xref: /linux/net/ipv4/ip_sockglue.c (revision c4101e55974cc7d835fbd2d8e01553a3f61e9e75)
1 // SPDX-License-Identifier: GPL-2.0
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  *		The IP to API glue.
8  *
9  * Authors:	see ip.c
10  *
11  * Fixes:
12  *		Many		:	Split from ip.c , see ip.c for history.
13  *		Martin Mares	:	TOS setting fixed.
14  *		Alan Cox	:	Fixed a couple of oopses in Martin's
15  *					TOS tweaks.
16  *		Mike McLagan	:	Routing by source
17  */
18 
19 #include <linux/module.h>
20 #include <linux/types.h>
21 #include <linux/mm.h>
22 #include <linux/skbuff.h>
23 #include <linux/ip.h>
24 #include <linux/icmp.h>
25 #include <linux/inetdevice.h>
26 #include <linux/netdevice.h>
27 #include <linux/slab.h>
28 #include <net/sock.h>
29 #include <net/ip.h>
30 #include <net/icmp.h>
31 #include <net/tcp_states.h>
32 #include <linux/udp.h>
33 #include <linux/igmp.h>
34 #include <linux/netfilter.h>
35 #include <linux/route.h>
36 #include <linux/mroute.h>
37 #include <net/inet_ecn.h>
38 #include <net/route.h>
39 #include <net/xfrm.h>
40 #include <net/compat.h>
41 #include <net/checksum.h>
42 #if IS_ENABLED(CONFIG_IPV6)
43 #include <net/transp_v6.h>
44 #endif
45 #include <net/ip_fib.h>
46 
47 #include <linux/errqueue.h>
48 #include <linux/uaccess.h>
49 
50 /*
51  *	SOL_IP control messages.
52  */
53 
54 static void ip_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
55 {
56 	struct in_pktinfo info = *PKTINFO_SKB_CB(skb);
57 
58 	info.ipi_addr.s_addr = ip_hdr(skb)->daddr;
59 
60 	put_cmsg(msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
61 }
62 
63 static void ip_cmsg_recv_ttl(struct msghdr *msg, struct sk_buff *skb)
64 {
65 	int ttl = ip_hdr(skb)->ttl;
66 	put_cmsg(msg, SOL_IP, IP_TTL, sizeof(int), &ttl);
67 }
68 
69 static void ip_cmsg_recv_tos(struct msghdr *msg, struct sk_buff *skb)
70 {
71 	put_cmsg(msg, SOL_IP, IP_TOS, 1, &ip_hdr(skb)->tos);
72 }
73 
74 static void ip_cmsg_recv_opts(struct msghdr *msg, struct sk_buff *skb)
75 {
76 	if (IPCB(skb)->opt.optlen == 0)
77 		return;
78 
79 	put_cmsg(msg, SOL_IP, IP_RECVOPTS, IPCB(skb)->opt.optlen,
80 		 ip_hdr(skb) + 1);
81 }
82 
83 
84 static void ip_cmsg_recv_retopts(struct net *net, struct msghdr *msg,
85 				 struct sk_buff *skb)
86 {
87 	unsigned char optbuf[sizeof(struct ip_options) + 40];
88 	struct ip_options *opt = (struct ip_options *)optbuf;
89 
90 	if (IPCB(skb)->opt.optlen == 0)
91 		return;
92 
93 	if (ip_options_echo(net, opt, skb)) {
94 		msg->msg_flags |= MSG_CTRUNC;
95 		return;
96 	}
97 	ip_options_undo(opt);
98 
99 	put_cmsg(msg, SOL_IP, IP_RETOPTS, opt->optlen, opt->__data);
100 }
101 
102 static void ip_cmsg_recv_fragsize(struct msghdr *msg, struct sk_buff *skb)
103 {
104 	int val;
105 
106 	if (IPCB(skb)->frag_max_size == 0)
107 		return;
108 
109 	val = IPCB(skb)->frag_max_size;
110 	put_cmsg(msg, SOL_IP, IP_RECVFRAGSIZE, sizeof(val), &val);
111 }
112 
113 static void ip_cmsg_recv_checksum(struct msghdr *msg, struct sk_buff *skb,
114 				  int tlen, int offset)
115 {
116 	__wsum csum = skb->csum;
117 
118 	if (skb->ip_summed != CHECKSUM_COMPLETE)
119 		return;
120 
121 	if (offset != 0) {
122 		int tend_off = skb_transport_offset(skb) + tlen;
123 		csum = csum_sub(csum, skb_checksum(skb, tend_off, offset, 0));
124 	}
125 
126 	put_cmsg(msg, SOL_IP, IP_CHECKSUM, sizeof(__wsum), &csum);
127 }
128 
129 static void ip_cmsg_recv_security(struct msghdr *msg, struct sk_buff *skb)
130 {
131 	char *secdata;
132 	u32 seclen, secid;
133 	int err;
134 
135 	err = security_socket_getpeersec_dgram(NULL, skb, &secid);
136 	if (err)
137 		return;
138 
139 	err = security_secid_to_secctx(secid, &secdata, &seclen);
140 	if (err)
141 		return;
142 
143 	put_cmsg(msg, SOL_IP, SCM_SECURITY, seclen, secdata);
144 	security_release_secctx(secdata, seclen);
145 }
146 
147 static void ip_cmsg_recv_dstaddr(struct msghdr *msg, struct sk_buff *skb)
148 {
149 	__be16 _ports[2], *ports;
150 	struct sockaddr_in sin;
151 
152 	/* All current transport protocols have the port numbers in the
153 	 * first four bytes of the transport header and this function is
154 	 * written with this assumption in mind.
155 	 */
156 	ports = skb_header_pointer(skb, skb_transport_offset(skb),
157 				   sizeof(_ports), &_ports);
158 	if (!ports)
159 		return;
160 
161 	sin.sin_family = AF_INET;
162 	sin.sin_addr.s_addr = ip_hdr(skb)->daddr;
163 	sin.sin_port = ports[1];
164 	memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
165 
166 	put_cmsg(msg, SOL_IP, IP_ORIGDSTADDR, sizeof(sin), &sin);
167 }
168 
169 void ip_cmsg_recv_offset(struct msghdr *msg, struct sock *sk,
170 			 struct sk_buff *skb, int tlen, int offset)
171 {
172 	unsigned long flags = inet_cmsg_flags(inet_sk(sk));
173 
174 	if (!flags)
175 		return;
176 
177 	/* Ordered by supposed usage frequency */
178 	if (flags & IP_CMSG_PKTINFO) {
179 		ip_cmsg_recv_pktinfo(msg, skb);
180 
181 		flags &= ~IP_CMSG_PKTINFO;
182 		if (!flags)
183 			return;
184 	}
185 
186 	if (flags & IP_CMSG_TTL) {
187 		ip_cmsg_recv_ttl(msg, skb);
188 
189 		flags &= ~IP_CMSG_TTL;
190 		if (!flags)
191 			return;
192 	}
193 
194 	if (flags & IP_CMSG_TOS) {
195 		ip_cmsg_recv_tos(msg, skb);
196 
197 		flags &= ~IP_CMSG_TOS;
198 		if (!flags)
199 			return;
200 	}
201 
202 	if (flags & IP_CMSG_RECVOPTS) {
203 		ip_cmsg_recv_opts(msg, skb);
204 
205 		flags &= ~IP_CMSG_RECVOPTS;
206 		if (!flags)
207 			return;
208 	}
209 
210 	if (flags & IP_CMSG_RETOPTS) {
211 		ip_cmsg_recv_retopts(sock_net(sk), msg, skb);
212 
213 		flags &= ~IP_CMSG_RETOPTS;
214 		if (!flags)
215 			return;
216 	}
217 
218 	if (flags & IP_CMSG_PASSSEC) {
219 		ip_cmsg_recv_security(msg, skb);
220 
221 		flags &= ~IP_CMSG_PASSSEC;
222 		if (!flags)
223 			return;
224 	}
225 
226 	if (flags & IP_CMSG_ORIGDSTADDR) {
227 		ip_cmsg_recv_dstaddr(msg, skb);
228 
229 		flags &= ~IP_CMSG_ORIGDSTADDR;
230 		if (!flags)
231 			return;
232 	}
233 
234 	if (flags & IP_CMSG_CHECKSUM)
235 		ip_cmsg_recv_checksum(msg, skb, tlen, offset);
236 
237 	if (flags & IP_CMSG_RECVFRAGSIZE)
238 		ip_cmsg_recv_fragsize(msg, skb);
239 }
240 EXPORT_SYMBOL(ip_cmsg_recv_offset);
241 
242 int ip_cmsg_send(struct sock *sk, struct msghdr *msg, struct ipcm_cookie *ipc,
243 		 bool allow_ipv6)
244 {
245 	int err, val;
246 	struct cmsghdr *cmsg;
247 	struct net *net = sock_net(sk);
248 
249 	for_each_cmsghdr(cmsg, msg) {
250 		if (!CMSG_OK(msg, cmsg))
251 			return -EINVAL;
252 #if IS_ENABLED(CONFIG_IPV6)
253 		if (allow_ipv6 &&
254 		    cmsg->cmsg_level == SOL_IPV6 &&
255 		    cmsg->cmsg_type == IPV6_PKTINFO) {
256 			struct in6_pktinfo *src_info;
