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>
47 #include <net/ndisc.h>
48 #include <net/protocol.h>
49 #include <net/ip6_route.h>
50 #include <net/addrconf.h>
51 #include <net/rawv6.h>
54 #include <net/checksum.h>
55 #include <linux/mroute6.h>
56 #include <net/l3mdev.h>
57 #include <net/lwtunnel.h>
58 #include <net/ip_tunnels.h>
60 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
62 struct dst_entry *dst = skb_dst(skb);
63 struct net_device *dev = dst->dev;
64 struct inet6_dev *idev = ip6_dst_idev(dst);
65 unsigned int hh_len = LL_RESERVED_SPACE(dev);
66 const struct in6_addr *daddr, *nexthop;
68 struct neighbour *neigh;
71 /* Be paranoid, rather than too clever. */
72 if (unlikely(hh_len > skb_headroom(skb)) && dev->header_ops) {
73 skb = skb_expand_head(skb, hh_len);
75 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
82 if (ipv6_addr_is_multicast(daddr)) {
83 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
84 ((mroute6_is_socket(net, skb) &&
85 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
86 ipv6_chk_mcast_addr(dev, daddr, &hdr->saddr))) {
87 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
89 /* Do not check for IFF_ALLMULTI; multicast routing
90 is not supported in any case.
93 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
94 net, sk, newskb, NULL, newskb->dev,
97 if (hdr->hop_limit == 0) {
98 IP6_INC_STATS(net, idev,
99 IPSTATS_MIB_OUTDISCARDS);
105 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
106 if (IPV6_ADDR_MC_SCOPE(daddr) <= IPV6_ADDR_SCOPE_NODELOCAL &&
107 !(dev->flags & IFF_LOOPBACK)) {
113 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
114 int res = lwtunnel_xmit(skb);
116 if (res != LWTUNNEL_XMIT_CONTINUE)
121 nexthop = rt6_nexthop((struct rt6_info *)dst, daddr);
122 neigh = __ipv6_neigh_lookup_noref(dev, nexthop);
124 if (unlikely(IS_ERR_OR_NULL(neigh))) {
125 if (unlikely(!neigh))
126 neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
129 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
130 kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
134 sock_confirm_neigh(skb, neigh);
135 ret = neigh_output(neigh, skb, false);
141 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
142 struct sk_buff *skb, unsigned int mtu)
144 struct sk_buff *segs, *nskb;
145 netdev_features_t features;
148 /* Please see corresponding comment in ip_finish_output_gso
149 * describing the cases where GSO segment length exceeds the
152 features = netif_skb_features(skb);
153 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
154 if (IS_ERR_OR_NULL(segs)) {
161 skb_list_walk_safe(segs, segs, nskb) {
164 skb_mark_not_on_list(segs);
165 err = ip6_fragment(net, sk, segs, ip6_finish_output2);
173 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
177 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
178 /* Policy lookup after SNAT yielded a new policy */
179 if (skb_dst(skb)->xfrm) {
180 IP6CB(skb)->flags |= IP6SKB_REROUTED;
181 return dst_output(net, sk, skb);
185 mtu = ip6_skb_dst_mtu(skb);
186 if (skb_is_gso(skb) &&
187 !(IP6CB(skb)->flags & IP6SKB_FAKEJUMBO) &&
188 !skb_gso_validate_network_len(skb, mtu))
189 return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
191 if ((skb->len > mtu && !skb_is_gso(skb)) ||
192 dst_allfrag(skb_dst(skb)) ||
193 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
194 return ip6_fragment(net, sk, skb, ip6_finish_output2);
196 return ip6_finish_output2(net, sk, skb);
199 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
203 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
205 case NET_XMIT_SUCCESS:
207 return __ip6_finish_output(net, sk, skb) ? : ret;
209 kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
214 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
216 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
217 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
219 skb->protocol = htons(ETH_P_IPV6);
222 if (unlikely(idev->cnf.disable_ipv6)) {
223 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
224 kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
228 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
229 net, sk, skb, indev, dev,
231 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
233 EXPORT_SYMBOL(ip6_output);
235 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
237 if (!np->autoflowlabel_set)
238 return ip6_default_np_autolabel(net);
240 return np->autoflowlabel;
244 * xmit an sk_buff (used by TCP, SCTP and DCCP)
245 * Note : socket lock is not held for SYNACK packets, but might be modified
246 * by calls to skb_set_owner_w() and ipv6_local_error(),
247 * which are using proper atomic operations or spinlocks.
249 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
250 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
252 struct net *net = sock_net(sk);
253 const struct ipv6_pinfo *np = inet6_sk(sk);
254 struct in6_addr *first_hop = &fl6->daddr;
255 struct dst_entry *dst = skb_dst(skb);
256 struct net_device *dev = dst->dev;
257 struct inet6_dev *idev = ip6_dst_idev(dst);
258 struct hop_jumbo_hdr *hop_jumbo;
259 int hoplen = sizeof(*hop_jumbo);
260 unsigned int head_room;
262 u8 proto = fl6->flowi6_proto;
263 int seg_len = skb->len;
267 head_room = sizeof(struct ipv6hdr) + hoplen + LL_RESERVED_SPACE(dev);
269 head_room += opt->opt_nflen + opt->opt_flen;
271 if (unlikely(head_room > skb_headroom(skb))) {
272 skb = skb_expand_head(skb, head_room);
274 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
280 seg_len += opt->opt_nflen + opt->opt_flen;
283 ipv6_push_frag_opts(skb, opt, &proto);
286 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
290 if (unlikely(seg_len > IPV6_MAXPLEN)) {
291 hop_jumbo = skb_push(skb, hoplen);
293 hop_jumbo->nexthdr = proto;
294 hop_jumbo->hdrlen = 0;
295 hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
296 hop_jumbo->tlv_len = 4;
297 hop_jumbo->jumbo_payload_len = htonl(seg_len + hoplen);
299 proto = IPPROTO_HOPOPTS;
301 IP6CB(skb)->flags |= IP6SKB_FAKEJUMBO;
304 skb_push(skb, sizeof(struct ipv6hdr));
305 skb_reset_network_header(skb);
309 * Fill in the IPv6 header
312 hlimit = np->hop_limit;
314 hlimit = ip6_dst_hoplimit(dst);
316 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
317 ip6_autoflowlabel(net, np), fl6));
319 hdr->payload_len = htons(seg_len);
320 hdr->nexthdr = proto;
321 hdr->hop_limit = hlimit;
323 hdr->saddr = fl6->saddr;
324 hdr->daddr = *first_hop;
326 skb->protocol = htons(ETH_P_IPV6);
327 skb->priority = priority;
331 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
332 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
334 /* if egress device is enslaved to an L3 master device pass the
335 * skb to its handler for processing
337 skb = l3mdev_ip6_out((struct sock *)sk, skb);
341 /* hooks should never assume socket lock is held.
