projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[deliverable/linux.git] / 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/mroute.h>
80 #include <linux/netlink.h>
81 #include <linux/tcp.h>
82
83 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
84 EXPORT_SYMBOL(sysctl_ip_default_ttl);
85
86 static int ip_fragment(struct sock *sk, struct sk_buff *skb,
87 unsigned int mtu,
88 int (*output)(struct sock *, struct sk_buff *));
89
90 /* Generate a checksum for an outgoing IP datagram. */
91 void ip_send_check(struct iphdr *iph)
92 {
93 iph->check = 0;
94 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
95 }
96 EXPORT_SYMBOL(ip_send_check);
97
98 static int __ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
99 {
100 struct iphdr *iph = ip_hdr(skb);
101
102 iph->tot_len = htons(skb->len);
103 ip_send_check(iph);
104 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, sk, skb, NULL,
105 skb_dst(skb)->dev, dst_output_sk);
106 }
107
108 int __ip_local_out(struct sk_buff *skb)
109 {
110 return __ip_local_out_sk(skb->sk, skb);
111 }
112
113 int ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
114 {
115 int err;
116
117 err = __ip_local_out(skb);
118 if (likely(err == 1))
119 err = dst_output_sk(sk, skb);
120
121 return err;
122 }
123 EXPORT_SYMBOL_GPL(ip_local_out_sk);
124
125 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
126 {
127 int ttl = inet->uc_ttl;
128
129 if (ttl < 0)
130 ttl = ip4_dst_hoplimit(dst);
131 return ttl;
132 }
133
134 /*
135 * Add an ip header to a skbuff and send it out.
136 *
137 */
138 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
139 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
140 {
141 struct inet_sock *inet = inet_sk(sk);
142 struct rtable *rt = skb_rtable(skb);
143 struct iphdr *iph;
144
145 /* Build the IP header. */
146 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
147 skb_reset_network_header(skb);
148 iph = ip_hdr(skb);
149 iph->version = 4;
150 iph->ihl = 5;
151 iph->tos = inet->tos;
152 if (ip_dont_fragment(sk, &rt->dst))
153 iph->frag_off = htons(IP_DF);
154 else
155 iph->frag_off = 0;
156 iph->ttl = ip_select_ttl(inet, &rt->dst);
157 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
158 iph->saddr = saddr;
159 iph->protocol = sk->sk_protocol;
160 ip_select_ident(sock_net(sk), skb, sk);
161
162 if (opt && opt->opt.optlen) {
163 iph->ihl += opt->opt.optlen>>2;
164 ip_options_build(skb, &opt->opt, daddr, rt, 0);
165 }
166
167 skb->priority = sk->sk_priority;
168 skb->mark = sk->sk_mark;
169
170 /* Send it out. */
171 return ip_local_out(skb);
172 }
173 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
174
175 static int ip_finish_output2(struct sock *sk, struct sk_buff *skb)
176 {
177 struct dst_entry *dst = skb_dst(skb);
178 struct rtable *rt = (struct rtable *)dst;
179 struct net_device *dev = dst->dev;
180 unsigned int hh_len = LL_RESERVED_SPACE(dev);
181 struct neighbour *neigh;
182 u32 nexthop;
183
184 if (rt->rt_type == RTN_MULTICAST) {
185 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
186 } else if (rt->rt_type == RTN_BROADCAST)
187 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
188
189 /* Be paranoid, rather than too clever. */
190 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
191 struct sk_buff *skb2;
192
193 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
194 if (!skb2) {
195 kfree_skb(skb);
196 return -ENOMEM;
197 }
198 if (skb->sk)
199 skb_set_owner_w(skb2, skb->sk);
200 consume_skb(skb);
201 skb = skb2;
202 }
203
204 rcu_read_lock_bh();
205 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
206 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
207 if (unlikely(!neigh))
208 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
209 if (!IS_ERR(neigh)) {
210 int res = dst_neigh_output(dst, neigh, skb);
211
212 rcu_read_unlock_bh();
213 return res;
214 }
215 rcu_read_unlock_bh();
216
217 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
218 __func__);
219 kfree_skb(skb);
220 return -EINVAL;
221 }
222
223 static int ip_finish_output_gso(struct sock *sk, struct sk_buff *skb,
224 unsigned int mtu)
225 {
226 netdev_features_t features;
227 struct sk_buff *segs;
228 int ret = 0;
229
230 /* common case: locally created skb or seglen is <= mtu */
231 if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) ||
232 skb_gso_network_seglen(skb) <= mtu)
233 return ip_finish_output2(sk, skb);
234
235 /* Slowpath - GSO segment length is exceeding the dst MTU.
236 *
237 * This can happen in two cases:
238 * 1) TCP GRO packet, DF bit not set
239 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
240 * from host network stack.
241 */
242 features = netif_skb_features(skb);
243 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
244 if (IS_ERR_OR_NULL(segs)) {
245 kfree_skb(skb);
246 return -ENOMEM;
247 }
248
249 consume_skb(skb);
250
251 do {
252 struct sk_buff *nskb = segs->next;
253 int err;
254
255 segs->next = NULL;
256 err = ip_fragment(sk, segs, mtu, ip_finish_output2);
257
258 if (err && ret == 0)
259 ret = err;
260 segs = nskb;
261 } while (segs);
262
263 return ret;
264 }
265
266 static int ip_finish_output(struct sock *sk, struct sk_buff *skb)
267 {
268 unsigned int mtu;
269
270 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
271 /* Policy lookup after SNAT yielded a new policy */
272 if (skb_dst(skb)->xfrm) {
273 IPCB(skb)->flags |= IPSKB_REROUTED;
274 return dst_output_sk(sk, skb);
275 }
276 #endif
277 mtu = ip_skb_dst_mtu(skb);
278 if (skb_is_gso(skb))
279 return ip_finish_output_gso(sk, skb, mtu);
280
281 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
282 return ip_fragment(sk, skb, mtu, ip_finish_output2);
283
284 return ip_finish_output2(sk, skb);
285 }
286
287 int ip_mc_output(struct sock *sk, struct sk_buff *skb)
288 {
289 struct rtable *rt = skb_rtable(skb);
290 struct net_device *dev = rt->dst.dev;
291
292 /*
293 * If the indicated interface is up and running, send the packet.
