projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[SK_BUFF]: Introduce skb_network_offset()
[linux-block.git] / net / ipv4 / ip_sockglue.c
CommitLineData
1da177e4
LT		 1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The IP to API glue.
e905a9ed 7 *
1da177e4
LT		 8 * Version: $Id: ip_sockglue.c,v 1.62 2002/02/01 22:01:04 davem Exp $
9 *
10 * Authors: see ip.c
11 *
12 * Fixes:
13 * Many : Split from ip.c , see ip.c for history.
14 * Martin Mares : TOS setting fixed.
e905a9ed 15 * Alan Cox : Fixed a couple of oopses in Martin's
1da177e4
LT		 16 * TOS tweaks.
17 * Mike McLagan : Routing by source
18 */
19
1da177e4
LT		 20#include <linux/module.h>
21#include <linux/types.h>
22#include <linux/mm.h>
1da177e4
LT		 23#include <linux/skbuff.h>
24#include <linux/ip.h>
25#include <linux/icmp.h>
14c85021 26#include <linux/inetdevice.h>
1da177e4
LT		 27#include <linux/netdevice.h>
28#include <net/sock.h>
29#include <net/ip.h>
30#include <net/icmp.h>
d83d8461 31#include <net/tcp_states.h>
1da177e4
LT		 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/route.h>
38#include <net/xfrm.h>
39#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
40#include <net/transp_v6.h>
41#endif
42
43#include <linux/errqueue.h>
44#include <asm/uaccess.h>
45
46#define IP_CMSG_PKTINFO 1
47#define IP_CMSG_TTL 2
48#define IP_CMSG_TOS 4
49#define IP_CMSG_RECVOPTS 8
50#define IP_CMSG_RETOPTS 16
2c7946a7 51#define IP_CMSG_PASSSEC 32
1da177e4
LT		 52
53/*
54 * SOL_IP control messages.
55 */
56
57static void ip_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
58{
59 struct in_pktinfo info;
60 struct rtable *rt = (struct rtable *)skb->dst;
61
62 info.ipi_addr.s_addr = skb->nh.iph->daddr;
63 if (rt) {
64 info.ipi_ifindex = rt->rt_iif;
65 info.ipi_spec_dst.s_addr = rt->rt_spec_dst;
66 } else {
67 info.ipi_ifindex = 0;
68 info.ipi_spec_dst.s_addr = 0;
69 }
70
71 put_cmsg(msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
72}
73
74static void ip_cmsg_recv_ttl(struct msghdr *msg, struct sk_buff *skb)
75{
76 int ttl = skb->nh.iph->ttl;
77 put_cmsg(msg, SOL_IP, IP_TTL, sizeof(int), &ttl);
78}
79
80static void ip_cmsg_recv_tos(struct msghdr *msg, struct sk_buff *skb)
81{
82 put_cmsg(msg, SOL_IP, IP_TOS, 1, &skb->nh.iph->tos);
83}
84
85static void ip_cmsg_recv_opts(struct msghdr *msg, struct sk_buff *skb)
86{
87 if (IPCB(skb)->opt.optlen == 0)
88 return;
89
90 put_cmsg(msg, SOL_IP, IP_RECVOPTS, IPCB(skb)->opt.optlen, skb->nh.iph+1);
91}
92
93
94static void ip_cmsg_recv_retopts(struct msghdr *msg, struct sk_buff *skb)
95{
96 unsigned char optbuf[sizeof(struct ip_options) + 40];
97 struct ip_options * opt = (struct ip_options*)optbuf;
98
99 if (IPCB(skb)->opt.optlen == 0)
100 return;
101
102 if (ip_options_echo(opt, skb)) {
103 msg->msg_flags |= MSG_CTRUNC;
104 return;
105 }
106 ip_options_undo(opt);
107
108 put_cmsg(msg, SOL_IP, IP_RETOPTS, opt->optlen, opt->__data);
109}
110
2c7946a7
CZ		 111static void ip_cmsg_recv_security(struct msghdr *msg, struct sk_buff *skb)
112{
113 char *secdata;
dc49c1f9 114 u32 seclen, secid;
2c7946a7
CZ		 115 int err;
116
dc49c1f9
CZ		 117 err = security_socket_getpeersec_dgram(NULL, skb, &secid);
118 if (err)
119 return;
120
121 err = security_secid_to_secctx(secid, &secdata, &seclen);
2c7946a7
CZ		 122 if (err)
123 return;
124
125 put_cmsg(msg, SOL_IP, SCM_SECURITY, seclen, secdata);
dc49c1f9 126 security_release_secctx(secdata, seclen);
2c7946a7
CZ		 127}
128
1da177e4
LT		 129
130void ip_cmsg_recv(struct msghdr *msg, struct sk_buff *skb)
131{
132 struct inet_sock *inet = inet_sk(skb->sk);
133 unsigned flags = inet->cmsg_flags;
134
135 /* Ordered by supposed usage frequency */
136 if (flags & 1)
137 ip_cmsg_recv_pktinfo(msg, skb);
138 if ((flags>>=1) == 0)
139 return;
140
141 if (flags & 1)
142 ip_cmsg_recv_ttl(msg, skb);
143 if ((flags>>=1) == 0)
144 return;
145
146 if (flags & 1)
147 ip_cmsg_recv_tos(msg, skb);
148 if ((flags>>=1) == 0)
149 return;
150
151 if (flags & 1)
152 ip_cmsg_recv_opts(msg, skb);
153 if ((flags>>=1) == 0)
154 return;
155
156 if (flags & 1)
157 ip_cmsg_recv_retopts(msg, skb);
2c7946a7
CZ		 158 if ((flags>>=1) == 0)
159 return;
160
161 if (flags & 1)
162 ip_cmsg_recv_security(msg, skb);
1da177e4
LT		 163}
164
165int ip_cmsg_send(struct msghdr *msg, struct ipcm_cookie *ipc)
166{
167 int err;
168 struct cmsghdr *cmsg;
169
170 for (cmsg = CMSG_FIRSTHDR(msg); cmsg; cmsg = CMSG_NXTHDR(msg, cmsg)) {
171 if (!CMSG_OK(msg, cmsg))
172 return -EINVAL;
173 if (cmsg->cmsg_level != SOL_IP)
174 continue;
175 switch (cmsg->cmsg_type) {
176 case IP_RETOPTS:
177 err = cmsg->cmsg_len - CMSG_ALIGN(sizeof(struct cmsghdr));
4c6ea29d 178 err = ip_options_get(&ipc->opt, CMSG_DATA(cmsg), err < 40 ? err : 40);
1da177e4
LT		 179 if (err)
180 return err;
181 break;
182 case IP_PKTINFO:
183 {
184 struct in_pktinfo *info;
185 if (cmsg->cmsg_len != CMSG_LEN(sizeof(struct in_pktinfo)))
186 return -EINVAL;
187 info = (struct in_pktinfo *)CMSG_DATA(cmsg);
188 ipc->oif = info->ipi_ifindex;
189 ipc->addr = info->ipi_spec_dst.s_addr;
190 break;
191 }
192 default:
193 return -EINVAL;
194 }
195 }
196 return 0;
197}
198
199
200/* Special input handler for packets caught by router alert option.
