Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * INET     An implementation of the TCP/IP protocol suite for the LINUX
0004  *      operating system.  INET is implemented using the  BSD Socket
0005  *      interface as the means of communication with the user level.
0006  *
0007  *      The Internet Protocol (IP) output module.
0008  *
0009  * Authors: Ross Biro
0010  *      Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
0011  *      Donald Becker, <becker@super.org>
0012  *      Alan Cox, <Alan.Cox@linux.org>
0013  *      Richard Underwood
0014  *      Stefan Becker, <stefanb@yello.ping.de>
0015  *      Jorge Cwik, <jorge@laser.satlink.net>
0016  *      Arnt Gulbrandsen, <agulbra@nvg.unit.no>
0017  *      Hirokazu Takahashi, <taka@valinux.co.jp>
0018  *
0019  *  See ip_input.c for original log
0020  *
0021  *  Fixes:
0022  *      Alan Cox    :   Missing nonblock feature in ip_build_xmit.
0023  *      Mike Kilburn    :   htons() missing in ip_build_xmit.
0024  *      Bradford Johnson:   Fix faulty handling of some frames when
0025  *                  no route is found.
0026  *      Alexander Demenshin:    Missing sk/skb free in ip_queue_xmit
0027  *                  (in case if packet not accepted by
0028  *                  output firewall rules)
0029  *      Mike McLagan    :   Routing by source
0030  *      Alexey Kuznetsov:   use new route cache
0031  *      Andi Kleen:     Fix broken PMTU recovery and remove
0032  *                  some redundant tests.
0033  *  Vitaly E. Lavrov    :   Transparent proxy revived after year coma.
0034  *      Andi Kleen  :   Replace ip_reply with ip_send_reply.
0035  *      Andi Kleen  :   Split fast and slow ip_build_xmit path
0036  *                  for decreased register pressure on x86
0037  *                  and more readability.
0038  *      Marc Boucher    :   When call_out_firewall returns FW_QUEUE,
0039  *                  silently drop skb instead of failing with -EPERM.
0040  *      Detlev Wengorz  :   Copy protocol for fragments.
0041  *      Hirokazu Takahashi: HW checksumming for outgoing UDP
0042  *                  datagrams.
0043  *      Hirokazu Takahashi: sendfile() on UDP works now.
0044  */
0045 
0046 #include <linux/uaccess.h>
0047 #include <linux/module.h>
0048 #include <linux/types.h>
0049 #include <linux/kernel.h>
0050 #include <linux/mm.h>
0051 #include <linux/string.h>
0052 #include <linux/errno.h>
0053 #include <linux/highmem.h>
0054 #include <linux/slab.h>
0055 
0056 #include <linux/socket.h>
0057 #include <linux/sockios.h>
0058 #include <linux/in.h>
0059 #include <linux/inet.h>
0060 #include <linux/netdevice.h>
0061 #include <linux/etherdevice.h>
0062 #include <linux/proc_fs.h>
0063 #include <linux/stat.h>
0064 #include <linux/init.h>
0065 
0066 #include <net/snmp.h>
0067 #include <net/ip.h>
0068 #include <net/protocol.h>
0069 #include <net/route.h>
0070 #include <net/xfrm.h>
0071 #include <linux/skbuff.h>
0072 #include <net/sock.h>
0073 #include <net/arp.h>
0074 #include <net/icmp.h>
0075 #include <net/checksum.h>
0076 #include <net/inetpeer.h>
0077 #include <net/inet_ecn.h>
0078 #include <net/lwtunnel.h>
0079 #include <linux/bpf-cgroup.h>
0080 #include <linux/igmp.h>
0081 #include <linux/netfilter_ipv4.h>
0082 #include <linux/netfilter_bridge.h>
0083 #include <linux/netlink.h>
0084 #include <linux/tcp.h>
0085 
0086 static int
0087 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
0088         unsigned int mtu,
0089         int (*output)(struct net *, struct sock *, struct sk_buff *));
0090 
0091 /* Generate a checksum for an outgoing IP datagram. */
0092 void ip_send_check(struct iphdr *iph)
0093 {
0094     iph->check = 0;
0095     iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
0096 }
0097 EXPORT_SYMBOL(ip_send_check);
0098 
0099 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
0100 {
0101     struct iphdr *iph = ip_hdr(skb);
0102 
0103     iph->tot_len = htons(skb->len);
0104     ip_send_check(iph);
0105 
0106     /* if egress device is enslaved to an L3 master device pass the
0107      * skb to its handler for processing
0108      */
0109     skb = l3mdev_ip_out(sk, skb);
0110     if (unlikely(!skb))
0111         return 0;
0112 
0113     skb->protocol = htons(ETH_P_IP);
0114 
0115     return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
0116                net, sk, skb, NULL, skb_dst(skb)->dev,
0117                dst_output);
0118 }
0119 
0120 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
0121 {
0122     int err;
0123 
0124     err = __ip_local_out(net, sk, skb);
0125     if (likely(err == 1))
0126         err = dst_output(net, sk, skb);
0127 
0128     return err;
0129 }
0130 EXPORT_SYMBOL_GPL(ip_local_out);
0131 
0132 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
0133 {
0134     int ttl = inet->uc_ttl;
0135 
0136     if (ttl < 0)
0137         ttl = ip4_dst_hoplimit(dst);
0138     return ttl;
0139 }
0140 
0141 /*
0142  *      Add an ip header to a skbuff and send it out.
0143  *
0144  */
0145 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
0146               __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
0147               u8 tos)
0148 {
0149     struct inet_sock *inet = inet_sk(sk);
0150     struct rtable *rt = skb_rtable(skb);
0151     struct net *net = sock_net(sk);
0152     struct iphdr *iph;
0153 
0154     /* Build the IP header. */
0155     skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
0156     skb_reset_network_header(skb);
0157     iph = ip_hdr(skb);
0158     iph->version  = 4;
0159     iph->ihl      = 5;
0160     iph->tos      = tos;
0161     iph->ttl      = ip_select_ttl(inet, &rt->dst);
0162     iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
0163     iph->saddr    = saddr;
0164     iph->protocol = sk->sk_protocol;
0165     /* Do not bother generating IPID for small packets (eg SYNACK) */
0166     if (skb->len <= IPV4_MIN_MTU || ip_dont_fragment(sk, &rt->dst)) {
0167         iph->frag_off = htons(IP_DF);
0168         iph->id = 0;
0169     } else {
0170         iph->frag_off = 0;
0171         /* TCP packets here are SYNACK with fat IPv4/TCP options.
0172          * Avoid using the hashed IP ident generator.