294 */
295 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
296
297 skb->dev = dev;
298 skb->protocol = htons(ETH_P_IP);
299
300 /*
301 * Multicasts are looped back for other local users
302 */
303
304 if (rt->rt_flags&RTCF_MULTICAST) {
305 if (sk_mc_loop(sk)
306 #ifdef CONFIG_IP_MROUTE
307 /* Small optimization: do not loopback not local frames,
308 which returned after forwarding; they will be dropped
309 by ip_mr_input in any case.
310 Note, that local frames are looped back to be delivered
311 to local recipients.
312
313 This check is duplicated in ip_mr_input at the moment.
314 */
315 &&
316 ((rt->rt_flags & RTCF_LOCAL) ||
317 !(IPCB(skb)->flags & IPSKB_FORWARDED))
318 #endif
319 ) {
320 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
321 if (newskb)
322 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
323 sk, newskb, NULL, newskb->dev,
324 dev_loopback_xmit);
325 }
326
327 /* Multicasts with ttl 0 must not go beyond the host */
328
329 if (ip_hdr(skb)->ttl == 0) {
330 kfree_skb(skb);
331 return 0;
332 }
333 }
334
335 if (rt->rt_flags&RTCF_BROADCAST) {
336 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
337 if (newskb)
338 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, newskb,
339 NULL, newskb->dev, dev_loopback_xmit);
340 }
341
342 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, skb, NULL,
343 skb->dev, ip_finish_output,
344 !(IPCB(skb)->flags & IPSKB_REROUTED));
345 }
346
347 int ip_output(struct sock *sk, struct sk_buff *skb)
348 {
349 struct net_device *dev = skb_dst(skb)->dev;
350
351 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
352
353 skb->dev = dev;
354 skb->protocol = htons(ETH_P_IP);
355
356 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, skb,
357 NULL, dev,
358 ip_finish_output,
359 !(IPCB(skb)->flags & IPSKB_REROUTED));
360 }
361
362 /*
363 * copy saddr and daddr, possibly using 64bit load/stores
364 * Equivalent to :
365 * iph->saddr = fl4->saddr;
366 * iph->daddr = fl4->daddr;
367 */
368 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
369 {
370 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
371 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
372 memcpy(&iph->saddr, &fl4->saddr,
373 sizeof(fl4->saddr) + sizeof(fl4->daddr));
374 }
375
376 /* Note: skb->sk can be different from sk, in case of tunnels */
377 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
378 {
379 struct inet_sock *inet = inet_sk(sk);
380 struct ip_options_rcu *inet_opt;
381 struct flowi4 *fl4;
382 struct rtable *rt;
383 struct iphdr *iph;
384 int res;
385
386 /* Skip all of this if the packet is already routed,
387 * f.e. by something like SCTP.
388 */
389 rcu_read_lock();
390 inet_opt = rcu_dereference(inet->inet_opt);
391 fl4 = &fl->u.ip4;
392 rt = skb_rtable(skb);
393 if (rt)
394 goto packet_routed;
395
396 /* Make sure we can route this packet. */
397 rt = (struct rtable *)__sk_dst_check(sk, 0);
398 if (!rt) {
399 __be32 daddr;
400
401 /* Use correct destination address if we have options. */
402 daddr = inet->inet_daddr;
403 if (inet_opt && inet_opt->opt.srr)
404 daddr = inet_opt->opt.faddr;
405
406 /* If this fails, retransmit mechanism of transport layer will
407 * keep trying until route appears or the connection times
408 * itself out.
409 */
410 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
411 daddr, inet->inet_saddr,
412 inet->inet_dport,
413 inet->inet_sport,
414 sk->sk_protocol,
415 RT_CONN_FLAGS(sk),
416 sk->sk_bound_dev_if);
417 if (IS_ERR(rt))
418 goto no_route;
419 sk_setup_caps(sk, &rt->dst);
420 }
421 skb_dst_set_noref(skb, &rt->dst);
422
423 packet_routed:
424 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
425 goto no_route;
426
427 /* OK, we know where to send it, allocate and build IP header. */
428 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
429 skb_reset_network_header(skb);
430 iph = ip_hdr(skb);
431 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
432 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
433 iph->frag_off = htons(IP_DF);
434 else
435 iph->frag_off = 0;
436 iph->ttl = ip_select_ttl(inet, &rt->dst);
437 iph->protocol = sk->sk_protocol;
438 ip_copy_addrs(iph, fl4);
439
440 /* Transport layer set skb->h.foo itself. */
441
442 if (inet_opt && inet_opt->opt.optlen) {
443 iph->ihl += inet_opt->opt.optlen >> 2;
444 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
445 }
446
447 ip_select_ident_segs(sock_net(sk), skb, sk,
448 skb_shinfo(skb)->gso_segs ?: 1);
449
450 /* TODO : should we use skb->sk here instead of sk ? */
451 skb->priority = sk->sk_priority;
452 skb->mark = sk->sk_mark;
453
454 res = ip_local_out(skb);
455 rcu_read_unlock();
456 return res;
457
458 no_route:
459 rcu_read_unlock();
460 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
461 kfree_skb(skb);
462 return -EHOSTUNREACH;
463 }
464 EXPORT_SYMBOL(ip_queue_xmit);
465
466 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
467 {
468 to->pkt_type = from->pkt_type;
469 to->priority = from->priority;
470 to->protocol = from->protocol;
471 skb_dst_drop(to);
472 skb_dst_copy(to, from);
473 to->dev = from->dev;
474 to->mark = from->mark;
475
476 /* Copy the flags to each fragment. */
477 IPCB(to)->flags = IPCB(from)->flags;
478
479 #ifdef CONFIG_NET_SCHED
480 to->tc_index = from->tc_index;
481 #endif
482 nf_copy(to, from);
483 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
484 to->ipvs_property = from->ipvs_property;
485 #endif
486 skb_copy_secmark(to, from);
487 }
488
489 static int ip_fragment(struct sock *sk, struct sk_buff *skb,
490 unsigned int mtu,
491 int (*output)(struct sock *, struct sk_buff *))
492 {
493 struct iphdr *iph = ip_hdr(skb);
494
495 if ((iph->frag_off & htons(IP_DF)) == 0)
496 return ip_do_fragment(sk, skb, output);
497
498 if (unlikely(!skb->ignore_df ||
499 (IPCB(skb)->frag_max_size &&
500 IPCB(skb)->frag_max_size > mtu))) {
501 struct rtable *rt = skb_rtable(skb);
502 struct net_device *dev = rt->dst.dev;
503
504 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
505 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
506 htonl(mtu));
507 kfree_skb(skb);
508 return -EMSGSIZE;
509 }
510
511 return ip_do_fragment(sk, skb, output);
512 }
513
514 /*
515 * This IP datagram is too large to be sent in one piece. Break it up into
516 * smaller pieces (each of size equal to IP header plus
517 * a block of the data of the original IP data part) that will yet fit in a
518 * single device frame, and queue such a frame for sending.