342 * we promote our socket to non const
344 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
345 net, (struct sock *)sk, skb, NULL, dev,
350 /* ipv6_local_error() does not require socket lock,
351 * we promote our socket to non const
353 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
355 IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
359 EXPORT_SYMBOL(ip6_xmit);
361 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
363 struct ip6_ra_chain *ra;
364 struct sock *last = NULL;
366 read_lock(&ip6_ra_lock);
367 for (ra = ip6_ra_chain; ra; ra = ra->next) {
368 struct sock *sk = ra->sk;
369 if (sk && ra->sel == sel &&
370 (!sk->sk_bound_dev_if ||
371 sk->sk_bound_dev_if == skb->dev->ifindex)) {
372 struct ipv6_pinfo *np = inet6_sk(sk);
374 if (np && np->rtalert_isolate &&
375 !net_eq(sock_net(sk), dev_net(skb->dev))) {
379 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
381 rawv6_rcv(last, skb2);
388 rawv6_rcv(last, skb);
389 read_unlock(&ip6_ra_lock);
392 read_unlock(&ip6_ra_lock);
396 static int ip6_forward_proxy_check(struct sk_buff *skb)
398 struct ipv6hdr *hdr = ipv6_hdr(skb);
399 u8 nexthdr = hdr->nexthdr;
403 if (ipv6_ext_hdr(nexthdr)) {
404 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
408 offset = sizeof(struct ipv6hdr);
410 if (nexthdr == IPPROTO_ICMPV6) {
411 struct icmp6hdr *icmp6;
413 if (!pskb_may_pull(skb, (skb_network_header(skb) +
414 offset + 1 - skb->data)))
417 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
419 switch (icmp6->icmp6_type) {
420 case NDISC_ROUTER_SOLICITATION:
421 case NDISC_ROUTER_ADVERTISEMENT:
422 case NDISC_NEIGHBOUR_SOLICITATION:
423 case NDISC_NEIGHBOUR_ADVERTISEMENT:
425 /* For reaction involving unicast neighbor discovery
426 * message destined to the proxied address, pass it to
436 * The proxying router can't forward traffic sent to a link-local
437 * address, so signal the sender and discard the packet. This
438 * behavior is clarified by the MIPv6 specification.
440 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
441 dst_link_failure(skb);
448 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
451 struct dst_entry *dst = skb_dst(skb);
453 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
455 #ifdef CONFIG_NET_SWITCHDEV
456 if (skb->offload_l3_fwd_mark) {
462 skb_clear_tstamp(skb);
463 return dst_output(net, sk, skb);
466 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
471 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
472 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
478 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
484 int ip6_forward(struct sk_buff *skb)
486 struct dst_entry *dst = skb_dst(skb);
487 struct ipv6hdr *hdr = ipv6_hdr(skb);
488 struct inet6_skb_parm *opt = IP6CB(skb);
489 struct net *net = dev_net(dst->dev);
490 struct inet6_dev *idev;
494 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
495 if (net->ipv6.devconf_all->forwarding == 0)
498 if (skb->pkt_type != PACKET_HOST)
501 if (unlikely(skb->sk))
504 if (skb_warn_if_lro(skb))
507 if (!net->ipv6.devconf_all->disable_policy &&
508 (!idev || !idev->cnf.disable_policy) &&
509 !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
510 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
514 skb_forward_csum(skb);
517 * We DO NOT make any processing on
518 * RA packets, pushing them to user level AS IS
519 * without ane WARRANTY that application will be able
520 * to interpret them. The reason is that we
521 * cannot make anything clever here.
523 * We are not end-node, so that if packet contains
524 * AH/ESP, we cannot make anything.
525 * Defragmentation also would be mistake, RA packets
526 * cannot be fragmented, because there is no warranty
527 * that different fragments will go along one path. --ANK
529 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
530 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
535 * check and decrement ttl
537 if (hdr->hop_limit <= 1) {
538 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
539 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
541 kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
545 /* XXX: idev->cnf.proxy_ndp? */
546 if (net->ipv6.devconf_all->proxy_ndp &&
547 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
548 int proxied = ip6_forward_proxy_check(skb);
550 /* It's tempting to decrease the hop limit
551 * here by 1, as we do at the end of the
554 * But that would be incorrect, as proxying is
555 * not forwarding. The ip6_input function
556 * will handle this packet locally, and it
557 * depends on the hop limit being unchanged.
559 * One example is the NDP hop limit, that
560 * always has to stay 255, but other would be
561 * similar checks around RA packets, where the
562 * user can even change the desired limit.
564 return ip6_input(skb);
565 } else if (proxied < 0) {
566 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
571 if (!xfrm6_route_forward(skb)) {
572 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
573 SKB_DR_SET(reason, XFRM_POLICY);
578 /* IPv6 specs say nothing about it, but it is clear that we cannot
579 send redirects to source routed frames.