0173          */
0174         if (sk->sk_protocol == IPPROTO_TCP)
0175             iph->id = (__force __be16)prandom_u32();
0176         else
0177             __ip_select_ident(net, iph, 1);
0178     }
0179 
0180     if (opt && opt->opt.optlen) {
0181         iph->ihl += opt->opt.optlen>>2;
0182         ip_options_build(skb, &opt->opt, daddr, rt);
0183     }
0184 
0185     skb->priority = sk->sk_priority;
0186     if (!skb->mark)
0187         skb->mark = sk->sk_mark;
0188 
0189     /* Send it out. */
0190     return ip_local_out(net, skb->sk, skb);
0191 }
0192 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
0193 
0194 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
0195 {
0196     struct dst_entry *dst = skb_dst(skb);
0197     struct rtable *rt = (struct rtable *)dst;
0198     struct net_device *dev = dst->dev;
0199     unsigned int hh_len = LL_RESERVED_SPACE(dev);
0200     struct neighbour *neigh;
0201     bool is_v6gw = false;
0202 
0203     if (rt->rt_type == RTN_MULTICAST) {
0204         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
0205     } else if (rt->rt_type == RTN_BROADCAST)
0206         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
0207 
0208     if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
0209         skb = skb_expand_head(skb, hh_len);
0210         if (!skb)
0211             return -ENOMEM;
0212     }
0213 
0214     if (lwtunnel_xmit_redirect(dst->lwtstate)) {
0215         int res = lwtunnel_xmit(skb);
0216 
0217         if (res < 0 || res == LWTUNNEL_XMIT_DONE)
0218             return res;
0219     }
0220 
0221     rcu_read_lock_bh();
0222     neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
0223     if (!IS_ERR(neigh)) {
0224         int res;
0225 
0226         sock_confirm_neigh(skb, neigh);
0227         /* if crossing protocols, can not use the cached header */
0228         res = neigh_output(neigh, skb, is_v6gw);
0229         rcu_read_unlock_bh();
0230         return res;
0231     }
0232     rcu_read_unlock_bh();
0233 
0234     net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
0235                 __func__);
0236     kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
0237     return -EINVAL;
0238 }
0239 
0240 static int ip_finish_output_gso(struct net *net, struct sock *sk,
0241                 struct sk_buff *skb, unsigned int mtu)
0242 {
0243     struct sk_buff *segs, *nskb;
0244     netdev_features_t features;
0245     int ret = 0;
0246 
0247     /* common case: seglen is <= mtu
0248      */
0249     if (skb_gso_validate_network_len(skb, mtu))
0250         return ip_finish_output2(net, sk, skb);
0251 
0252     /* Slowpath -  GSO segment length exceeds the egress MTU.
0253      *
0254      * This can happen in several cases:
0255      *  - Forwarding of a TCP GRO skb, when DF flag is not set.
0256      *  - Forwarding of an skb that arrived on a virtualization interface
0257      *    (virtio-net/vhost/tap) with TSO/GSO size set by other network
0258      *    stack.
0259      *  - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
0260      *    interface with a smaller MTU.
0261      *  - Arriving GRO skb (or GSO skb in a virtualized environment) that is
0262      *    bridged to a NETIF_F_TSO tunnel stacked over an interface with an
0263      *    insufficient MTU.
0264      */
0265     features = netif_skb_features(skb);
0266     BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
0267     segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
0268     if (IS_ERR_OR_NULL(segs)) {
0269         kfree_skb(skb);
0270         return -ENOMEM;
0271     }
0272 
0273     consume_skb(skb);
0274 
0275     skb_list_walk_safe(segs, segs, nskb) {
0276         int err;
0277 
0278         skb_mark_not_on_list(segs);
0279         err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
0280 
0281         if (err && ret == 0)
0282             ret = err;
0283     }
0284 
0285     return ret;
0286 }
0287 
0288 static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
0289 {
0290     unsigned int mtu;
0291 
0292 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
0293     /* Policy lookup after SNAT yielded a new policy */
0294     if (skb_dst(skb)->xfrm) {
0295         IPCB(skb)->flags |= IPSKB_REROUTED;
0296         return dst_output(net, sk, skb);
0297     }
0298 #endif
0299     mtu = ip_skb_dst_mtu(sk, skb);
0300     if (skb_is_gso(skb))
0301         return ip_finish_output_gso(net, sk, skb, mtu);
0302 
0303     if (skb->len > mtu || IPCB(skb)->frag_max_size)
0304         return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
0305 
0306     return ip_finish_output2(net, sk, skb);
0307 }
0308 
0309 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
0310 {
0311     int ret;
0312 
0313     ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
0314     switch (ret) {
0315     case NET_XMIT_SUCCESS:
0316         return __ip_finish_output(net, sk, skb);
0317     case NET_XMIT_CN:
0318         return __ip_finish_output(net, sk, skb) ? : ret;
0319     default:
0320         kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
0321         return ret;
0322     }
0323 }
0324 
0325 static int ip_mc_finish_output(struct net *net, struct sock *sk,
0326                    struct sk_buff *skb)
0327 {
0328     struct rtable *new_rt;
0329     bool do_cn = false;
0330     int ret, err;
0331 
0332     ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
0333     switch (ret) {
0334     case NET_XMIT_CN:
0335         do_cn = true;
0336         fallthrough;
0337     case NET_XMIT_SUCCESS:
0338         break;
0339     default:
0340         kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
0341         return ret;
0342     }
0343 
0344     /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
0345      * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
0346      * see ipv4_pktinfo_prepare().
0347      */
0348     new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
0349     if (new_rt) {
0350         new_rt->rt_iif = 0;
0351         skb_dst_drop(skb);
0352         skb_dst_set(skb, &new_rt->dst);
0353     }
0354 
0355     err = dev_loopback_xmit(net, sk, skb);
0356     return (do_cn && err) ? ret : err;
0357 }
0358 
0359 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
0360 {
0361     struct rtable *rt = skb_rtable(skb);
0362     struct net_device *dev = rt->dst.dev;
0363 
0364     /*
0365      *  If the indicated interface is up and running, send the packet.
0366      */
0367     IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
0368 
0369     skb->dev = dev;
0370     skb->protocol = htons(ETH_P_IP);
0371 
0372     /*
0373      *  Multicasts are looped back for other local users
0374      */
0375 
0376     if (rt->rt_flags&RTCF_MULTICAST) {
0377         if (sk_mc_loop(sk)
0378 #ifdef CONFIG_IP_MROUTE
0379         /* Small optimization: do not loopback not local frames,
0380            which returned after forwarding; they will be  dropped
0381            by ip_mr_input in any case.
0382            Note, that local frames are looped back to be delivered
0383            to local recipients.
0384 
0385            This check is duplicated in ip_mr_input at the moment.
0386          */
0387             &&
0388             ((rt->rt_flags & RTCF_LOCAL) ||
0389              !(IPCB(skb)->flags & IPSKB_FORWARDED))
0390 #endif
0391            ) {
0392             struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
0393             if (newskb)
0394                 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
0395                     net, sk, newskb, NULL, newskb->dev,
0396                     ip_mc_finish_output);
0397         }
0398 
0399         /* Multicasts with ttl 0 must not go beyond the host */
0400 
0401         if (ip_hdr(skb)->ttl == 0) {
0402             kfree_skb(skb);
0403             return 0;
0404         }
0405     }
0406 
0407     if (rt->rt_flags&RTCF_BROADCAST) {
0408         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
0409         if (newskb)
0410             NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
0411                 net, sk, newskb, NULL, newskb->dev,
0412                 ip_mc_finish_output);
0413     }
0414 
0415     return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
0416                 net, sk, skb, NULL, skb->dev,
0417                 ip_finish_output,
0418                 !(IPCB(skb)->flags & IPSKB_REROUTED));
0419 }
0420 
0421 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
0422 {
0423     struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
0424 
0425     IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
0426 
0427     skb->dev = dev;
0428     skb->protocol = htons(ETH_P_IP);
0429 
0430     return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
0431                 net, sk, skb, indev, dev,
0432                 ip_finish_output,
0433                 !(IPCB(skb)->flags & IPSKB_REROUTED));
0434 }
0435 EXPORT_SYMBOL(ip_output);
0436 
0437 /*
0438  * copy saddr and daddr, possibly using 64bit load/stores
0439  * Equivalent to :
0440  *   iph->saddr = fl4->saddr;
0441  *   iph->daddr = fl4->daddr;
0442  */
0443 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
0444 {
0445     BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
0446              offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
0447 
0448     iph->saddr = fl4->saddr;
0449     iph->daddr = fl4->daddr;
0450 }
0451 
0452 /* Note: skb->sk can be different from sk, in case of tunnels */
0453 int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
0454             __u8 tos)
0455 {
0456     struct inet_sock *inet = inet_sk(sk);
0457     struct net *net = sock_net(sk);
0458     struct ip_options_rcu *inet_opt;
0459     struct flowi4 *fl4;
0460     struct rtable *rt;
0461     struct iphdr *iph;
0462     int res;
0463 
0464     /* Skip all of this if the packet is already routed,
0465      * f.e. by something like SCTP.
