projects / deliverable / linux.git / blob
? search:


66c627b85a91dcc5e49cba3714df5d795954f32f
[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
87 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
88 unsigned int mtu,
89 int (*output)(struct net *, struct sock *, struct sk_buff *));
90
91 /* Generate a checksum for an outgoing IP datagram. */
92 void ip_send_check(struct iphdr *iph)
93 {
94 iph->check = 0;
95 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
96 }
97 EXPORT_SYMBOL(ip_send_check);
98
99 int __ip_local_out(struct sock *sk, struct sk_buff *skb)
100 {
101 struct net *net = dev_net(skb_dst(skb)->dev);
102 struct iphdr *iph = ip_hdr(skb);
103
104 iph->tot_len = htons(skb->len);
105 ip_send_check(iph);
106 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
107 net, sk, skb, NULL, skb_dst(skb)->dev,
108 dst_output);
109 }
110
111 int ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
112 {
113 struct net *net = dev_net(skb_dst(skb)->dev);
114 int err;
115
116 err = __ip_local_out(sk, skb);
117 if (likely(err == 1))
118 err = dst_output(net, sk, skb);
119
120 return err;
121 }
122 EXPORT_SYMBOL_GPL(ip_local_out_sk);
123
124 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
125 {
126 int ttl = inet->uc_ttl;
127
128 if (ttl < 0)
129 ttl = ip4_dst_hoplimit(dst);
130 return ttl;
131 }
132
133 /*
134 * Add an ip header to a skbuff and send it out.
135 *
136 */
137 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
138 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
139 {
140 struct inet_sock *inet = inet_sk(sk);
141 struct rtable *rt = skb_rtable(skb);
142 struct iphdr *iph;
143
144 /* Build the IP header. */
145 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
146 skb_reset_network_header(skb);
147 iph = ip_hdr(skb);
148 iph->version = 4;
149 iph->ihl = 5;
150 iph->tos = inet->tos;
151 iph->ttl = ip_select_ttl(inet, &rt->dst);
152 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
153 iph->saddr = saddr;
154 iph->protocol = sk->sk_protocol;
155 if (ip_dont_fragment(sk, &rt->dst)) {
156 iph->frag_off = htons(IP_DF);
157 iph->id = 0;
158 } else {
159 iph->frag_off = 0;
160 __ip_select_ident(sock_net(sk), iph, 1);
161 }
162
163 if (opt && opt->opt.optlen) {
164 iph->ihl += opt->opt.optlen>>2;
165 ip_options_build(skb, &opt->opt, daddr, rt, 0);
166 }
167
168 skb->priority = sk->sk_priority;
169 skb->mark = sk->sk_mark;
170
171 /* Send it out. */
172 return ip_local_out(skb);
173 }
174 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
175
176 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
177 {
178 struct dst_entry *dst = skb_dst(skb);
179 struct rtable *rt = (struct rtable *)dst;
180 struct net_device *dev = dst->dev;
181 unsigned int hh_len = LL_RESERVED_SPACE(dev);
182 struct neighbour *neigh;
183 u32 nexthop;
184
185 if (rt->rt_type == RTN_MULTICAST) {
186 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
187 } else if (rt->rt_type == RTN_BROADCAST)
188 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
189
190 /* Be paranoid, rather than too clever. */
191 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
192 struct sk_buff *skb2;
193
194 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
195 if (!skb2) {
196 kfree_skb(skb);
197 return -ENOMEM;
198 }
199 if (skb->sk)
200 skb_set_owner_w(skb2, skb->sk);
201 consume_skb(skb);
202 skb = skb2;
203 }
204
205 rcu_read_lock_bh();
206 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
207 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
208 if (unlikely(!neigh))
209 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
210 if (!IS_ERR(neigh)) {
211 int res = dst_neigh_output(dst, neigh, skb);
212
213 rcu_read_unlock_bh();
214 return res;
215 }
216 rcu_read_unlock_bh();
217
218 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
219 __func__);
220 kfree_skb(skb);
221 return -EINVAL;
222 }
223
224 static int ip_finish_output_gso(struct net *net, struct sock *sk,
225 struct sk_buff *skb, unsigned int mtu)
226 {
227 netdev_features_t features;
228 struct sk_buff *segs;
229 int ret = 0;
230
231 /* common case: locally created skb or seglen is <= mtu */
232 if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) ||
233 skb_gso_network_seglen(skb) <= mtu)
234 return ip_finish_output2(net, sk, skb);
235
236 /* Slowpath - GSO segment length is exceeding the dst MTU.
