ia64/linux-2.6.18-xen.hg

view net/ipv6/ip6_output.c @ 673:3161879fdf22

[IA64] xencomm: support XENMEM_add_to_physmap and XENMEM_remove_from_phsymap

support XENMEM_add_to_physmap and XENMEM_remove_from_phsymap.

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
author Isaku Yamahata <yamahata@valinux.co.jp>
date Tue Sep 16 21:26:15 2008 +0900 (2008-09-16)
parents 831230e53067
children
line source
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 */
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>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
46 #include <net/sock.h>
47 #include <net/snmp.h>
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>
59 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
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);
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 }
73 static inline int ip6_output_finish(struct sk_buff *skb)
74 {
76 struct dst_entry *dst = skb->dst;
77 struct hh_cache *hh = dst->hh;
79 if (hh) {
80 int hh_alen;
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);
91 IP6_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
92 kfree_skb(skb);
93 return -EINVAL;
95 }
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);
106 netif_rx(newskb);
107 return 0;
108 }
111 static int ip6_output2(struct sk_buff *skb)
112 {
113 struct dst_entry *dst = skb->dst;
114 struct net_device *dev = dst->dev;
116 skb->protocol = htons(ETH_P_IPV6);
117 skb->dev = dev;
119 if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
120 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
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);
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);
135 if (skb->nh.ipv6h->hop_limit == 0) {
136 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
137 kfree_skb(skb);
138 return 0;
139 }
140 }
142 IP6_INC_STATS(IPSTATS_MIB_OUTMCASTPKTS);
143 }
145 return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
146 }
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 }
157 /*
158 * xmit an sk_buff (used by TCP)
159 */
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;
173 if (opt) {
174 int head_room;
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);
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 }
200 hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
202 /*
203 * Fill in the IPv6 header
204 */
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);
214 tclass = -1;
215 if (np)
216 tclass = np->tclass;
217 if (tclass < 0)
218 tclass = 0;
220 *(u32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
222 hdr->payload_len = htons(seg_len);
223 hdr->nexthdr = proto;
224 hdr->hop_limit = hlimit;
226 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
227 ipv6_addr_copy(&hdr->daddr, first_hop);
229 skb->priority = sk->sk_priority;
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 }
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 }
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 */
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;
262 skb->protocol = htons(ETH_P_IPV6);
263 skb->dev = dev;
265 totlen = len + sizeof(struct ipv6hdr);
267 hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
268 skb->nh.ipv6h = hdr;
270 *(u32*)hdr = htonl(0x60000000);
272 hdr->payload_len = htons(len);
273 hdr->nexthdr = proto;
274 hdr->hop_limit = np->hop_limit;
276 ipv6_addr_copy(&hdr->saddr, saddr);
277 ipv6_addr_copy(&hdr->daddr, daddr);
279 return 0;
280 }
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;
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 }
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 }
311 static inline int ip6_forward_finish(struct sk_buff *skb)
312 {
313 return dst_output(skb);
314 }
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);
322 if (ipv6_devconf.forwarding == 0)
323 goto error;
325 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
326 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
327 goto drop;
328 }
330 skb->ip_summed = CHECKSUM_NONE;
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 }
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 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
361 kfree_skb(skb);
362 return -ETIMEDOUT;
363 }
365 if (!xfrm6_route_forward(skb)) {
366 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
367 goto drop;
368 }
369 dst = skb->dst;
371 /* IPv6 specs say nothing about it, but it is clear that we cannot
372 send redirects to source routed frames.
373 */
374 if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
375 struct in6_addr *target = NULL;
376 struct rt6_info *rt;
377 struct neighbour *n = dst->neighbour;
379 /*
380 * incoming and outgoing devices are the same
381 * send a redirect.
