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);
133 kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
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;
205 kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
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);
220 kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
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) {
443 skb_clear_tstamp(skb);
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 bool mono_delivery_time = skb->mono_delivery_time;
817 struct ip6_frag_state state;
818 unsigned int mtu, hlen, nexthdr_offset;
819 ktime_t tstamp = skb->tstamp;
822 u8 *prevhdr, nexthdr = 0;
824 err = ip6_find_1stfragopt(skb, &prevhdr);
829 nexthdr_offset = prevhdr - skb_network_header(skb);
831 mtu = ip6_skb_dst_mtu(skb);
833 /* We must not fragment if the socket is set to force MTU discovery
834 * or if the skb it not generated by a local socket.
836 if (unlikely(!skb->ignore_df && skb->len > mtu))
839 if (IP6CB(skb)->frag_max_size) {
840 if (IP6CB(skb)->frag_max_size > mtu)
843 /* don't send fragments larger than what we received */
844 mtu = IP6CB(skb)->frag_max_size;
845 if (mtu < IPV6_MIN_MTU)
849 if (np && np->frag_size < mtu) {
853 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
855 mtu -= hlen + sizeof(struct frag_hdr);
857 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
858 &ipv6_hdr(skb)->saddr);
860 if (skb->ip_summed == CHECKSUM_PARTIAL &&
861 (err = skb_checksum_help(skb)))
864 prevhdr = skb_network_header(skb) + nexthdr_offset;
865 hroom = LL_RESERVED_SPACE(rt->dst.dev);
866 if (skb_has_frag_list(skb)) {
867 unsigned int first_len = skb_pagelen(skb);
868 struct ip6_fraglist_iter iter;
869 struct sk_buff *frag2;
871 if (first_len - hlen > mtu ||
872 ((first_len - hlen) & 7) ||
874 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
877 skb_walk_frags(skb, frag) {
878 /* Correct geometry. */
879 if (frag->len > mtu ||
880 ((frag->len & 7) && frag->next) ||
881 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
882 goto slow_path_clean;
884 /* Partially cloned skb? */
885 if (skb_shared(frag))
886 goto slow_path_clean;
891 frag->destructor = sock_wfree;
893 skb->truesize -= frag->truesize;
896 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
902 /* Prepare header of the next frame,
903 * before previous one went down. */
905 ip6_fraglist_prepare(skb, &iter);
907 skb_set_delivery_time(skb, tstamp, mono_delivery_time);
908 err = output(net, sk, skb);
910 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
911 IPSTATS_MIB_FRAGCREATES);
913 if (err || !iter.frag)
916 skb = ip6_fraglist_next(&iter);
922 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
923 IPSTATS_MIB_FRAGOKS);
927 kfree_skb_list(iter.frag);
929 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
930 IPSTATS_MIB_FRAGFAILS);
934 skb_walk_frags(skb, frag2) {
938 frag2->destructor = NULL;
939 skb->truesize += frag2->truesize;
945 * Fragment the datagram.
948 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
949 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
953 * Keep copying data until we run out.
956 while (state.left > 0) {
957 frag = ip6_frag_next(skb, &state);
964 * Put this fragment into the sending queue.
966 skb_set_delivery_time(frag, tstamp, mono_delivery_time);
967 err = output(net, sk, frag);
971 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
972 IPSTATS_MIB_FRAGCREATES);
974 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
975 IPSTATS_MIB_FRAGOKS);
980 if (skb->sk && dst_allfrag(skb_dst(skb)))
981 sk_gso_disable(skb->sk);
983 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
987 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
988 IPSTATS_MIB_FRAGFAILS);
993 static inline int ip6_rt_check(const struct rt6key *rt_key,
994 const struct in6_addr *fl_addr,
995 const struct in6_addr *addr_cache)
997 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
998 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1001 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1002 struct dst_entry *dst,
1003 const struct flowi6 *fl6)
1005 struct ipv6_pinfo *np = inet6_sk(sk);
1006 struct rt6_info *rt;
1011 if (dst->ops->family != AF_INET6) {
1016 rt = (struct rt6_info *)dst;
1017 /* Yes, checking route validity in not connected
1018 * case is not very simple. Take into account,
1019 * that we do not support routing by source, TOS,
1020 * and MSG_DONTROUTE --ANK (980726)
1022 * 1. ip6_rt_check(): If route was host route,
1023 * check that cached destination is current.
