscuity(dev, what);
3632 	case PACKET_MR_ALLMULTI:
3633 		return dev_set_allmulti(dev, what);
3634 	case PACKET_MR_UNICAST:
3635 		if (i->alen != dev->addr_len)
3636 			return -EINVAL;
3637 		if (what > 0)
3638 			return dev_uc_add(dev, i->addr);
3639 		else
3640 			return dev_uc_del(dev, i->addr);
3641 		break;
3642 	default:
3643 		break;
3644 	}
3645 	return 0;
3646 }
3647 
3648 static void packet_dev_mclist_delete(struct net_device *dev,
3649 				     struct packet_mclist **mlp)
3650 {
3651 	struct packet_mclist *ml;
3652 
3653 	while ((ml = *mlp) != NULL) {
3654 		if (ml->ifindex == dev->ifindex) {
3655 			packet_dev_mc(dev, ml, -1);
3656 			*mlp = ml->next;
3657 			kfree(ml);
3658 		} else
3659 			mlp = &ml->next;
3660 	}
3661 }
3662 
3663 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3664 {
3665 	struct packet_sock *po = pkt_sk(sk);
3666 	struct packet_mclist *ml, *i;
3667 	struct net_device *dev;
3668 	int err;
3669 
3670 	rtnl_lock();
3671 
3672 	err = -ENODEV;
3673 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3674 	if (!dev)
3675 		goto done;
3676 
3677 	err = -EINVAL;
3678 	if (mreq->mr_alen > dev->addr_len)
3679 		goto done;
3680 
3681 	err = -ENOBUFS;
3682 	i = kmalloc(sizeof(*i), GFP_KERNEL);
3683 	if (i == NULL)
3684 		goto done;
3685 
3686 	err = 0;
3687 	for (ml = po->mclist; ml; ml = ml->next) {
3688 		if (ml->ifindex == mreq->mr_ifindex &&
3689 		    ml->type == mreq->mr_type &&
3690 		    ml->alen == mreq->mr_alen &&
3691 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3692 			ml->count++;
3693 			/* Free the new element ... */
3694 			kfree(i);
3695 			goto done;
3696 		}
3697 	}
3698 
3699 	i->type = mreq->mr_type;
3700 	i->ifindex = mreq->mr_ifindex;
3701 	i->alen = mreq->mr_alen;
3702 	memcpy(i->addr, mreq->mr_address, i->alen);
3703 	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3704 	i->count = 1;
3705 	i->next = po->mclist;
3706 	po->mclist = i;
3707 	err = packet_dev_mc(dev, i, 1);
3708 	if (err) {
3709 		po->mclist = i->next;
3710 		kfree(i);
3711 	}
3712 
3713 done:
3714 	rtnl_unlock();
3715 	return err;
3716 }
3717 
3718 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3719 {
3720 	struct packet_mclist *ml, **mlp;
3721 
3722 	rtnl_lock();
3723 
3724 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3725 		if (ml->ifindex == mreq->mr_ifindex &&
3726 		    ml->type == mreq->mr_type &&
3727 		    ml->alen == mreq->mr_alen &&
3728 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3729 			if (--ml->count == 0) {
3730 				struct net_device *dev;
3731 				*mlp = ml->next;
3732 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3733 				if (dev)
3734 					packet_dev_mc(dev, ml, -1);
3735 				kfree(ml);
3736 			}
3737 			break;
3738 		}
3739 	}
3740 	rtnl_unlock();
3741 	return 0;
3742 }
3743 
3744 static void packet_flush_mclist(struct sock *sk)
3745 {
3746 	struct packet_sock *po = pkt_sk(sk);
3747 	struct packet_mclist *ml;
3748 
3749 	if (!po->mclist)
3750 		return;
3751 
3752 	rtnl_lock();
3753 	while ((ml = po->mclist) != NULL) {
3754 		struct net_device *dev;
3755 
3756 		po->mclist = ml->next;
3757 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3758 		if (dev != NULL)
3759 			packet_dev_mc(dev, ml, -1);
3760 		kfree(ml);
3761 	}
3762 	rtnl_unlock();
3763 }
3764 
3765 static int
3766 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3767 		  unsigned int optlen)
3768 {
3769 	struct sock *sk = sock->sk;
3770 	struct packet_sock *po = pkt_sk(sk);
3771 	int ret;
3772 
3773 	if (level != SOL_PACKET)
3774 		return -ENOPROTOOPT;
3775 
3776 	switch (optname) {
3777 	case PACKET_ADD_MEMBERSHIP:
3778 	case PACKET_DROP_MEMBERSHIP:
3779 	{
3780 		struct packet_mreq_max mreq;
3781 		int len = optlen;
3782 		memset(&mreq, 0, sizeof(mreq));
3783 		if (len < sizeof(struct packet_mreq))
3784 			return -EINVAL;
3785 		if (len > sizeof(mreq))
3786 			len = sizeof(mreq);
3787 		if (copy_from_sockptr(&mreq, optval, len))
3788 			return -EFAULT;
3789 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3790 			return -EINVAL;
3791 		if (optname == PACKET_ADD_MEMBERSHIP)
3792 			ret = packet_mc_add(sk, &mreq);
3793 		else
3794 			ret = packet_mc_drop(sk, &mreq);
3795 		return ret;
3796 	}
3797 
3798 	case PACKET_RX_RING:
3799 	case PACKET_TX_RING:
3800 	{
3801 		union tpacket_req_u req_u;
3802 
3803 		ret = -EINVAL;
3804 		lock_sock(sk);
3805 		switch (po->tp_version) {
3806 		case TPACKET_V1:
3807 		case TPACKET_V2:
3808 			if (optlen < sizeof(req_u.req))
3809 				break;
3810 			ret = copy_from_sockptr(&req_u.req, optval,
3811 						sizeof(req_u.req)) ?