580 We don't send redirects to frames decapsulated from IPsec.
582 if (IP6CB(skb)->iif == dst->dev->ifindex &&
583 opt->srcrt == 0 && !skb_sec_path(skb)) {
584 struct in6_addr *target = NULL;
585 struct inet_peer *peer;
589 * incoming and outgoing devices are the same
593 rt = (struct rt6_info *) dst;
594 if (rt->rt6i_flags & RTF_GATEWAY)
595 target = &rt->rt6i_gateway;
597 target = &hdr->daddr;
599 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
601 /* Limit redirects both by destination (here)
602 and by source (inside ndisc_send_redirect)
604 if (inet_peer_xrlim_allow(peer, 1*HZ))
605 ndisc_send_redirect(skb, target);
609 int addrtype = ipv6_addr_type(&hdr->saddr);
611 /* This check is security critical. */
612 if (addrtype == IPV6_ADDR_ANY ||
613 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
615 if (addrtype & IPV6_ADDR_LINKLOCAL) {
616 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
617 ICMPV6_NOT_NEIGHBOUR, 0);
622 mtu = ip6_dst_mtu_maybe_forward(dst, true);
623 if (mtu < IPV6_MIN_MTU)
626 if (ip6_pkt_too_big(skb, mtu)) {
627 /* Again, force OUTPUT device used as source address */
629 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
630 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
631 __IP6_INC_STATS(net, ip6_dst_idev(dst),
632 IPSTATS_MIB_FRAGFAILS);
633 kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
637 if (skb_cow(skb, dst->dev->hard_header_len)) {
638 __IP6_INC_STATS(net, ip6_dst_idev(dst),
639 IPSTATS_MIB_OUTDISCARDS);
645 /* Mangling hops number delayed to point after skb COW */
649 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
650 net, NULL, skb, skb->dev, dst->dev,
654 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
655 SKB_DR_SET(reason, IP_INADDRERRORS);
657 kfree_skb_reason(skb, reason);
661 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
663 to->pkt_type = from->pkt_type;
664 to->priority = from->priority;
665 to->protocol = from->protocol;
667 skb_dst_set(to, dst_clone(skb_dst(from)));
669 to->mark = from->mark;
671 skb_copy_hash(to, from);
673 #ifdef CONFIG_NET_SCHED
674 to->tc_index = from->tc_index;
677 skb_ext_copy(to, from);
678 skb_copy_secmark(to, from);
681 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
682 u8 nexthdr, __be32 frag_id,
683 struct ip6_fraglist_iter *iter)
685 unsigned int first_len;
689 *prevhdr = NEXTHDR_FRAGMENT;
690 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
694 iter->frag = skb_shinfo(skb)->frag_list;
695 skb_frag_list_init(skb);
699 iter->frag_id = frag_id;
700 iter->nexthdr = nexthdr;
702 __skb_pull(skb, hlen);
703 fh = __skb_push(skb, sizeof(struct frag_hdr));
704 __skb_push(skb, hlen);
705 skb_reset_network_header(skb);
706 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
708 fh->nexthdr = nexthdr;
710 fh->frag_off = htons(IP6_MF);
711 fh->identification = frag_id;
713 first_len = skb_pagelen(skb);
714 skb->data_len = first_len - skb_headlen(skb);
715 skb->len = first_len;
716 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
720 EXPORT_SYMBOL(ip6_fraglist_init);
722 void ip6_fraglist_prepare(struct sk_buff *skb,
723 struct ip6_fraglist_iter *iter)
725 struct sk_buff *frag = iter->frag;
726 unsigned int hlen = iter->hlen;
729 frag->ip_summed = CHECKSUM_NONE;
730 skb_reset_transport_header(frag);
731 fh = __skb_push(frag, sizeof(struct frag_hdr));
732 __skb_push(frag, hlen);
733 skb_reset_network_header(frag);
734 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
735 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
736 fh->nexthdr = iter->nexthdr;
738 fh->frag_off = htons(iter->offset);
740 fh->frag_off |= htons(IP6_MF);
741 fh->identification = iter->frag_id;
742 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
743 ip6_copy_metadata(frag, skb);
745 EXPORT_SYMBOL(ip6_fraglist_prepare);
747 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
748 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
749 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
751 state->prevhdr = prevhdr;
752 state->nexthdr = nexthdr;
753 state->frag_id = frag_id;
758 state->left = skb->len - hlen; /* Space per frame */
759 state->ptr = hlen; /* Where to start from */
761 state->hroom = hdr_room;
762 state->troom = needed_tailroom;
766 EXPORT_SYMBOL(ip6_frag_init);
768 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
770 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
771 struct sk_buff *frag;
776 /* IF: it doesn't fit, use 'mtu' - the data space left */
777 if (len > state->mtu)
779 /* IF: we are not sending up to and including the packet end
780 then align the next start on an eight byte boundary */
781 if (len < state->left)
784 /* Allocate buffer */
785 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
786 state->hroom + state->troom, GFP_ATOMIC);
788 return ERR_PTR(-ENOMEM);
791 * Set up data on packet
794 ip6_copy_metadata(frag, skb);
795 skb_reserve(frag, state->hroom);
796 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
797 skb_reset_network_header(frag);
798 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
799 frag->transport_header = (frag->network_header + state->hlen +
800 sizeof(struct frag_hdr));
803 * Charge the memory for the fragment to any owner
807 skb_set_owner_w(frag, skb->sk);
810 * Copy the packet header into the new buffer.
812 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
814 fragnexthdr_offset = skb_network_header(frag);
815 fragnexthdr_offset += prevhdr - skb_network_header(skb);
816 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
819 * Build fragment header.
821 fh->nexthdr = state->nexthdr;
823 fh->identification = state->frag_id;
826 * Copy a block of the IP datagram.
