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 unsigned int hh_len = LL_RESERVED_SPACE(dev);
64 int delta = hh_len - skb_headroom(skb);
65 const struct in6_addr *nexthop;
66 struct neighbour *neigh;
69 /* Be paranoid, rather than too clever. */
70 if (unlikely(delta > 0) && dev->header_ops) {
71 /* pskb_expand_head() might crash, if skb is shared */
72 if (skb_shared(skb)) {
73 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
77 skb_set_owner_w(nskb, skb->sk);
85 pskb_expand_head(skb, SKB_DATA_ALIGN(delta), 0, GFP_ATOMIC)) {
90 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
95 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
96 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
98 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
99 ((mroute6_is_socket(net, skb) &&
100 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
101 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
102 &ipv6_hdr(skb)->saddr))) {
103 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
105 /* Do not check for IFF_ALLMULTI; multicast routing
106 is not supported in any case.
109 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
110 net, sk, newskb, NULL, newskb->dev,
113 if (ipv6_hdr(skb)->hop_limit == 0) {
114 IP6_INC_STATS(net, idev,
115 IPSTATS_MIB_OUTDISCARDS);
121 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
123 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
124 IPV6_ADDR_SCOPE_NODELOCAL &&
125 !(dev->flags & IFF_LOOPBACK)) {
131 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
132 int res = lwtunnel_xmit(skb);
134 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
139 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
140 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
141 if (unlikely(!neigh))
142 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
143 if (!IS_ERR(neigh)) {
144 sock_confirm_neigh(skb, neigh);
145 ret = neigh_output(neigh, skb, false);
146 rcu_read_unlock_bh();
149 rcu_read_unlock_bh();
151 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
157 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
158 struct sk_buff *skb, unsigned int mtu)
160 struct sk_buff *segs, *nskb;
161 netdev_features_t features;
164 /* Please see corresponding comment in ip_finish_output_gso
165 * describing the cases where GSO segment length exceeds the
168 features = netif_skb_features(skb);
169 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
170 if (IS_ERR_OR_NULL(segs)) {
177 skb_list_walk_safe(segs, segs, nskb) {
180 skb_mark_not_on_list(segs);
181 err = ip6_fragment(net, sk, segs, ip6_finish_output2);
189 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
193 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
194 /* Policy lookup after SNAT yielded a new policy */
195 if (skb_dst(skb)->xfrm) {
196 IPCB(skb)->flags |= IPSKB_REROUTED;
197 return dst_output(net, sk, skb);
201 mtu = ip6_skb_dst_mtu(skb);
202 if (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu))
203 return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
205 if ((skb->len > mtu && !skb_is_gso(skb)) ||
206 dst_allfrag(skb_dst(skb)) ||
207 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
208 return ip6_fragment(net, sk, skb, ip6_finish_output2);
210 return ip6_finish_output2(net, sk, skb);
213 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
217 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
219 case NET_XMIT_SUCCESS:
220 return __ip6_finish_output(net, sk, skb);
222 return __ip6_finish_output(net, sk, skb) ? : ret;
229 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
231 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
232 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
234 skb->protocol = htons(ETH_P_IPV6);
237 if (unlikely(idev->cnf.disable_ipv6)) {
238 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
243 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
244 net, sk, skb, indev, dev,
246 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
248 EXPORT_SYMBOL(ip6_output);
250 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
252 if (!np->autoflowlabel_set)
253 return ip6_default_np_autolabel(net);
255 return np->autoflowlabel;
259 * xmit an sk_buff (used by TCP, SCTP and DCCP)
260 * Note : socket lock is not held for SYNACK packets, but might be modified
261 * by calls to skb_set_owner_w() and ipv6_local_error(),
262 * which are using proper atomic operations or spinlocks.
264 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
265 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
267 struct net *net = sock_net(sk);
268 const struct ipv6_pinfo *np = inet6_sk(sk);
269 struct in6_addr *first_hop = &fl6->daddr;
270 struct dst_entry *dst = skb_dst(skb);
271 unsigned int head_room;
273 u8 proto = fl6->flowi6_proto;
274 int seg_len = skb->len;
278 head_room = sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
280 head_room += opt->opt_nflen + opt->opt_flen;
282 if (unlikely(skb_headroom(skb) < head_room)) {
283 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
285 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
286 IPSTATS_MIB_OUTDISCARDS);
291 skb_set_owner_w(skb2, skb->sk);
297 seg_len += opt->opt_nflen + opt->opt_flen;
300 ipv6_push_frag_opts(skb, opt, &proto);
303 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
307 skb_push(skb, sizeof(struct ipv6hdr));
308 skb_reset_network_header(skb);
312 * Fill in the IPv6 header
315 hlimit = np->hop_limit;
317 hlimit = ip6_dst_hoplimit(dst);
319 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
320 ip6_autoflowlabel(net, np), fl6));
322 hdr->payload_len = htons(seg_len);
323 hdr->nexthdr = proto;
324 hdr->hop_limit = hlimit;
326 hdr->saddr = fl6->saddr;
327 hdr->daddr = *first_hop;
329 skb->protocol = htons(ETH_P_IPV6);
330 skb->priority = priority;
334 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
335 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
336 IPSTATS_MIB_OUT, skb->len);
338 /* if egress device is enslaved to an L3 master device pass the
339 * skb to its handler for processing
341 skb = l3mdev_ip6_out((struct sock *)sk, skb);
345 /* hooks should never assume socket lock is held.