0466      */
0467     rcu_read_lock();
0468     inet_opt = rcu_dereference(inet->inet_opt);
0469     fl4 = &fl->u.ip4;
0470     rt = skb_rtable(skb);
0471     if (rt)
0472         goto packet_routed;
0473 
0474     /* Make sure we can route this packet. */
0475     rt = (struct rtable *)__sk_dst_check(sk, 0);
0476     if (!rt) {
0477         __be32 daddr;
0478 
0479         /* Use correct destination address if we have options. */
0480         daddr = inet->inet_daddr;
0481         if (inet_opt && inet_opt->opt.srr)
0482             daddr = inet_opt->opt.faddr;
0483 
0484         /* If this fails, retransmit mechanism of transport layer will
0485          * keep trying until route appears or the connection times
0486          * itself out.
0487          */
0488         rt = ip_route_output_ports(net, fl4, sk,
0489                        daddr, inet->inet_saddr,
0490                        inet->inet_dport,
0491                        inet->inet_sport,
0492                        sk->sk_protocol,
0493                        RT_CONN_FLAGS_TOS(sk, tos),
0494                        sk->sk_bound_dev_if);
0495         if (IS_ERR(rt))
0496             goto no_route;
0497         sk_setup_caps(sk, &rt->dst);
0498     }
0499     skb_dst_set_noref(skb, &rt->dst);
0500 
0501 packet_routed:
0502     if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
0503         goto no_route;
0504 
0505     /* OK, we know where to send it, allocate and build IP header. */
0506     skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
0507     skb_reset_network_header(skb);
0508     iph = ip_hdr(skb);
0509     *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
0510     if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
0511         iph->frag_off = htons(IP_DF);
0512     else
0513         iph->frag_off = 0;
0514     iph->ttl      = ip_select_ttl(inet, &rt->dst);
0515     iph->protocol = sk->sk_protocol;
0516     ip_copy_addrs(iph, fl4);
0517 
0518     /* Transport layer set skb->h.foo itself. */
0519 
0520     if (inet_opt && inet_opt->opt.optlen) {
0521         iph->ihl += inet_opt->opt.optlen >> 2;
0522         ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt);
0523     }
0524 
0525     ip_select_ident_segs(net, skb, sk,
0526                  skb_shinfo(skb)->gso_segs ?: 1);
0527 
0528     /* TODO : should we use skb->sk here instead of sk ? */
0529     skb->priority = sk->sk_priority;
0530     skb->mark = sk->sk_mark;
0531 
0532     res = ip_local_out(net, sk, skb);
0533     rcu_read_unlock();
0534     return res;
0535 
0536 no_route:
0537     rcu_read_unlock();
0538     IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
0539     kfree_skb_reason(skb, SKB_DROP_REASON_IP_OUTNOROUTES);
0540     return -EHOSTUNREACH;
0541 }
0542 EXPORT_SYMBOL(__ip_queue_xmit);
0543 
0544 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
0545 {
0546     return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
0547 }
0548 EXPORT_SYMBOL(ip_queue_xmit);
0549 
0550 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
0551 {
0552     to->pkt_type = from->pkt_type;
0553     to->priority = from->priority;
0554     to->protocol = from->protocol;
0555     to->skb_iif = from->skb_iif;
0556     skb_dst_drop(to);
0557     skb_dst_copy(to, from);
0558     to->dev = from->dev;
0559     to->mark = from->mark;
0560 
0561     skb_copy_hash(to, from);
0562 
0563 #ifdef CONFIG_NET_SCHED
0564     to->tc_index = from->tc_index;
0565 #endif
0566     nf_copy(to, from);
0567     skb_ext_copy(to, from);
0568 #if IS_ENABLED(CONFIG_IP_VS)
0569     to->ipvs_property = from->ipvs_property;
0570 #endif
0571     skb_copy_secmark(to, from);
0572 }
0573 
0574 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
0575                unsigned int mtu,
0576                int (*output)(struct net *, struct sock *, struct sk_buff *))
0577 {
0578     struct iphdr *iph = ip_hdr(skb);
0579 
0580     if ((iph->frag_off & htons(IP_DF)) == 0)
0581         return ip_do_fragment(net, sk, skb, output);
0582 
0583     if (unlikely(!skb->ignore_df ||
0584              (IPCB(skb)->frag_max_size &&
0585               IPCB(skb)->frag_max_size > mtu))) {
0586         IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
0587         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
0588               htonl(mtu));
0589         kfree_skb(skb);
0590         return -EMSGSIZE;
0591     }
0592 
0593     return ip_do_fragment(net, sk, skb, output);
0594 }
0595 
0596 void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
0597               unsigned int hlen, struct ip_fraglist_iter *iter)
0598 {
0599     unsigned int first_len = skb_pagelen(skb);
0600 
0601     iter->frag = skb_shinfo(skb)->frag_list;
0602     skb_frag_list_init(skb);
0603 
0604     iter->offset = 0;
0605     iter->iph = iph;
0606     iter->hlen = hlen;
0607 
0608     skb->data_len = first_len - skb_headlen(skb);
0609     skb->len = first_len;
0610     iph->tot_len = htons(first_len);
0611     iph->frag_off = htons(IP_MF);
0612     ip_send_check(iph);
0613 }
0614 EXPORT_SYMBOL(ip_fraglist_init);
0615 
0616 void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
0617 {
0618     unsigned int hlen = iter->hlen;
0619     struct iphdr *iph = iter->iph;
0620     struct sk_buff *frag;
0621 
0622     frag = iter->frag;
0623     frag->ip_summed = CHECKSUM_NONE;
0624     skb_reset_transport_header(frag);
0625     __skb_push(frag, hlen);
0626     skb_reset_network_header(frag);
0627     memcpy(skb_network_header(frag), iph, hlen);
0628     iter->iph = ip_hdr(frag);
0629     iph = iter->iph;
0630     iph->tot_len = htons(frag->len);
0631     ip_copy_metadata(frag, skb);
0632     iter->offset += skb->len - hlen;
0633     iph->frag_off = htons(iter->offset >> 3);
0634     if (frag->next)
0635         iph->frag_off |= htons(IP_MF);
0636     /* Ready, complete checksum */
0637     ip_send_check(iph);
0638 }
0639 EXPORT_SYMBOL(ip_fraglist_prepare);
0640 
0641 void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
0642           unsigned int ll_rs, unsigned int mtu, bool DF,
0643           struct ip_frag_state *state)
0644 {
0645     struct iphdr *iph = ip_hdr(skb);
0646 
0647     state->DF = DF;
0648     state->hlen = hlen;
0649     state->ll_rs = ll_rs;
0650     state->mtu = mtu;
0651 
0652     state->left = skb->len - hlen;  /* Space per frame */
0653     state->ptr = hlen;      /* Where to start from */
0654 
0655     state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
0656     state->not_last_frag = iph->frag_off & htons(IP_MF);
0657 }
0658 EXPORT_SYMBOL(ip_frag_init);
0659 
0660 static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
0661              bool first_frag)
0662 {
0663     /* Copy the flags to each fragment. */
0664     IPCB(to)->flags = IPCB(from)->flags;
0665 
0666     /* ANK: dirty, but effective trick. Upgrade options only if
0667      * the segment to be fragmented was THE FIRST (otherwise,
0668      * options are already fixed) and make it ONCE
0669      * on the initial skb, so that all the following fragments
0670      * will inherit fixed options.