201 They are selected only by protocol field, and then processed likely
202 local ones; but only if someone wants them! Otherwise, router
203 not running rsvpd will kill RSVP.
204
205 It is user level problem, what it will make with them.
206 I have no idea, how it will masquearde or NAT them (it is joke, joke :-)),
207 but receiver should be enough clever f.e. to forward mtrace requests,
208 sent to multicast group to reach destination designated router.
209 */
210struct ip_ra_chain *ip_ra_chain;
211DEFINE_RWLOCK(ip_ra_lock);
212
213int ip_ra_control(struct sock *sk, unsigned char on, void (*destructor)(struct sock *))
214{
215 struct ip_ra_chain *ra, *new_ra, **rap;
216
217 if (sk->sk_type != SOCK_RAW || inet_sk(sk)->num == IPPROTO_RAW)
218 return -EINVAL;
219
220 new_ra = on ? kmalloc(sizeof(*new_ra), GFP_KERNEL) : NULL;
221
222 write_lock_bh(&ip_ra_lock);
223 for (rap = &ip_ra_chain; (ra=*rap) != NULL; rap = &ra->next) {
224 if (ra->sk == sk) {
225 if (on) {
226 write_unlock_bh(&ip_ra_lock);
a51482bd 227 kfree(new_ra);
1da177e4
LT		 228 return -EADDRINUSE;
229 }
230 *rap = ra->next;
231 write_unlock_bh(&ip_ra_lock);
232
233 if (ra->destructor)
234 ra->destructor(sk);
235 sock_put(sk);
236 kfree(ra);
237 return 0;
238 }
239 }
240 if (new_ra == NULL) {
241 write_unlock_bh(&ip_ra_lock);
242 return -ENOBUFS;
243 }
244 new_ra->sk = sk;
245 new_ra->destructor = destructor;
246
247 new_ra->next = ra;
248 *rap = new_ra;
249 sock_hold(sk);
250 write_unlock_bh(&ip_ra_lock);
251
252 return 0;
253}
254
e905a9ed 255void ip_icmp_error(struct sock *sk, struct sk_buff *skb, int err,
35986b32 256 __be16 port, u32 info, u8 *payload)
1da177e4
LT		 257{
258 struct inet_sock *inet = inet_sk(sk);
259 struct sock_exterr_skb *serr;
260
261 if (!inet->recverr)
262 return;
263
264 skb = skb_clone(skb, GFP_ATOMIC);
265 if (!skb)
266 return;
267
e905a9ed 268 serr = SKB_EXT_ERR(skb);
1da177e4
LT		 269 serr->ee.ee_errno = err;
270 serr->ee.ee_origin = SO_EE_ORIGIN_ICMP;
e905a9ed 271 serr->ee.ee_type = skb->h.icmph->type;
1da177e4
LT		 272 serr->ee.ee_code = skb->h.icmph->code;
273 serr->ee.ee_pad = 0;
274 serr->ee.ee_info = info;
275 serr->ee.ee_data = 0;
276 serr->addr_offset = (u8*)&(((struct iphdr*)(skb->h.icmph+1))->daddr) - skb->nh.raw;
277 serr->port = port;
278
279 skb->h.raw = payload;
280 if (!skb_pull(skb, payload - skb->data) ||
281 sock_queue_err_skb(sk, skb))
282 kfree_skb(skb);
283}
284
0579016e 285void ip_local_error(struct sock *sk, int err, __be32 daddr, __be16 port, u32 info)
1da177e4
LT		 286{
287 struct inet_sock *inet = inet_sk(sk);
288 struct sock_exterr_skb *serr;
289 struct iphdr *iph;
290 struct sk_buff *skb;
291
292 if (!inet->recverr)
293 return;
294
295 skb = alloc_skb(sizeof(struct iphdr), GFP_ATOMIC);
296 if (!skb)
297 return;
298
2ca9e6f2
ACM		 299 skb_put(skb, sizeof(struct iphdr));
300 skb_reset_network_header(skb);
301 iph = skb->nh.iph;
1da177e4
LT		 302 iph->daddr = daddr;
303
e905a9ed 304 serr = SKB_EXT_ERR(skb);
1da177e4
LT		 305 serr->ee.ee_errno = err;
306 serr->ee.ee_origin = SO_EE_ORIGIN_LOCAL;
e905a9ed 307 serr->ee.ee_type = 0;
1da177e4
LT		 308 serr->ee.ee_code = 0;
309 serr->ee.ee_pad = 0;
310 serr->ee.ee_info = info;
311 serr->ee.ee_data = 0;
312 serr->addr_offset = (u8*)&iph->daddr - skb->nh.raw;
313 serr->port = port;
314
315 skb->h.raw = skb->tail;
316 __skb_pull(skb, skb->tail - skb->data);
317
318 if (sock_queue_err_skb(sk, skb))
319 kfree_skb(skb);
320}
321
e905a9ed 322/*
1da177e4
LT		 323 * Handle MSG_ERRQUEUE
324 */
325int ip_recv_error(struct sock *sk, struct msghdr *msg, int len)
326{
327 struct sock_exterr_skb *serr;
328 struct sk_buff *skb, *skb2;
329 struct sockaddr_in *sin;
330 struct {
331 struct sock_extended_err ee;
332 struct sockaddr_in offender;
333 } errhdr;
334 int err;
335 int copied;
336
337 err = -EAGAIN;
338 skb = skb_dequeue(&sk->sk_error_queue);
339 if (skb == NULL)
340 goto out;
341
342 copied = skb->len;
343 if (copied > len) {
344 msg->msg_flags |= MSG_TRUNC;
345 copied = len;
346 }
347 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