346 * we promote our socket to non const
348 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
349 net, (struct sock *)sk, skb, NULL, dst->dev,
354 /* ipv6_local_error() does not require socket lock,
355 * we promote our socket to non const
357 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
359 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
363 EXPORT_SYMBOL(ip6_xmit);
365 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
367 struct ip6_ra_chain *ra;
368 struct sock *last = NULL;
370 read_lock(&ip6_ra_lock);
371 for (ra = ip6_ra_chain; ra; ra = ra->next) {
372 struct sock *sk = ra->sk;
373 if (sk && ra->sel == sel &&
374 (!sk->sk_bound_dev_if ||
375 sk->sk_bound_dev_if == skb->dev->ifindex)) {
376 struct ipv6_pinfo *np = inet6_sk(sk);
378 if (np && np->rtalert_isolate &&
379 !net_eq(sock_net(sk), dev_net(skb->dev))) {
383 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
385 rawv6_rcv(last, skb2);
392 rawv6_rcv(last, skb);
393 read_unlock(&ip6_ra_lock);
396 read_unlock(&ip6_ra_lock);
400 static int ip6_forward_proxy_check(struct sk_buff *skb)
402 struct ipv6hdr *hdr = ipv6_hdr(skb);
403 u8 nexthdr = hdr->nexthdr;
407 if (ipv6_ext_hdr(nexthdr)) {
408 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
412 offset = sizeof(struct ipv6hdr);
414 if (nexthdr == IPPROTO_ICMPV6) {
415 struct icmp6hdr *icmp6;
417 if (!pskb_may_pull(skb, (skb_network_header(skb) +
418 offset + 1 - skb->data)))
421 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
423 switch (icmp6->icmp6_type) {
424 case NDISC_ROUTER_SOLICITATION:
425 case NDISC_ROUTER_ADVERTISEMENT:
426 case NDISC_NEIGHBOUR_SOLICITATION:
427 case NDISC_NEIGHBOUR_ADVERTISEMENT:
429 /* For reaction involving unicast neighbor discovery
430 * message destined to the proxied address, pass it to
440 * The proxying router can't forward traffic sent to a link-local
441 * address, so signal the sender and discard the packet. This
442 * behavior is clarified by the MIPv6 specification.
444 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
445 dst_link_failure(skb);
452 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
455 struct dst_entry *dst = skb_dst(skb);
457 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
458 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
460 #ifdef CONFIG_NET_SWITCHDEV
461 if (skb->offload_l3_fwd_mark) {
468 return dst_output(net, sk, skb);
471 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
476 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
477 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
483 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
489 int ip6_forward(struct sk_buff *skb)
491 struct inet6_dev *idev = __in6_dev_get_safely(skb->dev);
492 struct dst_entry *dst = skb_dst(skb);
493 struct ipv6hdr *hdr = ipv6_hdr(skb);
494 struct inet6_skb_parm *opt = IP6CB(skb);
495 struct net *net = dev_net(dst->dev);
498 if (net->ipv6.devconf_all->forwarding == 0)
501 if (skb->pkt_type != PACKET_HOST)
504 if (unlikely(skb->sk))
507 if (skb_warn_if_lro(skb))
510 if (!net->ipv6.devconf_all->disable_policy &&
511 !idev->cnf.disable_policy &&
512 !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
513 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
517 skb_forward_csum(skb);
520 * We DO NOT make any processing on
521 * RA packets, pushing them to user level AS IS
522 * without ane WARRANTY that application will be able
523 * to interpret them. The reason is that we
524 * cannot make anything clever here.
526 * We are not end-node, so that if packet contains
527 * AH/ESP, we cannot make anything.
528 * Defragmentation also would be mistake, RA packets
529 * cannot be fragmented, because there is no warranty
530 * that different fragments will go along one path. --ANK
532 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
533 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
538 * check and decrement ttl
540 if (hdr->hop_limit <= 1) {
541 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
542 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
548 /* XXX: idev->cnf.proxy_ndp? */
549 if (net->ipv6.devconf_all->proxy_ndp &&
550 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
551 int proxied = ip6_forward_proxy_check(skb);
554 return ip6_input(skb);
555 } else if (proxied < 0) {
556 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
561 if (!xfrm6_route_forward(skb)) {
562 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
567 /* IPv6 specs say nothing about it, but it is clear that we cannot
568 send redirects to source routed frames.
