xref: /linux/net/ipv6/ip6_output.c (revision 14b42963f64b98ab61fa9723c03d71aa5ef4f862)
1 /*
2  *	IPv6 output functions
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	$Id: ip6_output.c,v 1.34 2002/02/01 22:01:04 davem Exp $
9  *
10  *	Based on linux/net/ipv4/ip_output.c
11  *
12  *	This program is free software; you can redistribute it and/or
13  *      modify it under the terms of the GNU General Public License
14  *      as published by the Free Software Foundation; either version
15  *      2 of the License, or (at your option) any later version.
16  *
17  *	Changes:
18  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
19  *				extension headers are implemented.
20  *				route changes now work.
21  *				ip6_forward does not confuse sniffers.
22  *				etc.
23  *
24  *      H. von Brand    :       Added missing #include <linux/string.h>
25  *	Imran Patel	: 	frag id should be in NBO
26  *      Kazunori MIYAZAWA @USAGI
27  *			:       add ip6_append_data and related functions
28  *				for datagram xmit
29  */
30 
31 #include <linux/errno.h>
32 #include <linux/types.h>
33 #include <linux/string.h>
34 #include <linux/socket.h>
35 #include <linux/net.h>
36 #include <linux/netdevice.h>
37 #include <linux/if_arp.h>
38 #include <linux/in6.h>
39 #include <linux/tcp.h>
40 #include <linux/route.h>
41 #include <linux/module.h>
42 
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45 
46 #include <net/sock.h>
47 #include <net/snmp.h>
48 
49 #include <net/ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
55 #include <net/icmp.h>
56 #include <net/xfrm.h>
57 #include <net/checksum.h>
58 
59 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
60 
61 static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr)
62 {
63 	static u32 ipv6_fragmentation_id = 1;
64 	static DEFINE_SPINLOCK(ip6_id_lock);
65 
66 	spin_lock_bh(&ip6_id_lock);
67 	fhdr->identification = htonl(ipv6_fragmentation_id);
68 	if (++ipv6_fragmentation_id == 0)
69 		ipv6_fragmentation_id = 1;
70 	spin_unlock_bh(&ip6_id_lock);
71 }
72 
73 static inline int ip6_output_finish(struct sk_buff *skb)
74 {
75 
76 	struct dst_entry *dst = skb->dst;
77 	struct hh_cache *hh = dst->hh;
78 
79 	if (hh) {
80 		int hh_alen;
81 
82 		read_lock_bh(&hh->hh_lock);
83 		hh_alen = HH_DATA_ALIGN(hh->hh_len);
84 		memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
85 		read_unlock_bh(&hh->hh_lock);
86 	        skb_push(skb, hh->hh_len);
87 		return hh->hh_output(skb);
88 	} else if (dst->neighbour)
89 		return dst->neighbour->output(skb);
90 
91 	IP6_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
92 	kfree_skb(skb);
93 	return -EINVAL;
94 
95 }
96 
97 /* dev_loopback_xmit for use with netfilter. */
98 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
99 {
100 	newskb->mac.raw = newskb->data;
101 	__skb_pull(newskb, newskb->nh.raw - newskb->data);
102 	newskb->pkt_type = PACKET_LOOPBACK;
103 	newskb->ip_summed = CHECKSUM_UNNECESSARY;
104 	BUG_TRAP(newskb->dst);
105 
106 	netif_rx(newskb);
107 	return 0;
108 }
109 
110 
111 static int ip6_output2(struct sk_buff *skb)
112 {
113 	struct dst_entry *dst = skb->dst;
114 	struct net_device *dev = dst->dev;
115 
116 	skb->protocol = htons(ETH_P_IPV6);
117 	skb->dev = dev;
118 
119 	if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
120 		struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
121 
122 		if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
123 		    ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
124 				&skb->nh.ipv6h->saddr)) {
125 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
126 
127 			/* Do not check for IFF_ALLMULTI; multicast routing
128 			   is not supported in any case.
129 			 */
130 			if (newskb)
131 				NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
132 					newskb->dev,
133 					ip6_dev_loopback_xmit);
134 
135 			if (skb->nh.ipv6h->hop_limit == 0) {
136 				IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
137 				kfree_skb(skb);
138 				return 0;
139 			}
140 		}
141 
142 		IP6_INC_STATS(IPSTATS_MIB_OUTMCASTPKTS);
143 	}
144 
145 	return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
146 }
147 
148 int ip6_output(struct sk_buff *skb)
149 {
150 	if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
151 				dst_allfrag(skb->dst))
152 		return ip6_fragment(skb, ip6_output2);
153 	else
154 		return ip6_output2(skb);
155 }
156 
157 /*
158  *	xmit an sk_buff (used by TCP)
159  */
160 
161 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
162 	     struct ipv6_txoptions *opt, int ipfragok)
163 {
164 	struct ipv6_pinfo *np = inet6_sk(sk);
165 	struct in6_addr *first_hop = &fl->fl6_dst;
166 	struct dst_entry *dst = skb->dst;
167 	struct ipv6hdr *hdr;
168 	u8  proto = fl->proto;
169 	int seg_len = skb->len;
170 	int hlimit, tclass;
171 	u32 mtu;
172 
173 	if (opt) {
174 		int head_room;
175 
176 		/* First: exthdrs may take lots of space (~8K for now)
