Version:  2.0.40 2.2.26 2.4.37 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8

Linux/net/ipv4/ip_output.c

  1 /*
  2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
  3  *              operating system.  INET is implemented using the  BSD Socket
  4  *              interface as the means of communication with the user level.
  5  *
  6  *              The Internet Protocol (IP) output module.
  7  *
  8  * Authors:     Ross Biro
  9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 10  *              Donald Becker, <becker@super.org>
 11  *              Alan Cox, <Alan.Cox@linux.org>
 12  *              Richard Underwood
 13  *              Stefan Becker, <stefanb@yello.ping.de>
 14  *              Jorge Cwik, <jorge@laser.satlink.net>
 15  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 16  *              Hirokazu Takahashi, <taka@valinux.co.jp>
 17  *
 18  *      See ip_input.c for original log
 19  *
 20  *      Fixes:
 21  *              Alan Cox        :       Missing nonblock feature in ip_build_xmit.
 22  *              Mike Kilburn    :       htons() missing in ip_build_xmit.
 23  *              Bradford Johnson:       Fix faulty handling of some frames when
 24  *                                      no route is found.
 25  *              Alexander Demenshin:    Missing sk/skb free in ip_queue_xmit
 26  *                                      (in case if packet not accepted by
 27  *                                      output firewall rules)
 28  *              Mike McLagan    :       Routing by source
 29  *              Alexey Kuznetsov:       use new route cache
 30  *              Andi Kleen:             Fix broken PMTU recovery and remove
 31  *                                      some redundant tests.
 32  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
 33  *              Andi Kleen      :       Replace ip_reply with ip_send_reply.
 34  *              Andi Kleen      :       Split fast and slow ip_build_xmit path
 35  *                                      for decreased register pressure on x86
 36  *                                      and more readibility.
 37  *              Marc Boucher    :       When call_out_firewall returns FW_QUEUE,
 38  *                                      silently drop skb instead of failing with -EPERM.
 39  *              Detlev Wengorz  :       Copy protocol for fragments.
 40  *              Hirokazu Takahashi:     HW checksumming for outgoing UDP
 41  *                                      datagrams.
 42  *              Hirokazu Takahashi:     sendfile() on UDP works now.
 43  */
 44 
 45 #include <asm/uaccess.h>
 46 #include <linux/module.h>
 47 #include <linux/types.h>
 48 #include <linux/kernel.h>
 49 #include <linux/mm.h>
 50 #include <linux/string.h>
 51 #include <linux/errno.h>
 52 #include <linux/highmem.h>
 53 #include <linux/slab.h>
 54 
 55 #include <linux/socket.h>
 56 #include <linux/sockios.h>
 57 #include <linux/in.h>
 58 #include <linux/inet.h>
 59 #include <linux/netdevice.h>
 60 #include <linux/etherdevice.h>
 61 #include <linux/proc_fs.h>
 62 #include <linux/stat.h>
 63 #include <linux/init.h>
 64 
 65 #include <net/snmp.h>
 66 #include <net/ip.h>
 67 #include <net/protocol.h>
 68 #include <net/route.h>
 69 #include <net/xfrm.h>
 70 #include <linux/skbuff.h>
 71 #include <net/sock.h>
 72 #include <net/arp.h>
 73 #include <net/icmp.h>
 74 #include <net/checksum.h>
 75 #include <net/inetpeer.h>
 76 #include <linux/igmp.h>
 77 #include <linux/netfilter_ipv4.h>
 78 #include <linux/netfilter_bridge.h>
 79 #include <linux/netlink.h>
 80 #include <linux/tcp.h>
 81 
 82 static int
 83 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 84             unsigned int mtu,
 85             int (*output)(struct net *, struct sock *, struct sk_buff *));
 86 
 87 /* Generate a checksum for an outgoing IP datagram. */
 88 void ip_send_check(struct iphdr *iph)
 89 {
 90         iph->check = 0;
 91         iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
 92 }
 93 EXPORT_SYMBOL(ip_send_check);
 94 
 95 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
 96 {
 97         struct iphdr *iph = ip_hdr(skb);
 98 
 99         iph->tot_len = htons(skb->len);
100         ip_send_check(iph);
101         return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
102                        net, sk, skb, NULL, skb_dst(skb)->dev,
103                        dst_output);
104 }
105 
106 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
107 {
108         int err;
109 
110         err = __ip_local_out(net, sk, skb);
111         if (likely(err == 1))
112                 err = dst_output(net, sk, skb);
113 
114         return err;
115 }
116 EXPORT_SYMBOL_GPL(ip_local_out);
117 
118 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
119 {
120         int ttl = inet->uc_ttl;
121 
122         if (ttl < 0)
123                 ttl = ip4_dst_hoplimit(dst);
124         return ttl;
125 }
126 
127 /*
128  *              Add an ip header to a skbuff and send it out.
129  *
130  */
131 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
132                           __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
133 {
134         struct inet_sock *inet = inet_sk(sk);
135         struct rtable *rt = skb_rtable(skb);
136         struct net *net = sock_net(sk);
137         struct iphdr *iph;
138 
139         /* Build the IP header. */
140         skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
141         skb_reset_network_header(skb);
142         iph = ip_hdr(skb);
143         iph->version  = 4;
144         iph->ihl      = 5;
145         iph->tos      = inet->tos;
146         iph->ttl      = ip_select_ttl(inet, &rt->dst);
147         iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
148         iph->saddr    = saddr;
149         iph->protocol = sk->sk_protocol;
150         if (ip_dont_fragment(sk, &rt->dst)) {
151                 iph->frag_off = htons(IP_DF);
152                 iph->id = 0;
153         } else {
154                 iph->frag_off = 0;
155                 __ip_select_ident(net, iph, 1);
156         }
157 
158         if (opt && opt->opt.optlen) {
159                 iph->ihl += opt->opt.optlen>>2;
160                 ip_options_build(skb, &opt->opt, daddr, rt, 0);
161         }
162 
163         skb->priority = sk->sk_priority;
164         skb->mark = sk->sk_mark;
165 
166         /* Send it out. */
167         return ip_local_out(net, skb->sk, skb);
168 }
169 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
170 
171 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
172 {
173         struct dst_entry *dst = skb_dst(skb);
174         struct rtable *rt = (struct rtable *)dst;
175         struct net_device *dev = dst->dev;
176         unsigned int hh_len = LL_RESERVED_SPACE(dev);
177         struct neighbour *neigh;
178         u32 nexthop;
179 
180         if (rt->rt_type == RTN_MULTICAST) {
181                 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
182         } else if (rt->rt_type == RTN_BROADCAST)
183                 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
184 
185         /* Be paranoid, rather than too clever. */
186         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
187                 struct sk_buff *skb2;
188 
189                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
190                 if (!skb2) {
191                         kfree_skb(skb);
192                         return -ENOMEM;
193                 }
194                 if (skb->sk)
195                         skb_set_owner_w(skb2, skb->sk);
196                 consume_skb(skb);
197                 skb = skb2;
198         }
199 
200         rcu_read_lock_bh();
201         nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
202         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
203         if (unlikely(!neigh))
204                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
205         if (!IS_ERR(neigh)) {
206                 int res = dst_neigh_output(dst, neigh, skb);
207 
208                 rcu_read_unlock_bh();
209                 return res;
210         }
211         rcu_read_unlock_bh();
212 
213         net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
214                             __func__);
215         kfree_skb(skb);
216         return -EINVAL;
217 }
218 
219 static int ip_finish_output_gso(struct net *net, struct sock *sk,
220                                 struct sk_buff *skb, unsigned int mtu)
221 {
222         netdev_features_t features;
223         struct sk_buff *segs;
224         int ret = 0;
225 
226         /* common case: fragmentation of segments is not allowed,
227          * or seglen is <= mtu
228          */
229         if (((IPCB(skb)->flags & IPSKB_FRAG_SEGS) == 0) ||
230               skb_gso_validate_mtu(skb, mtu))
231                 return ip_finish_output2(net, sk, skb);
232 
233         /* Slowpath -  GSO segment length is exceeding the dst MTU.