237 *
238 * This can happen in two cases:
239 * 1) TCP GRO packet, DF bit not set
240 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
241 * from host network stack.
242 */
243 features = netif_skb_features(skb);
244 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
245 if (IS_ERR_OR_NULL(segs)) {
246 kfree_skb(skb);
247 return -ENOMEM;
248 }
249
250 consume_skb(skb);
251
252 do {
253 struct sk_buff *nskb = segs->next;
254 int err;
255
256 segs->next = NULL;
257 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
258
259 if (err && ret == 0)
260 ret = err;
261 segs = nskb;
262 } while (segs);
263
264 return ret;
265 }
266
267 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
268 {
269 unsigned int mtu;
270
271 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
272 /* Policy lookup after SNAT yielded a new policy */
273 if (skb_dst(skb)->xfrm) {
274 IPCB(skb)->flags |= IPSKB_REROUTED;
275 return dst_output(net, sk, skb);
276 }
277 #endif
278 mtu = ip_skb_dst_mtu(skb);
279 if (skb_is_gso(skb))
280 return ip_finish_output_gso(net, sk, skb, mtu);
281
282 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
283 return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
284
285 return ip_finish_output2(net, sk, skb);
286 }
287
288 int ip_mc_output(struct sock *sk, struct sk_buff *skb)
289 {
290 struct rtable *rt = skb_rtable(skb);
291 struct net_device *dev = rt->dst.dev;
292 struct net *net = dev_net(dev);
293
294 /*
295 * If the indicated interface is up and running, send the packet.
296 */
297 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
298
299 skb->dev = dev;
300 skb->protocol = htons(ETH_P_IP);
301
302 /*
303 * Multicasts are looped back for other local users
304 */
305
306 if (rt->rt_flags&RTCF_MULTICAST) {
307 if (sk_mc_loop(sk)
308 #ifdef CONFIG_IP_MROUTE
309 /* Small optimization: do not loopback not local frames,
310 which returned after forwarding; they will be dropped
311 by ip_mr_input in any case.
312 Note, that local frames are looped back to be delivered
313 to local recipients.
314
315 This check is duplicated in ip_mr_input at the moment.
316 */
317 &&
318 ((rt->rt_flags & RTCF_LOCAL) ||
319 !(IPCB(skb)->flags & IPSKB_FORWARDED))
320 #endif
321 ) {
322 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
323 if (newskb)
324 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
325 net, sk, newskb, NULL, newskb->dev,
326 dev_loopback_xmit);
327 }
328
329 /* Multicasts with ttl 0 must not go beyond the host */
330
331 if (ip_hdr(skb)->ttl == 0) {
332 kfree_skb(skb);
333 return 0;
334 }
335 }
336
337 if (rt->rt_flags&RTCF_BROADCAST) {
338 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
339 if (newskb)
340 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
341 net, sk, newskb, NULL, newskb->dev,
342 dev_loopback_xmit);
343 }
344
345 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
346 net, sk, skb, NULL, skb->dev,
347 ip_finish_output,
348 !(IPCB(skb)->flags & IPSKB_REROUTED));
349 }
350
351 int ip_output(struct sock *sk, struct sk_buff *skb)
352 {
353 struct net_device *dev = skb_dst(skb)->dev;
354 struct net *net = dev_net(dev);
355
356 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
357
358 skb->dev = dev;
359 skb->protocol = htons(ETH_P_IP);
360
361 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
362 net, sk, skb, NULL, dev,
363 ip_finish_output,
364 !(IPCB(skb)->flags & IPSKB_REROUTED));
365 }
366
367 /*
368 * copy saddr and daddr, possibly using 64bit load/stores
369 * Equivalent to :
370 * iph->saddr = fl4->saddr;
371 * iph->daddr = fl4->daddr;
372 */
373 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
374 {
375 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
376 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
377 memcpy(&iph->saddr, &fl4->saddr,
378 sizeof(fl4->saddr) + sizeof(fl4->daddr));
379 }
380
381 /* Note: skb->sk can be different from sk, in case of tunnels */
382 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
383 {
384 struct inet_sock *inet = inet_sk(sk);
385 struct ip_options_rcu *inet_opt;
386 struct flowi4 *fl4;
387 struct rtable *rt;
388 struct iphdr *iph;
389 int res;
390
391 /* Skip all of this if the packet is already routed,
392 * f.e. by something like SCTP.