3812 						-EINVAL : 0;
3813 			break;
3814 		case TPACKET_V3:
3815 		default:
3816 			if (optlen < sizeof(req_u.req3))
3817 				break;
3818 			ret = copy_from_sockptr(&req_u.req3, optval,
3819 						sizeof(req_u.req3)) ?
3820 						-EINVAL : 0;
3821 			break;
3822 		}
3823 		if (!ret)
3824 			ret = packet_set_ring(sk, &req_u, 0,
3825 					      optname == PACKET_TX_RING);
3826 		release_sock(sk);
3827 		return ret;
3828 	}
3829 	case PACKET_COPY_THRESH:
3830 	{
3831 		int val;
3832 
3833 		if (optlen != sizeof(val))
3834 			return -EINVAL;
3835 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3836 			return -EFAULT;
3837 
3838 		WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3839 		return 0;
3840 	}
3841 	case PACKET_VERSION:
3842 	{
3843 		int val;
3844 
3845 		if (optlen != sizeof(val))
3846 			return -EINVAL;
3847 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3848 			return -EFAULT;
3849 		switch (val) {
3850 		case TPACKET_V1:
3851 		case TPACKET_V2:
3852 		case TPACKET_V3:
3853 			break;
3854 		default:
3855 			return -EINVAL;
3856 		}
3857 		lock_sock(sk);
3858 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3859 			ret = -EBUSY;
3860 		} else {
3861 			po->tp_version = val;
3862 			ret = 0;
3863 		}
3864 		release_sock(sk);
3865 		return ret;
3866 	}
3867 	case PACKET_RESERVE:
3868 	{
3869 		unsigned int val;
3870 
3871 		if (optlen != sizeof(val))
3872 			return -EINVAL;
3873 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3874 			return -EFAULT;
3875 		if (val > INT_MAX)
3876 			return -EINVAL;
3877 		lock_sock(sk);
3878 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3879 			ret = -EBUSY;
3880 		} else {
3881 			po->tp_reserve = val;
3882 			ret = 0;
3883 		}
3884 		release_sock(sk);
3885 		return ret;
3886 	}
3887 	case PACKET_LOSS:
3888 	{
3889 		unsigned int val;
3890 
3891 		if (optlen != sizeof(val))
3892 			return -EINVAL;
3893 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3894 			return -EFAULT;
3895 
3896 		lock_sock(sk);
3897 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3898 			ret = -EBUSY;
3899 		} else {
3900 			packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3901 			ret = 0;
3902 		}
3903 		release_sock(sk);
3904 		return ret;
3905 	}
3906 	case PACKET_AUXDATA:
3907 	{
3908 		int val;
3909 
3910 		if (optlen < sizeof(val))
3911 			return -EINVAL;
3912 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3913 			return -EFAULT;
3914 
3915 		packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3916 		return 0;
3917 	}
3918 	case PACKET_ORIGDEV:
3919 	{
3920 		int val;
3921 
3922 		if (optlen < sizeof(val))
3923 			return -EINVAL;
3924 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3925 			return -EFAULT;
3926 
3927 		packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3928 		return 0;
3929 	}
3930 	case PACKET_VNET_HDR:
3931 	case PACKET_VNET_HDR_SZ:
3932 	{
3933 		int val, hdr_len;
3934 
3935 		if (sock->type != SOCK_RAW)
3936 			return -EINVAL;
3937 		if (optlen < sizeof(val))
3938 			return -EINVAL;
3939 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3940 			return -EFAULT;
3941 
3942 		if (optname == PACKET_VNET_HDR_SZ) {
3943 			if (val && val != sizeof(struct virtio_net_hdr) &&
3944 			    val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
3945 				return -EINVAL;
3946 			hdr_len = val;
3947 		} else {
3948 			hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
3949 		}
3950 		lock_sock(sk);
3951 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3952 			ret = -EBUSY;
3953 		} else {
3954 			WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
3955 			ret = 0;
3956 		}
3957 		release_sock(sk);
3958 		return ret;
3959 	}
3960 	case PACKET_TIMESTAMP:
3961 	{
3962 		int val;
3963 
3964 		if (optlen != sizeof(val))
3965 			return -EINVAL;
3966 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3967 			return -EFAULT;
3968 
3969 		WRITE_ONCE(po->tp_tstamp, val);
3970 		return 0;
3971 	}
3972 	case PACKET_FANOUT:
3973 	{
3974 		struct fanout_args args = { 0 };
3975 
3976 		if (optlen != sizeof(int) && optlen != sizeof(args))
3977 			return -EINVAL;
