11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * INET An implementation of the TCP/IP protocol suite for the LINUX 31da177e4SLinus Torvalds * operating system. INET is implemented using the BSD Socket 41da177e4SLinus Torvalds * interface as the means of communication with the user level. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * The Internet Protocol (IP) output module. 71da177e4SLinus Torvalds * 802c30a84SJesper Juhl * Authors: Ross Biro 91da177e4SLinus Torvalds * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 101da177e4SLinus Torvalds * Donald Becker, <becker@super.org> 111da177e4SLinus Torvalds * Alan Cox, <Alan.Cox@linux.org> 121da177e4SLinus Torvalds * Richard Underwood 131da177e4SLinus Torvalds * Stefan Becker, <stefanb@yello.ping.de> 141da177e4SLinus Torvalds * Jorge Cwik, <jorge@laser.satlink.net> 151da177e4SLinus Torvalds * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 161da177e4SLinus Torvalds * Hirokazu Takahashi, <taka@valinux.co.jp> 171da177e4SLinus Torvalds * 181da177e4SLinus Torvalds * See ip_input.c for original log 191da177e4SLinus Torvalds * 201da177e4SLinus Torvalds * Fixes: 211da177e4SLinus Torvalds * Alan Cox : Missing nonblock feature in ip_build_xmit. 221da177e4SLinus Torvalds * Mike Kilburn : htons() missing in ip_build_xmit. 231da177e4SLinus Torvalds * Bradford Johnson: Fix faulty handling of some frames when 241da177e4SLinus Torvalds * no route is found. 251da177e4SLinus Torvalds * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit 261da177e4SLinus Torvalds * (in case if packet not accepted by 271da177e4SLinus Torvalds * output firewall rules) 281da177e4SLinus Torvalds * Mike McLagan : Routing by source 291da177e4SLinus Torvalds * Alexey Kuznetsov: use new route cache 301da177e4SLinus Torvalds * Andi Kleen: Fix broken PMTU recovery and remove 311da177e4SLinus Torvalds * some redundant tests. 321da177e4SLinus Torvalds * Vitaly E. Lavrov : Transparent proxy revived after year coma. 331da177e4SLinus Torvalds * Andi Kleen : Replace ip_reply with ip_send_reply. 341da177e4SLinus Torvalds * Andi Kleen : Split fast and slow ip_build_xmit path 351da177e4SLinus Torvalds * for decreased register pressure on x86 361da177e4SLinus Torvalds * and more readibility. 371da177e4SLinus Torvalds * Marc Boucher : When call_out_firewall returns FW_QUEUE, 381da177e4SLinus Torvalds * silently drop skb instead of failing with -EPERM. 391da177e4SLinus Torvalds * Detlev Wengorz : Copy protocol for fragments. 401da177e4SLinus Torvalds * Hirokazu Takahashi: HW checksumming for outgoing UDP 411da177e4SLinus Torvalds * datagrams. 421da177e4SLinus Torvalds * Hirokazu Takahashi: sendfile() on UDP works now. 431da177e4SLinus Torvalds */ 441da177e4SLinus Torvalds 451da177e4SLinus Torvalds #include <asm/uaccess.h> 461da177e4SLinus Torvalds #include <asm/system.h> 471da177e4SLinus Torvalds #include <linux/module.h> 481da177e4SLinus Torvalds #include <linux/types.h> 491da177e4SLinus Torvalds #include <linux/kernel.h> 501da177e4SLinus Torvalds #include <linux/mm.h> 511da177e4SLinus Torvalds #include <linux/string.h> 521da177e4SLinus Torvalds #include <linux/errno.h> 53a1f8e7f7SAl Viro #include <linux/highmem.h> 545a0e3ad6STejun Heo #include <linux/slab.h> 551da177e4SLinus Torvalds 561da177e4SLinus Torvalds #include <linux/socket.h> 571da177e4SLinus Torvalds #include <linux/sockios.h> 581da177e4SLinus Torvalds #include <linux/in.h> 591da177e4SLinus Torvalds #include <linux/inet.h> 601da177e4SLinus Torvalds #include <linux/netdevice.h> 611da177e4SLinus Torvalds #include <linux/etherdevice.h> 621da177e4SLinus Torvalds #include <linux/proc_fs.h> 631da177e4SLinus Torvalds #include <linux/stat.h> 641da177e4SLinus Torvalds #include <linux/init.h> 651da177e4SLinus Torvalds 661da177e4SLinus Torvalds #include <net/snmp.h> 671da177e4SLinus Torvalds #include <net/ip.h> 681da177e4SLinus Torvalds #include <net/protocol.h> 691da177e4SLinus Torvalds #include <net/route.h> 70cfacb057SPatrick McHardy #include <net/xfrm.h> 711da177e4SLinus Torvalds #include <linux/skbuff.h> 721da177e4SLinus Torvalds #include <net/sock.h> 731da177e4SLinus Torvalds #include <net/arp.h> 741da177e4SLinus Torvalds #include <net/icmp.h> 751da177e4SLinus Torvalds #include <net/checksum.h> 761da177e4SLinus Torvalds #include <net/inetpeer.h> 771da177e4SLinus Torvalds #include <linux/igmp.h> 781da177e4SLinus Torvalds #include <linux/netfilter_ipv4.h> 791da177e4SLinus Torvalds #include <linux/netfilter_bridge.h> 801da177e4SLinus Torvalds #include <linux/mroute.h> 811da177e4SLinus Torvalds #include <linux/netlink.h> 826cbb0df7SArnaldo Carvalho de Melo #include <linux/tcp.h> 831da177e4SLinus Torvalds 84ab32ea5dSBrian Haley int sysctl_ip_default_ttl __read_mostly = IPDEFTTL; 85323e126fSDavid S. Miller EXPORT_SYMBOL(sysctl_ip_default_ttl); 861da177e4SLinus Torvalds 871da177e4SLinus Torvalds /* Generate a checksum for an outgoing IP datagram. */ 881da177e4SLinus Torvalds __inline__ void ip_send_check(struct iphdr *iph) 891da177e4SLinus Torvalds { 901da177e4SLinus Torvalds iph->check = 0; 911da177e4SLinus Torvalds iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 921da177e4SLinus Torvalds } 934bc2f18bSEric Dumazet EXPORT_SYMBOL(ip_send_check); 941da177e4SLinus Torvalds 95c439cb2eSHerbert Xu int __ip_local_out(struct sk_buff *skb) 96c439cb2eSHerbert Xu { 97c439cb2eSHerbert Xu struct iphdr *iph = ip_hdr(skb); 98c439cb2eSHerbert Xu 99c439cb2eSHerbert Xu iph->tot_len = htons(skb->len); 100c439cb2eSHerbert Xu ip_send_check(iph); 1019bbc768aSJan Engelhardt return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL, 1029bbc768aSJan Engelhardt skb_dst(skb)->dev, dst_output); 103c439cb2eSHerbert Xu } 104c439cb2eSHerbert Xu 105c439cb2eSHerbert Xu int ip_local_out(struct sk_buff *skb) 106c439cb2eSHerbert Xu { 107c439cb2eSHerbert Xu int err; 108c439cb2eSHerbert Xu 109c439cb2eSHerbert Xu err = __ip_local_out(skb); 110c439cb2eSHerbert Xu if (likely(err == 1)) 111c439cb2eSHerbert Xu err = dst_output(skb); 112c439cb2eSHerbert Xu 113c439cb2eSHerbert Xu return err; 114c439cb2eSHerbert Xu } 115c439cb2eSHerbert Xu EXPORT_SYMBOL_GPL(ip_local_out); 116c439cb2eSHerbert Xu 1171da177e4SLinus Torvalds /* dev_loopback_xmit for use with netfilter. */ 1181da177e4SLinus Torvalds static int ip_dev_loopback_xmit(struct sk_buff *newskb) 1191da177e4SLinus Torvalds { 120459a98edSArnaldo Carvalho de Melo skb_reset_mac_header(newskb); 121bbe735e4SArnaldo Carvalho de Melo __skb_pull(newskb, skb_network_offset(newskb)); 1221da177e4SLinus Torvalds newskb->pkt_type = PACKET_LOOPBACK; 1231da177e4SLinus Torvalds newskb->ip_summed = CHECKSUM_UNNECESSARY; 124adf30907SEric Dumazet WARN_ON(!skb_dst(newskb)); 125e30b38c2SEric Dumazet netif_rx_ni(newskb); 1261da177e4SLinus Torvalds return 0; 1271da177e4SLinus Torvalds } 1281da177e4SLinus Torvalds 1291da177e4SLinus Torvalds static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) 1301da177e4SLinus Torvalds { 1311da177e4SLinus Torvalds int ttl = inet->uc_ttl; 1321da177e4SLinus Torvalds 1331da177e4SLinus Torvalds if (ttl < 0) 134323e126fSDavid S. Miller ttl = ip4_dst_hoplimit(dst); 1351da177e4SLinus Torvalds return ttl; 1361da177e4SLinus Torvalds } 1371da177e4SLinus Torvalds 1381da177e4SLinus Torvalds /* 1391da177e4SLinus Torvalds * Add an ip header to a skbuff and send it out. 1401da177e4SLinus Torvalds * 1411da177e4SLinus Torvalds */ 1421da177e4SLinus Torvalds int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk, 143f6d8bd05SEric Dumazet __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) 1441da177e4SLinus Torvalds { 1451da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 146511c3f92SEric Dumazet struct rtable *rt = skb_rtable(skb); 1471da177e4SLinus Torvalds struct iphdr *iph; 1481da177e4SLinus Torvalds 1491da177e4SLinus Torvalds /* Build the IP header. */ 150f6d8bd05SEric Dumazet skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); 1518856dfa3SArnaldo Carvalho de Melo skb_reset_network_header(skb); 152eddc9ec5SArnaldo Carvalho de Melo iph = ip_hdr(skb); 1531da177e4SLinus Torvalds iph->version = 4; 1541da177e4SLinus Torvalds iph->ihl = 5; 1551da177e4SLinus Torvalds iph->tos = inet->tos; 156d8d1f30bSChangli Gao if (ip_dont_fragment(sk, &rt->dst)) 1571da177e4SLinus Torvalds iph->frag_off = htons(IP_DF); 1581da177e4SLinus Torvalds else 1591da177e4SLinus Torvalds iph->frag_off = 0; 160d8d1f30bSChangli Gao iph->ttl = ip_select_ttl(inet, &rt->dst); 161dd927a26SDavid S. Miller iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); 162dd927a26SDavid S. Miller iph->saddr = saddr; 1631da177e4SLinus Torvalds iph->protocol = sk->sk_protocol; 164d8d1f30bSChangli Gao ip_select_ident(iph, &rt->dst, sk); 1651da177e4SLinus Torvalds 166f6d8bd05SEric Dumazet if (opt && opt->opt.optlen) { 167f6d8bd05SEric Dumazet iph->ihl += opt->opt.optlen>>2; 168f6d8bd05SEric Dumazet ip_options_build(skb, &opt->opt, daddr, rt, 0); 1691da177e4SLinus Torvalds } 1701da177e4SLinus Torvalds 1711da177e4SLinus Torvalds skb->priority = sk->sk_priority; 1724a19ec58SLaszlo Attila Toth skb->mark = sk->sk_mark; 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds /* Send it out. */ 175c439cb2eSHerbert Xu return ip_local_out(skb); 1761da177e4SLinus Torvalds } 177d8c97a94SArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); 178d8c97a94SArnaldo Carvalho de Melo 1791da177e4SLinus Torvalds static inline int ip_finish_output2(struct sk_buff *skb) 1801da177e4SLinus Torvalds { 181adf30907SEric Dumazet struct dst_entry *dst = skb_dst(skb); 