348 if (err)
349 goto out_free_skb;
350
351 sock_recv_timestamp(msg, sk, skb);
352
353 serr = SKB_EXT_ERR(skb);
354
355 sin = (struct sockaddr_in *)msg->msg_name;
356 if (sin) {
357 sin->sin_family = AF_INET;
714e85be 358 sin->sin_addr.s_addr = *(__be32*)(skb->nh.raw + serr->addr_offset);
1da177e4
LT		 359 sin->sin_port = serr->port;
360 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
361 }
362
363 memcpy(&errhdr.ee, &serr->ee, sizeof(struct sock_extended_err));
364 sin = &errhdr.offender;
365 sin->sin_family = AF_UNSPEC;
366 if (serr->ee.ee_origin == SO_EE_ORIGIN_ICMP) {
367 struct inet_sock *inet = inet_sk(sk);
368
369 sin->sin_family = AF_INET;
370 sin->sin_addr.s_addr = skb->nh.iph->saddr;
371 sin->sin_port = 0;
372 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
373 if (inet->cmsg_flags)
374 ip_cmsg_recv(msg, skb);
375 }
376
377 put_cmsg(msg, SOL_IP, IP_RECVERR, sizeof(errhdr), &errhdr);
378
379 /* Now we could try to dump offended packet options */
380
381 msg->msg_flags |= MSG_ERRQUEUE;
382 err = copied;
383
384 /* Reset and regenerate socket error */
e0f9f858 385 spin_lock_bh(&sk->sk_error_queue.lock);
1da177e4
LT		 386 sk->sk_err = 0;
387 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
388 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
e0f9f858 389 spin_unlock_bh(&sk->sk_error_queue.lock);
1da177e4
LT		 390 sk->sk_error_report(sk);
391 } else
e0f9f858 392 spin_unlock_bh(&sk->sk_error_queue.lock);
1da177e4 393
e905a9ed 394out_free_skb:
1da177e4
LT		 395 kfree_skb(skb);
396out:
397 return err;
398}
399
400
401/*
402 * Socket option code for IP. This is the end of the line after any TCP,UDP etc options on
403 * an IP socket.
404 */
405
3fdadf7d 406static int do_ip_setsockopt(struct sock *sk, int level,
132adf54 407 int optname, char __user *optval, int optlen)
1da177e4
LT		 408{
409 struct inet_sock *inet = inet_sk(sk);
410 int val=0,err;
411
e905a9ed 412 if (((1<<optname) & ((1<<IP_PKTINFO) | (1<<IP_RECVTTL) |
132adf54
SH		 413 (1<<IP_RECVOPTS) | (1<<IP_RECVTOS) |
414 (1<<IP_RETOPTS) | (1<<IP_TOS) |
415 (1<<IP_TTL) | (1<<IP_HDRINCL) |
416 (1<<IP_MTU_DISCOVER) | (1<<IP_RECVERR) |
417 (1<<IP_ROUTER_ALERT) | (1<<IP_FREEBIND) |
418 (1<<IP_PASSSEC))) ||
419 optname == IP_MULTICAST_TTL ||
420 optname == IP_MULTICAST_LOOP) {
1da177e4
LT		 421 if (optlen >= sizeof(int)) {
422 if (get_user(val, (int __user *) optval))
423 return -EFAULT;
424 } else if (optlen >= sizeof(char)) {
425 unsigned char ucval;
426
427 if (get_user(ucval, (unsigned char __user *) optval))
428 return -EFAULT;
429 val = (int) ucval;
430 }
431 }
432
433 /* If optlen==0, it is equivalent to val == 0 */
434
435#ifdef CONFIG_IP_MROUTE
436 if (optname >= MRT_BASE && optname <= (MRT_BASE + 10))
437 return ip_mroute_setsockopt(sk,optname,optval,optlen);
438#endif
439
440 err = 0;
441 lock_sock(sk);
442
443 switch (optname) {
132adf54
SH		 444 case IP_OPTIONS:
445 {
446 struct ip_options * opt = NULL;
447 if (optlen > 40 || optlen < 0)
448 goto e_inval;
449 err = ip_options_get_from_user(&opt, optval, optlen);
450 if (err)
451 break;
452 if (inet->is_icsk) {
453 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4 454#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
132adf54
SH		 455 if (sk->sk_family == PF_INET ||
456 (!((1 << sk->sk_state) &
457 (TCPF_LISTEN | TCPF_CLOSE)) &&
458 inet->daddr != LOOPBACK4_IPV6)) {
1da177e4 459#endif
132adf54
SH		 460 if (inet->opt)
461 icsk->icsk_ext_hdr_len -= inet->opt->optlen;
462 if (opt)
463 icsk->icsk_ext_hdr_len += opt->optlen;
464 icsk->icsk_sync_mss(sk, icsk->icsk_pmtu_cookie);
1da177e4 465#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1da177e4 466 }
132adf54 467#endif
1da177e4 468 }
132adf54
SH		 469 opt = xchg(&inet->opt, opt);
470 kfree(opt);
471 break;
472 }
473 case IP_PKTINFO:
474 if (val)
475 inet->cmsg_flags |= IP_CMSG_PKTINFO;
476 else
477 inet->cmsg_flags &= ~IP_CMSG_PKTINFO;
478 break;
479 case IP_RECVTTL:
480 if (val)
481 inet->cmsg_flags |= IP_CMSG_TTL;
482 else
483 inet->cmsg_flags &= ~IP_CMSG_TTL;
484 break;
485 case IP_RECVTOS:
