projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'v4.4.38' into android-4.4.y
[GitHub/exynos8895/android_kernel_samsung_universal8895.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 net *net, struct sock *sk, struct sk_buff *skb)
100 {
101 struct iphdr *iph = ip_hdr(skb);
102
103 iph->tot_len = htons(skb->len);
104 ip_send_check(iph);
105
106 skb->protocol = htons(ETH_P_IP);
107
108 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
109 net, sk, skb, NULL, skb_dst(skb)->dev,
110 dst_output);
111 }
112
113 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
114 {
115 int err;
116
117 err = __ip_local_out(net, sk, skb);
118 if (likely(err == 1))
119 err = dst_output(net, sk, skb);
120
121 return err;
122 }
123 EXPORT_SYMBOL_GPL(ip_local_out);
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, const 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 net *net = sock_net(sk);
144 struct iphdr *iph;
145
146 /* Build the IP header. */
147 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
148 skb_reset_network_header(skb);
149 iph = ip_hdr(skb);
150 iph->version = 4;
151 iph->ihl = 5;
152 iph->tos = inet->tos;
153 iph->ttl = ip_select_ttl(inet, &rt->dst);
154 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
155 iph->saddr = saddr;
156 iph->protocol = sk->sk_protocol;
157 if (ip_dont_fragment(sk, &rt->dst)) {
158 iph->frag_off = htons(IP_DF);
159 iph->id = 0;
160 } else {
161 iph->frag_off = 0;
162 __ip_select_ident(net, iph, 1);
163 }
164
165 if (opt && opt->opt.optlen) {
166 iph->ihl += opt->opt.optlen>>2;
167 ip_options_build(skb, &opt->opt, daddr, rt, 0);
168 }
169
170 skb->priority = sk->sk_priority;
171 skb->mark = sk->sk_mark;
172
173 /* Send it out. */
174 return ip_local_out(net, skb->sk, skb);
175 }
176 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
177
178 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
179 {
180 struct dst_entry *dst = skb_dst(skb);
181 struct rtable *rt = (struct rtable *)dst;
182 struct net_device *dev = dst->dev;
183 unsigned int hh_len = LL_RESERVED_SPACE(dev);
184 struct neighbour *neigh;
185 u32 nexthop;
186
187 if (rt->rt_type == RTN_MULTICAST) {
188 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
189 } else if (rt->rt_type == RTN_BROADCAST)
190 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
191
192 /* Be paranoid, rather than too clever. */
193 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
194 struct sk_buff *skb2;
195
196 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
197 if (!skb2) {
198 kfree_skb(skb);
199 return -ENOMEM;
200 }
201 if (skb->sk)
202 skb_set_owner_w(skb2, skb->sk);
203 consume_skb(skb);
204 skb = skb2;
205 }
206
207 rcu_read_lock_bh();
208 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
209 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
210 if (unlikely(!neigh))
211 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
212 if (!IS_ERR(neigh)) {
213 int res = dst_neigh_output(dst, neigh, skb);
214
215 rcu_read_unlock_bh();
216 return res;
217 }
218 rcu_read_unlock_bh();
219
220 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
221 __func__);
222 kfree_skb(skb);
223 return -EINVAL;
224 }
225
226 static int ip_finish_output_gso(struct net *net, struct sock *sk,
227 struct sk_buff *skb, unsigned int mtu)
228 {
229 netdev_features_t features;
230 struct sk_buff *segs;
231 int ret = 0;
232
233 /* common case: locally created skb or seglen is <= mtu */
234 if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) ||
235 skb_gso_network_seglen(skb) <= mtu)
236 return ip_finish_output2(net, sk, skb);
237
238 /* Slowpath - GSO segment length is exceeding the dst MTU.
239 *
240 * This can happen in two cases:
241 * 1) TCP GRO packet, DF bit not set
242 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
243 * from host network stack.
