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 const struct in6_addr *nexthop;
64 struct neighbour *neigh;
67 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
68 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
70 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
71 ((mroute6_is_socket(net, skb) &&
72 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
73 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
74 &ipv6_hdr(skb)->saddr))) {
75 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
77 /* Do not check for IFF_ALLMULTI; multicast routing
78 is not supported in any case.
81 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
82 net, sk, newskb, NULL, newskb->dev,
85 if (ipv6_hdr(skb)->hop_limit == 0) {
86 IP6_INC_STATS(net, idev,
87 IPSTATS_MIB_OUTDISCARDS);
93 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
95 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
96 IPV6_ADDR_SCOPE_NODELOCAL &&
97 !(dev->flags & IFF_LOOPBACK)) {
103 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
104 int res = lwtunnel_xmit(skb);
106 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
111 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
112 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
113 if (unlikely(!neigh))
114 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
115 if (!IS_ERR(neigh)) {
116 sock_confirm_neigh(skb, neigh);
117 ret = neigh_output(neigh, skb, false);
118 rcu_read_unlock_bh();
121 rcu_read_unlock_bh();
123 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
128 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
130 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
131 /* Policy lookup after SNAT yielded a new policy */
132 if (skb_dst(skb)->xfrm) {
133 IPCB(skb)->flags |= IPSKB_REROUTED;
134 return dst_output(net, sk, skb);
138 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
139 dst_allfrag(skb_dst(skb)) ||
140 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
141 return ip6_fragment(net, sk, skb, ip6_finish_output2);
143 return ip6_finish_output2(net, sk, skb);
146 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
150 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
152 case NET_XMIT_SUCCESS:
153 return __ip6_finish_output(net, sk, skb);
155 return __ip6_finish_output(net, sk, skb) ? : ret;
162 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
164 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
165 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
167 skb->protocol = htons(ETH_P_IPV6);
170 if (unlikely(idev->cnf.disable_ipv6)) {
171 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
176 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
177 net, sk, skb, indev, dev,
179 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
182 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
184 if (!np->autoflowlabel_set)
185 return ip6_default_np_autolabel(net);
187 return np->autoflowlabel;
191 * xmit an sk_buff (used by TCP, SCTP and DCCP)
192 * Note : socket lock is not held for SYNACK packets, but might be modified
193 * by calls to skb_set_owner_w() and ipv6_local_error(),
194 * which are using proper atomic operations or spinlocks.
196 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
197 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
199 struct net *net = sock_net(sk);
200 const struct ipv6_pinfo *np = inet6_sk(sk);
201 struct in6_addr *first_hop = &fl6->daddr;
202 struct dst_entry *dst = skb_dst(skb);
203 unsigned int head_room;
205 u8 proto = fl6->flowi6_proto;
206 int seg_len = skb->len;
210 head_room = sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
212 head_room += opt->opt_nflen + opt->opt_flen;
214 if (unlikely(skb_headroom(skb) < head_room)) {
215 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
217 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
218 IPSTATS_MIB_OUTDISCARDS);
223 skb_set_owner_w(skb2, skb->sk);
229 seg_len += opt->opt_nflen + opt->opt_flen;
232 ipv6_push_frag_opts(skb, opt, &proto);
235 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
239 skb_push(skb, sizeof(struct ipv6hdr));
240 skb_reset_network_header(skb);
244 * Fill in the IPv6 header
247 hlimit = np->hop_limit;
249 hlimit = ip6_dst_hoplimit(dst);
251 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
252 ip6_autoflowlabel(net, np), fl6));
254 hdr->payload_len = htons(seg_len);
255 hdr->nexthdr = proto;
256 hdr->hop_limit = hlimit;
258 hdr->saddr = fl6->saddr;
259 hdr->daddr = *first_hop;
261 skb->protocol = htons(ETH_P_IPV6);
262 skb->priority = priority;
266 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
267 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
268 IPSTATS_MIB_OUT, skb->len);
270 /* if egress device is enslaved to an L3 master device pass the
271 * skb to its handler for processing
273 skb = l3mdev_ip6_out((struct sock *)sk, skb);
277 /* hooks should never assume socket lock is held.
