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 /* Last GSO segment can be smaller than gso_size (and MTU).
166 * Adding a fragment header would produce an "atomic fragment",
167 * which is considered harmful (RFC-8021). Avoid that.
169 err = segs->len > mtu ?
170 ip6_fragment(net, sk, segs, ip6_finish_output2) :
171 ip6_finish_output2(net, sk, segs);
179 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
183 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
184 /* Policy lookup after SNAT yielded a new policy */
185 if (skb_dst(skb)->xfrm) {
186 IP6CB(skb)->flags |= IP6SKB_REROUTED;
187 return dst_output(net, sk, skb);
191 mtu = ip6_skb_dst_mtu(skb);
192 if (skb_is_gso(skb) &&
193 !(IP6CB(skb)->flags & IP6SKB_FAKEJUMBO) &&
194 !skb_gso_validate_network_len(skb, mtu))
195 return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
197 if ((skb->len > mtu && !skb_is_gso(skb)) ||
198 dst_allfrag(skb_dst(skb)) ||
199 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
200 return ip6_fragment(net, sk, skb, ip6_finish_output2);
202 return ip6_finish_output2(net, sk, skb);
205 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
209 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
211 case NET_XMIT_SUCCESS:
213 return __ip6_finish_output(net, sk, skb) ? : ret;
215 kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
220 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
222 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
223 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
225 skb->protocol = htons(ETH_P_IPV6);
228 if (unlikely(idev->cnf.disable_ipv6)) {
229 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
230 kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
234 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
235 net, sk, skb, indev, dev,
237 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
239 EXPORT_SYMBOL(ip6_output);
241 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
243 if (!np->autoflowlabel_set)
244 return ip6_default_np_autolabel(net);
246 return np->autoflowlabel;
250 * xmit an sk_buff (used by TCP, SCTP and DCCP)
251 * Note : socket lock is not held for SYNACK packets, but might be modified
252 * by calls to skb_set_owner_w() and ipv6_local_error(),
253 * which are using proper atomic operations or spinlocks.
255 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
256 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
258 struct net *net = sock_net(sk);
259 const struct ipv6_pinfo *np = inet6_sk(sk);
260 struct in6_addr *first_hop = &fl6->daddr;
261 struct dst_entry *dst = skb_dst(skb);
262 struct net_device *dev = dst->dev;
263 struct inet6_dev *idev = ip6_dst_idev(dst);
264 struct hop_jumbo_hdr *hop_jumbo;
265 int hoplen = sizeof(*hop_jumbo);
266 unsigned int head_room;
268 u8 proto = fl6->flowi6_proto;
269 int seg_len = skb->len;
273 head_room = sizeof(struct ipv6hdr) + hoplen + LL_RESERVED_SPACE(dev);
275 head_room += opt->opt_nflen + opt->opt_flen;
277 if (unlikely(head_room > skb_headroom(skb))) {
278 skb = skb_expand_head(skb, head_room);
280 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
286 seg_len += opt->opt_nflen + opt->opt_flen;
289 ipv6_push_frag_opts(skb, opt, &proto);
292 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
296 if (unlikely(seg_len > IPV6_MAXPLEN)) {
297 hop_jumbo = skb_push(skb, hoplen);
299 hop_jumbo->nexthdr = proto;
300 hop_jumbo->hdrlen = 0;
301 hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
302 hop_jumbo->tlv_len = 4;
303 hop_jumbo->jumbo_payload_len = htonl(seg_len + hoplen);
305 proto = IPPROTO_HOPOPTS;
307 IP6CB(skb)->flags |= IP6SKB_FAKEJUMBO;
310 skb_push(skb, sizeof(struct ipv6hdr));
311 skb_reset_network_header(skb);
315 * Fill in the IPv6 header
318 hlimit = np->hop_limit;
320 hlimit = ip6_dst_hoplimit(dst);
322 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
323 ip6_autoflowlabel(net, np), fl6));
325 hdr->payload_len = htons(seg_len);
326 hdr->nexthdr = proto;
327 hdr->hop_limit = hlimit;
329 hdr->saddr = fl6->saddr;
330 hdr->daddr = *first_hop;
332 skb->protocol = htons(ETH_P_IPV6);
333 skb->priority = priority;
337 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
338 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
340 /* if egress device is enslaved to an L3 master device pass the
341 * skb to its handler for processing
343 skb = l3mdev_ip6_out((struct sock *)sk, skb);
347 /* hooks should never assume socket lock is held.