382 */
384 rt = (struct rt6_info *) dst;
385 if ((rt->rt6i_flags & RTF_GATEWAY))
386 target = (struct in6_addr*)&n->primary_key;
387 else
388 target = &hdr->daddr;
390 /* Limit redirects both by destination (here)
391 and by source (inside ndisc_send_redirect)
392 */
393 if (xrlim_allow(dst, 1*HZ))
394 ndisc_send_redirect(skb, n, target);
395 } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
396 |IPV6_ADDR_LINKLOCAL)) {
397 /* This check is security critical. */
398 goto error;
399 }
401 if (skb->len > dst_mtu(dst)) {
402 /* Again, force OUTPUT device used as source address */
403 skb->dev = dst->dev;
404 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
405 IP6_INC_STATS_BH(IPSTATS_MIB_INTOOBIGERRORS);
406 IP6_INC_STATS_BH(IPSTATS_MIB_FRAGFAILS);
407 kfree_skb(skb);
408 return -EMSGSIZE;
409 }
411 if (skb_cow(skb, dst->dev->hard_header_len)) {
412 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
413 goto drop;
414 }
416 hdr = skb->nh.ipv6h;
418 /* Mangling hops number delayed to point after skb COW */
420 hdr->hop_limit--;
422 IP6_INC_STATS_BH(IPSTATS_MIB_OUTFORWDATAGRAMS);
423 return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
425 error:
426 IP6_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
427 drop:
428 kfree_skb(skb);
429 return -EINVAL;
430 }
432 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
433 {
434 to->pkt_type = from->pkt_type;
435 to->priority = from->priority;
436 to->protocol = from->protocol;
437 dst_release(to->dst);
438 to->dst = dst_clone(from->dst);
439 to->dev = from->dev;
441 #ifdef CONFIG_NET_SCHED
442 to->tc_index = from->tc_index;
443 #endif
444 #ifdef CONFIG_NETFILTER
445 to->nfmark = from->nfmark;
446 /* Connection association is same as pre-frag packet */
447 nf_conntrack_put(to->nfct);
448 to->nfct = from->nfct;
449 nf_conntrack_get(to->nfct);
450 to->nfctinfo = from->nfctinfo;
451 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
452 nf_conntrack_put_reasm(to->nfct_reasm);
453 to->nfct_reasm = from->nfct_reasm;
454 nf_conntrack_get_reasm(to->nfct_reasm);
455 #endif
456 #ifdef CONFIG_BRIDGE_NETFILTER
457 nf_bridge_put(to->nf_bridge);
458 to->nf_bridge = from->nf_bridge;
459 nf_bridge_get(to->nf_bridge);
460 #endif
461 #endif
462 skb_copy_secmark(to, from);
463 }
465 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
466 {
467 u16 offset = sizeof(struct ipv6hdr);
468 struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
469 unsigned int packet_len = skb->tail - skb->nh.raw;
470 int found_rhdr = 0;
471 *nexthdr = &skb->nh.ipv6h->nexthdr;
473 while (offset + 1 <= packet_len) {
475 switch (**nexthdr) {
477 case NEXTHDR_HOP:
478 case NEXTHDR_ROUTING:
479 case NEXTHDR_DEST:
480 if (**nexthdr == NEXTHDR_ROUTING) found_rhdr = 1;
481 if (**nexthdr == NEXTHDR_DEST && found_rhdr) return offset;
482 offset += ipv6_optlen(exthdr);
483 *nexthdr = &exthdr->nexthdr;
484 exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
485 break;
486 default :
487 return offset;
488 }
489 }
491 return offset;
492 }
493 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
495 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
496 {
497 struct net_device *dev;
498 struct sk_buff *frag;
499 struct rt6_info *rt = (struct rt6_info*)skb->dst;
500 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
501 struct ipv6hdr *tmp_hdr;
502 struct frag_hdr *fh;
503 unsigned int mtu, hlen, left, len;
504 u32 frag_id = 0;
505 int ptr, offset = 0, err=0;
506 u8 *prevhdr, nexthdr = 0;
508 dev = rt->u.dst.dev;
509 hlen = ip6_find_1stfragopt(skb, &prevhdr);
510 nexthdr = *prevhdr;
512 mtu = dst_mtu(&rt->u.dst);
513 if (np && np->frag_size < mtu) {
514 if (np->frag_size)
515 mtu = np->frag_size;
516 }
517 mtu -= hlen + sizeof(struct frag_hdr);
519 if (skb_shinfo(skb)->frag_list) {
520 int first_len = skb_pagelen(skb);
522 if (first_len - hlen > mtu ||
523 ((first_len - hlen) & 7) ||
524 skb_cloned(skb))
525 goto slow_path;
527 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
528 /* Correct geometry. */
529 if (frag->len > mtu ||
530 ((frag->len & 7) && frag->next) ||
531 skb_headroom(frag) < hlen)
532 goto slow_path;
534 /* Partially cloned skb? */