393 */
394 rcu_read_lock();
395 inet_opt = rcu_dereference(inet->inet_opt);
396 fl4 = &fl->u.ip4;
397 rt = skb_rtable(skb);
398 if (rt)
399 goto packet_routed;
400
401 /* Make sure we can route this packet. */
402 rt = (struct rtable *)__sk_dst_check(sk, 0);
403 if (!rt) {
404 __be32 daddr;
405
406 /* Use correct destination address if we have options. */
407 daddr = inet->inet_daddr;
408 if (inet_opt && inet_opt->opt.srr)
409 daddr = inet_opt->opt.faddr;
410
411 /* If this fails, retransmit mechanism of transport layer will
412 * keep trying until route appears or the connection times
413 * itself out.
414 */
415 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
416 daddr, inet->inet_saddr,
417 inet->inet_dport,
418 inet->inet_sport,
419 sk->sk_protocol,
420 RT_CONN_FLAGS(sk),
421 sk->sk_bound_dev_if);
422 if (IS_ERR(rt))
423 goto no_route;
424 sk_setup_caps(sk, &rt->dst);
425 }
426 skb_dst_set_noref(skb, &rt->dst);
427
428 packet_routed:
429 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
430 goto no_route;
431
432 /* OK, we know where to send it, allocate and build IP header. */
433 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
434 skb_reset_network_header(skb);
435 iph = ip_hdr(skb);
436 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
437 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
438 iph->frag_off = htons(IP_DF);
439 else
440 iph->frag_off = 0;
441 iph->ttl = ip_select_ttl(inet, &rt->dst);
442 iph->protocol = sk->sk_protocol;
443 ip_copy_addrs(iph, fl4);
444
445 /* Transport layer set skb->h.foo itself. */
446
447 if (inet_opt && inet_opt->opt.optlen) {
448 iph->ihl += inet_opt->opt.optlen >> 2;
449 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
450 }
451
452 ip_select_ident_segs(sock_net(sk), skb, sk,
453 skb_shinfo(skb)->gso_segs ?: 1);
454
455 /* TODO : should we use skb->sk here instead of sk ? */
456 skb->priority = sk->sk_priority;
457 skb->mark = sk->sk_mark;
458
459 res = ip_local_out_sk(sk, skb);
460 rcu_read_unlock();
461 return res;
462
463 no_route:
464 rcu_read_unlock();
465 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
466 kfree_skb(skb);
467 return -EHOSTUNREACH;
468 }
469 EXPORT_SYMBOL(ip_queue_xmit);
470
471 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
472 {
473 to->pkt_type = from->pkt_type;
474 to->priority = from->priority;
475 to->protocol = from->protocol;
476 skb_dst_drop(to);
477 skb_dst_copy(to, from);
478 to->dev = from->dev;
479 to->mark = from->mark;
480
481 /* Copy the flags to each fragment. */
482 IPCB(to)->flags = IPCB(from)->flags;
483
484 #ifdef CONFIG_NET_SCHED
485 to->tc_index = from->tc_index;
486 #endif
487 nf_copy(to, from);
488 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
489 to->ipvs_property = from->ipvs_property;
490 #endif
491 skb_copy_secmark(to, from);
492 }
493
494 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
495 unsigned int mtu,
496 int (*output)(struct net *, struct sock *, struct sk_buff *))
497 {
498 struct iphdr *iph = ip_hdr(skb);
499
500 if ((iph->frag_off & htons(IP_DF)) == 0)
501 return ip_do_fragment(net, sk, skb, output);
502
503 if (unlikely(!skb->ignore_df ||
504 (IPCB(skb)->frag_max_size &&
505 IPCB(skb)->frag_max_size > mtu))) {
506 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
507 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
508 htonl(mtu));
509 kfree_skb(skb);
510 return -EMSGSIZE;
511 }
512
513 return ip_do_fragment(net, sk, skb, output);
514 }
515
516 /*
517 * This IP datagram is too large to be sent in one piece. Break it up into
518 * smaller pieces (each of size equal to IP header plus
519 * a block of the data of the original IP data part) that will yet fit in a
520 * single device frame, and queue such a frame for sending.
521 */
522
523 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
524 int (*output)(struct net *, struct sock *, struct sk_buff *))
525 {
526 struct iphdr *iph;
527 int ptr;
528 struct net_device *dev;
529 struct sk_buff *skb2;
530 unsigned int mtu, hlen, left, len, ll_rs;
531 int offset;
532 __be16 not_last_frag;
533 struct rtable *rt = skb_rtable(skb);
534 int err = 0;
535
536 dev = rt->dst.dev;
537
538 /*
539 * Point into the IP datagram header.