257 
258 			if (cmsg->cmsg_len < CMSG_LEN(sizeof(*src_info)))
259 				return -EINVAL;
260 			src_info = (struct in6_pktinfo *)CMSG_DATA(cmsg);
261 			if (!ipv6_addr_v4mapped(&src_info->ipi6_addr))
262 				return -EINVAL;
263 			if (src_info->ipi6_ifindex)
264 				ipc->oif = src_info->ipi6_ifindex;
265 			ipc->addr = src_info->ipi6_addr.s6_addr32[3];
266 			continue;
267 		}
268 #endif
269 		if (cmsg->cmsg_level == SOL_SOCKET) {
270 			err = __sock_cmsg_send(sk, cmsg, &ipc->sockc);
271 			if (err)
272 				return err;
273 			continue;
274 		}
275 
276 		if (cmsg->cmsg_level != SOL_IP)
277 			continue;
278 		switch (cmsg->cmsg_type) {
279 		case IP_RETOPTS:
280 			err = cmsg->cmsg_len - sizeof(struct cmsghdr);
281 
282 			/* Our caller is responsible for freeing ipc->opt */
283 			err = ip_options_get(net, &ipc->opt,
284 					     KERNEL_SOCKPTR(CMSG_DATA(cmsg)),
285 					     err < 40 ? err : 40);
286 			if (err)
287 				return err;
288 			break;
289 		case IP_PKTINFO:
290 		{
291 			struct in_pktinfo *info;
292 			if (cmsg->cmsg_len != CMSG_LEN(sizeof(struct in_pktinfo)))
293 				return -EINVAL;
294 			info = (struct in_pktinfo *)CMSG_DATA(cmsg);
295 			if (info->ipi_ifindex)
296 				ipc->oif = info->ipi_ifindex;
297 			ipc->addr = info->ipi_spec_dst.s_addr;
298 			break;
299 		}
300 		case IP_TTL:
301 			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
302 				return -EINVAL;
303 			val = *(int *)CMSG_DATA(cmsg);
304 			if (val < 1 || val > 255)
305 				return -EINVAL;
306 			ipc->ttl = val;
307 			break;
308 		case IP_TOS:
309 			if (cmsg->cmsg_len == CMSG_LEN(sizeof(int)))
310 				val = *(int *)CMSG_DATA(cmsg);
311 			else if (cmsg->cmsg_len == CMSG_LEN(sizeof(u8)))
312 				val = *(u8 *)CMSG_DATA(cmsg);
313 			else
314 				return -EINVAL;
315 			if (val < 0 || val > 255)
316 				return -EINVAL;
317 			ipc->tos = val;
318 			ipc->priority = rt_tos2priority(ipc->tos);
319 			break;
320 		case IP_PROTOCOL:
321 			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
322 				return -EINVAL;
323 			val = *(int *)CMSG_DATA(cmsg);
324 			if (val < 1 || val > 255)
325 				return -EINVAL;
326 			ipc->protocol = val;
327 			break;
328 		default:
329 			return -EINVAL;
330 		}
331 	}
332 	return 0;
333 }
334 
335 static void ip_ra_destroy_rcu(struct rcu_head *head)
336 {
337 	struct ip_ra_chain *ra = container_of(head, struct ip_ra_chain, rcu);
338 
339 	sock_put(ra->saved_sk);
340 	kfree(ra);
341 }
342 
343 int ip_ra_control(struct sock *sk, unsigned char on,
344 		  void (*destructor)(struct sock *))
345 {
346 	struct ip_ra_chain *ra, *new_ra;
347 	struct ip_ra_chain __rcu **rap;
348 	struct net *net = sock_net(sk);
349 
350 	if (sk->sk_type != SOCK_RAW || inet_sk(sk)->inet_num == IPPROTO_RAW)
351 		return -EINVAL;
352 
353 	new_ra = on ? kmalloc(sizeof(*new_ra), GFP_KERNEL) : NULL;
354 	if (on && !new_ra)
355 		return -ENOMEM;
356 
357 	mutex_lock(&net->ipv4.ra_mutex);
358 	for (rap = &net->ipv4.ra_chain;
359 	     (ra = rcu_dereference_protected(*rap,
360 			lockdep_is_held(&net->ipv4.ra_mutex))) != NULL;
361 	     rap = &ra->next) {
362 		if (ra->sk == sk) {
363 			if (on) {
364 				mutex_unlock(&net->ipv4.ra_mutex);
365 				kfree(new_ra);
366 				return -EADDRINUSE;
367 			}
368 			/* dont let ip_call_ra_chain() use sk again */
369 			ra->sk = NULL;
370 			RCU_INIT_POINTER(*rap, ra->next);
371 			mutex_unlock(&net->ipv4.ra_mutex);
372 
373 			if (ra->destructor)
374 				ra->destructor(sk);
375 			/*
376 			 * Delay sock_put(sk) and kfree(ra) after one rcu grace
377 			 * period. This guarantee ip_call_ra_chain() dont need
378 			 * to mess with socket refcounts.
379 			 */
380 			ra->saved_sk = sk;
381 			call_rcu(&ra->rcu, ip_ra_destroy_rcu);
382 			return 0;
383 		}
384 	}
385 	if (!new_ra) {
386 		mutex_unlock(&net->ipv4.ra_mutex);
387 		return -ENOBUFS;
388 	}
389 	new_ra->sk = sk;
390 	new_ra->destructor = destructor;
391 
392 	RCU_INIT_POINTER(new_ra->next, ra);
393 	rcu_assign_pointer(*rap, new_ra);
394 	sock_hold(sk);
395 	mutex_unlock(&net->ipv4.ra_mutex);
396 
397 	return 0;
398 }
399 
400 static void ipv4_icmp_error_rfc4884(const struct sk_buff *skb,
401 				    struct sock_ee_data_rfc4884 *out)
402 {
403 	switch (icmp_hdr(skb)->type) {
404 	case ICMP_DEST_UNREACH:
405 	case ICMP_TIME_EXCEEDED:
406 	case ICMP_PARAMETERPROB:
407 		ip_icmp_error_rfc4884(skb, out, sizeof(struct icmphdr),
408 				      icmp_hdr(skb)->un.reserved[1] * 4);
409 	}
410 }
411 
412 void ip_icmp_error(struct sock *sk, struct sk_buff *skb, int err,
413 		   __be16 port, u32 info, u8 *payload)
414 {
415 	struct sock_exterr_skb *serr;
416 
417 	skb = skb_clone(skb, GFP_ATOMIC);
418 	if (!skb)
419 		return;
420 
421 	serr = SKB_EXT_ERR(skb);
422 	serr->ee.ee_errno = err;
423 	serr->ee.ee_origin = SO_EE_ORIGIN_ICMP;
424 	serr->ee.ee_type = icmp_hdr(skb)->type;
425 	serr->ee.ee_code = icmp_hdr(skb)->code;
426 	serr->ee.ee_pad = 0;
427 	serr->ee.ee_info = info;
428 	serr->ee.ee_data = 0;
429 	serr->addr_offset = (u8 *)&(((struct iphdr *)(icmp_hdr(skb) + 1))->daddr) -
430 				   skb_network_header(skb);
431 	serr->port = port;
432 
433 	if (skb_pull(skb, payload - skb->data)) {
434 		if (inet_test_bit(RECVERR_RFC4884, sk))
435 			ipv4_icmp_error_rfc4884(skb, &serr->ee.ee_rfc4884);
436 
437 		skb_reset_transport_header(skb);
438 		if (sock_queue_err_skb(sk, skb) == 0)
439 			return;
440 	}
441 	kfree_skb(skb);
442 }
443 EXPORT_SYMBOL_GPL(ip_icmp_error);
444 
445 void ip_local_error(struct sock *sk, int err, __be32 daddr, __be16 port, u32 info)
446 {
447 	struct sock_exterr_skb *serr;
448 	struct iphdr *iph;
449 	struct sk_buff *skb;
450 
451 	if (!inet_test_bit(RECVERR, sk))
452 		return;
453 
454 	skb = alloc_skb(sizeof(struct iphdr), GFP_ATOMIC);
455 	if (!skb)
456 		return;
457 
458 	skb_put(skb, sizeof(struct iphdr));
459 	skb_reset_network_header(skb);
460 	iph = ip_hdr(skb);
461 	iph->daddr = daddr;
462 
463 	serr = SKB_EXT_ERR(skb);
464 	serr->ee.ee_errno = err;
465 	serr->ee.ee_origin = SO_EE_ORIGIN_LOCAL;
466 	serr->ee.ee_type = 0;
467 	serr->ee.ee_code = 0;
468 	serr->ee.ee_pad = 0;
469 	serr->ee.ee_info = info;
470 	serr->ee.ee_data = 0;
471 	serr->addr_offset = (u8 *)&iph->daddr - skb_network_header(skb);
472 	serr->port = port;
473 
474 	__skb_pull(skb, skb_tail_pointer(skb) - skb->data);
475 	skb_reset_transport_header(skb);
476 
477 	if (sock_queue_err_skb(sk, skb))
478 		kfree_skb(skb);
479 }
480 
481 /* For some errors we have valid addr_offset even with zero payload and
482  * zero port. Also, addr_offset should be supported if port is set.