348 * we promote our socket to non const
350 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
351 net, (struct sock *)sk, skb, NULL, dev,
356 /* ipv6_local_error() does not require socket lock,
357 * we promote our socket to non const
359 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
361 IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
365 EXPORT_SYMBOL(ip6_xmit);
367 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
369 struct ip6_ra_chain *ra;
370 struct sock *last = NULL;
372 read_lock(&ip6_ra_lock);
373 for (ra = ip6_ra_chain; ra; ra = ra->next) {
374 struct sock *sk = ra->sk;
375 if (sk && ra->sel == sel &&
376 (!sk->sk_bound_dev_if ||
377 sk->sk_bound_dev_if == skb->dev->ifindex)) {
378 struct ipv6_pinfo *np = inet6_sk(sk);
380 if (np && np->rtalert_isolate &&
381 !net_eq(sock_net(sk), dev_net(skb->dev))) {
385 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
387 rawv6_rcv(last, skb2);
394 rawv6_rcv(last, skb);
395 read_unlock(&ip6_ra_lock);
398 read_unlock(&ip6_ra_lock);
402 static int ip6_forward_proxy_check(struct sk_buff *skb)
404 struct ipv6hdr *hdr = ipv6_hdr(skb);
405 u8 nexthdr = hdr->nexthdr;
409 if (ipv6_ext_hdr(nexthdr)) {
410 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
414 offset = sizeof(struct ipv6hdr);
416 if (nexthdr == IPPROTO_ICMPV6) {
417 struct icmp6hdr *icmp6;
419 if (!pskb_may_pull(skb, (skb_network_header(skb) +
420 offset + 1 - skb->data)))
423 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
425 switch (icmp6->icmp6_type) {
426 case NDISC_ROUTER_SOLICITATION:
427 case NDISC_ROUTER_ADVERTISEMENT:
428 case NDISC_NEIGHBOUR_SOLICITATION:
429 case NDISC_NEIGHBOUR_ADVERTISEMENT:
431 /* For reaction involving unicast neighbor discovery
432 * message destined to the proxied address, pass it to
442 * The proxying router can't forward traffic sent to a link-local
443 * address, so signal the sender and discard the packet. This
444 * behavior is clarified by the MIPv6 specification.
446 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
447 dst_link_failure(skb);
454 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
457 struct dst_entry *dst = skb_dst(skb);
459 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
461 #ifdef CONFIG_NET_SWITCHDEV
462 if (skb->offload_l3_fwd_mark) {
468 skb_clear_tstamp(skb);
469 return dst_output(net, sk, skb);
472 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
477 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
478 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
484 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
490 int ip6_forward(struct sk_buff *skb)
492 struct dst_entry *dst = skb_dst(skb);
493 struct ipv6hdr *hdr = ipv6_hdr(skb);
494 struct inet6_skb_parm *opt = IP6CB(skb);
495 struct net *net = dev_net(dst->dev);
496 struct inet6_dev *idev;
500 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
501 if (net->ipv6.devconf_all->forwarding == 0)
504 if (skb->pkt_type != PACKET_HOST)
507 if (unlikely(skb->sk))
510 if (skb_warn_if_lro(skb))
513 if (!net->ipv6.devconf_all->disable_policy &&
514 (!idev || !idev->cnf.disable_policy) &&
515 !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
516 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
520 skb_forward_csum(skb);
523 * We DO NOT make any processing on
524 * RA packets, pushing them to user level AS IS
525 * without ane WARRANTY that application will be able
526 * to interpret them. The reason is that we
527 * cannot make anything clever here.
529 * We are not end-node, so that if packet contains
530 * AH/ESP, we cannot make anything.
531 * Defragmentation also would be mistake, RA packets
532 * cannot be fragmented, because there is no warranty
533 * that different fragments will go along one path. --ANK
535 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
536 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
541 * check and decrement ttl
543 if (hdr->hop_limit <= 1) {
544 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
545 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
547 kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
551 /* XXX: idev->cnf.proxy_ndp? */
552 if (net->ipv6.devconf_all->proxy_ndp &&
553 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
554 int proxied = ip6_forward_proxy_check(skb);
556 /* It's tempting to decrease the hop limit
557 * here by 1, as we do at the end of the
560 * But that would be incorrect, as proxying is
561 * not forwarding. The ip6_input function
562 * will handle this packet locally, and it
563 * depends on the hop limit being unchanged.
565 * One example is the NDP hop limit, that
566 * always has to stay 255, but other would be
567 * similar checks around RA packets, where the
568 * user can even change the desired limit.
570 return ip6_input(skb);
571 } else if (proxied < 0) {
572 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
577 if (!xfrm6_route_forward(skb)) {
578 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
579 SKB_DR_SET(reason, XFRM_POLICY);
584 /* IPv6 specs say nothing about it, but it is clear that we cannot
585 send redirects to source routed frames.
586 We don't send redirects to frames decapsulated from IPsec.
588 if (IP6CB(skb)->iif == dst->dev->ifindex &&
589 opt->srcrt == 0 && !skb_sec_path(skb)) {
590 struct in6_addr *target = NULL;
591 struct inet_peer *peer;
595 * incoming and outgoing devices are the same
599 rt = (struct rt6_info *) dst;
600 if (rt->rt6i_flags & RTF_GATEWAY)
601 target = &rt->rt6i_gateway;
603 target = &hdr->daddr;
605 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
607 /* Limit redirects both by destination (here)
608 and by source (inside ndisc_send_redirect)
610 if (inet_peer_xrlim_allow(peer, 1*HZ))
611 ndisc_send_redirect(skb, target);
615 int addrtype = ipv6_addr_type(&hdr->saddr);
617 /* This check is security critical. */
618 if (addrtype == IPV6_ADDR_ANY ||