535 if (skb_shared(frag))
536 goto slow_path;
538 BUG_ON(frag->sk);
539 if (skb->sk) {
540 sock_hold(skb->sk);
541 frag->sk = skb->sk;
542 frag->destructor = sock_wfree;
543 skb->truesize -= frag->truesize;
544 }
545 }
547 err = 0;
548 offset = 0;
549 frag = skb_shinfo(skb)->frag_list;
550 skb_shinfo(skb)->frag_list = NULL;
551 /* BUILD HEADER */
553 tmp_hdr = kmalloc(hlen, GFP_ATOMIC);
554 if (!tmp_hdr) {
555 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
556 return -ENOMEM;
557 }
559 *prevhdr = NEXTHDR_FRAGMENT;
560 memcpy(tmp_hdr, skb->nh.raw, hlen);
561 __skb_pull(skb, hlen);
562 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
563 skb->nh.raw = __skb_push(skb, hlen);
564 memcpy(skb->nh.raw, tmp_hdr, hlen);
566 ipv6_select_ident(skb, fh);
567 fh->nexthdr = nexthdr;
568 fh->reserved = 0;
569 fh->frag_off = htons(IP6_MF);
570 frag_id = fh->identification;
572 first_len = skb_pagelen(skb);
573 skb->data_len = first_len - skb_headlen(skb);
574 skb->len = first_len;
575 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
578 for (;;) {
579 /* Prepare header of the next frame,
580 * before previous one went down. */
581 if (frag) {
582 frag->ip_summed = CHECKSUM_NONE;
583 frag->h.raw = frag->data;
584 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
585 frag->nh.raw = __skb_push(frag, hlen);
586 memcpy(frag->nh.raw, tmp_hdr, hlen);
587 offset += skb->len - hlen - sizeof(struct frag_hdr);
588 fh->nexthdr = nexthdr;
589 fh->reserved = 0;
590 fh->frag_off = htons(offset);
591 if (frag->next != NULL)
592 fh->frag_off |= htons(IP6_MF);
593 fh->identification = frag_id;
594 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
595 ip6_copy_metadata(frag, skb);
596 }
598 err = output(skb);
599 if(!err)
600 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
602 if (err || !frag)
603 break;
605 skb = frag;
606 frag = skb->next;
607 skb->next = NULL;
608 }
610 kfree(tmp_hdr);
612 if (err == 0) {
613 IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
614 return 0;
615 }
617 while (frag) {
618 skb = frag->next;
619 kfree_skb(frag);
620 frag = skb;
621 }
623 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
624 return err;
625 }
627 slow_path:
628 left = skb->len - hlen; /* Space per frame */
629 ptr = hlen; /* Where to start from */
631 /*
632 * Fragment the datagram.
633 */
635 *prevhdr = NEXTHDR_FRAGMENT;
637 /*
638 * Keep copying data until we run out.
639 */
640 while(left > 0) {
641 len = left;
642 /* IF: it doesn't fit, use 'mtu' - the data space left */
643 if (len > mtu)
644 len = mtu;
645 /* IF: we are not sending upto and including the packet end
646 then align the next start on an eight byte boundary */
647 if (len < left) {
648 len &= ~7;
649 }
650 /*
651 * Allocate buffer.
652 */
654 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
655 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
656 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
657 err = -ENOMEM;
658 goto fail;
659 }
661 /*
662 * Set up data on packet
663 */
665 ip6_copy_metadata(frag, skb);
666 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
667 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
668 frag->nh.raw = frag->data;
669 fh = (struct frag_hdr*)(frag->data + hlen);
670 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
672 /*
673 * Charge the memory for the fragment to any owner
674 * it might possess
675 */
676 if (skb->sk)
677 skb_set_owner_w(frag, skb->sk);
679 /*
680 * Copy the packet header into the new buffer.
681 */
682 memcpy(frag->nh.raw, skb->data, hlen);
684 /*
685 * Build fragment header.
686 */
687 fh->nexthdr = nexthdr;
688 fh->reserved = 0;
689 if (!frag_id) {
690 ipv6_select_ident(skb, fh);
691 frag_id = fh->identification;
692 } else
693 fh->identification = frag_id;
695 /*
696 * Copy a block of the IP datagram.
697 */
698 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
699 BUG();
700 left -= len;
702 fh->frag_off = htons(offset);
703 if (left > 0)
704 fh->frag_off |= htons(IP6_MF);
705 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
707 ptr += len;
708 offset += len;
710 /*
711 * Put this fragment into the sending queue.