540 */
541
542 iph = ip_hdr(skb);
543
544 mtu = ip_skb_dst_mtu(skb);
545 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
546 mtu = IPCB(skb)->frag_max_size;
547
548 /*
549 * Setup starting values.
550 */
551
552 hlen = iph->ihl * 4;
553 mtu = mtu - hlen; /* Size of data space */
554 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
555
556 /* When frag_list is given, use it. First, check its validity:
557 * some transformers could create wrong frag_list or break existing
558 * one, it is not prohibited. In this case fall back to copying.
559 *
560 * LATER: this step can be merged to real generation of fragments,
561 * we can switch to copy when see the first bad fragment.
562 */
563 if (skb_has_frag_list(skb)) {
564 struct sk_buff *frag, *frag2;
565 int first_len = skb_pagelen(skb);
566
567 if (first_len - hlen > mtu ||
568 ((first_len - hlen) & 7) ||
569 ip_is_fragment(iph) ||
570 skb_cloned(skb))
571 goto slow_path;
572
573 skb_walk_frags(skb, frag) {
574 /* Correct geometry. */
575 if (frag->len > mtu ||
576 ((frag->len & 7) && frag->next) ||
577 skb_headroom(frag) < hlen)
578 goto slow_path_clean;
579
580 /* Partially cloned skb? */
581 if (skb_shared(frag))
582 goto slow_path_clean;
583
584 BUG_ON(frag->sk);
585 if (skb->sk) {
586 frag->sk = skb->sk;
587 frag->destructor = sock_wfree;
588 }
589 skb->truesize -= frag->truesize;
590 }
591
592 /* Everything is OK. Generate! */
593
594 err = 0;
595 offset = 0;
596 frag = skb_shinfo(skb)->frag_list;
597 skb_frag_list_init(skb);
598 skb->data_len = first_len - skb_headlen(skb);
599 skb->len = first_len;
600 iph->tot_len = htons(first_len);
601 iph->frag_off = htons(IP_MF);
602 ip_send_check(iph);
603
604 for (;;) {
605 /* Prepare header of the next frame,
606 * before previous one went down. */
607 if (frag) {
608 frag->ip_summed = CHECKSUM_NONE;
609 skb_reset_transport_header(frag);
610 __skb_push(frag, hlen);
611 skb_reset_network_header(frag);
612 memcpy(skb_network_header(frag), iph, hlen);
613 iph = ip_hdr(frag);
614 iph->tot_len = htons(frag->len);
615 ip_copy_metadata(frag, skb);
616 if (offset == 0)
617 ip_options_fragment(frag);
618 offset += skb->len - hlen;
619 iph->frag_off = htons(offset>>3);
620 if (frag->next)
621 iph->frag_off |= htons(IP_MF);
622 /* Ready, complete checksum */
623 ip_send_check(iph);
624 }
625
626 err = output(net, sk, skb);
627
628 if (!err)
629 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
630 if (err || !frag)
631 break;
632
633 skb = frag;
634 frag = skb->next;
635 skb->next = NULL;
636 }
637
638 if (err == 0) {
639 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
640 return 0;
641 }
642
643 while (frag) {
644 skb = frag->next;
645 kfree_skb(frag);
646 frag = skb;
647 }
648 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
649 return err;
650
651 slow_path_clean:
652 skb_walk_frags(skb, frag2) {
653 if (frag2 == frag)
654 break;
655 frag2->sk = NULL;
656 frag2->destructor = NULL;
657 skb->truesize += frag2->truesize;
658 }
659 }
660
661 slow_path:
662 /* for offloaded checksums cleanup checksum before fragmentation */
663 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
664 goto fail;
665 iph = ip_hdr(skb);
666
667 left = skb->len - hlen; /* Space per frame */
668 ptr = hlen; /* Where to start from */
669
670 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
671
672 /*
673 * Fragment the datagram.
674 */
675
676 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
677 not_last_frag = iph->frag_off & htons(IP_MF);
678
679 /*
680 * Keep copying data until we run out.
681 */
682
683 while (left > 0) {
684 len = left;
685 /* IF: it doesn't fit, use 'mtu' - the data space left */
686 if (len > mtu)
687 len = mtu;
688 /* IF: we are not sending up to and including the packet end
689 then align the next start on an eight byte boundary */
690 if (len < left) {
691 len &= ~7;
692 }
693
694 /* Allocate buffer */
695 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
696 if (!skb2) {
697 err = -ENOMEM;
698 goto fail;
699 }
700
701 /*
702 * Set up data on packet
703 */
704
705 ip_copy_metadata(skb2, skb);
706 skb_reserve(skb2, ll_rs);
707 skb_put(skb2, len + hlen);
708 skb_reset_network_header(skb2);
709 skb2->transport_header = skb2->network_header + hlen;
710
711 /*
712 * Charge the memory for the fragment to any owner
713 * it might possess
714 */
715
716 if (skb->sk)
717 skb_set_owner_w(skb2, skb->sk);
718
719 /*
720 * Copy the packet header into the new buffer.
