1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * IPv6 output functions
4 * Linux INET6 implementation
7 * Pedro Roque <roque@di.fc.ul.pt>
9 * Based on linux/net/ipv4/ip_output.c
12 * A.N.Kuznetsov : airthmetics in fragmentation.
13 * extension headers are implemented.
14 * route changes now work.
15 * ip6_forward does not confuse sniffers.
18 * H. von Brand : Added missing #include <linux/string.h>
19 * Imran Patel : frag id should be in NBO
20 * Kazunori MIYAZAWA @USAGI
21 * : add ip6_append_data and related functions
25 #include <linux/errno.h>
26 #include <linux/kernel.h>
27 #include <linux/string.h>
28 #include <linux/socket.h>
29 #include <linux/net.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/in6.h>
33 #include <linux/tcp.h>
34 #include <linux/route.h>
35 #include <linux/module.h>
36 #include <linux/slab.h>
38 #include <linux/bpf-cgroup.h>
39 #include <linux/netfilter.h>
40 #include <linux/netfilter_ipv6.h>
46 #include <net/ndisc.h>
47 #include <net/protocol.h>
48 #include <net/ip6_route.h>
49 #include <net/addrconf.h>
50 #include <net/rawv6.h>
53 #include <net/checksum.h>
54 #include <linux/mroute6.h>
55 #include <net/l3mdev.h>
56 #include <net/lwtunnel.h>
57 #include <net/ip_tunnels.h>
59 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
61 struct dst_entry *dst = skb_dst(skb);
62 struct net_device *dev = dst->dev;
63 struct inet6_dev *idev = ip6_dst_idev(dst);
64 unsigned int hh_len = LL_RESERVED_SPACE(dev);
65 const struct in6_addr *daddr, *nexthop;
67 struct neighbour *neigh;
70 /* Be paranoid, rather than too clever. */
71 if (unlikely(hh_len > skb_headroom(skb)) && dev->header_ops) {
72 skb = skb_expand_head(skb, hh_len);
74 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
81 if (ipv6_addr_is_multicast(daddr)) {
82 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
83 ((mroute6_is_socket(net, skb) &&
84 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
85 ipv6_chk_mcast_addr(dev, daddr, &hdr->saddr))) {
86 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
88 /* Do not check for IFF_ALLMULTI; multicast routing
89 is not supported in any case.
92 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
93 net, sk, newskb, NULL, newskb->dev,
96 if (hdr->hop_limit == 0) {
97 IP6_INC_STATS(net, idev,
98 IPSTATS_MIB_OUTDISCARDS);
104 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
105 if (IPV6_ADDR_MC_SCOPE(daddr) <= IPV6_ADDR_SCOPE_NODELOCAL &&
106 !(dev->flags & IFF_LOOPBACK)) {
112 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
113 int res = lwtunnel_xmit(skb);
115 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
120 nexthop = rt6_nexthop((struct rt6_info *)dst, daddr);
121 neigh = __ipv6_neigh_lookup_noref(dev, nexthop);
122 if (unlikely(!neigh))
123 neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
124 if (!IS_ERR(neigh)) {
125 sock_confirm_neigh(skb, neigh);
126 ret = neigh_output(neigh, skb, false);
127 rcu_read_unlock_bh();
130 rcu_read_unlock_bh();
132 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
138 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
139 struct sk_buff *skb, unsigned int mtu)
141 struct sk_buff *segs, *nskb;
142 netdev_features_t features;
145 /* Please see corresponding comment in ip_finish_output_gso
146 * describing the cases where GSO segment length exceeds the
149 features = netif_skb_features(skb);
150 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
151 if (IS_ERR_OR_NULL(segs)) {
158 skb_list_walk_safe(segs, segs, nskb) {
161 skb_mark_not_on_list(segs);
162 err = ip6_fragment(net, sk, segs, ip6_finish_output2);
170 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
174 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
175 /* Policy lookup after SNAT yielded a new policy */
176 if (skb_dst(skb)->xfrm) {
177 IP6CB(skb)->flags |= IP6SKB_REROUTED;
178 return dst_output(net, sk, skb);
182 mtu = ip6_skb_dst_mtu(skb);
183 if (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu))
184 return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
186 if ((skb->len > mtu && !skb_is_gso(skb)) ||
187 dst_allfrag(skb_dst(skb)) ||
188 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
189 return ip6_fragment(net, sk, skb, ip6_finish_output2);
191 return ip6_finish_output2(net, sk, skb);
194 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
198 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
200 case NET_XMIT_SUCCESS:
201 return __ip6_finish_output(net, sk, skb);
203 return __ip6_finish_output(net, sk, skb) ? : ret;
210 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
212 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
213 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
215 skb->protocol = htons(ETH_P_IPV6);
218 if (unlikely(idev->cnf.disable_ipv6)) {
219 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
224 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
225 net, sk, skb, indev, dev,
227 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
229 EXPORT_SYMBOL(ip6_output);
231 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
233 if (!np->autoflowlabel_set)
234 return ip6_default_np_autolabel(net);
236 return np->autoflowlabel;
240 * xmit an sk_buff (used by TCP, SCTP and DCCP)
241 * Note : socket lock is not held for SYNACK packets, but might be modified
242 * by calls to skb_set_owner_w() and ipv6_local_error(),
243 * which are using proper atomic operations or spinlocks.