3978 		if (copy_from_sockptr(&args, optval, optlen))
3979 			return -EFAULT;
3980 
3981 		return fanout_add(sk, &args);
3982 	}
3983 	case PACKET_FANOUT_DATA:
3984 	{
3985 		/* Paired with the WRITE_ONCE() in fanout_add() */
3986 		if (!READ_ONCE(po->fanout))
3987 			return -EINVAL;
3988 
3989 		return fanout_set_data(po, optval, optlen);
3990 	}
3991 	case PACKET_IGNORE_OUTGOING:
3992 	{
3993 		int val;
3994 
3995 		if (optlen != sizeof(val))
3996 			return -EINVAL;
3997 		if (copy_from_sockptr(&val, optval, sizeof(val)))
3998 			return -EFAULT;
3999 		if (val < 0 || val > 1)
4000 			return -EINVAL;
4001 
4002 		WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4003 		return 0;
4004 	}
4005 	case PACKET_TX_HAS_OFF:
4006 	{
4007 		unsigned int val;
4008 
4009 		if (optlen != sizeof(val))
4010 			return -EINVAL;
4011 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4012 			return -EFAULT;
4013 
4014 		lock_sock(sk);
4015 		if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4016 			packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4017 
4018 		release_sock(sk);
4019 		return 0;
4020 	}
4021 	case PACKET_QDISC_BYPASS:
4022 	{
4023 		int val;
4024 
4025 		if (optlen != sizeof(val))
4026 			return -EINVAL;
4027 		if (copy_from_sockptr(&val, optval, sizeof(val)))
4028 			return -EFAULT;
4029 
4030 		packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4031 		return 0;
4032 	}
4033 	default:
4034 		return -ENOPROTOOPT;
4035 	}
4036 }
4037 
4038 static int packet_getsockopt(struct socket *sock, int level, int optname,
4039 			     char __user *optval, int __user *optlen)
4040 {
4041 	int len;
4042 	int val, lv = sizeof(val);
4043 	struct sock *sk = sock->sk;
4044 	struct packet_sock *po = pkt_sk(sk);
4045 	void *data = &val;
4046 	union tpacket_stats_u st;
4047 	struct tpacket_rollover_stats rstats;
4048 	int drops;
4049 
4050 	if (level != SOL_PACKET)
4051 		return -ENOPROTOOPT;
4052 
4053 	if (get_user(len, optlen))
4054 		return -EFAULT;
4055 
4056 	if (len < 0)
4057 		return -EINVAL;
4058 
4059 	switch (optname) {
4060 	case PACKET_STATISTICS:
4061 		spin_lock_bh(&sk->sk_receive_queue.lock);
4062 		memcpy(&st, &po->stats, sizeof(st));
4063 		memset(&po->stats, 0, sizeof(po->stats));
4064 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4065 		drops = atomic_xchg(&po->tp_drops, 0);
4066 
4067 		if (po->tp_version == TPACKET_V3) {
4068 			lv = sizeof(struct tpacket_stats_v3);
4069 			st.stats3.tp_drops = drops;
4070 			st.stats3.tp_packets += drops;
4071 			data = &st.stats3;
4072 		} else {
4073 			lv = sizeof(struct tpacket_stats);
4074 			st.stats1.tp_drops = drops;
4075 			st.stats1.tp_packets += drops;
4076 			data = &st.stats1;
4077 		}
4078 
4079 		break;
4080 	case PACKET_AUXDATA:
4081 		val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4082 		break;
4083 	case PACKET_ORIGDEV:
4084 		val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4085 		break;
4086 	case PACKET_VNET_HDR:
4087 		val = !!READ_ONCE(po->vnet_hdr_sz);
4088 		break;
4089 	case PACKET_VNET_HDR_SZ:
4090 		val = READ_ONCE(po->vnet_hdr_sz);
4091 		break;
4092 	case PACKET_COPY_THRESH:
4093 		val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4094 		break;
4095 	case PACKET_VERSION:
4096 		val = po->tp_version;
4097 		break;
4098 	case PACKET_HDRLEN:
4099 		if (len > sizeof(int))
4100 			len = sizeof(int);
4101 		if (len < sizeof(int))
4102 			return -EINVAL;
4103 		if (copy_from_user(&val, optval, len))
4104 			return -EFAULT;
4105 		switch (val) {
4106 		case TPACKET_V1:
4107 			val = sizeof(struct tpacket_hdr);
4108 			break;
4109 		case TPACKET_V2:
4110 			val = sizeof(struct tpacket2_hdr);
4111 			break;
4112 		case TPACKET_V3:
4113 			val = sizeof(struct tpacket3_hdr);
4114 			break;
4115 		default:
4116 			return -EINVAL;
4117 		}
4118 		break;
4119 	case PACKET_RESERVE:
4120 		val = po->tp_reserve;
4121 		break;
4122 	case PACKET_LOSS:
4123 		val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4124 		break;
4125 	case PACKET_TIMESTAMP:
4126 		val = READ_ONCE(po->tp_tstamp);
4127 		break;
4128 	case PACKET_FANOUT:
4129 		val = (po->fanout ?