18280787ebcSMitsuru Chinen struct rtable *rt = (struct rtable *)dst; 1831da177e4SLinus Torvalds struct net_device *dev = dst->dev; 184c2636b4dSChuck Lever unsigned int hh_len = LL_RESERVED_SPACE(dev); 185*f6b72b62SDavid S. Miller struct neighbour *neigh; 1861da177e4SLinus Torvalds 187edf391ffSNeil Horman if (rt->rt_type == RTN_MULTICAST) { 188edf391ffSNeil Horman IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len); 189edf391ffSNeil Horman } else if (rt->rt_type == RTN_BROADCAST) 190edf391ffSNeil Horman IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len); 19180787ebcSMitsuru Chinen 1921da177e4SLinus Torvalds /* Be paranoid, rather than too clever. */ 1933b04dddeSStephen Hemminger if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 1941da177e4SLinus Torvalds struct sk_buff *skb2; 1951da177e4SLinus Torvalds 1961da177e4SLinus Torvalds skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 1971da177e4SLinus Torvalds if (skb2 == NULL) { 1981da177e4SLinus Torvalds kfree_skb(skb); 1991da177e4SLinus Torvalds return -ENOMEM; 2001da177e4SLinus Torvalds } 2011da177e4SLinus Torvalds if (skb->sk) 2021da177e4SLinus Torvalds skb_set_owner_w(skb2, skb->sk); 2031da177e4SLinus Torvalds kfree_skb(skb); 2041da177e4SLinus Torvalds skb = skb2; 2051da177e4SLinus Torvalds } 2061da177e4SLinus Torvalds 207*f6b72b62SDavid S. Miller neigh = dst->neighbour; 208*f6b72b62SDavid S. Miller if (neigh) { 209*f6b72b62SDavid S. Miller struct hh_cache *hh = &neigh->hh; 210*f6b72b62SDavid S. Miller if (hh->hh_len) 211*f6b72b62SDavid S. Miller return neigh_hh_output(hh, skb); 212*f6b72b62SDavid S. Miller else 2131da177e4SLinus Torvalds return dst->neighbour->output(skb); 214*f6b72b62SDavid S. Miller } 2151da177e4SLinus Torvalds if (net_ratelimit()) 2161da177e4SLinus Torvalds printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n"); 2171da177e4SLinus Torvalds kfree_skb(skb); 2181da177e4SLinus Torvalds return -EINVAL; 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds 221628a5c56SJohn Heffner static inline int ip_skb_dst_mtu(struct sk_buff *skb) 222628a5c56SJohn Heffner { 223628a5c56SJohn Heffner struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL; 224628a5c56SJohn Heffner 225628a5c56SJohn Heffner return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ? 226adf30907SEric Dumazet skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb)); 227628a5c56SJohn Heffner } 228628a5c56SJohn Heffner 229861d0486SPatrick McHardy static int ip_finish_output(struct sk_buff *skb) 2301da177e4SLinus Torvalds { 2315c901daaSPatrick McHardy #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 2325c901daaSPatrick McHardy /* Policy lookup after SNAT yielded a new policy */ 233adf30907SEric Dumazet if (skb_dst(skb)->xfrm != NULL) { 23448d5cad8SPatrick McHardy IPCB(skb)->flags |= IPSKB_REROUTED; 23548d5cad8SPatrick McHardy return dst_output(skb); 23648d5cad8SPatrick McHardy } 2375c901daaSPatrick McHardy #endif 238628a5c56SJohn Heffner if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb)) 2391bd9bef6SPatrick McHardy return ip_fragment(skb, ip_finish_output2); 2401bd9bef6SPatrick McHardy else 2411bd9bef6SPatrick McHardy return ip_finish_output2(skb); 2421da177e4SLinus Torvalds } 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds int ip_mc_output(struct sk_buff *skb) 2451da177e4SLinus Torvalds { 2461da177e4SLinus Torvalds struct sock *sk = skb->sk; 247511c3f92SEric Dumazet struct rtable *rt = skb_rtable(skb); 248d8d1f30bSChangli Gao struct net_device *dev = rt->dst.dev; 2491da177e4SLinus Torvalds 2501da177e4SLinus Torvalds /* 2511da177e4SLinus Torvalds * If the indicated interface is up and running, send the packet. 2521da177e4SLinus Torvalds */ 253edf391ffSNeil Horman IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); 2541da177e4SLinus Torvalds 2551da177e4SLinus Torvalds skb->dev = dev; 2561da177e4SLinus Torvalds skb->protocol = htons(ETH_P_IP); 2571da177e4SLinus Torvalds 2581da177e4SLinus Torvalds /* 2591da177e4SLinus Torvalds * Multicasts are looped back for other local users 2601da177e4SLinus Torvalds */ 2611da177e4SLinus Torvalds 2621da177e4SLinus Torvalds if (rt->rt_flags&RTCF_MULTICAST) { 2637ad6848cSOctavian Purdila if (sk_mc_loop(sk) 2641da177e4SLinus Torvalds #ifdef CONFIG_IP_MROUTE 2651da177e4SLinus Torvalds /* Small optimization: do not loopback not local frames, 2661da177e4SLinus Torvalds which returned after forwarding; they will be dropped 2671da177e4SLinus Torvalds by ip_mr_input in any case. 2681da177e4SLinus Torvalds Note, that local frames are looped back to be delivered 2691da177e4SLinus Torvalds to local recipients. 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds This check is duplicated in ip_mr_input at the moment. 2721da177e4SLinus Torvalds */ 2739d4fb27dSJoe Perches && 2749d4fb27dSJoe Perches ((rt->rt_flags & RTCF_LOCAL) || 2759d4fb27dSJoe Perches !(IPCB(skb)->flags & IPSKB_FORWARDED)) 2761da177e4SLinus Torvalds #endif 2771da177e4SLinus Torvalds ) { 2781da177e4SLinus Torvalds struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 2791da177e4SLinus Torvalds if (newskb) 2809bbc768aSJan Engelhardt NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 2819bbc768aSJan Engelhardt newskb, NULL, newskb->dev, 2821da177e4SLinus Torvalds ip_dev_loopback_xmit); 2831da177e4SLinus Torvalds } 2841da177e4SLinus Torvalds 2851da177e4SLinus Torvalds /* Multicasts with ttl 0 must not go beyond the host */ 2861da177e4SLinus Torvalds 287eddc9ec5SArnaldo Carvalho de Melo if (ip_hdr(skb)->ttl == 0) { 2881da177e4SLinus Torvalds kfree_skb(skb); 2891da177e4SLinus Torvalds return 0; 2901da177e4SLinus Torvalds } 2911da177e4SLinus Torvalds } 2921da177e4SLinus Torvalds 2931da177e4SLinus Torvalds if (rt->rt_flags&RTCF_BROADCAST) { 2941da177e4SLinus Torvalds struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 2951da177e4SLinus Torvalds if (newskb) 2969bbc768aSJan Engelhardt NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb, 2979bbc768aSJan Engelhardt NULL, newskb->dev, ip_dev_loopback_xmit); 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds 3009bbc768aSJan Engelhardt return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, 3019bbc768aSJan Engelhardt skb->dev, ip_finish_output, 30248d5cad8SPatrick McHardy !(IPCB(skb)->flags & IPSKB_REROUTED)); 3031da177e4SLinus Torvalds } 3041da177e4SLinus Torvalds 3051da177e4SLinus Torvalds int ip_output(struct sk_buff *skb) 3061da177e4SLinus Torvalds { 307adf30907SEric Dumazet struct net_device *dev = skb_dst(skb)->dev; 3081bd9bef6SPatrick McHardy 309edf391ffSNeil Horman IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); 3101da177e4SLinus Torvalds 3111bd9bef6SPatrick McHardy skb->dev = dev; 3121bd9bef6SPatrick McHardy skb->protocol = htons(ETH_P_IP); 3131bd9bef6SPatrick McHardy 3149bbc768aSJan Engelhardt return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev, 31548d5cad8SPatrick McHardy ip_finish_output, 31648d5cad8SPatrick McHardy !(IPCB(skb)->flags & IPSKB_REROUTED)); 3171da177e4SLinus Torvalds } 3181da177e4SLinus Torvalds 319d9d8da80SDavid S. Miller int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl) 3201da177e4SLinus Torvalds { 321e89862f4SDavid S. Miller struct sock *sk = skb->sk; 3221da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 323f6d8bd05SEric Dumazet struct ip_options_rcu *inet_opt; 324b57ae01aSDavid S. Miller struct flowi4 *fl4; 3251da177e4SLinus Torvalds struct rtable *rt; 3261da177e4SLinus Torvalds struct iphdr *iph; 327ab6e3febSEric Dumazet int res; 3281da177e4SLinus Torvalds 3291da177e4SLinus Torvalds /* Skip all of this if the packet is already routed, 3301da177e4SLinus Torvalds * f.e. by something like SCTP. 3311da177e4SLinus Torvalds */ 332ab6e3febSEric Dumazet rcu_read_lock(); 333f6d8bd05SEric Dumazet inet_opt = rcu_dereference(inet->inet_opt); 334ea4fc0d6SDavid S. Miller fl4 = &fl->u.ip4; 335511c3f92SEric Dumazet rt = skb_rtable(skb); 3361da177e4SLinus Torvalds if (rt != NULL) 3371da177e4SLinus Torvalds goto packet_routed; 3381da177e4SLinus Torvalds 3391da177e4SLinus Torvalds /* Make sure we can route this packet. */ 3401da177e4SLinus Torvalds rt = (struct rtable *)__sk_dst_check(sk, 0); 3411da177e4SLinus Torvalds if (rt == NULL) { 3423ca3c68eSAl Viro __be32 daddr; 3431da177e4SLinus Torvalds 3441da177e4SLinus Torvalds /* Use correct destination address if we have options. */ 345c720c7e8SEric Dumazet daddr = inet->inet_daddr; 346f6d8bd05SEric Dumazet if (inet_opt && inet_opt->opt.srr) 347f6d8bd05SEric Dumazet daddr = inet_opt->opt.faddr; 3481da177e4SLinus Torvalds 3491da177e4SLinus Torvalds /* If this fails, retransmit mechanism of transport layer will 3501da177e4SLinus Torvalds * keep trying until route appears or the connection times 3511da177e4SLinus Torvalds * itself out. 3521da177e4SLinus Torvalds */ 353b57ae01aSDavid S. Miller rt = ip_route_output_ports(sock_net(sk), fl4, sk, 35478fbfd8aSDavid S. Miller daddr, inet->inet_saddr, 35578fbfd8aSDavid S. Miller inet->inet_dport, 35678fbfd8aSDavid S. Miller inet->inet_sport, 35778fbfd8aSDavid S. Miller sk->sk_protocol, 35878fbfd8aSDavid S. Miller RT_CONN_FLAGS(sk), 35978fbfd8aSDavid S. Miller sk->sk_bound_dev_if); 360b23dd4feSDavid S. Miller if (IS_ERR(rt)) 3611da177e4SLinus Torvalds goto no_route; 362d8d1f30bSChangli Gao sk_setup_caps(sk, &rt->dst); 3631da177e4SLinus Torvalds } 364d8d1f30bSChangli Gao skb_dst_set_noref(skb, &rt->dst); 3651da177e4SLinus Torvalds 3661da177e4SLinus Torvalds packet_routed: 367ea4fc0d6SDavid S. Miller if (inet_opt && inet_opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway) 3681da177e4SLinus Torvalds goto no_route; 3691da177e4SLinus Torvalds 3701da177e4SLinus Torvalds /* OK, we know where to send it, allocate and build IP header. */ 371f6d8bd05SEric Dumazet skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 3728856dfa3SArnaldo Carvalho de Melo skb_reset_network_header(skb); 373eddc9ec5SArnaldo Carvalho de Melo iph = ip_hdr(skb); 374714e85beSAl Viro *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); 375d8d1f30bSChangli Gao if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df) 3761da177e4SLinus Torvalds iph->frag_off = htons(IP_DF); 3771da177e4SLinus Torvalds else 3781da177e4SLinus Torvalds iph->frag_off = 0; 379d8d1f30bSChangli Gao iph->ttl = ip_select_ttl(inet, &rt->dst); 3801da177e4SLinus Torvalds iph->protocol = sk->sk_protocol; 381ea4fc0d6SDavid S. Miller iph->saddr = fl4->saddr; 382ea4fc0d6SDavid S. Miller iph->daddr = fl4->daddr; 3831da177e4SLinus Torvalds /* Transport layer set skb->h.foo itself. */ 3841da177e4SLinus Torvalds 385f6d8bd05SEric Dumazet if (inet_opt && inet_opt->opt.optlen) { 386f6d8bd05SEric Dumazet iph->ihl += inet_opt->opt.optlen >> 2; 387f6d8bd05SEric Dumazet ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 3881da177e4SLinus Torvalds } 3891da177e4SLinus Torvalds 390d8d1f30bSChangli Gao ip_select_ident_more(iph, &rt->dst, sk, 3917967168cSHerbert Xu (skb_shinfo(skb)->gso_segs ?: 1) - 1); 3921da177e4SLinus Torvalds 3931da177e4SLinus Torvalds skb->priority = sk->sk_priority; 3944a19ec58SLaszlo Attila Toth skb->mark = sk->sk_mark; 3951da177e4SLinus Torvalds 396ab6e3febSEric Dumazet res = ip_local_out(skb); 397ab6e3febSEric Dumazet rcu_read_unlock(); 398ab6e3febSEric Dumazet return res; 3991da177e4SLinus Torvalds 4001da177e4SLinus Torvalds no_route: 401ab6e3febSEric Dumazet rcu_read_unlock(); 4025e38e270SPavel Emelyanov IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); 4031da177e4SLinus Torvalds kfree_skb(skb); 4041da177e4SLinus Torvalds return -EHOSTUNREACH; 4051da177e4SLinus Torvalds } 4064bc2f18bSEric Dumazet EXPORT_SYMBOL(ip_queue_xmit); 4071da177e4SLinus Torvalds 4081da177e4SLinus Torvalds 4091da177e4SLinus Torvalds static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 4101da177e4SLinus Torvalds { 4111da177e4SLinus Torvalds to->pkt_type = from->pkt_type; 4121da177e4SLinus Torvalds to->priority = from->priority; 4131da177e4SLinus Torvalds to->protocol = from->protocol; 414adf30907SEric Dumazet skb_dst_drop(to); 415fe76cda3SEric Dumazet skb_dst_copy(to, from); 4161da177e4SLinus Torvalds to->dev = from->dev; 41782e91ffeSThomas Graf to->mark = from->mark; 4181da177e4SLinus Torvalds 4191da177e4SLinus Torvalds /* Copy the flags to each fragment. */ 4201da177e4SLinus Torvalds IPCB(to)->flags = IPCB(from)->flags; 4211da177e4SLinus Torvalds 4221da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 4231da177e4SLinus Torvalds to->tc_index = from->tc_index; 4241da177e4SLinus Torvalds #endif 425e7ac05f3SYasuyuki Kozakai nf_copy(to, from); 426ba9dda3aSJozsef Kadlecsik #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \ 427ba9dda3aSJozsef Kadlecsik defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE) 428ba9dda3aSJozsef Kadlecsik to->nf_trace = from->nf_trace; 429ba9dda3aSJozsef Kadlecsik #endif 430c98d80edSJulian Anastasov #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 431c98d80edSJulian Anastasov to->ipvs_property = from->ipvs_property; 432c98d80edSJulian Anastasov #endif 433984bc16cSJames Morris skb_copy_secmark(to, from); 4341da177e4SLinus Torvalds } 4351da177e4SLinus Torvalds 4361da177e4SLinus Torvalds /* 4371da177e4SLinus Torvalds * This IP datagram is too large to be sent in one piece. Break it up into 4381da177e4SLinus Torvalds * smaller pieces (each of size equal to IP header plus 4391da177e4SLinus Torvalds * a block of the data of the original IP data part) that will yet fit in a 4401da177e4SLinus Torvalds * single device frame, and queue such a frame for sending. 4411da177e4SLinus Torvalds */ 4421da177e4SLinus Torvalds 4432e2f7aefSPatrick McHardy int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)) 4441da177e4SLinus Torvalds { 4451da177e4SLinus Torvalds struct iphdr *iph; 4461da177e4SLinus Torvalds int ptr; 4471da177e4SLinus Torvalds struct net_device *dev; 4481da177e4SLinus Torvalds struct sk_buff *skb2; 449c893b806SChangli Gao unsigned int mtu, hlen, left, len, ll_rs; 4501da177e4SLinus Torvalds int offset; 45176ab608dSAlexey Dobriyan __be16 not_last_frag; 452511c3f92SEric Dumazet struct rtable *rt = skb_rtable(skb); 4531da177e4SLinus Torvalds int err = 0; 4541da177e4SLinus Torvalds 455d8d1f30bSChangli Gao dev = rt->dst.dev; 4561da177e4SLinus Torvalds 4571da177e4SLinus Torvalds /* 4581da177e4SLinus Torvalds * Point into the IP datagram header. 4591da177e4SLinus Torvalds */ 4601da177e4SLinus Torvalds 461eddc9ec5SArnaldo Carvalho de Melo iph = ip_hdr(skb); 4621da177e4SLinus Torvalds 4631da177e4SLinus Torvalds if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) { 4645e38e270SPavel Emelyanov IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 4651da177e4SLinus Torvalds icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 466628a5c56SJohn Heffner htonl(ip_skb_dst_mtu(skb))); 4671da177e4SLinus Torvalds kfree_skb(skb); 4681da177e4SLinus Torvalds return -EMSGSIZE; 4691da177e4SLinus Torvalds } 4701da177e4SLinus Torvalds 4711da177e4SLinus Torvalds /* 4721da177e4SLinus Torvalds * Setup starting values. 4731da177e4SLinus Torvalds */ 4741da177e4SLinus Torvalds 4751da177e4SLinus Torvalds hlen = iph->ihl * 4; 476d8d1f30bSChangli Gao mtu = dst_mtu(&rt->dst) - hlen; /* Size of data space */ 4776c79bf0fSBart De Schuymer #ifdef CONFIG_BRIDGE_NETFILTER 4786c79bf0fSBart De Schuymer if (skb->nf_bridge) 4796c79bf0fSBart De Schuymer mtu -= nf_bridge_mtu_reduction(skb); 4806c79bf0fSBart De Schuymer #endif 48189cee8b1SHerbert Xu IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 4821da177e4SLinus Torvalds 4831da177e4SLinus Torvalds /* When frag_list is given, use it. First, check its validity: 4841da177e4SLinus Torvalds * some transformers could create wrong frag_list or break existing 4851da177e4SLinus Torvalds * one, it is not prohibited. In this case fall back to copying. 4861da177e4SLinus Torvalds * 4871da177e4SLinus Torvalds * LATER: this step can be merged to real generation of fragments, 4881da177e4SLinus Torvalds * we can switch to copy when see the first bad fragment. 4891da177e4SLinus Torvalds */ 49021dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 4913d13008eSEric Dumazet struct sk_buff *frag, *frag2; 4921da177e4SLinus Torvalds int first_len = skb_pagelen(skb); 4931da177e4SLinus Torvalds 4941da177e4SLinus Torvalds if (first_len - hlen > mtu || 4951da177e4SLinus Torvalds ((first_len - hlen) & 7) || 49656f8a75cSPaul Gortmaker ip_is_fragment(iph) || 4971da177e4SLinus Torvalds skb_cloned(skb)) 4981da177e4SLinus Torvalds goto slow_path; 4991da177e4SLinus Torvalds 500d7fcf1a5SDavid S. Miller skb_walk_frags(skb, frag) { 5011da177e4SLinus Torvalds /* Correct geometry. */ 5021da177e4SLinus Torvalds if (frag->len > mtu || 5031da177e4SLinus Torvalds ((frag->len & 7) && frag->next) || 5041da177e4SLinus Torvalds skb_headroom(frag) < hlen) 5053d13008eSEric Dumazet goto slow_path_clean; 5061da177e4SLinus Torvalds 5071da177e4SLinus Torvalds /* Partially cloned skb? */ 5081da177e4SLinus Torvalds if (skb_shared(frag)) 5093d13008eSEric Dumazet goto slow_path_clean; 5102fdba6b0SHerbert Xu 5112fdba6b0SHerbert Xu BUG_ON(frag->sk); 5122fdba6b0SHerbert Xu if (skb->sk) { 5132fdba6b0SHerbert Xu frag->sk = skb->sk; 5142fdba6b0SHerbert Xu frag->destructor = sock_wfree; 5152fdba6b0SHerbert Xu } 5163d13008eSEric Dumazet skb->truesize -= frag->truesize; 5171da177e4SLinus Torvalds } 5181da177e4SLinus Torvalds 5191da177e4SLinus Torvalds /* Everything is OK. Generate! */ 5201da177e4SLinus Torvalds 5211da177e4SLinus Torvalds err = 0; 5221da177e4SLinus Torvalds offset = 0; 5231da177e4SLinus Torvalds frag = skb_shinfo(skb)->frag_list; 524d7fcf1a5SDavid S. Miller skb_frag_list_init(skb); 5251da177e4SLinus Torvalds skb->data_len = first_len - skb_headlen(skb); 5261da177e4SLinus Torvalds skb->len = first_len; 5271da177e4SLinus Torvalds iph->tot_len = htons(first_len); 5281da177e4SLinus Torvalds iph->frag_off = htons(IP_MF); 5291da177e4SLinus Torvalds ip_send_check(iph); 5301da177e4SLinus Torvalds 5311da177e4SLinus Torvalds for (;;) { 5321da177e4SLinus Torvalds /* Prepare header of the next frame, 5331da177e4SLinus Torvalds * before previous one went down. */ 5341da177e4SLinus Torvalds if (frag) { 5351da177e4SLinus Torvalds frag->ip_summed = CHECKSUM_NONE; 536badff6d0SArnaldo Carvalho de Melo skb_reset_transport_header(frag); 537e2d1bca7SArnaldo Carvalho de Melo __skb_push(frag, hlen); 538e2d1bca7SArnaldo Carvalho de Melo skb_reset_network_header(frag); 539d56f90a7SArnaldo Carvalho de Melo memcpy(skb_network_header(frag), iph, hlen); 540eddc9ec5SArnaldo Carvalho de Melo iph = ip_hdr(frag); 5411da177e4SLinus Torvalds iph->tot_len = htons(frag->len); 5421da177e4SLinus Torvalds ip_copy_metadata(frag, skb); 5431da177e4SLinus Torvalds if (offset == 0) 5441da177e4SLinus Torvalds ip_options_fragment(frag); 5451da177e4SLinus Torvalds offset += skb->len - hlen; 5461da177e4SLinus Torvalds iph->frag_off = htons(offset>>3); 5471da177e4SLinus Torvalds if (frag->next != NULL) 5481da177e4SLinus Torvalds iph->frag_off |= htons(IP_MF); 5491da177e4SLinus Torvalds /* Ready, complete checksum */ 5501da177e4SLinus Torvalds ip_send_check(iph); 5511da177e4SLinus Torvalds } 5521da177e4SLinus Torvalds 5531da177e4SLinus Torvalds err = output(skb); 5541da177e4SLinus Torvalds 555dafee490SWei Dong if (!err) 5565e38e270SPavel Emelyanov IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); 5571da177e4SLinus Torvalds if (err || !frag) 5581da177e4SLinus Torvalds break; 5591da177e4SLinus Torvalds 5601da177e4SLinus Torvalds skb = frag; 5611da177e4SLinus Torvalds frag = skb->next; 5621da177e4SLinus Torvalds skb->next = NULL; 5631da177e4SLinus Torvalds } 5641da177e4SLinus Torvalds 5651da177e4SLinus Torvalds if (err == 0) { 5665e38e270SPavel Emelyanov IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); 5671da177e4SLinus Torvalds return 0; 5681da177e4SLinus Torvalds } 5691da177e4SLinus Torvalds 5701da177e4SLinus Torvalds while (frag) { 5711da177e4SLinus Torvalds skb = frag->next; 5721da177e4SLinus Torvalds kfree_skb(frag); 5731da177e4SLinus Torvalds frag = skb; 5741da177e4SLinus Torvalds } 5755e38e270SPavel Emelyanov IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 5761da177e4SLinus Torvalds return err; 5773d13008eSEric Dumazet 5783d13008eSEric Dumazet slow_path_clean: 5793d13008eSEric Dumazet skb_walk_frags(skb, frag2) { 5803d13008eSEric Dumazet if (frag2 == frag) 5813d13008eSEric Dumazet break; 5823d13008eSEric Dumazet frag2->sk = NULL; 5833d13008eSEric Dumazet frag2->destructor = NULL; 5843d13008eSEric Dumazet skb->truesize += frag2->truesize; 5853d13008eSEric Dumazet } 5861da177e4SLinus Torvalds } 5871da177e4SLinus Torvalds 5881da177e4SLinus Torvalds slow_path: 5891da177e4SLinus Torvalds left = skb->len - hlen; /* Space per frame */ 59049085bd7SGeorge Kadianakis ptr = hlen; /* Where to start from */ 5911da177e4SLinus Torvalds 5921da177e4SLinus Torvalds /* for bridged IP traffic encapsulated inside f.e. a vlan header, 5939bcfcaf5SStephen Hemminger * we need to make room for the encapsulating header 5949bcfcaf5SStephen Hemminger */ 595c893b806SChangli Gao ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb)); 5969bcfcaf5SStephen Hemminger 5971da177e4SLinus Torvalds /* 5981da177e4SLinus Torvalds * Fragment the datagram. 5991da177e4SLinus Torvalds */ 6001da177e4SLinus Torvalds 6011da177e4SLinus Torvalds offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 6021da177e4SLinus Torvalds not_last_frag = iph->frag_off & htons(IP_MF); 6031da177e4SLinus Torvalds 6041da177e4SLinus Torvalds /* 6051da177e4SLinus Torvalds * Keep copying data until we run out. 6061da177e4SLinus Torvalds */ 6071da177e4SLinus Torvalds 6081da177e4SLinus Torvalds while (left > 0) { 6091da177e4SLinus Torvalds len = left; 6101da177e4SLinus Torvalds /* IF: it doesn't fit, use 'mtu' - the data space left */ 6111da177e4SLinus Torvalds if (len > mtu) 6121da177e4SLinus Torvalds len = mtu; 6131da177e4SLinus Torvalds /* IF: we are not sending up to and including the packet end 6141da177e4SLinus Torvalds then align the next start on an eight byte boundary */ 6151da177e4SLinus Torvalds if (len < left) { 6161da177e4SLinus Torvalds len &= ~7; 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds /* 6191da177e4SLinus Torvalds * Allocate buffer. 6201da177e4SLinus Torvalds */ 6211da177e4SLinus Torvalds 6221da177e4SLinus Torvalds if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { 62364ce2073SPatrick McHardy NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); 6241da177e4SLinus Torvalds err = -ENOMEM; 6251da177e4SLinus Torvalds goto fail; 6261da177e4SLinus Torvalds } 6271da177e4SLinus Torvalds 6281da177e4SLinus Torvalds /* 6291da177e4SLinus Torvalds * Set up data on packet 6301da177e4SLinus Torvalds */ 6311da177e4SLinus Torvalds 6321da177e4SLinus Torvalds ip_copy_metadata(skb2, skb); 6331da177e4SLinus Torvalds skb_reserve(skb2, ll_rs); 6341da177e4SLinus Torvalds skb_put(skb2, len + hlen); 635c1d2bbe1SArnaldo Carvalho de Melo skb_reset_network_header(skb2); 636b0e380b1SArnaldo Carvalho de Melo skb2->transport_header = skb2->network_header + hlen; 6371da177e4SLinus Torvalds 6381da177e4SLinus Torvalds /* 6391da177e4SLinus Torvalds * Charge the memory for the fragment to any owner 6401da177e4SLinus Torvalds * it might possess 6411da177e4SLinus Torvalds */ 6421da177e4SLinus Torvalds 6431da177e4SLinus Torvalds if (skb->sk) 6441da177e4SLinus Torvalds skb_set_owner_w(skb2, skb->sk); 6451da177e4SLinus Torvalds 6461da177e4SLinus Torvalds /* 6471da177e4SLinus Torvalds * Copy the packet header into the new buffer. 6481da177e4SLinus Torvalds */ 6491da177e4SLinus Torvalds 650d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds /* 6531da177e4SLinus Torvalds * Copy a block of the IP datagram. 6541da177e4SLinus Torvalds */ 655bff9b61cSArnaldo Carvalho de Melo if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) 6561da177e4SLinus Torvalds BUG(); 6571da177e4SLinus Torvalds left -= len; 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds /* 6601da177e4SLinus Torvalds * Fill in the new header fields. 6611da177e4SLinus Torvalds */ 662eddc9ec5SArnaldo Carvalho de Melo iph = ip_hdr(skb2); 6631da177e4SLinus Torvalds iph->frag_off = htons((offset >> 3)); 6641da177e4SLinus Torvalds 6651da177e4SLinus Torvalds /* ANK: dirty, but effective trick. Upgrade options only if 6661da177e4SLinus Torvalds * the segment to be fragmented was THE FIRST (otherwise, 6671da177e4SLinus Torvalds * options are already fixed) and make it ONCE 6681da177e4SLinus Torvalds * on the initial skb, so that all the following fragments 6691da177e4SLinus Torvalds * will inherit fixed options. 6701da177e4SLinus Torvalds */ 6711da177e4SLinus Torvalds if (offset == 0) 6721da177e4SLinus Torvalds ip_options_fragment(skb); 6731da177e4SLinus Torvalds 6741da177e4SLinus Torvalds /* 6751da177e4SLinus Torvalds * Added AC : If we are fragmenting a fragment that's not the 6761da177e4SLinus Torvalds * last fragment then keep MF on each bit 6771da177e4SLinus Torvalds */ 6781da177e4SLinus Torvalds if (left > 0 || not_last_frag) 6791da177e4SLinus Torvalds iph->frag_off |= htons(IP_MF); 6801da177e4SLinus Torvalds ptr += len; 6811da177e4SLinus Torvalds offset += len; 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds /* 6841da177e4SLinus Torvalds * Put this fragment into the sending queue. 6851da177e4SLinus Torvalds */ 6861da177e4SLinus Torvalds iph->tot_len = htons(len + hlen); 6871da177e4SLinus Torvalds 6881da177e4SLinus Torvalds ip_send_check(iph); 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds err = output(skb2); 6911da177e4SLinus Torvalds if (err) 6921da177e4SLinus Torvalds goto fail; 693dafee490SWei Dong 6945e38e270SPavel Emelyanov IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); 6951da177e4SLinus Torvalds } 6961da177e4SLinus Torvalds kfree_skb(skb); 6975e38e270SPavel Emelyanov IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); 6981da177e4SLinus Torvalds return err; 6991da177e4SLinus Torvalds 7001da177e4SLinus Torvalds fail: 7011da177e4SLinus Torvalds kfree_skb(skb); 7025e38e270SPavel Emelyanov IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 7031da177e4SLinus Torvalds return err; 7041da177e4SLinus Torvalds } 7052e2f7aefSPatrick McHardy EXPORT_SYMBOL(ip_fragment); 7062e2f7aefSPatrick McHardy 7071da177e4SLinus Torvalds int 7081da177e4SLinus Torvalds ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 7091da177e4SLinus Torvalds { 7101da177e4SLinus Torvalds struct iovec *iov = from; 7111da177e4SLinus Torvalds 71284fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 7131da177e4SLinus Torvalds if (memcpy_fromiovecend(to, iov, offset, len) < 0) 7141da177e4SLinus Torvalds return -EFAULT; 7151da177e4SLinus Torvalds } else { 71644bb9363SAl Viro __wsum csum = 0; 7171da177e4SLinus Torvalds if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) 7181da177e4SLinus Torvalds return -EFAULT; 7191da177e4SLinus Torvalds skb->csum = csum_block_add(skb->csum, csum, odd); 7201da177e4SLinus Torvalds } 7211da177e4SLinus Torvalds return 0; 7221da177e4SLinus Torvalds } 7234bc2f18bSEric Dumazet EXPORT_SYMBOL(ip_generic_getfrag); 7241da177e4SLinus Torvalds 72544bb9363SAl Viro static inline __wsum 7261da177e4SLinus Torvalds csum_page(struct page *page, int offset, int copy) 7271da177e4SLinus Torvalds { 7281da177e4SLinus Torvalds char *kaddr; 72944bb9363SAl Viro __wsum csum; 7301da177e4SLinus Torvalds kaddr = kmap(page); 7311da177e4SLinus Torvalds csum = csum_partial(kaddr + offset, copy, 0); 7321da177e4SLinus Torvalds kunmap(page); 7331da177e4SLinus Torvalds return csum; 7341da177e4SLinus Torvalds } 7351da177e4SLinus Torvalds 7364b30b1c6SAdrian Bunk static inline int ip_ufo_append_data(struct sock *sk, 7371470ddf7SHerbert Xu struct sk_buff_head *queue, 738e89e9cf5SAnanda Raju int getfrag(void *from, char *to, int offset, int len, 739e89e9cf5SAnanda Raju int odd, struct sk_buff *skb), 740e89e9cf5SAnanda Raju void *from, int length, int hh_len, int fragheaderlen, 741e89e9cf5SAnanda Raju int transhdrlen, int mtu, unsigned int flags) 742e89e9cf5SAnanda Raju { 743e89e9cf5SAnanda Raju struct sk_buff *skb; 744e89e9cf5SAnanda Raju int err; 745e89e9cf5SAnanda Raju 746e89e9cf5SAnanda Raju /* There is support for UDP fragmentation offload by network 747e89e9cf5SAnanda Raju * device, so create one single skb packet containing complete 748e89e9cf5SAnanda Raju * udp datagram 749e89e9cf5SAnanda Raju */ 7501470ddf7SHerbert Xu if ((skb = skb_peek_tail(queue)) == NULL) { 751e89e9cf5SAnanda Raju skb = sock_alloc_send_skb(sk, 752e89e9cf5SAnanda Raju hh_len + fragheaderlen + transhdrlen + 20, 753e89e9cf5SAnanda Raju (flags & MSG_DONTWAIT), &err); 754e89e9cf5SAnanda Raju 755e89e9cf5SAnanda Raju if (skb == NULL) 756e89e9cf5SAnanda Raju return err; 757e89e9cf5SAnanda Raju 758e89e9cf5SAnanda Raju /* reserve space for Hardware header */ 759e89e9cf5SAnanda Raju skb_reserve(skb, hh_len); 760e89e9cf5SAnanda Raju 761e89e9cf5SAnanda Raju /* create space for UDP/IP header */ 762e89e9cf5SAnanda Raju skb_put(skb, fragheaderlen + transhdrlen); 763e89e9cf5SAnanda Raju 764e89e9cf5SAnanda Raju /* initialize network header pointer */ 765c1d2bbe1SArnaldo Carvalho de Melo skb_reset_network_header(skb); 766e89e9cf5SAnanda Raju 767e89e9cf5SAnanda Raju /* initialize protocol header pointer */ 768b0e380b1SArnaldo Carvalho de Melo skb->transport_header = skb->network_header + fragheaderlen; 769e89e9cf5SAnanda Raju 77084fa7933SPatrick McHardy skb->ip_summed = CHECKSUM_PARTIAL; 771e89e9cf5SAnanda Raju skb->csum = 0; 772e89e9cf5SAnanda Raju 773e89e9cf5SAnanda Raju /* specify the length of each IP datagram fragment */ 7747967168cSHerbert Xu skb_shinfo(skb)->gso_size = mtu - fragheaderlen; 775f83ef8c0SHerbert Xu skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 7761470ddf7SHerbert Xu __skb_queue_tail(queue, skb); 777e89e9cf5SAnanda Raju } 778be9164e7SKostya B 779be9164e7SKostya B return skb_append_datato_frags(sk, skb, getfrag, from, 780be9164e7SKostya B (length - transhdrlen)); 781e89e9cf5SAnanda Raju } 782e89e9cf5SAnanda Raju 783f5fca608SDavid S. Miller static int __ip_append_data(struct sock *sk, 784f5fca608SDavid S. Miller struct flowi4 *fl4, 785f5fca608SDavid S. Miller struct sk_buff_head *queue, 7861470ddf7SHerbert Xu struct inet_cork *cork, 7871470ddf7SHerbert Xu int getfrag(void *from, char *to, int offset, 7881470ddf7SHerbert Xu int len, int odd, struct sk_buff *skb), 7891da177e4SLinus Torvalds void *from, int length, int transhdrlen, 7901da177e4SLinus Torvalds unsigned int flags) 7911da177e4SLinus Torvalds { 7921da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 7931da177e4SLinus Torvalds struct sk_buff *skb; 7941da177e4SLinus Torvalds 79507df5294SHerbert Xu struct ip_options *opt = cork->opt; 7961da177e4SLinus Torvalds int hh_len; 7971da177e4SLinus Torvalds int exthdrlen; 7981da177e4SLinus Torvalds int mtu; 7991da177e4SLinus Torvalds int copy; 8001da177e4SLinus Torvalds int err; 8011da177e4SLinus Torvalds int offset = 0; 8021da177e4SLinus Torvalds unsigned int maxfraglen, fragheaderlen; 8031da177e4SLinus Torvalds int csummode = CHECKSUM_NONE; 8041470ddf7SHerbert Xu struct rtable *rt = (struct rtable *)cork->dst; 8051da177e4SLinus Torvalds 80696d7303eSSteffen Klassert skb = skb_peek_tail(queue); 80796d7303eSSteffen Klassert 80896d7303eSSteffen Klassert exthdrlen = !skb ? rt->dst.header_len : 0; 80907df5294SHerbert Xu mtu = cork->fragsize; 8101470ddf7SHerbert Xu 811d8d1f30bSChangli Gao hh_len = LL_RESERVED_SPACE(rt->dst.dev); 8121da177e4SLinus Torvalds 8131da177e4SLinus Torvalds fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 8141da177e4SLinus Torvalds maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 8151da177e4SLinus Torvalds 8161470ddf7SHerbert Xu if (cork->length + length > 0xFFFF - fragheaderlen) { 817f5fca608SDavid S. Miller ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 818c720c7e8SEric Dumazet mtu-exthdrlen); 8191da177e4SLinus Torvalds return -EMSGSIZE; 8201da177e4SLinus Torvalds } 8211da177e4SLinus Torvalds 8221da177e4SLinus Torvalds /* 8231da177e4SLinus Torvalds * transhdrlen > 0 means that this is the first fragment and we wish 8241da177e4SLinus Torvalds * it won't be fragmented in the future. 8251da177e4SLinus Torvalds */ 8261da177e4SLinus Torvalds if (transhdrlen && 8271da177e4SLinus Torvalds length + fragheaderlen <= mtu && 828d8d1f30bSChangli Gao rt->dst.dev->features & NETIF_F_V4_CSUM && 8291da177e4SLinus Torvalds !exthdrlen) 83084fa7933SPatrick McHardy csummode = CHECKSUM_PARTIAL; 8311da177e4SLinus Torvalds 8321470ddf7SHerbert Xu cork->length += length; 83326cde9f7SHerbert Xu if (((length > mtu) || (skb && skb_is_gso(skb))) && 834be9164e7SKostya B (sk->sk_protocol == IPPROTO_UDP) && 835c146066aSSteffen Klassert (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) { 8361470ddf7SHerbert Xu err = ip_ufo_append_data(sk, queue, getfrag, from, length, 8371470ddf7SHerbert Xu hh_len, fragheaderlen, transhdrlen, 8381470ddf7SHerbert Xu mtu, flags); 839baa829d8SPatrick McHardy if (err) 840e89e9cf5SAnanda Raju goto error; 841e89e9cf5SAnanda Raju return 0; 842e89e9cf5SAnanda Raju } 8431da177e4SLinus Torvalds 8441da177e4SLinus Torvalds /* So, what's going on in the loop below? 8451da177e4SLinus Torvalds * 8461da177e4SLinus Torvalds * We use calculated fragment length to generate chained skb, 8471da177e4SLinus Torvalds * each of segments is IP fragment ready for sending to network after 8481da177e4SLinus Torvalds * adding appropriate IP header. 8491da177e4SLinus Torvalds */ 8501da177e4SLinus Torvalds 85126cde9f7SHerbert Xu if (!skb) 8521da177e4SLinus Torvalds goto alloc_new_skb; 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds while (length > 0) { 8551da177e4SLinus Torvalds /* Check if the remaining data fits into current packet. */ 8561da177e4SLinus Torvalds copy = mtu - skb->len; 8571da177e4SLinus Torvalds if (copy < length) 8581da177e4SLinus Torvalds copy = maxfraglen - skb->len; 8591da177e4SLinus Torvalds if (copy <= 0) { 8601da177e4SLinus Torvalds char *data; 8611da177e4SLinus Torvalds unsigned int datalen; 8621da177e4SLinus Torvalds unsigned int fraglen; 8631da177e4SLinus Torvalds unsigned int fraggap; 8641da177e4SLinus Torvalds unsigned int alloclen; 8651da177e4SLinus Torvalds struct sk_buff *skb_prev; 8661da177e4SLinus Torvalds alloc_new_skb: 8671da177e4SLinus Torvalds skb_prev = skb; 8681da177e4SLinus Torvalds if (skb_prev) 8691da177e4SLinus Torvalds fraggap = skb_prev->len - maxfraglen; 8701da177e4SLinus Torvalds else 8711da177e4SLinus Torvalds fraggap = 0; 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds /* 8741da177e4SLinus Torvalds * If remaining data exceeds the mtu, 8751da177e4SLinus Torvalds * we know we need more fragment(s). 8761da177e4SLinus Torvalds */ 8771da177e4SLinus Torvalds datalen = length + fraggap; 8781da177e4SLinus Torvalds if (datalen > mtu - fragheaderlen) 8791da177e4SLinus Torvalds datalen = maxfraglen - fragheaderlen; 8801da177e4SLinus Torvalds fraglen = datalen + fragheaderlen; 8811da177e4SLinus Torvalds 8821da177e4SLinus Torvalds if ((flags & MSG_MORE) && 883d8d1f30bSChangli Gao !(rt->dst.dev->features&NETIF_F_SG)) 8841da177e4SLinus Torvalds alloclen = mtu; 8851da177e4SLinus Torvalds else 88659104f06SEric Dumazet alloclen = fraglen; 8871da177e4SLinus Torvalds 888353e5c9aSSteffen Klassert alloclen += exthdrlen; 889353e5c9aSSteffen Klassert 8901da177e4SLinus Torvalds /* The last fragment gets additional space at tail. 8911da177e4SLinus Torvalds * Note, with MSG_MORE we overallocate on fragments, 8921da177e4SLinus Torvalds * because we have no idea what fragment will be 8931da177e4SLinus Torvalds * the last. 8941da177e4SLinus Torvalds */ 89533f99dc7SSteffen Klassert if (datalen == length + fraggap) 896d8d1f30bSChangli Gao alloclen += rt->dst.trailer_len; 89733f99dc7SSteffen Klassert 8981da177e4SLinus Torvalds if (transhdrlen) { 8991da177e4SLinus Torvalds skb = sock_alloc_send_skb(sk, 9001da177e4SLinus Torvalds alloclen + hh_len + 15, 9011da177e4SLinus Torvalds (flags & MSG_DONTWAIT), &err); 9021da177e4SLinus Torvalds } else { 9031da177e4SLinus Torvalds skb = NULL; 9041da177e4SLinus Torvalds if (atomic_read(&sk->sk_wmem_alloc) <= 9051da177e4SLinus Torvalds 2 * sk->sk_sndbuf) 9061da177e4SLinus Torvalds skb = sock_wmalloc(sk, 9071da177e4SLinus Torvalds alloclen + hh_len + 15, 1, 9081da177e4SLinus Torvalds sk->sk_allocation); 9091da177e4SLinus Torvalds if (unlikely(skb == NULL)) 9101da177e4SLinus Torvalds err = -ENOBUFS; 91151f31cabSPatrick Ohly else 91251f31cabSPatrick Ohly /* only the initial fragment is 91351f31cabSPatrick Ohly time stamped */ 9141470ddf7SHerbert Xu cork->tx_flags = 0; 9151da177e4SLinus Torvalds } 9161da177e4SLinus Torvalds if (skb == NULL) 9171da177e4SLinus Torvalds goto error; 9181da177e4SLinus Torvalds 9191da177e4SLinus Torvalds /* 9201da177e4SLinus Torvalds * Fill in the control structures 9211da177e4SLinus Torvalds */ 9221da177e4SLinus Torvalds skb->ip_summed = csummode; 9231da177e4SLinus Torvalds skb->csum = 0; 9241da177e4SLinus Torvalds skb_reserve(skb, hh_len); 9251470ddf7SHerbert Xu skb_shinfo(skb)->tx_flags = cork->tx_flags; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds /* 9281da177e4SLinus Torvalds * Find where to start putting bytes. 9291da177e4SLinus Torvalds */ 930353e5c9aSSteffen Klassert data = skb_put(skb, fraglen + exthdrlen); 931c14d2450SArnaldo Carvalho de Melo skb_set_network_header(skb, exthdrlen); 932b0e380b1SArnaldo Carvalho de Melo skb->transport_header = (skb->network_header + 933b0e380b1SArnaldo Carvalho de Melo fragheaderlen); 934353e5c9aSSteffen Klassert data += fragheaderlen + exthdrlen; 9351da177e4SLinus Torvalds 9361da177e4SLinus Torvalds if (fraggap) { 9371da177e4SLinus Torvalds skb->csum = skb_copy_and_csum_bits( 9381da177e4SLinus Torvalds skb_prev, maxfraglen, 9391da177e4SLinus Torvalds data + transhdrlen, fraggap, 0); 9401da177e4SLinus Torvalds skb_prev->csum = csum_sub(skb_prev->csum, 9411da177e4SLinus Torvalds skb->csum); 9421da177e4SLinus Torvalds data += fraggap; 943e9fa4f7bSHerbert Xu pskb_trim_unique(skb_prev, maxfraglen); 9441da177e4SLinus Torvalds } 9451da177e4SLinus Torvalds 9461da177e4SLinus Torvalds copy = datalen - transhdrlen - fraggap; 9471da177e4SLinus Torvalds if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 9481da177e4SLinus Torvalds err = -EFAULT; 9491da177e4SLinus Torvalds kfree_skb(skb); 9501da177e4SLinus Torvalds goto error; 9511da177e4SLinus Torvalds } 9521da177e4SLinus Torvalds 9531da177e4SLinus Torvalds offset += copy; 9541da177e4SLinus Torvalds length -= datalen - fraggap; 9551da177e4SLinus Torvalds transhdrlen = 0; 9561da177e4SLinus Torvalds exthdrlen = 0; 9571da177e4SLinus Torvalds csummode = CHECKSUM_NONE; 9581da177e4SLinus Torvalds 9591da177e4SLinus Torvalds /* 9601da177e4SLinus Torvalds * Put the packet on the pending queue. 