245 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
246 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
248 struct net *net = sock_net(sk);
249 const struct ipv6_pinfo *np = inet6_sk(sk);
250 struct in6_addr *first_hop = &fl6->daddr;
251 struct dst_entry *dst = skb_dst(skb);
252 struct net_device *dev = dst->dev;
253 struct inet6_dev *idev = ip6_dst_idev(dst);
254 unsigned int head_room;
256 u8 proto = fl6->flowi6_proto;
257 int seg_len = skb->len;
261 head_room = sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dev);
263 head_room += opt->opt_nflen + opt->opt_flen;
265 if (unlikely(head_room > skb_headroom(skb))) {
266 skb = skb_expand_head(skb, head_room);
268 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
274 seg_len += opt->opt_nflen + opt->opt_flen;
277 ipv6_push_frag_opts(skb, opt, &proto);
280 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
284 skb_push(skb, sizeof(struct ipv6hdr));
285 skb_reset_network_header(skb);
289 * Fill in the IPv6 header
292 hlimit = np->hop_limit;
294 hlimit = ip6_dst_hoplimit(dst);
296 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
297 ip6_autoflowlabel(net, np), fl6));
299 hdr->payload_len = htons(seg_len);
300 hdr->nexthdr = proto;
301 hdr->hop_limit = hlimit;
303 hdr->saddr = fl6->saddr;
304 hdr->daddr = *first_hop;
306 skb->protocol = htons(ETH_P_IPV6);
307 skb->priority = priority;
311 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
312 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
314 /* if egress device is enslaved to an L3 master device pass the
315 * skb to its handler for processing
317 skb = l3mdev_ip6_out((struct sock *)sk, skb);
321 /* hooks should never assume socket lock is held.
322 * we promote our socket to non const
324 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
325 net, (struct sock *)sk, skb, NULL, dev,
330 /* ipv6_local_error() does not require socket lock,
331 * we promote our socket to non const
333 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
335 IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
339 EXPORT_SYMBOL(ip6_xmit);
341 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
343 struct ip6_ra_chain *ra;
344 struct sock *last = NULL;
346 read_lock(&ip6_ra_lock);
347 for (ra = ip6_ra_chain; ra; ra = ra->next) {
348 struct sock *sk = ra->sk;
349 if (sk && ra->sel == sel &&
350 (!sk->sk_bound_dev_if ||
351 sk->sk_bound_dev_if == skb->dev->ifindex)) {
352 struct ipv6_pinfo *np = inet6_sk(sk);
354 if (np && np->rtalert_isolate &&
355 !net_eq(sock_net(sk), dev_net(skb->dev))) {
359 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
361 rawv6_rcv(last, skb2);
368 rawv6_rcv(last, skb);
369 read_unlock(&ip6_ra_lock);
372 read_unlock(&ip6_ra_lock);
376 static int ip6_forward_proxy_check(struct sk_buff *skb)
378 struct ipv6hdr *hdr = ipv6_hdr(skb);
379 u8 nexthdr = hdr->nexthdr;
383 if (ipv6_ext_hdr(nexthdr)) {
384 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
388 offset = sizeof(struct ipv6hdr);
390 if (nexthdr == IPPROTO_ICMPV6) {
391 struct icmp6hdr *icmp6;
393 if (!pskb_may_pull(skb, (skb_network_header(skb) +
394 offset + 1 - skb->data)))
397 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
399 switch (icmp6->icmp6_type) {
400 case NDISC_ROUTER_SOLICITATION:
401 case NDISC_ROUTER_ADVERTISEMENT:
402 case NDISC_NEIGHBOUR_SOLICITATION:
403 case NDISC_NEIGHBOUR_ADVERTISEMENT:
405 /* For reaction involving unicast neighbor discovery
406 * message destined to the proxied address, pass it to
416 * The proxying router can't forward traffic sent to a link-local
417 * address, so signal the sender and discard the packet. This
418 * behavior is clarified by the MIPv6 specification.
420 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
421 dst_link_failure(skb);
428 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
431 struct dst_entry *dst = skb_dst(skb);
433 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
434 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
436 #ifdef CONFIG_NET_SWITCHDEV
437 if (skb->offload_l3_fwd_mark) {
444 return dst_output(net, sk, skb);
447 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
452 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
453 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
459 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
465 int ip6_forward(struct sk_buff *skb)
467 struct dst_entry *dst = skb_dst(skb);
468 struct ipv6hdr *hdr = ipv6_hdr(skb);
469 struct inet6_skb_parm *opt = IP6CB(skb);
470 struct net *net = dev_net(dst->dev);
471 struct inet6_dev *idev;
474 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
475 if (net->ipv6.devconf_all->forwarding == 0)
478 if (skb->pkt_type != PACKET_HOST)
481 if (unlikely(skb->sk))
484 if (skb_warn_if_lro(skb))
487 if (!net->ipv6.devconf_all->disable_policy &&
488 (!idev || !idev->cnf.disable_policy) &&
489 !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
490 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
494 skb_forward_csum(skb);
497 * We DO NOT make any processing on
498 * RA packets, pushing them to user level AS IS
499 * without ane WARRANTY that application will be able
500 * to interpret them. The reason is that we
501 * cannot make anything clever here.
503 * We are not end-node, so that if packet contains
504 * AH/ESP, we cannot make anything.
505 * Defragmentation also would be mistake, RA packets
506 * cannot be fragmented, because there is no warranty
507 * that different fragments will go along one path. --ANK
509 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
510 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
515 * check and decrement ttl
517 if (hdr->hop_limit <= 1) {
518 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
519 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
525 /* XXX: idev->cnf.proxy_ndp? */
526 if (net->ipv6.devconf_all->proxy_ndp &&
527 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
528 int proxied = ip6_forward_proxy_check(skb);
531 return ip6_input(skb);
532 } else if (proxied < 0) {
533 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
538 if (!xfrm6_route_forward(skb)) {
539 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
544 /* IPv6 specs say nothing about it, but it is clear that we cannot
545 send redirects to source routed frames.