244 */
245 features = netif_skb_features(skb);
246 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
247 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
248 if (IS_ERR_OR_NULL(segs)) {
249 kfree_skb(skb);
250 return -ENOMEM;
251 }
252
253 consume_skb(skb);
254
255 do {
256 struct sk_buff *nskb = segs->next;
257 int err;
258
259 segs->next = NULL;
260 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
261
262 if (err && ret == 0)
263 ret = err;
264 segs = nskb;
265 } while (segs);
266
267 return ret;
268 }
269
270 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
271 {
272 unsigned int mtu;
273
274 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
275 /* Policy lookup after SNAT yielded a new policy */
276 if (skb_dst(skb)->xfrm) {
277 IPCB(skb)->flags |= IPSKB_REROUTED;
278 return dst_output(net, sk, skb);
279 }
280 #endif
281 mtu = ip_skb_dst_mtu(skb);
282 if (skb_is_gso(skb))
283 return ip_finish_output_gso(net, sk, skb, mtu);
284
285 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
286 return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
287
288 return ip_finish_output2(net, sk, skb);
289 }
290
291 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
292 {
293 struct rtable *rt = skb_rtable(skb);
294 struct net_device *dev = rt->dst.dev;
295
296 /*
297 * If the indicated interface is up and running, send the packet.
298 */
299 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
300
301 skb->dev = dev;
302 skb->protocol = htons(ETH_P_IP);
303
304 /*
305 * Multicasts are looped back for other local users
306 */
307
308 if (rt->rt_flags&RTCF_MULTICAST) {
309 if (sk_mc_loop(sk)
310 #ifdef CONFIG_IP_MROUTE
311 /* Small optimization: do not loopback not local frames,
312 which returned after forwarding; they will be dropped
313 by ip_mr_input in any case.
314 Note, that local frames are looped back to be delivered
315 to local recipients.
316
317 This check is duplicated in ip_mr_input at the moment.
318 */
319 &&
320 ((rt->rt_flags & RTCF_LOCAL) ||
321 !(IPCB(skb)->flags & IPSKB_FORWARDED))
322 #endif
323 ) {
324 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
325 if (newskb)
326 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
327 net, sk, newskb, NULL, newskb->dev,
328 dev_loopback_xmit);
329 }
330
331 /* Multicasts with ttl 0 must not go beyond the host */
332
333 if (ip_hdr(skb)->ttl == 0) {
334 kfree_skb(skb);
335 return 0;
336 }
337 }
338
339 if (rt->rt_flags&RTCF_BROADCAST) {
340 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
341 if (newskb)
342 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
343 net, sk, newskb, NULL, newskb->dev,
344 dev_loopback_xmit);
345 }
346
347 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
348 net, sk, skb, NULL, skb->dev,
349 ip_finish_output,
350 !(IPCB(skb)->flags & IPSKB_REROUTED));
351 }
352
353 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
354 {
355 struct net_device *dev = skb_dst(skb)->dev;
356
357 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
358
359 skb->dev = dev;
360 skb->protocol = htons(ETH_P_IP);
361
362 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
363 net, sk, skb, NULL, dev,
364 ip_finish_output,
365 !(IPCB(skb)->flags & IPSKB_REROUTED));
366 }
367
368 /*
369 * copy saddr and daddr, possibly using 64bit load/stores
370 * Equivalent to :
371 * iph->saddr = fl4->saddr;
372 * iph->daddr = fl4->daddr;
373 */
374 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
375 {
376 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
377 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
378 memcpy(&iph->saddr, &fl4->saddr,
379 sizeof(fl4->saddr) + sizeof(fl4->daddr));
380 }
381
382 /* Note: skb->sk can be different from sk, in case of tunnels */
383 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
384 {
385 struct inet_sock *inet = inet_sk(sk);
386 struct net *net = sock_net(sk);
387 struct ip_options_rcu *inet_opt;
388 struct flowi4 *fl4;
389 struct rtable *rt;
390 struct iphdr *iph;
391 int res;
392
393 /* Skip all of this if the packet is already routed,
394 * f.e. by something like SCTP.
395 */
396 rcu_read_lock();
397 inet_opt = rcu_dereference(inet->inet_opt);
398 fl4 = &fl->u.ip4;
399 rt = skb_rtable(skb);
400 if (rt)
401 goto packet_routed;
402
403 /* Make sure we can route this packet. */
404 rt = (struct rtable *)__sk_dst_check(sk, 0);
405 if (!rt) {
406 __be32 daddr;
407
408 /* Use correct destination address if we have options. */
409 daddr = inet->inet_daddr;
410 if (inet_opt && inet_opt->opt.srr)
411 daddr = inet_opt->opt.faddr;
412
413 /* If this fails, retransmit mechanism of transport layer will
414 * keep trying until route appears or the connection times
415 * itself out.