712 */
713 err = output(frag);
714 if (err)
715 goto fail;
717 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
718 }
719 kfree_skb(skb);
720 IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
721 return err;
723 fail:
724 kfree_skb(skb);
725 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
726 return err;
727 }
729 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
730 struct dst_entry *dst,
731 struct flowi *fl)
732 {
733 struct ipv6_pinfo *np = inet6_sk(sk);
734 struct rt6_info *rt = (struct rt6_info *)dst;
736 if (!dst)
737 goto out;
739 /* Yes, checking route validity in not connected
740 * case is not very simple. Take into account,
741 * that we do not support routing by source, TOS,
742 * and MSG_DONTROUTE --ANK (980726)
743 *
744 * 1. If route was host route, check that
745 * cached destination is current.
746 * If it is network route, we still may
747 * check its validity using saved pointer
748 * to the last used address: daddr_cache.
749 * We do not want to save whole address now,
750 * (because main consumer of this service
751 * is tcp, which has not this problem),
752 * so that the last trick works only on connected
753 * sockets.
754 * 2. oif also should be the same.
755 */
756 if (((rt->rt6i_dst.plen != 128 ||
757 !ipv6_addr_equal(&fl->fl6_dst, &rt->rt6i_dst.addr))
758 && (np->daddr_cache == NULL ||
759 !ipv6_addr_equal(&fl->fl6_dst, np->daddr_cache)))
760 || (fl->oif && fl->oif != dst->dev->ifindex)) {
761 dst_release(dst);
762 dst = NULL;
763 }
765 out:
766 return dst;
767 }
769 static int ip6_dst_lookup_tail(struct sock *sk,
770 struct dst_entry **dst, struct flowi *fl)
771 {
772 int err;
774 if (*dst == NULL)
775 *dst = ip6_route_output(sk, fl);
777 if ((err = (*dst)->error))
778 goto out_err_release;
780 if (ipv6_addr_any(&fl->fl6_src)) {
781 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
782 if (err)
783 goto out_err_release;
784 }
786 return 0;
788 out_err_release:
789 dst_release(*dst);
790 *dst = NULL;
791 return err;
792 }
794 /**
795 * ip6_dst_lookup - perform route lookup on flow
796 * @sk: socket which provides route info
797 * @dst: pointer to dst_entry * for result
798 * @fl: flow to lookup
799 *
800 * This function performs a route lookup on the given flow.
801 *
802 * It returns zero on success, or a standard errno code on error.
803 */
804 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
805 {
806 *dst = NULL;
807 return ip6_dst_lookup_tail(sk, dst, fl);
808 }
809 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
811 /**
812 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
813 * @sk: socket which provides the dst cache and route info
814 * @dst: pointer to dst_entry * for result
815 * @fl: flow to lookup
816 *
817 * This function performs a route lookup on the given flow with the
818 * possibility of using the cached route in the socket if it is valid.
819 * It will take the socket dst lock when operating on the dst cache.
820 * As a result, this function can only be used in process context.
821 *
822 * It returns zero on success, or a standard errno code on error.
823 */
824 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
825 {
826 *dst = NULL;
827 if (sk) {
828 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
829 *dst = ip6_sk_dst_check(sk, *dst, fl);
830 }
832 return ip6_dst_lookup_tail(sk, dst, fl);
833 }
834 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
836 static inline int ip6_ufo_append_data(struct sock *sk,
837 int getfrag(void *from, char *to, int offset, int len,
838 int odd, struct sk_buff *skb),
839 void *from, int length, int hh_len, int fragheaderlen,
840 int transhdrlen, int mtu,unsigned int flags)
842 {
843 struct sk_buff *skb;
844 int err;
846 /* There is support for UDP large send offload by network
847 * device, so create one single skb packet containing complete
848 * udp datagram
849 */
850 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
851 skb = sock_alloc_send_skb(sk,
852 hh_len + fragheaderlen + transhdrlen + 20,
853 (flags & MSG_DONTWAIT), &err);
854 if (skb == NULL)
855 return -ENOMEM;
857 /* reserve space for Hardware header */
858 skb_reserve(skb, hh_len);
860 /* create space for UDP/IP header */
861 skb_put(skb,fragheaderlen + transhdrlen);
863 /* initialize network header pointer */
864 skb->nh.raw = skb->data;
866 /* initialize protocol header pointer */
867 skb->h.raw = skb->data + fragheaderlen;
869 skb->ip_summed = CHECKSUM_HW;
870 skb->csum = 0;