177 		   MAX_HEADER is not enough.
178 		 */
179 		head_room = opt->opt_nflen + opt->opt_flen;
180 		seg_len += head_room;
181 		head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
182 
183 		if (skb_headroom(skb) < head_room) {
184 			struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
185 			kfree_skb(skb);
186 			skb = skb2;
187 			if (skb == NULL) {
188 				IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
189 				return -ENOBUFS;
190 			}
191 			if (sk)
192 				skb_set_owner_w(skb, sk);
193 		}
194 		if (opt->opt_flen)
195 			ipv6_push_frag_opts(skb, opt, &proto);
196 		if (opt->opt_nflen)
197 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
198 	}
199 
200 	hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
201 
202 	/*
203 	 *	Fill in the IPv6 header
204 	 */
205 
206 	hlimit = -1;
207 	if (np)
208 		hlimit = np->hop_limit;
209 	if (hlimit < 0)
210 		hlimit = dst_metric(dst, RTAX_HOPLIMIT);
211 	if (hlimit < 0)
212 		hlimit = ipv6_get_hoplimit(dst->dev);
213 
214 	tclass = -1;
215 	if (np)
216 		tclass = np->tclass;
217 	if (tclass < 0)
218 		tclass = 0;
219 
220 	*(u32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
221 
222 	hdr->payload_len = htons(seg_len);
223 	hdr->nexthdr = proto;
224 	hdr->hop_limit = hlimit;
225 
226 	ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
227 	ipv6_addr_copy(&hdr->daddr, first_hop);
228 
229 	skb->priority = sk->sk_priority;
230 
231 	mtu = dst_mtu(dst);
232 	if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
233 		IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
234 		return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
235 				dst_output);
236 	}
237 
238 	if (net_ratelimit())
239 		printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
240 	skb->dev = dst->dev;
241 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
242 	IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
243 	kfree_skb(skb);
244 	return -EMSGSIZE;
245 }
246 
247 /*
248  *	To avoid extra problems ND packets are send through this
249  *	routine. It's code duplication but I really want to avoid
250  *	extra checks since ipv6_build_header is used by TCP (which
251  *	is for us performance critical)
252  */
253 
254 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
255 	       struct in6_addr *saddr, struct in6_addr *daddr,
256 	       int proto, int len)
257 {
258 	struct ipv6_pinfo *np = inet6_sk(sk);
259 	struct ipv6hdr *hdr;
260 	int totlen;
261 
262 	skb->protocol = htons(ETH_P_IPV6);
263 	skb->dev = dev;
264 
265 	totlen = len + sizeof(struct ipv6hdr);
266 
267 	hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
268 	skb->nh.ipv6h = hdr;
269 
270 	*(u32*)hdr = htonl(0x60000000);
271 
272 	hdr->payload_len = htons(len);
273 	hdr->nexthdr = proto;
274 	hdr->hop_limit = np->hop_limit;
275 
276 	ipv6_addr_copy(&hdr->saddr, saddr);
277 	ipv6_addr_copy(&hdr->daddr, daddr);
278 
279 	return 0;
280 }
281 
282 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
283 {
284 	struct ip6_ra_chain *ra;
285 	struct sock *last = NULL;
286 
287 	read_lock(&ip6_ra_lock);
288 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
289 		struct sock *sk = ra->sk;
290 		if (sk && ra->sel == sel &&
291 		    (!sk->sk_bound_dev_if ||
292 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
293 			if (last) {
294 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
295 				if (skb2)
296 					rawv6_rcv(last, skb2);
297 			}
298 			last = sk;
299 		}
300 	}
301 
302 	if (last) {
303 		rawv6_rcv(last, skb);
304 		read_unlock(&ip6_ra_lock);
305 		return 1;
306 	}
307 	read_unlock(&ip6_ra_lock);
308 	return 0;
309 }
310 
311 static inline int ip6_forward_finish(struct sk_buff *skb)
312 {
313 	return dst_output(skb);
314 }
315 
316 int ip6_forward(struct sk_buff *skb)
317 {
318 	struct dst_entry *dst = skb->dst;
319 	struct ipv6hdr *hdr = skb->nh.ipv6h;
320 	struct inet6_skb_parm *opt = IP6CB(skb);
321 
322 	if (ipv6_devconf.forwarding == 0)
323 		goto error;
324 
325 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
326 		IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
327 		goto drop;
328 	}
329 
330 	skb->ip_summed = CHECKSUM_NONE;
331 
332 	/*
333 	 *	We DO NOT make any processing on
334 	 *	RA packets, pushing them to user level AS IS
335 	 *	without ane WARRANTY that application will be able
336 	 *	to interpret them. The reason is that we
337 	 *	cannot make anything clever here.
338 	 *
339 	 *	We are not end-node, so that if packet contains
340 	 *	AH/ESP, we cannot make anything.