4130 		       ((u32)po->fanout->id |
4131 			((u32)po->fanout->type << 16) |
4132 			((u32)po->fanout->flags << 24)) :
4133 		       0);
4134 		break;
4135 	case PACKET_IGNORE_OUTGOING:
4136 		val = READ_ONCE(po->prot_hook.ignore_outgoing);
4137 		break;
4138 	case PACKET_ROLLOVER_STATS:
4139 		if (!po->rollover)
4140 			return -EINVAL;
4141 		rstats.tp_all = atomic_long_read(&po->rollover->num);
4142 		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4143 		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4144 		data = &rstats;
4145 		lv = sizeof(rstats);
4146 		break;
4147 	case PACKET_TX_HAS_OFF:
4148 		val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4149 		break;
4150 	case PACKET_QDISC_BYPASS:
4151 		val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4152 		break;
4153 	default:
4154 		return -ENOPROTOOPT;
4155 	}
4156 
4157 	if (len > lv)
4158 		len = lv;
4159 	if (put_user(len, optlen))
4160 		return -EFAULT;
4161 	if (copy_to_user(optval, data, len))
4162 		return -EFAULT;
4163 	return 0;
4164 }
4165 
4166 static int packet_notifier(struct notifier_block *this,
4167 			   unsigned long msg, void *ptr)
4168 {
4169 	struct sock *sk;
4170 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4171 	struct net *net = dev_net(dev);
4172 
4173 	rcu_read_lock();
4174 	sk_for_each_rcu(sk, &net->packet.sklist) {
4175 		struct packet_sock *po = pkt_sk(sk);
4176 
4177 		switch (msg) {
4178 		case NETDEV_UNREGISTER:
4179 			if (po->mclist)
4180 				packet_dev_mclist_delete(dev, &po->mclist);
4181 			fallthrough;
4182 
4183 		case NETDEV_DOWN:
4184 			if (dev->ifindex == po->ifindex) {
4185 				spin_lock(&po->bind_lock);
4186 				if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4187 					__unregister_prot_hook(sk, false);
4188 					sk->sk_err = ENETDOWN;
4189 					if (!sock_flag(sk, SOCK_DEAD))
4190 						sk_error_report(sk);
4191 				}
4192 				if (msg == NETDEV_UNREGISTER) {
4193 					packet_cached_dev_reset(po);
4194 					WRITE_ONCE(po->ifindex, -1);
4195 					netdev_put(po->prot_hook.dev,
4196 						   &po->prot_hook.dev_tracker);
4197 					po->prot_hook.dev = NULL;
4198 				}
4199 				spin_unlock(&po->bind_lock);
4200 			}
4201 			break;
4202 		case NETDEV_UP:
4203 			if (dev->ifindex == po->ifindex) {
4204 				spin_lock(&po->bind_lock);
4205 				if (po->num)
4206 					register_prot_hook(sk);
4207 				spin_unlock(&po->bind_lock);
4208 			}
4209 			break;
4210 		}
4211 	}
4212 	rcu_read_unlock();
4213 	return NOTIFY_DONE;
4214 }
4215 
4216 
4217 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4218 			unsigned long arg)
4219 {
4220 	struct sock *sk = sock->sk;
4221 
4222 	switch (cmd) {
4223 	case SIOCOUTQ:
4224 	{
4225 		int amount = sk_wmem_alloc_get(sk);
4226 
4227 		return put_user(amount, (int __user *)arg);
4228 	}
4229 	case SIOCINQ:
4230 	{
4231 		struct sk_buff *skb;
4232 		int amount = 0;
4233 
4234 		spin_lock_bh(&sk->sk_receive_queue.lock);
4235 		skb = skb_peek(&sk->sk_receive_queue);
4236 		if (skb)
4237 			amount = skb->len;
4238 		spin_unlock_bh(&sk->sk_receive_queue.lock);
4239 		return put_user(amount, (int __user *)arg);
4240 	}
4241 #ifdef CONFIG_INET
4242 	case SIOCADDRT:
4243 	case SIOCDELRT:
4244 	case SIOCDARP:
4245 	case SIOCGARP:
4246 	case SIOCSARP:
4247 	case SIOCGIFADDR:
4248 	case SIOCSIFADDR:
4249 	case SIOCGIFBRDADDR:
4250 	case SIOCSIFBRDADDR:
4251 	case SIOCGIFNETMASK:
4252 	case SIOCSIFNETMASK:
4253 	case SIOCGIFDSTADDR:
4254 	case SIOCSIFDSTADDR:
4255 	case SIOCSIFFLAGS:
4256 		return inet_dgram_ops.ioctl(sock, cmd, arg);
4257 #endif
4258 
4259 	default:
4260 		return -ENOIOCTLCMD;
4261 	}
4262 	return 0;
4263 }
4264 
4265 static __poll_t packet_poll(struct file *file, struct socket *sock,