234          *
235          * This can happen in two cases:
236          * 1) TCP GRO packet, DF bit not set
237          * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
238          * from host network stack.
239          */
240         features = netif_skb_features(skb);
241         BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
242         segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
243         if (IS_ERR_OR_NULL(segs)) {
244                 kfree_skb(skb);
245                 return -ENOMEM;
246         }
247 
248         consume_skb(skb);
249 
250         do {
251                 struct sk_buff *nskb = segs->next;
252                 int err;
253 
254                 segs->next = NULL;
255                 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
256 
257                 if (err && ret == 0)
258                         ret = err;
259                 segs = nskb;
260         } while (segs);
261 
262         return ret;
263 }
264 
265 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
266 {
267         unsigned int mtu;
268 
269 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
270         /* Policy lookup after SNAT yielded a new policy */
271         if (skb_dst(skb)->xfrm) {
272                 IPCB(skb)->flags |= IPSKB_REROUTED;
273                 return dst_output(net, sk, skb);
274         }
275 #endif
276         mtu = ip_skb_dst_mtu(sk, skb);
277         if (skb_is_gso(skb))
278                 return ip_finish_output_gso(net, sk, skb, mtu);
279 
280         if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
281                 return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
282 
283         return ip_finish_output2(net, sk, skb);
284 }
285 
286 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
287 {
288         struct rtable *rt = skb_rtable(skb);
289         struct net_device *dev = rt->dst.dev;
290 
291         /*
292          *      If the indicated interface is up and running, send the packet.
293          */
294         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
295 
296         skb->dev = dev;
297         skb->protocol = htons(ETH_P_IP);
298 
299         /*
300          *      Multicasts are looped back for other local users
301          */
302 
303         if (rt->rt_flags&RTCF_MULTICAST) {
304                 if (sk_mc_loop(sk)
305 #ifdef CONFIG_IP_MROUTE
306                 /* Small optimization: do not loopback not local frames,
307                    which returned after forwarding; they will be  dropped
308                    by ip_mr_input in any case.
309                    Note, that local frames are looped back to be delivered
310                    to local recipients.
311 
312                    This check is duplicated in ip_mr_input at the moment.
313                  */
314                     &&
315                     ((rt->rt_flags & RTCF_LOCAL) ||
316                      !(IPCB(skb)->flags & IPSKB_FORWARDED))
317 #endif
318                    ) {
319                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
320                         if (newskb)
321                                 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
322                                         net, sk, newskb, NULL, newskb->dev,
323                                         dev_loopback_xmit);
324                 }
325 
326                 /* Multicasts with ttl 0 must not go beyond the host */
327 
328                 if (ip_hdr(skb)->ttl == 0) {
329                         kfree_skb(skb);
330                         return 0;
331                 }
332         }
333 
334         if (rt->rt_flags&RTCF_BROADCAST) {
335                 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
336                 if (newskb)
337                         NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
338                                 net, sk, newskb, NULL, newskb->dev,
339                                 dev_loopback_xmit);
340         }
341 
342         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
343                             net, sk, skb, NULL, skb->dev,
344                             ip_finish_output,
345                             !(IPCB(skb)->flags & IPSKB_REROUTED));
346 }
347 
348 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
349 {
350         struct net_device *dev = skb_dst(skb)->dev;
351 
352         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
353 
354         skb->dev = dev;
355         skb->protocol = htons(ETH_P_IP);
356 
357         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
358                             net, sk, skb, NULL, dev,
359                             ip_finish_output,
360                             !(IPCB(skb)->flags & IPSKB_REROUTED));
361 }
362 
363 /*
364  * copy saddr and daddr, possibly using 64bit load/stores
365  * Equivalent to :
366  *   iph->saddr = fl4->saddr;
367  *   iph->daddr = fl4->daddr;
368  */
369 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
370 {
371         BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
372                      offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
373         memcpy(&iph->saddr, &fl4->saddr,
374                sizeof(fl4->saddr) + sizeof(fl4->daddr));
375 }
376 
377 /* Note: skb->sk can be different from sk, in case of tunnels */
378 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
379 {
380         struct inet_sock *inet = inet_sk(sk);
381         struct net *net = sock_net(sk);
382         struct ip_options_rcu *inet_opt;
383         struct flowi4 *fl4;
384         struct rtable *rt;
385         struct iphdr *iph;
386         int res;
387 
388         /* Skip all of this if the packet is already routed,
389          * f.e. by something like SCTP.
390          */
391         rcu_read_lock();
392         inet_opt = rcu_dereference(inet->inet_opt);
393         fl4 = &fl->u.ip4;
394         rt = skb_rtable(skb);
395         if (rt)
396                 goto packet_routed;
397 
398         /* Make sure we can route this packet. */
399         rt = (struct rtable *)__sk_dst_check(sk, 0);
400         if (!rt) {
401                 __be32 daddr;
402 
403                 /* Use correct destination address if we have options. */
404                 daddr = inet->inet_daddr;
405                 if (inet_opt && inet_opt->opt.srr)
406                         daddr = inet_opt->opt.faddr;
407 
408                 /* If this fails, retransmit mechanism of transport layer will
409                  * keep trying until route appears or the connection times
410                  * itself out.