341 	 *	Defragmentation also would be mistake, RA packets
342 	 *	cannot be fragmented, because there is no warranty
343 	 *	that different fragments will go along one path. --ANK
344 	 */
345 	if (opt->ra) {
346 		u8 *ptr = skb->nh.raw + opt->ra;
347 		if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
348 			return 0;
349 	}
350 
351 	/*
352 	 *	check and decrement ttl
353 	 */
354 	if (hdr->hop_limit <= 1) {
355 		/* Force OUTPUT device used as source address */
356 		skb->dev = dst->dev;
357 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
358 			    0, skb->dev);
359 
360 		kfree_skb(skb);
361 		return -ETIMEDOUT;
362 	}
363 
364 	if (!xfrm6_route_forward(skb)) {
365 		IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
366 		goto drop;
367 	}
368 	dst = skb->dst;
369 
370 	/* IPv6 specs say nothing about it, but it is clear that we cannot
371 	   send redirects to source routed frames.
372 	 */
373 	if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
374 		struct in6_addr *target = NULL;
375 		struct rt6_info *rt;
376 		struct neighbour *n = dst->neighbour;
377 
378 		/*
379 		 *	incoming and outgoing devices are the same
380 		 *	send a redirect.
381 		 */
382 
383 		rt = (struct rt6_info *) dst;
384 		if ((rt->rt6i_flags & RTF_GATEWAY))
385 			target = (struct in6_addr*)&n->primary_key;
386 		else
387 			target = &hdr->daddr;
388 
389 		/* Limit redirects both by destination (here)
390 		   and by source (inside ndisc_send_redirect)
391 		 */
392 		if (xrlim_allow(dst, 1*HZ))
393 			ndisc_send_redirect(skb, n, target);
394 	} else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
395 						|IPV6_ADDR_LINKLOCAL)) {
396 		/* This check is security critical. */
397 		goto error;
398 	}
399 
400 	if (skb->len > dst_mtu(dst)) {
401 		/* Again, force OUTPUT device used as source address */
402 		skb->dev = dst->dev;
403 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
404 		IP6_INC_STATS_BH(IPSTATS_MIB_INTOOBIGERRORS);
405 		IP6_INC_STATS_BH(IPSTATS_MIB_FRAGFAILS);
406 		kfree_skb(skb);
407 		return -EMSGSIZE;
408 	}
409 
410 	if (skb_cow(skb, dst->dev->hard_header_len)) {
411 		IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
412 		goto drop;
413 	}
414 
415 	hdr = skb->nh.ipv6h;
416 
417 	/* Mangling hops number delayed to point after skb COW */
418 
419 	hdr->hop_limit--;
420 
421 	IP6_INC_STATS_BH(IPSTATS_MIB_OUTFORWDATAGRAMS);
422 	return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
423 
424 error:
425 	IP6_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
426 drop:
427 	kfree_skb(skb);
428 	return -EINVAL;
429 }
430 
431 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
432 {
433 	to->pkt_type = from->pkt_type;
434 	to->priority = from->priority;
435 	to->protocol = from->protocol;
436 	dst_release(to->dst);
437 	to->dst = dst_clone(from->dst);
438 	to->dev = from->dev;
439 
440 #ifdef CONFIG_NET_SCHED
441 	to->tc_index = from->tc_index;
442 #endif
443 #ifdef CONFIG_NETFILTER
444 	to->nfmark = from->nfmark;
445 	/* Connection association is same as pre-frag packet */
446 	nf_conntrack_put(to->nfct);
447 	to->nfct = from->nfct;
448 	nf_conntrack_get(to->nfct);
449 	to->nfctinfo = from->nfctinfo;
450 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
451 	nf_conntrack_put_reasm(to->nfct_reasm);
452 	to->nfct_reasm = from->nfct_reasm;
453 	nf_conntrack_get_reasm(to->nfct_reasm);
454 #endif
455 #ifdef CONFIG_BRIDGE_NETFILTER
456 	nf_bridge_put(to->nf_bridge);
457 	to->nf_bridge = from->nf_bridge;
458 	nf_bridge_get(to->nf_bridge);
459 #endif
460 #endif
461 	skb_copy_secmark(to, from);
462 }
463 
464 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
465 {
466 	u16 offset = sizeof(struct ipv6hdr);
467 	struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
468 	unsigned int packet_len = skb->tail - skb->nh.raw;
469 	int found_rhdr = 0;
470 	*nexthdr = &skb->nh.ipv6h->nexthdr;
471 
472 	while (offset + 1 <= packet_len) {
473 
474 		switch (**nexthdr) {
475 
476 		case NEXTHDR_HOP:
477 		case NEXTHDR_ROUTING:
478 		case NEXTHDR_DEST:
479 			if (**nexthdr == NEXTHDR_ROUTING) found_rhdr = 1;
480 			if (**nexthdr == NEXTHDR_DEST && found_rhdr) return offset;
481 			offset += ipv6_optlen(exthdr);
482 			*nexthdr = &exthdr->nexthdr;
483 			exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
484 			break;
485 		default :