828 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
832 fh->frag_off = htons(state->offset);
834 fh->frag_off |= htons(IP6_MF);
835 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
838 state->offset += len;
842 EXPORT_SYMBOL(ip6_frag_next);
844 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
845 int (*output)(struct net *, struct sock *, struct sk_buff *))
847 struct sk_buff *frag;
848 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
849 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
850 inet6_sk(skb->sk) : NULL;
851 bool mono_delivery_time = skb->mono_delivery_time;
852 struct ip6_frag_state state;
853 unsigned int mtu, hlen, nexthdr_offset;
854 ktime_t tstamp = skb->tstamp;
857 u8 *prevhdr, nexthdr = 0;
859 err = ip6_find_1stfragopt(skb, &prevhdr);
864 nexthdr_offset = prevhdr - skb_network_header(skb);
866 mtu = ip6_skb_dst_mtu(skb);
868 /* We must not fragment if the socket is set to force MTU discovery
869 * or if the skb it not generated by a local socket.
871 if (unlikely(!skb->ignore_df && skb->len > mtu))
874 if (IP6CB(skb)->frag_max_size) {
875 if (IP6CB(skb)->frag_max_size > mtu)
878 /* don't send fragments larger than what we received */
879 mtu = IP6CB(skb)->frag_max_size;
880 if (mtu < IPV6_MIN_MTU)
884 if (np && np->frag_size < mtu) {
888 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
890 mtu -= hlen + sizeof(struct frag_hdr);
892 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
893 &ipv6_hdr(skb)->saddr);
895 if (skb->ip_summed == CHECKSUM_PARTIAL &&
896 (err = skb_checksum_help(skb)))
899 prevhdr = skb_network_header(skb) + nexthdr_offset;
900 hroom = LL_RESERVED_SPACE(rt->dst.dev);
901 if (skb_has_frag_list(skb)) {
902 unsigned int first_len = skb_pagelen(skb);
903 struct ip6_fraglist_iter iter;
904 struct sk_buff *frag2;
906 if (first_len - hlen > mtu ||
907 ((first_len - hlen) & 7) ||
909 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
912 skb_walk_frags(skb, frag) {
913 /* Correct geometry. */
914 if (frag->len > mtu ||
915 ((frag->len & 7) && frag->next) ||
916 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
917 goto slow_path_clean;
919 /* Partially cloned skb? */
920 if (skb_shared(frag))
921 goto slow_path_clean;
926 frag->destructor = sock_wfree;
928 skb->truesize -= frag->truesize;
931 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
936 /* We prevent @rt from being freed. */
940 /* Prepare header of the next frame,
941 * before previous one went down. */
943 ip6_fraglist_prepare(skb, &iter);
945 skb_set_delivery_time(skb, tstamp, mono_delivery_time);
946 err = output(net, sk, skb);
948 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
949 IPSTATS_MIB_FRAGCREATES);
951 if (err || !iter.frag)
954 skb = ip6_fraglist_next(&iter);
960 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
961 IPSTATS_MIB_FRAGOKS);
966 kfree_skb_list(iter.frag);
968 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
969 IPSTATS_MIB_FRAGFAILS);
974 skb_walk_frags(skb, frag2) {
978 frag2->destructor = NULL;
979 skb->truesize += frag2->truesize;
985 * Fragment the datagram.
988 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
989 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
993 * Keep copying data until we run out.
996 while (state.left > 0) {
997 frag = ip6_frag_next(skb, &state);
1004 * Put this fragment into the sending queue.
1006 skb_set_delivery_time(frag, tstamp, mono_delivery_time);
1007 err = output(net, sk, frag);
1011 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1012 IPSTATS_MIB_FRAGCREATES);
1014 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1015 IPSTATS_MIB_FRAGOKS);
1020 if (skb->sk && dst_allfrag(skb_dst(skb)))
1021 sk_gso_disable(skb->sk);
1023 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1027 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1028 IPSTATS_MIB_FRAGFAILS);
1033 static inline int ip6_rt_check(const struct rt6key *rt_key,
1034 const struct in6_addr *fl_addr,
1035 const struct in6_addr *addr_cache)
1037 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1038 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1041 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1042 struct dst_entry *dst,
1043 const struct flowi6 *fl6)
1045 struct ipv6_pinfo *np = inet6_sk(sk);
1046 struct rt6_info *rt;
1051 if (dst->ops->family != AF_INET6) {
1056 rt = (struct rt6_info *)dst;
1057 /* Yes, checking route validity in not connected
1058 * case is not very simple. Take into account,
1059 * that we do not support routing by source, TOS,
1060 * and MSG_DONTROUTE --ANK (980726)
1062 * 1. ip6_rt_check(): If route was host route,
1063 * check that cached destination is current.
1064 * If it is network route, we still may
1065 * check its validity using saved pointer
1066 * to the last used address: daddr_cache.
1067 * We do not want to save whole address now,
1068 * (because main consumer of this service
1069 * is tcp, which has not this problem),
1070 * so that the last trick works only on connected
1072 * 2. oif also should be the same.
1074 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1075 #ifdef CONFIG_IPV6_SUBTREES
1076 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1078 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1087 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1088 struct dst_entry **dst, struct flowi6 *fl6)
1090 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1091 struct neighbour *n;
1092 struct rt6_info *rt;
1097 /* The correct way to handle this would be to do
1098 * ip6_route_get_saddr, and then ip6_route_output; however,
1099 * the route-specific preferred source forces the
1100 * ip6_route_output call _before_ ip6_route_get_saddr.
1102 * In source specific routing (no src=any default route),
1103 * ip6_route_output will fail given src=any saddr, though, so
1104 * that's why we try it again later.