1024 * If it is network route, we still may
1025 * check its validity using saved pointer
1026 * to the last used address: daddr_cache.
1027 * We do not want to save whole address now,
1028 * (because main consumer of this service
1029 * is tcp, which has not this problem),
1030 * so that the last trick works only on connected
1032 * 2. oif also should be the same.
1034 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1035 #ifdef CONFIG_IPV6_SUBTREES
1036 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1038 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1047 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1048 struct dst_entry **dst, struct flowi6 *fl6)
1050 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1051 struct neighbour *n;
1052 struct rt6_info *rt;
1057 /* The correct way to handle this would be to do
1058 * ip6_route_get_saddr, and then ip6_route_output; however,
1059 * the route-specific preferred source forces the
1060 * ip6_route_output call _before_ ip6_route_get_saddr.
1062 * In source specific routing (no src=any default route),
1063 * ip6_route_output will fail given src=any saddr, though, so
1064 * that's why we try it again later.
1066 if (ipv6_addr_any(&fl6->saddr)) {
1067 struct fib6_info *from;
1068 struct rt6_info *rt;
1070 *dst = ip6_route_output(net, sk, fl6);
1071 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1074 from = rt ? rcu_dereference(rt->from) : NULL;
1075 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1076 sk ? inet6_sk(sk)->srcprefs : 0,
1081 goto out_err_release;
1083 /* If we had an erroneous initial result, pretend it
1084 * never existed and let the SA-enabled version take
1087 if ((*dst)->error) {
1092 if (fl6->flowi6_oif)
1093 flags |= RT6_LOOKUP_F_IFACE;
1097 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1099 err = (*dst)->error;
1101 goto out_err_release;
1103 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1105 * Here if the dst entry we've looked up
1106 * has a neighbour entry that is in the INCOMPLETE
1107 * state and the src address from the flow is
1108 * marked as OPTIMISTIC, we release the found
1109 * dst entry and replace it instead with the
1110 * dst entry of the nexthop router
1112 rt = (struct rt6_info *) *dst;
1114 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1115 rt6_nexthop(rt, &fl6->daddr));
1116 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1117 rcu_read_unlock_bh();
1120 struct inet6_ifaddr *ifp;
1121 struct flowi6 fl_gw6;
1124 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1127 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1133 * We need to get the dst entry for the
1134 * default router instead
1137 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1138 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1139 *dst = ip6_route_output(net, sk, &fl_gw6);
1140 err = (*dst)->error;
1142 goto out_err_release;
1146 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1147 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1148 err = -EAFNOSUPPORT;
1149 goto out_err_release;
1158 if (err == -ENETUNREACH)
1159 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1164 * ip6_dst_lookup - perform route lookup on flow
1165 * @net: Network namespace to perform lookup in
1166 * @sk: socket which provides route info
1167 * @dst: pointer to dst_entry * for result
1168 * @fl6: flow to lookup
1170 * This function performs a route lookup on the given flow.
1172 * It returns zero on success, or a standard errno code on error.
1174 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1178 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1180 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1183 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1184 * @net: Network namespace to perform lookup in
1185 * @sk: socket which provides route info
1186 * @fl6: flow to lookup
1187 * @final_dst: final destination address for ipsec lookup
1189 * This function performs a route lookup on the given flow.