0671      */
0672     if (first_frag)
0673         ip_options_fragment(from);
0674 }
0675 
0676 struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
0677 {
0678     unsigned int len = state->left;
0679     struct sk_buff *skb2;
0680     struct iphdr *iph;
0681 
0682     /* IF: it doesn't fit, use 'mtu' - the data space left */
0683     if (len > state->mtu)
0684         len = state->mtu;
0685     /* IF: we are not sending up to and including the packet end
0686        then align the next start on an eight byte boundary */
0687     if (len < state->left)  {
0688         len &= ~7;
0689     }
0690 
0691     /* Allocate buffer */
0692     skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
0693     if (!skb2)
0694         return ERR_PTR(-ENOMEM);
0695 
0696     /*
0697      *  Set up data on packet
0698      */
0699 
0700     ip_copy_metadata(skb2, skb);
0701     skb_reserve(skb2, state->ll_rs);
0702     skb_put(skb2, len + state->hlen);
0703     skb_reset_network_header(skb2);
0704     skb2->transport_header = skb2->network_header + state->hlen;
0705 
0706     /*
0707      *  Charge the memory for the fragment to any owner
0708      *  it might possess
0709      */
0710 
0711     if (skb->sk)
0712         skb_set_owner_w(skb2, skb->sk);
0713 
0714     /*
0715      *  Copy the packet header into the new buffer.
0716      */
0717 
0718     skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
0719 
0720     /*
0721      *  Copy a block of the IP datagram.
0722      */
0723     if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
0724         BUG();
0725     state->left -= len;
0726 
0727     /*
0728      *  Fill in the new header fields.
0729      */
0730     iph = ip_hdr(skb2);
0731     iph->frag_off = htons((state->offset >> 3));
0732     if (state->DF)
0733         iph->frag_off |= htons(IP_DF);
0734 
0735     /*
0736      *  Added AC : If we are fragmenting a fragment that's not the
0737      *         last fragment then keep MF on each bit
0738      */
0739     if (state->left > 0 || state->not_last_frag)
0740         iph->frag_off |= htons(IP_MF);
0741     state->ptr += len;
0742     state->offset += len;
0743 
0744     iph->tot_len = htons(len + state->hlen);
0745 
0746     ip_send_check(iph);
0747 
0748     return skb2;
0749 }
0750 EXPORT_SYMBOL(ip_frag_next);
0751 
0752 /*
0753  *  This IP datagram is too large to be sent in one piece.  Break it up into
0754  *  smaller pieces (each of size equal to IP header plus
0755  *  a block of the data of the original IP data part) that will yet fit in a
0756  *  single device frame, and queue such a frame for sending.
0757  */
0758 
0759 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
0760            int (*output)(struct net *, struct sock *, struct sk_buff *))
0761 {
0762     struct iphdr *iph;
0763     struct sk_buff *skb2;
0764     bool mono_delivery_time = skb->mono_delivery_time;
0765     struct rtable *rt = skb_rtable(skb);
0766     unsigned int mtu, hlen, ll_rs;
0767     struct ip_fraglist_iter iter;
0768     ktime_t tstamp = skb->tstamp;
0769     struct ip_frag_state state;
0770     int err = 0;
0771 
0772     /* for offloaded checksums cleanup checksum before fragmentation */
0773     if (skb->ip_summed == CHECKSUM_PARTIAL &&
0774         (err = skb_checksum_help(skb)))
0775         goto fail;
0776 
0777     /*
0778      *  Point into the IP datagram header.
0779      */
0780 
0781     iph = ip_hdr(skb);
0782 
0783     mtu = ip_skb_dst_mtu(sk, skb);
0784     if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
0785         mtu = IPCB(skb)->frag_max_size;
0786 
0787     /*
0788      *  Setup starting values.
0789      */
0790 
0791     hlen = iph->ihl * 4;
0792     mtu = mtu - hlen;   /* Size of data space */
0793     IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
0794     ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
0795 
0796     /* When frag_list is given, use it. First, check its validity:
0797      * some transformers could create wrong frag_list or break existing
0798      * one, it is not prohibited. In this case fall back to copying.
0799      *
0800      * LATER: this step can be merged to real generation of fragments,
0801      * we can switch to copy when see the first bad fragment.
0802      */
0803     if (skb_has_frag_list(skb)) {
0804         struct sk_buff *frag, *frag2;
0805         unsigned int first_len = skb_pagelen(skb);
0806 
0807         if (first_len - hlen > mtu ||
0808             ((first_len - hlen) & 7) ||
0809             ip_is_fragment(iph) ||
0810             skb_cloned(skb) ||
0811             skb_headroom(skb) < ll_rs)
0812             goto slow_path;
0813 
0814         skb_walk_frags(skb, frag) {
0815             /* Correct geometry. */
0816             if (frag->len > mtu ||
0817                 ((frag->len & 7) && frag->next) ||
0818                 skb_headroom(frag) < hlen + ll_rs)
0819                 goto slow_path_clean;
0820 
0821             /* Partially cloned skb? */
0822             if (skb_shared(frag))
0823                 goto slow_path_clean;
0824 
0825             BUG_ON(frag->sk);
0826             if (skb->sk) {
0827                 frag->sk = skb->sk;
0828                 frag->destructor = sock_wfree;
0829             }
0830             skb->truesize -= frag->truesize;
0831         }
0832 
0833         /* Everything is OK. Generate! */
0834         ip_fraglist_init(skb, iph, hlen, &iter);
0835 
0836         for (;;) {
0837             /* Prepare header of the next frame,
0838              * before previous one went down. */
0839             if (iter.frag) {
0840                 bool first_frag = (iter.offset == 0);
0841 
0842                 IPCB(iter.frag)->flags = IPCB(skb)->flags;
0843                 ip_fraglist_prepare(skb, &iter);
0844                 if (first_frag && IPCB(skb)->opt.optlen) {
0845                     /* ipcb->opt is not populated for frags
0846                      * coming from __ip_make_skb(),
0847                      * ip_options_fragment() needs optlen
0848                      */
0849                     IPCB(iter.frag)->opt.optlen =
0850                         IPCB(skb)->opt.optlen;
0851                     ip_options_fragment(iter.frag);
0852                     ip_send_check(iter.iph);
0853                 }
0854             }
0855 
0856             skb_set_delivery_time(skb, tstamp, mono_delivery_time);
0857             err = output(net, sk, skb);
0858 
0859             if (!err)
0860                 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
0861             if (err || !iter.frag)
0862                 break;
0863 
0864             skb = ip_fraglist_next(&iter);
0865         }
0866 
0867         if (err == 0) {
0868             IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
0869             return 0;
0870         }
0871 
0872         kfree_skb_list(iter.frag);
0873 
0874         IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
0875         return err;
0876 
0877 slow_path_clean:
0878         skb_walk_frags(skb, frag2) {
0879             if (frag2 == frag)
0880                 break;
0881             frag2->sk = NULL;
0882             frag2->destructor = NULL;
0883             skb->truesize += frag2->truesize;
0884         }
0885     }
0886 
0887 slow_path:
0888     /*
0889      *  Fragment the datagram.