619 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
621 if (addrtype & IPV6_ADDR_LINKLOCAL) {
622 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
623 ICMPV6_NOT_NEIGHBOUR, 0);
628 mtu = ip6_dst_mtu_maybe_forward(dst, true);
629 if (mtu < IPV6_MIN_MTU)
632 if (ip6_pkt_too_big(skb, mtu)) {
633 /* Again, force OUTPUT device used as source address */
635 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
636 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
637 __IP6_INC_STATS(net, ip6_dst_idev(dst),
638 IPSTATS_MIB_FRAGFAILS);
639 kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
643 if (skb_cow(skb, dst->dev->hard_header_len)) {
644 __IP6_INC_STATS(net, ip6_dst_idev(dst),
645 IPSTATS_MIB_OUTDISCARDS);
651 /* Mangling hops number delayed to point after skb COW */
655 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
656 net, NULL, skb, skb->dev, dst->dev,
660 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
661 SKB_DR_SET(reason, IP_INADDRERRORS);
663 kfree_skb_reason(skb, reason);
667 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
669 to->pkt_type = from->pkt_type;
670 to->priority = from->priority;
671 to->protocol = from->protocol;
673 skb_dst_set(to, dst_clone(skb_dst(from)));
675 to->mark = from->mark;
677 skb_copy_hash(to, from);
679 #ifdef CONFIG_NET_SCHED
680 to->tc_index = from->tc_index;
683 skb_ext_copy(to, from);
684 skb_copy_secmark(to, from);
687 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
688 u8 nexthdr, __be32 frag_id,
689 struct ip6_fraglist_iter *iter)
691 unsigned int first_len;
695 *prevhdr = NEXTHDR_FRAGMENT;
696 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
700 iter->frag = skb_shinfo(skb)->frag_list;
701 skb_frag_list_init(skb);
705 iter->frag_id = frag_id;
706 iter->nexthdr = nexthdr;
708 __skb_pull(skb, hlen);
709 fh = __skb_push(skb, sizeof(struct frag_hdr));
710 __skb_push(skb, hlen);
711 skb_reset_network_header(skb);
712 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
714 fh->nexthdr = nexthdr;
716 fh->frag_off = htons(IP6_MF);
717 fh->identification = frag_id;
719 first_len = skb_pagelen(skb);
720 skb->data_len = first_len - skb_headlen(skb);
721 skb->len = first_len;
722 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
726 EXPORT_SYMBOL(ip6_fraglist_init);
728 void ip6_fraglist_prepare(struct sk_buff *skb,
729 struct ip6_fraglist_iter *iter)
731 struct sk_buff *frag = iter->frag;
732 unsigned int hlen = iter->hlen;
735 frag->ip_summed = CHECKSUM_NONE;
736 skb_reset_transport_header(frag);
737 fh = __skb_push(frag, sizeof(struct frag_hdr));
738 __skb_push(frag, hlen);
739 skb_reset_network_header(frag);
740 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
741 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
742 fh->nexthdr = iter->nexthdr;
744 fh->frag_off = htons(iter->offset);
746 fh->frag_off |= htons(IP6_MF);
747 fh->identification = iter->frag_id;
748 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
749 ip6_copy_metadata(frag, skb);
751 EXPORT_SYMBOL(ip6_fraglist_prepare);
753 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
754 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
755 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
757 state->prevhdr = prevhdr;
758 state->nexthdr = nexthdr;
759 state->frag_id = frag_id;
764 state->left = skb->len - hlen; /* Space per frame */
765 state->ptr = hlen; /* Where to start from */
767 state->hroom = hdr_room;
768 state->troom = needed_tailroom;
772 EXPORT_SYMBOL(ip6_frag_init);
774 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
776 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
777 struct sk_buff *frag;
782 /* IF: it doesn't fit, use 'mtu' - the data space left */
783 if (len > state->mtu)
785 /* IF: we are not sending up to and including the packet end
786 then align the next start on an eight byte boundary */
787 if (len < state->left)
790 /* Allocate buffer */
791 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
792 state->hroom + state->troom, GFP_ATOMIC);
794 return ERR_PTR(-ENOMEM);
797 * Set up data on packet
800 ip6_copy_metadata(frag, skb);
801 skb_reserve(frag, state->hroom);
802 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
803 skb_reset_network_header(frag);
804 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
805 frag->transport_header = (frag->network_header + state->hlen +
806 sizeof(struct frag_hdr));
809 * Charge the memory for the fragment to any owner
813 skb_set_owner_w(frag, skb->sk);
816 * Copy the packet header into the new buffer.
818 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
820 fragnexthdr_offset = skb_network_header(frag);
821 fragnexthdr_offset += prevhdr - skb_network_header(skb);
822 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
825 * Build fragment header.
827 fh->nexthdr = state->nexthdr;
829 fh->identification = state->frag_id;
832 * Copy a block of the IP datagram.
834 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
838 fh->frag_off = htons(state->offset);
840 fh->frag_off |= htons(IP6_MF);
841 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
844 state->offset += len;
848 EXPORT_SYMBOL(ip6_frag_next);
850 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
851 int (*output)(struct net *, struct sock *, struct sk_buff *))
853 struct sk_buff *frag;
854 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
855 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
856 inet6_sk(skb->sk) : NULL;
857 bool mono_delivery_time = skb->mono_delivery_time;
858 struct ip6_frag_state state;
859 unsigned int mtu, hlen, nexthdr_offset;
860 ktime_t tstamp = skb->tstamp;
863 u8 *prevhdr, nexthdr = 0;
865 err = ip6_find_1stfragopt(skb, &prevhdr);
870 nexthdr_offset = prevhdr - skb_network_header(skb);
872 mtu = ip6_skb_dst_mtu(skb);
874 /* We must not fragment if the socket is set to force MTU discovery
875 * or if the skb it not generated by a local socket.