411                  */
412                 rt = ip_route_output_ports(net, fl4, sk,
413                                            daddr, inet->inet_saddr,
414                                            inet->inet_dport,
415                                            inet->inet_sport,
416                                            sk->sk_protocol,
417                                            RT_CONN_FLAGS(sk),
418                                            sk->sk_bound_dev_if);
419                 if (IS_ERR(rt))
420                         goto no_route;
421                 sk_setup_caps(sk, &rt->dst);
422         }
423         skb_dst_set_noref(skb, &rt->dst);
424 
425 packet_routed:
426         if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
427                 goto no_route;
428 
429         /* OK, we know where to send it, allocate and build IP header. */
430         skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
431         skb_reset_network_header(skb);
432         iph = ip_hdr(skb);
433         *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
434         if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
435                 iph->frag_off = htons(IP_DF);
436         else
437                 iph->frag_off = 0;
438         iph->ttl      = ip_select_ttl(inet, &rt->dst);
439         iph->protocol = sk->sk_protocol;
440         ip_copy_addrs(iph, fl4);
441 
442         /* Transport layer set skb->h.foo itself. */
443 
444         if (inet_opt && inet_opt->opt.optlen) {
445                 iph->ihl += inet_opt->opt.optlen >> 2;
446                 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
447         }
448 
449         ip_select_ident_segs(net, skb, sk,
450                              skb_shinfo(skb)->gso_segs ?: 1);
451 
452         /* TODO : should we use skb->sk here instead of sk ? */
453         skb->priority = sk->sk_priority;
454         skb->mark = sk->sk_mark;
455 
456         res = ip_local_out(net, sk, skb);
457         rcu_read_unlock();
458         return res;
459 
460 no_route:
461         rcu_read_unlock();
462         IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
463         kfree_skb(skb);
464         return -EHOSTUNREACH;
465 }
466 EXPORT_SYMBOL(ip_queue_xmit);
467 
468 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
469 {
470         to->pkt_type = from->pkt_type;
471         to->priority = from->priority;
472         to->protocol = from->protocol;
473         skb_dst_drop(to);
474         skb_dst_copy(to, from);
475         to->dev = from->dev;
476         to->mark = from->mark;
477 
478         /* Copy the flags to each fragment. */
479         IPCB(to)->flags = IPCB(from)->flags;
480 
481 #ifdef CONFIG_NET_SCHED
482         to->tc_index = from->tc_index;
483 #endif
484         nf_copy(to, from);
485 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
486         to->ipvs_property = from->ipvs_property;
487 #endif
488         skb_copy_secmark(to, from);
489 }
490 
491 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
492                        unsigned int mtu,
493                        int (*output)(struct net *, struct sock *, struct sk_buff *))
494 {
495         struct iphdr *iph = ip_hdr(skb);
496 
497         if ((iph->frag_off & htons(IP_DF)) == 0)
498                 return ip_do_fragment(net, sk, skb, output);
499 
500         if (unlikely(!skb->ignore_df ||
501                      (IPCB(skb)->frag_max_size &&
502                       IPCB(skb)->frag_max_size > mtu))) {
503                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
504                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
505                           htonl(mtu));
506                 kfree_skb(skb);
507                 return -EMSGSIZE;
508         }
509 
510         return ip_do_fragment(net, sk, skb, output);
511 }
512 
513 /*
514  *      This IP datagram is too large to be sent in one piece.  Break it up into
515  *      smaller pieces (each of size equal to IP header plus
516  *      a block of the data of the original IP data part) that will yet fit in a
517  *      single device frame, and queue such a frame for sending.
518  */
519 
520 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
521                    int (*output)(struct net *, struct sock *, struct sk_buff *))
522 {
523         struct iphdr *iph;
524         int ptr;
525         struct net_device *dev;
526         struct sk_buff *skb2;
527         unsigned int mtu, hlen, left, len, ll_rs;
528         int offset;
529         __be16 not_last_frag;
530         struct rtable *rt = skb_rtable(skb);
531         int err = 0;
532 
533         dev = rt->dst.dev;
534 
535         /* for offloaded checksums cleanup checksum before fragmentation */
536         if (skb->ip_summed == CHECKSUM_PARTIAL &&
537             (err = skb_checksum_help(skb)))
538                 goto fail;
539 
540         /*
541          *      Point into the IP datagram header.
542          */
543 
544         iph = ip_hdr(skb);
545 
546         mtu = ip_skb_dst_mtu(sk, skb);
547         if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
548                 mtu = IPCB(skb)->frag_max_size;
549 
550         /*
551          *      Setup starting values.
552          */
553 
554         hlen = iph->ihl * 4;
555         mtu = mtu - hlen;       /* Size of data space */
556         IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
557 
558         /* When frag_list is given, use it. First, check its validity:
559          * some transformers could create wrong frag_list or break existing
560          * one, it is not prohibited. In this case fall back to copying.
561          *
562          * LATER: this step can be merged to real generation of fragments,
563          * we can switch to copy when see the first bad fragment.