9611da177e4SLinus Torvalds */ 9621470ddf7SHerbert Xu __skb_queue_tail(queue, skb); 9631da177e4SLinus Torvalds continue; 9641da177e4SLinus Torvalds } 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds if (copy > length) 9671da177e4SLinus Torvalds copy = length; 9681da177e4SLinus Torvalds 969d8d1f30bSChangli Gao if (!(rt->dst.dev->features&NETIF_F_SG)) { 9701da177e4SLinus Torvalds unsigned int off; 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds off = skb->len; 9731da177e4SLinus Torvalds if (getfrag(from, skb_put(skb, copy), 9741da177e4SLinus Torvalds offset, copy, off, skb) < 0) { 9751da177e4SLinus Torvalds __skb_trim(skb, off); 9761da177e4SLinus Torvalds err = -EFAULT; 9771da177e4SLinus Torvalds goto error; 9781da177e4SLinus Torvalds } 9791da177e4SLinus Torvalds } else { 9801da177e4SLinus Torvalds int i = skb_shinfo(skb)->nr_frags; 9811da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1]; 9821470ddf7SHerbert Xu struct page *page = cork->page; 9831470ddf7SHerbert Xu int off = cork->off; 9841da177e4SLinus Torvalds unsigned int left; 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds if (page && (left = PAGE_SIZE - off) > 0) { 9871da177e4SLinus Torvalds if (copy >= left) 9881da177e4SLinus Torvalds copy = left; 9891da177e4SLinus Torvalds if (page != frag->page) { 9901da177e4SLinus Torvalds if (i == MAX_SKB_FRAGS) { 9911da177e4SLinus Torvalds err = -EMSGSIZE; 9921da177e4SLinus Torvalds goto error; 9931da177e4SLinus Torvalds } 9941da177e4SLinus Torvalds get_page(page); 9951470ddf7SHerbert Xu skb_fill_page_desc(skb, i, page, off, 0); 9961da177e4SLinus Torvalds frag = &skb_shinfo(skb)->frags[i]; 9971da177e4SLinus Torvalds } 9981da177e4SLinus Torvalds } else if (i < MAX_SKB_FRAGS) { 9991da177e4SLinus Torvalds if (copy > PAGE_SIZE) 10001da177e4SLinus Torvalds copy = PAGE_SIZE; 10011da177e4SLinus Torvalds page = alloc_pages(sk->sk_allocation, 0); 10021da177e4SLinus Torvalds if (page == NULL) { 10031da177e4SLinus Torvalds err = -ENOMEM; 10041da177e4SLinus Torvalds goto error; 10051da177e4SLinus Torvalds } 10061470ddf7SHerbert Xu cork->page = page; 10071470ddf7SHerbert Xu cork->off = 0; 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds skb_fill_page_desc(skb, i, page, 0, 0); 10101da177e4SLinus Torvalds frag = &skb_shinfo(skb)->frags[i]; 10111da177e4SLinus Torvalds } else { 10121da177e4SLinus Torvalds err = -EMSGSIZE; 10131da177e4SLinus Torvalds goto error; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) { 10161da177e4SLinus Torvalds err = -EFAULT; 10171da177e4SLinus Torvalds goto error; 10181da177e4SLinus Torvalds } 10191470ddf7SHerbert Xu cork->off += copy; 10201da177e4SLinus Torvalds frag->size += copy; 10211da177e4SLinus Torvalds skb->len += copy; 10221da177e4SLinus Torvalds skb->data_len += copy; 1023f945fa7aSHerbert Xu skb->truesize += copy; 1024f945fa7aSHerbert Xu atomic_add(copy, &sk->sk_wmem_alloc); 10251da177e4SLinus Torvalds } 10261da177e4SLinus Torvalds offset += copy; 10271da177e4SLinus Torvalds length -= copy; 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds 10301da177e4SLinus Torvalds return 0; 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds error: 10331470ddf7SHerbert Xu cork->length -= length; 10345e38e270SPavel Emelyanov IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 10351da177e4SLinus Torvalds return err; 10361da177e4SLinus Torvalds } 10371da177e4SLinus Torvalds 10381470ddf7SHerbert Xu static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 10391470ddf7SHerbert Xu struct ipcm_cookie *ipc, struct rtable **rtp) 10401470ddf7SHerbert Xu { 10411470ddf7SHerbert Xu struct inet_sock *inet = inet_sk(sk); 1042f6d8bd05SEric Dumazet struct ip_options_rcu *opt; 10431470ddf7SHerbert Xu struct rtable *rt; 10441470ddf7SHerbert Xu 10451470ddf7SHerbert Xu /* 10461470ddf7SHerbert Xu * setup for corking. 10471470ddf7SHerbert Xu */ 10481470ddf7SHerbert Xu opt = ipc->opt; 10491470ddf7SHerbert Xu if (opt) { 10501470ddf7SHerbert Xu if (cork->opt == NULL) { 10511470ddf7SHerbert Xu cork->opt = kmalloc(sizeof(struct ip_options) + 40, 10521470ddf7SHerbert Xu sk->sk_allocation); 10531470ddf7SHerbert Xu if (unlikely(cork->opt == NULL)) 10541470ddf7SHerbert Xu return -ENOBUFS; 10551470ddf7SHerbert Xu } 1056f6d8bd05SEric Dumazet memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 10571470ddf7SHerbert Xu cork->flags |= IPCORK_OPT; 10581470ddf7SHerbert Xu cork->addr = ipc->addr; 10591470ddf7SHerbert Xu } 10601470ddf7SHerbert Xu rt = *rtp; 10611470ddf7SHerbert Xu if (unlikely(!rt)) 10621470ddf7SHerbert Xu return -EFAULT; 10631470ddf7SHerbert Xu /* 10641470ddf7SHerbert Xu * We steal reference to this route, caller should not release it 10651470ddf7SHerbert Xu */ 10661470ddf7SHerbert Xu *rtp = NULL; 10671470ddf7SHerbert Xu cork->fragsize = inet->pmtudisc == IP_PMTUDISC_PROBE ? 1068353e5c9aSSteffen Klassert rt->dst.dev->mtu : dst_mtu(&rt->dst); 10691470ddf7SHerbert Xu cork->dst = &rt->dst; 10701470ddf7SHerbert Xu cork->length = 0; 10711470ddf7SHerbert Xu cork->tx_flags = ipc->tx_flags; 10721470ddf7SHerbert Xu cork->page = NULL; 10731470ddf7SHerbert Xu cork->off = 0; 10741470ddf7SHerbert Xu 10751470ddf7SHerbert Xu return 0; 10761470ddf7SHerbert Xu } 10771470ddf7SHerbert Xu 10781470ddf7SHerbert Xu /* 10791470ddf7SHerbert Xu * ip_append_data() and ip_append_page() can make one large IP datagram 10801470ddf7SHerbert Xu * from many pieces of data. Each pieces will be holded on the socket 10811470ddf7SHerbert Xu * until ip_push_pending_frames() is called. Each piece can be a page 10821470ddf7SHerbert Xu * or non-page data. 10831470ddf7SHerbert Xu * 10841470ddf7SHerbert Xu * Not only UDP, other transport protocols - e.g. raw sockets - can use 10851470ddf7SHerbert Xu * this interface potentially. 10861470ddf7SHerbert Xu * 10871470ddf7SHerbert Xu * LATER: length must be adjusted by pad at tail, when it is required. 10881470ddf7SHerbert Xu */ 1089f5fca608SDavid S. Miller int ip_append_data(struct sock *sk, struct flowi4 *fl4, 10901470ddf7SHerbert Xu int getfrag(void *from, char *to, int offset, int len, 10911470ddf7SHerbert Xu int odd, struct sk_buff *skb), 10921470ddf7SHerbert Xu void *from, int length, int transhdrlen, 10931470ddf7SHerbert Xu struct ipcm_cookie *ipc, struct rtable **rtp, 10941470ddf7SHerbert Xu unsigned int flags) 10951470ddf7SHerbert Xu { 10961470ddf7SHerbert Xu struct inet_sock *inet = inet_sk(sk); 10971470ddf7SHerbert Xu int err; 10981470ddf7SHerbert Xu 10991470ddf7SHerbert Xu if (flags&MSG_PROBE) 11001470ddf7SHerbert Xu return 0; 11011470ddf7SHerbert Xu 11021470ddf7SHerbert Xu if (skb_queue_empty(&sk->sk_write_queue)) { 1103bdc712b4SDavid S. Miller err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 11041470ddf7SHerbert Xu if (err) 11051470ddf7SHerbert Xu return err; 11061470ddf7SHerbert Xu } else { 11071470ddf7SHerbert Xu transhdrlen = 0; 11081470ddf7SHerbert Xu } 11091470ddf7SHerbert Xu 1110f5fca608SDavid S. Miller return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, getfrag, 11111470ddf7SHerbert Xu from, length, transhdrlen, flags); 11121470ddf7SHerbert Xu } 11131470ddf7SHerbert Xu 1114f5fca608SDavid S. Miller ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 11151da177e4SLinus Torvalds int offset, size_t size, int flags) 11161da177e4SLinus Torvalds { 11171da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 11181da177e4SLinus Torvalds struct sk_buff *skb; 11191da177e4SLinus Torvalds struct rtable *rt; 11201da177e4SLinus Torvalds struct ip_options *opt = NULL; 1121bdc712b4SDavid S. Miller struct inet_cork *cork; 11221da177e4SLinus Torvalds int hh_len; 11231da177e4SLinus Torvalds int mtu; 11241da177e4SLinus Torvalds int len; 11251da177e4SLinus Torvalds int err; 11261da177e4SLinus Torvalds unsigned int maxfraglen, fragheaderlen, fraggap; 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds if (inet->hdrincl) 11291da177e4SLinus Torvalds return -EPERM; 11301da177e4SLinus Torvalds 11311da177e4SLinus Torvalds if (flags&MSG_PROBE) 11321da177e4SLinus Torvalds return 0; 11331da177e4SLinus Torvalds 11341da177e4SLinus Torvalds if (skb_queue_empty(&sk->sk_write_queue)) 11351da177e4SLinus Torvalds return -EINVAL; 11361da177e4SLinus Torvalds 1137bdc712b4SDavid S. Miller cork = &inet->cork.base; 1138bdc712b4SDavid S. Miller rt = (struct rtable *)cork->dst; 1139bdc712b4SDavid S. Miller if (cork->flags & IPCORK_OPT) 1140bdc712b4SDavid S. Miller opt = cork->opt; 11411da177e4SLinus Torvalds 1142d8d1f30bSChangli Gao if (!