877 if (unlikely(!skb->ignore_df && skb->len > mtu))
880 if (IP6CB(skb)->frag_max_size) {
881 if (IP6CB(skb)->frag_max_size > mtu)
884 /* don't send fragments larger than what we received */
885 mtu = IP6CB(skb)->frag_max_size;
886 if (mtu < IPV6_MIN_MTU)
890 if (np && np->frag_size < mtu) {
894 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
896 mtu -= hlen + sizeof(struct frag_hdr);
898 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
899 &ipv6_hdr(skb)->saddr);
901 if (skb->ip_summed == CHECKSUM_PARTIAL &&
902 (err = skb_checksum_help(skb)))
905 prevhdr = skb_network_header(skb) + nexthdr_offset;
906 hroom = LL_RESERVED_SPACE(rt->dst.dev);
907 if (skb_has_frag_list(skb)) {
908 unsigned int first_len = skb_pagelen(skb);
909 struct ip6_fraglist_iter iter;
910 struct sk_buff *frag2;
912 if (first_len - hlen > mtu ||
913 ((first_len - hlen) & 7) ||
915 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
918 skb_walk_frags(skb, frag) {
919 /* Correct geometry. */
920 if (frag->len > mtu ||
921 ((frag->len & 7) && frag->next) ||
922 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
923 goto slow_path_clean;
925 /* Partially cloned skb? */
926 if (skb_shared(frag))
927 goto slow_path_clean;
932 frag->destructor = sock_wfree;
934 skb->truesize -= frag->truesize;
937 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
942 /* We prevent @rt from being freed. */
946 /* Prepare header of the next frame,
947 * before previous one went down. */
949 ip6_fraglist_prepare(skb, &iter);
951 skb_set_delivery_time(skb, tstamp, mono_delivery_time);
952 err = output(net, sk, skb);
954 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
955 IPSTATS_MIB_FRAGCREATES);
957 if (err || !iter.frag)
960 skb = ip6_fraglist_next(&iter);
966 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
967 IPSTATS_MIB_FRAGOKS);
972 kfree_skb_list(iter.frag);
974 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
975 IPSTATS_MIB_FRAGFAILS);
980 skb_walk_frags(skb, frag2) {
984 frag2->destructor = NULL;
985 skb->truesize += frag2->truesize;
991 * Fragment the datagram.
994 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
995 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
999 * Keep copying data until we run out.
1002 while (state.left > 0) {
1003 frag = ip6_frag_next(skb, &state);
1005 err = PTR_ERR(frag);
1010 * Put this fragment into the sending queue.
1012 skb_set_delivery_time(frag, tstamp, mono_delivery_time);
1013 err = output(net, sk, frag);
1017 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1018 IPSTATS_MIB_FRAGCREATES);
1020 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1021 IPSTATS_MIB_FRAGOKS);
1026 if (skb->sk && dst_allfrag(skb_dst(skb)))
1027 sk_gso_disable(skb->sk);
1029 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1033 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1034 IPSTATS_MIB_FRAGFAILS);
1039 static inline int ip6_rt_check(const struct rt6key *rt_key,
1040 const struct in6_addr *fl_addr,
1041 const struct in6_addr *addr_cache)
1043 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1044 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1047 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1048 struct dst_entry *dst,
1049 const struct flowi6 *fl6)
1051 struct ipv6_pinfo *np = inet6_sk(sk);
1052 struct rt6_info *rt;
1057 if (dst->ops->family != AF_INET6) {
1062 rt = (struct rt6_info *)dst;
1063 /* Yes, checking route validity in not connected
1064 * case is not very simple. Take into account,
1065 * that we do not support routing by source, TOS,
1066 * and MSG_DONTROUTE --ANK (980726)
1068 * 1. ip6_rt_check(): If route was host route,
1069 * check that cached destination is current.
1070 * If it is network route, we still may
1071 * check its validity using saved pointer
1072 * to the last used address: daddr_cache.
1073 * We do not want to save whole address now,
1074 * (because main consumer of this service
1075 * is tcp, which has not this problem),
1076 * so that the last trick works only on connected
1078 * 2. oif also should be the same.
1080 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1081 #ifdef CONFIG_IPV6_SUBTREES
1082 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1084 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1093 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1094 struct dst_entry **dst, struct flowi6 *fl6)
1096 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1097 struct neighbour *n;
1098 struct rt6_info *rt;
1103 /* The correct way to handle this would be to do
1104 * ip6_route_get_saddr, and then ip6_route_output; however,
1105 * the route-specific preferred source forces the
1106 * ip6_route_output call _before_ ip6_route_get_saddr.
1108 * In source specific routing (no src=any default route),
1109 * ip6_route_output will fail given src=any saddr, though, so
1110 * that's why we try it again later.
1112 if (ipv6_addr_any(&fl6->saddr)) {
1113 struct fib6_info *from;
1114 struct rt6_info *rt;
1116 *dst = ip6_route_output(net, sk, fl6);
1117 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1120 from = rt ? rcu_dereference(rt->from) : NULL;
1121 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1122 sk ? inet6_sk(sk)->srcprefs : 0,
1127 goto out_err_release;
1129 /* If we had an erroneous initial result, pretend it
1130 * never existed and let the SA-enabled version take
1133 if ((*dst)->error) {
1138 if (fl6->flowi6_oif)
1139 flags |= RT6_LOOKUP_F_IFACE;
1143 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1145 err = (*dst)->error;
1147 goto out_err_release;
1149 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1151 * Here if the dst entry we've looked up
1152 * has a neighbour entry that is in the INCOMPLETE
1153 * state and the src address from the flow is
1154 * marked as OPTIMISTIC, we release the found
1155 * dst entry and replace it instead with the
1156 * dst entry of the nexthop router
1158 rt = (struct rt6_info *) *dst;
1160 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1161 rt6_nexthop(rt, &fl6->daddr));
1162 err = n && !(READ_ONCE(n->nud_state) & NUD_VALID) ? -EINVAL : 0;
1166 struct inet6_ifaddr *ifp;
1167 struct flowi6 fl_gw6;
1170 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1173 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1179 * We need to get the dst entry for the
1180 * default router instead
1183 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1184 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1185 *dst = ip6_route_output(net, sk, &fl_gw6);
1186 err = (*dst)->error;
1188 goto out_err_release;
1192 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1193 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1194 err = -EAFNOSUPPORT;
1195 goto out_err_release;
1204 if (err == -ENETUNREACH)
1205 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1210 * ip6_dst_lookup - perform route lookup on flow
1211 * @net: Network namespace to perform lookup in
1212 * @sk: socket which provides route info
1213 * @dst: pointer to dst_entry * for result
1214 * @fl6: flow to lookup
1216 * This function performs a route lookup on the given flow.