519 */
520
521 int ip_do_fragment(struct sock *sk, struct sk_buff *skb,
522 int (*output)(struct sock *, struct sk_buff *))
523 {
524 struct iphdr *iph;
525 int ptr;
526 struct net_device *dev;
527 struct sk_buff *skb2;
528 unsigned int mtu, hlen, left, len, ll_rs;
529 int offset;
530 __be16 not_last_frag;
531 struct rtable *rt = skb_rtable(skb);
532 int err = 0;
533
534 dev = rt->dst.dev;
535
536 /*
537 * Point into the IP datagram header.
538 */
539
540 iph = ip_hdr(skb);
541
542 mtu = ip_skb_dst_mtu(skb);
543 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
544 mtu = IPCB(skb)->frag_max_size;
545
546 /*
547 * Setup starting values.
548 */
549
550 hlen = iph->ihl * 4;
551 mtu = mtu - hlen; /* Size of data space */
552 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
553
554 /* When frag_list is given, use it. First, check its validity:
555 * some transformers could create wrong frag_list or break existing
556 * one, it is not prohibited. In this case fall back to copying.
557 *
558 * LATER: this step can be merged to real generation of fragments,
559 * we can switch to copy when see the first bad fragment.
560 */
561 if (skb_has_frag_list(skb)) {
562 struct sk_buff *frag, *frag2;
563 int first_len = skb_pagelen(skb);
564
565 if (first_len - hlen > mtu ||
566 ((first_len - hlen) & 7) ||
567 ip_is_fragment(iph) ||
568 skb_cloned(skb))
569 goto slow_path;
570
571 skb_walk_frags(skb, frag) {
572 /* Correct geometry. */
573 if (frag->len > mtu ||
574 ((frag->len & 7) && frag->next) ||
575 skb_headroom(frag) < hlen)
576 goto slow_path_clean;
577
578 /* Partially cloned skb? */
579 if (skb_shared(frag))
580 goto slow_path_clean;
581
582 BUG_ON(frag->sk);
583 if (skb->sk) {
584 frag->sk = skb->sk;
585 frag->destructor = sock_wfree;
586 }
587 skb->truesize -= frag->truesize;
588 }
589
590 /* Everything is OK. Generate! */
591
592 err = 0;
593 offset = 0;
594 frag = skb_shinfo(skb)->frag_list;
595 skb_frag_list_init(skb);
596 skb->data_len = first_len - skb_headlen(skb);
597 skb->len = first_len;
598 iph->tot_len = htons(first_len);
599 iph->frag_off = htons(IP_MF);
600 ip_send_check(iph);
601
602 for (;;) {
603 /* Prepare header of the next frame,
604 * before previous one went down. */
605 if (frag) {
606 frag->ip_summed = CHECKSUM_NONE;
607 skb_reset_transport_header(frag);
608 __skb_push(frag, hlen);
609 skb_reset_network_header(frag);
610 memcpy(skb_network_header(frag), iph, hlen);
611 iph = ip_hdr(frag);
612 iph->tot_len = htons(frag->len);
613 ip_copy_metadata(frag, skb);
614 if (offset == 0)
615 ip_options_fragment(frag);
616 offset += skb->len - hlen;
617 iph->frag_off = htons(offset>>3);
618 if (frag->next)
619 iph->frag_off |= htons(IP_MF);
620 /* Ready, complete checksum */
621 ip_send_check(iph);
622 }
623
624 err = output(sk, skb);
625
626 if (!err)
627 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
628 if (err || !frag)
629 break;
630
631 skb = frag;
632 frag = skb->next;
633 skb->next = NULL;
634 }
635
636 if (err == 0) {
637 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
638 return 0;
639 }
640
641 while (frag) {
642 skb = frag->next;
643 kfree_skb(frag);
644 frag = skb;
645 }
646 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
647 return err;
648
649 slow_path_clean:
650 skb_walk_frags(skb, frag2) {
651 if (frag2 == frag)
652 break;
653 frag2->sk = NULL;
654 frag2->destructor = NULL;
655 skb->truesize += frag2->truesize;
656 }
657 }
658
659 slow_path:
660 /* for offloaded checksums cleanup checksum before fragmentation */
661 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
662 goto fail;
663 iph = ip_hdr(skb);
664
665 left = skb->len - hlen; /* Space per frame */
666 ptr = hlen; /* Where to start from */
667
668 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
669
670 /*
671 * Fragment the datagram.