1106 if (ipv6_addr_any(&fl6->saddr)) {
1107 struct fib6_info *from;
1108 struct rt6_info *rt;
1110 *dst = ip6_route_output(net, sk, fl6);
1111 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1114 from = rt ? rcu_dereference(rt->from) : NULL;
1115 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1116 sk ? inet6_sk(sk)->srcprefs : 0,
1121 goto out_err_release;
1123 /* If we had an erroneous initial result, pretend it
1124 * never existed and let the SA-enabled version take
1127 if ((*dst)->error) {
1132 if (fl6->flowi6_oif)
1133 flags |= RT6_LOOKUP_F_IFACE;
1137 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1139 err = (*dst)->error;
1141 goto out_err_release;
1143 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1145 * Here if the dst entry we've looked up
1146 * has a neighbour entry that is in the INCOMPLETE
1147 * state and the src address from the flow is
1148 * marked as OPTIMISTIC, we release the found
1149 * dst entry and replace it instead with the
1150 * dst entry of the nexthop router
1152 rt = (struct rt6_info *) *dst;
1154 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1155 rt6_nexthop(rt, &fl6->daddr));
1156 err = n && !(READ_ONCE(n->nud_state) & NUD_VALID) ? -EINVAL : 0;
1160 struct inet6_ifaddr *ifp;
1161 struct flowi6 fl_gw6;
1164 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1167 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1173 * We need to get the dst entry for the
1174 * default router instead
1177 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1178 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1179 *dst = ip6_route_output(net, sk, &fl_gw6);
1180 err = (*dst)->error;
1182 goto out_err_release;
1186 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1187 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1188 err = -EAFNOSUPPORT;
1189 goto out_err_release;
1198 if (err == -ENETUNREACH)
1199 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1204 * ip6_dst_lookup - perform route lookup on flow
1205 * @net: Network namespace to perform lookup in
1206 * @sk: socket which provides route info
1207 * @dst: pointer to dst_entry * for result
1208 * @fl6: flow to lookup
1210 * This function performs a route lookup on the given flow.
1212 * It returns zero on success, or a standard errno code on error.
1214 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1218 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1220 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1223 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1224 * @net: Network namespace to perform lookup in
1225 * @sk: socket which provides route info
1226 * @fl6: flow to lookup
1227 * @final_dst: final destination address for ipsec lookup
1229 * This function performs a route lookup on the given flow.
1231 * It returns a valid dst pointer on success, or a pointer encoded
1234 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1235 const struct in6_addr *final_dst)
1237 struct dst_entry *dst = NULL;
1240 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1242 return ERR_PTR(err);
1244 fl6->daddr = *final_dst;
1246 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1248 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1251 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1252 * @sk: socket which provides the dst cache and route info
1253 * @fl6: flow to lookup
1254 * @final_dst: final destination address for ipsec lookup
1255 * @connected: whether @sk is connected or not
1257 * This function performs a route lookup on the given flow with the
1258 * possibility of using the cached route in the socket if it is valid.
1259 * It will take the socket dst lock when operating on the dst cache.
1260 * As a result, this function can only be used in process context.
1262 * In addition, for a connected socket, cache the dst in the socket
1263 * if the current cache is not valid.
1265 * It returns a valid dst pointer on success, or a pointer encoded
1268 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1269 const struct in6_addr *final_dst,
1272 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1274 dst = ip6_sk_dst_check(sk, dst, fl6);
1278 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1279 if (connected && !IS_ERR(dst))
1280 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1284 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1287 * ip6_dst_lookup_tunnel - perform route lookup on tunnel
1288 * @skb: Packet for which lookup is done
1289 * @dev: Tunnel device
1290 * @net: Network namespace of tunnel device
1291 * @sock: Socket which provides route info
1292 * @saddr: Memory to store the src ip address
1293 * @info: Tunnel information
1294 * @protocol: IP protocol
1295 * @use_cache: Flag to enable cache usage
1296 * This function performs a route lookup on a tunnel
1298 * It returns a valid dst pointer and stores src address to be used in
1299 * tunnel in param saddr on success, else a pointer encoded error code.
1302 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1303 struct net_device *dev,
1305 struct socket *sock,
1306 struct in6_addr *saddr,
1307 const struct ip_tunnel_info *info,
1311 struct dst_entry *dst = NULL;
1312 #ifdef CONFIG_DST_CACHE
1313 struct dst_cache *dst_cache;
1318 #ifdef CONFIG_DST_CACHE
1319 dst_cache = (struct dst_cache *)&info->dst_cache;
1321 dst = dst_cache_get_ip6(dst_cache, saddr);
1326 memset(&fl6, 0, sizeof(fl6));
1327 fl6.flowi6_mark = skb->mark;
1328 fl6.flowi6_proto = protocol;
1329 fl6.daddr = info->key.u.ipv6.dst;
1330 fl6.saddr = info->key.u.ipv6.src;
1331 prio = info->key.tos;
1332 fl6.flowlabel = ip6_make_flowinfo(prio, info->key.label);
1334 dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1337 netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1338 return ERR_PTR(-ENETUNREACH);
1340 if (dst->dev == dev) { /* is this necessary? */
1341 netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1343 return ERR_PTR(-ELOOP);
1345 #ifdef CONFIG_DST_CACHE
1347 dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1352 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1354 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1357 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1360 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1363 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1366 static void ip6_append_data_mtu(unsigned int *mtu,
1368 unsigned int fragheaderlen,
1369 struct sk_buff *skb,
1370 struct rt6_info *rt,
1371 unsigned int orig_mtu)
1373 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1375 /* first fragment, reserve header_len */
1376 *mtu = orig_mtu - rt->dst.header_len;
1380 * this fragment is not first, the headers
1381 * space is regarded as data space.
1385 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1386 + fragheaderlen - sizeof(struct frag_hdr);
1390 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1391 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1392 struct rt6_info *rt)
1394 struct ipv6_pinfo *np = inet6_sk(sk);
1396 struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1398 /* callers pass dst together with a reference, set it first so
1399 * ip6_cork_release() can put it down even in case of an error.