564          */
565         if (skb_has_frag_list(skb)) {
566                 struct sk_buff *frag, *frag2;
567                 int first_len = skb_pagelen(skb);
568 
569                 if (first_len - hlen > mtu ||
570                     ((first_len - hlen) & 7) ||
571                     ip_is_fragment(iph) ||
572                     skb_cloned(skb))
573                         goto slow_path;
574 
575                 skb_walk_frags(skb, frag) {
576                         /* Correct geometry. */
577                         if (frag->len > mtu ||
578                             ((frag->len & 7) && frag->next) ||
579                             skb_headroom(frag) < hlen)
580                                 goto slow_path_clean;
581 
582                         /* Partially cloned skb? */
583                         if (skb_shared(frag))
584                                 goto slow_path_clean;
585 
586                         BUG_ON(frag->sk);
587                         if (skb->sk) {
588                                 frag->sk = skb->sk;
589                                 frag->destructor = sock_wfree;
590                         }
591                         skb->truesize -= frag->truesize;
592                 }
593 
594                 /* Everything is OK. Generate! */
595 
596                 err = 0;
597                 offset = 0;
598                 frag = skb_shinfo(skb)->frag_list;
599                 skb_frag_list_init(skb);
600                 skb->data_len = first_len - skb_headlen(skb);
601                 skb->len = first_len;
602                 iph->tot_len = htons(first_len);
603                 iph->frag_off = htons(IP_MF);
604                 ip_send_check(iph);
605 
606                 for (;;) {
607                         /* Prepare header of the next frame,
608                          * before previous one went down. */
609                         if (frag) {
610                                 frag->ip_summed = CHECKSUM_NONE;
611                                 skb_reset_transport_header(frag);
612                                 __skb_push(frag, hlen);
613                                 skb_reset_network_header(frag);
614                                 memcpy(skb_network_header(frag), iph, hlen);
615                                 iph = ip_hdr(frag);
616                                 iph->tot_len = htons(frag->len);
617                                 ip_copy_metadata(frag, skb);
618                                 if (offset == 0)
619                                         ip_options_fragment(frag);
620                                 offset += skb->len - hlen;
621                                 iph->frag_off = htons(offset>>3);
622                                 if (frag->next)
623                                         iph->frag_off |= htons(IP_MF);
624                                 /* Ready, complete checksum */
625                                 ip_send_check(iph);
626                         }
627 
628                         err = output(net, sk, skb);
629 
630                         if (!err)
631                                 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
632                         if (err || !frag)
633                                 break;
634 
635                         skb = frag;
636                         frag = skb->next;
637                         skb->next = NULL;
638                 }
639 
640                 if (err == 0) {
641                         IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
642                         return 0;
643                 }
644 
645                 while (frag) {
646                         skb = frag->next;
647                         kfree_skb(frag);
648                         frag = skb;
649                 }
650                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
651                 return err;
652 
653 slow_path_clean:
654                 skb_walk_frags(skb, frag2) {
655                         if (frag2 == frag)
656                                 break;
657                         frag2->sk = NULL;
658                         frag2->destructor = NULL;
659                         skb->truesize += frag2->truesize;
660                 }
661         }
662 
663 slow_path:
664         iph = ip_hdr(skb);
665 
666         left = skb->len - hlen;         /* Space per frame */
667         ptr = hlen;             /* Where to start from */
668 
669         ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
670 
671         /*
672          *      Fragment the datagram.
673          */
674 
675         offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
676         not_last_frag = iph->frag_off & htons(IP_MF);
677 
678         /*
679          *      Keep copying data until we run out.
680          */
681 
682         while (left > 0) {
683                 len = left;
684                 /* IF: it doesn't fit, use 'mtu' - the data space left */
685                 if (len > mtu)
686                         len = mtu;
687                 /* IF: we are not sending up to and including the packet end
688                    then align the next start on an eight byte boundary */
689                 if (len < left) {
690                         len &= ~7;
691                 }
692 
693                 /* Allocate buffer */
694                 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
695                 if (!skb2) {
696                         err = -ENOMEM;
697                         goto fail;
698                 }
699 
700                 /*
701                  *      Set up data on packet
702                  */
703 
704                 ip_copy_metadata(skb2, skb);
705                 skb_reserve(skb2, ll_rs);
706                 skb_put(skb2, len + hlen);
707                 skb_reset_network_header(skb2);
708                 skb2->transport_header = skb2->network_header + hlen;
709 
710                 /*
711                  *      Charge the memory for the fragment to any owner
712                  *      it might possess
713                  */
714 
715                 if (skb->sk)
716                         skb_set_owner_w(skb2, skb->sk);
717 
718                 /*
719                  *      Copy the packet header into the new buffer.
720                  */
721 
722                 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
723 
724                 /*
725                  *      Copy a block of the IP datagram.
726                  */
727                 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
728                         BUG();
729                 left -= len;
730 
731                 /*
732                  *      Fill in the new header fields.
733                  */
734                 iph = ip_hdr(skb2);
735                 iph->frag_off = htons((offset >> 3));
736 
737                 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
738                         iph->frag_off |= htons(IP_DF);
739 
740                 /* ANK: dirty, but effective trick. Upgrade options only if
741                  * the segment to be fragmented was THE FIRST (otherwise,
742                  * options are already fixed) and make it ONCE
743                  * on the initial skb, so that all the following fragments
744                  * will inherit fixed options.
745                  */
746                 if (offset == 0)
747                         ip_options_fragment(skb);
748 
749                 /*
750                  *      Added AC : If we are fragmenting a fragment that's not the
751                  *                 last fragment then keep MF on each bit
752                  */
753                 if (left > 0 || not_last_frag)
754                         iph->frag_off |= htons(IP_MF);
755                 ptr += len;
756                 offset += len;
757 
758                 /*
759                  *      Put this fragment into the sending queue.
760                  */
761                 iph->tot_len = htons(len + hlen);
762 
763                 ip_send_check(iph);
764 
765                 err = output(net, sk, skb2);
766                 if (err)
767                         goto fail;
768 
769                 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
770         }
771         consume_skb(skb);
772         IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
773         return err;
774 
775 fail:
776         kfree_skb(skb);
777         IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
778         return err;
779 }
780 EXPORT_SYMBOL(ip_do_fragment);
781 
782 int
783 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
784 {
785         struct msghdr *msg = from;
786 
787         if (skb->ip_summed == CHECKSUM_PARTIAL) {
788                 if (copy_from_iter(to, len, &msg->msg_iter) != len)
789                         return -EFAULT;
790         } else {
791                 __wsum csum = 0;
792                 if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len)
793                         return -EFAULT;
794                 skb->csum = csum_block_add(skb->csum, csum, odd);
795         }
796         return 0;
797 }
798 EXPORT_SYMBOL(ip_generic_getfrag);
799 
800 static inline __wsum
801 csum_page(struct page *page, int offset, int copy)
802 {
803         char *kaddr;