4266 				poll_table *wait)
4267 {
4268 	struct sock *sk = sock->sk;
4269 	struct packet_sock *po = pkt_sk(sk);
4270 	__poll_t mask = datagram_poll(file, sock, wait);
4271 
4272 	spin_lock_bh(&sk->sk_receive_queue.lock);
4273 	if (po->rx_ring.pg_vec) {
4274 		if (!packet_previous_rx_frame(po, &po->rx_ring,
4275 			TP_STATUS_KERNEL))
4276 			mask |= EPOLLIN | EPOLLRDNORM;
4277 	}
4278 	packet_rcv_try_clear_pressure(po);
4279 	spin_unlock_bh(&sk->sk_receive_queue.lock);
4280 	spin_lock_bh(&sk->sk_write_queue.lock);
4281 	if (po->tx_ring.pg_vec) {
4282 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4283 			mask |= EPOLLOUT | EPOLLWRNORM;
4284 	}
4285 	spin_unlock_bh(&sk->sk_write_queue.lock);
4286 	return mask;
4287 }
4288 
4289 
4290 /* Dirty? Well, I still did not learn better way to account
4291  * for user mmaps.
4292  */
4293 
4294 static void packet_mm_open(struct vm_area_struct *vma)
4295 {
4296 	struct file *file = vma->vm_file;
4297 	struct socket *sock = file->private_data;
4298 	struct sock *sk = sock->sk;
4299 
4300 	if (sk)
4301 		atomic_long_inc(&pkt_sk(sk)->mapped);
4302 }
4303 
4304 static void packet_mm_close(struct vm_area_struct *vma)
4305 {
4306 	struct file *file = vma->vm_file;
4307 	struct socket *sock = file->private_data;
4308 	struct sock *sk = sock->sk;
4309 
4310 	if (sk)
4311 		atomic_long_dec(&pkt_sk(sk)->mapped);
4312 }
4313 
4314 static const struct vm_operations_struct packet_mmap_ops = {
4315 	.open	=	packet_mm_open,
4316 	.close	=	packet_mm_close,
4317 };
4318 
4319 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4320 			unsigned int len)
4321 {
4322 	int i;
4323 
4324 	for (i = 0; i < len; i++) {
4325 		if (likely(pg_vec[i].buffer)) {
4326 			if (is_vmalloc_addr(pg_vec[i].buffer))
4327 				vfree(pg_vec[i].buffer);
4328 			else
4329 				free_pages((unsigned long)pg_vec[i].buffer,
4330 					   order);
4331 			pg_vec[i].buffer = NULL;
4332 		}
4333 	}
4334 	kfree(pg_vec);
4335 }
4336 
4337 static char *alloc_one_pg_vec_page(unsigned long order)
4338 {
4339 	char *buffer;
4340 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4341 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4342 
4343 	buffer = (char *) __get_free_pages(gfp_flags, order);
4344 	if (buffer)
4345 		return buffer;
4346 
4347 	/* __get_free_pages failed, fall back to vmalloc */
4348 	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4349 	if (buffer)
4350 		return buffer;
4351 
4352 	/* vmalloc failed, lets dig into swap here */
4353 	gfp_flags &= ~__GFP_NORETRY;
4354 	buffer = (char *) __get_free_pages(gfp_flags, order);
4355 	if (buffer)
4356 		return buffer;
4357 
4358 	/* complete and utter failure */
4359 	return NULL;
4360 }
4361 
4362 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4363 {
4364 	unsigned int block_nr = req->tp_block_nr;
4365 	struct pgv *pg_vec;
4366 	int i;
4367 
4368 	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4369 	if (unlikely(!pg_vec))
4370 		goto out;
4371 
4372 	for (i = 0; i < block_nr; i++) {
4373 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4374 		if (unlikely(!pg_vec[i].buffer))
4375 			goto out_free_pgvec;
4376 	}
4377 
4378 out:
4379 	return pg_vec;
4380 
4381 out_free_pgvec:
4382 	free_pg_vec(pg_vec, order, block_nr);
4383 	pg_vec = NULL;
4384 	goto out;
4385 }
4386 
4387 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4388 		int closing, int tx_ring)
4389 {
4390 	struct pgv *pg_vec = NULL;
4391 	struct packet_sock *po = pkt_sk(sk);
4392 	unsigned long *rx_owner_map = NULL;
4393 	int was_running, order = 0;
4394 	struct packet_ring_buffer *rb;
4395 	struct sk_buff_head *rb_queue;
4396 	__be16 num;
4397 	int err;