871 sk->sk_sndmsg_off = 0;
872 }
874 err = skb_append_datato_frags(sk,skb, getfrag, from,
875 (length - transhdrlen));
876 if (!err) {
877 struct frag_hdr fhdr;
879 /* specify the length of each IP datagram fragment*/
880 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
881 sizeof(struct frag_hdr);
882 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
883 ipv6_select_ident(skb, &fhdr);
884 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
885 __skb_queue_tail(&sk->sk_write_queue, skb);
887 return 0;
888 }
889 /* There is not enough support do UPD LSO,
890 * so follow normal path
891 */
892 kfree_skb(skb);
894 return err;
895 }
897 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
898 int offset, int len, int odd, struct sk_buff *skb),
899 void *from, int length, int transhdrlen,
900 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
901 struct rt6_info *rt, unsigned int flags)
902 {
903 struct inet_sock *inet = inet_sk(sk);
904 struct ipv6_pinfo *np = inet6_sk(sk);
905 struct sk_buff *skb;
906 unsigned int maxfraglen, fragheaderlen;
907 int exthdrlen;
908 int hh_len;
909 int mtu;
910 int copy;
911 int err;
912 int offset = 0;
913 int csummode = CHECKSUM_NONE;
915 if (flags&MSG_PROBE)
916 return 0;
917 if (skb_queue_empty(&sk->sk_write_queue)) {
918 /*
919 * setup for corking
920 */
921 if (opt) {
922 if (np->cork.opt == NULL) {
923 np->cork.opt = kmalloc(opt->tot_len,
924 sk->sk_allocation);
925 if (unlikely(np->cork.opt == NULL))
926 return -ENOBUFS;
927 } else if (np->cork.opt->tot_len < opt->tot_len) {
928 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
929 return -EINVAL;
930 }
931 memcpy(np->cork.opt, opt, opt->tot_len);
932 inet->cork.flags |= IPCORK_OPT;
933 /* need source address above miyazawa*/
934 }
935 dst_hold(&rt->u.dst);
936 np->cork.rt = rt;
937 inet->cork.fl = *fl;
938 np->cork.hop_limit = hlimit;
939 np->cork.tclass = tclass;
940 mtu = dst_mtu(rt->u.dst.path);
941 if (np->frag_size < mtu) {
942 if (np->frag_size)
943 mtu = np->frag_size;
944 }
945 inet->cork.fragsize = mtu;
946 if (dst_allfrag(rt->u.dst.path))
947 inet->cork.flags |= IPCORK_ALLFRAG;
948 inet->cork.length = 0;
949 sk->sk_sndmsg_page = NULL;
950 sk->sk_sndmsg_off = 0;
951 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
952 length += exthdrlen;
953 transhdrlen += exthdrlen;
954 } else {
955 rt = np->cork.rt;
956 fl = &inet->cork.fl;
957 if (inet->cork.flags & IPCORK_OPT)
958 opt = np->cork.opt;
959 transhdrlen = 0;
960 exthdrlen = 0;
961 mtu = inet->cork.fragsize;
962 }
964 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
966 fragheaderlen = sizeof(struct ipv6hdr) + (opt ? opt->opt_nflen : 0);
967 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
969 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
970 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
971 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
972 return -EMSGSIZE;
973 }
974 }
976 /*
977 * Let's try using as much space as possible.
978 * Use MTU if total length of the message fits into the MTU.
979 * Otherwise, we need to reserve fragment header and
980 * fragment alignment (= 8-15 octects, in total).
981 *
982 * Note that we may need to "move" the data from the tail of
983 * of the buffer to the new fragment when we split
984 * the message.
985 *
986 * FIXME: It may be fragmented into multiple chunks
987 * at once if non-fragmentable extension headers
988 * are too large.
989 * --yoshfuji
990 */
992 inet->cork.length += length;
993 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
994 (rt->u.dst.dev->features & NETIF_F_UFO)) {
996 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
997 fragheaderlen, transhdrlen, mtu,
998 flags);
999 if (err)
1000 goto error;
1001 return 0;
1004 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1005 goto alloc_new_skb;
1007 while (length > 0) {
1008 /* Check if the remaining data fits into current packet. */
1009 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1010 if (copy < length)
1011 copy = maxfraglen - skb->len;
1013 if (copy <= 0) {
1014 char *data;
1015 unsigned int datalen;
1016 unsigned int fraglen;
1017 unsigned int fraggap;
1018 unsigned int alloclen;
1019 struct sk_buff *skb_prev;
1020 alloc_new_skb:
1021 skb_prev = skb;
1023 /* There's no room in the current skb */
1024 if (skb_prev)
1025 fraggap = skb_prev->len - maxfraglen;
1026 else
1027 fraggap = 0;
1029 /*
1030 * If remaining data exceeds the mtu,
1031 * we know we need more fragment(s).