1401 cork->base.dst = &rt->dst;
1407 if (WARN_ON(v6_cork->opt))
1410 nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1411 if (unlikely(!nopt))
1414 nopt->tot_len = sizeof(*opt);
1415 nopt->opt_flen = opt->opt_flen;
1416 nopt->opt_nflen = opt->opt_nflen;
1418 nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1419 if (opt->dst0opt && !nopt->dst0opt)
1422 nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1423 if (opt->dst1opt && !nopt->dst1opt)
1426 nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1427 if (opt->hopopt && !nopt->hopopt)
1430 nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1431 if (opt->srcrt && !nopt->srcrt)
1434 /* need source address above miyazawa*/
1436 v6_cork->hop_limit = ipc6->hlimit;
1437 v6_cork->tclass = ipc6->tclass;
1438 if (rt->dst.flags & DST_XFRM_TUNNEL)
1439 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1440 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1442 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1443 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1444 if (np->frag_size < mtu) {
1446 mtu = np->frag_size;
1448 cork->base.fragsize = mtu;
1449 cork->base.gso_size = ipc6->gso_size;
1450 cork->base.tx_flags = 0;
1451 cork->base.mark = ipc6->sockc.mark;
1452 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1454 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1455 cork->base.flags |= IPCORK_ALLFRAG;
1456 cork->base.length = 0;
1458 cork->base.transmit_time = ipc6->sockc.transmit_time;
1463 static int __ip6_append_data(struct sock *sk,
1464 struct sk_buff_head *queue,
1465 struct inet_cork_full *cork_full,
1466 struct inet6_cork *v6_cork,
1467 struct page_frag *pfrag,
1468 int getfrag(void *from, char *to, int offset,
1469 int len, int odd, struct sk_buff *skb),
1470 void *from, size_t length, int transhdrlen,
1471 unsigned int flags, struct ipcm6_cookie *ipc6)
1473 struct sk_buff *skb, *skb_prev = NULL;
1474 struct inet_cork *cork = &cork_full->base;
1475 struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1476 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1477 struct ubuf_info *uarg = NULL;
1479 int dst_exthdrlen = 0;
1486 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1487 struct ipv6_txoptions *opt = v6_cork->opt;
1488 int csummode = CHECKSUM_NONE;
1489 unsigned int maxnonfragsize, headersize;
1490 unsigned int wmem_alloc_delta = 0;
1491 bool paged, extra_uref = false;
1493 skb = skb_peek_tail(queue);
1495 exthdrlen = opt ? opt->opt_flen : 0;
1496 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1499 paged = !!cork->gso_size;
1500 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1503 if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
1504 READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID)
1505 tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1507 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1509 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1510 (opt ? opt->opt_nflen : 0);
1512 headersize = sizeof(struct ipv6hdr) +
1513 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1514 (dst_allfrag(&rt->dst) ?
1515 sizeof(struct frag_hdr) : 0) +
1516 rt->rt6i_nfheader_len;
1518 if (mtu <= fragheaderlen ||
1519 ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1522 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1523 sizeof(struct frag_hdr);
1525 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1526 * the first fragment
1528 if (headersize + transhdrlen > mtu)
1531 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1532 (sk->sk_protocol == IPPROTO_UDP ||
1533 sk->sk_protocol == IPPROTO_ICMPV6 ||
1534 sk->sk_protocol == IPPROTO_RAW)) {
1535 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1536 sizeof(struct ipv6hdr));
1540 if (ip6_sk_ignore_df(sk))
1541 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1543 maxnonfragsize = mtu;
1545 if (cork->length + length > maxnonfragsize - headersize) {
1547 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1548 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1552 /* CHECKSUM_PARTIAL only with no extension headers and when
1553 * we are not going to fragment
1555 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1556 headersize == sizeof(struct ipv6hdr) &&
1557 length <= mtu - headersize &&
1558 (!(flags & MSG_MORE) || cork->gso_size) &&
1559 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1560 csummode = CHECKSUM_PARTIAL;
1562 if ((flags & MSG_ZEROCOPY) && length) {
1563 struct msghdr *msg = from;
1565 if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1566 if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1569 /* Leave uarg NULL if can't zerocopy, callers should
1570 * be able to handle it.
1572 if ((rt->dst.dev->features & NETIF_F_SG) &&
1573 csummode == CHECKSUM_PARTIAL) {
1576 uarg = msg->msg_ubuf;
1578 } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1579 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1582 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1583 if (rt->dst.dev->features & NETIF_F_SG &&
1584 csummode == CHECKSUM_PARTIAL) {
1588 uarg_to_msgzc(uarg)->zerocopy = 0;
1589 skb_zcopy_set(skb, uarg, &extra_uref);
1592 } else if ((flags & MSG_SPLICE_PAGES) && length) {
1593 if (inet_test_bit(HDRINCL, sk))
1595 if (rt->dst.dev->features & NETIF_F_SG &&
1596 getfrag == ip_generic_getfrag)
1597 /* We need an empty buffer to attach stuff to */
1600 flags &= ~MSG_SPLICE_PAGES;
1604 * Let's try using as much space as possible.
1605 * Use MTU if total length of the message fits into the MTU.
1606 * Otherwise, we need to reserve fragment header and
1607 * fragment alignment (= 8-15 octects, in total).
1609 * Note that we may need to "move" the data from the tail
1610 * of the buffer to the new fragment when we split
1613 * FIXME: It may be fragmented into multiple chunks
1614 * at once if non-fragmentable extension headers
1619 cork->length += length;
1623 while (length > 0) {
1624 /* Check if the remaining data fits into current packet. */
1625 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1627 copy = maxfraglen - skb->len;
1631 unsigned int datalen;
1632 unsigned int fraglen;
1633 unsigned int fraggap;
1634 unsigned int alloclen, alloc_extra;
1635 unsigned int pagedlen;
1637 /* There's no room in the current skb */
1639 fraggap = skb->len - maxfraglen;
1642 /* update mtu and maxfraglen if necessary */
1643 if (!skb || !skb_prev)
1644 ip6_append_data_mtu(&mtu, &maxfraglen,
1645 fragheaderlen, skb, rt,
1651 * If remaining data exceeds the mtu,
1652 * we know we need more fragment(s).