1191 * It returns a valid dst pointer on success, or a pointer encoded
1194 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1195 const struct in6_addr *final_dst)
1197 struct dst_entry *dst = NULL;
1200 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1202 return ERR_PTR(err);
1204 fl6->daddr = *final_dst;
1206 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1208 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1211 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1212 * @sk: socket which provides the dst cache and route info
1213 * @fl6: flow to lookup
1214 * @final_dst: final destination address for ipsec lookup
1215 * @connected: whether @sk is connected or not
1217 * This function performs a route lookup on the given flow with the
1218 * possibility of using the cached route in the socket if it is valid.
1219 * It will take the socket dst lock when operating on the dst cache.
1220 * As a result, this function can only be used in process context.
1222 * In addition, for a connected socket, cache the dst in the socket
1223 * if the current cache is not valid.
1225 * It returns a valid dst pointer on success, or a pointer encoded
1228 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1229 const struct in6_addr *final_dst,
1232 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1234 dst = ip6_sk_dst_check(sk, dst, fl6);
1238 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1239 if (connected && !IS_ERR(dst))
1240 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1244 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1247 * ip6_dst_lookup_tunnel - perform route lookup on tunnel
1248 * @skb: Packet for which lookup is done
1249 * @dev: Tunnel device
1250 * @net: Network namespace of tunnel device
1251 * @sock: Socket which provides route info
1252 * @saddr: Memory to store the src ip address
1253 * @info: Tunnel information
1254 * @protocol: IP protocol
1255 * @use_cache: Flag to enable cache usage
1256 * This function performs a route lookup on a tunnel
1258 * It returns a valid dst pointer and stores src address to be used in
1259 * tunnel in param saddr on success, else a pointer encoded error code.
1262 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1263 struct net_device *dev,
1265 struct socket *sock,
1266 struct in6_addr *saddr,
1267 const struct ip_tunnel_info *info,
1271 struct dst_entry *dst = NULL;
1272 #ifdef CONFIG_DST_CACHE
1273 struct dst_cache *dst_cache;
1278 #ifdef CONFIG_DST_CACHE
1279 dst_cache = (struct dst_cache *)&info->dst_cache;
1281 dst = dst_cache_get_ip6(dst_cache, saddr);
1286 memset(&fl6, 0, sizeof(fl6));
1287 fl6.flowi6_mark = skb->mark;
1288 fl6.flowi6_proto = protocol;
1289 fl6.daddr = info->key.u.ipv6.dst;
1290 fl6.saddr = info->key.u.ipv6.src;
1291 prio = info->key.tos;
1292 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(prio),
1295 dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1298 netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1299 return ERR_PTR(-ENETUNREACH);
1301 if (dst->dev == dev) { /* is this necessary? */
1302 netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1304 return ERR_PTR(-ELOOP);
1306 #ifdef CONFIG_DST_CACHE
1308 dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1313 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1315 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1318 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1321 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1324 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1327 static void ip6_append_data_mtu(unsigned int *mtu,
1329 unsigned int fragheaderlen,
1330 struct sk_buff *skb,
1331 struct rt6_info *rt,
1332 unsigned int orig_mtu)
1334 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1336 /* first fragment, reserve header_len */
1337 *mtu = orig_mtu - rt->dst.header_len;
1341 * this fragment is not first, the headers
1342 * space is regarded as data space.
1346 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1347 + fragheaderlen - sizeof(struct frag_hdr);
1351 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1352 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1353 struct rt6_info *rt)
1355 struct ipv6_pinfo *np = inet6_sk(sk);
1357 struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1359 /* callers pass dst together with a reference, set it first so
1360 * ip6_cork_release() can put it down even in case of an error.