486 			return offset;
487 		}
488 	}
489 
490 	return offset;
491 }
492 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
493 
494 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
495 {
496 	struct net_device *dev;
497 	struct sk_buff *frag;
498 	struct rt6_info *rt = (struct rt6_info*)skb->dst;
499 	struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
500 	struct ipv6hdr *tmp_hdr;
501 	struct frag_hdr *fh;
502 	unsigned int mtu, hlen, left, len;
503 	u32 frag_id = 0;
504 	int ptr, offset = 0, err=0;
505 	u8 *prevhdr, nexthdr = 0;
506 
507 	dev = rt->u.dst.dev;
508 	hlen = ip6_find_1stfragopt(skb, &prevhdr);
509 	nexthdr = *prevhdr;
510 
511 	mtu = dst_mtu(&rt->u.dst);
512 	if (np && np->frag_size < mtu) {
513 		if (np->frag_size)
514 			mtu = np->frag_size;
515 	}
516 	mtu -= hlen + sizeof(struct frag_hdr);
517 
518 	if (skb_shinfo(skb)->frag_list) {
519 		int first_len = skb_pagelen(skb);
520 
521 		if (first_len - hlen > mtu ||
522 		    ((first_len - hlen) & 7) ||
523 		    skb_cloned(skb))
524 			goto slow_path;
525 
526 		for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
527 			/* Correct geometry. */
528 			if (frag->len > mtu ||
529 			    ((frag->len & 7) && frag->next) ||
530 			    skb_headroom(frag) < hlen)
531 			    goto slow_path;
532 
533 			/* Partially cloned skb? */
534 			if (skb_shared(frag))
535 				goto slow_path;
536 
537 			BUG_ON(frag->sk);
538 			if (skb->sk) {
539 				sock_hold(skb->sk);
540 				frag->sk = skb->sk;
541 				frag->destructor = sock_wfree;
542 				skb->truesize -= frag->truesize;
543 			}
544 		}
545 
546 		err = 0;
547 		offset = 0;
548 		frag = skb_shinfo(skb)->frag_list;
549 		skb_shinfo(skb)->frag_list = NULL;
550 		/* BUILD HEADER */
551 
552 		tmp_hdr = kmalloc(hlen, GFP_ATOMIC);
553 		if (!tmp_hdr) {
554 			IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
555 			return -ENOMEM;
556 		}
557 
558 		*prevhdr = NEXTHDR_FRAGMENT;
559 		memcpy(tmp_hdr, skb->nh.raw, hlen);
560 		__skb_pull(skb, hlen);
561 		fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
562 		skb->nh.raw = __skb_push(skb, hlen);
563 		memcpy(skb->nh.raw, tmp_hdr, hlen);
564 
565 		ipv6_select_ident(skb, fh);
566 		fh->nexthdr = nexthdr;
567 		fh->reserved = 0;
568 		fh->frag_off = htons(IP6_MF);
569 		frag_id = fh->identification;
570 
571 		first_len = skb_pagelen(skb);
572 		skb->data_len = first_len - skb_headlen(skb);
573 		skb->len = first_len;
574 		skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
575 
576 
577 		for (;;) {
578 			/* Prepare header of the next frame,
579 			 * before previous one went down. */
580 			if (frag) {
581 				frag->ip_summed = CHECKSUM_NONE;
582 				frag->h.raw = frag->data;
583 				fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
584 				frag->nh.raw = __skb_push(frag, hlen);
585 				memcpy(frag->nh.raw, tmp_hdr, hlen);
586 				offset += skb->len - hlen - sizeof(struct frag_hdr);
587 				fh->nexthdr = nexthdr;
588 				fh->reserved = 0;
589 				fh->frag_off = htons(offset);
590 				if (frag->next != NULL)
591 					fh->frag_off |= htons(IP6_MF);
592 				fh->identification = frag_id;
593 				frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
594 				ip6_copy_metadata(frag, skb);
595 			}
596 
597 			err = output(skb);
598 			if (err || !frag)
599 				break;
600 
601 			skb = frag;
602 			frag = skb->next;
603 			skb->next = NULL;
604 		}
605 
606 		kfree(tmp_hdr);
607 
608 		if (err == 0) {
609 			IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
610 			return 0;
611 		}
612 
613 		while (frag) {
614 			skb = frag->next;
615 			kfree_skb(frag);
616 			frag = skb;
617 		}
618 
619 		IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
620 		return err;
621 	}
622 
623 slow_path:
624 	left = skb->len - hlen;		/* Space per frame */
625 	ptr = hlen;			/* Where to start from */
626 
627 	/*
628 	 *	Fragment the datagram.
629 	 */
630 
631 	*prevhdr = NEXTHDR_FRAGMENT;
632 
633 	/*
634 	 *	Keep copying data until we run out.
635 	 */
636 	while(left > 0)	{
637 		len = left;
638 		/* IF: it doesn't fit, use 'mtu' - the data space left */
639 		if (len > mtu)
640 			len = mtu;
641 		/* IF: we are not sending upto and including the packet end
642 		   then align the next start on an eight byte boundary */
643 		if (len < left)	{
644 			len &= ~7;
645 		}
646 		/*
647 		 *	Allocate buffer.