672 */
673
674 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
675 not_last_frag = iph->frag_off & htons(IP_MF);
676
677 /*
678 * Keep copying data until we run out.
679 */
680
681 while (left > 0) {
682 len = left;
683 /* IF: it doesn't fit, use 'mtu' - the data space left */
684 if (len > mtu)
685 len = mtu;
686 /* IF: we are not sending up to and including the packet end
687 then align the next start on an eight byte boundary */
688 if (len < left) {
689 len &= ~7;
690 }
691
692 /* Allocate buffer */
693 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
694 if (!skb2) {
695 err = -ENOMEM;
696 goto fail;
697 }
698
699 /*
700 * Set up data on packet
701 */
702
703 ip_copy_metadata(skb2, skb);
704 skb_reserve(skb2, ll_rs);
705 skb_put(skb2, len + hlen);
706 skb_reset_network_header(skb2);
707 skb2->transport_header = skb2->network_header + hlen;
708
709 /*
710 * Charge the memory for the fragment to any owner
711 * it might possess
712 */
713
714 if (skb->sk)
715 skb_set_owner_w(skb2, skb->sk);
716
717 /*
718 * Copy the packet header into the new buffer.
719 */
720
721 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
722
723 /*
724 * Copy a block of the IP datagram.
725 */
726 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
727 BUG();
728 left -= len;
729
730 /*
731 * Fill in the new header fields.
732 */
733 iph = ip_hdr(skb2);
734 iph->frag_off = htons((offset >> 3));
735
736 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
737 iph->frag_off |= htons(IP_DF);
738
739 /* ANK: dirty, but effective trick. Upgrade options only if
740 * the segment to be fragmented was THE FIRST (otherwise,
741 * options are already fixed) and make it ONCE
742 * on the initial skb, so that all the following fragments
743 * will inherit fixed options.
744 */
745 if (offset == 0)
746 ip_options_fragment(skb);
747
748 /*
749 * Added AC : If we are fragmenting a fragment that's not the
750 * last fragment then keep MF on each bit
751 */
752 if (left > 0 || not_last_frag)
753 iph->frag_off |= htons(IP_MF);
754 ptr += len;
755 offset += len;
756
757 /*
758 * Put this fragment into the sending queue.
759 */
760 iph->tot_len = htons(len + hlen);
761
762 ip_send_check(iph);
763
764 err = output(sk, skb2);
765 if (err)
766 goto fail;
767
768 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
769 }
770 consume_skb(skb);
771 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
772 return err;
773
774 fail:
775 kfree_skb(skb);
776 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
777 return err;
778 }
779 EXPORT_SYMBOL(ip_do_fragment);
780
781 int
782 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
783 {
784 struct msghdr *msg = from;
785
786 if (skb->ip_summed == CHECKSUM_PARTIAL) {
787 if (copy_from_iter(to, len, &msg->msg_iter) != len)
788 return -EFAULT;
789 } else {
790 __wsum csum = 0;
791 if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len)
792 return -EFAULT;
793 skb->csum = csum_block_add(skb->csum, csum, odd);
794 }
795 return 0;
796 }
797 EXPORT_SYMBOL(ip_generic_getfrag);
798
799 static inline __wsum
800 csum_page(struct page *page, int offset, int copy)
801 {
802 char *kaddr;
803 __wsum csum;
804 kaddr = kmap(page);
805 csum = csum_partial(kaddr + offset, copy, 0);
806 kunmap(page);
807 return csum;
808 }
809
810 static inline int ip_ufo_append_data(struct sock *sk,
811 struct sk_buff_head *queue,
812 int getfrag(void *from, char *to, int offset, int len,
813 int odd, struct sk_buff *skb),
814 void *from, int length, int hh_len, int fragheaderlen,
815 int transhdrlen, int maxfraglen, unsigned int flags)
816 {
817 struct sk_buff *skb;
818 int err;
819
820 /* There is support for UDP fragmentation offload by network
821 * device, so create one single skb packet containing complete
822 * udp datagram
823 */
824 skb = skb_peek_tail(queue);
825 if (!skb) {
826 skb = sock_alloc_send_skb(sk,
827 hh_len + fragheaderlen + transhdrlen + 20,
828 (flags & MSG_DONTWAIT), &err);
829
830 if (!skb)
831 return err;
832
833 /* reserve space for Hardware header */
834 skb_reserve(skb, hh_len);
835
836 /* create space for UDP/IP header */
837 skb_put(skb, fragheaderlen + transhdrlen);
838
839 /* initialize network header pointer */
840 skb_reset_network_header(skb);
841
842 /* initialize protocol header pointer */
843 skb->transport_header = skb->network_header + fragheaderlen;
844
845 skb->csum = 0;
846
847 __skb_queue_tail(queue, skb);
848 } else if (skb_is_gso(skb)) {
849 goto append;
850 }
851
852 skb->ip_summed = CHECKSUM_PARTIAL;
853 /* specify the length of each IP datagram fragment */
854 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
855 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
856
857 append:
858 return skb_append_datato_frags(sk, skb, getfrag, from,
859 (length - transhdrlen));
860 }
861
862 static int __ip_append_data(struct sock *sk,
863 struct flowi4 *fl4,
864 struct sk_buff_head *queue,
865 struct inet_cork *cork,
866 struct page_frag *pfrag,
867 int getfrag(void *from, char *to, int offset,
868 int len, int odd, struct sk_buff *skb),
869 void *from, int length, int transhdrlen,
870 unsigned int flags)
871 {
872 struct inet_sock *inet = inet_sk(sk);
873 struct sk_buff *skb;
874
875 struct ip_options *opt = cork->opt;
876 int hh_len;
877 int exthdrlen;
878 int mtu;
879 int copy;
880 int err;
881 int offset = 0;
882 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
883 int csummode = CHECKSUM_NONE;
884 struct rtable *rt = (struct rtable *)cork->dst;
885 u32 tskey = 0;
886
887 skb = skb_peek_tail(queue);
888
889 exthdrlen = !skb ? rt->dst.header_len : 0;
890 mtu = cork->fragsize;
891 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
892 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
893 tskey = sk->sk_tskey++;
894
895 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
896
897 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
898 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
899 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
900
901 if (cork->length + length > maxnonfragsize - fragheaderlen) {
902 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
903 mtu - (opt ? opt->optlen : 0));
904 return -EMSGSIZE;
905 }
906
907 /*
908 * transhdrlen > 0 means that this is the first fragment and we wish
909 * it won't be fragmented in the future.