1218 * It returns zero on success, or a standard errno code on error.
1220 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1224 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1226 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1229 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1230 * @net: Network namespace to perform lookup in
1231 * @sk: socket which provides route info
1232 * @fl6: flow to lookup
1233 * @final_dst: final destination address for ipsec lookup
1235 * This function performs a route lookup on the given flow.
1237 * It returns a valid dst pointer on success, or a pointer encoded
1240 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1241 const struct in6_addr *final_dst)
1243 struct dst_entry *dst = NULL;
1246 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1248 return ERR_PTR(err);
1250 fl6->daddr = *final_dst;
1252 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1254 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1257 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1258 * @sk: socket which provides the dst cache and route info
1259 * @fl6: flow to lookup
1260 * @final_dst: final destination address for ipsec lookup
1261 * @connected: whether @sk is connected or not
1263 * This function performs a route lookup on the given flow with the
1264 * possibility of using the cached route in the socket if it is valid.
1265 * It will take the socket dst lock when operating on the dst cache.
1266 * As a result, this function can only be used in process context.
1268 * In addition, for a connected socket, cache the dst in the socket
1269 * if the current cache is not valid.
1271 * It returns a valid dst pointer on success, or a pointer encoded
1274 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1275 const struct in6_addr *final_dst,
1278 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1280 dst = ip6_sk_dst_check(sk, dst, fl6);
1284 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1285 if (connected && !IS_ERR(dst))
1286 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1290 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1293 * ip6_dst_lookup_tunnel - perform route lookup on tunnel
1294 * @skb: Packet for which lookup is done
1295 * @dev: Tunnel device
1296 * @net: Network namespace of tunnel device
1297 * @sock: Socket which provides route info
1298 * @saddr: Memory to store the src ip address
1299 * @info: Tunnel information
1300 * @protocol: IP protocol
1301 * @use_cache: Flag to enable cache usage
1302 * This function performs a route lookup on a tunnel
1304 * It returns a valid dst pointer and stores src address to be used in
1305 * tunnel in param saddr on success, else a pointer encoded error code.
1308 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1309 struct net_device *dev,
1311 struct socket *sock,
1312 struct in6_addr *saddr,
1313 const struct ip_tunnel_info *info,
1317 struct dst_entry *dst = NULL;
1318 #ifdef CONFIG_DST_CACHE
1319 struct dst_cache *dst_cache;
1324 #ifdef CONFIG_DST_CACHE
1325 dst_cache = (struct dst_cache *)&info->dst_cache;
1327 dst = dst_cache_get_ip6(dst_cache, saddr);
1332 memset(&fl6, 0, sizeof(fl6));
1333 fl6.flowi6_mark = skb->mark;
1334 fl6.flowi6_proto = protocol;
1335 fl6.daddr = info->key.u.ipv6.dst;
1336 fl6.saddr = info->key.u.ipv6.src;
1337 prio = info->key.tos;
1338 fl6.flowlabel = ip6_make_flowinfo(prio, info->key.label);
1340 dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1343 netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1344 return ERR_PTR(-ENETUNREACH);
1346 if (dst->dev == dev) { /* is this necessary? */
1347 netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1349 return ERR_PTR(-ELOOP);
1351 #ifdef CONFIG_DST_CACHE
1353 dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1358 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1360 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1363 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1366 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1369 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1372 static void ip6_append_data_mtu(unsigned int *mtu,
1374 unsigned int fragheaderlen,
1375 struct sk_buff *skb,
1376 struct rt6_info *rt,
1377 unsigned int orig_mtu)
1379 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1381 /* first fragment, reserve header_len */
1382 *mtu = orig_mtu - rt->dst.header_len;
1386 * this fragment is not first, the headers
1387 * space is regarded as data space.
1391 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1392 + fragheaderlen - sizeof(struct frag_hdr);
1396 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1397 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1398 struct rt6_info *rt)
1400 struct ipv6_pinfo *np = inet6_sk(sk);
1402 struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1404 /* callers pass dst together with a reference, set it first so
1405 * ip6_cork_release() can put it down even in case of an error.