4398 	/* Added to avoid minimal code churn */
4399 	struct tpacket_req *req = &req_u->req;
4400 
4401 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4402 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4403 
4404 	err = -EBUSY;
4405 	if (!closing) {
4406 		if (atomic_long_read(&po->mapped))
4407 			goto out;
4408 		if (packet_read_pending(rb))
4409 			goto out;
4410 	}
4411 
4412 	if (req->tp_block_nr) {
4413 		unsigned int min_frame_size;
4414 
4415 		/* Sanity tests and some calculations */
4416 		err = -EBUSY;
4417 		if (unlikely(rb->pg_vec))
4418 			goto out;
4419 
4420 		switch (po->tp_version) {
4421 		case TPACKET_V1:
4422 			po->tp_hdrlen = TPACKET_HDRLEN;
4423 			break;
4424 		case TPACKET_V2:
4425 			po->tp_hdrlen = TPACKET2_HDRLEN;
4426 			break;
4427 		case TPACKET_V3:
4428 			po->tp_hdrlen = TPACKET3_HDRLEN;
4429 			break;
4430 		}
4431 
4432 		err = -EINVAL;
4433 		if (unlikely((int)req->tp_block_size <= 0))
4434 			goto out;
4435 		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4436 			goto out;
4437 		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4438 		if (po->tp_version >= TPACKET_V3 &&
4439 		    req->tp_block_size <
4440 		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4441 			goto out;
4442 		if (unlikely(req->tp_frame_size < min_frame_size))
4443 			goto out;
4444 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4445 			goto out;
4446 
4447 		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4448 		if (unlikely(rb->frames_per_block == 0))
4449 			goto out;
4450 		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4451 			goto out;
4452 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4453 					req->tp_frame_nr))
4454 			goto out;
4455 
4456 		err = -ENOMEM;
4457 		order = get_order(req->tp_block_size);
4458 		pg_vec = alloc_pg_vec(req, order);
4459 		if (unlikely(!pg_vec))
4460 			goto out;
4461 		switch (po->tp_version) {
4462 		case TPACKET_V3:
4463 			/* Block transmit is not supported yet */
4464 			if (!tx_ring) {
4465 				init_prb_bdqc(po, rb, pg_vec, req_u);
4466 			} else {
4467 				struct tpacket_req3 *req3 = &req_u->req3;
4468 
4469 				if (req3->tp_retire_blk_tov ||
4470 				    req3->tp_sizeof_priv ||
4471 				    req3->tp_feature_req_word) {
4472 					err = -EINVAL;
4473 					goto out_free_pg_vec;
4474 				}
4475 			}
4476 			break;
4477 		default:
4478 			if (!tx_ring) {
4479 				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4480 					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4481 				if (!rx_owner_map)
4482 					goto out_free_pg_vec;
4483 			}
4484 			break;
4485 		}
4486 	}
4487 	/* Done */
4488 	else {
4489 		err = -EINVAL;
4490 		if (unlikely(req->tp_frame_nr))
4491 			goto out;
4492 	}
4493 
4494 
4495 	/* Detach socket from network */
4496 	spin_lock(&po->bind_lock);
4497 	was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4498 	num = po->num;
4499 	if (was_running) {
4500 		WRITE_ONCE(po->num, 0);
4501 		__unregister_prot_hook(sk, false);
4502 	}
4503 	spin_unlock(&po->bind_lock);
4504 
4505 	synchronize_net();
4506 
4507 	err = -EBUSY;
4508 	mutex_lock(&po->pg_vec_lock);
4509 	if (closing || atomic_long_read(&po->mapped) == 0) {
4510 		err = 0;
4511 		spin_lock_bh(&rb_queue->lock);
4512 		swap(rb->pg_vec, pg_vec);
4513 		if (po->tp_version <= TPACKET_V2)
4514 			swap(rb->rx_owner_map, rx_owner_map);
4515 		rb->frame_max = (req->tp_frame_nr - 1);
4516 		rb->head = 0;
4517 		rb->frame_size = req->tp_frame_size;
4518 		spin_unlock_bh(&rb_queue->lock);
4519 
4520 		swap(rb->pg_vec_order, order);
4521 		swap(rb->pg_vec_len, req->tp_block_nr);
4522 