648 		 */
649 
650 		if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
651 			NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
652 			IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
653 			err = -ENOMEM;
654 			goto fail;
655 		}
656 
657 		/*
658 		 *	Set up data on packet
659 		 */
660 
661 		ip6_copy_metadata(frag, skb);
662 		skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
663 		skb_put(frag, len + hlen + sizeof(struct frag_hdr));
664 		frag->nh.raw = frag->data;
665 		fh = (struct frag_hdr*)(frag->data + hlen);
666 		frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
667 
668 		/*
669 		 *	Charge the memory for the fragment to any owner
670 		 *	it might possess
671 		 */
672 		if (skb->sk)
673 			skb_set_owner_w(frag, skb->sk);
674 
675 		/*
676 		 *	Copy the packet header into the new buffer.
677 		 */
678 		memcpy(frag->nh.raw, skb->data, hlen);
679 
680 		/*
681 		 *	Build fragment header.
682 		 */
683 		fh->nexthdr = nexthdr;
684 		fh->reserved = 0;
685 		if (!frag_id) {
686 			ipv6_select_ident(skb, fh);
687 			frag_id = fh->identification;
688 		} else
689 			fh->identification = frag_id;
690 
691 		/*
692 		 *	Copy a block of the IP datagram.
693 		 */
694 		if (skb_copy_bits(skb, ptr, frag->h.raw, len))
695 			BUG();
696 		left -= len;
697 
698 		fh->frag_off = htons(offset);
699 		if (left > 0)
700 			fh->frag_off |= htons(IP6_MF);
701 		frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
702 
703 		ptr += len;
704 		offset += len;
705 
706 		/*
707 		 *	Put this fragment into the sending queue.
708 		 */
709 
710 		IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
711 
712 		err = output(frag);
713 		if (err)
714 			goto fail;
715 	}
716 	kfree_skb(skb);
717 	IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
718 	return err;
719 
720 fail:
721 	kfree_skb(skb);
722 	IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
723 	return err;
724 }
725 
726 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
727 {
728 	int err = 0;
729 
730 	*dst = NULL;
731 	if (sk) {
732 		struct ipv6_pinfo *np = inet6_sk(sk);
733 
734 		*dst = sk_dst_check(sk, np->dst_cookie);
735 		if (*dst) {
736 			struct rt6_info *rt = (struct rt6_info*)*dst;
737 
738 			/* Yes, checking route validity in not connected
739 			 * case is not very simple. Take into account,
740 			 * that we do not support routing by source, TOS,
741 			 * and MSG_DONTROUTE 		--ANK (980726)
742 			 *
743 			 * 1. If route was host route, check that
744 			 *    cached destination is current.
745 			 *    If it is network route, we still may
746 			 *    check its validity using saved pointer
747 			 *    to the last used address: daddr_cache.
748 			 *    We do not want to save whole address now,
749 			 *    (because main consumer of this service
750 			 *    is tcp, which has not this problem),
751 			 *    so that the last trick works only on connected
752 			 *    sockets.
753 			 * 2. oif also should be the same.
754 			 */
755 			if (((rt->rt6i_dst.plen != 128 ||
756 			      !ipv6_addr_equal(&fl->fl6_dst,
757 					       &rt->rt6i_dst.addr))
758 			     && (np->daddr_cache == NULL ||
759 				 !ipv6_addr_equal(&fl->fl6_dst,
760 						  np->daddr_cache)))
761 			    || (fl->oif && fl->oif != (*dst)->dev->ifindex)) {
762 				dst_release(*dst);
763 				*dst = NULL;
764 			}
765 		}
766 	}
767 
768 	if (*dst == NULL)
769 		*dst = ip6_route_output(sk, fl);
770 
771 	if ((err = (*dst)->error))
772 		goto out_err_release;
773 
774 	if (ipv6_addr_any(&fl->fl6_src)) {
775 		err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
776 
777 		if (err)
778 			goto out_err_release;
779 	}
780 
781 	return 0;
782 
783 out_err_release:
784 	dst_release(*dst);
785 	*dst = NULL;
786 	return err;
787 }
788 
789 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
790 
791 static inline int ip6_ufo_append_data(struct sock *sk,
792 			int getfrag(void *from, char *to, int offset, int len,
793 			int odd, struct sk_buff *skb),
794 			void *from, int length, int hh_len, int fragheaderlen,
795 			int transhdrlen, int mtu,unsigned int flags)
796 
797 {
798 	struct sk_buff *skb;
799 	int err;
800 
801 	/* There is support for UDP large send offload by network
802 	 * device, so create one single skb packet containing complete
803 	 * udp datagram
804 	 */
805 	if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