486 if (val)
487 inet->cmsg_flags |= IP_CMSG_TOS;
488 else
489 inet->cmsg_flags &= ~IP_CMSG_TOS;
490 break;
491 case IP_RECVOPTS:
492 if (val)
493 inet->cmsg_flags |= IP_CMSG_RECVOPTS;
494 else
495 inet->cmsg_flags &= ~IP_CMSG_RECVOPTS;
496 break;
497 case IP_RETOPTS:
498 if (val)
499 inet->cmsg_flags |= IP_CMSG_RETOPTS;
500 else
501 inet->cmsg_flags &= ~IP_CMSG_RETOPTS;
502 break;
503 case IP_PASSSEC:
504 if (val)
505 inet->cmsg_flags |= IP_CMSG_PASSSEC;
506 else
507 inet->cmsg_flags &= ~IP_CMSG_PASSSEC;
508 break;
509 case IP_TOS: /* This sets both TOS and Precedence */
510 if (sk->sk_type == SOCK_STREAM) {
511 val &= ~3;
512 val |= inet->tos & 3;
513 }
514 if (IPTOS_PREC(val) >= IPTOS_PREC_CRITIC_ECP &&
515 !capable(CAP_NET_ADMIN)) {
516 err = -EPERM;
1da177e4 517 break;
132adf54
SH		 518 }
519 if (inet->tos != val) {
520 inet->tos = val;
521 sk->sk_priority = rt_tos2priority(val);
522 sk_dst_reset(sk);
523 }
524 break;
525 case IP_TTL:
526 if (optlen<1)
527 goto e_inval;
528 if (val != -1 && (val < 1 || val>255))
529 goto e_inval;
530 inet->uc_ttl = val;
531 break;
532 case IP_HDRINCL:
533 if (sk->sk_type != SOCK_RAW) {
534 err = -ENOPROTOOPT;
2c7946a7 535 break;
132adf54
SH		 536 }
537 inet->hdrincl = val ? 1 : 0;
538 break;
539 case IP_MTU_DISCOVER:
540 if (val<0 || val>2)
541 goto e_inval;
542 inet->pmtudisc = val;
543 break;
544 case IP_RECVERR:
545 inet->recverr = !!val;
546 if (!val)
547 skb_queue_purge(&sk->sk_error_queue);
548 break;
549 case IP_MULTICAST_TTL:
550 if (sk->sk_type == SOCK_STREAM)
551 goto e_inval;
552 if (optlen<1)
553 goto e_inval;
554 if (val==-1)
555 val = 1;
556 if (val < 0 || val > 255)
557 goto e_inval;
558 inet->mc_ttl = val;
559 break;
560 case IP_MULTICAST_LOOP:
561 if (optlen<1)
562 goto e_inval;
563 inet->mc_loop = !!val;
564 break;
565 case IP_MULTICAST_IF:
566 {
567 struct ip_mreqn mreq;
568 struct net_device *dev = NULL;
569
570 if (sk->sk_type == SOCK_STREAM)
571 goto e_inval;
572 /*
573 * Check the arguments are allowable
574 */
575
576 err = -EFAULT;
577 if (optlen >= sizeof(struct ip_mreqn)) {
578 if (copy_from_user(&mreq,optval,sizeof(mreq)))
1da177e4 579 break;
132adf54
SH		 580 } else {
581 memset(&mreq, 0, sizeof(mreq));
582 if (optlen >= sizeof(struct in_addr) &&
583 copy_from_user(&mreq.imr_address,optval,sizeof(struct in_addr)))
584 break;
585 }
586
587 if (!mreq.imr_ifindex) {
588 if (mreq.imr_address.s_addr == INADDR_ANY) {
589 inet->mc_index = 0;
590 inet->mc_addr = 0;
591 err = 0;
1da177e4
LT		 592 break;
593 }
132adf54
SH		 594 dev = ip_dev_find(mreq.imr_address.s_addr);
595 if (dev) {
596 mreq.imr_ifindex = dev->ifindex;
597 dev_put(dev);
598 }
599 } else
600 dev = __dev_get_by_index(mreq.imr_ifindex);
1da177e4 601
1da177e4 602
132adf54
SH		 603 err = -EADDRNOTAVAIL;
604 if (!dev)
605 break;
606
607 err = -EINVAL;
608 if (sk->sk_bound_dev_if &&
609 mreq.imr_ifindex != sk->sk_bound_dev_if)
610 break;
1da177e4 611
132adf54
SH		 612 inet->mc_index = mreq.imr_ifindex;
613 inet->mc_addr = mreq.imr_address.s_addr;
614 err = 0;
615 break;
616 }
1da177e4 617
132adf54
SH		 618 case IP_ADD_MEMBERSHIP:
619 case IP_DROP_MEMBERSHIP:
620 {
621 struct ip_mreqn mreq;
1da177e4 622
132adf54
SH		 623 if (optlen < sizeof(struct ip_mreq))
624 goto e_inval;
625 err = -EFAULT;
626 if (optlen >= sizeof(struct ip_mreqn)) {
627 if (copy_from_user(&mreq,optval,sizeof(mreq)))
1da177e4 628 break;
132adf54
SH		 629 } else {
630 memset(&mreq, 0, sizeof(mreq));
631 if (copy_from_user(&mreq,optval,sizeof(struct ip_mreq)))
1da177e4 632 break;
132adf54 633 }
1da177e4 634
132adf54
SH		 635 if (optname == IP_ADD_MEMBERSHIP)
636 err = ip_mc_join_group(sk, &mreq);
637 else
638 err = ip_mc_leave_group(sk, &mreq);
639 break;
640 }
641 case IP_MSFILTER:
642 {
643 extern int sysctl_igmp_max_msf;
644 struct ip_msfilter *msf;
645
646 if (optlen < IP_MSFILTER_SIZE(0))
647 goto e_inval;
648 if (optlen > sysctl_optmem_max) {
649 err = -ENOBUFS;
1da177e4
LT		 650 break;
651 }
132adf54
SH		 652 msf = kmalloc(optlen, GFP_KERNEL);
653 if (msf == 0) {
654 err = -ENOBUFS;
655 break;