4523 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4524 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4525 						tpacket_rcv : packet_rcv;
4526 		skb_queue_purge(rb_queue);
4527 		if (atomic_long_read(&po->mapped))
4528 			pr_err("packet_mmap: vma is busy: %ld\n",
4529 			       atomic_long_read(&po->mapped));
4530 	}
4531 	mutex_unlock(&po->pg_vec_lock);
4532 
4533 	spin_lock(&po->bind_lock);
4534 	if (was_running) {
4535 		WRITE_ONCE(po->num, num);
4536 		register_prot_hook(sk);
4537 	}
4538 	spin_unlock(&po->bind_lock);
4539 	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4540 		/* Because we don't support block-based V3 on tx-ring */
4541 		if (!tx_ring)
4542 			prb_shutdown_retire_blk_timer(po, rb_queue);
4543 	}
4544 
4545 out_free_pg_vec:
4546 	if (pg_vec) {
4547 		bitmap_free(rx_owner_map);
4548 		free_pg_vec(pg_vec, order, req->tp_block_nr);
4549 	}
4550 out:
4551 	return err;
4552 }
4553 
4554 static int packet_mmap(struct file *file, struct socket *sock,
4555 		struct vm_area_struct *vma)
4556 {
4557 	struct sock *sk = sock->sk;
4558 	struct packet_sock *po = pkt_sk(sk);
4559 	unsigned long size, expected_size;
4560 	struct packet_ring_buffer *rb;
4561 	unsigned long start;
4562 	int err = -EINVAL;
4563 	int i;
4564 
4565 	if (vma->vm_pgoff)
4566 		return -EINVAL;
4567 
4568 	mutex_lock(&po->pg_vec_lock);
4569 
4570 	expected_size = 0;
4571 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4572 		if (rb->pg_vec) {
4573 			expected_size += rb->pg_vec_len
4574 						* rb->pg_vec_pages
4575 						* PAGE_SIZE;
4576 		}
4577 	}
4578 
4579 	if (expected_size == 0)
4580 		goto out;
4581 
4582 	size = vma->vm_end - vma->vm_start;
4583 	if (size != expected_size)
4584 		goto out;
4585 
4586 	start = vma->vm_start;
4587 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4588 		if (rb->pg_vec == NULL)
4589 			continue;
4590 
4591 		for (i = 0; i < rb->pg_vec_len; i++) {
4592 			struct page *page;
4593 			void *kaddr = rb->pg_vec[i].buffer;
4594 			int pg_num;
4595 
4596 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4597 				page = pgv_to_page(kaddr);
4598 				err = vm_insert_page(vma, start, page);
4599 				if (unlikely(err))
4600 					goto out;
4601 				start += PAGE_SIZE;
4602 				kaddr += PAGE_SIZE;
4603 			}
4604 		}
4605 	}
4606 
4607 	atomic_long_inc(&po->mapped);
4608 	vma->vm_ops = &packet_mmap_ops;
4609 	err = 0;
4610 
4611 out:
4612 	mutex_unlock(&po->pg_vec_lock);
4613 	return err;
4614 }
4615 
4616 static const struct proto_ops packet_ops_spkt = {
4617 	.family =	PF_PACKET,
4618 	.owner =	THIS_MODULE,
4619 	.release =	packet_release,
4620 	.bind =		packet_bind_spkt,
4621 	.connect =	sock_no_connect,
4622 	.socketpair =	sock_no_socketpair,
4623 	.accept =	sock_no_accept,
4624 	.getname =	packet_getname_spkt,
4625 	.poll =		datagram_poll,
4626 	.ioctl =	packet_ioctl,
4627 	.gettstamp =	sock_gettstamp,
4628 	.listen =	sock_no_listen,
4629 	.shutdown =	sock_no_shutdown,
4630 	.sendmsg =	packet_sendmsg_spkt,
4631 	.recvmsg =	packet_recvmsg,
4632 	.mmap =		sock_no_mmap,
4633 };
4634 
4635 static const struct proto_ops packet_ops = {
4636 	.family =	PF_PACKET,
4637 	.owner =	THIS_MODULE,
4638 	.release =	packet_release,
4639 	.bind =		packet_bind,
4640 	.connect =	sock_no_connect,
4641 	.socketpair =	sock_no_socketpair,
4642 	.accept =	sock_no_accept,
4643 	.getname =	packet_getname,
4644 	.poll =		packet_poll,
4645 	.ioctl =	packet_ioctl,
4646 	.gettstamp =	sock_gettstamp,
4647 	.listen =	sock_no_listen,
4648 	.shutdown =	sock_no_shutdown,
4649 	.setsockopt =	packet_setsockopt,
4650 	.getsockopt =	packet_getsockopt,
4651 	.sendmsg =	packet_sendmsg,
4652 	.recvmsg =	packet_recvmsg,
4653 	.mmap =		packet_mmap,
4654 };
4655 