1362 cork->base.dst = &rt->dst;
1368 if (WARN_ON(v6_cork->opt))
1371 nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1372 if (unlikely(!nopt))
1375 nopt->tot_len = sizeof(*opt);
1376 nopt->opt_flen = opt->opt_flen;
1377 nopt->opt_nflen = opt->opt_nflen;
1379 nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1380 if (opt->dst0opt && !nopt->dst0opt)
1383 nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1384 if (opt->dst1opt && !nopt->dst1opt)
1387 nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1388 if (opt->hopopt && !nopt->hopopt)
1391 nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1392 if (opt->srcrt && !nopt->srcrt)
1395 /* need source address above miyazawa*/
1397 v6_cork->hop_limit = ipc6->hlimit;
1398 v6_cork->tclass = ipc6->tclass;
1399 if (rt->dst.flags & DST_XFRM_TUNNEL)
1400 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1401 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1403 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1404 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1405 if (np->frag_size < mtu) {
1407 mtu = np->frag_size;
1409 cork->base.fragsize = mtu;
1410 cork->base.gso_size = ipc6->gso_size;
1411 cork->base.tx_flags = 0;
1412 cork->base.mark = ipc6->sockc.mark;
1413 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1415 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1416 cork->base.flags |= IPCORK_ALLFRAG;
1417 cork->base.length = 0;
1419 cork->base.transmit_time = ipc6->sockc.transmit_time;
1424 static int __ip6_append_data(struct sock *sk,
1425 struct sk_buff_head *queue,
1426 struct inet_cork_full *cork_full,
1427 struct inet6_cork *v6_cork,
1428 struct page_frag *pfrag,
1429 int getfrag(void *from, char *to, int offset,
1430 int len, int odd, struct sk_buff *skb),
1431 void *from, int length, int transhdrlen,
1432 unsigned int flags, struct ipcm6_cookie *ipc6)
1434 struct sk_buff *skb, *skb_prev = NULL;
1435 struct inet_cork *cork = &cork_full->base;
1436 struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1437 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1438 struct ubuf_info *uarg = NULL;
1440 int dst_exthdrlen = 0;
1446 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1447 struct ipv6_txoptions *opt = v6_cork->opt;
1448 int csummode = CHECKSUM_NONE;
1449 unsigned int maxnonfragsize, headersize;
1450 unsigned int wmem_alloc_delta = 0;
1451 bool paged, extra_uref = false;
1453 skb = skb_peek_tail(queue);
1455 exthdrlen = opt ? opt->opt_flen : 0;
1456 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1459 paged = !!cork->gso_size;
1460 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1463 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1464 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1465 tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1467 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1469 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1470 (opt ? opt->opt_nflen : 0);
1472 headersize = sizeof(struct ipv6hdr) +
1473 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1474 (dst_allfrag(&rt->dst) ?
1475 sizeof(struct frag_hdr) : 0) +
1476 rt->rt6i_nfheader_len;
1478 if (mtu <= fragheaderlen ||
1479 ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1482 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1483 sizeof(struct frag_hdr);
1485 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1486 * the first fragment
1488 if (headersize + transhdrlen > mtu)
1491 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1492 (sk->sk_protocol == IPPROTO_UDP ||
1493 sk->sk_protocol == IPPROTO_ICMPV6 ||
1494 sk->sk_protocol == IPPROTO_RAW)) {
1495 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1496 sizeof(struct ipv6hdr));
1500 if (ip6_sk_ignore_df(sk))
1501 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1503 maxnonfragsize = mtu;
1505 if (cork->length + length > maxnonfragsize - headersize) {
1507 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1508 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1512 /* CHECKSUM_PARTIAL only with no extension headers and when
1513 * we are not going to fragment
1515 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1516 headersize == sizeof(struct ipv6hdr) &&
1517 length <= mtu - headersize &&
1518 (!(flags & MSG_MORE) || cork->gso_size) &&
1519 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1520 csummode = CHECKSUM_PARTIAL;
1522 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1523 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1526 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1527 if (rt->dst.dev->features & NETIF_F_SG &&
1528 csummode == CHECKSUM_PARTIAL) {
1532 skb_zcopy_set(skb, uarg, &extra_uref);
1537 * Let's try using as much space as possible.