656 }
657 err = -EFAULT;
658 if (copy_from_user(msf, optval, optlen)) {
659 kfree(msf);
660 break;
661 }
662 /* numsrc >= (1G-4) overflow in 32 bits */
663 if (msf->imsf_numsrc >= 0x3ffffffcU ||
664 msf->imsf_numsrc > sysctl_igmp_max_msf) {
665 kfree(msf);
666 err = -ENOBUFS;
667 break;
668 }
669 if (IP_MSFILTER_SIZE(msf->imsf_numsrc) > optlen) {
670 kfree(msf);
671 err = -EINVAL;
672 break;
673 }
674 err = ip_mc_msfilter(sk, msf, 0);
675 kfree(msf);
676 break;
677 }
678 case IP_BLOCK_SOURCE:
679 case IP_UNBLOCK_SOURCE:
680 case IP_ADD_SOURCE_MEMBERSHIP:
681 case IP_DROP_SOURCE_MEMBERSHIP:
682 {
683 struct ip_mreq_source mreqs;
684 int omode, add;
1da177e4 685
132adf54
SH		 686 if (optlen != sizeof(struct ip_mreq_source))
687 goto e_inval;
688 if (copy_from_user(&mreqs, optval, sizeof(mreqs))) {
1da177e4 689 err = -EFAULT;
1da177e4
LT		 690 break;
691 }
132adf54
SH		 692 if (optname == IP_BLOCK_SOURCE) {
693 omode = MCAST_EXCLUDE;
694 add = 1;
695 } else if (optname == IP_UNBLOCK_SOURCE) {
696 omode = MCAST_EXCLUDE;
697 add = 0;
698 } else if (optname == IP_ADD_SOURCE_MEMBERSHIP) {
699 struct ip_mreqn mreq;
1da177e4 700
132adf54
SH		 701 mreq.imr_multiaddr.s_addr = mreqs.imr_multiaddr;
702 mreq.imr_address.s_addr = mreqs.imr_interface;
703 mreq.imr_ifindex = 0;
704 err = ip_mc_join_group(sk, &mreq);
705 if (err && err != -EADDRINUSE)
1da177e4 706 break;
132adf54
SH		 707 omode = MCAST_INCLUDE;
708 add = 1;
709 } else /* IP_DROP_SOURCE_MEMBERSHIP */ {
710 omode = MCAST_INCLUDE;
711 add = 0;
712 }
713 err = ip_mc_source(add, omode, sk, &mreqs, 0);
714 break;
715 }
716 case MCAST_JOIN_GROUP:
717 case MCAST_LEAVE_GROUP:
718 {
719 struct group_req greq;
720 struct sockaddr_in *psin;
721 struct ip_mreqn mreq;
722
723 if (optlen < sizeof(struct group_req))
724 goto e_inval;
725 err = -EFAULT;
726 if (copy_from_user(&greq, optval, sizeof(greq)))
727 break;
728 psin = (struct sockaddr_in *)&greq.gr_group;
729 if (psin->sin_family != AF_INET)
730 goto e_inval;
731 memset(&mreq, 0, sizeof(mreq));
732 mreq.imr_multiaddr = psin->sin_addr;
733 mreq.imr_ifindex = greq.gr_interface;
734
735 if (optname == MCAST_JOIN_GROUP)
736 err = ip_mc_join_group(sk, &mreq);
737 else
738 err = ip_mc_leave_group(sk, &mreq);
739 break;
740 }
741 case MCAST_JOIN_SOURCE_GROUP:
742 case MCAST_LEAVE_SOURCE_GROUP:
743 case MCAST_BLOCK_SOURCE:
744 case MCAST_UNBLOCK_SOURCE:
745 {
746 struct group_source_req greqs;
747 struct ip_mreq_source mreqs;
748 struct sockaddr_in *psin;
749 int omode, add;
750
751 if (optlen != sizeof(struct group_source_req))
752 goto e_inval;
753 if (copy_from_user(&greqs, optval, sizeof(greqs))) {
1da177e4 754 err = -EFAULT;
1da177e4
LT		 755 break;
756 }
132adf54
SH		 757 if (greqs.gsr_group.ss_family != AF_INET ||
758 greqs.gsr_source.ss_family != AF_INET) {
759 err = -EADDRNOTAVAIL;
1da177e4
LT		 760 break;
761 }
132adf54
SH		 762 psin = (struct sockaddr_in *)&greqs.gsr_group;
763 mreqs.imr_multiaddr = psin->sin_addr.s_addr;
764 psin = (struct sockaddr_in *)&greqs.gsr_source;
765 mreqs.imr_sourceaddr = psin->sin_addr.s_addr;
766 mreqs.imr_interface = 0; /* use index for mc_source */
767
768 if (optname == MCAST_BLOCK_SOURCE) {
769 omode = MCAST_EXCLUDE;
770 add = 1;
771 } else if (optname == MCAST_UNBLOCK_SOURCE) {
772 omode = MCAST_EXCLUDE;
773 add = 0;
774 } else if (optname == MCAST_JOIN_SOURCE_GROUP) {
1da177e4
LT		 775 struct ip_mreqn mreq;
776
132adf54 777 psin = (struct sockaddr_in *)&greqs.gsr_group;
1da177e4 778 mreq.imr_multiaddr = psin->sin_addr;
132adf54
SH		 779 mreq.imr_address.s_addr = 0;
780 mreq.imr_ifindex = greqs.gsr_interface;
781 err = ip_mc_join_group(sk, &mreq);
782 if (err && err != -EADDRINUSE)
783 break;
784 greqs.gsr_interface = mreq.imr_ifindex;
785 omode = MCAST_INCLUDE;
786 add = 1;
787 } else /* MCAST_LEAVE_SOURCE_GROUP */ {
788 omode = MCAST_INCLUDE;
789 add = 0;
790 }
791 err = ip_mc_source(add, omode, sk, &mreqs,
792 greqs.gsr_interface);
793 break;
794 }
795 case MCAST_MSFILTER:
796 {
797 extern int sysctl_igmp_max_msf;
798 struct sockaddr_in *psin;