0890      */
0891 
0892     ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
0893              &state);
0894 
0895     /*
0896      *  Keep copying data until we run out.
0897      */
0898 
0899     while (state.left > 0) {
0900         bool first_frag = (state.offset == 0);
0901 
0902         skb2 = ip_frag_next(skb, &state);
0903         if (IS_ERR(skb2)) {
0904             err = PTR_ERR(skb2);
0905             goto fail;
0906         }
0907         ip_frag_ipcb(skb, skb2, first_frag);
0908 
0909         /*
0910          *  Put this fragment into the sending queue.
0911          */
0912         skb_set_delivery_time(skb2, tstamp, mono_delivery_time);
0913         err = output(net, sk, skb2);
0914         if (err)
0915             goto fail;
0916 
0917         IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
0918     }
0919     consume_skb(skb);
0920     IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
0921     return err;
0922 
0923 fail:
0924     kfree_skb(skb);
0925     IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
0926     return err;
0927 }
0928 EXPORT_SYMBOL(ip_do_fragment);
0929 
0930 int
0931 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
0932 {
0933     struct msghdr *msg = from;
0934 
0935     if (skb->ip_summed == CHECKSUM_PARTIAL) {
0936         if (!copy_from_iter_full(to, len, &msg->msg_iter))
0937             return -EFAULT;
0938     } else {
0939         __wsum csum = 0;
0940         if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
0941             return -EFAULT;
0942         skb->csum = csum_block_add(skb->csum, csum, odd);
0943     }
0944     return 0;
0945 }
0946 EXPORT_SYMBOL(ip_generic_getfrag);
0947 
0948 static inline __wsum
0949 csum_page(struct page *page, int offset, int copy)
0950 {
0951     char *kaddr;
0952     __wsum csum;
0953     kaddr = kmap(page);
0954     csum = csum_partial(kaddr + offset, copy, 0);
0955     kunmap(page);
0956     return csum;
0957 }
0958 
0959 static int __ip_append_data(struct sock *sk,
0960                 struct flowi4 *fl4,
0961                 struct sk_buff_head *queue,
0962                 struct inet_cork *cork,
0963                 struct page_frag *pfrag,
0964                 int getfrag(void *from, char *to, int offset,
0965                     int len, int odd, struct sk_buff *skb),
0966                 void *from, int length, int transhdrlen,
0967                 unsigned int flags)
0968 {
0969     struct inet_sock *inet = inet_sk(sk);
0970     struct ubuf_info *uarg = NULL;
0971     struct sk_buff *skb;
0972     struct ip_options *opt = cork->opt;
0973     int hh_len;
0974     int exthdrlen;
0975     int mtu;
0976     int copy;
0977     int err;
0978     int offset = 0;
0979     bool zc = false;
0980     unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
0981     int csummode = CHECKSUM_NONE;
0982     struct rtable *rt = (struct rtable *)cork->dst;
0983     unsigned int wmem_alloc_delta = 0;
0984     bool paged, extra_uref = false;
0985     u32 tskey = 0;
0986 
0987     skb = skb_peek_tail(queue);
0988 
0989     exthdrlen = !skb ? rt->dst.header_len : 0;
0990     mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
0991     paged = !!cork->gso_size;
0992 
0993     if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
0994         sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
0995         tskey = atomic_inc_return(&sk->sk_tskey) - 1;
0996 
0997     hh_len = LL_RESERVED_SPACE(rt->dst.dev);
0998 
0999     fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1000     maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1001     maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
1002 
1003     if (cork->length + length > maxnonfragsize - fragheaderlen) {
1004         ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1005                    mtu - (opt ? opt->optlen : 0));
1006         return -EMSGSIZE;
1007     }
1008 
1009     /*
1010      * transhdrlen > 0 means that this is the first fragment and we wish
1011      * it won't be fragmented in the future.
1012      */
1013     if (transhdrlen &&
1014         length + fragheaderlen <= mtu &&
1015         rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1016         (!(flags & MSG_MORE) || cork->gso_size) &&
1017         (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1018         csummode = CHECKSUM_PARTIAL;
1019 
1020     if ((flags & MSG_ZEROCOPY) && length) {
1021         struct msghdr *msg = from;
1022 
1023         if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1024             if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1025                 return -EINVAL;
1026 
1027             /* Leave uarg NULL if can't zerocopy, callers should
1028              * be able to handle it.
1029              */
1030             if ((rt->dst.dev->features & NETIF_F_SG) &&
1031                 csummode == CHECKSUM_PARTIAL) {
1032                 paged = true;
1033                 zc = true;
1034                 uarg = msg->msg_ubuf;
1035             }
1036         } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1037             uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1038             if (!uarg)
1039                 return -ENOBUFS;
1040             extra_uref = !skb_zcopy(skb);   /* only ref on new uarg */
1041             if (rt->dst.dev->features & NETIF_F_SG &&
1042                 csummode == CHECKSUM_PARTIAL) {
1043                 paged = true;
1044                 zc = true;
1045             } else {
1046                 uarg->zerocopy = 0;
1047                 skb_zcopy_set(skb, uarg, &extra_uref);
1048             }
1049         }
1050     }
1051 
1052     cork->length += length;
1053 
1054     /* So, what's going on in the loop below?
1055      *
1056      * We use calculated fragment length to generate chained skb,
1057      * each of segments is IP fragment ready for sending to network after
1058      * adding appropriate IP header.
1059      */
1060 
1061     if (!skb)
1062         goto alloc_new_skb;
1063 
1064     while (length > 0) {
1065         /* Check if the remaining data fits into current packet. */
1066         copy = mtu - skb->len;
1067         if (copy < length)
1068             copy = maxfraglen - skb->len;
1069         if (copy <= 0) {
1070             char *data;
1071             unsigned int datalen;
1072             unsigned int fraglen;
1073             unsigned int fraggap;
1074             unsigned int alloclen, alloc_extra;
1075             unsigned int pagedlen;
1076             struct sk_buff *skb_prev;
1077 alloc_new_skb:
1078             skb_prev = skb;
1079             if (skb_prev)
1080                 fraggap = skb_prev->len - maxfraglen;
1081             else
1082                 fraggap = 0;
1083 
1084             /*
1085              * If remaining data exceeds the mtu,
1086              * we know we need more fragment(s).
1087              */
1088             datalen = length + fraggap;
1089             if (datalen > mtu - fragheaderlen)
1090                 datalen = maxfraglen - fragheaderlen;
1091             fraglen = datalen + fragheaderlen;
1092             pagedlen = 0;
1093 
1094             alloc_extra = hh_len + 15;
1095             alloc_extra += exthdrlen;
1096 
1097             /* The last fragment gets additional space at tail.