4656 static const struct net_proto_family packet_family_ops = {
4657 	.family =	PF_PACKET,
4658 	.create =	packet_create,
4659 	.owner	=	THIS_MODULE,
4660 };
4661 
4662 static struct notifier_block packet_netdev_notifier = {
4663 	.notifier_call =	packet_notifier,
4664 };
4665 
4666 #ifdef CONFIG_PROC_FS
4667 
4668 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4669 	__acquires(RCU)
4670 {
4671 	struct net *net = seq_file_net(seq);
4672 
4673 	rcu_read_lock();
4674 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4675 }
4676 
4677 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4678 {
4679 	struct net *net = seq_file_net(seq);
4680 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4681 }
4682 
4683 static void packet_seq_stop(struct seq_file *seq, void *v)
4684 	__releases(RCU)
4685 {
4686 	rcu_read_unlock();
4687 }
4688 
4689 static int packet_seq_show(struct seq_file *seq, void *v)
4690 {
4691 	if (v == SEQ_START_TOKEN)
4692 		seq_printf(seq,
4693 			   "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4694 			   IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4695 	else {
4696 		struct sock *s = sk_entry(v);
4697 		const struct packet_sock *po = pkt_sk(s);
4698 
4699 		seq_printf(seq,
4700 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4701 			   s,
4702 			   refcount_read(&s->sk_refcnt),
4703 			   s->sk_type,
4704 			   ntohs(READ_ONCE(po->num)),
4705 			   READ_ONCE(po->ifindex),
4706 			   packet_sock_flag(po, PACKET_SOCK_RUNNING),
4707 			   atomic_read(&s->sk_rmem_alloc),
4708 			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4709 			   sock_i_ino(s));
4710 	}
4711 
4712 	return 0;
4713 }
4714 
4715 static const struct seq_operations packet_seq_ops = {
4716 	.start	= packet_seq_start,
4717 	.next	= packet_seq_next,
4718 	.stop	= packet_seq_stop,
4719 	.show	= packet_seq_show,
4720 };
4721 #endif
4722 
4723 static int __net_init packet_net_init(struct net *net)
4724 {
4725 	mutex_init(&net->packet.sklist_lock);
4726 	INIT_HLIST_HEAD(&net->packet.sklist);
4727 
4728 #ifdef CONFIG_PROC_FS
4729 	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4730 			sizeof(struct seq_net_private)))
4731 		return -ENOMEM;
4732 #endif /* CONFIG_PROC_FS */
4733 
4734 	return 0;
4735 }
4736 
4737 static void __net_exit packet_net_exit(struct net *net)
4738 {
4739 	remove_proc_entry("packet", net->proc_net);
4740 	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4741 }
4742 
4743 static struct pernet_operations packet_net_ops = {
4744 	.init = packet_net_init,
4745 	.exit = packet_net_exit,
4746 };
4747 
4748 
4749 static void __exit packet_exit(void)
4750 {
4751 	sock_unregister(PF_PACKET);
4752 	proto_unregister(&packet_proto);
4753 	unregister_netdevice_notifier(&packet_netdev_notifier);
4754 	unregister_pernet_subsys(&packet_net_ops);
4755 }
4756 
4757 static int __init packet_init(void)
4758 {
4759 	int rc;
4760 
4761 	rc = register_pernet_subsys(&packet_net_ops);
4762 	if (rc)
4763 		goto out;
4764 	rc = register_netdevice_notifier(&packet_netdev_notifier);
4765 	if (rc)
4766 		goto out_pernet;
4767 	rc = proto_register(&packet_proto, 0);
4768 	if (rc)
4769 		goto out_notifier;
4770 	rc = sock_register(&packet_family_ops);
4771 	if (rc)
4772 		goto out_proto;
4773 
4774 	return 0;
4775 
4776 out_proto:
4777 	proto_unregister(&packet_proto);
4778 out_notifier:
4779 	unregister_netdevice_notifier(&packet_netdev_notifier);
4780 out_pernet:
4781 	unregister_pernet_subsys(&packet_net_ops);
4782 out:
4783 	return rc;
4784 }
4785 
4786 module_init(packet_init);
4787 module_exit(packet_exit);
4788 MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4789 MODULE_LICENSE("GPL");
4790 MODULE_ALIAS_NETPROTO(PF_PACKET);
4791