799 struct ip_msfilter *msf = NULL;
800 struct group_filter *gsf = NULL;
801 int msize, i, ifindex;
802
803 if (optlen < GROUP_FILTER_SIZE(0))
804 goto e_inval;
805 if (optlen > sysctl_optmem_max) {
806 err = -ENOBUFS;
1da177e4
LT		 807 break;
808 }
132adf54
SH		 809 gsf = kmalloc(optlen,GFP_KERNEL);
810 if (gsf == 0) {
811 err = -ENOBUFS;
1da177e4
LT		 812 break;
813 }
132adf54
SH		 814 err = -EFAULT;
815 if (copy_from_user(gsf, optval, optlen)) {
816 goto mc_msf_out;
817 }
818 /* numsrc >= (4G-140)/128 overflow in 32 bits */
819 if (gsf->gf_numsrc >= 0x1ffffff ||
820 gsf->gf_numsrc > sysctl_igmp_max_msf) {
821 err = -ENOBUFS;
822 goto mc_msf_out;
823 }
824 if (GROUP_FILTER_SIZE(gsf->gf_numsrc) > optlen) {
825 err = -EINVAL;
826 goto mc_msf_out;
827 }
828 msize = IP_MSFILTER_SIZE(gsf->gf_numsrc);
829 msf = kmalloc(msize,GFP_KERNEL);
830 if (msf == 0) {
831 err = -ENOBUFS;
832 goto mc_msf_out;
833 }
834 ifindex = gsf->gf_interface;
835 psin = (struct sockaddr_in *)&gsf->gf_group;
836 if (psin->sin_family != AF_INET) {
1da177e4 837 err = -EADDRNOTAVAIL;
132adf54 838 goto mc_msf_out;
1da177e4 839 }
132adf54
SH		 840 msf->imsf_multiaddr = psin->sin_addr.s_addr;
841 msf->imsf_interface = 0;
842 msf->imsf_fmode = gsf->gf_fmode;
843 msf->imsf_numsrc = gsf->gf_numsrc;
844 err = -EADDRNOTAVAIL;
845 for (i=0; i<gsf->gf_numsrc; ++i) {
846 psin = (struct sockaddr_in *)&gsf->gf_slist[i];
e905a9ed 847
132adf54
SH		 848 if (psin->sin_family != AF_INET)
849 goto mc_msf_out;
850 msf->imsf_slist[i] = psin->sin_addr.s_addr;
851 }
852 kfree(gsf);
853 gsf = NULL;
854
855 err = ip_mc_msfilter(sk, msf, ifindex);
856 mc_msf_out:
857 kfree(msf);
858 kfree(gsf);
859 break;
860 }
861 case IP_ROUTER_ALERT:
862 err = ip_ra_control(sk, val ? 1 : 0, NULL);
863 break;
864
865 case IP_FREEBIND:
866 if (optlen<1)
867 goto e_inval;
868 inet->freebind = !!val;
869 break;
870
871 case IP_IPSEC_POLICY:
872 case IP_XFRM_POLICY:
873 err = -EPERM;
874 if (!capable(CAP_NET_ADMIN))
1da177e4 875 break;
132adf54
SH		 876 err = xfrm_user_policy(sk, optname, optval, optlen);
877 break;
1da177e4 878
132adf54
SH		 879 default:
880 err = -ENOPROTOOPT;
881 break;
1da177e4
LT		 882 }
883 release_sock(sk);
884 return err;
885
886e_inval:
887 release_sock(sk);
888 return -EINVAL;
889}
890
3fdadf7d
DM		 891int ip_setsockopt(struct sock *sk, int level,
892 int optname, char __user *optval, int optlen)
893{
894 int err;
895
896 if (level != SOL_IP)
897 return -ENOPROTOOPT;
898
899 err = do_ip_setsockopt(sk, level, optname, optval, optlen);
900#ifdef CONFIG_NETFILTER
901 /* we need to exclude all possible ENOPROTOOPTs except default case */
902 if (err == -ENOPROTOOPT && optname != IP_HDRINCL &&
903 optname != IP_IPSEC_POLICY && optname != IP_XFRM_POLICY
904#ifdef CONFIG_IP_MROUTE
905 && (optname < MRT_BASE || optname > (MRT_BASE + 10))
906#endif
907 ) {
908 lock_sock(sk);
909 err = nf_setsockopt(sk, PF_INET, optname, optval, optlen);
910 release_sock(sk);
911 }
912#endif
913 return err;
914}
915
916#ifdef CONFIG_COMPAT
543d9cfe
ACM		 917int compat_ip_setsockopt(struct sock *sk, int level, int optname,
918 char __user *optval, int optlen)
3fdadf7d
DM		 919{
920 int err;
921
922 if (level != SOL_IP)
923 return -ENOPROTOOPT;
924
925 err = do_ip_setsockopt(sk, level, optname, optval, optlen);
926#ifdef CONFIG_NETFILTER
927 /* we need to exclude all possible ENOPROTOOPTs except default case */
928 if (err == -ENOPROTOOPT && optname != IP_HDRINCL &&
543d9cfe 929 optname != IP_IPSEC_POLICY && optname != IP_XFRM_POLICY
3fdadf7d 930#ifdef CONFIG_IP_MROUTE
543d9cfe 931 && (optname < MRT_BASE || optname > (MRT_BASE + 10))
3fdadf7d
DM		 932#endif
933 ) {
934 lock_sock(sk);
543d9cfe
ACM		 935 err = compat_nf_setsockopt(sk, PF_INET, optname,
936 optval, optlen);
3fdadf7d
DM		 937 release_sock(sk);
938 }
939#endif
940 return err;
941}
543d9cfe
ACM		 942
943EXPORT_SYMBOL(compat_ip_setsockopt);
3fdadf7d
DM		 944#endif
945
1da177e4
LT		 946/*
947 * Get the options. Note for future reference. The GET of IP options gets the
948 * _received_ ones. The set sets the _sent_ ones.