569 We don't send redirects to frames decapsulated from IPsec.
571 if (IP6CB(skb)->iif == dst->dev->ifindex &&
572 opt->srcrt == 0 && !skb_sec_path(skb)) {
573 struct in6_addr *target = NULL;
574 struct inet_peer *peer;
578 * incoming and outgoing devices are the same
582 rt = (struct rt6_info *) dst;
583 if (rt->rt6i_flags & RTF_GATEWAY)
584 target = &rt->rt6i_gateway;
586 target = &hdr->daddr;
588 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
590 /* Limit redirects both by destination (here)
591 and by source (inside ndisc_send_redirect)
593 if (inet_peer_xrlim_allow(peer, 1*HZ))
594 ndisc_send_redirect(skb, target);
598 int addrtype = ipv6_addr_type(&hdr->saddr);
600 /* This check is security critical. */
601 if (addrtype == IPV6_ADDR_ANY ||
602 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
604 if (addrtype & IPV6_ADDR_LINKLOCAL) {
605 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
606 ICMPV6_NOT_NEIGHBOUR, 0);
611 mtu = ip6_dst_mtu_forward(dst);
612 if (mtu < IPV6_MIN_MTU)
615 if (ip6_pkt_too_big(skb, mtu)) {
616 /* Again, force OUTPUT device used as source address */
618 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
619 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
620 __IP6_INC_STATS(net, ip6_dst_idev(dst),
621 IPSTATS_MIB_FRAGFAILS);
626 if (skb_cow(skb, dst->dev->hard_header_len)) {
627 __IP6_INC_STATS(net, ip6_dst_idev(dst),
628 IPSTATS_MIB_OUTDISCARDS);
634 /* Mangling hops number delayed to point after skb COW */
638 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
639 net, NULL, skb, skb->dev, dst->dev,
643 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
649 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
651 to->pkt_type = from->pkt_type;
652 to->priority = from->priority;
653 to->protocol = from->protocol;
655 skb_dst_set(to, dst_clone(skb_dst(from)));
657 to->mark = from->mark;
659 skb_copy_hash(to, from);
661 #ifdef CONFIG_NET_SCHED
662 to->tc_index = from->tc_index;
665 skb_ext_copy(to, from);
666 skb_copy_secmark(to, from);
669 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
670 u8 nexthdr, __be32 frag_id,
671 struct ip6_fraglist_iter *iter)
673 unsigned int first_len;
677 *prevhdr = NEXTHDR_FRAGMENT;
678 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
682 iter->frag = skb_shinfo(skb)->frag_list;
683 skb_frag_list_init(skb);
687 iter->frag_id = frag_id;
688 iter->nexthdr = nexthdr;
690 __skb_pull(skb, hlen);
691 fh = __skb_push(skb, sizeof(struct frag_hdr));
692 __skb_push(skb, hlen);
693 skb_reset_network_header(skb);
694 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
696 fh->nexthdr = nexthdr;
698 fh->frag_off = htons(IP6_MF);
699 fh->identification = frag_id;
701 first_len = skb_pagelen(skb);
702 skb->data_len = first_len - skb_headlen(skb);
703 skb->len = first_len;
704 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
708 EXPORT_SYMBOL(ip6_fraglist_init);
710 void ip6_fraglist_prepare(struct sk_buff *skb,
711 struct ip6_fraglist_iter *iter)
713 struct sk_buff *frag = iter->frag;
714 unsigned int hlen = iter->hlen;
717 frag->ip_summed = CHECKSUM_NONE;
718 skb_reset_transport_header(frag);
719 fh = __skb_push(frag, sizeof(struct frag_hdr));
720 __skb_push(frag, hlen);
721 skb_reset_network_header(frag);
722 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
723 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
724 fh->nexthdr = iter->nexthdr;
726 fh->frag_off = htons(iter->offset);
728 fh->frag_off |= htons(IP6_MF);
729 fh->identification = iter->frag_id;
730 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
731 ip6_copy_metadata(frag, skb);
733 EXPORT_SYMBOL(ip6_fraglist_prepare);
735 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
736 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
737 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
739 state->prevhdr = prevhdr;
740 state->nexthdr = nexthdr;
741 state->frag_id = frag_id;
746 state->left = skb->len - hlen; /* Space per frame */
747 state->ptr = hlen; /* Where to start from */
749 state->hroom = hdr_room;
750 state->troom = needed_tailroom;
754 EXPORT_SYMBOL(ip6_frag_init);
756 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
758 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
759 struct sk_buff *frag;
764 /* IF: it doesn't fit, use 'mtu' - the data space left */
765 if (len > state->mtu)
767 /* IF: we are not sending up to and including the packet end
768 then align the next start on an eight byte boundary */
769 if (len < state->left)
772 /* Allocate buffer */
773 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
774 state->hroom + state->troom, GFP_ATOMIC);
776 return ERR_PTR(-ENOMEM);
779 * Set up data on packet
782 ip6_copy_metadata(frag, skb);
783 skb_reserve(frag, state->hroom);
784 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
785 skb_reset_network_header(frag);
786 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
787 frag->transport_header = (frag->network_header + state->hlen +
788 sizeof(struct frag_hdr));
791 * Charge the memory for the fragment to any owner
795 skb_set_owner_w(frag, skb->sk);
798 * Copy the packet header into the new buffer.
800 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
802 fragnexthdr_offset = skb_network_header(frag);
803 fragnexthdr_offset += prevhdr - skb_network_header(skb);
804 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
807 * Build fragment header.
809 fh->nexthdr = state->nexthdr;
811 fh->identification = state->frag_id;
814 * Copy a block of the IP datagram.
816 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
820 fh->frag_off = htons(state->offset);
822 fh->frag_off |= htons(IP6_MF);
823 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
826 state->offset += len;
830 EXPORT_SYMBOL(ip6_frag_next);
832 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
833 int (*output)(struct net *, struct sock *, struct sk_buff *))
835 struct sk_buff *frag;
836 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
837 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
838 inet6_sk(skb->sk) : NULL;
839 struct ip6_frag_state state;
840 unsigned int mtu, hlen, nexthdr_offset;
841 ktime_t tstamp = skb->tstamp;
844 u8 *prevhdr, nexthdr = 0;
846 err = ip6_find_1stfragopt(skb, &prevhdr);
851 nexthdr_offset = prevhdr - skb_network_header(skb);
853 mtu = ip6_skb_dst_mtu(skb);
855 /* We must not fragment if the socket is set to force MTU discovery
856 * or if the skb it not generated by a local socket.