546 We don't send redirects to frames decapsulated from IPsec.
548 if (IP6CB(skb)->iif == dst->dev->ifindex &&
549 opt->srcrt == 0 && !skb_sec_path(skb)) {
550 struct in6_addr *target = NULL;
551 struct inet_peer *peer;
555 * incoming and outgoing devices are the same
559 rt = (struct rt6_info *) dst;
560 if (rt->rt6i_flags & RTF_GATEWAY)
561 target = &rt->rt6i_gateway;
563 target = &hdr->daddr;
565 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
567 /* Limit redirects both by destination (here)
568 and by source (inside ndisc_send_redirect)
570 if (inet_peer_xrlim_allow(peer, 1*HZ))
571 ndisc_send_redirect(skb, target);
575 int addrtype = ipv6_addr_type(&hdr->saddr);
577 /* This check is security critical. */
578 if (addrtype == IPV6_ADDR_ANY ||
579 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
581 if (addrtype & IPV6_ADDR_LINKLOCAL) {
582 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
583 ICMPV6_NOT_NEIGHBOUR, 0);
588 mtu = ip6_dst_mtu_maybe_forward(dst, true);
589 if (mtu < IPV6_MIN_MTU)
592 if (ip6_pkt_too_big(skb, mtu)) {
593 /* Again, force OUTPUT device used as source address */
595 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
596 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
597 __IP6_INC_STATS(net, ip6_dst_idev(dst),
598 IPSTATS_MIB_FRAGFAILS);
603 if (skb_cow(skb, dst->dev->hard_header_len)) {
604 __IP6_INC_STATS(net, ip6_dst_idev(dst),
605 IPSTATS_MIB_OUTDISCARDS);
611 /* Mangling hops number delayed to point after skb COW */
615 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
616 net, NULL, skb, skb->dev, dst->dev,
620 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
626 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
628 to->pkt_type = from->pkt_type;
629 to->priority = from->priority;
630 to->protocol = from->protocol;
632 skb_dst_set(to, dst_clone(skb_dst(from)));
634 to->mark = from->mark;
636 skb_copy_hash(to, from);
638 #ifdef CONFIG_NET_SCHED
639 to->tc_index = from->tc_index;
642 skb_ext_copy(to, from);
643 skb_copy_secmark(to, from);
646 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
647 u8 nexthdr, __be32 frag_id,
648 struct ip6_fraglist_iter *iter)
650 unsigned int first_len;
654 *prevhdr = NEXTHDR_FRAGMENT;
655 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
659 iter->frag = skb_shinfo(skb)->frag_list;
660 skb_frag_list_init(skb);
664 iter->frag_id = frag_id;
665 iter->nexthdr = nexthdr;
667 __skb_pull(skb, hlen);
668 fh = __skb_push(skb, sizeof(struct frag_hdr));
669 __skb_push(skb, hlen);
670 skb_reset_network_header(skb);
671 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
673 fh->nexthdr = nexthdr;
675 fh->frag_off = htons(IP6_MF);
676 fh->identification = frag_id;
678 first_len = skb_pagelen(skb);
679 skb->data_len = first_len - skb_headlen(skb);
680 skb->len = first_len;
681 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
685 EXPORT_SYMBOL(ip6_fraglist_init);
687 void ip6_fraglist_prepare(struct sk_buff *skb,
688 struct ip6_fraglist_iter *iter)
690 struct sk_buff *frag = iter->frag;
691 unsigned int hlen = iter->hlen;
694 frag->ip_summed = CHECKSUM_NONE;
695 skb_reset_transport_header(frag);
696 fh = __skb_push(frag, sizeof(struct frag_hdr));
697 __skb_push(frag, hlen);
698 skb_reset_network_header(frag);
699 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
700 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
701 fh->nexthdr = iter->nexthdr;
703 fh->frag_off = htons(iter->offset);
705 fh->frag_off |= htons(IP6_MF);
706 fh->identification = iter->frag_id;
707 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
708 ip6_copy_metadata(frag, skb);
710 EXPORT_SYMBOL(ip6_fraglist_prepare);
712 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
713 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
714 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
716 state->prevhdr = prevhdr;
717 state->nexthdr = nexthdr;
718 state->frag_id = frag_id;
723 state->left = skb->len - hlen; /* Space per frame */
724 state->ptr = hlen; /* Where to start from */
726 state->hroom = hdr_room;
727 state->troom = needed_tailroom;
731 EXPORT_SYMBOL(ip6_frag_init);
733 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
735 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
736 struct sk_buff *frag;
741 /* IF: it doesn't fit, use 'mtu' - the data space left */
742 if (len > state->mtu)
744 /* IF: we are not sending up to and including the packet end
745 then align the next start on an eight byte boundary */
746 if (len < state->left)
749 /* Allocate buffer */
750 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
751 state->hroom + state->troom, GFP_ATOMIC);
753 return ERR_PTR(-ENOMEM);
756 * Set up data on packet
759 ip6_copy_metadata(frag, skb);
760 skb_reserve(frag, state->hroom);
761 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
762 skb_reset_network_header(frag);
763 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
764 frag->transport_header = (frag->network_header + state->hlen +
765 sizeof(struct frag_hdr));
768 * Charge the memory for the fragment to any owner
772 skb_set_owner_w(frag, skb->sk);
775 * Copy the packet header into the new buffer.