(rt->dst.dev->features&NETIF_F_SG)) 11431da177e4SLinus Torvalds return -EOPNOTSUPP; 11441da177e4SLinus Torvalds 1145d8d1f30bSChangli Gao hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1146bdc712b4SDavid S. Miller mtu = cork->fragsize; 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 11491da177e4SLinus Torvalds maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 11501da177e4SLinus Torvalds 1151bdc712b4SDavid S. Miller if (cork->length + size > 0xFFFF - fragheaderlen) { 1152f5fca608SDavid S. Miller ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, mtu); 11531da177e4SLinus Torvalds return -EMSGSIZE; 11541da177e4SLinus Torvalds } 11551da177e4SLinus Torvalds 11561da177e4SLinus Torvalds if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) 11571da177e4SLinus Torvalds return -EINVAL; 11581da177e4SLinus Torvalds 1159bdc712b4SDavid S. Miller cork->length += size; 116026cde9f7SHerbert Xu if ((size + skb->len > mtu) && 116126cde9f7SHerbert Xu (sk->sk_protocol == IPPROTO_UDP) && 1162d8d1f30bSChangli Gao (rt->dst.dev->features & NETIF_F_UFO)) { 11637967168cSHerbert Xu skb_shinfo(skb)->gso_size = mtu - fragheaderlen; 1164f83ef8c0SHerbert Xu skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 11657967168cSHerbert Xu } 1166e89e9cf5SAnanda Raju 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds while (size > 0) { 11691da177e4SLinus Torvalds int i; 11701da177e4SLinus Torvalds 117189114afdSHerbert Xu if (skb_is_gso(skb)) 1172e89e9cf5SAnanda Raju len = size; 1173e89e9cf5SAnanda Raju else { 1174e89e9cf5SAnanda Raju 11751da177e4SLinus Torvalds /* Check if the remaining data fits into current packet. */ 11761da177e4SLinus Torvalds len = mtu - skb->len; 11771da177e4SLinus Torvalds if (len < size) 11781da177e4SLinus Torvalds len = maxfraglen - skb->len; 1179e89e9cf5SAnanda Raju } 11801da177e4SLinus Torvalds if (len <= 0) { 11811da177e4SLinus Torvalds struct sk_buff *skb_prev; 11821da177e4SLinus Torvalds int alloclen; 11831da177e4SLinus Torvalds 11841da177e4SLinus Torvalds skb_prev = skb; 11851da177e4SLinus Torvalds fraggap = skb_prev->len - maxfraglen; 11861da177e4SLinus Torvalds 11871da177e4SLinus Torvalds alloclen = fragheaderlen + hh_len + fraggap + 15; 11881da177e4SLinus Torvalds skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 11891da177e4SLinus Torvalds if (unlikely(!skb)) { 11901da177e4SLinus Torvalds err = -ENOBUFS; 11911da177e4SLinus Torvalds goto error; 11921da177e4SLinus Torvalds } 11931da177e4SLinus Torvalds 11941da177e4SLinus Torvalds /* 11951da177e4SLinus Torvalds * Fill in the control structures 11961da177e4SLinus Torvalds */ 11971da177e4SLinus Torvalds skb->ip_summed = CHECKSUM_NONE; 11981da177e4SLinus Torvalds skb->csum = 0; 11991da177e4SLinus Torvalds skb_reserve(skb, hh_len); 12001da177e4SLinus Torvalds 12011da177e4SLinus Torvalds /* 12021da177e4SLinus Torvalds * Find where to start putting bytes. 12031da177e4SLinus Torvalds */ 1204967b05f6SArnaldo Carvalho de Melo skb_put(skb, fragheaderlen + fraggap); 12052ca9e6f2SArnaldo Carvalho de Melo skb_reset_network_header(skb); 1206b0e380b1SArnaldo Carvalho de Melo skb->transport_header = (skb->network_header + 1207b0e380b1SArnaldo Carvalho de Melo fragheaderlen); 12081da177e4SLinus Torvalds if (fraggap) { 1209967b05f6SArnaldo Carvalho de Melo skb->csum = skb_copy_and_csum_bits(skb_prev, 1210967b05f6SArnaldo Carvalho de Melo maxfraglen, 12119c70220bSArnaldo Carvalho de Melo skb_transport_header(skb), 1212967b05f6SArnaldo Carvalho de Melo fraggap, 0); 12131da177e4SLinus Torvalds skb_prev->csum = csum_sub(skb_prev->csum, 12141da177e4SLinus Torvalds skb->csum); 1215e9fa4f7bSHerbert Xu pskb_trim_unique(skb_prev, maxfraglen); 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds 12181da177e4SLinus Torvalds /* 12191da177e4SLinus Torvalds * Put the packet on the pending queue. 12201da177e4SLinus Torvalds */ 12211da177e4SLinus Torvalds __skb_queue_tail(&sk->sk_write_queue, skb); 12221da177e4SLinus Torvalds continue; 12231da177e4SLinus Torvalds } 12241da177e4SLinus Torvalds 12251da177e4SLinus Torvalds i = skb_shinfo(skb)->nr_frags; 12261da177e4SLinus Torvalds if (len > size) 12271da177e4SLinus Torvalds len = size; 12281da177e4SLinus Torvalds if (skb_can_coalesce(skb, i, page, offset)) { 12291da177e4SLinus Torvalds skb_shinfo(skb)->frags[i-1].size += len; 12301da177e4SLinus Torvalds } else if (i < MAX_SKB_FRAGS) { 12311da177e4SLinus Torvalds get_page(page); 12321da177e4SLinus Torvalds skb_fill_page_desc(skb, i, page, offset, len); 12331da177e4SLinus Torvalds } else { 12341da177e4SLinus Torvalds err = -EMSGSIZE; 12351da177e4SLinus Torvalds goto error; 12361da177e4SLinus Torvalds } 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds if (skb->ip_summed == CHECKSUM_NONE) { 123944bb9363SAl Viro __wsum csum; 12401da177e4SLinus Torvalds csum = csum_page(page, offset, len); 12411da177e4SLinus Torvalds skb->csum = csum_block_add(skb->csum, csum, skb->len); 12421da177e4SLinus Torvalds } 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds skb->len += len; 12451da177e4SLinus Torvalds skb->data_len += len; 12461e34a11dSDavid S. Miller skb->truesize += len; 12471e34a11dSDavid S. Miller atomic_add(len, &sk->sk_wmem_alloc); 12481da177e4SLinus Torvalds offset += len; 12491da177e4SLinus Torvalds size -= len; 12501da177e4SLinus Torvalds } 12511da177e4SLinus Torvalds return 0; 12521da177e4SLinus Torvalds 12531da177e4SLinus Torvalds error: 1254bdc712b4SDavid S. Miller cork->length -= size; 12555e38e270SPavel Emelyanov IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 12561da177e4SLinus Torvalds return err; 12571da177e4SLinus Torvalds } 12581da177e4SLinus Torvalds 12591470ddf7SHerbert Xu static void ip_cork_release(struct inet_cork *cork) 1260429f08e9SPavel Emelyanov { 12611470ddf7SHerbert Xu cork->flags &= ~IPCORK_OPT; 12621470ddf7SHerbert Xu kfree(cork->opt); 12631470ddf7SHerbert Xu cork->opt = NULL; 12641470ddf7SHerbert Xu dst_release(cork->dst); 12651470ddf7SHerbert Xu cork->dst = NULL; 1266429f08e9SPavel Emelyanov } 1267429f08e9SPavel Emelyanov 12681da177e4SLinus Torvalds /* 12691da177e4SLinus Torvalds * Combined all pending IP fragments on the socket as one IP datagram 12701da177e4SLinus Torvalds * and push them out. 12711da177e4SLinus Torvalds */ 12721c32c5adSHerbert Xu struct sk_buff *__ip_make_skb(struct sock *sk, 127377968b78SDavid S. Miller struct flowi4 *fl4, 12741470ddf7SHerbert Xu struct sk_buff_head *queue, 12751470ddf7SHerbert Xu struct inet_cork *cork) 12761da177e4SLinus Torvalds { 12771da177e4SLinus Torvalds struct sk_buff *skb, *tmp_skb; 12781da177e4SLinus Torvalds struct sk_buff **tail_skb; 12791da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 12800388b004SPavel Emelyanov struct net *net = sock_net(sk); 12811da177e4SLinus Torvalds struct ip_options *opt = NULL; 12821470ddf7SHerbert Xu struct rtable *rt = (struct rtable *)cork->dst; 12831da177e4SLinus Torvalds struct iphdr *iph; 128476ab608dSAlexey Dobriyan __be16 df = 0; 12851da177e4SLinus Torvalds __u8 ttl; 12861da177e4SLinus Torvalds 12871470ddf7SHerbert Xu if ((skb = __skb_dequeue(queue)) == NULL) 12881da177e4SLinus Torvalds goto out; 12891da177e4SLinus Torvalds tail_skb = &(skb_shinfo(skb)->frag_list); 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds /* move skb->data to ip header from ext header */ 1292d56f90a7SArnaldo Carvalho de Melo if (skb->data < skb_network_header(skb)) 1293bbe735e4SArnaldo Carvalho de Melo __skb_pull(skb, skb_network_offset(skb)); 12941470ddf7SHerbert Xu while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1295cfe1fc77SArnaldo Carvalho de Melo __skb_pull(tmp_skb, skb_network_header_len(skb)); 12961da177e4SLinus Torvalds *tail_skb = tmp_skb; 12971da177e4SLinus Torvalds tail_skb = &(tmp_skb->next); 12981da177e4SLinus Torvalds skb->len += tmp_skb->len; 12991da177e4SLinus Torvalds skb->data_len += tmp_skb->len; 13001da177e4SLinus Torvalds skb->truesize += tmp_skb->truesize; 13011da177e4SLinus Torvalds tmp_skb->destructor = NULL; 13021da177e4SLinus Torvalds tmp_skb->sk = NULL; 13031da177e4SLinus Torvalds } 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 13061da177e4SLinus Torvalds * to fragment the frame generated here. No matter, what transforms 13071da177e4SLinus Torvalds * how transforms change size of the packet, it will come out. 13081da177e4SLinus Torvalds */ 1309628a5c56SJohn Heffner if (inet->pmtudisc < IP_PMTUDISC_DO) 13101da177e4SLinus Torvalds skb->local_df = 1; 13111da177e4SLinus Torvalds 13121da177e4SLinus Torvalds /* DF bit is set when we want to see DF on outgoing frames. 