1538 * Use MTU if total length of the message fits into the MTU.
1539 * Otherwise, we need to reserve fragment header and
1540 * fragment alignment (= 8-15 octects, in total).
1542 * Note that we may need to "move" the data from the tail
1543 * of the buffer to the new fragment when we split
1546 * FIXME: It may be fragmented into multiple chunks
1547 * at once if non-fragmentable extension headers
1552 cork->length += length;
1556 while (length > 0) {
1557 /* Check if the remaining data fits into current packet. */
1558 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1560 copy = maxfraglen - skb->len;
1564 unsigned int datalen;
1565 unsigned int fraglen;
1566 unsigned int fraggap;
1567 unsigned int alloclen, alloc_extra;
1568 unsigned int pagedlen;
1570 /* There's no room in the current skb */
1572 fraggap = skb->len - maxfraglen;
1575 /* update mtu and maxfraglen if necessary */
1576 if (!skb || !skb_prev)
1577 ip6_append_data_mtu(&mtu, &maxfraglen,
1578 fragheaderlen, skb, rt,
1584 * If remaining data exceeds the mtu,
1585 * we know we need more fragment(s).
1587 datalen = length + fraggap;
1589 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1590 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1591 fraglen = datalen + fragheaderlen;
1594 alloc_extra = hh_len;
1595 alloc_extra += dst_exthdrlen;
1596 alloc_extra += rt->dst.trailer_len;
1598 /* We just reserve space for fragment header.
1599 * Note: this may be overallocation if the message
1600 * (without MSG_MORE) fits into the MTU.
1602 alloc_extra += sizeof(struct frag_hdr);
1604 if ((flags & MSG_MORE) &&
1605 !(rt->dst.dev->features&NETIF_F_SG))
1608 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1609 !(rt->dst.dev->features & NETIF_F_SG)))
1612 alloclen = min_t(int, fraglen, MAX_HEADER);
1613 pagedlen = fraglen - alloclen;
1615 alloclen += alloc_extra;
1617 if (datalen != length + fraggap) {
1619 * this is not the last fragment, the trailer
1620 * space is regarded as data space.
1622 datalen += rt->dst.trailer_len;
1625 fraglen = datalen + fragheaderlen;
1627 copy = datalen - transhdrlen - fraggap - pagedlen;
1633 skb = sock_alloc_send_skb(sk, alloclen,
1634 (flags & MSG_DONTWAIT), &err);
1637 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1639 skb = alloc_skb(alloclen,
1647 * Fill in the control structures
1649 skb->protocol = htons(ETH_P_IPV6);
1650 skb->ip_summed = csummode;
1652 /* reserve for fragmentation and ipsec header */
1653 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1657 * Find where to start putting bytes
1659 data = skb_put(skb, fraglen - pagedlen);
1660 skb_set_network_header(skb, exthdrlen);
1661 data += fragheaderlen;
1662 skb->transport_header = (skb->network_header +
1665 skb->csum = skb_copy_and_csum_bits(
1666 skb_prev, maxfraglen,
1667 data + transhdrlen, fraggap);
1668 skb_prev->csum = csum_sub(skb_prev->csum,
1671 pskb_trim_unique(skb_prev, maxfraglen);
1674 getfrag(from, data + transhdrlen, offset,
1675 copy, fraggap, skb) < 0) {
1682 length -= copy + transhdrlen;
1687 /* Only the initial fragment is time stamped */
1688 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1690 skb_shinfo(skb)->tskey = tskey;
1692 skb_zcopy_set(skb, uarg, &extra_uref);
1694 if ((flags & MSG_CONFIRM) && !skb_prev)
1695 skb_set_dst_pending_confirm(skb, 1);
1698 * Put the packet on the pending queue
1700 if (!skb->destructor) {
1701 skb->destructor = sock_wfree;
1703 wmem_alloc_delta += skb->truesize;
1705 __skb_queue_tail(queue, skb);