483  */
484 static inline bool ipv4_datagram_support_addr(struct sock_exterr_skb *serr)
485 {
486 	return serr->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
487 	       serr->ee.ee_origin == SO_EE_ORIGIN_LOCAL || serr->port;
488 }
489 
490 /* IPv4 supports cmsg on all imcp errors and some timestamps
491  *
492  * Timestamp code paths do not initialize the fields expected by cmsg:
493  * the PKTINFO fields in skb->cb[]. Fill those in here.
494  */
495 static bool ipv4_datagram_support_cmsg(const struct sock *sk,
496 				       struct sk_buff *skb,
497 				       int ee_origin)
498 {
499 	struct in_pktinfo *info;
500 
501 	if (ee_origin == SO_EE_ORIGIN_ICMP)
502 		return true;
503 
504 	if (ee_origin == SO_EE_ORIGIN_LOCAL)
505 		return false;
506 
507 	/* Support IP_PKTINFO on tstamp packets if requested, to correlate
508 	 * timestamp with egress dev. Not possible for packets without iif
509 	 * or without payload (SOF_TIMESTAMPING_OPT_TSONLY).
510 	 */
511 	info = PKTINFO_SKB_CB(skb);
512 	if (!(READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_CMSG) ||
513 	    !info->ipi_ifindex)
514 		return false;
515 
516 	info->ipi_spec_dst.s_addr = ip_hdr(skb)->saddr;
517 	return true;
518 }
519 
520 /*
521  *	Handle MSG_ERRQUEUE
522  */
523 int ip_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
524 {
525 	struct sock_exterr_skb *serr;
526 	struct sk_buff *skb;
527 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
528 	struct {
529 		struct sock_extended_err ee;
530 		struct sockaddr_in	 offender;
531 	} errhdr;
532 	int err;
533 	int copied;
534 
535 	err = -EAGAIN;
536 	skb = sock_dequeue_err_skb(sk);
537 	if (!skb)
538 		goto out;
539 
540 	copied = skb->len;
541 	if (copied > len) {
542 		msg->msg_flags |= MSG_TRUNC;
543 		copied = len;
544 	}
545 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
546 	if (unlikely(err)) {
547 		kfree_skb(skb);
548 		return err;
549 	}
550 	sock_recv_timestamp(msg, sk, skb);
551 
552 	serr = SKB_EXT_ERR(skb);
553 
554 	if (sin && ipv4_datagram_support_addr(serr)) {
555 		sin->sin_family = AF_INET;
556 		sin->sin_addr.s_addr = *(__be32 *)(skb_network_header(skb) +
557 						   serr->addr_offset);
558 		sin->sin_port = serr->port;
559 		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
560 		*addr_len = sizeof(*sin);
561 	}
562 
563 	memcpy(&errhdr.ee, &serr->ee, sizeof(struct sock_extended_err));
564 	sin = &errhdr.offender;
565 	memset(sin, 0, sizeof(*sin));
566 
567 	if (ipv4_datagram_support_cmsg(sk, skb, serr->ee.ee_origin)) {
568 		sin->sin_family = AF_INET;
569 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
570 		if (inet_cmsg_flags(inet_sk(sk)))
571 			ip_cmsg_recv(msg, skb);
572 	}
573 
574 	put_cmsg(msg, SOL_IP, IP_RECVERR, sizeof(errhdr), &errhdr);
575 
576 	/* Now we could try to dump offended packet options */
577 
578 	msg->msg_flags |= MSG_ERRQUEUE;
579 	err = copied;
580 
581 	consume_skb(skb);
582 out:
583 	return err;
584 }
585 
586 void __ip_sock_set_tos(struct sock *sk, int val)
587 {
588 	u8 old_tos = inet_sk(sk)->tos;
589 
590 	if (sk->sk_type == SOCK_STREAM) {
591 		val &= ~INET_ECN_MASK;
592 		val |= old_tos & INET_ECN_MASK;
593 	}
594 	if (old_tos != val) {
595 		WRITE_ONCE(inet_sk(sk)->tos, val);
596 		WRITE_ONCE(sk->sk_priority, rt_tos2priority(val));
597 		sk_dst_reset(sk);
598 	}
599 }
600 
601 void ip_sock_set_tos(struct sock *sk, int val)
602 {
603 	sockopt_lock_sock(sk);
604 	__ip_sock_set_tos(sk, val);
605 	sockopt_release_sock(sk);
606 }
607 EXPORT_SYMBOL(ip_sock_set_tos);
608 
609 void ip_sock_set_freebind(struct sock *sk)
610 {
611 	inet_set_bit(FREEBIND, sk);
612 }
613 EXPORT_SYMBOL(ip_sock_set_freebind);
614 
615 void ip_sock_set_recverr(struct sock *sk)
616 {
617 	inet_set_bit(RECVERR, sk);
618 }
619 EXPORT_SYMBOL(ip_sock_set_recverr);
620 
621 int ip_sock_set_mtu_discover(struct sock *sk, int val)
622 {
623 	if (val < IP_PMTUDISC_DONT || val > IP_PMTUDISC_OMIT)
624 		return -EINVAL;
625 	WRITE_ONCE(inet_sk(sk)->pmtudisc, val);
626 	return 0;
627 }
628 EXPORT_SYMBOL(ip_sock_set_mtu_discover);
629 
630 void ip_sock_set_pktinfo(struct sock *sk)
631 {
632 	inet_set_bit(PKTINFO, sk);
633 }
634 EXPORT_SYMBOL(ip_sock_set_pktinfo);
635 
636 /*
637  *	Socket option code for IP. This is the end of the line after any
638  *	TCP,UDP etc options on an IP socket.