858 if (unlikely(!skb->ignore_df && skb->len > mtu))
861 if (IP6CB(skb)->frag_max_size) {
862 if (IP6CB(skb)->frag_max_size > mtu)
865 /* don't send fragments larger than what we received */
866 mtu = IP6CB(skb)->frag_max_size;
867 if (mtu < IPV6_MIN_MTU)
871 if (np && np->frag_size < mtu) {
875 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
877 mtu -= hlen + sizeof(struct frag_hdr);
879 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
880 &ipv6_hdr(skb)->saddr);
882 if (skb->ip_summed == CHECKSUM_PARTIAL &&
883 (err = skb_checksum_help(skb)))
886 prevhdr = skb_network_header(skb) + nexthdr_offset;
887 hroom = LL_RESERVED_SPACE(rt->dst.dev);
888 if (skb_has_frag_list(skb)) {
889 unsigned int first_len = skb_pagelen(skb);
890 struct ip6_fraglist_iter iter;
891 struct sk_buff *frag2;
893 if (first_len - hlen > mtu ||
894 ((first_len - hlen) & 7) ||
896 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
899 skb_walk_frags(skb, frag) {
900 /* Correct geometry. */
901 if (frag->len > mtu ||
902 ((frag->len & 7) && frag->next) ||
903 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
904 goto slow_path_clean;
906 /* Partially cloned skb? */
907 if (skb_shared(frag))
908 goto slow_path_clean;
913 frag->destructor = sock_wfree;
915 skb->truesize -= frag->truesize;
918 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
924 /* Prepare header of the next frame,
925 * before previous one went down. */
927 ip6_fraglist_prepare(skb, &iter);
929 skb->tstamp = tstamp;
930 err = output(net, sk, skb);
932 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
933 IPSTATS_MIB_FRAGCREATES);
935 if (err || !iter.frag)
938 skb = ip6_fraglist_next(&iter);
944 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
945 IPSTATS_MIB_FRAGOKS);
949 kfree_skb_list(iter.frag);
951 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
952 IPSTATS_MIB_FRAGFAILS);
956 skb_walk_frags(skb, frag2) {
960 frag2->destructor = NULL;
961 skb->truesize += frag2->truesize;
967 * Fragment the datagram.
970 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
971 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
975 * Keep copying data until we run out.
978 while (state.left > 0) {
979 frag = ip6_frag_next(skb, &state);
986 * Put this fragment into the sending queue.
988 frag->tstamp = tstamp;
989 err = output(net, sk, frag);
993 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
994 IPSTATS_MIB_FRAGCREATES);
996 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
997 IPSTATS_MIB_FRAGOKS);
1002 if (skb->sk && dst_allfrag(skb_dst(skb)))
1003 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
1005 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1009 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1010 IPSTATS_MIB_FRAGFAILS);
1015 static inline int ip6_rt_check(const struct rt6key *rt_key,
1016 const struct in6_addr *fl_addr,
1017 const struct in6_addr *addr_cache)
1019 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1020 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1023 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1024 struct dst_entry *dst,
1025 const struct flowi6 *fl6)
1027 struct ipv6_pinfo *np = inet6_sk(sk);
1028 struct rt6_info *rt;
1033 if (dst->ops->family != AF_INET6) {
1038 rt = (struct rt6_info *)dst;
1039 /* Yes, checking route validity in not connected
1040 * case is not very simple. Take into account,
1041 * that we do not support routing by source, TOS,
1042 * and MSG_DONTROUTE --ANK (980726)
1044 * 1. ip6_rt_check(): If route was host route,
1045 * check that cached destination is current.
1046 * If it is network route, we still may
1047 * check its validity using saved pointer
1048 * to the last used address: daddr_cache.
1049 * We do not want to save whole address now,
1050 * (because main consumer of this service
1051 * is tcp, which has not this problem),
1052 * so that the last trick works only on connected
1054 * 2. oif also should be the same.
1056 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1057 #ifdef CONFIG_IPV6_SUBTREES
1058 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1060 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
1061 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
1070 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1071 struct dst_entry **dst, struct flowi6 *fl6)
1073 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1074 struct neighbour *n;
1075 struct rt6_info *rt;
1080 /* The correct way to handle this would be to do
1081 * ip6_route_get_saddr, and then ip6_route_output; however,
1082 * the route-specific preferred source forces the
1083 * ip6_route_output call _before_ ip6_route_get_saddr.
1085 * In source specific routing (no src=any default route),
1086 * ip6_route_output will fail given src=any saddr, though, so
1087 * that's why we try it again later.