910 */
911 if (transhdrlen &&
912 length + fragheaderlen <= mtu &&
913 rt->dst.dev->features & NETIF_F_V4_CSUM &&
914 !exthdrlen)
915 csummode = CHECKSUM_PARTIAL;
916
917 cork->length += length;
918 if (((length > mtu) || (skb && skb_is_gso(skb))) &&
919 (sk->sk_protocol == IPPROTO_UDP) &&
920 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len &&
921 (sk->sk_type == SOCK_DGRAM)) {
922 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
923 hh_len, fragheaderlen, transhdrlen,
924 maxfraglen, flags);
925 if (err)
926 goto error;
927 return 0;
928 }
929
930 /* So, what's going on in the loop below?
931 *
932 * We use calculated fragment length to generate chained skb,
933 * each of segments is IP fragment ready for sending to network after
934 * adding appropriate IP header.
935 */
936
937 if (!skb)
938 goto alloc_new_skb;
939
940 while (length > 0) {
941 /* Check if the remaining data fits into current packet. */
942 copy = mtu - skb->len;
943 if (copy < length)
944 copy = maxfraglen - skb->len;
945 if (copy <= 0) {
946 char *data;
947 unsigned int datalen;
948 unsigned int fraglen;
949 unsigned int fraggap;
950 unsigned int alloclen;
951 struct sk_buff *skb_prev;
952 alloc_new_skb:
953 skb_prev = skb;
954 if (skb_prev)
955 fraggap = skb_prev->len - maxfraglen;
956 else
957 fraggap = 0;
958
959 /*
960 * If remaining data exceeds the mtu,
961 * we know we need more fragment(s).
962 */
963 datalen = length + fraggap;
964 if (datalen > mtu - fragheaderlen)
965 datalen = maxfraglen - fragheaderlen;
966 fraglen = datalen + fragheaderlen;
967
968 if ((flags & MSG_MORE) &&
969 !(rt->dst.dev->features&NETIF_F_SG))
970 alloclen = mtu;
971 else
972 alloclen = fraglen;
973
974 alloclen += exthdrlen;
975
976 /* The last fragment gets additional space at tail.
977 * Note, with MSG_MORE we overallocate on fragments,
978 * because we have no idea what fragment will be
979 * the last.
980 */
981 if (datalen == length + fraggap)
982 alloclen += rt->dst.trailer_len;
983
984 if (transhdrlen) {
985 skb = sock_alloc_send_skb(sk,
986 alloclen + hh_len + 15,
987 (flags & MSG_DONTWAIT), &err);
988 } else {
989 skb = NULL;
990 if (atomic_read(&sk->sk_wmem_alloc) <=
991 2 * sk->sk_sndbuf)
992 skb = sock_wmalloc(sk,
993 alloclen + hh_len + 15, 1,
994 sk->sk_allocation);
995 if (unlikely(!skb))
996 err = -ENOBUFS;
997 }
998 if (!skb)
999 goto error;
1000
1001 /*
1002 * Fill in the control structures
1003 */
1004 skb->ip_summed = csummode;
1005 skb->csum = 0;
1006 skb_reserve(skb, hh_len);
1007
1008 /* only the initial fragment is time stamped */
1009 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1010 cork->tx_flags = 0;
1011 skb_shinfo(skb)->tskey = tskey;
1012 tskey = 0;
1013
1014 /*
1015 * Find where to start putting bytes.
1016 */
1017 data = skb_put(skb, fraglen + exthdrlen);
1018 skb_set_network_header(skb, exthdrlen);
1019 skb->transport_header = (skb->network_header +
1020 fragheaderlen);
1021 data += fragheaderlen + exthdrlen;
1022
1023 if (fraggap) {
1024 skb->csum = skb_copy_and_csum_bits(
1025 skb_prev, maxfraglen,
1026 data + transhdrlen, fraggap, 0);
1027 skb_prev->csum = csum_sub(skb_prev->csum,
1028 skb->csum);
1029 data += fraggap;
1030 pskb_trim_unique(skb_prev, maxfraglen);
1031 }
1032
1033 copy = datalen - transhdrlen - fraggap;
1034 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1035 err = -EFAULT;
1036 kfree_skb(skb);
1037 goto error;
1038 }
1039
1040 offset += copy;
1041 length -= datalen - fraggap;
1042 transhdrlen = 0;
1043 exthdrlen = 0;
1044 csummode = CHECKSUM_NONE;
1045
1046 /*
1047 * Put the packet on the pending queue.