949 */
950
3fdadf7d 951static int do_ip_getsockopt(struct sock *sk, int level, int optname,
132adf54 952 char __user *optval, int __user *optlen)
1da177e4
LT		 953{
954 struct inet_sock *inet = inet_sk(sk);
955 int val;
956 int len;
e905a9ed 957
132adf54 958 if (level != SOL_IP)
1da177e4
LT		 959 return -EOPNOTSUPP;
960
961#ifdef CONFIG_IP_MROUTE
132adf54 962 if (optname >= MRT_BASE && optname <= MRT_BASE+10) {
1da177e4
LT		 963 return ip_mroute_getsockopt(sk,optname,optval,optlen);
964 }
965#endif
966
132adf54 967 if (get_user(len,optlen))
1da177e4 968 return -EFAULT;
132adf54 969 if (len < 0)
1da177e4 970 return -EINVAL;
e905a9ed 971
1da177e4
LT		 972 lock_sock(sk);
973
132adf54
SH		 974 switch (optname) {
975 case IP_OPTIONS:
976 {
977 unsigned char optbuf[sizeof(struct ip_options)+40];
978 struct ip_options * opt = (struct ip_options*)optbuf;
979 opt->optlen = 0;
980 if (inet->opt)
981 memcpy(optbuf, inet->opt,
982 sizeof(struct ip_options)+
983 inet->opt->optlen);
984 release_sock(sk);
985
986 if (opt->optlen == 0)
987 return put_user(0, optlen);
988
989 ip_options_undo(opt);
990
991 len = min_t(unsigned int, len, opt->optlen);
992 if (put_user(len, optlen))
993 return -EFAULT;
994 if (copy_to_user(optval, opt->__data, len))
995 return -EFAULT;
996 return 0;
997 }
998 case IP_PKTINFO:
999 val = (inet->cmsg_flags & IP_CMSG_PKTINFO) != 0;
1000 break;
1001 case IP_RECVTTL:
1002 val = (inet->cmsg_flags & IP_CMSG_TTL) != 0;
1003 break;
1004 case IP_RECVTOS:
1005 val = (inet->cmsg_flags & IP_CMSG_TOS) != 0;
1006 break;
1007 case IP_RECVOPTS:
1008 val = (inet->cmsg_flags & IP_CMSG_RECVOPTS) != 0;
1009 break;
1010 case IP_RETOPTS:
1011 val = (inet->cmsg_flags & IP_CMSG_RETOPTS) != 0;
1012 break;
1013 case IP_PASSSEC:
1014 val = (inet->cmsg_flags & IP_CMSG_PASSSEC) != 0;
1015 break;
1016 case IP_TOS:
1017 val = inet->tos;
1018 break;
1019 case IP_TTL:
1020 val = (inet->uc_ttl == -1 ?
1021 sysctl_ip_default_ttl :
1022 inet->uc_ttl);
1023 break;
1024 case IP_HDRINCL:
1025 val = inet->hdrincl;
1026 break;
1027 case IP_MTU_DISCOVER:
1028 val = inet->pmtudisc;
1029 break;
1030 case IP_MTU:
1031 {
1032 struct dst_entry *dst;
1033 val = 0;
1034 dst = sk_dst_get(sk);
1035 if (dst) {
1036 val = dst_mtu(dst);
1037 dst_release(dst);
1da177e4 1038 }
132adf54 1039 if (!val) {
1da177e4 1040 release_sock(sk);
132adf54 1041 return -ENOTCONN;
1da177e4 1042 }
132adf54
SH		 1043 break;
1044 }
1045 case IP_RECVERR:
1046 val = inet->recverr;
1047 break;
1048 case IP_MULTICAST_TTL:
1049 val = inet->mc_ttl;
1050 break;
1051 case IP_MULTICAST_LOOP:
1052 val = inet->mc_loop;
1053 break;
1054 case IP_MULTICAST_IF:
1055 {
1056 struct in_addr addr;
1057 len = min_t(unsigned int, len, sizeof(struct in_addr));
1058 addr.s_addr = inet->mc_addr;
1059 release_sock(sk);
1da177e4 1060
132adf54
SH		 1061 if (put_user(len, optlen))
1062 return -EFAULT;
1063 if (copy_to_user(optval, &addr, len))
1064 return -EFAULT;
1065 return 0;
1066 }
1067 case IP_MSFILTER:
1068 {
1069 struct ip_msfilter msf;
1070 int err;
1071
1072 if (len < IP_MSFILTER_SIZE(0)) {
1da177e4 1073 release_sock(sk);
132adf54 1074 return -EINVAL;
1da177e4 1075 }
132adf54 1076 if (copy_from_user(&msf, optval, IP_MSFILTER_SIZE(0))) {
1da177e4 1077 release_sock(sk);
132adf54 1078 return -EFAULT;
1da177e4 1079 }
132adf54
SH		 1080 err = ip_mc_msfget(sk, &msf,
1081 (struct ip_msfilter __user *)optval, optlen);
1082 release_sock(sk);
1083 return err;
1084 }
1085 case MCAST_MSFILTER:
1086 {
1087 struct group_filter gsf;
1088 int err;
1da177e4 1089
132adf54 1090 if (len < GROUP_FILTER_SIZE(0)) {
1da177e4 1091 release_sock(sk);
132adf54
SH		 1092 return -EINVAL;
1093 }
1094 if (copy_from_user(&gsf, optval, GROUP_FILTER_SIZE(0))) {
1095 release_sock(sk);
1096 return -EFAULT;