1089 if (ipv6_addr_any(&fl6->saddr)) {
1090 struct fib6_info *from;
1091 struct rt6_info *rt;
1093 *dst = ip6_route_output(net, sk, fl6);
1094 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1097 from = rt ? rcu_dereference(rt->from) : NULL;
1098 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1099 sk ? inet6_sk(sk)->srcprefs : 0,
1104 goto out_err_release;
1106 /* If we had an erroneous initial result, pretend it
1107 * never existed and let the SA-enabled version take
1110 if ((*dst)->error) {
1115 if (fl6->flowi6_oif)
1116 flags |= RT6_LOOKUP_F_IFACE;
1120 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1122 err = (*dst)->error;
1124 goto out_err_release;
1126 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1128 * Here if the dst entry we've looked up
1129 * has a neighbour entry that is in the INCOMPLETE
1130 * state and the src address from the flow is
1131 * marked as OPTIMISTIC, we release the found
1132 * dst entry and replace it instead with the
1133 * dst entry of the nexthop router
1135 rt = (struct rt6_info *) *dst;
1137 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1138 rt6_nexthop(rt, &fl6->daddr));
1139 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1140 rcu_read_unlock_bh();
1143 struct inet6_ifaddr *ifp;
1144 struct flowi6 fl_gw6;
1147 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1150 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1156 * We need to get the dst entry for the
1157 * default router instead
1160 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1161 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1162 *dst = ip6_route_output(net, sk, &fl_gw6);
1163 err = (*dst)->error;
1165 goto out_err_release;
1169 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1170 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1171 err = -EAFNOSUPPORT;
1172 goto out_err_release;
1181 if (err == -ENETUNREACH)
1182 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1187 * ip6_dst_lookup - perform route lookup on flow
1188 * @net: Network namespace to perform lookup in
1189 * @sk: socket which provides route info
1190 * @dst: pointer to dst_entry * for result
1191 * @fl6: flow to lookup
1193 * This function performs a route lookup on the given flow.
1195 * It returns zero on success, or a standard errno code on error.
1197 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1201 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1203 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1206 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1207 * @net: Network namespace to perform lookup in
1208 * @sk: socket which provides route info
1209 * @fl6: flow to lookup
1210 * @final_dst: final destination address for ipsec lookup
1212 * This function performs a route lookup on the given flow.
1214 * It returns a valid dst pointer on success, or a pointer encoded
1217 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1218 const struct in6_addr *final_dst)
1220 struct dst_entry *dst = NULL;
1223 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1225 return ERR_PTR(err);
1227 fl6->daddr = *final_dst;
1229 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1231 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1234 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1235 * @sk: socket which provides the dst cache and route info
1236 * @fl6: flow to lookup
1237 * @final_dst: final destination address for ipsec lookup
1238 * @connected: whether @sk is connected or not
1240 * This function performs a route lookup on the given flow with the
1241 * possibility of using the cached route in the socket if it is valid.
1242 * It will take the socket dst lock when operating on the dst cache.
1243 * As a result, this function can only be used in process context.
1245 * In addition, for a connected socket, cache the dst in the socket
1246 * if the current cache is not valid.
1248 * It returns a valid dst pointer on success, or a pointer encoded
1251 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1252 const struct in6_addr *final_dst,
1255 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1257 dst = ip6_sk_dst_check(sk, dst, fl6);
1261 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1262 if (connected && !IS_ERR(dst))
1263 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1267 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1270 * ip6_dst_lookup_tunnel - perform route lookup on tunnel
1271 * @skb: Packet for which lookup is done
1272 * @dev: Tunnel device
1273 * @net: Network namespace of tunnel device
1274 * @sock: Socket which provides route info
1275 * @saddr: Memory to store the src ip address
1276 * @info: Tunnel information
1277 * @protocol: IP protocol
1278 * @use_cache: Flag to enable cache usage
1279 * This function performs a route lookup on a tunnel
1281 * It returns a valid dst pointer and stores src address to be used in
1282 * tunnel in param saddr on success, else a pointer encoded error code.
1285 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1286 struct net_device *dev,
1288 struct socket *sock,
1289 struct in6_addr *saddr,
1290 const struct ip_tunnel_info *info,
1294 struct dst_entry *dst = NULL;
1295 #ifdef CONFIG_DST_CACHE
1296 struct dst_cache *dst_cache;
1301 #ifdef CONFIG_DST_CACHE
1302 dst_cache = (struct dst_cache *)&info->dst_cache;
1304 dst = dst_cache_get_ip6(dst_cache, saddr);
1309 memset(&fl6, 0, sizeof(fl6));
1310 fl6.flowi6_mark = skb->mark;
1311 fl6.flowi6_proto = protocol;
1312 fl6.daddr = info->key.u.ipv6.dst;
1313 fl6.saddr = info->key.u.ipv6.src;
1314 prio = info->key.tos;
1315 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(prio),
1318 dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1321 netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1322 return ERR_PTR(-ENETUNREACH);
1324 if (dst->dev == dev) { /* is this necessary? */
1325 netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1327 return ERR_PTR(-ELOOP);
1329 #ifdef CONFIG_DST_CACHE
1331 dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1336 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1338 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1341 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1344 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1347 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1350 static void ip6_append_data_mtu(unsigned int *mtu,
1352 unsigned int fragheaderlen,
1353 struct sk_buff *skb,
1354 struct rt6_info *rt,
1355 unsigned int orig_mtu)
1357 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1359 /* first fragment, reserve header_len */
1360 *mtu = orig_mtu - rt->dst.header_len;
1364 * this fragment is not first, the headers
1365 * space is regarded as data space.