1098              * Note, with MSG_MORE we overallocate on fragments,
1099              * because we have no idea what fragment will be
1100              * the last.
1101              */
1102             if (datalen == length + fraggap)
1103                 alloc_extra += rt->dst.trailer_len;
1104 
1105             if ((flags & MSG_MORE) &&
1106                 !(rt->dst.dev->features&NETIF_F_SG))
1107                 alloclen = mtu;
1108             else if (!paged &&
1109                  (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1110                   !(rt->dst.dev->features & NETIF_F_SG)))
1111                 alloclen = fraglen;
1112             else if (!zc) {
1113                 alloclen = min_t(int, fraglen, MAX_HEADER);
1114                 pagedlen = fraglen - alloclen;
1115             } else {
1116                 alloclen = fragheaderlen + transhdrlen;
1117                 pagedlen = datalen - transhdrlen;
1118             }
1119 
1120             alloclen += alloc_extra;
1121 
1122             if (transhdrlen) {
1123                 skb = sock_alloc_send_skb(sk, alloclen,
1124                         (flags & MSG_DONTWAIT), &err);
1125             } else {
1126                 skb = NULL;
1127                 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1128                     2 * sk->sk_sndbuf)
1129                     skb = alloc_skb(alloclen,
1130                             sk->sk_allocation);
1131                 if (unlikely(!skb))
1132                     err = -ENOBUFS;
1133             }
1134             if (!skb)
1135                 goto error;
1136 
1137             /*
1138              *  Fill in the control structures
1139              */
1140             skb->ip_summed = csummode;
1141             skb->csum = 0;
1142             skb_reserve(skb, hh_len);
1143 
1144             /*
1145              *  Find where to start putting bytes.
1146              */
1147             data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1148             skb_set_network_header(skb, exthdrlen);
1149             skb->transport_header = (skb->network_header +
1150                          fragheaderlen);
1151             data += fragheaderlen + exthdrlen;
1152 
1153             if (fraggap) {
1154                 skb->csum = skb_copy_and_csum_bits(
1155                     skb_prev, maxfraglen,
1156                     data + transhdrlen, fraggap);
1157                 skb_prev->csum = csum_sub(skb_prev->csum,
1158                               skb->csum);
1159                 data += fraggap;
1160                 pskb_trim_unique(skb_prev, maxfraglen);
1161             }
1162 
1163             copy = datalen - transhdrlen - fraggap - pagedlen;
1164             if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1165                 err = -EFAULT;
1166                 kfree_skb(skb);
1167                 goto error;
1168             }
1169 
1170             offset += copy;
1171             length -= copy + transhdrlen;
1172             transhdrlen = 0;
1173             exthdrlen = 0;
1174             csummode = CHECKSUM_NONE;
1175 
1176             /* only the initial fragment is time stamped */
1177             skb_shinfo(skb)->tx_flags = cork->tx_flags;
1178             cork->tx_flags = 0;
1179             skb_shinfo(skb)->tskey = tskey;
1180             tskey = 0;
1181             skb_zcopy_set(skb, uarg, &extra_uref);
1182 
1183             if ((flags & MSG_CONFIRM) && !skb_prev)
1184                 skb_set_dst_pending_confirm(skb, 1);
1185 
1186             /*
1187              * Put the packet on the pending queue.
1188              */
1189             if (!skb->destructor) {
1190                 skb->destructor = sock_wfree;
1191                 skb->sk = sk;
1192                 wmem_alloc_delta += skb->truesize;
1193             }
1194             __skb_queue_tail(queue, skb);
1195             continue;
1196         }
1197 
1198         if (copy > length)
1199             copy = length;
1200 
1201         if (!(rt->dst.dev->features&NETIF_F_SG) &&
1202             skb_tailroom(skb) >= copy) {
1203             unsigned int off;
1204 
1205             off = skb->len;
1206             if (getfrag(from, skb_put(skb, copy),
1207                     offset, copy, off, skb) < 0) {
1208                 __skb_trim(skb, off);
1209                 err = -EFAULT;
1210                 goto error;
1211             }
1212         } else if (!zc) {
1213             int i = skb_shinfo(skb)->nr_frags;
1214 
1215             err = -ENOMEM;
1216             if (!sk_page_frag_refill(sk, pfrag))
1217                 goto error;
1218 
1219             skb_zcopy_downgrade_managed(skb);
1220             if (!skb_can_coalesce(skb, i, pfrag->page,
1221                           pfrag->offset)) {
1222                 err = -EMSGSIZE;
1223                 if (i == MAX_SKB_FRAGS)
1224                     goto error;
1225 
1226                 __skb_fill_page_desc(skb, i, pfrag->page,
1227                              pfrag->offset, 0);
1228                 skb_shinfo(skb)->nr_frags = ++i;
1229                 get_page(pfrag->page);
1230             }
1231             copy = min_t(int, copy, pfrag->size - pfrag->offset);
1232             if (getfrag(from,
1233                     page_address(pfrag->page) + pfrag->offset,
1234                     offset, copy, skb->len, skb) < 0)
1235                 goto error_efault;
1236 
1237             pfrag->offset += copy;
1238             skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1239             skb_len_add(skb, copy);
1240             wmem_alloc_delta += copy;
1241         } else {
1242             err = skb_zerocopy_iter_dgram(skb, from, copy);
1243             if (err < 0)
1244                 goto error;
1245         }
1246         offset += copy;
1247         length -= copy;
1248     }
1249 
1250     if (wmem_alloc_delta)
1251         refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1252     return 0;
1253 
1254 error_efault:
1255     err = -EFAULT;
1256 error:
1257     net_zcopy_put_abort(uarg, extra_uref);
1258     cork->length -= length;
1259     IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1260     refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1261     return err;
1262 }
1263 
1264 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1265              struct ipcm_cookie *ipc, struct rtable **rtp)
1266 {
1267     struct ip_options_rcu *opt;
1268     struct rtable *rt;
1269 
1270     rt = *rtp;
1271     if (unlikely(!rt))
1272         return -EFAULT;
1273 
1274     /*
1275      * setup for corking.
1276      */
1277     opt = ipc->opt;
1278     if (opt) {
1279         if (!cork->opt) {
1280             cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1281                         sk->sk_allocation);
1282             if (unlikely(!cork->opt))
1283                 return -ENOBUFS;
1284         }
1285         memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1286         cork->flags |= IPCORK_OPT;
1287         cork->addr = ipc->addr;
1288     }
1289 
1290     cork->fragsize = ip_sk_use_pmtu(sk) ?
1291              dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1292 
1293     if (!inetdev_valid_mtu(cork->fragsize))
1294         return -ENETUNREACH;
1295 
1296     cork->gso_size = ipc->gso_size;
1297 
1298     cork->dst = &rt->dst;
1299     /* We stole this route, caller should not release it. */
1300     *rtp = NULL;
1301 
1302     cork->length = 0;
1303     cork->ttl = ipc->ttl;
1304     cork->tos = ipc->tos;
1305     cork->mark = ipc->sockc.mark;
1306     cork->priority = ipc->priority;
1307     cork->transmit_time = ipc->sockc.transmit_time;
1308     cork->tx_flags = 0;
1309     sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags);
1310 
1311     return 0;
1312 }
1313 
1314 /*
1315  *  ip_append_data() and ip_append_page() can make one large IP datagram
1316  *  from many pieces of data. Each pieces will be holded on the socket
1317  *  until ip_push_pending_frames() is called. Each piece can be a page
1318  *  or non-page data.