639  */
640 static bool setsockopt_needs_rtnl(int optname)
641 {
642 	switch (optname) {
643 	case IP_ADD_MEMBERSHIP:
644 	case IP_ADD_SOURCE_MEMBERSHIP:
645 	case IP_BLOCK_SOURCE:
646 	case IP_DROP_MEMBERSHIP:
647 	case IP_DROP_SOURCE_MEMBERSHIP:
648 	case IP_MSFILTER:
649 	case IP_UNBLOCK_SOURCE:
650 	case MCAST_BLOCK_SOURCE:
651 	case MCAST_MSFILTER:
652 	case MCAST_JOIN_GROUP:
653 	case MCAST_JOIN_SOURCE_GROUP:
654 	case MCAST_LEAVE_GROUP:
655 	case MCAST_LEAVE_SOURCE_GROUP:
656 	case MCAST_UNBLOCK_SOURCE:
657 		return true;
658 	}
659 	return false;
660 }
661 
662 static int set_mcast_msfilter(struct sock *sk, int ifindex,
663 			      int numsrc, int fmode,
664 			      struct sockaddr_storage *group,
665 			      struct sockaddr_storage *list)
666 {
667 	struct ip_msfilter *msf;
668 	struct sockaddr_in *psin;
669 	int err, i;
670 
671 	msf = kmalloc(IP_MSFILTER_SIZE(numsrc), GFP_KERNEL);
672 	if (!msf)
673 		return -ENOBUFS;
674 
675 	psin = (struct sockaddr_in *)group;
676 	if (psin->sin_family != AF_INET)
677 		goto Eaddrnotavail;
678 	msf->imsf_multiaddr = psin->sin_addr.s_addr;
679 	msf->imsf_interface = 0;
680 	msf->imsf_fmode = fmode;
681 	msf->imsf_numsrc = numsrc;
682 	for (i = 0; i < numsrc; ++i) {
683 		psin = (struct sockaddr_in *)&list[i];
684 
685 		if (psin->sin_family != AF_INET)
686 			goto Eaddrnotavail;
687 		msf->imsf_slist_flex[i] = psin->sin_addr.s_addr;
688 	}
689 	err = ip_mc_msfilter(sk, msf, ifindex);
690 	kfree(msf);
691 	return err;
692 
693 Eaddrnotavail:
694 	kfree(msf);
695 	return -EADDRNOTAVAIL;
696 }
697 
698 static int copy_group_source_from_sockptr(struct group_source_req *greqs,
699 		sockptr_t optval, int optlen)
700 {
701 	if (in_compat_syscall()) {
702 		struct compat_group_source_req gr32;
703 
704 		if (optlen != sizeof(gr32))
705 			return -EINVAL;
706 		if (copy_from_sockptr(&gr32, optval, sizeof(gr32)))
707 			return -EFAULT;
708 		greqs->gsr_interface = gr32.gsr_interface;
709 		greqs->gsr_group = gr32.gsr_group;
710 		greqs->gsr_source = gr32.gsr_source;
711 	} else {
712 		if (optlen != sizeof(*greqs))
713 			return -EINVAL;
714 		if (copy_from_sockptr(greqs, optval, sizeof(*greqs)))
715 			return -EFAULT;
716 	}
717 
718 	return 0;
719 }
720 
721 static int do_mcast_group_source(struct sock *sk, int optname,
722 		sockptr_t optval, int optlen)
723 {
724 	struct group_source_req greqs;
725 	struct ip_mreq_source mreqs;
726 	struct sockaddr_in *psin;
727 	int omode, add, err;
728 
729 	err = copy_group_source_from_sockptr(&greqs, optval, optlen);
730 	if (err)
731 		return err;
732 
733 	if (greqs.gsr_group.ss_family != AF_INET ||
734 	    greqs.gsr_source.ss_family != AF_INET)
735 		return -EADDRNOTAVAIL;
736 
737 	psin = (struct sockaddr_in *)&greqs.gsr_group;
738 	mreqs.imr_multiaddr = psin->sin_addr.s_addr;
739 	psin = (struct sockaddr_in *)&greqs.gsr_source;
740 	mreqs.imr_sourceaddr = psin->sin_addr.s_addr;
741 	mreqs.imr_interface = 0; /* use index for mc_source */
742 
743 	if (optname == MCAST_BLOCK_SOURCE) {
744 		omode = MCAST_EXCLUDE;
745 		add = 1;
746 	} else if (optname == MCAST_UNBLOCK_SOURCE) {
747 		omode = MCAST_EXCLUDE;
748 		add = 0;
749 	} else if (optname == MCAST_JOIN_SOURCE_GROUP) {
750 		struct ip_mreqn mreq;
751 
752 		psin = (struct sockaddr_in *)&greqs.gsr_group;
753 		mreq.imr_multiaddr = psin->sin_addr;
754 		mreq.imr_address.s_addr = 0;
755 		mreq.imr_ifindex = greqs.gsr_interface;
756 		err = ip_mc_join_group_ssm(sk, &mreq, MCAST_INCLUDE);
757 		if (err && err != -EADDRINUSE)
758 			return err;
759 		greqs.gsr_interface = mreq.imr_ifindex;
760 		omode = MCAST_INCLUDE;
761 		add = 1;
762 	} else /* MCAST_LEAVE_SOURCE_GROUP */ {
763 		omode = MCAST_INCLUDE;
764 		add = 0;
765 	}
766 	return ip_mc_source(add, omode, sk, &mreqs, greqs.gsr_interface);
767 }
768 
769 static int ip_set_mcast_msfilter(struct sock *sk, sockptr_t optval, int optlen)
770 {
771 	struct group_filter *gsf = NULL;
772 	int err;
773 
774 	if (optlen < GROUP_FILTER_SIZE(0))
775 		return -EINVAL;
776 	if (optlen > READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
777 		return -ENOBUFS;
778 
779 	gsf = memdup_sockptr(optval, optlen);
780 	if (IS_ERR(gsf))
781 		return PTR_ERR(gsf);
782 
783 	/* numsrc >= (4G-140)/128 overflow in 32 bits */
784 	err = -ENOBUFS;
785 	if (gsf->gf_numsrc >= 0x1ffffff ||
786 	    gsf->gf_numsrc > READ_ONCE(sock_net(sk)->ipv4.sysctl_igmp_max_msf))
787 		goto out_free_gsf;
788 
789 	err = -EINVAL;
790 	if (GROUP_FILTER_SIZE(gsf->gf_numsrc) > optlen)
791 		goto out_free_gsf;
792 
793 	err = set_mcast_msfilter(sk, gsf->gf_interface, gsf->gf_numsrc,
794 				 gsf->gf_fmode, &gsf->gf_group,
795 				 gsf->gf_slist_flex);
796 out_free_gsf:
797 	kfree(gsf);
798 	return err;
799 }
800 
801 static int compat_ip_set_mcast_msfilter(struct sock *sk, sockptr_t optval,
802 		int optlen)
803 {
804 	const int size0 = offsetof(struct compat_group_filter, gf_slist_flex);
805 	struct compat_group_filter *gf32;
806 	unsigned int n;
807 	void *p;
808 	int err;
809 
810 	if (optlen < size0)
811 		return -EINVAL;
812 	if (optlen > READ_ONCE(sock_net(sk)->core.sysctl_optmem_max) - 4)
813 		return -ENOBUFS;
814 
815 	p = kmalloc(optlen + 4, GFP_KERNEL);
816 	if (!p)
817 		return -ENOMEM;
818 	gf32 = p + 4; /* we want ->gf_group and ->gf_slist_flex aligned */
819 
820 	err = -EFAULT;
821 	if (copy_from_sockptr(gf32, optval, optlen))
822 		goto out_free_gsf;
823 
824 	/* numsrc >= (4G-140)/128 overflow in 32 bits */
825 	n = gf32->gf_numsrc;
826 	err = -ENOBUFS;
827 	if (n >= 0x1ffffff)
828 		goto out_free_gsf;
829 
830 	err = -EINVAL;
831 	if (offsetof(struct compat_group_filter, gf_slist_flex[n]) > optlen)
832 		goto out_free_gsf;
833 
834 	/* numsrc >= (4G-140)/128 overflow in 32 bits */
835 	err = -ENOBUFS;
836 	if (n > READ_ONCE(sock_net(sk)->ipv4.sysctl_igmp_max_msf))
837 		goto out_free_gsf;
838 	err = set_mcast_msfilter(sk, gf32->gf_interface, n, gf32->gf_fmode,
839 				 &gf32->gf_group, gf32->gf_slist_flex);
840 out_free_gsf:
841 	kfree(p);
842 	return err;
843 }
844 
845 static int ip_mcast_join_leave(struct sock *sk, int optname,
846 		sockptr_t optval, int optlen)
847 {
848 	struct ip_mreqn mreq = { };
849 	struct sockaddr_in *psin;
850 	struct group_req greq;
851 
852 	if (optlen < sizeof(struct group_req))
853 		return -EINVAL;
854 	if (copy_from_sockptr(&greq, optval, sizeof(greq)))
855 		return -EFAULT;
856 
857 	psin = (struct sockaddr_in *)&greq.gr_group;
858 	if (psin->sin_family != AF_INET)
859 		return -EINVAL;
860 	mreq.imr_multiaddr = psin->sin_addr;
861 	mreq.imr_ifindex = greq.gr_interface;
862 	if (optname == MCAST_JOIN_GROUP)
863 		return ip_mc_join_group(sk, &mreq);
864 	return ip_mc_leave_group(sk, &mreq);
865 }
866 
867 static int compat_ip_mcast_join_leave(struct sock *sk, int optname,
868 		sockptr_t optval, int optlen)
869 {
870 	struct compat_group_req greq;