1032 */
1033 datalen = length + fraggap;
1034 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1035 datalen = maxfraglen - fragheaderlen;
1037 fraglen = datalen + fragheaderlen;
1038 if ((flags & MSG_MORE) &&
1039 !(rt->u.dst.dev->features&NETIF_F_SG))
1040 alloclen = mtu;
1041 else
1042 alloclen = datalen + fragheaderlen;
1044 /*
1045 * The last fragment gets additional space at tail.
1046 * Note: we overallocate on fragments with MSG_MODE
1047 * because we have no idea if we're the last one.
1048 */
1049 if (datalen == length + fraggap)
1050 alloclen += rt->u.dst.trailer_len;
1052 /*
1053 * We just reserve space for fragment header.
1054 * Note: this may be overallocation if the message
1055 * (without MSG_MORE) fits into the MTU.
1056 */
1057 alloclen += sizeof(struct frag_hdr);
1059 if (transhdrlen) {
1060 skb = sock_alloc_send_skb(sk,
1061 alloclen + hh_len,
1062 (flags & MSG_DONTWAIT), &err);
1063 } else {
1064 skb = NULL;
1065 if (atomic_read(&sk->sk_wmem_alloc) <=
1066 2 * sk->sk_sndbuf)
1067 skb = sock_wmalloc(sk,
1068 alloclen + hh_len, 1,
1069 sk->sk_allocation);
1070 if (unlikely(skb == NULL))
1071 err = -ENOBUFS;
1073 if (skb == NULL)
1074 goto error;
1075 /*
1076 * Fill in the control structures
1077 */
1078 skb->ip_summed = csummode;
1079 skb->csum = 0;
1080 /* reserve for fragmentation */
1081 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1083 /*
1084 * Find where to start putting bytes
1085 */
1086 data = skb_put(skb, fraglen);
1087 skb->nh.raw = data + exthdrlen;
1088 data += fragheaderlen;
1089 skb->h.raw = data + exthdrlen;
1091 if (fraggap) {
1092 skb->csum = skb_copy_and_csum_bits(
1093 skb_prev, maxfraglen,
1094 data + transhdrlen, fraggap, 0);
1095 skb_prev->csum = csum_sub(skb_prev->csum,
1096 skb->csum);
1097 data += fraggap;
1098 pskb_trim_unique(skb_prev, maxfraglen);
1100 copy = datalen - transhdrlen - fraggap;
1101 if (copy < 0) {
1102 err = -EINVAL;
1103 kfree_skb(skb);
1104 goto error;
1105 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1106 err = -EFAULT;
1107 kfree_skb(skb);
1108 goto error;
1111 offset += copy;
1112 length -= datalen - fraggap;
1113 transhdrlen = 0;
1114 exthdrlen = 0;
1115 csummode = CHECKSUM_NONE;
1117 /*
1118 * Put the packet on the pending queue
1119 */
1120 __skb_queue_tail(&sk->sk_write_queue, skb);
1121 continue;
1124 if (copy > length)
1125 copy = length;
1127 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1128 unsigned int off;
1130 off = skb->len;
1131 if (getfrag(from, skb_put(skb, copy),
1132 offset, copy, off, skb) < 0) {
1133 __skb_trim(skb, off);
1134 err = -EFAULT;
1135 goto error;
1137 } else {
1138 int i = skb_shinfo(skb)->nr_frags;
1139 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1140 struct page *page = sk->sk_sndmsg_page;
1141 int off = sk->sk_sndmsg_off;
1142 unsigned int left;
1144 if (page && (left = PAGE_SIZE - off) > 0) {
1145 if (copy >= left)
1146 copy = left;
1147 if (page != frag->page) {
1148 if (i == MAX_SKB_FRAGS) {
1149 err = -EMSGSIZE;
1150 goto error;
1152 get_page(page);
1153 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1154 frag = &skb_shinfo(skb)->frags[i];
1156 } else if(i < MAX_SKB_FRAGS) {
1157 if (copy > PAGE_SIZE)
1158 copy = PAGE_SIZE;
1159 page = alloc_pages(sk->sk_allocation, 0);
1160 if (page == NULL) {
1161 err = -ENOMEM;
1162 goto error;
1164 sk->sk_sndmsg_page = page;
1165 sk->sk_sndmsg_off = 0;
1167 skb_fill_page_desc(skb, i, page, 0, 0);
1168 frag = &skb_shinfo(skb)->frags[i];
1169 skb->truesize += PAGE_SIZE;