804         __wsum csum;
805         kaddr = kmap(page);
806         csum = csum_partial(kaddr + offset, copy, 0);
807         kunmap(page);
808         return csum;
809 }
810 
811 static inline int ip_ufo_append_data(struct sock *sk,
812                         struct sk_buff_head *queue,
813                         int getfrag(void *from, char *to, int offset, int len,
814                                int odd, struct sk_buff *skb),
815                         void *from, int length, int hh_len, int fragheaderlen,
816                         int transhdrlen, int maxfraglen, unsigned int flags)
817 {
818         struct sk_buff *skb;
819         int err;
820 
821         /* There is support for UDP fragmentation offload by network
822          * device, so create one single skb packet containing complete
823          * udp datagram
824          */
825         skb = skb_peek_tail(queue);
826         if (!skb) {
827                 skb = sock_alloc_send_skb(sk,
828                         hh_len + fragheaderlen + transhdrlen + 20,
829                         (flags & MSG_DONTWAIT), &err);
830 
831                 if (!skb)
832                         return err;
833 
834                 /* reserve space for Hardware header */
835                 skb_reserve(skb, hh_len);
836 
837                 /* create space for UDP/IP header */
838                 skb_put(skb, fragheaderlen + transhdrlen);
839 
840                 /* initialize network header pointer */
841                 skb_reset_network_header(skb);
842 
843                 /* initialize protocol header pointer */
844                 skb->transport_header = skb->network_header + fragheaderlen;
845 
846                 skb->csum = 0;
847 
848                 __skb_queue_tail(queue, skb);
849         } else if (skb_is_gso(skb)) {
850                 goto append;
851         }
852 
853         skb->ip_summed = CHECKSUM_PARTIAL;
854         /* specify the length of each IP datagram fragment */
855         skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
856         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
857 
858 append:
859         return skb_append_datato_frags(sk, skb, getfrag, from,
860                                        (length - transhdrlen));
861 }
862 
863 static int __ip_append_data(struct sock *sk,
864                             struct flowi4 *fl4,
865                             struct sk_buff_head *queue,
866                             struct inet_cork *cork,
867                             struct page_frag *pfrag,
868                             int getfrag(void *from, char *to, int offset,
869                                         int len, int odd, struct sk_buff *skb),
870                             void *from, int length, int transhdrlen,
871                             unsigned int flags)
872 {
873         struct inet_sock *inet = inet_sk(sk);
874         struct sk_buff *skb;
875 
876         struct ip_options *opt = cork->opt;
877         int hh_len;
878         int exthdrlen;
879         int mtu;
880         int copy;
881         int err;
882         int offset = 0;
883         unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
884         int csummode = CHECKSUM_NONE;
885         struct rtable *rt = (struct rtable *)cork->dst;
886         u32 tskey = 0;
887 
888         skb = skb_peek_tail(queue);
889 
890         exthdrlen = !skb ? rt->dst.header_len : 0;
891         mtu = cork->fragsize;
892         if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
893             sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
894                 tskey = sk->sk_tskey++;
895 
896         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
897 
898         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
899         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
900         maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
901 
902         if (cork->length + length > maxnonfragsize - fragheaderlen) {
903                 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
904                                mtu - (opt ? opt->optlen : 0));
905                 return -EMSGSIZE;
906         }
907 
908         /*
909          * transhdrlen > 0 means that this is the first fragment and we wish
910          * it won't be fragmented in the future.
911          */
912         if (transhdrlen &&
913             length + fragheaderlen <= mtu &&
914             rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
915             !(flags & MSG_MORE) &&
916             !exthdrlen)
917                 csummode = CHECKSUM_PARTIAL;
918 
919         cork->length += length;
920         if (((length > mtu) || (skb && skb_is_gso(skb))) &&
921             (sk->sk_protocol == IPPROTO_UDP) &&
922             (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len &&
923             (sk->sk_type == SOCK_DGRAM) && !sk->sk_no_check_tx) {
924                 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
925                                          hh_len, fragheaderlen, transhdrlen,
926                                          maxfraglen, flags);
927                 if (err)
928                         goto error;
929                 return 0;
930         }
931 
932         /* So, what's going on in the loop below?
933          *
934          * We use calculated fragment length to generate chained skb,
935          * each of segments is IP fragment ready for sending to network after
936          * adding appropriate IP header.
937          */
938 
939         if (!skb)
940                 goto alloc_new_skb;
941 
942         while (length > 0) {
943                 /* Check if the remaining data fits into current packet. */
944                 copy = mtu - skb->len;
945                 if (copy < length)
946                         copy = maxfraglen - skb->len;
947                 if (copy <= 0) {
948                         char *data;
949                         unsigned int datalen;
950                         unsigned int fraglen;
951                         unsigned int fraggap;
952                         unsigned int alloclen;
953                         struct sk_buff *skb_prev;
954 alloc_new_skb:
955                         skb_prev = skb;
956                         if (skb_prev)
957                                 fraggap = skb_prev->len - maxfraglen;
958                         else
959                                 fraggap = 0;
960 
961                         /*
962                          * If remaining data exceeds the mtu,
963                          * we know we need more fragment(s).
964                          */
965                         datalen = length + fraggap;
966                         if (datalen > mtu - fragheaderlen)
967                                 datalen = maxfraglen - fragheaderlen;
968                         fraglen = datalen + fragheaderlen;
969 
970                         if ((flags & MSG_MORE) &&
971                             !(rt->dst.dev->features&NETIF_F_SG))
972                                 alloclen = mtu;
973                         else
974                                 alloclen = fraglen;
975 
976                         alloclen += exthdrlen;
977 
978                         /* The last fragment gets additional space at tail.
979                          * Note, with MSG_MORE we overallocate on fragments,
980                          * because we have no idea what fragment will be
981                          * the last.
982                          */
983                         if (datalen == length + fraggap)
984                                 alloclen += rt->dst.trailer_len;
985 
986                         if (transhdrlen) {
987                                 skb = sock_alloc_send_skb(sk,
988                                                 alloclen + hh_len + 15,
989                                                 (flags & MSG_DONTWAIT), &err);
990                         } else {
991                                 skb = NULL;
992                                 if (atomic_read(&sk->sk_wmem_alloc) <=
993                                     2 * sk->sk_sndbuf)
994                                         skb = sock_wmalloc(sk,
995                                                            alloclen + hh_len + 15, 1,
996                                                            sk->sk_allocation);
997                                 if (unlikely(!skb))
998                                         err = -ENOBUFS;
999                         }
1000                         if (!skb)
1001                                 goto error;
1002 
1003                         /*
1004                          *      Fill in the control structures
1005                          */
1006                         skb->ip_summed = csummode;
1007                         skb->csum = 0;
1008                         skb_reserve(skb, hh_len);
1009 
1010                         /* only the initial fragment is time stamped */
1011                         skb_shinfo(skb)->tx_flags = cork->tx_flags;
1012                         cork->tx_flags = 0;
1013                         skb_shinfo(skb)->tskey = tskey;
1014                         tskey = 0;
1015 
1016                         /*
1017                          *      Find where to start putting bytes.