806 		skb = sock_alloc_send_skb(sk,
807 			hh_len + fragheaderlen + transhdrlen + 20,
808 			(flags & MSG_DONTWAIT), &err);
809 		if (skb == NULL)
810 			return -ENOMEM;
811 
812 		/* reserve space for Hardware header */
813 		skb_reserve(skb, hh_len);
814 
815 		/* create space for UDP/IP header */
816 		skb_put(skb,fragheaderlen + transhdrlen);
817 
818 		/* initialize network header pointer */
819 		skb->nh.raw = skb->data;
820 
821 		/* initialize protocol header pointer */
822 		skb->h.raw = skb->data + fragheaderlen;
823 
824 		skb->ip_summed = CHECKSUM_HW;
825 		skb->csum = 0;
826 		sk->sk_sndmsg_off = 0;
827 	}
828 
829 	err = skb_append_datato_frags(sk,skb, getfrag, from,
830 				      (length - transhdrlen));
831 	if (!err) {
832 		struct frag_hdr fhdr;
833 
834 		/* specify the length of each IP datagram fragment*/
835 		skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
836 					    sizeof(struct frag_hdr);
837 		skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
838 		ipv6_select_ident(skb, &fhdr);
839 		skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
840 		__skb_queue_tail(&sk->sk_write_queue, skb);
841 
842 		return 0;
843 	}
844 	/* There is not enough support do UPD LSO,
845 	 * so follow normal path
846 	 */
847 	kfree_skb(skb);
848 
849 	return err;
850 }
851 
852 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
853 	int offset, int len, int odd, struct sk_buff *skb),
854 	void *from, int length, int transhdrlen,
855 	int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
856 	struct rt6_info *rt, unsigned int flags)
857 {
858 	struct inet_sock *inet = inet_sk(sk);
859 	struct ipv6_pinfo *np = inet6_sk(sk);
860 	struct sk_buff *skb;
861 	unsigned int maxfraglen, fragheaderlen;
862 	int exthdrlen;
863 	int hh_len;
864 	int mtu;
865 	int copy;
866 	int err;
867 	int offset = 0;
868 	int csummode = CHECKSUM_NONE;
869 
870 	if (flags&MSG_PROBE)
871 		return 0;
872 	if (skb_queue_empty(&sk->sk_write_queue)) {
873 		/*
874 		 * setup for corking
875 		 */
876 		if (opt) {
877 			if (np->cork.opt == NULL) {
878 				np->cork.opt = kmalloc(opt->tot_len,
879 						       sk->sk_allocation);
880 				if (unlikely(np->cork.opt == NULL))
881 					return -ENOBUFS;
882 			} else if (np->cork.opt->tot_len < opt->tot_len) {
883 				printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
884 				return -EINVAL;
885 			}
886 			memcpy(np->cork.opt, opt, opt->tot_len);
887 			inet->cork.flags |= IPCORK_OPT;
888 			/* need source address above miyazawa*/
889 		}
890 		dst_hold(&rt->u.dst);
891 		np->cork.rt = rt;
892 		inet->cork.fl = *fl;
893 		np->cork.hop_limit = hlimit;
894 		np->cork.tclass = tclass;
895 		mtu = dst_mtu(rt->u.dst.path);
896 		if (np->frag_size < mtu) {
897 			if (np->frag_size)
898 				mtu = np->frag_size;
899 		}
900 		inet->cork.fragsize = mtu;
901 		if (dst_allfrag(rt->u.dst.path))
902 			inet->cork.flags |= IPCORK_ALLFRAG;
903 		inet->cork.length = 0;
904 		sk->sk_sndmsg_page = NULL;
905 		sk->sk_sndmsg_off = 0;
906 		exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
907 		length += exthdrlen;
908 		transhdrlen += exthdrlen;
909 	} else {
910 		rt = np->cork.rt;
911 		fl = &inet->cork.fl;
912 		if (inet->cork.flags & IPCORK_OPT)
913 			opt = np->cork.opt;
914 		transhdrlen = 0;
915 		exthdrlen = 0;
916 		mtu = inet->cork.fragsize;
917 	}
918 
919 	hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
920 
921 	fragheaderlen = sizeof(struct ipv6hdr) + (opt ? opt->opt_nflen : 0);
922 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
923 
924 	if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
925 		if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
926 			ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
927 			return -EMSGSIZE;
928 		}
929 	}
930 
931 	/*
932 	 * Let's try using as much space as possible.
933 	 * Use MTU if total length of the message fits into the MTU.
934 	 * Otherwise, we need to reserve fragment header and
935 	 * fragment alignment (= 8-15 octects, in total).
936 	 *
937 	 * Note that we may need to "move" the data from the tail of
938 	 * of the buffer to the new fragment when we split
939 	 * the message.
940 	 *
941 	 * FIXME: It may be fragmented into multiple chunks
942 	 *        at once if non-fragmentable extension headers
943 	 *        are too large.