13131da177e4SLinus Torvalds * If local_df is set too, we still allow to fragment this frame 13141da177e4SLinus Torvalds * locally. */ 1315628a5c56SJohn Heffner if (inet->pmtudisc >= IP_PMTUDISC_DO || 1316d8d1f30bSChangli Gao (skb->len <= dst_mtu(&rt->dst) && 1317d8d1f30bSChangli Gao ip_dont_fragment(sk, &rt->dst))) 13181da177e4SLinus Torvalds df = htons(IP_DF); 13191da177e4SLinus Torvalds 13201470ddf7SHerbert Xu if (cork->flags & IPCORK_OPT) 13211470ddf7SHerbert Xu opt = cork->opt; 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds if (rt->rt_type == RTN_MULTICAST) 13241da177e4SLinus Torvalds ttl = inet->mc_ttl; 13251da177e4SLinus Torvalds else 1326d8d1f30bSChangli Gao ttl = ip_select_ttl(inet, &rt->dst); 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds iph = (struct iphdr *)skb->data; 13291da177e4SLinus Torvalds iph->version = 4; 13301da177e4SLinus Torvalds iph->ihl = 5; 13311da177e4SLinus Torvalds iph->tos = inet->tos; 13321da177e4SLinus Torvalds iph->frag_off = df; 1333d8d1f30bSChangli Gao ip_select_ident(iph, &rt->dst, sk); 13341da177e4SLinus Torvalds iph->ttl = ttl; 13351da177e4SLinus Torvalds iph->protocol = sk->sk_protocol; 133677968b78SDavid S. Miller iph->saddr = fl4->saddr; 133777968b78SDavid S. Miller iph->daddr = fl4->daddr; 13381da177e4SLinus Torvalds 133922f728f8SDavid S. Miller if (opt) { 134022f728f8SDavid S. Miller iph->ihl += opt->optlen>>2; 134122f728f8SDavid S. Miller ip_options_build(skb, opt, cork->addr, rt, 0); 134222f728f8SDavid S. Miller } 134322f728f8SDavid S. Miller 13441da177e4SLinus Torvalds skb->priority = sk->sk_priority; 13454a19ec58SLaszlo Attila Toth skb->mark = sk->sk_mark; 1346a21bba94SEric Dumazet /* 1347a21bba94SEric Dumazet * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1348a21bba94SEric Dumazet * on dst refcount 1349a21bba94SEric Dumazet */ 13501470ddf7SHerbert Xu cork->dst = NULL; 1351d8d1f30bSChangli Gao skb_dst_set(skb, &rt->dst); 13521da177e4SLinus Torvalds 135396793b48SDavid L Stevens if (iph->protocol == IPPROTO_ICMP) 13540388b004SPavel Emelyanov icmp_out_count(net, ((struct icmphdr *) 135596793b48SDavid L Stevens skb_transport_header(skb))->type); 135696793b48SDavid L Stevens 13571c32c5adSHerbert Xu ip_cork_release(cork); 13581c32c5adSHerbert Xu out: 13591c32c5adSHerbert Xu return skb; 13601c32c5adSHerbert Xu } 13611c32c5adSHerbert Xu 13621c32c5adSHerbert Xu int ip_send_skb(struct sk_buff *skb) 13631c32c5adSHerbert Xu { 13641c32c5adSHerbert Xu struct net *net = sock_net(skb->sk); 13651c32c5adSHerbert Xu int err; 13661c32c5adSHerbert Xu 1367c439cb2eSHerbert Xu err = ip_local_out(skb); 13681da177e4SLinus Torvalds if (err) { 13691da177e4SLinus Torvalds if (err > 0) 13706ce9e7b5SEric Dumazet err = net_xmit_errno(err); 13711da177e4SLinus Torvalds if (err) 13721c32c5adSHerbert Xu IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds return err; 13761da177e4SLinus Torvalds } 13771da177e4SLinus Torvalds 137877968b78SDavid S. Miller int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 13791470ddf7SHerbert Xu { 13801c32c5adSHerbert Xu struct sk_buff *skb; 13811c32c5adSHerbert Xu 138277968b78SDavid S. Miller skb = ip_finish_skb(sk, fl4); 13831c32c5adSHerbert Xu if (!skb) 13841c32c5adSHerbert Xu return 0; 13851c32c5adSHerbert Xu 13861c32c5adSHerbert Xu /* Netfilter gets whole the not fragmented skb. */ 13871c32c5adSHerbert Xu return ip_send_skb(skb); 13881470ddf7SHerbert Xu } 13891470ddf7SHerbert Xu 13901da177e4SLinus Torvalds /* 13911da177e4SLinus Torvalds * Throw away all pending data on the socket. 13921da177e4SLinus Torvalds */ 13931470ddf7SHerbert Xu static void __ip_flush_pending_frames(struct sock *sk, 13941470ddf7SHerbert Xu struct sk_buff_head *queue, 13951470ddf7SHerbert Xu struct inet_cork *cork) 13961da177e4SLinus Torvalds { 13971da177e4SLinus Torvalds struct sk_buff *skb; 13981da177e4SLinus Torvalds 13991470ddf7SHerbert Xu while ((skb = __skb_dequeue_tail(queue)) != NULL) 14001da177e4SLinus Torvalds kfree_skb(skb); 14011da177e4SLinus Torvalds 14021470ddf7SHerbert Xu ip_cork_release(cork); 14031470ddf7SHerbert Xu } 14041470ddf7SHerbert Xu 14051470ddf7SHerbert Xu void ip_flush_pending_frames(struct sock *sk) 14061470ddf7SHerbert Xu { 1407bdc712b4SDavid S. Miller __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 14081da177e4SLinus Torvalds } 14091da177e4SLinus Torvalds 14101c32c5adSHerbert Xu struct sk_buff *ip_make_skb(struct sock *sk, 141177968b78SDavid S. Miller struct flowi4 *fl4, 14121c32c5adSHerbert Xu int getfrag(void *from, char *to, int offset, 14131c32c5adSHerbert Xu int len, int odd, struct sk_buff *skb), 14141c32c5adSHerbert Xu void *from, int length, int transhdrlen, 14151c32c5adSHerbert Xu struct ipcm_cookie *ipc, struct rtable **rtp, 14161c32c5adSHerbert Xu unsigned int flags) 14171c32c5adSHerbert Xu { 1418b80d7226SDavid S. Miller struct inet_cork cork; 14191c32c5adSHerbert Xu struct sk_buff_head queue; 14201c32c5adSHerbert Xu int err; 14211c32c5adSHerbert Xu 14221c32c5adSHerbert Xu if (flags & MSG_PROBE) 14231c32c5adSHerbert Xu return NULL; 14241c32c5adSHerbert Xu 14251c32c5adSHerbert Xu __skb_queue_head_init(&queue); 14261c32c5adSHerbert Xu 1427b80d7226SDavid S. Miller cork.flags = 0; 1428b80d7226SDavid S. Miller cork.addr = 0; 142970652728SDavid S. Miller cork.opt = NULL; 14301c32c5adSHerbert Xu err = ip_setup_cork(sk, &cork, ipc, rtp); 14311c32c5adSHerbert Xu if (err) 14321c32c5adSHerbert Xu return ERR_PTR(err); 14331c32c5adSHerbert Xu 1434f5fca608SDavid S. Miller err = __ip_append_data(sk, fl4, &queue, &cork, getfrag, 14351c32c5adSHerbert Xu from, length, transhdrlen, flags); 14361c32c5adSHerbert Xu if (err) { 14371c32c5adSHerbert Xu __ip_flush_pending_frames(sk, &queue, &cork); 14381c32c5adSHerbert Xu return ERR_PTR(err); 14391c32c5adSHerbert Xu } 14401c32c5adSHerbert Xu 144177968b78SDavid S. Miller return __ip_make_skb(sk, fl4, &queue, &cork); 14421c32c5adSHerbert Xu } 14431da177e4SLinus Torvalds 14441da177e4SLinus Torvalds /* 14451da177e4SLinus Torvalds * Fetch data from kernel space and fill in checksum if needed. 14461da177e4SLinus Torvalds */ 14471da177e4SLinus Torvalds static int ip_reply_glue_bits(void *dptr, char *to, int offset, 14481da177e4SLinus Torvalds int len, int odd, struct sk_buff *skb) 14491da177e4SLinus Torvalds { 14505084205fSAl Viro __wsum csum; 14511da177e4SLinus Torvalds 14521da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 14531da177e4SLinus Torvalds skb->csum = csum_block_add(skb->csum, csum, odd); 14541da177e4SLinus Torvalds return 0; 14551da177e4SLinus Torvalds } 14561da177e4SLinus Torvalds 14571da177e4SLinus Torvalds /* 14581da177e4SLinus Torvalds * Generic function to send a packet as reply to another packet. 14591da177e4SLinus Torvalds * Used to send TCP resets so far. ICMP should use this function too. 14601da177e4SLinus Torvalds * 14611da177e4SLinus Torvalds * Should run single threaded per socket because it uses the sock 14621da177e4SLinus Torvalds * structure to pass arguments. 14631da177e4SLinus Torvalds */ 14640a5ebb80SDavid S. Miller void ip_send_reply(struct sock *sk, struct sk_buff *skb, __be32 daddr, 14650a5ebb80SDavid S. Miller struct ip_reply_arg *arg, unsigned int len) 14661da177e4SLinus Torvalds { 14671da177e4SLinus Torvalds struct inet_sock *inet = inet_sk(sk); 1468f6d8bd05SEric Dumazet struct ip_options_data replyopts; 14691da177e4SLinus Torvalds struct ipcm_cookie ipc; 147077968b78SDavid S. Miller struct flowi4 fl4; 1471511c3f92SEric Dumazet struct rtable *rt = skb_rtable(skb); 14721da177e4SLinus Torvalds 1473f6d8bd05SEric Dumazet if (ip_options_echo(&replyopts.opt.opt, skb)) 14741da177e4SLinus Torvalds return; 14751da177e4SLinus Torvalds 14760a5ebb80SDavid S. Miller ipc.addr = daddr; 14771da177e4SLinus Torvalds ipc.opt = NULL; 14782244d07bSOliver Hartkopp ipc.tx_flags = 0; 14791da177e4SLinus Torvalds 1480f6d8bd05SEric Dumazet if (replyopts.opt.opt.optlen) { 14811da177e4SLinus Torvalds ipc.opt = &replyopts.opt; 14821da177e4SLinus Torvalds 1483f6d8bd05SEric Dumazet if (replyopts.opt.opt.srr) 1484f6d8bd05SEric Dumazet daddr = replyopts.opt.opt.faddr; 14851da177e4SLinus Torvalds } 14861da177e4SLinus Torvalds 1487538de0e0SDavid S. Miller flowi4_init_output(&fl4, arg->bound_dev_if, 0, 1488538de0e0SDavid S. Miller RT_TOS(ip_hdr(skb)->tos), 1489538de0e0SDavid S. Miller RT_SCOPE_UNIVERSE, sk->sk_protocol, 1490538de0e0SDavid S. Miller ip_reply_arg_flowi_flags(arg), 1491538de0e0SDavid S. Miller daddr, rt->rt_spec_dst, 1492538de0e0SDavid S. Miller tcp_hdr(skb)->source, tcp_hdr(skb)->dest); 14939d6ec938SDavid S. Miller security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 14949d6ec938SDavid S. Miller rt = ip_route_output_key(sock_net(sk), &fl4); 1495b23dd4feSDavid S. Miller if (IS_ERR(rt)) 14961da177e4SLinus Torvalds return; 14971da177e4SLinus Torvalds 14981da177e4SLinus Torvalds /* And let IP do all the hard work. 14991da177e4SLinus Torvalds 15001da177e4SLinus Torvalds This chunk is not reenterable, hence spinlock. 15011da177e4SLinus Torvalds Note that it uses the fact, that this function is called 15021da177e4SLinus Torvalds with locally disabled BH and that sk cannot be already spinlocked. 15031da177e4SLinus Torvalds */ 15041da177e4SLinus Torvalds bh_lock_sock(sk); 1505eddc9ec5SArnaldo Carvalho de Melo inet->tos = ip_hdr(skb)->tos; 15061da177e4SLinus Torvalds sk->sk_priority = skb->priority; 1507eddc9ec5SArnaldo Carvalho de Melo sk->sk_protocol = ip_hdr(skb)->protocol; 1508f0e48dbfSPatrick McHardy sk->sk_bound_dev_if = arg->bound_dev_if; 1509f5fca608SDavid S. Miller ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0, 15102e77d89bSEric Dumazet &ipc, &rt, MSG_DONTWAIT); 15111da177e4SLinus Torvalds if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 15121da177e4SLinus Torvalds if (arg->csumoffset >= 0) 15139c70220bSArnaldo Carvalho de Melo *((__sum16 *)skb_transport_header(skb) + 15149c70220bSArnaldo Carvalho de Melo arg->csumoffset) = csum_fold(csum_add(skb->csum, 15159c70220bSArnaldo Carvalho de Melo arg->csum)); 15161da177e4SLinus Torvalds skb->ip_summed = CHECKSUM_NONE; 151777968b78SDavid S. Miller ip_push_pending_frames(sk, &fl4); 15181da177e4SLinus Torvalds } 15191da177e4SLinus Torvalds 15201da177e4SLinus Torvalds bh_unlock_sock(sk); 15211da177e4SLinus Torvalds 15221da177e4SLinus Torvalds ip_rt_put(rt); 15231da177e4SLinus Torvalds } 15241da177e4SLinus Torvalds 15251da177e4SLinus Torvalds void __init ip_init(void) 15261da177e4SLinus Torvalds { 15271da177e4SLinus Torvalds ip_rt_init(); 15281da177e4SLinus Torvalds inet_initpeers(); 15291da177e4SLinus Torvalds 15301da177e4SLinus Torvalds #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS) 15311da177e4SLinus Torvalds igmp_mc_proc_init(); 15321da177e4SLinus Torvalds #endif 15331da177e4SLinus Torvalds } 1534