1319  *
1320  *  Not only UDP, other transport protocols - e.g. raw sockets - can use
1321  *  this interface potentially.
1322  *
1323  *  LATER: length must be adjusted by pad at tail, when it is required.
1324  */
1325 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1326            int getfrag(void *from, char *to, int offset, int len,
1327                    int odd, struct sk_buff *skb),
1328            void *from, int length, int transhdrlen,
1329            struct ipcm_cookie *ipc, struct rtable **rtp,
1330            unsigned int flags)
1331 {
1332     struct inet_sock *inet = inet_sk(sk);
1333     int err;
1334 
1335     if (flags&MSG_PROBE)
1336         return 0;
1337 
1338     if (skb_queue_empty(&sk->sk_write_queue)) {
1339         err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1340         if (err)
1341             return err;
1342     } else {
1343         transhdrlen = 0;
1344     }
1345 
1346     return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1347                 sk_page_frag(sk), getfrag,
1348                 from, length, transhdrlen, flags);
1349 }
1350 
1351 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1352                int offset, size_t size, int flags)
1353 {
1354     struct inet_sock *inet = inet_sk(sk);
1355     struct sk_buff *skb;
1356     struct rtable *rt;
1357     struct ip_options *opt = NULL;
1358     struct inet_cork *cork;
1359     int hh_len;
1360     int mtu;
1361     int len;
1362     int err;
1363     unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1364 
1365     if (inet->hdrincl)
1366         return -EPERM;
1367 
1368     if (flags&MSG_PROBE)
1369         return 0;
1370 
1371     if (skb_queue_empty(&sk->sk_write_queue))
1372         return -EINVAL;
1373 
1374     cork = &inet->cork.base;
1375     rt = (struct rtable *)cork->dst;
1376     if (cork->flags & IPCORK_OPT)
1377         opt = cork->opt;
1378 
1379     if (!(rt->dst.dev->features & NETIF_F_SG))
1380         return -EOPNOTSUPP;
1381 
1382     hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1383     mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
1384 
1385     fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1386     maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1387     maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1388 
1389     if (cork->length + size > maxnonfragsize - fragheaderlen) {
1390         ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1391                    mtu - (opt ? opt->optlen : 0));
1392         return -EMSGSIZE;
1393     }
1394 
1395     skb = skb_peek_tail(&sk->sk_write_queue);
1396     if (!skb)
1397         return -EINVAL;
1398 
1399     cork->length += size;
1400 
1401     while (size > 0) {
1402         /* Check if the remaining data fits into current packet. */
1403         len = mtu - skb->len;
1404         if (len < size)
1405             len = maxfraglen - skb->len;
1406 
1407         if (len <= 0) {
1408             struct sk_buff *skb_prev;
1409             int alloclen;
1410 
1411             skb_prev = skb;
1412             fraggap = skb_prev->len - maxfraglen;
1413 
1414             alloclen = fragheaderlen + hh_len + fraggap + 15;
1415             skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1416             if (unlikely(!skb)) {
1417                 err = -ENOBUFS;
1418                 goto error;
1419             }
1420 
1421             /*
1422              *  Fill in the control structures
1423              */
1424             skb->ip_summed = CHECKSUM_NONE;
1425             skb->csum = 0;
1426             skb_reserve(skb, hh_len);
1427 
1428             /*
1429              *  Find where to start putting bytes.
1430              */
1431             skb_put(skb, fragheaderlen + fraggap);
1432             skb_reset_network_header(skb);
1433             skb->transport_header = (skb->network_header +
1434                          fragheaderlen);
1435             if (fraggap) {
1436                 skb->csum = skb_copy_and_csum_bits(skb_prev,
1437                                    maxfraglen,
1438                             skb_transport_header(skb),
1439                                    fraggap);
1440                 skb_prev->csum = csum_sub(skb_prev->csum,
1441                               skb->csum);
1442                 pskb_trim_unique(skb_prev, maxfraglen);
1443             }
1444 
1445             /*
1446              * Put the packet on the pending queue.
1447              */
1448             __skb_queue_tail(&sk->sk_write_queue, skb);
1449             continue;
1450         }
1451 
1452         if (len > size)
1453             len = size;
1454 
1455         if (skb_append_pagefrags(skb, page, offset, len)) {
1456             err = -EMSGSIZE;
1457             goto error;
1458         }
1459 
1460         if (skb->ip_summed == CHECKSUM_NONE) {
1461             __wsum csum;
1462             csum = csum_page(page, offset, len);
1463             skb->csum = csum_block_add(skb->csum, csum, skb->len);
1464         }
1465 
1466         skb_len_add(skb, len);
1467         refcount_add(len, &sk->sk_wmem_alloc);
1468         offset += len;
1469         size -= len;
1470     }
1471     return 0;
1472 
1473 error:
1474     cork->length -= size;
1475     IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1476     return err;
1477 }
1478 
1479 static void ip_cork_release(struct inet_cork *cork)
1480 {
1481     cork->flags &= ~IPCORK_OPT;
1482     kfree(cork->opt);
1483     cork->opt = NULL;
1484     dst_release(cork->dst);
1485     cork->dst = NULL;
1486 }
1487 
1488 /*
1489  *  Combined all pending IP fragments on the socket as one IP datagram
1490  *  and push them out.
1491  */
1492 struct sk_buff *__ip_make_skb(struct sock *sk,
1493                   struct flowi4 *fl4,
1494                   struct sk_buff_head *queue,
1495                   struct inet_cork *cork)
1496 {
1497     struct sk_buff *skb, *tmp_skb;
1498     struct sk_buff **tail_skb;
1499     struct inet_sock *inet = inet_sk(sk);
1500     struct net *net = sock_net(sk);
1501     struct ip_options *opt = NULL;
1502     struct rtable *rt = (struct rtable *)cork->dst;
1503     struct iphdr *iph;
1504     __be16 df = 0;
1505     __u8 ttl;
1506 
1507     skb = __skb_dequeue(queue);
1508     if (!skb)
1509         goto out;
1510     tail_skb = &(skb_shinfo(skb)->frag_list);
1511 
1512     /* move skb->data to ip header from ext header */
1513     if (skb->data < skb_network_header(skb))
1514         __skb_pull(skb, skb_network_offset(skb));
1515     while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1516         __skb_pull(tmp_skb, skb_network_header_len(skb));
1517         *tail_skb = tmp_skb;
1518         tail_skb = &(tmp_skb->next);
1519         skb->len += tmp_skb->len;
1520         skb->data_len += tmp_skb->len;
1521         skb->truesize += tmp_skb->truesize;
1522         tmp_skb->destructor = NULL;
1523         tmp_skb->sk = NULL;
1524     }
1525 
1526     /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1527      * to fragment the frame generated here. No matter, what transforms
1528      * how transforms change size of the packet, it will come out.
1529      */
1530     skb->ignore_df = ip_sk_ignore_df(sk);
1531 
1532     /* DF bit is set when we want to see DF on outgoing frames.