1018                          */
1019                         data = skb_put(skb, fraglen + exthdrlen);
1020                         skb_set_network_header(skb, exthdrlen);
1021                         skb->transport_header = (skb->network_header +
1022                                                  fragheaderlen);
1023                         data += fragheaderlen + exthdrlen;
1024 
1025                         if (fraggap) {
1026                                 skb->csum = skb_copy_and_csum_bits(
1027                                         skb_prev, maxfraglen,
1028                                         data + transhdrlen, fraggap, 0);
1029                                 skb_prev->csum = csum_sub(skb_prev->csum,
1030                                                           skb->csum);
1031                                 data += fraggap;
1032                                 pskb_trim_unique(skb_prev, maxfraglen);
1033                         }
1034 
1035                         copy = datalen - transhdrlen - fraggap;
1036                         if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1037                                 err = -EFAULT;
1038                                 kfree_skb(skb);
1039                                 goto error;
1040                         }
1041 
1042                         offset += copy;
1043                         length -= datalen - fraggap;
1044                         transhdrlen = 0;
1045                         exthdrlen = 0;
1046                         csummode = CHECKSUM_NONE;
1047 
1048                         /*
1049                          * Put the packet on the pending queue.
1050                          */
1051                         __skb_queue_tail(queue, skb);
1052                         continue;
1053                 }
1054 
1055                 if (copy > length)
1056                         copy = length;
1057 
1058                 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1059                         unsigned int off;
1060 
1061                         off = skb->len;
1062                         if (getfrag(from, skb_put(skb, copy),
1063                                         offset, copy, off, skb) < 0) {
1064                                 __skb_trim(skb, off);
1065                                 err = -EFAULT;
1066                                 goto error;
1067                         }
1068                 } else {
1069                         int i = skb_shinfo(skb)->nr_frags;
1070 
1071                         err = -ENOMEM;
1072                         if (!sk_page_frag_refill(sk, pfrag))
1073                                 goto error;
1074 
1075                         if (!skb_can_coalesce(skb, i, pfrag->page,
1076                                               pfrag->offset)) {
1077                                 err = -EMSGSIZE;
1078                                 if (i == MAX_SKB_FRAGS)
1079                                         goto error;
1080 
1081                                 __skb_fill_page_desc(skb, i, pfrag->page,
1082                                                      pfrag->offset, 0);
1083                                 skb_shinfo(skb)->nr_frags = ++i;
1084                                 get_page(pfrag->page);
1085                         }
1086                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
1087                         if (getfrag(from,
1088                                     page_address(pfrag->page) + pfrag->offset,
1089                                     offset, copy, skb->len, skb) < 0)
1090                                 goto error_efault;
1091 
1092                         pfrag->offset += copy;
1093                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1094                         skb->len += copy;
1095                         skb->data_len += copy;
1096                         skb->truesize += copy;
1097                         atomic_add(copy, &sk->sk_wmem_alloc);
1098                 }
1099                 offset += copy;
1100                 length -= copy;
1101         }
1102 
1103         return 0;
1104 
1105 error_efault:
1106         err = -EFAULT;
1107 error:
1108         cork->length -= length;
1109         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1110         return err;
1111 }
1112 
1113 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1114                          struct ipcm_cookie *ipc, struct rtable **rtp)
1115 {
1116         struct ip_options_rcu *opt;
1117         struct rtable *rt;
1118 
1119         /*
1120          * setup for corking.
1121          */
1122         opt = ipc->opt;
1123         if (opt) {
1124                 if (!cork->opt) {
1125                         cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1126                                             sk->sk_allocation);
1127                         if (unlikely(!cork->opt))
1128                                 return -ENOBUFS;
1129                 }
1130                 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1131                 cork->flags |= IPCORK_OPT;
1132                 cork->addr = ipc->addr;
1133         }
1134         rt = *rtp;
1135         if (unlikely(!rt))
1136                 return -EFAULT;
1137         /*
1138          * We steal reference to this route, caller should not release it
1139          */
1140         *rtp = NULL;
1141         cork->fragsize = ip_sk_use_pmtu(sk) ?
1142                          dst_mtu(&rt->dst) : rt->dst.dev->mtu;
1143         cork->dst = &rt->dst;
1144         cork->length = 0;
1145         cork->ttl = ipc->ttl;
1146         cork->tos = ipc->tos;
1147         cork->priority = ipc->priority;
1148         cork->tx_flags = ipc->tx_flags;
1149 
1150         return 0;
1151 }
1152 
1153 /*
1154  *      ip_append_data() and ip_append_page() can make one large IP datagram
1155  *      from many pieces of data. Each pieces will be holded on the socket
1156  *      until ip_push_pending_frames() is called. Each piece can be a page
1157  *      or non-page data.
1158  *
1159  *      Not only UDP, other transport protocols - e.g. raw sockets - can use
1160  *      this interface potentially.
1161  *
1162  *      LATER: length must be adjusted by pad at tail, when it is required.
1163  */
1164 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1165                    int getfrag(void *from, char *to, int offset, int len,
1166                                int odd, struct sk_buff *skb),
1167                    void *from, int length, int transhdrlen,
1168                    struct ipcm_cookie *ipc, struct rtable **rtp,
1169                    unsigned int flags)
1170 {
1171         struct inet_sock *inet = inet_sk(sk);
1172         int err;
1173 
1174         if (flags&MSG_PROBE)
1175                 return 0;
1176 
1177         if (skb_queue_empty(&sk->sk_write_queue)) {
1178                 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1179                 if (err)
1180                         return err;
1181         } else {
1182                 transhdrlen = 0;
1183         }
1184 
1185         return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1186                                 sk_page_frag(sk), getfrag,
1187                                 from, length, transhdrlen, flags);
1188 }
1189 
1190 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1191                        int offset, size_t size, int flags)
1192 {
1193         struct inet_sock *inet = inet_sk(sk);
1194         struct sk_buff *skb;
1195         struct rtable *rt;
1196         struct ip_options *opt = NULL;
1197         struct inet_cork *cork;
1198         int hh_len;
1199         int mtu;
1200         int len;
1201         int err;
1202         unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1203 
1204         if (inet->hdrincl)
1205                 return -EPERM;
1206 
1207         if (flags&MSG_PROBE)
1208                 return 0;
1209 
1210         if (skb_queue_empty(&sk->sk_write_queue))
1211                 return -EINVAL;
1212 
1213         cork = &inet->cork.base;
1214         rt = (struct rtable *)cork->dst;
1215         if (cork->flags & IPCORK_OPT)
1216                 opt = cork->opt;
1217 
1218         if (!(rt->dst.dev->features&NETIF_F_SG))
1219                 return -EOPNOTSUPP;
1220 
1221         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1222         mtu = cork->fragsize;
1223 
1224         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1225         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1226         maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1227 
1228         if (cork->length + size > maxnonfragsize - fragheaderlen) {
1229                 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1230                                mtu - (opt ? opt->optlen : 0));
1231                 return -EMSGSIZE;
1232         }
1233 
1234         skb = skb_peek_tail(&sk->sk_write_queue);
1235         if (!skb)
1236                 return -EINVAL;
1237 
1238         if ((size + skb->len > mtu) &&
1239             (sk->sk_protocol == IPPROTO_UDP) &&
1240             (rt->dst.dev->features & NETIF_F_UFO)) {
1241                 if (skb->ip_summed != CHECKSUM_PARTIAL)
1242                         return -EOPNOTSUPP;
1243 
1244                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1245                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1246         }
1247         cork->length += size;
1248 
1249         while (size > 0) {
1250                 if (skb_is_gso(skb)) {
1251                         len = size;
1252                 } else {
1253 
1254                         /* Check if the remaining data fits into current packet. */
1255                         len = mtu - skb->len;
1256                         if (len < size)
1257                                 len = maxfraglen - skb->len;
1258                 }
1259                 if (len <= 0) {
1260                         struct sk_buff *skb_prev;
1261                         int alloclen;
1262 
1263                         skb_prev = skb;
1264                         fraggap = skb_prev->len - maxfraglen;
1265 
1266                         alloclen = fragheaderlen + hh_len + fraggap + 15;
1267                         skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1268                         if (unlikely(!skb)) {
1269                                 err = -ENOBUFS;
1270                                 goto error;
1271                         }
1272 
1273                         /*
1274                          *      Fill in the control structures
1275                          */
1276                         skb->ip_summed = CHECKSUM_NONE;
1277                         skb->csum = 0;
1278                         skb_reserve(skb, hh_len);
1279 
1280                         /*
1281                          *      Find where to start putting bytes.