278 * we promote our socket to non const
280 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
281 net, (struct sock *)sk, skb, NULL, dst->dev,
286 /* ipv6_local_error() does not require socket lock,
287 * we promote our socket to non const
289 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
291 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
295 EXPORT_SYMBOL(ip6_xmit);
297 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
299 struct ip6_ra_chain *ra;
300 struct sock *last = NULL;
302 read_lock(&ip6_ra_lock);
303 for (ra = ip6_ra_chain; ra; ra = ra->next) {
304 struct sock *sk = ra->sk;
305 if (sk && ra->sel == sel &&
306 (!sk->sk_bound_dev_if ||
307 sk->sk_bound_dev_if == skb->dev->ifindex)) {
308 struct ipv6_pinfo *np = inet6_sk(sk);
310 if (np && np->rtalert_isolate &&
311 !net_eq(sock_net(sk), dev_net(skb->dev))) {
315 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
317 rawv6_rcv(last, skb2);
324 rawv6_rcv(last, skb);
325 read_unlock(&ip6_ra_lock);
328 read_unlock(&ip6_ra_lock);
332 static int ip6_forward_proxy_check(struct sk_buff *skb)
334 struct ipv6hdr *hdr = ipv6_hdr(skb);
335 u8 nexthdr = hdr->nexthdr;
339 if (ipv6_ext_hdr(nexthdr)) {
340 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
344 offset = sizeof(struct ipv6hdr);
346 if (nexthdr == IPPROTO_ICMPV6) {
347 struct icmp6hdr *icmp6;
349 if (!pskb_may_pull(skb, (skb_network_header(skb) +
350 offset + 1 - skb->data)))
353 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
355 switch (icmp6->icmp6_type) {
356 case NDISC_ROUTER_SOLICITATION:
357 case NDISC_ROUTER_ADVERTISEMENT:
358 case NDISC_NEIGHBOUR_SOLICITATION:
359 case NDISC_NEIGHBOUR_ADVERTISEMENT:
361 /* For reaction involving unicast neighbor discovery
362 * message destined to the proxied address, pass it to
372 * The proxying router can't forward traffic sent to a link-local
373 * address, so signal the sender and discard the packet. This
374 * behavior is clarified by the MIPv6 specification.
376 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
377 dst_link_failure(skb);
384 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
387 struct dst_entry *dst = skb_dst(skb);
389 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
390 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
392 #ifdef CONFIG_NET_SWITCHDEV
393 if (skb->offload_l3_fwd_mark) {
400 return dst_output(net, sk, skb);
403 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
408 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
409 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
415 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
421 int ip6_forward(struct sk_buff *skb)
423 struct inet6_dev *idev = __in6_dev_get_safely(skb->dev);
424 struct dst_entry *dst = skb_dst(skb);
425 struct ipv6hdr *hdr = ipv6_hdr(skb);
426 struct inet6_skb_parm *opt = IP6CB(skb);
427 struct net *net = dev_net(dst->dev);
430 if (net->ipv6.devconf_all->forwarding == 0)
433 if (skb->pkt_type != PACKET_HOST)
436 if (unlikely(skb->sk))
439 if (skb_warn_if_lro(skb))
442 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
443 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
447 skb_forward_csum(skb);
450 * We DO NOT make any processing on
451 * RA packets, pushing them to user level AS IS
452 * without ane WARRANTY that application will be able
453 * to interpret them. The reason is that we
454 * cannot make anything clever here.
456 * We are not end-node, so that if packet contains
457 * AH/ESP, we cannot make anything.
458 * Defragmentation also would be mistake, RA packets
459 * cannot be fragmented, because there is no warranty
460 * that different fragments will go along one path. --ANK
462 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
463 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
468 * check and decrement ttl
470 if (hdr->hop_limit <= 1) {
471 /* Force OUTPUT device used as source address */
473 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
474 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
480 /* XXX: idev->cnf.proxy_ndp? */
481 if (net->ipv6.devconf_all->proxy_ndp &&
482 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
483 int proxied = ip6_forward_proxy_check(skb);
485 return ip6_input(skb);
486 else if (proxied < 0) {
487 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
492 if (!xfrm6_route_forward(skb)) {
493 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
498 /* IPv6 specs say nothing about it, but it is clear that we cannot
499 send redirects to source routed frames.
500 We don't send redirects to frames decapsulated from IPsec.
502 if (IP6CB(skb)->iif == dst->dev->ifindex &&
503 opt->srcrt == 0 && !skb_sec_path(skb)) {
504 struct in6_addr *target = NULL;
505 struct inet_peer *peer;
509 * incoming and outgoing devices are the same
513 rt = (struct rt6_info *) dst;
514 if (rt->rt6i_flags & RTF_GATEWAY)
515 target = &rt->rt6i_gateway;
517 target = &hdr->daddr;
519 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
521 /* Limit redirects both by destination (here)
522 and by source (inside ndisc_send_redirect)
524 if (inet_peer_xrlim_allow(peer, 1*HZ))
525 ndisc_send_redirect(skb, target);
529 int addrtype = ipv6_addr_type(&hdr->saddr);
531 /* This check is security critical. */
532 if (addrtype == IPV6_ADDR_ANY ||
533 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
535 if (addrtype & IPV6_ADDR_LINKLOCAL) {
536 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
537 ICMPV6_NOT_NEIGHBOUR, 0);
542 mtu = ip6_dst_mtu_forward(dst);