721 */
722
723 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
724
725 /*
726 * Copy a block of the IP datagram.
727 */
728 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
729 BUG();
730 left -= len;
731
732 /*
733 * Fill in the new header fields.
734 */
735 iph = ip_hdr(skb2);
736 iph->frag_off = htons((offset >> 3));
737
738 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
739 iph->frag_off |= htons(IP_DF);
740
741 /* ANK: dirty, but effective trick. Upgrade options only if
742 * the segment to be fragmented was THE FIRST (otherwise,
743 * options are already fixed) and make it ONCE
744 * on the initial skb, so that all the following fragments
745 * will inherit fixed options.
746 */
747 if (offset == 0)
748 ip_options_fragment(skb);
749
750 /*
751 * Added AC : If we are fragmenting a fragment that's not the
752 * last fragment then keep MF on each bit
753 */
754 if (left > 0 || not_last_frag)
755 iph->frag_off |= htons(IP_MF);
756 ptr += len;
757 offset += len;
758
759 /*
760 * Put this fragment into the sending queue.
761 */
762 iph->tot_len = htons(len + hlen);
763
764 ip_send_check(iph);
765
766 err = output(net, sk, skb2);
767 if (err)
768 goto fail;
769
770 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
771 }
772 consume_skb(skb);
773 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
774 return err;
775
776 fail:
777 kfree_skb(skb);
778 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
779 return err;
780 }
781 EXPORT_SYMBOL(ip_do_fragment);
782
783 int
784 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
785 {
786 struct msghdr *msg = from;
787
788 if (skb->ip_summed == CHECKSUM_PARTIAL) {
789 if (copy_from_iter(to, len, &msg->msg_iter) != len)
790 return -EFAULT;
791 } else {
792 __wsum csum = 0;
793 if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len)
794 return -EFAULT;
795 skb->csum = csum_block_add(skb->csum, csum, odd);
796 }
797 return 0;
798 }
799 EXPORT_SYMBOL(ip_generic_getfrag);
800
801 static inline __wsum
802 csum_page(struct page *page, int offset, int copy)
803 {
804 char *kaddr;
805 __wsum csum;
806 kaddr = kmap(page);
807 csum = csum_partial(kaddr + offset, copy, 0);
808 kunmap(page);
809 return csum;
810 }
811
812 static inline int ip_ufo_append_data(struct sock *sk,
813 struct sk_buff_head *queue,
814 int getfrag(void *from, char *to, int offset, int len,
815 int odd, struct sk_buff *skb),
816 void *from, int length, int hh_len, int fragheaderlen,
817 int transhdrlen, int maxfraglen, unsigned int flags)
818 {
819 struct sk_buff *skb;
820 int err;
821
822 /* There is support for UDP fragmentation offload by network
823 * device, so create one single skb packet containing complete
824 * udp datagram
825 */
826 skb = skb_peek_tail(queue);
827 if (!skb) {
828 skb = sock_alloc_send_skb(sk,
829 hh_len + fragheaderlen + transhdrlen + 20,
830 (flags & MSG_DONTWAIT), &err);
831
832 if (!skb)
833 return err;
834
835 /* reserve space for Hardware header */
836 skb_reserve(skb, hh_len);
837
838 /* create space for UDP/IP header */
839 skb_put(skb, fragheaderlen + transhdrlen);
840
841 /* initialize network header pointer */
842 skb_reset_network_header(skb);
843
844 /* initialize protocol header pointer */
845 skb->transport_header = skb->network_header + fragheaderlen;
846
847 skb->csum = 0;
848
849 __skb_queue_tail(queue, skb);
850 } else if (skb_is_gso(skb)) {
851 goto append;
852 }
853
854 skb->ip_summed = CHECKSUM_PARTIAL;
855 /* specify the length of each IP datagram fragment */
856 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
857 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
858
859 append:
860 return skb_append_datato_frags(sk, skb, getfrag, from,
861 (length - transhdrlen));
862 }
863
864 static int __ip_append_data(struct sock *sk,
865 struct flowi4 *fl4,
866 struct sk_buff_head *queue,
867 struct inet_cork *cork,
868 struct page_frag *pfrag,
869 int getfrag(void *from, char *to, int offset,
870 int len, int odd, struct sk_buff *skb),
871 void *from, int length, int transhdrlen,
872 unsigned int flags)
873 {
874 struct inet_sock *inet = inet_sk(sk);
875 struct sk_buff *skb;
876
877 struct ip_options *opt = cork->opt;
878 int hh_len;
879 int exthdrlen;
880 int mtu;
881 int copy;
882 int err;
883 int offset = 0;
884 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
885 int csummode = CHECKSUM_NONE;
886 struct rtable *rt = (struct rtable *)cork->dst;
887 u32 tskey = 0;
888
889 skb = skb_peek_tail(queue);
890
891 exthdrlen = !skb ? rt->dst.header_len : 0;
892 mtu = cork->fragsize;
893 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
894 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
895 tskey = sk->sk_tskey++;
896
897 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
898
899 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
900 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
901 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
902
903 if (cork->length + length > maxnonfragsize - fragheaderlen) {
904 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
905 mtu - (opt ? opt->optlen : 0));
906 return -EMSGSIZE;
907 }
908
909 /*
910 * transhdrlen > 0 means that this is the first fragment and we wish
911 * it won't be fragmented in the future.