777 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
779 fragnexthdr_offset = skb_network_header(frag);
780 fragnexthdr_offset += prevhdr - skb_network_header(skb);
781 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
784 * Build fragment header.
786 fh->nexthdr = state->nexthdr;
788 fh->identification = state->frag_id;
791 * Copy a block of the IP datagram.
793 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
797 fh->frag_off = htons(state->offset);
799 fh->frag_off |= htons(IP6_MF);
800 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
803 state->offset += len;
807 EXPORT_SYMBOL(ip6_frag_next);
809 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
810 int (*output)(struct net *, struct sock *, struct sk_buff *))
812 struct sk_buff *frag;
813 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
814 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
815 inet6_sk(skb->sk) : NULL;
816 struct ip6_frag_state state;
817 unsigned int mtu, hlen, nexthdr_offset;
818 ktime_t tstamp = skb->tstamp;
821 u8 *prevhdr, nexthdr = 0;
823 err = ip6_find_1stfragopt(skb, &prevhdr);
828 nexthdr_offset = prevhdr - skb_network_header(skb);
830 mtu = ip6_skb_dst_mtu(skb);
832 /* We must not fragment if the socket is set to force MTU discovery
833 * or if the skb it not generated by a local socket.
835 if (unlikely(!skb->ignore_df && skb->len > mtu))
838 if (IP6CB(skb)->frag_max_size) {
839 if (IP6CB(skb)->frag_max_size > mtu)
842 /* don't send fragments larger than what we received */
843 mtu = IP6CB(skb)->frag_max_size;
844 if (mtu < IPV6_MIN_MTU)
848 if (np && np->frag_size < mtu) {
852 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
854 mtu -= hlen + sizeof(struct frag_hdr);
856 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
857 &ipv6_hdr(skb)->saddr);
859 if (skb->ip_summed == CHECKSUM_PARTIAL &&
860 (err = skb_checksum_help(skb)))
863 prevhdr = skb_network_header(skb) + nexthdr_offset;
864 hroom = LL_RESERVED_SPACE(rt->dst.dev);
865 if (skb_has_frag_list(skb)) {
866 unsigned int first_len = skb_pagelen(skb);
867 struct ip6_fraglist_iter iter;
868 struct sk_buff *frag2;
870 if (first_len - hlen > mtu ||
871 ((first_len - hlen) & 7) ||
873 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
876 skb_walk_frags(skb, frag) {
877 /* Correct geometry. */
878 if (frag->len > mtu ||
879 ((frag->len & 7) && frag->next) ||
880 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
881 goto slow_path_clean;
883 /* Partially cloned skb? */
884 if (skb_shared(frag))
885 goto slow_path_clean;
890 frag->destructor = sock_wfree;
892 skb->truesize -= frag->truesize;
895 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
900 /* We prevent @rt from being freed. */
904 /* Prepare header of the next frame,
905 * before previous one went down. */
907 ip6_fraglist_prepare(skb, &iter);
909 skb->tstamp = tstamp;
910 err = output(net, sk, skb);
912 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
913 IPSTATS_MIB_FRAGCREATES);
915 if (err || !iter.frag)
918 skb = ip6_fraglist_next(&iter);
924 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
925 IPSTATS_MIB_FRAGOKS);
930 kfree_skb_list(iter.frag);
932 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
933 IPSTATS_MIB_FRAGFAILS);
938 skb_walk_frags(skb, frag2) {
942 frag2->destructor = NULL;
943 skb->truesize += frag2->truesize;
949 * Fragment the datagram.
952 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
953 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
957 * Keep copying data until we run out.
960 while (state.left > 0) {
961 frag = ip6_frag_next(skb, &state);
968 * Put this fragment into the sending queue.
970 frag->tstamp = tstamp;
971 err = output(net, sk, frag);
975 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
976 IPSTATS_MIB_FRAGCREATES);
978 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
979 IPSTATS_MIB_FRAGOKS);
984 if (skb->sk && dst_allfrag(skb_dst(skb)))
985 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
987 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
991 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
992 IPSTATS_MIB_FRAGFAILS);
997 static inline int ip6_rt_check(const struct rt6key *rt_key,
998 const struct in6_addr *fl_addr,
999 const struct in6_addr *addr_cache)
1001 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1002 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1005 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1006 struct dst_entry *dst,
1007 const struct flowi6 *fl6)
1009 struct ipv6_pinfo *np = inet6_sk(sk);
1010 struct rt6_info *rt;
1015 if (dst->ops->family != AF_INET6) {
1020 rt = (struct rt6_info *)dst;
1021 /* Yes, checking route validity in not connected
1022 * case is not very simple. Take into account,
1023 * that we do not support routing by source, TOS,
1024 * and MSG_DONTROUTE --ANK (980726)
1026 * 1. ip6_rt_check(): If route was host route,
1027 * check that cached destination is current.
1028 * If it is network route, we still may
1029 * check its validity using saved pointer
1030 * to the last used address: daddr_cache.
1031 * We do not want to save whole address now,
1032 * (because main consumer of this service
1033 * is tcp, which has not this problem),
1034 * so that the last trick works only on connected
1036 * 2. oif also should be the same.