543 if (mtu < IPV6_MIN_MTU)
546 if (ip6_pkt_too_big(skb, mtu)) {
547 /* Again, force OUTPUT device used as source address */
549 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
550 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
551 __IP6_INC_STATS(net, ip6_dst_idev(dst),
552 IPSTATS_MIB_FRAGFAILS);
557 if (skb_cow(skb, dst->dev->hard_header_len)) {
558 __IP6_INC_STATS(net, ip6_dst_idev(dst),
559 IPSTATS_MIB_OUTDISCARDS);
565 /* Mangling hops number delayed to point after skb COW */
569 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
570 net, NULL, skb, skb->dev, dst->dev,
574 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
580 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
582 to->pkt_type = from->pkt_type;
583 to->priority = from->priority;
584 to->protocol = from->protocol;
586 skb_dst_set(to, dst_clone(skb_dst(from)));
588 to->mark = from->mark;
590 skb_copy_hash(to, from);
592 #ifdef CONFIG_NET_SCHED
593 to->tc_index = from->tc_index;
596 skb_ext_copy(to, from);
597 skb_copy_secmark(to, from);
600 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
601 u8 nexthdr, __be32 frag_id,
602 struct ip6_fraglist_iter *iter)
604 unsigned int first_len;
608 *prevhdr = NEXTHDR_FRAGMENT;
609 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
613 iter->frag = skb_shinfo(skb)->frag_list;
614 skb_frag_list_init(skb);
618 iter->frag_id = frag_id;
619 iter->nexthdr = nexthdr;
621 __skb_pull(skb, hlen);
622 fh = __skb_push(skb, sizeof(struct frag_hdr));
623 __skb_push(skb, hlen);
624 skb_reset_network_header(skb);
625 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
627 fh->nexthdr = nexthdr;
629 fh->frag_off = htons(IP6_MF);
630 fh->identification = frag_id;
632 first_len = skb_pagelen(skb);
633 skb->data_len = first_len - skb_headlen(skb);
634 skb->len = first_len;
635 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
639 EXPORT_SYMBOL(ip6_fraglist_init);
641 void ip6_fraglist_prepare(struct sk_buff *skb,
642 struct ip6_fraglist_iter *iter)
644 struct sk_buff *frag = iter->frag;
645 unsigned int hlen = iter->hlen;
648 frag->ip_summed = CHECKSUM_NONE;
649 skb_reset_transport_header(frag);
650 fh = __skb_push(frag, sizeof(struct frag_hdr));
651 __skb_push(frag, hlen);
652 skb_reset_network_header(frag);
653 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
654 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
655 fh->nexthdr = iter->nexthdr;
657 fh->frag_off = htons(iter->offset);
659 fh->frag_off |= htons(IP6_MF);
660 fh->identification = iter->frag_id;
661 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
662 ip6_copy_metadata(frag, skb);
664 EXPORT_SYMBOL(ip6_fraglist_prepare);
666 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
667 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
668 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
670 state->prevhdr = prevhdr;
671 state->nexthdr = nexthdr;
672 state->frag_id = frag_id;
677 state->left = skb->len - hlen; /* Space per frame */
678 state->ptr = hlen; /* Where to start from */
680 state->hroom = hdr_room;
681 state->troom = needed_tailroom;
685 EXPORT_SYMBOL(ip6_frag_init);
687 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
689 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
690 struct sk_buff *frag;
695 /* IF: it doesn't fit, use 'mtu' - the data space left */
696 if (len > state->mtu)
698 /* IF: we are not sending up to and including the packet end
699 then align the next start on an eight byte boundary */
700 if (len < state->left)
703 /* Allocate buffer */
704 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
705 state->hroom + state->troom, GFP_ATOMIC);
707 return ERR_PTR(-ENOMEM);
710 * Set up data on packet
713 ip6_copy_metadata(frag, skb);
714 skb_reserve(frag, state->hroom);
715 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
716 skb_reset_network_header(frag);
717 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
718 frag->transport_header = (frag->network_header + state->hlen +
719 sizeof(struct frag_hdr));
722 * Charge the memory for the fragment to any owner
726 skb_set_owner_w(frag, skb->sk);
729 * Copy the packet header into the new buffer.
731 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
733 fragnexthdr_offset = skb_network_header(frag);
734 fragnexthdr_offset += prevhdr - skb_network_header(skb);
735 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
738 * Build fragment header.
740 fh->nexthdr = state->nexthdr;
742 fh->identification = state->frag_id;
745 * Copy a block of the IP datagram.
747 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
751 fh->frag_off = htons(state->offset);
753 fh->frag_off |= htons(IP6_MF);
754 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
757 state->offset += len;
761 EXPORT_SYMBOL(ip6_frag_next);
763 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
764 int (*output)(struct net *, struct sock *, struct sk_buff *))
766 struct sk_buff *frag;
767 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
768 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
769 inet6_sk(skb->sk) : NULL;
770 struct ip6_frag_state state;
771 unsigned int mtu, hlen, nexthdr_offset;
772 ktime_t tstamp = skb->tstamp;
775 u8 *prevhdr, nexthdr = 0;
777 err = ip6_find_1stfragopt(skb, &prevhdr);
782 nexthdr_offset = prevhdr - skb_network_header(skb);
784 mtu = ip6_skb_dst_mtu(skb);
786 /* We must not fragment if the socket is set to force MTU discovery
787 * or if the skb it not generated by a local socket.