912 */
913 if (transhdrlen &&
914 length + fragheaderlen <= mtu &&
915 rt->dst.dev->features & NETIF_F_V4_CSUM &&
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)) {
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 cork->length += size;
1239 if ((size + skb->len > mtu) &&
1240 (sk->sk_protocol == IPPROTO_UDP) &&
1241 (rt->dst.dev->features & NETIF_F_UFO)) {
1242 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1243 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1244 }
1245
1246 while (size > 0) {
1247 if (skb_is_gso(skb)) {
1248 len = size;
1249 } else {
1250
1251 /* Check if the remaining data fits into current packet. */
1252 len = mtu - skb->len;
1253 if (len < size)
1254 len = maxfraglen - skb->len;
1255 }
1256 if (len <= 0) {
1257 struct sk_buff *skb_prev;
1258 int alloclen;
1259
1260 skb_prev = skb;
1261 fraggap = skb_prev->len - maxfraglen;
1262
1263 alloclen = fragheaderlen + hh_len + fraggap + 15;
1264 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1265 if (unlikely(!skb)) {
1266 err = -ENOBUFS;
1267 goto error;
1268 }
1269
1270 /*
1271 * Fill in the control structures
1272 */
1273 skb->ip_summed = CHECKSUM_NONE;
1274 skb->csum = 0;
1275 skb_reserve(skb, hh_len);
1276
1277 /*
1278 * Find where to start putting bytes.
1279 */
1280 skb_put(skb, fragheaderlen + fraggap);
1281 skb_reset_network_header(skb);
1282 skb->transport_header = (skb->network_header +
1283 fragheaderlen);
1284 if (fraggap) {
1285 skb->csum = skb_copy_and_csum_bits(skb_prev,
1286 maxfraglen,
1287 skb_transport_header(skb),
1288 fraggap, 0);
1289 skb_prev->csum = csum_sub(skb_prev->csum,
1290 skb->csum);
1291 pskb_trim_unique(skb_prev, maxfraglen);
1292 }
1293
1294 /*
1295 * Put the packet on the pending queue.
1296 */
1297 __skb_queue_tail(&sk->sk_write_queue, skb);
1298 continue;
1299 }
1300
1301 if (len > size)
1302 len = size;
1303
1304 if (skb_append_pagefrags(skb, page, offset, len)) {
1305 err = -EMSGSIZE;
1306 goto error;
1307 }
1308
1309 if (skb->ip_summed == CHECKSUM_NONE) {
1310 __wsum csum;
1311 csum = csum_page(page, offset, len);
1312 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1313 }
1314
1315 skb->len += len;
1316 skb->data_len += len;
1317 skb->truesize += len;
1318 atomic_add(len, &sk->sk_wmem_alloc);
1319 offset += len;
1320 size -= len;
1321 }
1322 return 0;
1323
1324 error:
1325 cork->length -= size;
1326 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1327 return err;
1328 }
1329
1330 static void ip_cork_release(struct inet_cork *cork)
1331 {
1332 cork->flags &= ~IPCORK_OPT;
1333 kfree(cork->opt);
1334 cork->opt = NULL;
1335 dst_release(cork->dst);
1336 cork->dst = NULL;
1337 }
1338
1339 /*
1340 * Combined all pending IP fragments on the socket as one IP datagram
1341 * and push them out.