1712 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1713 skb_tailroom(skb) >= copy) {
1717 if (getfrag(from, skb_put(skb, copy),
1718 offset, copy, off, skb) < 0) {
1719 __skb_trim(skb, off);
1723 } else if (!uarg || !uarg->zerocopy) {
1724 int i = skb_shinfo(skb)->nr_frags;
1727 if (!sk_page_frag_refill(sk, pfrag))
1730 if (!skb_can_coalesce(skb, i, pfrag->page,
1733 if (i == MAX_SKB_FRAGS)
1736 __skb_fill_page_desc(skb, i, pfrag->page,
1738 skb_shinfo(skb)->nr_frags = ++i;
1739 get_page(pfrag->page);
1741 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1743 page_address(pfrag->page) + pfrag->offset,
1744 offset, copy, skb->len, skb) < 0)
1747 pfrag->offset += copy;
1748 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1750 skb->data_len += copy;
1751 skb->truesize += copy;
1752 wmem_alloc_delta += copy;
1754 err = skb_zerocopy_iter_dgram(skb, from, copy);
1762 if (wmem_alloc_delta)
1763 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1769 net_zcopy_put_abort(uarg, extra_uref);
1770 cork->length -= length;
1771 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1772 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1776 int ip6_append_data(struct sock *sk,
1777 int getfrag(void *from, char *to, int offset, int len,
1778 int odd, struct sk_buff *skb),
1779 void *from, int length, int transhdrlen,
1780 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1781 struct rt6_info *rt, unsigned int flags)
1783 struct inet_sock *inet = inet_sk(sk);
1784 struct ipv6_pinfo *np = inet6_sk(sk);
1788 if (flags&MSG_PROBE)
1790 if (skb_queue_empty(&sk->sk_write_queue)) {
1795 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1800 inet->cork.fl.u.ip6 = *fl6;
1801 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1802 length += exthdrlen;
1803 transhdrlen += exthdrlen;
1808 return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1809 &np->cork, sk_page_frag(sk), getfrag,
1810 from, length, transhdrlen, flags, ipc6);
1812 EXPORT_SYMBOL_GPL(ip6_append_data);
1814 static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1816 struct dst_entry *dst = cork->base.dst;
1818 cork->base.dst = NULL;
1819 cork->base.flags &= ~IPCORK_ALLFRAG;
1820 skb_dst_set(skb, dst);
1823 static void ip6_cork_release(struct inet_cork_full *cork,
1824 struct inet6_cork *v6_cork)
1827 struct ipv6_txoptions *opt = v6_cork->opt;
1829 kfree(opt->dst0opt);
1830 kfree(opt->dst1opt);
1834 v6_cork->opt = NULL;
1837 if (cork->base.dst) {
1838 dst_release(cork->base.dst);
1839 cork->base.dst = NULL;
1840 cork->base.flags &= ~IPCORK_ALLFRAG;
1844 struct sk_buff *__ip6_make_skb(struct sock *sk,
1845 struct sk_buff_head *queue,
1846 struct inet_cork_full *cork,
1847 struct inet6_cork *v6_cork)
1849 struct sk_buff *skb, *tmp_skb;
1850 struct sk_buff **tail_skb;
1851 struct in6_addr *final_dst;
1852 struct ipv6_pinfo *np = inet6_sk(sk);
1853 struct net *net = sock_net(sk);
1854 struct ipv6hdr *hdr;
1855 struct ipv6_txoptions *opt = v6_cork->opt;
1856 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1857 struct flowi6 *fl6 = &cork->fl.u.ip6;
1858 unsigned char proto = fl6->flowi6_proto;
1860 skb = __skb_dequeue(queue);
1863 tail_skb = &(skb_shinfo(skb)->frag_list);
1865 /* move skb->data to ip header from ext header */
1866 if (skb->data < skb_network_header(skb))
1867 __skb_pull(skb, skb_network_offset(skb));
1868 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1869 __skb_pull(tmp_skb, skb_network_header_len(skb));
1870 *tail_skb = tmp_skb;