1038 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1039 #ifdef CONFIG_IPV6_SUBTREES
1040 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1042 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
1043 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
1052 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1053 struct dst_entry **dst, struct flowi6 *fl6)
1055 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1056 struct neighbour *n;
1057 struct rt6_info *rt;
1062 /* The correct way to handle this would be to do
1063 * ip6_route_get_saddr, and then ip6_route_output; however,
1064 * the route-specific preferred source forces the
1065 * ip6_route_output call _before_ ip6_route_get_saddr.
1067 * In source specific routing (no src=any default route),
1068 * ip6_route_output will fail given src=any saddr, though, so
1069 * that's why we try it again later.
1071 if (ipv6_addr_any(&fl6->saddr)) {
1072 struct fib6_info *from;
1073 struct rt6_info *rt;
1075 *dst = ip6_route_output(net, sk, fl6);
1076 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1079 from = rt ? rcu_dereference(rt->from) : NULL;
1080 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1081 sk ? inet6_sk(sk)->srcprefs : 0,
1086 goto out_err_release;
1088 /* If we had an erroneous initial result, pretend it
1089 * never existed and let the SA-enabled version take
1092 if ((*dst)->error) {
1097 if (fl6->flowi6_oif)
1098 flags |= RT6_LOOKUP_F_IFACE;
1102 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1104 err = (*dst)->error;
1106 goto out_err_release;
1108 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1110 * Here if the dst entry we've looked up
1111 * has a neighbour entry that is in the INCOMPLETE
1112 * state and the src address from the flow is
1113 * marked as OPTIMISTIC, we release the found
1114 * dst entry and replace it instead with the
1115 * dst entry of the nexthop router
1117 rt = (struct rt6_info *) *dst;
1119 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1120 rt6_nexthop(rt, &fl6->daddr));
1121 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1122 rcu_read_unlock_bh();
1125 struct inet6_ifaddr *ifp;
1126 struct flowi6 fl_gw6;
1129 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1132 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1138 * We need to get the dst entry for the
1139 * default router instead
1142 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1143 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1144 *dst = ip6_route_output(net, sk, &fl_gw6);
1145 err = (*dst)->error;
1147 goto out_err_release;
1151 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1152 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1153 err = -EAFNOSUPPORT;
1154 goto out_err_release;
1163 if (err == -ENETUNREACH)
1164 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1169 * ip6_dst_lookup - perform route lookup on flow
1170 * @net: Network namespace to perform lookup in
1171 * @sk: socket which provides route info
1172 * @dst: pointer to dst_entry * for result
1173 * @fl6: flow to lookup
1175 * This function performs a route lookup on the given flow.
1177 * It returns zero on success, or a standard errno code on error.
1179 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1183 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1185 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1188 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1189 * @net: Network namespace to perform lookup in
1190 * @sk: socket which provides route info
1191 * @fl6: flow to lookup
1192 * @final_dst: final destination address for ipsec lookup
1194 * This function performs a route lookup on the given flow.
1196 * It returns a valid dst pointer on success, or a pointer encoded
1199 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1200 const struct in6_addr *final_dst)
1202 struct dst_entry *dst = NULL;
1205 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1207 return ERR_PTR(err);
1209 fl6->daddr = *final_dst;
1211 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1213 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1216 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1217 * @sk: socket which provides the dst cache and route info
1218 * @fl6: flow to lookup
1219 * @final_dst: final destination address for ipsec lookup
1220 * @connected: whether @sk is connected or not
1222 * This function performs a route lookup on the given flow with the
1223 * possibility of using the cached route in the socket if it is valid.
1224 * It will take the socket dst lock when operating on the dst cache.
1225 * As a result, this function can only be used in process context.
1227 * In addition, for a connected socket, cache the dst in the socket
1228 * if the current cache is not valid.
1230 * It returns a valid dst pointer on success, or a pointer encoded
1233 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1234 const struct in6_addr *final_dst,
1237 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1239 dst = ip6_sk_dst_check(sk, dst, fl6);
1243 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1244 if (connected && !IS_ERR(dst))
1245 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1249 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1252 * ip6_dst_lookup_tunnel - perform route lookup on tunnel
1253 * @skb: Packet for which lookup is done
1254 * @dev: Tunnel device
1255 * @net: Network namespace of tunnel device
1256 * @sock: Socket which provides route info
1257 * @saddr: Memory to store the src ip address
1258 * @info: Tunnel information
1259 * @protocol: IP protocol
1260 * @use_cache: Flag to enable cache usage
1261 * This function performs a route lookup on a tunnel
1263 * It returns a valid dst pointer and stores src address to be used in
1264 * tunnel in param saddr on success, else a pointer encoded error code.
1267 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1268 struct net_device *dev,
1270 struct socket *sock,
1271 struct in6_addr *saddr,
1272 const struct ip_tunnel_info *info,
1276 struct dst_entry *dst = NULL;
1277 #ifdef CONFIG_DST_CACHE
1278 struct dst_cache *dst_cache;
1283 #ifdef CONFIG_DST_CACHE
1284 dst_cache = (struct dst_cache *)&info->dst_cache;
1286 dst = dst_cache_get_ip6(dst_cache, saddr);
1291 memset(&fl6, 0, sizeof(fl6));
1292 fl6.flowi6_mark = skb->mark;
1293 fl6.flowi6_proto = protocol;
1294 fl6.daddr = info->key.u.ipv6.dst;
1295 fl6.saddr = info->key.u.ipv6.src;
1296 prio = info->key.tos;
1297 fl6.flowlabel = ip6_make_flowinfo(prio, info->key.label);
1299 dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1302 netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1303 return ERR_PTR(-ENETUNREACH);
1305 if (dst->dev == dev) { /* is this necessary? */
1306 netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1308 return ERR_PTR(-ELOOP);
1310 #ifdef CONFIG_DST_CACHE
1312 dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1317 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1319 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1322 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1325 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1328 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1331 static void ip6_append_data_mtu(unsigned int *mtu,
1333 unsigned int fragheaderlen,
1334 struct sk_buff *skb,
1335 struct rt6_info *rt,
1336 unsigned int orig_mtu)
1338 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1340 /* first fragment, reserve header_len */
1341 *mtu = orig_mtu - rt->dst.header_len;
1345 * this fragment is not first, the headers
1346 * space is regarded as data space.