789 if (unlikely(!skb->ignore_df && skb->len > mtu))
792 if (IP6CB(skb)->frag_max_size) {
793 if (IP6CB(skb)->frag_max_size > mtu)
796 /* don't send fragments larger than what we received */
797 mtu = IP6CB(skb)->frag_max_size;
798 if (mtu < IPV6_MIN_MTU)
802 if (np && np->frag_size < mtu) {
806 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
808 mtu -= hlen + sizeof(struct frag_hdr);
810 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
811 &ipv6_hdr(skb)->saddr);
813 if (skb->ip_summed == CHECKSUM_PARTIAL &&
814 (err = skb_checksum_help(skb)))
817 prevhdr = skb_network_header(skb) + nexthdr_offset;
818 hroom = LL_RESERVED_SPACE(rt->dst.dev);
819 if (skb_has_frag_list(skb)) {
820 unsigned int first_len = skb_pagelen(skb);
821 struct ip6_fraglist_iter iter;
822 struct sk_buff *frag2;
824 if (first_len - hlen > mtu ||
825 ((first_len - hlen) & 7) ||
827 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
830 skb_walk_frags(skb, frag) {
831 /* Correct geometry. */
832 if (frag->len > mtu ||
833 ((frag->len & 7) && frag->next) ||
834 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
835 goto slow_path_clean;
837 /* Partially cloned skb? */
838 if (skb_shared(frag))
839 goto slow_path_clean;
844 frag->destructor = sock_wfree;
846 skb->truesize -= frag->truesize;
849 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
855 /* Prepare header of the next frame,
856 * before previous one went down. */
858 ip6_fraglist_prepare(skb, &iter);
860 skb->tstamp = tstamp;
861 err = output(net, sk, skb);
863 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
864 IPSTATS_MIB_FRAGCREATES);
866 if (err || !iter.frag)
869 skb = ip6_fraglist_next(&iter);
875 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
876 IPSTATS_MIB_FRAGOKS);
880 kfree_skb_list(iter.frag);
882 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
883 IPSTATS_MIB_FRAGFAILS);
887 skb_walk_frags(skb, frag2) {
891 frag2->destructor = NULL;
892 skb->truesize += frag2->truesize;
898 * Fragment the datagram.
901 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
902 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
906 * Keep copying data until we run out.
909 while (state.left > 0) {
910 frag = ip6_frag_next(skb, &state);
917 * Put this fragment into the sending queue.
919 frag->tstamp = tstamp;
920 err = output(net, sk, frag);
924 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
925 IPSTATS_MIB_FRAGCREATES);
927 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
928 IPSTATS_MIB_FRAGOKS);
933 if (skb->sk && dst_allfrag(skb_dst(skb)))
934 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
936 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
940 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
941 IPSTATS_MIB_FRAGFAILS);
946 static inline int ip6_rt_check(const struct rt6key *rt_key,
947 const struct in6_addr *fl_addr,
948 const struct in6_addr *addr_cache)
950 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
951 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
954 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
955 struct dst_entry *dst,
956 const struct flowi6 *fl6)
958 struct ipv6_pinfo *np = inet6_sk(sk);
964 if (dst->ops->family != AF_INET6) {
969 rt = (struct rt6_info *)dst;
970 /* Yes, checking route validity in not connected
971 * case is not very simple. Take into account,
972 * that we do not support routing by source, TOS,
973 * and MSG_DONTROUTE --ANK (980726)
975 * 1. ip6_rt_check(): If route was host route,
976 * check that cached destination is current.
977 * If it is network route, we still may
978 * check its validity using saved pointer
979 * to the last used address: daddr_cache.
980 * We do not want to save whole address now,
981 * (because main consumer of this service
982 * is tcp, which has not this problem),
983 * so that the last trick works only on connected
985 * 2. oif also should be the same.
987 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
988 #ifdef CONFIG_IPV6_SUBTREES
989 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
991 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
992 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
1001 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1002 struct dst_entry **dst, struct flowi6 *fl6)
1004 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1005 struct neighbour *n;
1006 struct rt6_info *rt;
1011 /* The correct way to handle this would be to do
1012 * ip6_route_get_saddr, and then ip6_route_output; however,
1013 * the route-specific preferred source forces the
1014 * ip6_route_output call _before_ ip6_route_get_saddr.
1016 * In source specific routing (no src=any default route),
1017 * ip6_route_output will fail given src=any saddr, though, so
1018 * that's why we try it again later.