944 	 * --yoshfuji
945 	 */
946 
947 	inet->cork.length += length;
948 	if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
949 	    (rt->u.dst.dev->features & NETIF_F_UFO)) {
950 
951 		err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
952 					  fragheaderlen, transhdrlen, mtu,
953 					  flags);
954 		if (err)
955 			goto error;
956 		return 0;
957 	}
958 
959 	if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
960 		goto alloc_new_skb;
961 
962 	while (length > 0) {
963 		/* Check if the remaining data fits into current packet. */
964 		copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
965 		if (copy < length)
966 			copy = maxfraglen - skb->len;
967 
968 		if (copy <= 0) {
969 			char *data;
970 			unsigned int datalen;
971 			unsigned int fraglen;
972 			unsigned int fraggap;
973 			unsigned int alloclen;
974 			struct sk_buff *skb_prev;
975 alloc_new_skb:
976 			skb_prev = skb;
977 
978 			/* There's no room in the current skb */
979 			if (skb_prev)
980 				fraggap = skb_prev->len - maxfraglen;
981 			else
982 				fraggap = 0;
983 
984 			/*
985 			 * If remaining data exceeds the mtu,
986 			 * we know we need more fragment(s).
987 			 */
988 			datalen = length + fraggap;
989 			if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
990 				datalen = maxfraglen - fragheaderlen;
991 
992 			fraglen = datalen + fragheaderlen;
993 			if ((flags & MSG_MORE) &&
994 			    !(rt->u.dst.dev->features&NETIF_F_SG))
995 				alloclen = mtu;
996 			else
997 				alloclen = datalen + fragheaderlen;
998 
999 			/*
1000 			 * The last fragment gets additional space at tail.
1001 			 * Note: we overallocate on fragments with MSG_MODE
1002 			 * because we have no idea if we're the last one.
1003 			 */
1004 			if (datalen == length + fraggap)
1005 				alloclen += rt->u.dst.trailer_len;
1006 
1007 			/*
1008 			 * We just reserve space for fragment header.
1009 			 * Note: this may be overallocation if the message
1010 			 * (without MSG_MORE) fits into the MTU.
1011 			 */
1012 			alloclen += sizeof(struct frag_hdr);
1013 
1014 			if (transhdrlen) {
1015 				skb = sock_alloc_send_skb(sk,
1016 						alloclen + hh_len,
1017 						(flags & MSG_DONTWAIT), &err);
1018 			} else {
1019 				skb = NULL;
1020 				if (atomic_read(&sk->sk_wmem_alloc) <=
1021 				    2 * sk->sk_sndbuf)
1022 					skb = sock_wmalloc(sk,
1023 							   alloclen + hh_len, 1,
1024 							   sk->sk_allocation);
1025 				if (unlikely(skb == NULL))
1026 					err = -ENOBUFS;
1027 			}
1028 			if (skb == NULL)
1029 				goto error;
1030 			/*
1031 			 *	Fill in the control structures
1032 			 */
1033 			skb->ip_summed = csummode;
1034 			skb->csum = 0;
1035 			/* reserve for fragmentation */
1036 			skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1037 
1038 			/*
1039 			 *	Find where to start putting bytes
1040 			 */
1041 			data = skb_put(skb, fraglen);
1042 			skb->nh.raw = data + exthdrlen;
1043 			data += fragheaderlen;
1044 			skb->h.raw = data + exthdrlen;
1045 
1046 			if (fraggap) {
1047 				skb->csum = skb_copy_and_csum_bits(
1048 					skb_prev, maxfraglen,
1049 					data + transhdrlen, fraggap, 0);
1050 				skb_prev->csum = csum_sub(skb_prev->csum,
1051 							  skb->csum);
1052 				data += fraggap;
1053 				skb_trim(skb_prev, maxfraglen);
1054 			}
1055 			copy = datalen - transhdrlen - fraggap;
1056 			if (copy < 0) {
1057 				err = -EINVAL;
1058 				kfree_skb(skb);
1059 				goto error;
1060 			} else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1061 				err = -EFAULT;
1062 				kfree_skb(skb);
1063 				goto error;
1064 			}
1065 
1066 			offset += copy;
1067 			length -= datalen - fraggap;
1068 			transhdrlen = 0;
1069 			exthdrlen = 0;
1070 			csummode = CHECKSUM_NONE;
1071 
1072 			/*
1073 			 * Put the packet on the pending queue
1074 			 */
1075 			__skb_queue_tail(&sk->sk_write_queue, skb);
1076 			continue;
1077 		}
1078 
1079 		if (copy > length)
1080 			copy = length;
1081 
1082 		if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1083 			unsigned int off;
1084 
1085 			off = skb->len;
1086 			if (getfrag(from, skb_put(skb, copy),
1087 						offset, copy, off, skb) < 0) {
1088 				__skb_trim(skb, off);
1089 				err = -EFAULT;
1090 				goto error;
1091 			}
1092 		} else {
1093 			int i = skb_shinfo(skb)->nr_frags;
1094 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1095 			struct page *page = sk->sk_sndmsg_page;
1096 			int off = sk->sk_sndmsg_off;
1097 			unsigned int left;
1098 
1099 			if (page && (left = PAGE_SIZE - off) > 0) {
1100 				if (copy >= left)
1101 					copy = left;
1102 				if (page != frag->page) {