1407 cork->base.dst = &rt->dst;
1413 if (WARN_ON(v6_cork->opt))
1416 nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1417 if (unlikely(!nopt))
1420 nopt->tot_len = sizeof(*opt);
1421 nopt->opt_flen = opt->opt_flen;
1422 nopt->opt_nflen = opt->opt_nflen;
1424 nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1425 if (opt->dst0opt && !nopt->dst0opt)
1428 nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1429 if (opt->dst1opt && !nopt->dst1opt)
1432 nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1433 if (opt->hopopt && !nopt->hopopt)
1436 nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1437 if (opt->srcrt && !nopt->srcrt)
1440 /* need source address above miyazawa*/
1442 v6_cork->hop_limit = ipc6->hlimit;
1443 v6_cork->tclass = ipc6->tclass;
1444 if (rt->dst.flags & DST_XFRM_TUNNEL)
1445 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1446 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1448 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1449 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1450 if (np->frag_size < mtu) {
1452 mtu = np->frag_size;
1454 cork->base.fragsize = mtu;
1455 cork->base.gso_size = ipc6->gso_size;
1456 cork->base.tx_flags = 0;
1457 cork->base.mark = ipc6->sockc.mark;
1458 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1460 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1461 cork->base.flags |= IPCORK_ALLFRAG;
1462 cork->base.length = 0;
1464 cork->base.transmit_time = ipc6->sockc.transmit_time;
1469 static int __ip6_append_data(struct sock *sk,
1470 struct sk_buff_head *queue,
1471 struct inet_cork_full *cork_full,
1472 struct inet6_cork *v6_cork,
1473 struct page_frag *pfrag,
1474 int getfrag(void *from, char *to, int offset,
1475 int len, int odd, struct sk_buff *skb),
1476 void *from, size_t length, int transhdrlen,
1477 unsigned int flags, struct ipcm6_cookie *ipc6)
1479 struct sk_buff *skb, *skb_prev = NULL;
1480 struct inet_cork *cork = &cork_full->base;
1481 struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1482 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1483 struct ubuf_info *uarg = NULL;
1485 int dst_exthdrlen = 0;
1492 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1493 struct ipv6_txoptions *opt = v6_cork->opt;
1494 int csummode = CHECKSUM_NONE;
1495 unsigned int maxnonfragsize, headersize;
1496 unsigned int wmem_alloc_delta = 0;
1497 bool paged, extra_uref = false;
1499 skb = skb_peek_tail(queue);
1501 exthdrlen = opt ? opt->opt_flen : 0;
1502 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1505 paged = !!cork->gso_size;
1506 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1509 if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
1510 READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID)
1511 tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1513 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1515 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1516 (opt ? opt->opt_nflen : 0);
1518 headersize = sizeof(struct ipv6hdr) +
1519 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1520 (dst_allfrag(&rt->dst) ?
1521 sizeof(struct frag_hdr) : 0) +
1522 rt->rt6i_nfheader_len;
1524 if (mtu <= fragheaderlen ||
1525 ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1528 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1529 sizeof(struct frag_hdr);
1531 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1532 * the first fragment
1534 if (headersize + transhdrlen > mtu)
1537 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1538 (sk->sk_protocol == IPPROTO_UDP ||
1539 sk->sk_protocol == IPPROTO_ICMPV6 ||
1540 sk->sk_protocol == IPPROTO_RAW)) {
1541 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1542 sizeof(struct ipv6hdr));
1546 if (ip6_sk_ignore_df(sk))
1547 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1549 maxnonfragsize = mtu;
1551 if (cork->length + length > maxnonfragsize - headersize) {
1553 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1554 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1558 /* CHECKSUM_PARTIAL only with no extension headers and when
1559 * we are not going to fragment
1561 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1562 headersize == sizeof(struct ipv6hdr) &&
1563 length <= mtu - headersize &&
1564 (!(flags & MSG_MORE) || cork->gso_size) &&
1565 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1566 csummode = CHECKSUM_PARTIAL;
1568 if ((flags & MSG_ZEROCOPY) && length) {
1569 struct msghdr *msg = from;
1571 if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1572 if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1575 /* Leave uarg NULL if can't zerocopy, callers should
1576 * be able to handle it.
1578 if ((rt->dst.dev->features & NETIF_F_SG) &&
1579 csummode == CHECKSUM_PARTIAL) {
1582 uarg = msg->msg_ubuf;
1584 } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1585 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1588 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1589 if (rt->dst.dev->features & NETIF_F_SG &&
1590 csummode == CHECKSUM_PARTIAL) {
1594 uarg_to_msgzc(uarg)->zerocopy = 0;
1595 skb_zcopy_set(skb, uarg, &extra_uref);
1598 } else if ((flags & MSG_SPLICE_PAGES) && length) {
1599 if (inet_test_bit(HDRINCL, sk))
1601 if (rt->dst.dev->features & NETIF_F_SG &&
1602 getfrag == ip_generic_getfrag)
1603 /* We need an empty buffer to attach stuff to */
1606 flags &= ~MSG_SPLICE_PAGES;
1610 * Let's try using as much space as possible.
1611 * Use MTU if total length of the message fits into the MTU.
1612 * Otherwise, we need to reserve fragment header and
1613 * fragment alignment (= 8-15 octects, in total).
1615 * Note that we may need to "move" the data from the tail
1616 * of the buffer to the new fragment when we split
1619 * FIXME: It may be fragmented into multiple chunks
1620 * at once if non-fragmentable extension headers
1625 cork->length += length;
1629 while (length > 0) {
1630 /* Check if the remaining data fits into current packet. */
1631 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1633 copy = maxfraglen - skb->len;
1637 unsigned int datalen;
1638 unsigned int fraglen;
1639 unsigned int fraggap;
1640 unsigned int alloclen, alloc_extra;
1641 unsigned int pagedlen;
1643 /* There's no room in the current skb */
1645 fraggap = skb->len - maxfraglen;
1648 /* update mtu and maxfraglen if necessary */
1649 if (!skb || !skb_prev)
1650 ip6_append_data_mtu(&mtu, &maxfraglen,
1651 fragheaderlen, skb, rt,
1657 * If remaining data exceeds the mtu,
1658 * we know we need more fragment(s).
1660 datalen = length + fraggap;
1662 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1663 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1664 fraglen = datalen + fragheaderlen;
1667 alloc_extra = hh_len;
1668 alloc_extra += dst_exthdrlen;
1669 alloc_extra += rt->dst.trailer_len;
1671 /* We just reserve space for fragment header.