416 */
417 rt = ip_route_output_ports(net, fl4, sk,
418 daddr, inet->inet_saddr,
419 inet->inet_dport,
420 inet->inet_sport,
421 sk->sk_protocol,
422 RT_CONN_FLAGS(sk),
423 sk->sk_bound_dev_if);
424 if (IS_ERR(rt))
425 goto no_route;
426 sk_setup_caps(sk, &rt->dst);
427 }
428 skb_dst_set_noref(skb, &rt->dst);
429
430 packet_routed:
431 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
432 goto no_route;
433
434 /* OK, we know where to send it, allocate and build IP header. */
435 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
436 skb_reset_network_header(skb);
437 iph = ip_hdr(skb);
438 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
439 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
440 iph->frag_off = htons(IP_DF);
441 else
442 iph->frag_off = 0;
443 iph->ttl = ip_select_ttl(inet, &rt->dst);
444 iph->protocol = sk->sk_protocol;
445 ip_copy_addrs(iph, fl4);
446
447 /* Transport layer set skb->h.foo itself. */
448
449 if (inet_opt && inet_opt->opt.optlen) {
450 iph->ihl += inet_opt->opt.optlen >> 2;
451 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
452 }
453
454 ip_select_ident_segs(net, skb, sk,
455 skb_shinfo(skb)->gso_segs ?: 1);
456
457 /* TODO : should we use skb->sk here instead of sk ? */
458 skb->priority = sk->sk_priority;
459 skb->mark = sk->sk_mark;
460
461 res = ip_local_out(net, sk, skb);
462 rcu_read_unlock();
463 return res;
464
465 no_route:
466 rcu_read_unlock();
467 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
468 kfree_skb(skb);
469 return -EHOSTUNREACH;
470 }
471 EXPORT_SYMBOL(ip_queue_xmit);
472
473 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
474 {
475 to->pkt_type = from->pkt_type;
476 to->priority = from->priority;
477 to->protocol = from->protocol;
478 skb_dst_drop(to);
479 skb_dst_copy(to, from);
480 to->dev = from->dev;
481 to->mark = from->mark;
482
483 /* Copy the flags to each fragment. */
484 IPCB(to)->flags = IPCB(from)->flags;
485
486 #ifdef CONFIG_NET_SCHED
487 to->tc_index = from->tc_index;
488 #endif
489 nf_copy(to, from);
490 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
491 to->ipvs_property = from->ipvs_property;
492 #endif
493 skb_copy_secmark(to, from);
494 }
495
496 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
497 unsigned int mtu,
498 int (*output)(struct net *, struct sock *, struct sk_buff *))
499 {
500 struct iphdr *iph = ip_hdr(skb);
501
502 if ((iph->frag_off & htons(IP_DF)) == 0)
503 return ip_do_fragment(net, sk, skb, output);
504
505 if (unlikely(!skb->ignore_df ||
506 (IPCB(skb)->frag_max_size &&
507 IPCB(skb)->frag_max_size > mtu))) {
508 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
509 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
510 htonl(mtu));
511 kfree_skb(skb);
512 return -EMSGSIZE;
513 }
514
515 return ip_do_fragment(net, sk, skb, output);
516 }
517
518 /*
519 * This IP datagram is too large to be sent in one piece. Break it up into
520 * smaller pieces (each of size equal to IP header plus
521 * a block of the data of the original IP data part) that will yet fit in a
522 * single device frame, and queue such a frame for sending.
523 */
524
525 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
526 int (*output)(struct net *, struct sock *, struct sk_buff *))
527 {
528 struct iphdr *iph;
529 int ptr;
530 struct net_device *dev;
531 struct sk_buff *skb2;
532 unsigned int mtu, hlen, left, len, ll_rs;
533 int offset;
534 __be16 not_last_frag;
535 struct rtable *rt = skb_rtable(skb);
536 int err = 0;
537
538 dev = rt->dst.dev;
539
540 /* for offloaded checksums cleanup checksum before fragmentation */
541 if (skb->ip_summed == CHECKSUM_PARTIAL &&
542 (err = skb_checksum_help(skb)))
543 goto fail;
544
545 /*
546 * Point into the IP datagram header.
547 */
548
549 iph = ip_hdr(skb);
550
551 mtu = ip_skb_dst_mtu(skb);
552 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
553 mtu = IPCB(skb)->frag_max_size;
554
555 /*
556 * Setup starting values.
557 */
558
559 hlen = iph->ihl * 4;
560 mtu = mtu - hlen; /* Size of data space */
561 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
562
563 /* When frag_list is given, use it. First, check its validity:
564 * some transformers could create wrong frag_list or break existing
565 * one, it is not prohibited. In this case fall back to copying.