1048 */
1049 __skb_queue_tail(queue, skb);
1050 continue;
1051 }
1052
1053 if (copy > length)
1054 copy = length;
1055
1056 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1057 unsigned int off;
1058
1059 off = skb->len;
1060 if (getfrag(from, skb_put(skb, copy),
1061 offset, copy, off, skb) < 0) {
1062 __skb_trim(skb, off);
1063 err = -EFAULT;
1064 goto error;
1065 }
1066 } else {
1067 int i = skb_shinfo(skb)->nr_frags;
1068
1069 err = -ENOMEM;
1070 if (!sk_page_frag_refill(sk, pfrag))
1071 goto error;
1072
1073 if (!skb_can_coalesce(skb, i, pfrag->page,
1074 pfrag->offset)) {
1075 err = -EMSGSIZE;
1076 if (i == MAX_SKB_FRAGS)
1077 goto error;
1078
1079 __skb_fill_page_desc(skb, i, pfrag->page,
1080 pfrag->offset, 0);
1081 skb_shinfo(skb)->nr_frags = ++i;
1082 get_page(pfrag->page);
1083 }
1084 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1085 if (getfrag(from,
1086 page_address(pfrag->page) + pfrag->offset,
1087 offset, copy, skb->len, skb) < 0)
1088 goto error_efault;
1089
1090 pfrag->offset += copy;
1091 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1092 skb->len += copy;
1093 skb->data_len += copy;
1094 skb->truesize += copy;
1095 atomic_add(copy, &sk->sk_wmem_alloc);
1096 }
1097 offset += copy;
1098 length -= copy;
1099 }
1100
1101 return 0;
1102
1103 error_efault:
1104 err = -EFAULT;
1105 error:
1106 cork->length -= length;
1107 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1108 return err;
1109 }
1110
1111 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1112 struct ipcm_cookie *ipc, struct rtable **rtp)
1113 {
1114 struct ip_options_rcu *opt;
1115 struct rtable *rt;
1116
1117 /*
1118 * setup for corking.
1119 */
1120 opt = ipc->opt;
1121 if (opt) {
1122 if (!cork->opt) {
1123 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1124 sk->sk_allocation);
1125 if (unlikely(!cork->opt))
1126 return -ENOBUFS;
1127 }
1128 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1129 cork->flags |= IPCORK_OPT;
1130 cork->addr = ipc->addr;
1131 }
1132 rt = *rtp;
1133 if (unlikely(!rt))
1134 return -EFAULT;
1135 /*
1136 * We steal reference to this route, caller should not release it
1137 */
1138 *rtp = NULL;
1139 cork->fragsize = ip_sk_use_pmtu(sk) ?
1140 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
1141 cork->dst = &rt->dst;
1142 cork->length = 0;
1143 cork->ttl = ipc->ttl;
1144 cork->tos = ipc->tos;
1145 cork->priority = ipc->priority;
1146 cork->tx_flags = ipc->tx_flags;
1147
1148 return 0;
1149 }
1150
1151 /*
1152 * ip_append_data() and ip_append_page() can make one large IP datagram
1153 * from many pieces of data. Each pieces will be holded on the socket
1154 * until ip_push_pending_frames() is called. Each piece can be a page
1155 * or non-page data.
1156 *
1157 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1158 * this interface potentially.
1159 *
1160 * LATER: length must be adjusted by pad at tail, when it is required.
1161 */
1162 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1163 int getfrag(void *from, char *to, int offset, int len,
1164 int odd, struct sk_buff *skb),
1165 void *from, int length, int transhdrlen,
1166 struct ipcm_cookie *ipc, struct rtable **rtp,
1167 unsigned int flags)
1168 {
1169 struct inet_sock *inet = inet_sk(sk);
1170 int err;
1171
1172 if (flags&MSG_PROBE)
1173 return 0;
1174
1175 if (skb_queue_empty(&sk->sk_write_queue)) {
1176 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1177 if (err)
1178 return err;
1179 } else {
1180 transhdrlen = 0;
1181 }
1182
1183 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1184 sk_page_frag(sk), getfrag,
1185 from, length, transhdrlen, flags);
1186 }
1187
1188 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1189 int offset, size_t size, int flags)
1190 {
1191 struct inet_sock *inet = inet_sk(sk);
1192 struct sk_buff *skb;
1193 struct rtable *rt;
1194 struct ip_options *opt = NULL;
1195 struct inet_cork *cork;
1196 int hh_len;
1197 int mtu;
1198 int len;
1199 int err;
1200 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1201
1202 if (inet->hdrincl)
1203 return -EPERM;
1204
1205 if (flags&MSG_PROBE)
1206 return 0;
1207
1208 if (skb_queue_empty(&sk->sk_write_queue))
1209 return -EINVAL;
1210
1211 cork = &inet->cork.base;
1212 rt = (struct rtable *)cork->dst;
1213 if (cork->flags & IPCORK_OPT)
1214 opt = cork->opt;
1215
1216 if (!(rt->dst.dev->features&NETIF_F_SG))
1217 return -EOPNOTSUPP;
1218
1219 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1220 mtu = cork->fragsize;
1221
1222 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1223 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1224 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1225
1226 if (cork->length + size > maxnonfragsize - fragheaderlen) {
1227 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1228 mtu - (opt ? opt->optlen : 0));
1229 return -EMSGSIZE;
1230 }
1231
1232 skb = skb_peek_tail(&sk->sk_write_queue);
1233 if (!skb)
1234 return -EINVAL;
1235
1236 cork->length += size;
1237 if ((size + skb->len > mtu) &&
1238 (sk->sk_protocol == IPPROTO_UDP) &&
1239 (rt->dst.dev->features & NETIF_F_UFO)) {
1240 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1241 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1242 }
1243
1244 while (size > 0) {
1245 if (skb_is_gso(skb)) {
1246 len = size;
1247 } else {
1248
1249 /* Check if the remaining data fits into current packet. */
1250 len = mtu - skb->len;
1251 if (len < size)
1252 len = maxfraglen - skb->len;
1253 }
1254 if (len <= 0) {
1255 struct sk_buff *skb_prev;
1256 int alloclen;
1257
1258 skb_prev = skb;
1259 fraggap = skb_prev->len - maxfraglen;
1260
1261 alloclen = fragheaderlen + hh_len + fraggap + 15;
1262 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1263 if (unlikely(!skb)) {
1264 err = -ENOBUFS;
1265 goto error;
1266 }
1267
1268 /*
1269 * Fill in the control structures
1270 */
1271 skb->ip_summed = CHECKSUM_NONE;
1272 skb->csum = 0;
1273 skb_reserve(skb, hh_len);
1274
1275 /*
1276 * Find where to start putting bytes.