871 	struct ip_mreqn mreq = { };
872 	struct sockaddr_in *psin;
873 
874 	if (optlen < sizeof(struct compat_group_req))
875 		return -EINVAL;
876 	if (copy_from_sockptr(&greq, optval, sizeof(greq)))
877 		return -EFAULT;
878 
879 	psin = (struct sockaddr_in *)&greq.gr_group;
880 	if (psin->sin_family != AF_INET)
881 		return -EINVAL;
882 	mreq.imr_multiaddr = psin->sin_addr;
883 	mreq.imr_ifindex = greq.gr_interface;
884 
885 	if (optname == MCAST_JOIN_GROUP)
886 		return ip_mc_join_group(sk, &mreq);
887 	return ip_mc_leave_group(sk, &mreq);
888 }
889 
890 DEFINE_STATIC_KEY_FALSE(ip4_min_ttl);
891 
892 int do_ip_setsockopt(struct sock *sk, int level, int optname,
893 		     sockptr_t optval, unsigned int optlen)
894 {
895 	struct inet_sock *inet = inet_sk(sk);
896 	struct net *net = sock_net(sk);
897 	int val = 0, err;
898 	bool needs_rtnl = setsockopt_needs_rtnl(optname);
899 
900 	switch (optname) {
901 	case IP_PKTINFO:
902 	case IP_RECVTTL:
903 	case IP_RECVOPTS:
904 	case IP_RECVTOS:
905 	case IP_RETOPTS:
906 	case IP_TOS:
907 	case IP_TTL:
908 	case IP_HDRINCL:
909 	case IP_MTU_DISCOVER:
910 	case IP_RECVERR:
911 	case IP_ROUTER_ALERT:
912 	case IP_FREEBIND:
913 	case IP_PASSSEC:
914 	case IP_TRANSPARENT:
915 	case IP_MINTTL:
916 	case IP_NODEFRAG:
917 	case IP_BIND_ADDRESS_NO_PORT:
918 	case IP_UNICAST_IF:
919 	case IP_MULTICAST_TTL:
920 	case IP_MULTICAST_ALL:
921 	case IP_MULTICAST_LOOP:
922 	case IP_RECVORIGDSTADDR:
923 	case IP_CHECKSUM:
924 	case IP_RECVFRAGSIZE:
925 	case IP_RECVERR_RFC4884:
926 	case IP_LOCAL_PORT_RANGE:
927 		if (optlen >= sizeof(int)) {
928 			if (copy_from_sockptr(&val, optval, sizeof(val)))
929 				return -EFAULT;
930 		} else if (optlen >= sizeof(char)) {
931 			unsigned char ucval;
932 
933 			if (copy_from_sockptr(&ucval, optval, sizeof(ucval)))
934 				return -EFAULT;
935 			val = (int) ucval;
936 		}
937 	}
938 
939 	/* If optlen==0, it is equivalent to val == 0 */
940 
941 	if (optname == IP_ROUTER_ALERT)
942 		return ip_ra_control(sk, val ? 1 : 0, NULL);
943 	if (ip_mroute_opt(optname))
944 		return ip_mroute_setsockopt(sk, optname, optval, optlen);
945 
946 	/* Handle options that can be set without locking the socket. */
947 	switch (optname) {
948 	case IP_PKTINFO:
949 		inet_assign_bit(PKTINFO, sk, val);
950 		return 0;
951 	case IP_RECVTTL:
952 		inet_assign_bit(TTL, sk, val);
953 		return 0;
954 	case IP_RECVTOS:
955 		inet_assign_bit(TOS, sk, val);
956 		return 0;
957 	case IP_RECVOPTS:
958 		inet_assign_bit(RECVOPTS, sk, val);
959 		return 0;
960 	case IP_RETOPTS:
961 		inet_assign_bit(RETOPTS, sk, val);
962 		return 0;
963 	case IP_PASSSEC:
964 		inet_assign_bit(PASSSEC, sk, val);
965 		return 0;
966 	case IP_RECVORIGDSTADDR:
967 		inet_assign_bit(ORIGDSTADDR, sk, val);
968 		return 0;
969 	case IP_RECVFRAGSIZE:
970 		if (sk->sk_type != SOCK_RAW && sk->sk_type != SOCK_DGRAM)
971 			return -EINVAL;
972 		inet_assign_bit(RECVFRAGSIZE, sk, val);
973 		return 0;
974 	case IP_RECVERR:
975 		inet_assign_bit(RECVERR, sk, val);
976 		if (!val)
977 			skb_errqueue_purge(&sk->sk_error_queue);
978 		return 0;
979 	case IP_RECVERR_RFC4884:
980 		if (val < 0 || val > 1)
981 			return -EINVAL;
982 		inet_assign_bit(RECVERR_RFC4884, sk, val);
983 		return 0;
984 	case IP_FREEBIND:
985 		if (optlen < 1)
986 			return -EINVAL;
987 		inet_assign_bit(FREEBIND, sk, val);
988 		return 0;
989 	case IP_HDRINCL:
990 		if (sk->sk_type != SOCK_RAW)
991 			return -ENOPROTOOPT;
992 		inet_assign_bit(HDRINCL, sk, val);
993 		return 0;
994 	case IP_MULTICAST_LOOP:
995 		if (optlen < 1)
996 			return -EINVAL;
997 		inet_assign_bit(MC_LOOP, sk, val);
998 		return 0;
999 	case IP_MULTICAST_ALL:
1000 		if (optlen < 1)
1001 			return -EINVAL;
1002 		if (val != 0 && val != 1)
1003 			return -EINVAL;
1004 		inet_assign_bit(MC_ALL, sk, val);
1005 		return 0;
1006 	case IP_TRANSPARENT:
1007 		if (!!val && !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1008 		    !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1009 			return -EPERM;
1010 		if (optlen < 1)
1011 			return -EINVAL;
1012 		inet_assign_bit(TRANSPARENT, sk, val);
1013 		return 0;
1014 	case IP_NODEFRAG:
1015 		if (sk->sk_type != SOCK_RAW)
1016 			return -ENOPROTOOPT;
1017 		inet_assign_bit(NODEFRAG, sk, val);
1018 		return 0;
1019 	case IP_BIND_ADDRESS_NO_PORT:
1020 		inet_assign_bit(BIND_ADDRESS_NO_PORT, sk, val);
1021 		return 0;
1022 	case IP_TTL:
1023 		if (optlen < 1)
1024 			return -EINVAL;
1025 		if (val != -1 && (val < 1 || val > 255))
1026 			return -EINVAL;
1027 		WRITE_ONCE(inet->uc_ttl, val);
1028 		return 0;
1029 	case IP_MINTTL:
1030 		if (optlen < 1)
1031 			return -EINVAL;
1032 		if (val < 0 || val > 255)
1033 			return -EINVAL;
1034 
1035 		if (val)
1036 			static_branch_enable(&ip4_min_ttl);
1037 
1038 		WRITE_ONCE(inet->min_ttl, val);
1039 		return 0;
1040 	case IP_MULTICAST_TTL:
1041 		if (sk->sk_type == SOCK_STREAM)
1042 			return -EINVAL;
1043 		if (optlen < 1)
1044 			return -EINVAL;
1045 		if (val == -1)
1046 			val = 1;
1047 		if (val < 0 || val > 255)
1048 			return -EINVAL;
1049 		WRITE_ONCE(inet->mc_ttl, val);
1050 		return 0;
1051 	case IP_MTU_DISCOVER:
1052 		return ip_sock_set_mtu_discover(sk, val);
1053 	case IP_TOS:	/* This sets both TOS and Precedence */
1054 		ip_sock_set_tos(sk, val);
1055 		return 0;
1056 	case IP_LOCAL_PORT_RANGE:
1057 	{
1058 		u16 lo = val;
1059 		u16 hi = val >> 16;
1060 
1061 		if (optlen != sizeof(u32))
1062 			return -EINVAL;
1063 		if (lo != 0 && hi != 0 && lo > hi)
1064 			return -EINVAL;
1065 
1066 		WRITE_ONCE(inet->local_port_range, val);
1067 		return 0;
1068 	}
1069 	}
1070 
1071 	err = 0;
1072 	if (needs_rtnl)
1073 		rtnl_lock();
1074 	sockopt_lock_sock(sk);
1075 
1076 	switch (optname) {
1077 	case IP_OPTIONS:
1078 	{
1079 		struct ip_options_rcu *old, *opt = NULL;
1080 
1081 		if (optlen > 40)
1082 			goto e_inval;
1083 		err = ip_options_get(sock_net(sk), &opt, optval, optlen);
1084 		if (err)
1085 			break;
1086 		old = rcu_dereference_protected(inet->inet_opt,
1087 						lockdep_sock_is_held(sk));
1088 		if (inet_test_bit(IS_ICSK, sk)) {
1089 			struct inet_connection_sock *icsk = inet_csk(sk);
1090 #if IS_ENABLED(CONFIG_IPV6)
1091 			if (sk->sk_family == PF_INET ||
1092 			    (!((1 << sk->sk_state) &
1093 			       (TCPF_LISTEN | TCPF_CLOSE)) &&
1094 			     inet->inet_daddr != LOOPBACK4_IPV6)) {
1095 #endif
1096 				if (old)
1097 					icsk->icsk_ext_hdr_len -= old->opt.optlen;
1098 				if (opt)
1099 					icsk->icsk_ext_hdr_len += opt->opt.optlen;
1100 				icsk->icsk_sync_mss(sk, icsk->icsk_pmtu_cookie);
1101 #if IS_ENABLED(CONFIG_IPV6)
1102 			}
1103 #endif
1104 		}
1105 		rcu_assign_pointer(inet->inet_opt, opt);
1106 		if (old)
1107 			kfree_rcu(old, rcu);
1108 		break;
1109 	}
1110 	case IP_CHECKSUM:
1111 		if (val) {
1112 			if (!(inet_test_bit(CHECKSUM, sk))) {
1113 				inet_inc_convert_csum(sk);