1020 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
1021 struct fib6_info *from;
1022 struct rt6_info *rt;
1023 bool had_dst = *dst != NULL;
1026 *dst = ip6_route_output(net, sk, fl6);
1027 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1030 from = rt ? rcu_dereference(rt->from) : NULL;
1031 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1032 sk ? inet6_sk(sk)->srcprefs : 0,
1037 goto out_err_release;
1039 /* If we had an erroneous initial result, pretend it
1040 * never existed and let the SA-enabled version take
1043 if (!had_dst && (*dst)->error) {
1048 if (fl6->flowi6_oif)
1049 flags |= RT6_LOOKUP_F_IFACE;
1053 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1055 err = (*dst)->error;
1057 goto out_err_release;
1059 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1061 * Here if the dst entry we've looked up
1062 * has a neighbour entry that is in the INCOMPLETE
1063 * state and the src address from the flow is
1064 * marked as OPTIMISTIC, we release the found
1065 * dst entry and replace it instead with the
1066 * dst entry of the nexthop router
1068 rt = (struct rt6_info *) *dst;
1070 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1071 rt6_nexthop(rt, &fl6->daddr));
1072 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1073 rcu_read_unlock_bh();
1076 struct inet6_ifaddr *ifp;
1077 struct flowi6 fl_gw6;
1080 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1083 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1089 * We need to get the dst entry for the
1090 * default router instead
1093 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1094 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1095 *dst = ip6_route_output(net, sk, &fl_gw6);
1096 err = (*dst)->error;
1098 goto out_err_release;
1102 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1103 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1104 err = -EAFNOSUPPORT;
1105 goto out_err_release;
1114 if (err == -ENETUNREACH)
1115 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1120 * ip6_dst_lookup - perform route lookup on flow
1121 * @net: Network namespace to perform lookup in
1122 * @sk: socket which provides route info
1123 * @dst: pointer to dst_entry * for result
1124 * @fl6: flow to lookup
1126 * This function performs a route lookup on the given flow.
1128 * It returns zero on success, or a standard errno code on error.
1130 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1134 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1136 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1139 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1140 * @net: Network namespace to perform lookup in
1141 * @sk: socket which provides route info
1142 * @fl6: flow to lookup
1143 * @final_dst: final destination address for ipsec lookup
1145 * This function performs a route lookup on the given flow.
1147 * It returns a valid dst pointer on success, or a pointer encoded
1150 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1151 const struct in6_addr *final_dst)
1153 struct dst_entry *dst = NULL;
1156 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1158 return ERR_PTR(err);
1160 fl6->daddr = *final_dst;
1162 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1164 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1167 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1168 * @sk: socket which provides the dst cache and route info
1169 * @fl6: flow to lookup
1170 * @final_dst: final destination address for ipsec lookup
1171 * @connected: whether @sk is connected or not
1173 * This function performs a route lookup on the given flow with the
1174 * possibility of using the cached route in the socket if it is valid.
1175 * It will take the socket dst lock when operating on the dst cache.
1176 * As a result, this function can only be used in process context.
1178 * In addition, for a connected socket, cache the dst in the socket
1179 * if the current cache is not valid.
1181 * It returns a valid dst pointer on success, or a pointer encoded
1184 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1185 const struct in6_addr *final_dst,
1188 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1190 dst = ip6_sk_dst_check(sk, dst, fl6);
1194 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1195 if (connected && !IS_ERR(dst))
1196 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1200 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1203 * ip6_dst_lookup_tunnel - perform route lookup on tunnel
1204 * @skb: Packet for which lookup is done
1205 * @dev: Tunnel device
1206 * @net: Network namespace of tunnel device
1207 * @sock: Socket which provides route info
1208 * @saddr: Memory to store the src ip address
1209 * @info: Tunnel information
1210 * @protocol: IP protocol
1211 * @use_cache: Flag to enable cache usage
1212 * This function performs a route lookup on a tunnel
1214 * It returns a valid dst pointer and stores src address to be used in
1215 * tunnel in param saddr on success, else a pointer encoded error code.
1218 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1219 struct net_device *dev,
1221 struct socket *sock,
1222 struct in6_addr *saddr,
1223 const struct ip_tunnel_info *info,
1227 struct dst_entry *dst = NULL;
1228 #ifdef CONFIG_DST_CACHE
1229 struct dst_cache *dst_cache;
1234 #ifdef CONFIG_DST_CACHE
1235 dst_cache = (struct dst_cache *)&info->dst_cache;
1237 dst = dst_cache_get_ip6(dst_cache, saddr);
1242 memset(&fl6, 0, sizeof(fl6));
1243 fl6.flowi6_mark = skb->mark;
1244 fl6.flowi6_proto = protocol;
1245 fl6.daddr = info->key.u.ipv6.dst;
1246 fl6.saddr = info->key.u.ipv6.src;
1247 prio = info->key.tos;
1248 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(prio),
1251 dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1254 netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1255 return ERR_PTR(-ENETUNREACH);
1257 if (dst->dev == dev) { /* is this necessary? */
1258 netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1260 return ERR_PTR(-ELOOP);
1262 #ifdef CONFIG_DST_CACHE
1264 dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1269 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1271 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1274 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1277 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1280 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1283 static void ip6_append_data_mtu(unsigned int *mtu,
1285 unsigned int fragheaderlen,
1286 struct sk_buff *skb,
1287 struct rt6_info *rt,
1288 unsigned int orig_mtu)
1290 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1292 /* first fragment, reserve header_len */
1293 *mtu = orig_mtu - rt->dst.header_len;
1297 * this fragment is not first, the headers
1298 * space is regarded as data space.