1342 */
1343 struct sk_buff *__ip_make_skb(struct sock *sk,
1344 struct flowi4 *fl4,
1345 struct sk_buff_head *queue,
1346 struct inet_cork *cork)
1347 {
1348 struct sk_buff *skb, *tmp_skb;
1349 struct sk_buff **tail_skb;
1350 struct inet_sock *inet = inet_sk(sk);
1351 struct net *net = sock_net(sk);
1352 struct ip_options *opt = NULL;
1353 struct rtable *rt = (struct rtable *)cork->dst;
1354 struct iphdr *iph;
1355 __be16 df = 0;
1356 __u8 ttl;
1357
1358 skb = __skb_dequeue(queue);
1359 if (!skb)
1360 goto out;
1361 tail_skb = &(skb_shinfo(skb)->frag_list);
1362
1363 /* move skb->data to ip header from ext header */
1364 if (skb->data < skb_network_header(skb))
1365 __skb_pull(skb, skb_network_offset(skb));
1366 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1367 __skb_pull(tmp_skb, skb_network_header_len(skb));
1368 *tail_skb = tmp_skb;
1369 tail_skb = &(tmp_skb->next);
1370 skb->len += tmp_skb->len;
1371 skb->data_len += tmp_skb->len;
1372 skb->truesize += tmp_skb->truesize;
1373 tmp_skb->destructor = NULL;
1374 tmp_skb->sk = NULL;
1375 }
1376
1377 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1378 * to fragment the frame generated here. No matter, what transforms
1379 * how transforms change size of the packet, it will come out.
1380 */
1381 skb->ignore_df = ip_sk_ignore_df(sk);
1382
1383 /* DF bit is set when we want to see DF on outgoing frames.
1384 * If ignore_df is set too, we still allow to fragment this frame
1385 * locally. */
1386 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1387 inet->pmtudisc == IP_PMTUDISC_PROBE ||
1388 (skb->len <= dst_mtu(&rt->dst) &&
1389 ip_dont_fragment(sk, &rt->dst)))
1390 df = htons(IP_DF);
1391
1392 if (cork->flags & IPCORK_OPT)
1393 opt = cork->opt;
1394
1395 if (cork->ttl != 0)
1396 ttl = cork->ttl;
1397 else if (rt->rt_type == RTN_MULTICAST)
1398 ttl = inet->mc_ttl;
1399 else
1400 ttl = ip_select_ttl(inet, &rt->dst);
1401
1402 iph = ip_hdr(skb);
1403 iph->version = 4;
1404 iph->ihl = 5;
1405 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1406 iph->frag_off = df;
1407 iph->ttl = ttl;
1408 iph->protocol = sk->sk_protocol;
1409 ip_copy_addrs(iph, fl4);
1410 ip_select_ident(net, skb, sk);
1411
1412 if (opt) {
1413 iph->ihl += opt->optlen>>2;
1414 ip_options_build(skb, opt, cork->addr, rt, 0);
1415 }
1416
1417 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1418 skb->mark = sk->sk_mark;
1419 /*
1420 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1421 * on dst refcount
1422 */
1423 cork->dst = NULL;
1424 skb_dst_set(skb, &rt->dst);
1425
1426 if (iph->protocol == IPPROTO_ICMP)
1427 icmp_out_count(net, ((struct icmphdr *)
1428 skb_transport_header(skb))->type);
1429
1430 ip_cork_release(cork);
1431 out:
1432 return skb;
1433 }
1434
1435 int ip_send_skb(struct net *net, struct sk_buff *skb)
1436 {
1437 int err;
1438
1439 err = ip_local_out(skb);
1440 if (err) {
1441 if (err > 0)
1442 err = net_xmit_errno(err);
1443 if (err)
1444 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1445 }
1446
1447 return err;
1448 }
1449
1450 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1451 {
1452 struct sk_buff *skb;
1453
1454 skb = ip_finish_skb(sk, fl4);
1455 if (!skb)
1456 return 0;
1457
1458 /* Netfilter gets whole the not fragmented skb. */
1459 return ip_send_skb(sock_net(sk), skb);
1460 }
1461
1462 /*
1463 * Throw away all pending data on the socket.