1114 				inet_set_bit(CHECKSUM, sk);
1115 			}
1116 		} else {
1117 			if (inet_test_bit(CHECKSUM, sk)) {
1118 				inet_dec_convert_csum(sk);
1119 				inet_clear_bit(CHECKSUM, sk);
1120 			}
1121 		}
1122 		break;
1123 	case IP_UNICAST_IF:
1124 	{
1125 		struct net_device *dev = NULL;
1126 		int ifindex;
1127 		int midx;
1128 
1129 		if (optlen != sizeof(int))
1130 			goto e_inval;
1131 
1132 		ifindex = (__force int)ntohl((__force __be32)val);
1133 		if (ifindex == 0) {
1134 			WRITE_ONCE(inet->uc_index, 0);
1135 			err = 0;
1136 			break;
1137 		}
1138 
1139 		dev = dev_get_by_index(sock_net(sk), ifindex);
1140 		err = -EADDRNOTAVAIL;
1141 		if (!dev)
1142 			break;
1143 
1144 		midx = l3mdev_master_ifindex(dev);
1145 		dev_put(dev);
1146 
1147 		err = -EINVAL;
1148 		if (sk->sk_bound_dev_if && midx != sk->sk_bound_dev_if)
1149 			break;
1150 
1151 		WRITE_ONCE(inet->uc_index, ifindex);
1152 		err = 0;
1153 		break;
1154 	}
1155 	case IP_MULTICAST_IF:
1156 	{
1157 		struct ip_mreqn mreq;
1158 		struct net_device *dev = NULL;
1159 		int midx;
1160 
1161 		if (sk->sk_type == SOCK_STREAM)
1162 			goto e_inval;
1163 		/*
1164 		 *	Check the arguments are allowable
1165 		 */
1166 
1167 		if (optlen < sizeof(struct in_addr))
1168 			goto e_inval;
1169 
1170 		err = -EFAULT;
1171 		if (optlen >= sizeof(struct ip_mreqn)) {
1172 			if (copy_from_sockptr(&mreq, optval, sizeof(mreq)))
1173 				break;
1174 		} else {
1175 			memset(&mreq, 0, sizeof(mreq));
1176 			if (optlen >= sizeof(struct ip_mreq)) {
1177 				if (copy_from_sockptr(&mreq, optval,
1178 						      sizeof(struct ip_mreq)))
1179 					break;
1180 			} else if (optlen >= sizeof(struct in_addr)) {
1181 				if (copy_from_sockptr(&mreq.imr_address, optval,
1182 						      sizeof(struct in_addr)))
1183 					break;
1184 			}
1185 		}
1186 
1187 		if (!mreq.imr_ifindex) {
1188 			if (mreq.imr_address.s_addr == htonl(INADDR_ANY)) {
1189 				WRITE_ONCE(inet->mc_index, 0);
1190 				WRITE_ONCE(inet->mc_addr, 0);
1191 				err = 0;
1192 				break;
1193 			}
1194 			dev = ip_dev_find(sock_net(sk), mreq.imr_address.s_addr);
1195 			if (dev)
1196 				mreq.imr_ifindex = dev->ifindex;
1197 		} else
1198 			dev = dev_get_by_index(sock_net(sk), mreq.imr_ifindex);
1199 
1200 
1201 		err = -EADDRNOTAVAIL;
1202 		if (!dev)
1203 			break;
1204 
1205 		midx = l3mdev_master_ifindex(dev);
1206 
1207 		dev_put(dev);
1208 
1209 		err = -EINVAL;
1210 		if (sk->sk_bound_dev_if &&
1211 		    mreq.imr_ifindex != sk->sk_bound_dev_if &&
1212 		    midx != sk->sk_bound_dev_if)
1213 			break;
1214 
1215 		WRITE_ONCE(inet->mc_index, mreq.imr_ifindex);
1216 		WRITE_ONCE(inet->mc_addr, mreq.imr_address.s_addr);
1217 		err = 0;
1218 		break;
1219 	}
1220 
1221 	case IP_ADD_MEMBERSHIP:
1222 	case IP_DROP_MEMBERSHIP:
1223 	{
1224 		struct ip_mreqn mreq;
1225 
1226 		err = -EPROTO;
1227 		if (inet_test_bit(IS_ICSK, sk))
1228 			break;
1229 
1230 		if (optlen < sizeof(struct ip_mreq))
1231 			goto e_inval;
1232 		err = -EFAULT;
1233 		if (optlen >= sizeof(struct ip_mreqn)) {
1234 			if (copy_from_sockptr(&mreq, optval, sizeof(mreq)))
1235 				break;
1236 		} else {
1237 			memset(&mreq, 0, sizeof(mreq));
1238 			if (copy_from_sockptr(&mreq, optval,
1239 					      sizeof(struct ip_mreq)))
1240 				break;
1241 		}
1242 
1243 		if (optname == IP_ADD_MEMBERSHIP)
1244 			err = ip_mc_join_group(sk, &mreq);
1245 		else
1246 			err = ip_mc_leave_group(sk, &mreq);
1247 		break;
1248 	}
1249 	case IP_MSFILTER:
1250 	{
1251 		struct ip_msfilter *msf;
1252 
1253 		if (optlen < IP_MSFILTER_SIZE(0))
1254 			goto e_inval;
1255 		if (optlen > READ_ONCE(net->core.sysctl_optmem_max)) {
1256 			err = -ENOBUFS;
1257 			break;
1258 		}
1259 		msf = memdup_sockptr(optval, optlen);
1260 		if (IS_ERR(msf)) {
1261 			err = PTR_ERR(msf);
1262 			break;
1263 		}
1264 		/* numsrc >= (1G-4) overflow in 32 bits */
1265 		if (msf->imsf_numsrc >= 0x3ffffffcU ||
1266 		    msf->imsf_numsrc > READ_ONCE(net->ipv4.sysctl_igmp_max_msf)) {
1267 			kfree(msf);
1268 			err = -ENOBUFS;
1269 			break;
1270 		}
1271 		if (IP_MSFILTER_SIZE(msf->imsf_numsrc) > optlen) {
1272 			kfree(msf);
1273 			err = -EINVAL;
1274 			break;
1275 		}
1276 		err = ip_mc_msfilter(sk, msf, 0);
1277 		kfree(msf);
1278 		break;
1279 	}
1280 	case IP_BLOCK_SOURCE:
1281 	case IP_UNBLOCK_SOURCE:
1282 	case IP_ADD_SOURCE_MEMBERSHIP:
1283 	case IP_DROP_SOURCE_MEMBERSHIP:
1284 	{
1285 		struct ip_mreq_source mreqs;
1286 		int omode, add;
1287 
1288 		if (optlen != sizeof(struct ip_mreq_source))
1289 			goto e_inval;
1290 		if (copy_from_sockptr(&mreqs, optval, sizeof(mreqs))) {
1291 			err = -EFAULT;
1292 			break;
1293 		}
1294 		if (optname == IP_BLOCK_SOURCE) {
1295 			omode = MCAST_EXCLUDE;
1296 			add = 1;
1297 		} else if (optname == IP_UNBLOCK_SOURCE) {
1298 			omode = MCAST_EXCLUDE;
1299 			add = 0;
1300 		} else if (optname == IP_ADD_SOURCE_MEMBERSHIP) {
1301 			struct ip_mreqn mreq;
1302 
1303 			mreq.imr_multiaddr.s_addr = mreqs.imr_multiaddr;
1304 			mreq.imr_address.s_addr = mreqs.imr_interface;
1305 			mreq.imr_ifindex = 0;
1306 			err = ip_mc_join_group_ssm(sk, &mreq, MCAST_INCLUDE);
1307 			if (err && err != -EADDRINUSE)
1308 				break;
1309 			omode = MCAST_INCLUDE;
1310 			add = 1;
1311 		} else /* IP_DROP_SOURCE_MEMBERSHIP */ {
1312 			omode = MCAST_INCLUDE;
1313 			add = 0;
1314 		}
1315 		err = ip_mc_source(add, omode, sk, &mreqs, 0);
1316 		break;
1317 	}
1318 	case MCAST_JOIN_GROUP:
1319 	case MCAST_LEAVE_GROUP:
1320 		if (in_compat_syscall())
1321 			err = compat_ip_mcast_join_leave(sk, optname, optval,
1322 							 optlen);
1323 		else
1324 			err = ip_mcast_join_leave(sk, optname, optval, optlen);
1325 		break;
1326 	case MCAST_JOIN_SOURCE_GROUP:
1327 	case MCAST_LEAVE_SOURCE_GROUP:
1328 	case MCAST_BLOCK_SOURCE:
1329 	case MCAST_UNBLOCK_SOURCE:
1330 		err = do_mcast_group_source(sk, optname, optval, optlen);
1331 		break;
1332 	case MCAST_MSFILTER:
1333 		if (in_compat_syscall())
1334 			err = compat_ip_set_mcast_msfilter(sk, optval, optlen);
1335 		else
1336 			err = ip_set_mcast_msfilter(sk, optval, optlen);
1337 		break;
1338 	case IP_IPSEC_POLICY:
1339 	case IP_XFRM_POLICY:
1340 		err = -EPERM;
1341 		if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1342 			break;
1343 		err = xfrm_user_policy(sk, optname, optval, optlen);
1344 		break;
1345 
1346 	default:
1347 		err = -ENOPROTOOPT;
1348 		break;
1349 	}
1350 	sockopt_release_sock(sk);
1351 	if (needs_rtnl)
1352 		rtnl_unlock();
1353 	return err;
1354 
1355 e_inval:
1356 	sockopt_release_sock(sk);
1357 	if (needs_rtnl)
1358 		rtnl_unlock();
1359 	return -EINVAL;
1360 }
1361 
1362 /**
1363  * ipv4_pktinfo_prepare - transfer some info from rtable to skb
1364  * @sk: socket
1365  * @skb: buffer
1366  *
1367  * To support IP_CMSG_PKTINFO option, we store rt_iif and specific
1368  * destination in skb->cb[] before dst drop.