1672 * Note: this may be overallocation if the message
1673 * (without MSG_MORE) fits into the MTU.
1675 alloc_extra += sizeof(struct frag_hdr);
1677 if ((flags & MSG_MORE) &&
1678 !(rt->dst.dev->features&NETIF_F_SG))
1681 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1682 !(rt->dst.dev->features & NETIF_F_SG)))
1685 alloclen = fragheaderlen + transhdrlen;
1686 pagedlen = datalen - transhdrlen;
1688 alloclen += alloc_extra;
1690 if (datalen != length + fraggap) {
1692 * this is not the last fragment, the trailer
1693 * space is regarded as data space.
1695 datalen += rt->dst.trailer_len;
1698 fraglen = datalen + fragheaderlen;
1700 copy = datalen - transhdrlen - fraggap - pagedlen;
1701 /* [!] NOTE: copy may be negative if pagedlen>0
1702 * because then the equation may reduces to -fraggap.
1704 if (copy < 0 && !(flags & MSG_SPLICE_PAGES)) {
1709 skb = sock_alloc_send_skb(sk, alloclen,
1710 (flags & MSG_DONTWAIT), &err);
1713 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1715 skb = alloc_skb(alloclen,
1723 * Fill in the control structures
1725 skb->protocol = htons(ETH_P_IPV6);
1726 skb->ip_summed = csummode;
1728 /* reserve for fragmentation and ipsec header */
1729 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1733 * Find where to start putting bytes
1735 data = skb_put(skb, fraglen - pagedlen);
1736 skb_set_network_header(skb, exthdrlen);
1737 data += fragheaderlen;
1738 skb->transport_header = (skb->network_header +
1741 skb->csum = skb_copy_and_csum_bits(
1742 skb_prev, maxfraglen,
1743 data + transhdrlen, fraggap);
1744 skb_prev->csum = csum_sub(skb_prev->csum,
1747 pskb_trim_unique(skb_prev, maxfraglen);
1750 getfrag(from, data + transhdrlen, offset,
1751 copy, fraggap, skb) < 0) {
1755 } else if (flags & MSG_SPLICE_PAGES) {
1760 length -= copy + transhdrlen;
1765 /* Only the initial fragment is time stamped */
1766 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1768 skb_shinfo(skb)->tskey = tskey;
1770 skb_zcopy_set(skb, uarg, &extra_uref);
1772 if ((flags & MSG_CONFIRM) && !skb_prev)
1773 skb_set_dst_pending_confirm(skb, 1);
1776 * Put the packet on the pending queue
1778 if (!skb->destructor) {
1779 skb->destructor = sock_wfree;
1781 wmem_alloc_delta += skb->truesize;
1783 __skb_queue_tail(queue, skb);
1790 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1791 skb_tailroom(skb) >= copy) {
1795 if (getfrag(from, skb_put(skb, copy),
1796 offset, copy, off, skb) < 0) {
1797 __skb_trim(skb, off);
1801 } else if (flags & MSG_SPLICE_PAGES) {
1802 struct msghdr *msg = from;
1805 if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1808 err = skb_splice_from_iter(skb, &msg->msg_iter, copy,
1813 wmem_alloc_delta += copy;
1815 int i = skb_shinfo(skb)->nr_frags;
1818 if (!sk_page_frag_refill(sk, pfrag))
1821 skb_zcopy_downgrade_managed(skb);
1822 if (!skb_can_coalesce(skb, i, pfrag->page,
1825 if (i == MAX_SKB_FRAGS)
1828 __skb_fill_page_desc(skb, i, pfrag->page,
1830 skb_shinfo(skb)->nr_frags = ++i;
1831 get_page(pfrag->page);
1833 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1835 page_address(pfrag->page) + pfrag->offset,
1836 offset, copy, skb->len, skb) < 0)
1839 pfrag->offset += copy;
1840 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1842 skb->data_len += copy;
1843 skb->truesize += copy;
1844 wmem_alloc_delta += copy;
1846 err = skb_zerocopy_iter_dgram(skb, from, copy);
1854 if (wmem_alloc_delta)
1855 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1861 net_zcopy_put_abort(uarg, extra_uref);
1862 cork->length -= length;
1863 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1864 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1868 int ip6_append_data(struct sock *sk,
1869 int getfrag(void *from, char *to, int offset, int len,
1870 int odd, struct sk_buff *skb),
1871 void *from, size_t length, int transhdrlen,
1872 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1873 struct rt6_info *rt, unsigned int flags)
1875 struct inet_sock *inet = inet_sk(sk);
1876 struct ipv6_pinfo *np = inet6_sk(sk);
1880 if (flags&MSG_PROBE)
1882 if (skb_queue_empty(&sk->sk_write_queue)) {
1887 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1892 inet->cork.fl.u.ip6 = *fl6;
1893 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1894 length += exthdrlen;
1895 transhdrlen += exthdrlen;
1900 return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1901 &np->cork, sk_page_frag(sk), getfrag,
1902 from, length, transhdrlen, flags, ipc6);
1904 EXPORT_SYMBOL_GPL(ip6_append_data);
1906 static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1908 struct dst_entry *dst = cork->base.dst;
1910 cork->base.dst = NULL;
1911 cork->base.flags &= ~IPCORK_ALLFRAG;
1912 skb_dst_set(skb, dst);
1915 static void ip6_cork_release(struct inet_cork_full *cork,
1916 struct inet6_cork *v6_cork)
1919 struct ipv6_txoptions *opt = v6_cork->opt;