1464 */
1465 static void __ip_flush_pending_frames(struct sock *sk,
1466 struct sk_buff_head *queue,
1467 struct inet_cork *cork)
1468 {
1469 struct sk_buff *skb;
1470
1471 while ((skb = __skb_dequeue_tail(queue)) != NULL)
1472 kfree_skb(skb);
1473
1474 ip_cork_release(cork);
1475 }
1476
1477 void ip_flush_pending_frames(struct sock *sk)
1478 {
1479 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1480 }
1481
1482 struct sk_buff *ip_make_skb(struct sock *sk,
1483 struct flowi4 *fl4,
1484 int getfrag(void *from, char *to, int offset,
1485 int len, int odd, struct sk_buff *skb),
1486 void *from, int length, int transhdrlen,
1487 struct ipcm_cookie *ipc, struct rtable **rtp,
1488 unsigned int flags)
1489 {
1490 struct inet_cork cork;
1491 struct sk_buff_head queue;
1492 int err;
1493
1494 if (flags & MSG_PROBE)
1495 return NULL;
1496
1497 __skb_queue_head_init(&queue);
1498
1499 cork.flags = 0;
1500 cork.addr = 0;
1501 cork.opt = NULL;
1502 err = ip_setup_cork(sk, &cork, ipc, rtp);
1503 if (err)
1504 return ERR_PTR(err);
1505
1506 err = __ip_append_data(sk, fl4, &queue, &cork,
1507 &current->task_frag, getfrag,
1508 from, length, transhdrlen, flags);
1509 if (err) {
1510 __ip_flush_pending_frames(sk, &queue, &cork);
1511 return ERR_PTR(err);
1512 }
1513
1514 return __ip_make_skb(sk, fl4, &queue, &cork);
1515 }
1516
1517 /*
1518 * Fetch data from kernel space and fill in checksum if needed.
1519 */
1520 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1521 int len, int odd, struct sk_buff *skb)
1522 {
1523 __wsum csum;
1524
1525 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1526 skb->csum = csum_block_add(skb->csum, csum, odd);
1527 return 0;
1528 }
1529
1530 /*
1531 * Generic function to send a packet as reply to another packet.
1532 * Used to send some TCP resets/acks so far.
1533 */
1534 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1535 const struct ip_options *sopt,
1536 __be32 daddr, __be32 saddr,
1537 const struct ip_reply_arg *arg,
1538 unsigned int len)
1539 {
1540 struct ip_options_data replyopts;
1541 struct ipcm_cookie ipc;
1542 struct flowi4 fl4;
1543 struct rtable *rt = skb_rtable(skb);
1544 struct net *net = sock_net(sk);
1545 struct sk_buff *nskb;
1546 int err;
1547 int oif;
1548
1549 if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
1550 return;
1551
1552 ipc.addr = daddr;
1553 ipc.opt = NULL;
1554 ipc.tx_flags = 0;
1555 ipc.ttl = 0;
1556 ipc.tos = -1;
1557
1558 if (replyopts.opt.opt.optlen) {
1559 ipc.opt = &replyopts.opt;
1560
1561 if (replyopts.opt.opt.srr)
1562 daddr = replyopts.opt.opt.faddr;
1563 }
1564
1565 oif = arg->bound_dev_if;
1566 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1567 oif = skb->skb_iif;
1568
1569 flowi4_init_output(&fl4, oif,
1570 IP4_REPLY_MARK(net, skb->mark),
1571 RT_TOS(arg->tos),
1572 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1573 ip_reply_arg_flowi_flags(arg),
1574 daddr, saddr,
1575 tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1576 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1577 rt = ip_route_output_key(net, &fl4);
1578 if (IS_ERR(rt))
1579 return;
1580
1581 inet_sk(sk)->tos = arg->tos;
1582
1583 sk->sk_priority = skb->priority;
1584 sk->sk_protocol = ip_hdr(skb)->protocol;
1585 sk->sk_bound_dev_if = arg->bound_dev_if;
1586 sk->sk_sndbuf = sysctl_wmem_default;
1587 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1588 len, 0, &ipc, &rt, MSG_DONTWAIT);
1589 if (unlikely(err)) {
1590 ip_flush_pending_frames(sk);
1591 goto out;
1592 }
1593
1594 nskb = skb_peek(&sk->sk_write_queue);
1595 if (nskb) {
1596 if (arg->csumoffset >= 0)
1597 *((__sum16 *)skb_transport_header(nskb) +
1598 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1599 arg->csum));
1600 nskb->ip_summed = CHECKSUM_NONE;
1601 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
1602 ip_push_pending_frames(sk, &fl4);
1603 }
1604 out:
1605 ip_rt_put(rt);
1606 }
1607
1608 void __init ip_init(void)
1609 {
1610 ip_rt_init();
1611 inet_initpeers();
1612
1613 #if defined(CONFIG_IP_MULTICAST)
1614 igmp_mc_init();
1615 #endif
1616 }
Linux kernel - Deliverables
This page took 0.062472 seconds and 4 git commands to generate.