566 *
567 * LATER: this step can be merged to real generation of fragments,
568 * we can switch to copy when see the first bad fragment.
569 */
570 if (skb_has_frag_list(skb)) {
571 struct sk_buff *frag, *frag2;
572 int first_len = skb_pagelen(skb);
573
574 if (first_len - hlen > mtu ||
575 ((first_len - hlen) & 7) ||
576 ip_is_fragment(iph) ||
577 skb_cloned(skb))
578 goto slow_path;
579
580 skb_walk_frags(skb, frag) {
581 /* Correct geometry. */
582 if (frag->len > mtu ||
583 ((frag->len & 7) && frag->next) ||
584 skb_headroom(frag) < hlen)
585 goto slow_path_clean;
586
587 /* Partially cloned skb? */
588 if (skb_shared(frag))
589 goto slow_path_clean;
590
591 BUG_ON(frag->sk);
592 if (skb->sk) {
593 frag->sk = skb->sk;
594 frag->destructor = sock_wfree;
595 }
596 skb->truesize -= frag->truesize;
597 }
598
599 /* Everything is OK. Generate! */
600
601 err = 0;
602 offset = 0;
603 frag = skb_shinfo(skb)->frag_list;
604 skb_frag_list_init(skb);
605 skb->data_len = first_len - skb_headlen(skb);
606 skb->len = first_len;
607 iph->tot_len = htons(first_len);
608 iph->frag_off = htons(IP_MF);
609 ip_send_check(iph);
610
611 for (;;) {
612 /* Prepare header of the next frame,
613 * before previous one went down. */
614 if (frag) {
615 frag->ip_summed = CHECKSUM_NONE;
616 skb_reset_transport_header(frag);
617 __skb_push(frag, hlen);
618 skb_reset_network_header(frag);
619 memcpy(skb_network_header(frag), iph, hlen);
620 iph = ip_hdr(frag);
621 iph->tot_len = htons(frag->len);
622 ip_copy_metadata(frag, skb);
623 if (offset == 0)
624 ip_options_fragment(frag);
625 offset += skb->len - hlen;
626 iph->frag_off = htons(offset>>3);
627 if (frag->next)
628 iph->frag_off |= htons(IP_MF);
629 /* Ready, complete checksum */
630 ip_send_check(iph);
631 }
632
633 err = output(net, sk, skb);
634
635 if (!err)
636 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
637 if (err || !frag)
638 break;
639
640 skb = frag;
641 frag = skb->next;
642 skb->next = NULL;
643 }
644
645 if (err == 0) {
646 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
647 return 0;
648 }
649
650 while (frag) {
651 skb = frag->next;
652 kfree_skb(frag);
653 frag = skb;
654 }
655 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
656 return err;
657
658 slow_path_clean:
659 skb_walk_frags(skb, frag2) {
660 if (frag2 == frag)
661 break;
662 frag2->sk = NULL;
663 frag2->destructor = NULL;
664 skb->truesize += frag2->truesize;
665 }
666 }
667
668 slow_path:
669 iph = ip_hdr(skb);
670
671 left = skb->len - hlen; /* Space per frame */
672 ptr = hlen; /* Where to start from */
673
674 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
675
676 /*
677 * Fragment the datagram.
678 */
679
680 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
681 not_last_frag = iph->frag_off & htons(IP_MF);
682
683 /*
684 * Keep copying data until we run out.
685 */
686
687 while (left > 0) {
688 len = left;
689 /* IF: it doesn't fit, use 'mtu' - the data space left */
690 if (len > mtu)
691 len = mtu;
692 /* IF: we are not sending up to and including the packet end
693 then align the next start on an eight byte boundary */
694 if (len < left) {
695 len &= ~7;
696 }
697
698 /* Allocate buffer */
699 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
700 if (!skb2) {
701 err = -ENOMEM;
702 goto fail;
703 }
704
705 /*
706 * Set up data on packet
707 */
708
709 ip_copy_metadata(skb2, skb);
710 skb_reserve(skb2, ll_rs);
711 skb_put(skb2, len + hlen);
712 skb_reset_network_header(skb2);
713 skb2->transport_header = skb2->network_header + hlen;
714
715 /*
716 * Charge the memory for the fragment to any owner
717 * it might possess
718 */
719
720 if (skb->sk)
721 skb_set_owner_w(skb2, skb->sk);
722
723 /*
724 * Copy the packet header into the new buffer.