1350 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1351 + fragheaderlen - sizeof(struct frag_hdr);
1355 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1356 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1357 struct rt6_info *rt, struct flowi6 *fl6)
1359 struct ipv6_pinfo *np = inet6_sk(sk);
1361 struct ipv6_txoptions *opt = ipc6->opt;
1367 if (WARN_ON(v6_cork->opt))
1370 v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1371 if (unlikely(!v6_cork->opt))
1374 v6_cork->opt->tot_len = sizeof(*opt);
1375 v6_cork->opt->opt_flen = opt->opt_flen;
1376 v6_cork->opt->opt_nflen = opt->opt_nflen;
1378 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1380 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1383 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1385 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1388 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1390 if (opt->hopopt && !v6_cork->opt->hopopt)
1393 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1395 if (opt->srcrt && !v6_cork->opt->srcrt)
1398 /* need source address above miyazawa*/
1401 cork->base.dst = &rt->dst;
1402 cork->fl.u.ip6 = *fl6;
1403 v6_cork->hop_limit = ipc6->hlimit;
1404 v6_cork->tclass = ipc6->tclass;
1405 if (rt->dst.flags & DST_XFRM_TUNNEL)
1406 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1407 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1409 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1410 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1411 if (np->frag_size < mtu) {
1413 mtu = np->frag_size;
1415 cork->base.fragsize = mtu;
1416 cork->base.gso_size = ipc6->gso_size;
1417 cork->base.tx_flags = 0;
1418 cork->base.mark = ipc6->sockc.mark;
1419 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1421 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1422 cork->base.flags |= IPCORK_ALLFRAG;
1423 cork->base.length = 0;
1425 cork->base.transmit_time = ipc6->sockc.transmit_time;
1430 static int __ip6_append_data(struct sock *sk,
1432 struct sk_buff_head *queue,
1433 struct inet_cork *cork,
1434 struct inet6_cork *v6_cork,
1435 struct page_frag *pfrag,
1436 int getfrag(void *from, char *to, int offset,
1437 int len, int odd, struct sk_buff *skb),
1438 void *from, int length, int transhdrlen,
1439 unsigned int flags, struct ipcm6_cookie *ipc6)
1441 struct sk_buff *skb, *skb_prev = NULL;
1442 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1443 struct ubuf_info *uarg = NULL;
1445 int dst_exthdrlen = 0;
1451 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1452 struct ipv6_txoptions *opt = v6_cork->opt;
1453 int csummode = CHECKSUM_NONE;
1454 unsigned int maxnonfragsize, headersize;
1455 unsigned int wmem_alloc_delta = 0;
1456 bool paged, extra_uref = false;
1458 skb = skb_peek_tail(queue);
1460 exthdrlen = opt ? opt->opt_flen : 0;
1461 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1464 paged = !!cork->gso_size;
1465 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1468 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1469 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1470 tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1472 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1474 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1475 (opt ? opt->opt_nflen : 0);
1477 headersize = sizeof(struct ipv6hdr) +
1478 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1479 (dst_allfrag(&rt->dst) ?
1480 sizeof(struct frag_hdr) : 0) +
1481 rt->rt6i_nfheader_len;
1483 if (mtu <= fragheaderlen ||
1484 ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1487 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1488 sizeof(struct frag_hdr);
1490 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1491 * the first fragment
1493 if (headersize + transhdrlen > mtu)
1496 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1497 (sk->sk_protocol == IPPROTO_UDP ||
1498 sk->sk_protocol == IPPROTO_RAW)) {
1499 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1500 sizeof(struct ipv6hdr));
1504 if (ip6_sk_ignore_df(sk))
1505 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1507 maxnonfragsize = mtu;
1509 if (cork->length + length > maxnonfragsize - headersize) {
1511 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1512 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1516 /* CHECKSUM_PARTIAL only with no extension headers and when
1517 * we are not going to fragment
1519 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1520 headersize == sizeof(struct ipv6hdr) &&
1521 length <= mtu - headersize &&
1522 (!(flags & MSG_MORE) || cork->gso_size) &&
1523 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1524 csummode = CHECKSUM_PARTIAL;
1526 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1527 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1530 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1531 if (rt->dst.dev->features & NETIF_F_SG &&
1532 csummode == CHECKSUM_PARTIAL) {
1536 skb_zcopy_set(skb, uarg, &extra_uref);
1541 * Let's try using as much space as possible.
1542 * Use MTU if total length of the message fits into the MTU.
1543 * Otherwise, we need to reserve fragment header and
1544 * fragment alignment (= 8-15 octects, in total).