1170 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1171 } else {
1172 err = -EMSGSIZE;
1173 goto error;
1175 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1176 err = -EFAULT;
1177 goto error;
1179 sk->sk_sndmsg_off += copy;
1180 frag->size += copy;
1181 skb->len += copy;
1182 skb->data_len += copy;
1184 offset += copy;
1185 length -= copy;
1187 return 0;
1188 error:
1189 inet->cork.length -= length;
1190 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1191 return err;
1194 int ip6_push_pending_frames(struct sock *sk)
1196 struct sk_buff *skb, *tmp_skb;
1197 struct sk_buff **tail_skb;
1198 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1199 struct inet_sock *inet = inet_sk(sk);
1200 struct ipv6_pinfo *np = inet6_sk(sk);
1201 struct ipv6hdr *hdr;
1202 struct ipv6_txoptions *opt = np->cork.opt;
1203 struct rt6_info *rt = np->cork.rt;
1204 struct flowi *fl = &inet->cork.fl;
1205 unsigned char proto = fl->proto;
1206 int err = 0;
1208 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1209 goto out;
1210 tail_skb = &(skb_shinfo(skb)->frag_list);
1212 /* move skb->data to ip header from ext header */
1213 if (skb->data < skb->nh.raw)
1214 __skb_pull(skb, skb->nh.raw - skb->data);
1215 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1216 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1217 *tail_skb = tmp_skb;
1218 tail_skb = &(tmp_skb->next);
1219 skb->len += tmp_skb->len;
1220 skb->data_len += tmp_skb->len;
1221 skb->truesize += tmp_skb->truesize;
1222 __sock_put(tmp_skb->sk);
1223 tmp_skb->destructor = NULL;
1224 tmp_skb->sk = NULL;
1227 ipv6_addr_copy(final_dst, &fl->fl6_dst);
1228 __skb_pull(skb, skb->h.raw - skb->nh.raw);
1229 if (opt && opt->opt_flen)
1230 ipv6_push_frag_opts(skb, opt, &proto);
1231 if (opt && opt->opt_nflen)
1232 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1234 skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1236 *(u32*)hdr = fl->fl6_flowlabel |
1237 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1239 if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1240 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1241 else
1242 hdr->payload_len = 0;
1243 hdr->hop_limit = np->cork.hop_limit;
1244 hdr->nexthdr = proto;
1245 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1246 ipv6_addr_copy(&hdr->daddr, final_dst);
1248 skb->priority = sk->sk_priority;
1250 skb->dst = dst_clone(&rt->u.dst);
1251 IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
1252 err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1253 if (err) {
1254 if (err > 0)
1255 err = np->recverr ? net_xmit_errno(err) : 0;
1256 if (err)
1257 goto error;
1260 out:
1261 inet->cork.flags &= ~IPCORK_OPT;
1262 kfree(np->cork.opt);
1263 np->cork.opt = NULL;
1264 if (np->cork.rt) {
1265 dst_release(&np->cork.rt->u.dst);
1266 np->cork.rt = NULL;
1267 inet->cork.flags &= ~IPCORK_ALLFRAG;
1269 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1270 return err;
1271 error:
1272 goto out;
1275 void ip6_flush_pending_frames(struct sock *sk)
1277 struct inet_sock *inet = inet_sk(sk);
1278 struct ipv6_pinfo *np = inet6_sk(sk);
1279 struct sk_buff *skb;
1281 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1282 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1283 kfree_skb(skb);
1286 inet->cork.flags &= ~IPCORK_OPT;
1288 kfree(np->cork.opt);
1289 np->cork.opt = NULL;
1290 if (np->cork.rt) {
1291 dst_release(&np->cork.rt->u.dst);
1292 np->cork.rt = NULL;
1293 inet->cork.flags &= ~IPCORK_ALLFRAG;
1295 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));