1369 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1370 + fragheaderlen - sizeof(struct frag_hdr);
1374 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1375 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1376 struct rt6_info *rt, struct flowi6 *fl6)
1378 struct ipv6_pinfo *np = inet6_sk(sk);
1380 struct ipv6_txoptions *opt = ipc6->opt;
1386 if (WARN_ON(v6_cork->opt))
1389 v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1390 if (unlikely(!v6_cork->opt))
1393 v6_cork->opt->tot_len = sizeof(*opt);
1394 v6_cork->opt->opt_flen = opt->opt_flen;
1395 v6_cork->opt->opt_nflen = opt->opt_nflen;
1397 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1399 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1402 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1404 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1407 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1409 if (opt->hopopt && !v6_cork->opt->hopopt)
1412 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1414 if (opt->srcrt && !v6_cork->opt->srcrt)
1417 /* need source address above miyazawa*/
1420 cork->base.dst = &rt->dst;
1421 cork->fl.u.ip6 = *fl6;
1422 v6_cork->hop_limit = ipc6->hlimit;
1423 v6_cork->tclass = ipc6->tclass;
1424 if (rt->dst.flags & DST_XFRM_TUNNEL)
1425 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1426 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1428 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1429 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1430 if (np->frag_size < mtu) {
1432 mtu = np->frag_size;
1434 if (mtu < IPV6_MIN_MTU)
1436 cork->base.fragsize = mtu;
1437 cork->base.gso_size = ipc6->gso_size;
1438 cork->base.tx_flags = 0;
1439 cork->base.mark = ipc6->sockc.mark;
1440 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1442 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1443 cork->base.flags |= IPCORK_ALLFRAG;
1444 cork->base.length = 0;
1446 cork->base.transmit_time = ipc6->sockc.transmit_time;
1451 static int __ip6_append_data(struct sock *sk,
1453 struct sk_buff_head *queue,
1454 struct inet_cork *cork,
1455 struct inet6_cork *v6_cork,
1456 struct page_frag *pfrag,
1457 int getfrag(void *from, char *to, int offset,
1458 int len, int odd, struct sk_buff *skb),
1459 void *from, int length, int transhdrlen,
1460 unsigned int flags, struct ipcm6_cookie *ipc6)
1462 struct sk_buff *skb, *skb_prev = NULL;
1463 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1464 struct ubuf_info *uarg = NULL;
1466 int dst_exthdrlen = 0;
1472 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1473 struct ipv6_txoptions *opt = v6_cork->opt;
1474 int csummode = CHECKSUM_NONE;
1475 unsigned int maxnonfragsize, headersize;
1476 unsigned int wmem_alloc_delta = 0;
1477 bool paged, extra_uref = false;
1479 skb = skb_peek_tail(queue);
1481 exthdrlen = opt ? opt->opt_flen : 0;
1482 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1485 paged = !!cork->gso_size;
1486 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1489 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1490 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1491 tskey = sk->sk_tskey++;
1493 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1495 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1496 (opt ? opt->opt_nflen : 0);
1497 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1498 sizeof(struct frag_hdr);
1500 headersize = sizeof(struct ipv6hdr) +
1501 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1502 (dst_allfrag(&rt->dst) ?
1503 sizeof(struct frag_hdr) : 0) +
1504 rt->rt6i_nfheader_len;
1506 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1507 * the first fragment
1509 if (headersize + transhdrlen > mtu)
1512 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1513 (sk->sk_protocol == IPPROTO_UDP ||
1514 sk->sk_protocol == IPPROTO_RAW)) {
1515 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1516 sizeof(struct ipv6hdr));
1520 if (ip6_sk_ignore_df(sk))
1521 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1523 maxnonfragsize = mtu;
1525 if (cork->length + length > maxnonfragsize - headersize) {
1527 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1528 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1532 /* CHECKSUM_PARTIAL only with no extension headers and when
1533 * we are not going to fragment
1535 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1536 headersize == sizeof(struct ipv6hdr) &&
1537 length <= mtu - headersize &&
1538 (!(flags & MSG_MORE) || cork->gso_size) &&
1539 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1540 csummode = CHECKSUM_PARTIAL;
1542 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1543 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1546 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1547 if (rt->dst.dev->features & NETIF_F_SG &&
1548 csummode == CHECKSUM_PARTIAL) {
1552 skb_zcopy_set(skb, uarg, &extra_uref);
1557 * Let's try using as much space as possible.
1558 * Use MTU if total length of the message fits into the MTU.
1559 * Otherwise, we need to reserve fragment header and
1560 * fragment alignment (= 8-15 octects, in total).
1562 * Note that we may need to "move" the data from the tail
1563 * of the buffer to the new fragment when we split
1566 * FIXME: It may be fragmented into multiple chunks
1567 * at once if non-fragmentable extension headers
1572 cork->length += length;
1576 while (length > 0) {
1577 /* Check if the remaining data fits into current packet. */
1578 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1580 copy = maxfraglen - skb->len;
1584 unsigned int datalen;
1585 unsigned int fraglen;
1586 unsigned int fraggap;
1587 unsigned int alloclen, alloc_extra;
1588 unsigned int pagedlen;
1590 /* There's no room in the current skb */
1592 fraggap = skb->len - maxfraglen;
1595 /* update mtu and maxfraglen if necessary */
1596 if (!skb || !skb_prev)
1597 ip6_append_data_mtu(&mtu, &maxfraglen,
1598 fragheaderlen, skb, rt,
1604 * If remaining data exceeds the mtu,
1605 * we know we need more fragment(s).