1103 					if (i == MAX_SKB_FRAGS) {
1104 						err = -EMSGSIZE;
1105 						goto error;
1106 					}
1107 					get_page(page);
1108 					skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1109 					frag = &skb_shinfo(skb)->frags[i];
1110 				}
1111 			} else if(i < MAX_SKB_FRAGS) {
1112 				if (copy > PAGE_SIZE)
1113 					copy = PAGE_SIZE;
1114 				page = alloc_pages(sk->sk_allocation, 0);
1115 				if (page == NULL) {
1116 					err = -ENOMEM;
1117 					goto error;
1118 				}
1119 				sk->sk_sndmsg_page = page;
1120 				sk->sk_sndmsg_off = 0;
1121 
1122 				skb_fill_page_desc(skb, i, page, 0, 0);
1123 				frag = &skb_shinfo(skb)->frags[i];
1124 				skb->truesize += PAGE_SIZE;
1125 				atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1126 			} else {
1127 				err = -EMSGSIZE;
1128 				goto error;
1129 			}
1130 			if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1131 				err = -EFAULT;
1132 				goto error;
1133 			}
1134 			sk->sk_sndmsg_off += copy;
1135 			frag->size += copy;
1136 			skb->len += copy;
1137 			skb->data_len += copy;
1138 		}
1139 		offset += copy;
1140 		length -= copy;
1141 	}
1142 	return 0;
1143 error:
1144 	inet->cork.length -= length;
1145 	IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1146 	return err;
1147 }
1148 
1149 int ip6_push_pending_frames(struct sock *sk)
1150 {
1151 	struct sk_buff *skb, *tmp_skb;
1152 	struct sk_buff **tail_skb;
1153 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1154 	struct inet_sock *inet = inet_sk(sk);
1155 	struct ipv6_pinfo *np = inet6_sk(sk);
1156 	struct ipv6hdr *hdr;
1157 	struct ipv6_txoptions *opt = np->cork.opt;
1158 	struct rt6_info *rt = np->cork.rt;
1159 	struct flowi *fl = &inet->cork.fl;
1160 	unsigned char proto = fl->proto;
1161 	int err = 0;
1162 
1163 	if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1164 		goto out;
1165 	tail_skb = &(skb_shinfo(skb)->frag_list);
1166 
1167 	/* move skb->data to ip header from ext header */
1168 	if (skb->data < skb->nh.raw)
1169 		__skb_pull(skb, skb->nh.raw - skb->data);
1170 	while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1171 		__skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1172 		*tail_skb = tmp_skb;
1173 		tail_skb = &(tmp_skb->next);
1174 		skb->len += tmp_skb->len;
1175 		skb->data_len += tmp_skb->len;
1176 		skb->truesize += tmp_skb->truesize;
1177 		__sock_put(tmp_skb->sk);
1178 		tmp_skb->destructor = NULL;
1179 		tmp_skb->sk = NULL;
1180 	}
1181 
1182 	ipv6_addr_copy(final_dst, &fl->fl6_dst);
1183 	__skb_pull(skb, skb->h.raw - skb->nh.raw);
1184 	if (opt && opt->opt_flen)
1185 		ipv6_push_frag_opts(skb, opt, &proto);
1186 	if (opt && opt->opt_nflen)
1187 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1188 
1189 	skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1190 
1191 	*(u32*)hdr = fl->fl6_flowlabel |
1192 		     htonl(0x60000000 | ((int)np->cork.tclass << 20));
1193 
1194 	if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1195 		hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1196 	else
1197 		hdr->payload_len = 0;
1198 	hdr->hop_limit = np->cork.hop_limit;
1199 	hdr->nexthdr = proto;
1200 	ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1201 	ipv6_addr_copy(&hdr->daddr, final_dst);
1202 
1203 	skb->priority = sk->sk_priority;
1204 
1205 	skb->dst = dst_clone(&rt->u.dst);
1206 	IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
1207 	err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1208 	if (err) {
1209 		if (err > 0)
1210 			err = np->recverr ? net_xmit_errno(err) : 0;
1211 		if (err)
1212 			goto error;
1213 	}
1214 
1215 out:
1216 	inet->cork.flags &= ~IPCORK_OPT;
1217 	kfree(np->cork.opt);
1218 	np->cork.opt = NULL;
1219 	if (np->cork.rt) {
1220 		dst_release(&np->cork.rt->u.dst);
1221 		np->cork.rt = NULL;
1222 		inet->cork.flags &= ~IPCORK_ALLFRAG;
1223 	}
1224 	memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1225 	return err;
1226 error:
1227 	goto out;
1228 }
1229 
1230 void ip6_flush_pending_frames(struct sock *sk)
1231 {
1232 	struct inet_sock *inet = inet_sk(sk);
1233 	struct ipv6_pinfo *np = inet6_sk(sk);
1234 	struct sk_buff *skb;
1235 
1236 	while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1237 		IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1238 		kfree_skb(skb);
1239 	}
1240 
1241 	inet->cork.flags &= ~IPCORK_OPT;
1242 
1243 	kfree(np->cork.opt);
1244 	np->cork.opt = NULL;
1245 	if (np->cork.rt) {
1246 		dst_release(&np->cork.rt->u.dst);
1247 		np->cork.rt = NULL;
1248 		inet->cork.flags &= ~IPCORK_ALLFRAG;
1249 	}
1250 	memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1251 }
1252