1097 }
1098 err = ip_mc_gsfget(sk, &gsf,
1099 (struct group_filter __user *)optval, optlen);
1100 release_sock(sk);
1101 return err;
1102 }
1103 case IP_PKTOPTIONS:
1104 {
1105 struct msghdr msg;
1da177e4 1106
132adf54 1107 release_sock(sk);
1da177e4 1108
132adf54
SH		 1109 if (sk->sk_type != SOCK_STREAM)
1110 return -ENOPROTOOPT;
1da177e4 1111
132adf54
SH		 1112 msg.msg_control = optval;
1113 msg.msg_controllen = len;
1114 msg.msg_flags = 0;
1da177e4 1115
132adf54
SH		 1116 if (inet->cmsg_flags & IP_CMSG_PKTINFO) {
1117 struct in_pktinfo info;
1118
1119 info.ipi_addr.s_addr = inet->rcv_saddr;
1120 info.ipi_spec_dst.s_addr = inet->rcv_saddr;
1121 info.ipi_ifindex = inet->mc_index;
1122 put_cmsg(&msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
1da177e4 1123 }
132adf54
SH		 1124 if (inet->cmsg_flags & IP_CMSG_TTL) {
1125 int hlim = inet->mc_ttl;
1126 put_cmsg(&msg, SOL_IP, IP_TTL, sizeof(hlim), &hlim);
1127 }
1128 len -= msg.msg_controllen;
1129 return put_user(len, optlen);
1130 }
1131 case IP_FREEBIND:
1132 val = inet->freebind;
1133 break;
1134 default:
1135 release_sock(sk);
1136 return -ENOPROTOOPT;
1da177e4
LT		 1137 }
1138 release_sock(sk);
e905a9ed 1139
1da177e4
LT		 1140 if (len < sizeof(int) && len > 0 && val>=0 && val<255) {
1141 unsigned char ucval = (unsigned char)val;
1142 len = 1;
132adf54 1143 if (put_user(len, optlen))
1da177e4 1144 return -EFAULT;
132adf54 1145 if (copy_to_user(optval,&ucval,1))
1da177e4
LT		 1146 return -EFAULT;
1147 } else {
1148 len = min_t(unsigned int, sizeof(int), len);
132adf54 1149 if (put_user(len, optlen))
1da177e4 1150 return -EFAULT;
132adf54 1151 if (copy_to_user(optval,&val,len))
1da177e4
LT		 1152 return -EFAULT;
1153 }
1154 return 0;
1155}
1156
3fdadf7d 1157int ip_getsockopt(struct sock *sk, int level,
132adf54 1158 int optname, char __user *optval, int __user *optlen)
3fdadf7d
DM		 1159{
1160 int err;
1161
1162 err = do_ip_getsockopt(sk, level, optname, optval, optlen);
1163#ifdef CONFIG_NETFILTER
1164 /* we need to exclude all possible ENOPROTOOPTs except default case */
1165 if (err == -ENOPROTOOPT && optname != IP_PKTOPTIONS
1166#ifdef CONFIG_IP_MROUTE
1167 && (optname < MRT_BASE || optname > MRT_BASE+10)
1168#endif
1169 ) {
e905a9ed 1170 int len;
3fdadf7d 1171
132adf54 1172 if (get_user(len,optlen))
3fdadf7d
DM		 1173 return -EFAULT;
1174
1175 lock_sock(sk);
1176 err = nf_getsockopt(sk, PF_INET, optname, optval,
1177 &len);
1178 release_sock(sk);
1179 if (err >= 0)
1180 err = put_user(len, optlen);
1181 return err;
1182 }
1183#endif
1184 return err;
1185}
1186
1187#ifdef CONFIG_COMPAT
543d9cfe
ACM		 1188int compat_ip_getsockopt(struct sock *sk, int level, int optname,
1189 char __user *optval, int __user *optlen)
3fdadf7d 1190{
543d9cfe 1191 int err = do_ip_getsockopt(sk, level, optname, optval, optlen);
3fdadf7d
DM		 1192#ifdef CONFIG_NETFILTER
1193 /* we need to exclude all possible ENOPROTOOPTs except default case */
1194 if (err == -ENOPROTOOPT && optname != IP_PKTOPTIONS
1195#ifdef CONFIG_IP_MROUTE
543d9cfe 1196 && (optname < MRT_BASE || optname > MRT_BASE+10)
3fdadf7d
DM		 1197#endif
1198 ) {
e905a9ed 1199 int len;
3fdadf7d 1200
543d9cfe 1201 if (get_user(len, optlen))
3fdadf7d
DM		 1202 return -EFAULT;
1203
1204 lock_sock(sk);
543d9cfe 1205 err = compat_nf_getsockopt(sk, PF_INET, optname, optval, &len);
3fdadf7d
DM		 1206 release_sock(sk);
1207 if (err >= 0)
1208 err = put_user(len, optlen);
1209 return err;
1210 }
1211#endif
1212 return err;
1213}
543d9cfe
ACM		 1214
1215EXPORT_SYMBOL(compat_ip_getsockopt);
3fdadf7d
DM		 1216#endif
1217
1da177e4
LT		 1218EXPORT_SYMBOL(ip_cmsg_recv);
1219
1da177e4
LT		 1220EXPORT_SYMBOL(ip_getsockopt);
1221EXPORT_SYMBOL(ip_setsockopt);
Linux 6.x block layer and io_uring tree(s)