1302 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1303 + fragheaderlen - sizeof(struct frag_hdr);
1307 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1308 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1309 struct rt6_info *rt, struct flowi6 *fl6)
1311 struct ipv6_pinfo *np = inet6_sk(sk);
1313 struct ipv6_txoptions *opt = ipc6->opt;
1319 if (WARN_ON(v6_cork->opt))
1322 v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1323 if (unlikely(!v6_cork->opt))
1326 v6_cork->opt->tot_len = sizeof(*opt);
1327 v6_cork->opt->opt_flen = opt->opt_flen;
1328 v6_cork->opt->opt_nflen = opt->opt_nflen;
1330 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1332 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1335 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1337 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1340 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1342 if (opt->hopopt && !v6_cork->opt->hopopt)
1345 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1347 if (opt->srcrt && !v6_cork->opt->srcrt)
1350 /* need source address above miyazawa*/
1353 cork->base.dst = &rt->dst;
1354 cork->fl.u.ip6 = *fl6;
1355 v6_cork->hop_limit = ipc6->hlimit;
1356 v6_cork->tclass = ipc6->tclass;
1357 if (rt->dst.flags & DST_XFRM_TUNNEL)
1358 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1359 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1361 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1362 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1363 if (np->frag_size < mtu) {
1365 mtu = np->frag_size;
1367 if (mtu < IPV6_MIN_MTU)
1369 cork->base.fragsize = mtu;
1370 cork->base.gso_size = ipc6->gso_size;
1371 cork->base.tx_flags = 0;
1372 cork->base.mark = ipc6->sockc.mark;
1373 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1375 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1376 cork->base.flags |= IPCORK_ALLFRAG;
1377 cork->base.length = 0;
1379 cork->base.transmit_time = ipc6->sockc.transmit_time;
1384 static int __ip6_append_data(struct sock *sk,
1386 struct sk_buff_head *queue,
1387 struct inet_cork *cork,
1388 struct inet6_cork *v6_cork,
1389 struct page_frag *pfrag,
1390 int getfrag(void *from, char *to, int offset,
1391 int len, int odd, struct sk_buff *skb),
1392 void *from, int length, int transhdrlen,
1393 unsigned int flags, struct ipcm6_cookie *ipc6)
1395 struct sk_buff *skb, *skb_prev = NULL;
1396 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1397 struct ubuf_info *uarg = NULL;
1399 int dst_exthdrlen = 0;
1405 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1406 struct ipv6_txoptions *opt = v6_cork->opt;
1407 int csummode = CHECKSUM_NONE;
1408 unsigned int maxnonfragsize, headersize;
1409 unsigned int wmem_alloc_delta = 0;
1410 bool paged, extra_uref = false;
1412 skb = skb_peek_tail(queue);
1414 exthdrlen = opt ? opt->opt_flen : 0;
1415 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1418 paged = !!cork->gso_size;
1419 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1422 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1423 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1424 tskey = sk->sk_tskey++;
1426 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1428 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1429 (opt ? opt->opt_nflen : 0);
1430 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1431 sizeof(struct frag_hdr);
1433 headersize = sizeof(struct ipv6hdr) +
1434 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1435 (dst_allfrag(&rt->dst) ?
1436 sizeof(struct frag_hdr) : 0) +
1437 rt->rt6i_nfheader_len;
1439 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1440 * the first fragment
1442 if (headersize + transhdrlen > mtu)
1445 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1446 (sk->sk_protocol == IPPROTO_UDP ||
1447 sk->sk_protocol == IPPROTO_RAW)) {
1448 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1449 sizeof(struct ipv6hdr));
1453 if (ip6_sk_ignore_df(sk))
1454 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1456 maxnonfragsize = mtu;
1458 if (cork->length + length > maxnonfragsize - headersize) {
1460 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1461 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1465 /* CHECKSUM_PARTIAL only with no extension headers and when
1466 * we are not going to fragment
1468 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1469 headersize == sizeof(struct ipv6hdr) &&
1470 length <= mtu - headersize &&
1471 (!(flags & MSG_MORE) || cork->gso_size) &&
1472 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1473 csummode = CHECKSUM_PARTIAL;
1475 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1476 uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb));
1479 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1480 if (rt->dst.dev->features & NETIF_F_SG &&
1481 csummode == CHECKSUM_PARTIAL) {
1485 skb_zcopy_set(skb, uarg, &extra_uref);
1490 * Let's try using as much space as possible.
1491 * Use MTU if total length of the message fits into the MTU.
1492 * Otherwise, we need to reserve fragment header and
1493 * fragment alignment (= 8-15 octects, in total).
1495 * Note that we may need to "move" the data from the tail of
1496 * of the buffer to the new fragment when we split
1499 * FIXME: It may be fragmented into multiple chunks
1500 * at once if non-fragmentable extension headers
1505 cork->length += length;
1509 while (length > 0) {
1510 /* Check if the remaining data fits into current packet. */
1511 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1513 copy = maxfraglen - skb->len;
1517 unsigned int datalen;
1518 unsigned int fraglen;
1519 unsigned int fraggap;
1520 unsigned int alloclen;
1521 unsigned int pagedlen;
1523 /* There's no room in the current skb */
1525 fraggap = skb->len - maxfraglen;
1528 /* update mtu and maxfraglen if necessary */
1529 if (!skb || !skb_prev)
1530 ip6_append_data_mtu(&mtu, &maxfraglen,
1531 fragheaderlen, skb, rt,
1537 * If remaining data exceeds the mtu,
1538 * we know we need more fragment(s).