1282                          */
1283                         skb_put(skb, fragheaderlen + fraggap);
1284                         skb_reset_network_header(skb);
1285                         skb->transport_header = (skb->network_header +
1286                                                  fragheaderlen);
1287                         if (fraggap) {
1288                                 skb->csum = skb_copy_and_csum_bits(skb_prev,
1289                                                                    maxfraglen,
1290                                                     skb_transport_header(skb),
1291                                                                    fraggap, 0);
1292                                 skb_prev->csum = csum_sub(skb_prev->csum,
1293                                                           skb->csum);
1294                                 pskb_trim_unique(skb_prev, maxfraglen);
1295                         }
1296 
1297                         /*
1298                          * Put the packet on the pending queue.
1299                          */
1300                         __skb_queue_tail(&sk->sk_write_queue, skb);
1301                         continue;
1302                 }
1303 
1304                 if (len > size)
1305                         len = size;
1306 
1307                 if (skb_append_pagefrags(skb, page, offset, len)) {
1308                         err = -EMSGSIZE;
1309                         goto error;
1310                 }
1311 
1312                 if (skb->ip_summed == CHECKSUM_NONE) {
1313                         __wsum csum;
1314                         csum = csum_page(page, offset, len);
1315                         skb->csum = csum_block_add(skb->csum, csum, skb->len);
1316                 }
1317 
1318                 skb->len += len;
1319                 skb->data_len += len;
1320                 skb->truesize += len;
1321                 atomic_add(len, &sk->sk_wmem_alloc);
1322                 offset += len;
1323                 size -= len;
1324         }
1325         return 0;
1326 
1327 error:
1328         cork->length -= size;
1329         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1330         return err;
1331 }
1332 
1333 static void ip_cork_release(struct inet_cork *cork)
1334 {
1335         cork->flags &= ~IPCORK_OPT;
1336         kfree(cork->opt);
1337         cork->opt = NULL;
1338         dst_release(cork->dst);
1339         cork->dst = NULL;
1340 }
1341 
1342 /*
1343  *      Combined all pending IP fragments on the socket as one IP datagram
1344  *      and push them out.
1345  */
1346 struct sk_buff *__ip_make_skb(struct sock *sk,
1347                               struct flowi4 *fl4,
1348                               struct sk_buff_head *queue,
1349                               struct inet_cork *cork)
1350 {
1351         struct sk_buff *skb, *tmp_skb;
1352         struct sk_buff **tail_skb;
1353         struct inet_sock *inet = inet_sk(sk);
1354         struct net *net = sock_net(sk);
1355         struct ip_options *opt = NULL;
1356         struct rtable *rt = (struct rtable *)cork->dst;
1357         struct iphdr *iph;
1358         __be16 df = 0;
1359         __u8 ttl;
1360 
1361         skb = __skb_dequeue(queue);
1362         if (!skb)
1363                 goto out;
1364         tail_skb = &(skb_shinfo(skb)->frag_list);
1365 
1366         /* move skb->data to ip header from ext header */
1367         if (skb->data < skb_network_header(skb))
1368                 __skb_pull(skb, skb_network_offset(skb));
1369         while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1370                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1371                 *tail_skb = tmp_skb;
1372                 tail_skb = &(tmp_skb->next);
1373                 skb->len += tmp_skb->len;
1374                 skb->data_len += tmp_skb->len;
1375                 skb->truesize += tmp_skb->truesize;
1376                 tmp_skb->destructor = NULL;
1377                 tmp_skb->sk = NULL;
1378         }
1379 
1380         /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1381          * to fragment the frame generated here. No matter, what transforms
1382          * how transforms change size of the packet, it will come out.
1383          */
1384         skb->ignore_df = ip_sk_ignore_df(sk);
1385 
1386         /* DF bit is set when we want to see DF on outgoing frames.