725 */
726
727 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
728
729 /*
730 * Copy a block of the IP datagram.
731 */
732 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
733 BUG();
734 left -= len;
735
736 /*
737 * Fill in the new header fields.
738 */
739 iph = ip_hdr(skb2);
740 iph->frag_off = htons((offset >> 3));
741
742 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
743 iph->frag_off |= htons(IP_DF);
744
745 /* ANK: dirty, but effective trick. Upgrade options only if
746 * the segment to be fragmented was THE FIRST (otherwise,
747 * options are already fixed) and make it ONCE
748 * on the initial skb, so that all the following fragments
749 * will inherit fixed options.
750 */
751 if (offset == 0)
752 ip_options_fragment(skb);
753
754 /*
755 * Added AC : If we are fragmenting a fragment that's not the
756 * last fragment then keep MF on each bit
757 */
758 if (left > 0 || not_last_frag)
759 iph->frag_off |= htons(IP_MF);
760 ptr += len;
761 offset += len;
762
763 /*
764 * Put this fragment into the sending queue.
765 */
766 iph->tot_len = htons(len + hlen);
767
768 ip_send_check(iph);
769
770 err = output(net, sk, skb2);
771 if (err)
772 goto fail;
773
774 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
775 }
776 consume_skb(skb);
777 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
778 return err;
779
780 fail:
781 kfree_skb(skb);
782 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
783 return err;
784 }
785 EXPORT_SYMBOL(ip_do_fragment);
786
787 int
788 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
789 {
790 struct msghdr *msg = from;
791
792 if (skb->ip_summed == CHECKSUM_PARTIAL) {
793 if (copy_from_iter(to, len, &msg->msg_iter) != len)
794 return -EFAULT;
795 } else {
796 __wsum csum = 0;
797 if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len)
798 return -EFAULT;
799 skb->csum = csum_block_add(skb->csum, csum, odd);
800 }
801 return 0;
802 }
803 EXPORT_SYMBOL(ip_generic_getfrag);
804
805 static inline __wsum
806 csum_page(struct page *page, int offset, int copy)
807 {
808 char *kaddr;
809 __wsum csum;
810 kaddr = kmap(page);
811 csum = csum_partial(kaddr + offset, copy, 0);
812 kunmap(page);
813 return csum;
814 }
815
816 static inline int ip_ufo_append_data(struct sock *sk,
817 struct sk_buff_head *queue,
818 int getfrag(void *from, char *to, int offset, int len,
819 int odd, struct sk_buff *skb),
820 void *from, int length, int hh_len, int fragheaderlen,
821 int transhdrlen, int maxfraglen, unsigned int flags)
822 {
823 struct sk_buff *skb;
824 int err;
825
826 /* There is support for UDP fragmentation offload by network
827 * device, so create one single skb packet containing complete
828 * udp datagram
829 */
830 skb = skb_peek_tail(queue);
831 if (!skb) {
832 skb = sock_alloc_send_skb(sk,
833 hh_len + fragheaderlen + transhdrlen + 20,
834 (flags & MSG_DONTWAIT), &err);
835
836 if (!skb)
837 return err;
838
839 /* reserve space for Hardware header */
840 skb_reserve(skb, hh_len);
841
842 /* create space for UDP/IP header */
843 skb_put(skb, fragheaderlen + transhdrlen);
844
845 /* initialize network header pointer */
846 skb_reset_network_header(skb);
847
848 /* initialize protocol header pointer */
849 skb->transport_header = skb->network_header + fragheaderlen;
850
851 skb->csum = 0;
852
853 __skb_queue_tail(queue, skb);
854 } else if (skb_is_gso(skb)) {
855 goto append;
856 }
857
858 skb->ip_summed = CHECKSUM_PARTIAL;
859 /* specify the length of each IP datagram fragment */
860 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
861 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
862
863 append:
864 return skb_append_datato_frags(sk, skb, getfrag, from,
865 (length - transhdrlen));
866 }
867
868 static int __ip_append_data(struct sock *sk,
869 struct flowi4 *fl4,
870 struct sk_buff_head *queue,
871 struct inet_cork *cork,
872 struct page_frag *pfrag,
873 int getfrag(void *from, char *to, int offset,
874 int len, int odd, struct sk_buff *skb),
875 void *from, int length, int transhdrlen,
876 unsigned int flags)
877 {
878 struct inet_sock *inet = inet_sk(sk);
879 struct sk_buff *skb;
880
881 struct ip_options *opt = cork->opt;
882 int hh_len;
883 int exthdrlen;
884 int mtu;
885 int copy;
886 int err;
887 int offset = 0;
888 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
889 int csummode = CHECKSUM_NONE;
890 struct rtable *rt = (struct rtable *)cork->dst;
891 u32 tskey = 0;
892
893 skb = skb_peek_tail(queue);
894
895 exthdrlen = !skb ? rt->dst.header_len : 0;
896 mtu = cork->fragsize;
897 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
898 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
899 tskey = sk->sk_tskey++;
900
901 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
902
903 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
904 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
905 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
906
907 if (cork->length + length > maxnonfragsize - fragheaderlen) {
908 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
909 mtu - (opt ? opt->optlen : 0));
910 return -EMSGSIZE;
911 }
912
913 /*
914 * transhdrlen > 0 means that this is the first fragment and we wish
915 * it won't be fragmented in the future.