1540 datalen = length + fraggap;
1542 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1543 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1544 fraglen = datalen + fragheaderlen;
1547 if ((flags & MSG_MORE) &&
1548 !(rt->dst.dev->features&NETIF_F_SG))
1553 alloclen = min_t(int, fraglen, MAX_HEADER);
1554 pagedlen = fraglen - alloclen;
1557 alloclen += dst_exthdrlen;
1559 if (datalen != length + fraggap) {
1561 * this is not the last fragment, the trailer
1562 * space is regarded as data space.
1564 datalen += rt->dst.trailer_len;
1567 alloclen += rt->dst.trailer_len;
1568 fraglen = datalen + fragheaderlen;
1571 * We just reserve space for fragment header.
1572 * Note: this may be overallocation if the message
1573 * (without MSG_MORE) fits into the MTU.
1575 alloclen += sizeof(struct frag_hdr);
1577 copy = datalen - transhdrlen - fraggap - pagedlen;
1583 skb = sock_alloc_send_skb(sk,
1585 (flags & MSG_DONTWAIT), &err);
1588 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1590 skb = alloc_skb(alloclen + hh_len,
1598 * Fill in the control structures
1600 skb->protocol = htons(ETH_P_IPV6);
1601 skb->ip_summed = csummode;
1603 /* reserve for fragmentation and ipsec header */
1604 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1608 * Find where to start putting bytes
1610 data = skb_put(skb, fraglen - pagedlen);
1611 skb_set_network_header(skb, exthdrlen);
1612 data += fragheaderlen;
1613 skb->transport_header = (skb->network_header +
1616 skb->csum = skb_copy_and_csum_bits(
1617 skb_prev, maxfraglen,
1618 data + transhdrlen, fraggap);
1619 skb_prev->csum = csum_sub(skb_prev->csum,
1622 pskb_trim_unique(skb_prev, maxfraglen);
1625 getfrag(from, data + transhdrlen, offset,
1626 copy, fraggap, skb) < 0) {
1633 length -= copy + transhdrlen;
1638 /* Only the initial fragment is time stamped */
1639 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1641 skb_shinfo(skb)->tskey = tskey;
1643 skb_zcopy_set(skb, uarg, &extra_uref);
1645 if ((flags & MSG_CONFIRM) && !skb_prev)
1646 skb_set_dst_pending_confirm(skb, 1);
1649 * Put the packet on the pending queue
1651 if (!skb->destructor) {
1652 skb->destructor = sock_wfree;
1654 wmem_alloc_delta += skb->truesize;
1656 __skb_queue_tail(queue, skb);
1663 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1664 skb_tailroom(skb) >= copy) {
1668 if (getfrag(from, skb_put(skb, copy),
1669 offset, copy, off, skb) < 0) {
1670 __skb_trim(skb, off);
1674 } else if (!uarg || !uarg->zerocopy) {
1675 int i = skb_shinfo(skb)->nr_frags;
1678 if (!sk_page_frag_refill(sk, pfrag))
1681 if (!skb_can_coalesce(skb, i, pfrag->page,
1684 if (i == MAX_SKB_FRAGS)
1687 __skb_fill_page_desc(skb, i, pfrag->page,
1689 skb_shinfo(skb)->nr_frags = ++i;
1690 get_page(pfrag->page);
1692 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1694 page_address(pfrag->page) + pfrag->offset,
1695 offset, copy, skb->len, skb) < 0)
1698 pfrag->offset += copy;
1699 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1701 skb->data_len += copy;
1702 skb->truesize += copy;
1703 wmem_alloc_delta += copy;
1705 err = skb_zerocopy_iter_dgram(skb, from, copy);
1713 if (wmem_alloc_delta)
1714 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1721 sock_zerocopy_put_abort(uarg, extra_uref);
1722 cork->length -= length;
1723 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1724 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1728 int ip6_append_data(struct sock *sk,
1729 int getfrag(void *from, char *to, int offset, int len,
1730 int odd, struct sk_buff *skb),
1731 void *from, int length, int transhdrlen,
1732 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1733 struct rt6_info *rt, unsigned int flags)
1735 struct inet_sock *inet = inet_sk(sk);
1736 struct ipv6_pinfo *np = inet6_sk(sk);
1740 if (flags&MSG_PROBE)
1742 if (skb_queue_empty(&sk->sk_write_queue)) {
1746 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1751 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1752 length += exthdrlen;
1753 transhdrlen += exthdrlen;
1755 fl6 = &inet->cork.fl.u.ip6;
1759 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1760 &np->cork, sk_page_frag(sk), getfrag,
1761 from, length, transhdrlen, flags, ipc6);
1763 EXPORT_SYMBOL_GPL(ip6_append_data);
1765 static void ip6_cork_release(struct inet_cork_full *cork,
1766 struct inet6_cork *v6_cork)
1769 kfree(v6_cork->opt->dst0opt);
1770 kfree(v6_cork->opt->dst1opt);
1771 kfree(v6_cork->opt->hopopt);
1772 kfree(v6_cork->opt->srcrt);
1773 kfree(v6_cork->opt);
1774 v6_cork->opt = NULL;
1777 if (cork->base.dst) {
1778 dst_release(cork->base.dst);