1607 datalen = length + fraggap;
1609 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1610 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1611 fraglen = datalen + fragheaderlen;
1614 alloc_extra = hh_len;
1615 alloc_extra += dst_exthdrlen;
1616 alloc_extra += rt->dst.trailer_len;
1618 /* We just reserve space for fragment header.
1619 * Note: this may be overallocation if the message
1620 * (without MSG_MORE) fits into the MTU.
1622 alloc_extra += sizeof(struct frag_hdr);
1624 if ((flags & MSG_MORE) &&
1625 !(rt->dst.dev->features&NETIF_F_SG))
1628 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1629 !(rt->dst.dev->features & NETIF_F_SG)))
1632 alloclen = min_t(int, fraglen, MAX_HEADER);
1633 pagedlen = fraglen - alloclen;
1635 alloclen += alloc_extra;
1637 if (datalen != length + fraggap) {
1639 * this is not the last fragment, the trailer
1640 * space is regarded as data space.
1642 datalen += rt->dst.trailer_len;
1645 fraglen = datalen + fragheaderlen;
1647 copy = datalen - transhdrlen - fraggap - pagedlen;
1653 skb = sock_alloc_send_skb(sk, alloclen,
1654 (flags & MSG_DONTWAIT), &err);
1657 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1659 skb = alloc_skb(alloclen,
1667 * Fill in the control structures
1669 skb->protocol = htons(ETH_P_IPV6);
1670 skb->ip_summed = csummode;
1672 /* reserve for fragmentation and ipsec header */
1673 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1677 * Find where to start putting bytes
1679 data = skb_put(skb, fraglen - pagedlen);
1680 skb_set_network_header(skb, exthdrlen);
1681 data += fragheaderlen;
1682 skb->transport_header = (skb->network_header +
1685 skb->csum = skb_copy_and_csum_bits(
1686 skb_prev, maxfraglen,
1687 data + transhdrlen, fraggap);
1688 skb_prev->csum = csum_sub(skb_prev->csum,
1691 pskb_trim_unique(skb_prev, maxfraglen);
1694 getfrag(from, data + transhdrlen, offset,
1695 copy, fraggap, skb) < 0) {
1702 length -= copy + transhdrlen;
1707 /* Only the initial fragment is time stamped */
1708 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1710 skb_shinfo(skb)->tskey = tskey;
1712 skb_zcopy_set(skb, uarg, &extra_uref);
1714 if ((flags & MSG_CONFIRM) && !skb_prev)
1715 skb_set_dst_pending_confirm(skb, 1);
1718 * Put the packet on the pending queue
1720 if (!skb->destructor) {
1721 skb->destructor = sock_wfree;
1723 wmem_alloc_delta += skb->truesize;
1725 __skb_queue_tail(queue, skb);
1732 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1733 skb_tailroom(skb) >= copy) {
1737 if (getfrag(from, skb_put(skb, copy),
1738 offset, copy, off, skb) < 0) {
1739 __skb_trim(skb, off);
1743 } else if (!uarg || !uarg->zerocopy) {
1744 int i = skb_shinfo(skb)->nr_frags;
1747 if (!sk_page_frag_refill(sk, pfrag))
1750 if (!skb_can_coalesce(skb, i, pfrag->page,
1753 if (i == MAX_SKB_FRAGS)
1756 __skb_fill_page_desc(skb, i, pfrag->page,
1758 skb_shinfo(skb)->nr_frags = ++i;
1759 get_page(pfrag->page);
1761 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1763 page_address(pfrag->page) + pfrag->offset,
1764 offset, copy, skb->len, skb) < 0)
1767 pfrag->offset += copy;
1768 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1770 skb->data_len += copy;
1771 skb->truesize += copy;
1772 wmem_alloc_delta += copy;
1774 err = skb_zerocopy_iter_dgram(skb, from, copy);
1782 if (wmem_alloc_delta)
1783 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1789 net_zcopy_put_abort(uarg, extra_uref);
1790 cork->length -= length;
1791 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1792 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1796 int ip6_append_data(struct sock *sk,
1797 int getfrag(void *from, char *to, int offset, int len,
1798 int odd, struct sk_buff *skb),
1799 void *from, int length, int transhdrlen,
1800 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1801 struct rt6_info *rt, unsigned int flags)
1803 struct inet_sock *inet = inet_sk(sk);
1804 struct ipv6_pinfo *np = inet6_sk(sk);
1808 if (flags&MSG_PROBE)
1810 if (skb_queue_empty(&sk->sk_write_queue)) {
1814 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1819 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1820 length += exthdrlen;
1821 transhdrlen += exthdrlen;
1823 fl6 = &inet->cork.fl.u.ip6;
1827 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1828 &np->cork, sk_page_frag(sk), getfrag,
1829 from, length, transhdrlen, flags, ipc6);
1831 EXPORT_SYMBOL_GPL(ip6_append_data);
1833 static void ip6_cork_release(struct inet_cork_full *cork,
1834 struct inet6_cork *v6_cork)
1837 kfree(v6_cork->opt->dst0opt);
1838 kfree(v6_cork->opt->dst1opt);
1839 kfree(v6_cork->opt->hopopt);
1840 kfree(v6_cork->opt->srcrt);
1841 kfree(v6_cork->opt);