1277 */
1278 skb_put(skb, fragheaderlen + fraggap);
1279 skb_reset_network_header(skb);
1280 skb->transport_header = (skb->network_header +
1281 fragheaderlen);
1282 if (fraggap) {
1283 skb->csum = skb_copy_and_csum_bits(skb_prev,
1284 maxfraglen,
1285 skb_transport_header(skb),
1286 fraggap, 0);
1287 skb_prev->csum = csum_sub(skb_prev->csum,
1288 skb->csum);
1289 pskb_trim_unique(skb_prev, maxfraglen);
1290 }
1291
1292 /*
1293 * Put the packet on the pending queue.
1294 */
1295 __skb_queue_tail(&sk->sk_write_queue, skb);
1296 continue;
1297 }
1298
1299 if (len > size)
1300 len = size;
1301
1302 if (skb_append_pagefrags(skb, page, offset, len)) {
1303 err = -EMSGSIZE;
1304 goto error;
1305 }
1306
1307 if (skb->ip_summed == CHECKSUM_NONE) {
1308 __wsum csum;
1309 csum = csum_page(page, offset, len);
1310 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1311 }
1312
1313 skb->len += len;
1314 skb->data_len += len;
1315 skb->truesize += len;
1316 atomic_add(len, &sk->sk_wmem_alloc);
1317 offset += len;
1318 size -= len;
1319 }
1320 return 0;
1321
1322 error:
1323 cork->length -= size;
1324 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1325 return err;
1326 }
1327
1328 static void ip_cork_release(struct inet_cork *cork)
1329 {
1330 cork->flags &= ~IPCORK_OPT;
1331 kfree(cork->opt);
1332 cork->opt = NULL;
1333 dst_release(cork->dst);
1334 cork->dst = NULL;
1335 }
1336
1337 /*
1338 * Combined all pending IP fragments on the socket as one IP datagram
1339 * and push them out.
1340 */
1341 struct sk_buff *__ip_make_skb(struct sock *sk,
1342 struct flowi4 *fl4,
1343 struct sk_buff_head *queue,
1344 struct inet_cork *cork)
1345 {
1346 struct sk_buff *skb, *tmp_skb;
1347 struct sk_buff **tail_skb;
1348 struct inet_sock *inet = inet_sk(sk);
1349 struct net *net = sock_net(sk);
1350 struct ip_options *opt = NULL;
1351 struct rtable *rt = (struct rtable *)cork->dst;
1352 struct iphdr *iph;
1353 __be16 df = 0;
1354 __u8 ttl;
1355
1356 skb = __skb_dequeue(queue);
1357 if (!skb)
1358 goto out;
1359 tail_skb = &(skb_shinfo(skb)->frag_list);
1360
1361 /* move skb->data to ip header from ext header */
1362 if (skb->data < skb_network_header(skb))
1363 __skb_pull(skb, skb_network_offset(skb));
1364 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1365 __skb_pull(tmp_skb, skb_network_header_len(skb));
1366 *tail_skb = tmp_skb;
1367 tail_skb = &(tmp_skb->next);
1368 skb->len += tmp_skb->len;
1369 skb->data_len += tmp_skb->len;
1370 skb->truesize += tmp_skb->truesize;
1371 tmp_skb->destructor = NULL;
1372 tmp_skb->sk = NULL;
1373 }
1374
1375 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1376 * to fragment the frame generated here. No matter, what transforms
1377 * how transforms change size of the packet, it will come out.
1378 */
1379 skb->ignore_df = ip_sk_ignore_df(sk);
1380
1381 /* DF bit is set when we want to see DF on outgoing frames.
1382 * If ignore_df is set too, we still allow to fragment this frame
1383 * locally. */
1384 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1385 inet->pmtudisc == IP_PMTUDISC_PROBE ||
1386 (skb->len <= dst_mtu(&rt->dst) &&
1387 ip_dont_fragment(sk, &rt->dst)))
1388 df = htons(IP_DF);
1389
1390 if (cork->flags & IPCORK_OPT)
1391 opt = cork->opt;
1392
1393 if (cork->ttl != 0)
1394 ttl = cork->ttl;
1395 else if (rt->rt_type == RTN_MULTICAST)
1396 ttl = inet->mc_ttl;
1397 else
1398 ttl = ip_select_ttl(inet, &rt->dst);
1399
1400 iph = ip_hdr(skb);
1401 iph->version = 4;
1402 iph->ihl = 5;
1403 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1404 iph->frag_off = df;
1405 iph->ttl = ttl;
1406 iph->protocol = sk->sk_protocol;
1407 ip_copy_addrs(iph, fl4);
1408 ip_select_ident(net, skb, sk);
1409
1410 if (opt) {
1411 iph->ihl += opt->optlen>>2;
1412 ip_options_build(skb, opt, cork->addr, rt, 0);
1413 }
1414
1415 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1416 skb->mark = sk->sk_mark;
1417 /*
1418 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1419 * on dst refcount
1420 */
1421 cork->dst = NULL;
1422 skb_dst_set(skb, &rt->dst);
1423
1424 if (iph->protocol == IPPROTO_ICMP)
1425 icmp_out_count(net, ((struct icmphdr *)
1426 skb_transport_header(skb))->type);
1427
1428 ip_cork_release(cork);
1429 out:
1430 return skb;
1431 }
1432
1433 int ip_send_skb(struct net *net, struct sk_buff *skb)
1434 {
1435 int err;
1436
1437 err = ip_local_out(skb);
1438 if (err) {
1439 if (err > 0)
1440 err = net_xmit_errno(err);
1441 if (err)
1442 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1443 }
1444
1445 return err;
1446 }
1447
1448 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1449 {
1450 struct sk_buff *skb;
1451
1452 skb = ip_finish_skb(sk, fl4);
1453 if (!skb)
1454 return 0;
1455
1456 /* Netfilter gets whole the not fragmented skb. */
1457 return ip_send_skb(sock_net(sk), skb);
1458 }
1459
1460 /*
1461 * Throw away all pending data on the socket.