1369  * This way, receiver doesn't make cache line misses to read rtable.
1370  */
1371 void ipv4_pktinfo_prepare(const struct sock *sk, struct sk_buff *skb)
1372 {
1373 	struct in_pktinfo *pktinfo = PKTINFO_SKB_CB(skb);
1374 	bool prepare = inet_test_bit(PKTINFO, sk) ||
1375 		       ipv6_sk_rxinfo(sk);
1376 
1377 	if (prepare && skb_rtable(skb)) {
1378 		/* skb->cb is overloaded: prior to this point it is IP{6}CB
1379 		 * which has interface index (iif) as the first member of the
1380 		 * underlying inet{6}_skb_parm struct. This code then overlays
1381 		 * PKTINFO_SKB_CB and in_pktinfo also has iif as the first
1382 		 * element so the iif is picked up from the prior IPCB. If iif
1383 		 * is the loopback interface, then return the sending interface
1384 		 * (e.g., process binds socket to eth0 for Tx which is
1385 		 * redirected to loopback in the rtable/dst).
1386 		 */
1387 		struct rtable *rt = skb_rtable(skb);
1388 		bool l3slave = ipv4_l3mdev_skb(IPCB(skb)->flags);
1389 
1390 		if (pktinfo->ipi_ifindex == LOOPBACK_IFINDEX)
1391 			pktinfo->ipi_ifindex = inet_iif(skb);
1392 		else if (l3slave && rt && rt->rt_iif)
1393 			pktinfo->ipi_ifindex = rt->rt_iif;
1394 
1395 		pktinfo->ipi_spec_dst.s_addr = fib_compute_spec_dst(skb);
1396 	} else {
1397 		pktinfo->ipi_ifindex = 0;
1398 		pktinfo->ipi_spec_dst.s_addr = 0;
1399 	}
1400 	skb_dst_drop(skb);
1401 }
1402 
1403 int ip_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1404 		unsigned int optlen)
1405 {
1406 	int err;
1407 
1408 	if (level != SOL_IP)
1409 		return -ENOPROTOOPT;
1410 
1411 	err = do_ip_setsockopt(sk, level, optname, optval, optlen);
1412 #ifdef CONFIG_NETFILTER
1413 	/* we need to exclude all possible ENOPROTOOPTs except default case */
1414 	if (err == -ENOPROTOOPT && optname != IP_HDRINCL &&
1415 			optname != IP_IPSEC_POLICY &&
1416 			optname != IP_XFRM_POLICY &&
1417 			!ip_mroute_opt(optname))
1418 		err = nf_setsockopt(sk, PF_INET, optname, optval, optlen);
1419 #endif
1420 	return err;
1421 }
1422 EXPORT_SYMBOL(ip_setsockopt);
1423 
1424 /*
1425  *	Get the options. Note for future reference. The GET of IP options gets
1426  *	the _received_ ones. The set sets the _sent_ ones.
1427  */
1428 
1429 static bool getsockopt_needs_rtnl(int optname)
1430 {
1431 	switch (optname) {
1432 	case IP_MSFILTER:
1433 	case MCAST_MSFILTER:
1434 		return true;
1435 	}
1436 	return false;
1437 }
1438 
1439 static int ip_get_mcast_msfilter(struct sock *sk, sockptr_t optval,
1440 				 sockptr_t optlen, int len)
1441 {
1442 	const int size0 = offsetof(struct group_filter, gf_slist_flex);
1443 	struct group_filter gsf;
1444 	int num, gsf_size;
1445 	int err;
1446 
1447 	if (len < size0)
1448 		return -EINVAL;
1449 	if (copy_from_sockptr(&gsf, optval, size0))
1450 		return -EFAULT;
1451 
1452 	num = gsf.gf_numsrc;
1453 	err = ip_mc_gsfget(sk, &gsf, optval,
1454 			   offsetof(struct group_filter, gf_slist_flex));
1455 	if (err)
1456 		return err;
1457 	if (gsf.gf_numsrc < num)
1458 		num = gsf.gf_numsrc;
1459 	gsf_size = GROUP_FILTER_SIZE(num);
1460 	if (copy_to_sockptr(optlen, &gsf_size, sizeof(int)) ||
1461 	    copy_to_sockptr(optval, &gsf, size0))
1462 		return -EFAULT;
1463 	return 0;
1464 }
1465 
1466 static int compat_ip_get_mcast_msfilter(struct sock *sk, sockptr_t optval,
1467 					sockptr_t optlen, int len)
1468 {
1469 	const int size0 = offsetof(struct compat_group_filter, gf_slist_flex);
1470 	struct compat_group_filter gf32;
1471 	struct group_filter gf;
1472 	int num;
1473 	int err;
1474 
1475 	if (len < size0)
1476 		return -EINVAL;
1477 	if (copy_from_sockptr(&gf32, optval, size0))
1478 		return -EFAULT;
1479 
1480 	gf.gf_interface = gf32.gf_interface;
1481 	gf.gf_fmode = gf32.gf_fmode;
1482 	num = gf.gf_numsrc = gf32.gf_numsrc;
1483 	gf.gf_group = gf32.gf_group;
1484 
1485 	err = ip_mc_gsfget(sk, &gf, optval,
1486 			   offsetof(struct compat_group_filter, gf_slist_flex));
1487 	if (err)
1488 		return err;
1489 	if (gf.gf_numsrc < num)
1490 		num = gf.gf_numsrc;
1491 	len = GROUP_FILTER_SIZE(num) - (sizeof(gf) - sizeof(gf32));
1492 	if (copy_to_sockptr(optlen, &len, sizeof(int)) ||
1493 	    copy_to_sockptr_offset(optval, offsetof(struct compat_group_filter, gf_fmode),
1494 				   &gf.gf_fmode, sizeof(gf.gf_fmode)) ||
1495 	    copy_to_sockptr_offset(optval, offsetof(struct compat_group_filter, gf_numsrc),
1496 				   &gf.gf_numsrc, sizeof(gf.gf_numsrc)))
1497 		return -EFAULT;
1498 	return 0;
1499 }
1500 
1501 int do_ip_getsockopt(struct sock *sk, int level, int optname,
1502 		     sockptr_t optval, sockptr_t optlen)
1503 {
1504 	struct inet_sock *inet = inet_sk(sk);
1505 	bool needs_rtnl = getsockopt_needs_rtnl(optname);
1506 	int val, err = 0;
1507 	int len;
1508 
1509 	if (level != SOL_IP)
1510 		return -EOPNOTSUPP;
1511 
1512 	if (ip_mroute_opt(optname))
1513 		return ip_mroute_getsockopt(sk, optname, optval, optlen);
1514 
1515 	if (copy_from_sockptr(&len, optlen, sizeof(int)))
1516 		return -EFAULT;
1517 	if (len < 0)
1518 		return -EINVAL;
1519 
1520 	/* Handle options that can be read without locking the socket. */
1521 	switch (optname) {
1522 	case IP_PKTINFO:
1523 		val = inet_test_bit(PKTINFO, sk);
1524 		goto copyval;
1525 	case IP_RECVTTL:
1526 		val = inet_test_bit(TTL, sk);
1527 		goto copyval;
1528 	case IP_RECVTOS:
1529 		val = inet_test_bit(TOS, sk);
1530 		goto copyval;
1531 	case IP_RECVOPTS:
1532 		val = inet_test_bit(RECVOPTS, sk);
1533 		goto copyval;
1534 	case IP_RETOPTS:
1535 		val = inet_test_bit(RETOPTS, sk);
1536 		goto copyval;
1537 	case IP_PASSSEC:
1538 		val = inet_test_bit(PASSSEC, sk);
1539 		goto copyval;
1540 	case IP_RECVORIGDSTADDR:
1541 		val = inet_test_bit(ORIGDSTADDR, sk);
1542 		goto copyval;
1543 	case IP_CHECKSUM:
1544 		val = inet_test_bit(CHECKSUM, sk);
1545 		goto copyval;
1546 	case IP_RECVFRAGSIZE:
1547 		val = inet_test_bit(RECVFRAGSIZE, sk);
1548 		goto copyval;
1549 	case IP_RECVERR:
1550 		val = inet_test_bit(RECVERR, sk);
1551 		goto copyval;
1552 	case IP_RECVERR_RFC4884:
1553 		val = inet_test_bit(RECVERR_RFC4884, sk);
1554 		goto copyval;
1555 	case IP_FREEBIND:
1556 		val = inet_test_bit(FREEBIND, sk);
1557 		goto copyval;
1558 	case IP_HDRINCL:
1559 		val = inet_test_bit(HDRINCL, sk);