1387          * If ignore_df is set too, we still allow to fragment this frame
1388          * locally. */
1389         if (inet->pmtudisc == IP_PMTUDISC_DO ||
1390             inet->pmtudisc == IP_PMTUDISC_PROBE ||
1391             (skb->len <= dst_mtu(&rt->dst) &&
1392              ip_dont_fragment(sk, &rt->dst)))
1393                 df = htons(IP_DF);
1394 
1395         if (cork->flags & IPCORK_OPT)
1396                 opt = cork->opt;
1397 
1398         if (cork->ttl != 0)
1399                 ttl = cork->ttl;
1400         else if (rt->rt_type == RTN_MULTICAST)
1401                 ttl = inet->mc_ttl;
1402         else
1403                 ttl = ip_select_ttl(inet, &rt->dst);
1404 
1405         iph = ip_hdr(skb);
1406         iph->version = 4;
1407         iph->ihl = 5;
1408         iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1409         iph->frag_off = df;
1410         iph->ttl = ttl;
1411         iph->protocol = sk->sk_protocol;
1412         ip_copy_addrs(iph, fl4);
1413         ip_select_ident(net, skb, sk);
1414 
1415         if (opt) {
1416                 iph->ihl += opt->optlen>>2;
1417                 ip_options_build(skb, opt, cork->addr, rt, 0);
1418         }
1419 
1420         skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1421         skb->mark = sk->sk_mark;
1422         /*
1423          * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1424          * on dst refcount
1425          */
1426         cork->dst = NULL;
1427         skb_dst_set(skb, &rt->dst);
1428 
1429         if (iph->protocol == IPPROTO_ICMP)
1430                 icmp_out_count(net, ((struct icmphdr *)
1431                         skb_transport_header(skb))->type);
1432 
1433         ip_cork_release(cork);
1434 out:
1435         return skb;
1436 }
1437 
1438 int ip_send_skb(struct net *net, struct sk_buff *skb)
1439 {
1440         int err;
1441 
1442         err = ip_local_out(net, skb->sk, skb);
1443         if (err) {
1444                 if (err > 0)
1445                         err = net_xmit_errno(err);
1446                 if (err)
1447                         IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1448         }
1449 
1450         return err;
1451 }
1452 
1453 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1454 {
1455         struct sk_buff *skb;
1456 
1457         skb = ip_finish_skb(sk, fl4);
1458         if (!skb)
1459                 return 0;
1460 
1461         /* Netfilter gets whole the not fragmented skb. */
1462         return ip_send_skb(sock_net(sk), skb);
1463 }
1464 
1465 /*
1466  *      Throw away all pending data on the socket.
1467  */
1468 static void __ip_flush_pending_frames(struct sock *sk,
1469                                       struct sk_buff_head *queue,
1470                                       struct inet_cork *cork)
1471 {
1472         struct sk_buff *skb;
1473 
1474         while ((skb = __skb_dequeue_tail(queue)) != NULL)
1475                 kfree_skb(skb);
1476 
1477         ip_cork_release(cork);
1478 }
1479 
1480 void ip_flush_pending_frames(struct sock *sk)
1481 {
1482         __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1483 }
1484 
1485 struct sk_buff *ip_make_skb(struct sock *sk,
1486                             struct flowi4 *fl4,
1487                             int getfrag(void *from, char *to, int offset,
1488                                         int len, int odd, struct sk_buff *skb),
1489                             void *from, int length, int transhdrlen,
1490                             struct ipcm_cookie *ipc, struct rtable **rtp,
1491                             unsigned int flags)
1492 {
1493         struct inet_cork cork;
1494         struct sk_buff_head queue;
1495         int err;
1496 
1497         if (flags & MSG_PROBE)
1498                 return NULL;
1499 
1500         __skb_queue_head_init(&queue);
1501 
1502         cork.flags = 0;
1503         cork.addr = 0;
1504         cork.opt = NULL;
1505         err = ip_setup_cork(sk, &cork, ipc, rtp);
1506         if (err)
1507                 return ERR_PTR(err);
1508 
1509         err = __ip_append_data(sk, fl4, &queue, &cork,
1510                                &current->task_frag, getfrag,
1511                                from, length, transhdrlen, flags);
1512         if (err) {
1513                 __ip_flush_pending_frames(sk, &queue, &cork);
1514                 return ERR_PTR(err);
1515         }
1516 
1517         return __ip_make_skb(sk, fl4, &queue, &cork);
1518 }
1519 
1520 /*
1521  *      Fetch data from kernel space and fill in checksum if needed.
1522  */
1523 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1524                               int len, int odd, struct sk_buff *skb)
1525 {
1526         __wsum csum;
1527 
1528         csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1529         skb->csum = csum_block_add(skb->csum, csum, odd);
1530         return 0;
1531 }
1532 
1533 /*
1534  *      Generic function to send a packet as reply to another packet.
1535  *      Used to send some TCP resets/acks so far.
1536  */
1537 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1538                            const struct ip_options *sopt,
1539                            __be32 daddr, __be32 saddr,
1540                            const struct ip_reply_arg *arg,
1541                            unsigned int len)
1542 {
1543         struct ip_options_data replyopts;
1544         struct ipcm_cookie ipc;
1545         struct flowi4 fl4;
1546         struct rtable *rt = skb_rtable(skb);
1547         struct net *net = sock_net(sk);
1548         struct sk_buff *nskb;
1549         int err;
1550         int oif;
1551 
1552         if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
1553                 return;
1554 
1555         ipc.addr = daddr;
1556         ipc.opt = NULL;
1557         ipc.tx_flags = 0;
1558         ipc.ttl = 0;
1559         ipc.tos = -1;
1560 
1561         if (replyopts.opt.opt.optlen) {
1562                 ipc.opt = &replyopts.opt;
1563 
1564                 if (replyopts.opt.opt.srr)
1565                         daddr = replyopts.opt.opt.faddr;
1566         }
1567 
1568         oif = arg->bound_dev_if;
1569         if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1570                 oif = skb->skb_iif;
1571 
1572         flowi4_init_output(&fl4, oif,
1573                            IP4_REPLY_MARK(net, skb->mark),
1574                            RT_TOS(arg->tos),
1575                            RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1576                            ip_reply_arg_flowi_flags(arg),
1577                            daddr, saddr,
1578                            tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1579         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1580         rt = ip_route_output_key(net, &fl4);
1581         if (IS_ERR(rt))
1582                 return;
1583 
1584         inet_sk(sk)->tos = arg->tos;
1585 
1586         sk->sk_priority = skb->priority;
1587         sk->sk_protocol = ip_hdr(skb)->protocol;
1588         sk->sk_bound_dev_if = arg->bound_dev_if;
1589         sk->sk_sndbuf = sysctl_wmem_default;
1590         err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1591                              len, 0, &ipc, &rt, MSG_DONTWAIT);
1592         if (unlikely(err)) {
1593                 ip_flush_pending_frames(sk);
1594                 goto out;
1595         }
1596 
1597         nskb = skb_peek(&sk->sk_write_queue);
1598         if (nskb) {
1599                 if (arg->csumoffset >= 0)
1600                         *((__sum16 *)skb_transport_header(nskb) +
1601                           arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1602                                                                 arg->csum));
1603                 nskb->ip_summed = CHECKSUM_NONE;
1604                 ip_push_pending_frames(sk, &fl4);
1605         }
1606 out:
1607         ip_rt_put(rt);
1608 }
1609 
1610 void __init ip_init(void)
1611 {
1612         ip_rt_init();
1613         inet_initpeers();
1614 
1615 #if defined(CONFIG_IP_MULTICAST)
1616         igmp_mc_init();
1617 #endif
1618 }
1619 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us