1654 datalen = length + fraggap;
1656 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1657 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1658 fraglen = datalen + fragheaderlen;
1661 alloc_extra = hh_len;
1662 alloc_extra += dst_exthdrlen;
1663 alloc_extra += rt->dst.trailer_len;
1665 /* We just reserve space for fragment header.
1666 * Note: this may be overallocation if the message
1667 * (without MSG_MORE) fits into the MTU.
1669 alloc_extra += sizeof(struct frag_hdr);
1671 if ((flags & MSG_MORE) &&
1672 !(rt->dst.dev->features&NETIF_F_SG))
1675 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1676 !(rt->dst.dev->features & NETIF_F_SG)))
1679 alloclen = fragheaderlen + transhdrlen;
1680 pagedlen = datalen - transhdrlen;
1682 alloclen += alloc_extra;
1684 if (datalen != length + fraggap) {
1686 * this is not the last fragment, the trailer
1687 * space is regarded as data space.
1689 datalen += rt->dst.trailer_len;
1692 fraglen = datalen + fragheaderlen;
1694 copy = datalen - transhdrlen - fraggap - pagedlen;
1695 /* [!] NOTE: copy may be negative if pagedlen>0
1696 * because then the equation may reduces to -fraggap.
1698 if (copy < 0 && !(flags & MSG_SPLICE_PAGES)) {
1703 skb = sock_alloc_send_skb(sk, alloclen,
1704 (flags & MSG_DONTWAIT), &err);
1707 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1709 skb = alloc_skb(alloclen,
1717 * Fill in the control structures
1719 skb->protocol = htons(ETH_P_IPV6);
1720 skb->ip_summed = csummode;
1722 /* reserve for fragmentation and ipsec header */
1723 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1727 * Find where to start putting bytes
1729 data = skb_put(skb, fraglen - pagedlen);
1730 skb_set_network_header(skb, exthdrlen);
1731 data += fragheaderlen;
1732 skb->transport_header = (skb->network_header +
1735 skb->csum = skb_copy_and_csum_bits(
1736 skb_prev, maxfraglen,
1737 data + transhdrlen, fraggap);
1738 skb_prev->csum = csum_sub(skb_prev->csum,
1741 pskb_trim_unique(skb_prev, maxfraglen);
1744 getfrag(from, data + transhdrlen, offset,
1745 copy, fraggap, skb) < 0) {
1749 } else if (flags & MSG_SPLICE_PAGES) {
1754 length -= copy + transhdrlen;
1759 /* Only the initial fragment is time stamped */
1760 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1762 skb_shinfo(skb)->tskey = tskey;
1764 skb_zcopy_set(skb, uarg, &extra_uref);
1766 if ((flags & MSG_CONFIRM) && !skb_prev)
1767 skb_set_dst_pending_confirm(skb, 1);
1770 * Put the packet on the pending queue
1772 if (!skb->destructor) {
1773 skb->destructor = sock_wfree;
1775 wmem_alloc_delta += skb->truesize;
1777 __skb_queue_tail(queue, skb);
1784 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1785 skb_tailroom(skb) >= copy) {
1789 if (getfrag(from, skb_put(skb, copy),
1790 offset, copy, off, skb) < 0) {
1791 __skb_trim(skb, off);
1795 } else if (flags & MSG_SPLICE_PAGES) {
1796 struct msghdr *msg = from;
1799 if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1802 err = skb_splice_from_iter(skb, &msg->msg_iter, copy,
1807 wmem_alloc_delta += copy;
1809 int i = skb_shinfo(skb)->nr_frags;
1812 if (!sk_page_frag_refill(sk, pfrag))
1815 skb_zcopy_downgrade_managed(skb);
1816 if (!skb_can_coalesce(skb, i, pfrag->page,
1819 if (i == MAX_SKB_FRAGS)
1822 __skb_fill_page_desc(skb, i, pfrag->page,
1824 skb_shinfo(skb)->nr_frags = ++i;
1825 get_page(pfrag->page);
1827 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1829 page_address(pfrag->page) + pfrag->offset,
1830 offset, copy, skb->len, skb) < 0)
1833 pfrag->offset += copy;
1834 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1836 skb->data_len += copy;
1837 skb->truesize += copy;
1838 wmem_alloc_delta += copy;
1840 err = skb_zerocopy_iter_dgram(skb, from, copy);
1848 if (wmem_alloc_delta)
1849 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1855 net_zcopy_put_abort(uarg, extra_uref);
1856 cork->length -= length;
1857 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1858 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1862 int ip6_append_data(struct sock *sk,
1863 int getfrag(void *from, char *to, int offset, int len,
1864 int odd, struct sk_buff *skb),
1865 void *from, size_t length, int transhdrlen,
1866 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1867 struct rt6_info *rt, unsigned int flags)
1869 struct inet_sock *inet = inet_sk(sk);
1870 struct ipv6_pinfo *np = inet6_sk(sk);
1874 if (flags&MSG_PROBE)
1876 if (skb_queue_empty(&sk->sk_write_queue)) {
1881 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1886 inet->cork.fl.u.ip6 = *fl6;
1887 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1888 length += exthdrlen;
1889 transhdrlen += exthdrlen;
1894 return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1895 &np->cork, sk_page_frag(sk), getfrag,
1896 from, length, transhdrlen, flags, ipc6);
1898 EXPORT_SYMBOL_GPL(ip6_append_data);
1900 static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1902 struct dst_entry *dst = cork->base.dst;
1904 cork->base.dst = NULL;
1905 cork->base.flags &= ~IPCORK_ALLFRAG;