1546 * Note that we may need to "move" the data from the tail
1547 * of the buffer to the new fragment when we split
1550 * FIXME: It may be fragmented into multiple chunks
1551 * at once if non-fragmentable extension headers
1556 cork->length += length;
1560 while (length > 0) {
1561 /* Check if the remaining data fits into current packet. */
1562 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1564 copy = maxfraglen - skb->len;
1568 unsigned int datalen;
1569 unsigned int fraglen;
1570 unsigned int fraggap;
1571 unsigned int alloclen, alloc_extra;
1572 unsigned int pagedlen;
1574 /* There's no room in the current skb */
1576 fraggap = skb->len - maxfraglen;
1579 /* update mtu and maxfraglen if necessary */
1580 if (!skb || !skb_prev)
1581 ip6_append_data_mtu(&mtu, &maxfraglen,
1582 fragheaderlen, skb, rt,
1588 * If remaining data exceeds the mtu,
1589 * we know we need more fragment(s).
1591 datalen = length + fraggap;
1593 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1594 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1595 fraglen = datalen + fragheaderlen;
1598 alloc_extra = hh_len;
1599 alloc_extra += dst_exthdrlen;
1600 alloc_extra += rt->dst.trailer_len;
1602 /* We just reserve space for fragment header.
1603 * Note: this may be overallocation if the message
1604 * (without MSG_MORE) fits into the MTU.
1606 alloc_extra += sizeof(struct frag_hdr);
1608 if ((flags & MSG_MORE) &&
1609 !(rt->dst.dev->features&NETIF_F_SG))
1612 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1613 !(rt->dst.dev->features & NETIF_F_SG)))
1616 alloclen = min_t(int, fraglen, MAX_HEADER);
1617 pagedlen = fraglen - alloclen;
1619 alloclen += alloc_extra;
1621 if (datalen != length + fraggap) {
1623 * this is not the last fragment, the trailer
1624 * space is regarded as data space.
1626 datalen += rt->dst.trailer_len;
1629 fraglen = datalen + fragheaderlen;
1631 copy = datalen - transhdrlen - fraggap - pagedlen;
1637 skb = sock_alloc_send_skb(sk, alloclen,
1638 (flags & MSG_DONTWAIT), &err);
1641 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1643 skb = alloc_skb(alloclen,
1651 * Fill in the control structures
1653 skb->protocol = htons(ETH_P_IPV6);
1654 skb->ip_summed = csummode;
1656 /* reserve for fragmentation and ipsec header */
1657 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1661 * Find where to start putting bytes
1663 data = skb_put(skb, fraglen - pagedlen);
1664 skb_set_network_header(skb, exthdrlen);
1665 data += fragheaderlen;
1666 skb->transport_header = (skb->network_header +
1669 skb->csum = skb_copy_and_csum_bits(
1670 skb_prev, maxfraglen,
1671 data + transhdrlen, fraggap);
1672 skb_prev->csum = csum_sub(skb_prev->csum,
1675 pskb_trim_unique(skb_prev, maxfraglen);
1678 getfrag(from, data + transhdrlen, offset,
1679 copy, fraggap, skb) < 0) {
1686 length -= copy + transhdrlen;
1691 /* Only the initial fragment is time stamped */
1692 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1694 skb_shinfo(skb)->tskey = tskey;
1696 skb_zcopy_set(skb, uarg, &extra_uref);
1698 if ((flags & MSG_CONFIRM) && !skb_prev)
1699 skb_set_dst_pending_confirm(skb, 1);
1702 * Put the packet on the pending queue
1704 if (!skb->destructor) {
1705 skb->destructor = sock_wfree;
1707 wmem_alloc_delta += skb->truesize;
1709 __skb_queue_tail(queue, skb);
1716 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1717 skb_tailroom(skb) >= copy) {
1721 if (getfrag(from, skb_put(skb, copy),
1722 offset, copy, off, skb) < 0) {
1723 __skb_trim(skb, off);
1727 } else if (!uarg || !uarg->zerocopy) {
1728 int i = skb_shinfo(skb)->nr_frags;
1731 if (!sk_page_frag_refill(sk, pfrag))
1734 if (!skb_can_coalesce(skb, i, pfrag->page,
1737 if (i == MAX_SKB_FRAGS)
1740 __skb_fill_page_desc(skb, i, pfrag->page,
1742 skb_shinfo(skb)->nr_frags = ++i;
1743 get_page(pfrag->page);
1745 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1747 page_address(pfrag->page) + pfrag->offset,
1748 offset, copy, skb->len, skb) < 0)
1751 pfrag->offset += copy;
1752 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1754 skb->data_len += copy;
1755 skb->truesize += copy;
1756 wmem_alloc_delta += copy;
1758 err = skb_zerocopy_iter_dgram(skb, from, copy);
1766 if (wmem_alloc_delta)
1767 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1773 net_zcopy_put_abort(uarg, extra_uref);
1774 cork->length -= length;
1775 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1776 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1780 int ip6_append_data(struct sock *sk,
1781 int getfrag(void *from, char *to, int offset, int len,
1782 int odd, struct sk_buff *skb),
1783 void *from, int length, int transhdrlen,
1784 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1785 struct rt6_info *rt, unsigned int flags)
1787 struct inet_sock *inet = inet_sk(sk);
1788 struct ipv6_pinfo *np = inet6_sk(sk);
1792 if (flags&MSG_PROBE)
1794 if (skb_queue_empty(&sk->sk_write_queue)) {
1798 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1803 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1804 length += exthdrlen;
1805 transhdrlen += exthdrlen;
1807 fl6 = &inet->cork.fl.u.ip6;