1871 tail_skb = &(tmp_skb->next);
1872 skb->len += tmp_skb->len;
1873 skb->data_len += tmp_skb->len;
1874 skb->truesize += tmp_skb->truesize;
1875 tmp_skb->destructor = NULL;
1879 /* Allow local fragmentation. */
1880 skb->ignore_df = ip6_sk_ignore_df(sk);
1881 __skb_pull(skb, skb_network_header_len(skb));
1883 final_dst = &fl6->daddr;
1884 if (opt && opt->opt_flen)
1885 ipv6_push_frag_opts(skb, opt, &proto);
1886 if (opt && opt->opt_nflen)
1887 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1889 skb_push(skb, sizeof(struct ipv6hdr));
1890 skb_reset_network_header(skb);
1891 hdr = ipv6_hdr(skb);
1893 ip6_flow_hdr(hdr, v6_cork->tclass,
1894 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1895 ip6_autoflowlabel(net, np), fl6));
1896 hdr->hop_limit = v6_cork->hop_limit;
1897 hdr->nexthdr = proto;
1898 hdr->saddr = fl6->saddr;
1899 hdr->daddr = *final_dst;
1901 skb->priority = sk->sk_priority;
1902 skb->mark = cork->base.mark;
1903 skb->tstamp = cork->base.transmit_time;
1905 ip6_cork_steal_dst(skb, cork);
1906 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1907 if (proto == IPPROTO_ICMPV6) {
1908 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1910 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1911 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1914 ip6_cork_release(cork, v6_cork);
1919 int ip6_send_skb(struct sk_buff *skb)
1921 struct net *net = sock_net(skb->sk);
1922 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1925 err = ip6_local_out(net, skb->sk, skb);
1928 err = net_xmit_errno(err);
1930 IP6_INC_STATS(net, rt->rt6i_idev,
1931 IPSTATS_MIB_OUTDISCARDS);
1937 int ip6_push_pending_frames(struct sock *sk)
1939 struct sk_buff *skb;
1941 skb = ip6_finish_skb(sk);
1945 return ip6_send_skb(skb);
1947 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1949 static void __ip6_flush_pending_frames(struct sock *sk,
1950 struct sk_buff_head *queue,
1951 struct inet_cork_full *cork,
1952 struct inet6_cork *v6_cork)
1954 struct sk_buff *skb;
1956 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1958 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1959 IPSTATS_MIB_OUTDISCARDS);
1963 ip6_cork_release(cork, v6_cork);
1966 void ip6_flush_pending_frames(struct sock *sk)
1968 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1969 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1971 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1973 struct sk_buff *ip6_make_skb(struct sock *sk,
1974 int getfrag(void *from, char *to, int offset,
1975 int len, int odd, struct sk_buff *skb),
1976 void *from, int length, int transhdrlen,
1977 struct ipcm6_cookie *ipc6, struct rt6_info *rt,
1978 unsigned int flags, struct inet_cork_full *cork)
1980 struct inet6_cork v6_cork;
1981 struct sk_buff_head queue;
1982 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1985 if (flags & MSG_PROBE) {
1986 dst_release(&rt->dst);
1990 __skb_queue_head_init(&queue);
1992 cork->base.flags = 0;
1993 cork->base.addr = 0;
1994 cork->base.opt = NULL;
1996 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
1998 ip6_cork_release(cork, &v6_cork);
1999 return ERR_PTR(err);
2001 if (ipc6->dontfrag < 0)
2002 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2004 err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2005 ¤t->task_frag, getfrag, from,
2006 length + exthdrlen, transhdrlen + exthdrlen,
2009 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2010 return ERR_PTR(err);
2013 return __ip6_make_skb(sk, &queue, cork, &v6_cork);