1921 kfree(opt->dst0opt);
1922 kfree(opt->dst1opt);
1926 v6_cork->opt = NULL;
1929 if (cork->base.dst) {
1930 dst_release(cork->base.dst);
1931 cork->base.dst = NULL;
1932 cork->base.flags &= ~IPCORK_ALLFRAG;
1936 struct sk_buff *__ip6_make_skb(struct sock *sk,
1937 struct sk_buff_head *queue,
1938 struct inet_cork_full *cork,
1939 struct inet6_cork *v6_cork)
1941 struct sk_buff *skb, *tmp_skb;
1942 struct sk_buff **tail_skb;
1943 struct in6_addr *final_dst;
1944 struct ipv6_pinfo *np = inet6_sk(sk);
1945 struct net *net = sock_net(sk);
1946 struct ipv6hdr *hdr;
1947 struct ipv6_txoptions *opt = v6_cork->opt;
1948 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1949 struct flowi6 *fl6 = &cork->fl.u.ip6;
1950 unsigned char proto = fl6->flowi6_proto;
1952 skb = __skb_dequeue(queue);
1955 tail_skb = &(skb_shinfo(skb)->frag_list);
1957 /* move skb->data to ip header from ext header */
1958 if (skb->data < skb_network_header(skb))
1959 __skb_pull(skb, skb_network_offset(skb));
1960 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1961 __skb_pull(tmp_skb, skb_network_header_len(skb));
1962 *tail_skb = tmp_skb;
1963 tail_skb = &(tmp_skb->next);
1964 skb->len += tmp_skb->len;
1965 skb->data_len += tmp_skb->len;
1966 skb->truesize += tmp_skb->truesize;
1967 tmp_skb->destructor = NULL;
1971 /* Allow local fragmentation. */
1972 skb->ignore_df = ip6_sk_ignore_df(sk);
1973 __skb_pull(skb, skb_network_header_len(skb));
1975 final_dst = &fl6->daddr;
1976 if (opt && opt->opt_flen)
1977 ipv6_push_frag_opts(skb, opt, &proto);
1978 if (opt && opt->opt_nflen)
1979 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1981 skb_push(skb, sizeof(struct ipv6hdr));
1982 skb_reset_network_header(skb);
1983 hdr = ipv6_hdr(skb);
1985 ip6_flow_hdr(hdr, v6_cork->tclass,
1986 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1987 ip6_autoflowlabel(net, np), fl6));
1988 hdr->hop_limit = v6_cork->hop_limit;
1989 hdr->nexthdr = proto;
1990 hdr->saddr = fl6->saddr;
1991 hdr->daddr = *final_dst;
1993 skb->priority = sk->sk_priority;
1994 skb->mark = cork->base.mark;
1995 skb->tstamp = cork->base.transmit_time;
1997 ip6_cork_steal_dst(skb, cork);
1998 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1999 if (proto == IPPROTO_ICMPV6) {
2000 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
2003 if (sk->sk_socket->type == SOCK_RAW &&
2004 !inet_test_bit(HDRINCL, sk))
2005 icmp6_type = fl6->fl6_icmp_type;
2007 icmp6_type = icmp6_hdr(skb)->icmp6_type;
2008 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_type);
2009 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
2012 ip6_cork_release(cork, v6_cork);
2017 int ip6_send_skb(struct sk_buff *skb)
2019 struct net *net = sock_net(skb->sk);
2020 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
2023 err = ip6_local_out(net, skb->sk, skb);
2026 err = net_xmit_errno(err);
2028 IP6_INC_STATS(net, rt->rt6i_idev,
2029 IPSTATS_MIB_OUTDISCARDS);
2035 int ip6_push_pending_frames(struct sock *sk)
2037 struct sk_buff *skb;
2039 skb = ip6_finish_skb(sk);
2043 return ip6_send_skb(skb);
2045 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
2047 static void __ip6_flush_pending_frames(struct sock *sk,
2048 struct sk_buff_head *queue,
2049 struct inet_cork_full *cork,
2050 struct inet6_cork *v6_cork)
2052 struct sk_buff *skb;
2054 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
2056 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
2057 IPSTATS_MIB_OUTDISCARDS);
2061 ip6_cork_release(cork, v6_cork);
2064 void ip6_flush_pending_frames(struct sock *sk)
2066 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
2067 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
2069 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
2071 struct sk_buff *ip6_make_skb(struct sock *sk,
2072 int getfrag(void *from, char *to, int offset,
2073 int len, int odd, struct sk_buff *skb),
2074 void *from, size_t length, int transhdrlen,
2075 struct ipcm6_cookie *ipc6, struct rt6_info *rt,
2076 unsigned int flags, struct inet_cork_full *cork)
2078 struct inet6_cork v6_cork;
2079 struct sk_buff_head queue;
2080 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
2083 if (flags & MSG_PROBE) {
2084 dst_release(&rt->dst);
2088 __skb_queue_head_init(&queue);
2090 cork->base.flags = 0;
2091 cork->base.addr = 0;
2092 cork->base.opt = NULL;
2094 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2096 ip6_cork_release(cork, &v6_cork);
2097 return ERR_PTR(err);
2099 if (ipc6->dontfrag < 0)
2100 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2102 err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2103 ¤t->task_frag, getfrag, from,
2104 length + exthdrlen, transhdrlen + exthdrlen,
2107 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2108 return ERR_PTR(err);
2111 return __ip6_make_skb(sk, &queue, cork, &v6_cork);