1906 skb_dst_set(skb, dst);
1909 static void ip6_cork_release(struct inet_cork_full *cork,
1910 struct inet6_cork *v6_cork)
1913 struct ipv6_txoptions *opt = v6_cork->opt;
1915 kfree(opt->dst0opt);
1916 kfree(opt->dst1opt);
1920 v6_cork->opt = NULL;
1923 if (cork->base.dst) {
1924 dst_release(cork->base.dst);
1925 cork->base.dst = NULL;
1926 cork->base.flags &= ~IPCORK_ALLFRAG;
1930 struct sk_buff *__ip6_make_skb(struct sock *sk,
1931 struct sk_buff_head *queue,
1932 struct inet_cork_full *cork,
1933 struct inet6_cork *v6_cork)
1935 struct sk_buff *skb, *tmp_skb;
1936 struct sk_buff **tail_skb;
1937 struct in6_addr *final_dst;
1938 struct ipv6_pinfo *np = inet6_sk(sk);
1939 struct net *net = sock_net(sk);
1940 struct ipv6hdr *hdr;
1941 struct ipv6_txoptions *opt = v6_cork->opt;
1942 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1943 struct flowi6 *fl6 = &cork->fl.u.ip6;
1944 unsigned char proto = fl6->flowi6_proto;
1946 skb = __skb_dequeue(queue);
1949 tail_skb = &(skb_shinfo(skb)->frag_list);
1951 /* move skb->data to ip header from ext header */
1952 if (skb->data < skb_network_header(skb))
1953 __skb_pull(skb, skb_network_offset(skb));
1954 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1955 __skb_pull(tmp_skb, skb_network_header_len(skb));
1956 *tail_skb = tmp_skb;
1957 tail_skb = &(tmp_skb->next);
1958 skb->len += tmp_skb->len;
1959 skb->data_len += tmp_skb->len;
1960 skb->truesize += tmp_skb->truesize;
1961 tmp_skb->destructor = NULL;
1965 /* Allow local fragmentation. */
1966 skb->ignore_df = ip6_sk_ignore_df(sk);
1967 __skb_pull(skb, skb_network_header_len(skb));
1969 final_dst = &fl6->daddr;
1970 if (opt && opt->opt_flen)
1971 ipv6_push_frag_opts(skb, opt, &proto);
1972 if (opt && opt->opt_nflen)
1973 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1975 skb_push(skb, sizeof(struct ipv6hdr));
1976 skb_reset_network_header(skb);
1977 hdr = ipv6_hdr(skb);
1979 ip6_flow_hdr(hdr, v6_cork->tclass,
1980 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1981 ip6_autoflowlabel(net, np), fl6));
1982 hdr->hop_limit = v6_cork->hop_limit;
1983 hdr->nexthdr = proto;
1984 hdr->saddr = fl6->saddr;
1985 hdr->daddr = *final_dst;
1987 skb->priority = sk->sk_priority;
1988 skb->mark = cork->base.mark;
1989 skb->tstamp = cork->base.transmit_time;
1991 ip6_cork_steal_dst(skb, cork);
1992 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1993 if (proto == IPPROTO_ICMPV6) {
1994 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1997 if (sk->sk_socket->type == SOCK_RAW &&
1998 !inet_test_bit(HDRINCL, sk))
1999 icmp6_type = fl6->fl6_icmp_type;
2001 icmp6_type = icmp6_hdr(skb)->icmp6_type;
2002 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_type);
2003 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
2006 ip6_cork_release(cork, v6_cork);
2011 int ip6_send_skb(struct sk_buff *skb)
2013 struct net *net = sock_net(skb->sk);
2014 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
2017 err = ip6_local_out(net, skb->sk, skb);
2020 err = net_xmit_errno(err);
2022 IP6_INC_STATS(net, rt->rt6i_idev,
2023 IPSTATS_MIB_OUTDISCARDS);
2029 int ip6_push_pending_frames(struct sock *sk)
2031 struct sk_buff *skb;
2033 skb = ip6_finish_skb(sk);
2037 return ip6_send_skb(skb);
2039 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
2041 static void __ip6_flush_pending_frames(struct sock *sk,
2042 struct sk_buff_head *queue,
2043 struct inet_cork_full *cork,
2044 struct inet6_cork *v6_cork)
2046 struct sk_buff *skb;
2048 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
2050 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
2051 IPSTATS_MIB_OUTDISCARDS);
2055 ip6_cork_release(cork, v6_cork);
2058 void ip6_flush_pending_frames(struct sock *sk)
2060 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
2061 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
2063 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
2065 struct sk_buff *ip6_make_skb(struct sock *sk,
2066 int getfrag(void *from, char *to, int offset,
2067 int len, int odd, struct sk_buff *skb),
2068 void *from, size_t length, int transhdrlen,
2069 struct ipcm6_cookie *ipc6, struct rt6_info *rt,
2070 unsigned int flags, struct inet_cork_full *cork)
2072 struct inet6_cork v6_cork;
2073 struct sk_buff_head queue;
2074 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
2077 if (flags & MSG_PROBE) {
2078 dst_release(&rt->dst);
2082 __skb_queue_head_init(&queue);
2084 cork->base.flags = 0;
2085 cork->base.addr = 0;
2086 cork->base.opt = NULL;
2088 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2090 ip6_cork_release(cork, &v6_cork);
2091 return ERR_PTR(err);
2093 if (ipc6->dontfrag < 0)
2094 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2096 err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2097 ¤t->task_frag, getfrag, from,
2098 length + exthdrlen, transhdrlen + exthdrlen,
2101 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2102 return ERR_PTR(err);
2105 return __ip6_make_skb(sk, &queue, cork, &v6_cork);