1560 		goto copyval;
1561 	case IP_MULTICAST_LOOP:
1562 		val = inet_test_bit(MC_LOOP, sk);
1563 		goto copyval;
1564 	case IP_MULTICAST_ALL:
1565 		val = inet_test_bit(MC_ALL, sk);
1566 		goto copyval;
1567 	case IP_TRANSPARENT:
1568 		val = inet_test_bit(TRANSPARENT, sk);
1569 		goto copyval;
1570 	case IP_NODEFRAG:
1571 		val = inet_test_bit(NODEFRAG, sk);
1572 		goto copyval;
1573 	case IP_BIND_ADDRESS_NO_PORT:
1574 		val = inet_test_bit(BIND_ADDRESS_NO_PORT, sk);
1575 		goto copyval;
1576 	case IP_TTL:
1577 		val = READ_ONCE(inet->uc_ttl);
1578 		if (val < 0)
1579 			val = READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_default_ttl);
1580 		goto copyval;
1581 	case IP_MINTTL:
1582 		val = READ_ONCE(inet->min_ttl);
1583 		goto copyval;
1584 	case IP_MULTICAST_TTL:
1585 		val = READ_ONCE(inet->mc_ttl);
1586 		goto copyval;
1587 	case IP_MTU_DISCOVER:
1588 		val = READ_ONCE(inet->pmtudisc);
1589 		goto copyval;
1590 	case IP_TOS:
1591 		val = READ_ONCE(inet->tos);
1592 		goto copyval;
1593 	case IP_OPTIONS:
1594 	{
1595 		unsigned char optbuf[sizeof(struct ip_options)+40];
1596 		struct ip_options *opt = (struct ip_options *)optbuf;
1597 		struct ip_options_rcu *inet_opt;
1598 
1599 		rcu_read_lock();
1600 		inet_opt = rcu_dereference(inet->inet_opt);
1601 		opt->optlen = 0;
1602 		if (inet_opt)
1603 			memcpy(optbuf, &inet_opt->opt,
1604 			       sizeof(struct ip_options) +
1605 			       inet_opt->opt.optlen);
1606 		rcu_read_unlock();
1607 
1608 		if (opt->optlen == 0) {
1609 			len = 0;
1610 			return copy_to_sockptr(optlen, &len, sizeof(int));
1611 		}
1612 
1613 		ip_options_undo(opt);
1614 
1615 		len = min_t(unsigned int, len, opt->optlen);
1616 		if (copy_to_sockptr(optlen, &len, sizeof(int)))
1617 			return -EFAULT;
1618 		if (copy_to_sockptr(optval, opt->__data, len))
1619 			return -EFAULT;
1620 		return 0;
1621 	}
1622 	case IP_MTU:
1623 	{
1624 		struct dst_entry *dst;
1625 		val = 0;
1626 		dst = sk_dst_get(sk);
1627 		if (dst) {
1628 			val = dst_mtu(dst);
1629 			dst_release(dst);
1630 		}
1631 		if (!val)
1632 			return -ENOTCONN;
1633 		goto copyval;
1634 	}
1635 	case IP_PKTOPTIONS:
1636 	{
1637 		struct msghdr msg;
1638 
1639 		if (sk->sk_type != SOCK_STREAM)
1640 			return -ENOPROTOOPT;
1641 
1642 		if (optval.is_kernel) {
1643 			msg.msg_control_is_user = false;
1644 			msg.msg_control = optval.kernel;
1645 		} else {
1646 			msg.msg_control_is_user = true;
1647 			msg.msg_control_user = optval.user;
1648 		}
1649 		msg.msg_controllen = len;
1650 		msg.msg_flags = in_compat_syscall() ? MSG_CMSG_COMPAT : 0;
1651 
1652 		if (inet_test_bit(PKTINFO, sk)) {
1653 			struct in_pktinfo info;
1654 
1655 			info.ipi_addr.s_addr = READ_ONCE(inet->inet_rcv_saddr);
1656 			info.ipi_spec_dst.s_addr = READ_ONCE(inet->inet_rcv_saddr);
1657 			info.ipi_ifindex = READ_ONCE(inet->mc_index);
1658 			put_cmsg(&msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
1659 		}
1660 		if (inet_test_bit(TTL, sk)) {
1661 			int hlim = READ_ONCE(inet->mc_ttl);
1662 
1663 			put_cmsg(&msg, SOL_IP, IP_TTL, sizeof(hlim), &hlim);
1664 		}
1665 		if (inet_test_bit(TOS, sk)) {
1666 			int tos = READ_ONCE(inet->rcv_tos);
1667 			put_cmsg(&msg, SOL_IP, IP_TOS, sizeof(tos), &tos);
1668 		}
1669 		len -= msg.msg_controllen;
1670 		return copy_to_sockptr(optlen, &len, sizeof(int));
1671 	}
1672 	case IP_UNICAST_IF:
1673 		val = (__force int)htonl((__u32) READ_ONCE(inet->uc_index));
1674 		goto copyval;
1675 	case IP_MULTICAST_IF:
1676 	{
1677 		struct in_addr addr;
1678 		len = min_t(unsigned int, len, sizeof(struct in_addr));
1679 		addr.s_addr = READ_ONCE(inet->mc_addr);
1680 
1681 		if (copy_to_sockptr(optlen, &len, sizeof(int)))
1682 			return -EFAULT;
1683 		if (copy_to_sockptr(optval, &addr, len))
1684 			return -EFAULT;
1685 		return 0;
1686 	}
1687 	case IP_LOCAL_PORT_RANGE:
1688 		val = READ_ONCE(inet->local_port_range);
1689 		goto copyval;
1690 	}
1691 
1692 	if (needs_rtnl)
1693 		rtnl_lock();
1694 	sockopt_lock_sock(sk);
1695 
1696 	switch (optname) {
1697 	case IP_MSFILTER:
1698 	{
1699 		struct ip_msfilter msf;
1700 
1701 		if (len < IP_MSFILTER_SIZE(0)) {
1702 			err = -EINVAL;
1703 			goto out;
1704 		}
1705 		if (copy_from_sockptr(&msf, optval, IP_MSFILTER_SIZE(0))) {
1706 			err = -EFAULT;
1707 			goto out;
1708 		}
1709 		err = ip_mc_msfget(sk, &msf, optval, optlen);
1710 		goto out;
1711 	}
1712 	case MCAST_MSFILTER:
1713 		if (in_compat_syscall())
1714 			err = compat_ip_get_mcast_msfilter(sk, optval, optlen,
1715 							   len);
1716 		else
1717 			err = ip_get_mcast_msfilter(sk, optval, optlen, len);
1718 		goto out;
1719 	case IP_PROTOCOL:
1720 		val = inet_sk(sk)->inet_num;
1721 		break;
1722 	default:
1723 		sockopt_release_sock(sk);
1724 		return -ENOPROTOOPT;
1725 	}
1726 	sockopt_release_sock(sk);
1727 copyval:
1728 	if (len < sizeof(int) && len > 0 && val >= 0 && val <= 255) {
1729 		unsigned char ucval = (unsigned char)val;
1730 		len = 1;
1731 		if (copy_to_sockptr(optlen, &len, sizeof(int)))
1732 			return -EFAULT;
1733 		if (copy_to_sockptr(optval, &ucval, 1))
1734 			return -EFAULT;
1735 	} else {
1736 		len = min_t(unsigned int, sizeof(int), len);
1737 		if (copy_to_sockptr(optlen, &len, sizeof(int)))
1738 			return -EFAULT;
1739 		if (copy_to_sockptr(optval, &val, len))
1740 			return -EFAULT;
1741 	}
1742 	return 0;
1743 
1744 out:
1745 	sockopt_release_sock(sk);
1746 	if (needs_rtnl)
1747 		rtnl_unlock();
1748 	return err;
1749 }
1750 
1751 int ip_getsockopt(struct sock *sk, int level,
1752 		  int optname, char __user *optval, int __user *optlen)
1753 {
1754 	int err;
1755 
1756 	err = do_ip_getsockopt(sk, level, optname,
1757 			       USER_SOCKPTR(optval), USER_SOCKPTR(optlen));
1758 
1759 #ifdef CONFIG_NETFILTER
1760 	/* we need to exclude all possible ENOPROTOOPTs except default case */
1761 	if (err == -ENOPROTOOPT && optname != IP_PKTOPTIONS &&
1762 			!ip_mroute_opt(optname)) {
1763 		int len;
1764 
1765 		if (get_user(len, optlen))
1766 			return -EFAULT;
1767 
1768 		err = nf_getsockopt(sk, PF_INET, optname, optval, &len);
1769 		if (err >= 0)
1770 			err = put_user(len, optlen);
1771 		return err;
1772 	}
1773 #endif
1774 	return err;
1775 }
1776 EXPORT_SYMBOL(ip_getsockopt);
1777