1779 cork->base.dst = NULL;
1780 cork->base.flags &= ~IPCORK_ALLFRAG;
1782 memset(&cork->fl, 0, sizeof(cork->fl));
1785 struct sk_buff *__ip6_make_skb(struct sock *sk,
1786 struct sk_buff_head *queue,
1787 struct inet_cork_full *cork,
1788 struct inet6_cork *v6_cork)
1790 struct sk_buff *skb, *tmp_skb;
1791 struct sk_buff **tail_skb;
1792 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1793 struct ipv6_pinfo *np = inet6_sk(sk);
1794 struct net *net = sock_net(sk);
1795 struct ipv6hdr *hdr;
1796 struct ipv6_txoptions *opt = v6_cork->opt;
1797 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1798 struct flowi6 *fl6 = &cork->fl.u.ip6;
1799 unsigned char proto = fl6->flowi6_proto;
1801 skb = __skb_dequeue(queue);
1804 tail_skb = &(skb_shinfo(skb)->frag_list);
1806 /* move skb->data to ip header from ext header */
1807 if (skb->data < skb_network_header(skb))
1808 __skb_pull(skb, skb_network_offset(skb));
1809 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1810 __skb_pull(tmp_skb, skb_network_header_len(skb));
1811 *tail_skb = tmp_skb;
1812 tail_skb = &(tmp_skb->next);
1813 skb->len += tmp_skb->len;
1814 skb->data_len += tmp_skb->len;
1815 skb->truesize += tmp_skb->truesize;
1816 tmp_skb->destructor = NULL;
1820 /* Allow local fragmentation. */
1821 skb->ignore_df = ip6_sk_ignore_df(sk);
1823 *final_dst = fl6->daddr;
1824 __skb_pull(skb, skb_network_header_len(skb));
1825 if (opt && opt->opt_flen)
1826 ipv6_push_frag_opts(skb, opt, &proto);
1827 if (opt && opt->opt_nflen)
1828 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1830 skb_push(skb, sizeof(struct ipv6hdr));
1831 skb_reset_network_header(skb);
1832 hdr = ipv6_hdr(skb);
1834 ip6_flow_hdr(hdr, v6_cork->tclass,
1835 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1836 ip6_autoflowlabel(net, np), fl6));
1837 hdr->hop_limit = v6_cork->hop_limit;
1838 hdr->nexthdr = proto;
1839 hdr->saddr = fl6->saddr;
1840 hdr->daddr = *final_dst;
1842 skb->priority = sk->sk_priority;
1843 skb->mark = cork->base.mark;
1845 skb->tstamp = cork->base.transmit_time;
1847 skb_dst_set(skb, dst_clone(&rt->dst));
1848 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1849 if (proto == IPPROTO_ICMPV6) {
1850 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1852 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1853 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1856 ip6_cork_release(cork, v6_cork);
1861 int ip6_send_skb(struct sk_buff *skb)
1863 struct net *net = sock_net(skb->sk);
1864 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1867 err = ip6_local_out(net, skb->sk, skb);
1870 err = net_xmit_errno(err);
1872 IP6_INC_STATS(net, rt->rt6i_idev,
1873 IPSTATS_MIB_OUTDISCARDS);
1879 int ip6_push_pending_frames(struct sock *sk)
1881 struct sk_buff *skb;
1883 skb = ip6_finish_skb(sk);
1887 return ip6_send_skb(skb);
1889 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1891 static void __ip6_flush_pending_frames(struct sock *sk,
1892 struct sk_buff_head *queue,
1893 struct inet_cork_full *cork,
1894 struct inet6_cork *v6_cork)
1896 struct sk_buff *skb;
1898 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1900 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1901 IPSTATS_MIB_OUTDISCARDS);
1905 ip6_cork_release(cork, v6_cork);
1908 void ip6_flush_pending_frames(struct sock *sk)
1910 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1911 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1913 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1915 struct sk_buff *ip6_make_skb(struct sock *sk,
1916 int getfrag(void *from, char *to, int offset,
1917 int len, int odd, struct sk_buff *skb),
1918 void *from, int length, int transhdrlen,
1919 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1920 struct rt6_info *rt, unsigned int flags,
1921 struct inet_cork_full *cork)
1923 struct inet6_cork v6_cork;
1924 struct sk_buff_head queue;
1925 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1928 if (flags & MSG_PROBE)
1931 __skb_queue_head_init(&queue);
1933 cork->base.flags = 0;
1934 cork->base.addr = 0;
1935 cork->base.opt = NULL;
1936 cork->base.dst = NULL;
1938 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
1940 ip6_cork_release(cork, &v6_cork);
1941 return ERR_PTR(err);
1943 if (ipc6->dontfrag < 0)
1944 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1946 err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
1947 ¤t->task_frag, getfrag, from,
1948 length + exthdrlen, transhdrlen + exthdrlen,
1951 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
1952 return ERR_PTR(err);
1955 return __ip6_make_skb(sk, &queue, cork, &v6_cork);