1811 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1812 &np->cork, sk_page_frag(sk), getfrag,
1813 from, length, transhdrlen, flags, ipc6);
1815 EXPORT_SYMBOL_GPL(ip6_append_data);
1817 static void ip6_cork_release(struct inet_cork_full *cork,
1818 struct inet6_cork *v6_cork)
1821 kfree(v6_cork->opt->dst0opt);
1822 kfree(v6_cork->opt->dst1opt);
1823 kfree(v6_cork->opt->hopopt);
1824 kfree(v6_cork->opt->srcrt);
1825 kfree(v6_cork->opt);
1826 v6_cork->opt = NULL;
1829 if (cork->base.dst) {
1830 dst_release(cork->base.dst);
1831 cork->base.dst = NULL;
1832 cork->base.flags &= ~IPCORK_ALLFRAG;
1834 memset(&cork->fl, 0, sizeof(cork->fl));
1837 struct sk_buff *__ip6_make_skb(struct sock *sk,
1838 struct sk_buff_head *queue,
1839 struct inet_cork_full *cork,
1840 struct inet6_cork *v6_cork)
1842 struct sk_buff *skb, *tmp_skb;
1843 struct sk_buff **tail_skb;
1844 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1845 struct ipv6_pinfo *np = inet6_sk(sk);
1846 struct net *net = sock_net(sk);
1847 struct ipv6hdr *hdr;
1848 struct ipv6_txoptions *opt = v6_cork->opt;
1849 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1850 struct flowi6 *fl6 = &cork->fl.u.ip6;
1851 unsigned char proto = fl6->flowi6_proto;
1853 skb = __skb_dequeue(queue);
1856 tail_skb = &(skb_shinfo(skb)->frag_list);
1858 /* move skb->data to ip header from ext header */
1859 if (skb->data < skb_network_header(skb))
1860 __skb_pull(skb, skb_network_offset(skb));
1861 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1862 __skb_pull(tmp_skb, skb_network_header_len(skb));
1863 *tail_skb = tmp_skb;
1864 tail_skb = &(tmp_skb->next);
1865 skb->len += tmp_skb->len;
1866 skb->data_len += tmp_skb->len;
1867 skb->truesize += tmp_skb->truesize;
1868 tmp_skb->destructor = NULL;
1872 /* Allow local fragmentation. */
1873 skb->ignore_df = ip6_sk_ignore_df(sk);
1875 *final_dst = fl6->daddr;
1876 __skb_pull(skb, skb_network_header_len(skb));
1877 if (opt && opt->opt_flen)
1878 ipv6_push_frag_opts(skb, opt, &proto);
1879 if (opt && opt->opt_nflen)
1880 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1882 skb_push(skb, sizeof(struct ipv6hdr));
1883 skb_reset_network_header(skb);
1884 hdr = ipv6_hdr(skb);
1886 ip6_flow_hdr(hdr, v6_cork->tclass,
1887 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1888 ip6_autoflowlabel(net, np), fl6));
1889 hdr->hop_limit = v6_cork->hop_limit;
1890 hdr->nexthdr = proto;
1891 hdr->saddr = fl6->saddr;
1892 hdr->daddr = *final_dst;
1894 skb->priority = sk->sk_priority;
1895 skb->mark = cork->base.mark;
1897 skb->tstamp = cork->base.transmit_time;
1899 skb_dst_set(skb, dst_clone(&rt->dst));
1900 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1901 if (proto == IPPROTO_ICMPV6) {
1902 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1904 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1905 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1908 ip6_cork_release(cork, v6_cork);
1913 int ip6_send_skb(struct sk_buff *skb)
1915 struct net *net = sock_net(skb->sk);
1916 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1919 err = ip6_local_out(net, skb->sk, skb);
1922 err = net_xmit_errno(err);
1924 IP6_INC_STATS(net, rt->rt6i_idev,
1925 IPSTATS_MIB_OUTDISCARDS);
1931 int ip6_push_pending_frames(struct sock *sk)
1933 struct sk_buff *skb;
1935 skb = ip6_finish_skb(sk);
1939 return ip6_send_skb(skb);
1941 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1943 static void __ip6_flush_pending_frames(struct sock *sk,
1944 struct sk_buff_head *queue,
1945 struct inet_cork_full *cork,
1946 struct inet6_cork *v6_cork)
1948 struct sk_buff *skb;
1950 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1952 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1953 IPSTATS_MIB_OUTDISCARDS);
1957 ip6_cork_release(cork, v6_cork);
1960 void ip6_flush_pending_frames(struct sock *sk)
1962 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1963 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1965 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1967 struct sk_buff *ip6_make_skb(struct sock *sk,
1968 int getfrag(void *from, char *to, int offset,
1969 int len, int odd, struct sk_buff *skb),
1970 void *from, int length, int transhdrlen,
1971 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1972 struct rt6_info *rt, unsigned int flags,
1973 struct inet_cork_full *cork)
1975 struct inet6_cork v6_cork;
1976 struct sk_buff_head queue;
1977 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1980 if (flags & MSG_PROBE)
1983 __skb_queue_head_init(&queue);
1985 cork->base.flags = 0;
1986 cork->base.addr = 0;
1987 cork->base.opt = NULL;
1988 cork->base.dst = NULL;
1990 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
1992 ip6_cork_release(cork, &v6_cork);
1993 return ERR_PTR(err);
1995 if (ipc6->dontfrag < 0)
1996 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1998 err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
1999 ¤t->task_frag, getfrag, from,
2000 length + exthdrlen, transhdrlen + exthdrlen,
2003 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2004 return ERR_PTR(err);
2007 return __ip6_make_skb(sk, &queue, cork, &v6_cork);