1533      * If ignore_df is set too, we still allow to fragment this frame
1534      * locally. */
1535     if (inet->pmtudisc == IP_PMTUDISC_DO ||
1536         inet->pmtudisc == IP_PMTUDISC_PROBE ||
1537         (skb->len <= dst_mtu(&rt->dst) &&
1538          ip_dont_fragment(sk, &rt->dst)))
1539         df = htons(IP_DF);
1540 
1541     if (cork->flags & IPCORK_OPT)
1542         opt = cork->opt;
1543 
1544     if (cork->ttl != 0)
1545         ttl = cork->ttl;
1546     else if (rt->rt_type == RTN_MULTICAST)
1547         ttl = inet->mc_ttl;
1548     else
1549         ttl = ip_select_ttl(inet, &rt->dst);
1550 
1551     iph = ip_hdr(skb);
1552     iph->version = 4;
1553     iph->ihl = 5;
1554     iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1555     iph->frag_off = df;
1556     iph->ttl = ttl;
1557     iph->protocol = sk->sk_protocol;
1558     ip_copy_addrs(iph, fl4);
1559     ip_select_ident(net, skb, sk);
1560 
1561     if (opt) {
1562         iph->ihl += opt->optlen >> 2;
1563         ip_options_build(skb, opt, cork->addr, rt);
1564     }
1565 
1566     skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1567     skb->mark = cork->mark;
1568     skb->tstamp = cork->transmit_time;
1569     /*
1570      * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1571      * on dst refcount
1572      */
1573     cork->dst = NULL;
1574     skb_dst_set(skb, &rt->dst);
1575 
1576     if (iph->protocol == IPPROTO_ICMP)
1577         icmp_out_count(net, ((struct icmphdr *)
1578             skb_transport_header(skb))->type);
1579 
1580     ip_cork_release(cork);
1581 out:
1582     return skb;
1583 }
1584 
1585 int ip_send_skb(struct net *net, struct sk_buff *skb)
1586 {
1587     int err;
1588 
1589     err = ip_local_out(net, skb->sk, skb);
1590     if (err) {
1591         if (err > 0)
1592             err = net_xmit_errno(err);
1593         if (err)
1594             IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1595     }
1596 
1597     return err;
1598 }
1599 
1600 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1601 {
1602     struct sk_buff *skb;
1603 
1604     skb = ip_finish_skb(sk, fl4);
1605     if (!skb)
1606         return 0;
1607 
1608     /* Netfilter gets whole the not fragmented skb. */
1609     return ip_send_skb(sock_net(sk), skb);
1610 }
1611 
1612 /*
1613  *  Throw away all pending data on the socket.
1614  */
1615 static void __ip_flush_pending_frames(struct sock *sk,
1616                       struct sk_buff_head *queue,
1617                       struct inet_cork *cork)
1618 {
1619     struct sk_buff *skb;
1620 
1621     while ((skb = __skb_dequeue_tail(queue)) != NULL)
1622         kfree_skb(skb);
1623 
1624     ip_cork_release(cork);
1625 }
1626 
1627 void ip_flush_pending_frames(struct sock *sk)
1628 {
1629     __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1630 }
1631 
1632 struct sk_buff *ip_make_skb(struct sock *sk,
1633                 struct flowi4 *fl4,
1634                 int getfrag(void *from, char *to, int offset,
1635                     int len, int odd, struct sk_buff *skb),
1636                 void *from, int length, int transhdrlen,
1637                 struct ipcm_cookie *ipc, struct rtable **rtp,
1638                 struct inet_cork *cork, unsigned int flags)
1639 {
1640     struct sk_buff_head queue;
1641     int err;
1642 
1643     if (flags & MSG_PROBE)
1644         return NULL;
1645 
1646     __skb_queue_head_init(&queue);
1647 
1648     cork->flags = 0;
1649     cork->addr = 0;
1650     cork->opt = NULL;
1651     err = ip_setup_cork(sk, cork, ipc, rtp);
1652     if (err)
1653         return ERR_PTR(err);
1654 
1655     err = __ip_append_data(sk, fl4, &queue, cork,
1656                    &current->task_frag, getfrag,
1657                    from, length, transhdrlen, flags);
1658     if (err) {
1659         __ip_flush_pending_frames(sk, &queue, cork);
1660         return ERR_PTR(err);
1661     }
1662 
1663     return __ip_make_skb(sk, fl4, &queue, cork);
1664 }
1665 
1666 /*
1667  *  Fetch data from kernel space and fill in checksum if needed.
1668  */
1669 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1670                   int len, int odd, struct sk_buff *skb)
1671 {
1672     __wsum csum;
1673 
1674     csum = csum_partial_copy_nocheck(dptr+offset, to, len);
1675     skb->csum = csum_block_add(skb->csum, csum, odd);
1676     return 0;
1677 }
1678 
1679 /*
1680  *  Generic function to send a packet as reply to another packet.
1681  *  Used to send some TCP resets/acks so far.
1682  */
1683 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1684                const struct ip_options *sopt,
1685                __be32 daddr, __be32 saddr,
1686                const struct ip_reply_arg *arg,
1687                unsigned int len, u64 transmit_time)
1688 {
1689     struct ip_options_data replyopts;
1690     struct ipcm_cookie ipc;
1691     struct flowi4 fl4;
1692     struct rtable *rt = skb_rtable(skb);
1693     struct net *net = sock_net(sk);
1694     struct sk_buff *nskb;
1695     int err;
1696     int oif;
1697 
1698     if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
1699         return;
1700 
1701     ipcm_init(&ipc);
1702     ipc.addr = daddr;
1703     ipc.sockc.transmit_time = transmit_time;
1704 
1705     if (replyopts.opt.opt.optlen) {
1706         ipc.opt = &replyopts.opt;
1707 
1708         if (replyopts.opt.opt.srr)
1709             daddr = replyopts.opt.opt.faddr;
1710     }
1711 
1712     oif = arg->bound_dev_if;
1713     if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1714         oif = skb->skb_iif;
1715 
1716     flowi4_init_output(&fl4, oif,
1717                IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1718                RT_TOS(arg->tos),
1719                RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1720                ip_reply_arg_flowi_flags(arg),
1721                daddr, saddr,
1722                tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1723                arg->uid);
1724     security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
1725     rt = ip_route_output_flow(net, &fl4, sk);
1726     if (IS_ERR(rt))
1727         return;
1728 
1729     inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK;
1730 
1731     sk->sk_protocol = ip_hdr(skb)->protocol;
1732     sk->sk_bound_dev_if = arg->bound_dev_if;
1733     sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1734     ipc.sockc.mark = fl4.flowi4_mark;
1735     err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1736                  len, 0, &ipc, &rt, MSG_DONTWAIT);
1737     if (unlikely(err)) {
1738         ip_flush_pending_frames(sk);
1739         goto out;
1740     }
1741 
1742     nskb = skb_peek(&sk->sk_write_queue);
1743     if (nskb) {
1744         if (arg->csumoffset >= 0)
1745             *((__sum16 *)skb_transport_header(nskb) +
1746               arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1747                                 arg->csum));
1748         nskb->ip_summed = CHECKSUM_NONE;
1749         nskb->mono_delivery_time = !!transmit_time;
1750         ip_push_pending_frames(sk, &fl4);
1751     }
1752 out:
1753     ip_rt_put(rt);
1754 }
1755 
1756 void __init ip_init(void)
1757 {
1758     ip_rt_init();
1759     inet_initpeers();
1760 
1761 #if defined(CONFIG_IP_MULTICAST)
1762     igmp_mc_init();
1763 #endif
1764 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!