1842 v6_cork->opt = NULL;
1845 if (cork->base.dst) {
1846 dst_release(cork->base.dst);
1847 cork->base.dst = NULL;
1848 cork->base.flags &= ~IPCORK_ALLFRAG;
1850 memset(&cork->fl, 0, sizeof(cork->fl));
1853 struct sk_buff *__ip6_make_skb(struct sock *sk,
1854 struct sk_buff_head *queue,
1855 struct inet_cork_full *cork,
1856 struct inet6_cork *v6_cork)
1858 struct sk_buff *skb, *tmp_skb;
1859 struct sk_buff **tail_skb;
1860 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1861 struct ipv6_pinfo *np = inet6_sk(sk);
1862 struct net *net = sock_net(sk);
1863 struct ipv6hdr *hdr;
1864 struct ipv6_txoptions *opt = v6_cork->opt;
1865 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1866 struct flowi6 *fl6 = &cork->fl.u.ip6;
1867 unsigned char proto = fl6->flowi6_proto;
1869 skb = __skb_dequeue(queue);
1872 tail_skb = &(skb_shinfo(skb)->frag_list);
1874 /* move skb->data to ip header from ext header */
1875 if (skb->data < skb_network_header(skb))
1876 __skb_pull(skb, skb_network_offset(skb));
1877 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1878 __skb_pull(tmp_skb, skb_network_header_len(skb));
1879 *tail_skb = tmp_skb;
1880 tail_skb = &(tmp_skb->next);
1881 skb->len += tmp_skb->len;
1882 skb->data_len += tmp_skb->len;
1883 skb->truesize += tmp_skb->truesize;
1884 tmp_skb->destructor = NULL;
1888 /* Allow local fragmentation. */
1889 skb->ignore_df = ip6_sk_ignore_df(sk);
1891 *final_dst = fl6->daddr;
1892 __skb_pull(skb, skb_network_header_len(skb));
1893 if (opt && opt->opt_flen)
1894 ipv6_push_frag_opts(skb, opt, &proto);
1895 if (opt && opt->opt_nflen)
1896 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1898 skb_push(skb, sizeof(struct ipv6hdr));
1899 skb_reset_network_header(skb);
1900 hdr = ipv6_hdr(skb);
1902 ip6_flow_hdr(hdr, v6_cork->tclass,
1903 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1904 ip6_autoflowlabel(net, np), fl6));
1905 hdr->hop_limit = v6_cork->hop_limit;
1906 hdr->nexthdr = proto;
1907 hdr->saddr = fl6->saddr;
1908 hdr->daddr = *final_dst;
1910 skb->priority = sk->sk_priority;
1911 skb->mark = cork->base.mark;
1913 skb->tstamp = cork->base.transmit_time;
1915 skb_dst_set(skb, dst_clone(&rt->dst));
1916 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1917 if (proto == IPPROTO_ICMPV6) {
1918 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1920 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1921 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1924 ip6_cork_release(cork, v6_cork);
1929 int ip6_send_skb(struct sk_buff *skb)
1931 struct net *net = sock_net(skb->sk);
1932 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1935 err = ip6_local_out(net, skb->sk, skb);
1938 err = net_xmit_errno(err);
1940 IP6_INC_STATS(net, rt->rt6i_idev,
1941 IPSTATS_MIB_OUTDISCARDS);
1947 int ip6_push_pending_frames(struct sock *sk)
1949 struct sk_buff *skb;
1951 skb = ip6_finish_skb(sk);
1955 return ip6_send_skb(skb);
1957 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1959 static void __ip6_flush_pending_frames(struct sock *sk,
1960 struct sk_buff_head *queue,
1961 struct inet_cork_full *cork,
1962 struct inet6_cork *v6_cork)
1964 struct sk_buff *skb;
1966 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1968 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1969 IPSTATS_MIB_OUTDISCARDS);
1973 ip6_cork_release(cork, v6_cork);
1976 void ip6_flush_pending_frames(struct sock *sk)
1978 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1979 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1981 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1983 struct sk_buff *ip6_make_skb(struct sock *sk,
1984 int getfrag(void *from, char *to, int offset,
1985 int len, int odd, struct sk_buff *skb),
1986 void *from, int length, int transhdrlen,
1987 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1988 struct rt6_info *rt, unsigned int flags,
1989 struct inet_cork_full *cork)
1991 struct inet6_cork v6_cork;
1992 struct sk_buff_head queue;
1993 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1996 if (flags & MSG_PROBE)
1999 __skb_queue_head_init(&queue);
2001 cork->base.flags = 0;
2002 cork->base.addr = 0;
2003 cork->base.opt = NULL;
2004 cork->base.dst = NULL;
2006 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
2008 ip6_cork_release(cork, &v6_cork);
2009 return ERR_PTR(err);
2011 if (ipc6->dontfrag < 0)
2012 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2014 err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
2015 ¤t->task_frag, getfrag, from,
2016 length + exthdrlen, transhdrlen + exthdrlen,
2019 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2020 return ERR_PTR(err);
2023 return __ip6_make_skb(sk, &queue, cork, &v6_cork);