1462 */
1463 static void __ip_flush_pending_frames(struct sock *sk,
1464 struct sk_buff_head *queue,
1465 struct inet_cork *cork)
1466 {
1467 struct sk_buff *skb;
1468
1469 while ((skb = __skb_dequeue_tail(queue)) != NULL)
1470 kfree_skb(skb);
1471
1472 ip_cork_release(cork);
1473 }
1474
1475 void ip_flush_pending_frames(struct sock *sk)
1476 {
1477 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1478 }
1479
1480 struct sk_buff *ip_make_skb(struct sock *sk,
1481 struct flowi4 *fl4,
1482 int getfrag(void *from, char *to, int offset,
1483 int len, int odd, struct sk_buff *skb),
1484 void *from, int length, int transhdrlen,
1485 struct ipcm_cookie *ipc, struct rtable **rtp,
1486 unsigned int flags)
1487 {
1488 struct inet_cork cork;
1489 struct sk_buff_head queue;
1490 int err;
1491
1492 if (flags & MSG_PROBE)
1493 return NULL;
1494
1495 __skb_queue_head_init(&queue);
1496
1497 cork.flags = 0;
1498 cork.addr = 0;
1499 cork.opt = NULL;
1500 err = ip_setup_cork(sk, &cork, ipc, rtp);
1501 if (err)
1502 return ERR_PTR(err);
1503
1504 err = __ip_append_data(sk, fl4, &queue, &cork,
1505 &current->task_frag, getfrag,
1506 from, length, transhdrlen, flags);
1507 if (err) {
1508 __ip_flush_pending_frames(sk, &queue, &cork);
1509 return ERR_PTR(err);
1510 }
1511
1512 return __ip_make_skb(sk, fl4, &queue, &cork);
1513 }
1514
1515 /*
1516 * Fetch data from kernel space and fill in checksum if needed.
1517 */
1518 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1519 int len, int odd, struct sk_buff *skb)
1520 {
1521 __wsum csum;
1522
1523 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1524 skb->csum = csum_block_add(skb->csum, csum, odd);
1525 return 0;
1526 }
1527
1528 /*
1529 * Generic function to send a packet as reply to another packet.
1530 * Used to send some TCP resets/acks so far.
1531 */
1532 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1533 const struct ip_options *sopt,
1534 __be32 daddr, __be32 saddr,
1535 const struct ip_reply_arg *arg,
1536 unsigned int len)
1537 {
1538 struct ip_options_data replyopts;
1539 struct ipcm_cookie ipc;
1540 struct flowi4 fl4;
1541 struct rtable *rt = skb_rtable(skb);
1542 struct net *net = sock_net(sk);
1543 struct sk_buff *nskb;
1544 int err;
1545 int oif;
1546
1547 if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
1548 return;
1549
1550 ipc.addr = daddr;
1551 ipc.opt = NULL;
1552 ipc.tx_flags = 0;
1553 ipc.ttl = 0;
1554 ipc.tos = -1;
1555
1556 if (replyopts.opt.opt.optlen) {
1557 ipc.opt = &replyopts.opt;
1558
1559 if (replyopts.opt.opt.srr)
1560 daddr = replyopts.opt.opt.faddr;
1561 }
1562
1563 oif = arg->bound_dev_if;
1564 if (!oif && netif_index_is_vrf(net, skb->skb_iif))
1565 oif = skb->skb_iif;
1566
1567 flowi4_init_output(&fl4, oif,
1568 IP4_REPLY_MARK(net, skb->mark),
1569 RT_TOS(arg->tos),
1570 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1571 ip_reply_arg_flowi_flags(arg),
1572 daddr, saddr,
1573 tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1574 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1575 rt = ip_route_output_key(net, &fl4);
1576 if (IS_ERR(rt))
1577 return;
1578
1579 inet_sk(sk)->tos = arg->tos;
1580
1581 sk->sk_priority = skb->priority;
1582 sk->sk_protocol = ip_hdr(skb)->protocol;
1583 sk->sk_bound_dev_if = arg->bound_dev_if;
1584 sk->sk_sndbuf = sysctl_wmem_default;
1585 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1586 len, 0, &ipc, &rt, MSG_DONTWAIT);
1587 if (unlikely(err)) {
1588 ip_flush_pending_frames(sk);
1589 goto out;
1590 }
1591
1592 nskb = skb_peek(&sk->sk_write_queue);
1593 if (nskb) {
1594 if (arg->csumoffset >= 0)
1595 *((__sum16 *)skb_transport_header(nskb) +
1596 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1597 arg->csum));
1598 nskb->ip_summed = CHECKSUM_NONE;
1599 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
1600 ip_push_pending_frames(sk, &fl4);
1601 }
1602 out:
1603 ip_rt_put(rt);
1604 }
1605
1606 void __init ip_init(void)
1607 {
1608 ip_rt_init();
1609 inet_initpeers();
1610
1611 #if defined(CONFIG_IP_MULTICAST)
1612 igmp_mc_init();
1613 #endif
1614 }
Linux kernel - Deliverables
This page took 0.061128 seconds and 6 git commands to generate.