916 */
917 if (transhdrlen &&
918 length + fragheaderlen <= mtu &&
919 rt->dst.dev->features & NETIF_F_V4_CSUM &&
920 !(flags & MSG_MORE) &&
921 !exthdrlen)
922 csummode = CHECKSUM_PARTIAL;
923
924 cork->length += length;
925 if (((length > mtu) || (skb && skb_is_gso(skb))) &&
926 (sk->sk_protocol == IPPROTO_UDP) &&
927 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len &&
928 (sk->sk_type == SOCK_DGRAM) && !sk->sk_no_check_tx) {
929 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
930 hh_len, fragheaderlen, transhdrlen,
931 maxfraglen, flags);
932 if (err)
933 goto error;
934 return 0;
935 }
936
937 /* So, what's going on in the loop below?
938 *
939 * We use calculated fragment length to generate chained skb,
940 * each of segments is IP fragment ready for sending to network after
941 * adding appropriate IP header.
942 */
943
944 if (!skb)
945 goto alloc_new_skb;
946
947 while (length > 0) {
948 /* Check if the remaining data fits into current packet. */
949 copy = mtu - skb->len;
950 if (copy < length)
951 copy = maxfraglen - skb->len;
952 if (copy <= 0) {
953 char *data;
954 unsigned int datalen;
955 unsigned int fraglen;
956 unsigned int fraggap;
957 unsigned int alloclen;
958 struct sk_buff *skb_prev;
959 alloc_new_skb:
960 skb_prev = skb;
961 if (skb_prev)
962 fraggap = skb_prev->len - maxfraglen;
963 else
964 fraggap = 0;
965
966 /*
967 * If remaining data exceeds the mtu,
968 * we know we need more fragment(s).
969 */
970 datalen = length + fraggap;
971 if (datalen > mtu - fragheaderlen)
972 datalen = maxfraglen - fragheaderlen;
973 fraglen = datalen + fragheaderlen;
974
975 if ((flags & MSG_MORE) &&
976 !(rt->dst.dev->features&NETIF_F_SG))
977 alloclen = mtu;
978 else
979 alloclen = fraglen;
980
981 alloclen += exthdrlen;
982
983 /* The last fragment gets additional space at tail.
984 * Note, with MSG_MORE we overallocate on fragments,
985 * because we have no idea what fragment will be
986 * the last.
987 */
988 if (datalen == length + fraggap)
989 alloclen += rt->dst.trailer_len;
990
991 if (transhdrlen) {
992 skb = sock_alloc_send_skb(sk,
993 alloclen + hh_len + 15,
994 (flags & MSG_DONTWAIT), &err);
995 } else {
996 skb = NULL;
997 if (atomic_read(&sk->sk_wmem_alloc) <=
998 2 * sk->sk_sndbuf)
999 skb = sock_wmalloc(sk,
1000 alloclen + hh_len + 15, 1,
1001 sk->sk_allocation);
1002 if (unlikely(!skb))
1003 err = -ENOBUFS;
1004 }
1005 if (!skb)
1006 goto error;
1007
1008 /*
1009 * Fill in the control structures
1010 */
1011 skb->ip_summed = csummode;
1012 skb->csum = 0;
1013 skb_reserve(skb, hh_len);
1014
1015 /* only the initial fragment is time stamped */
1016 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1017 cork->tx_flags = 0;
1018 skb_shinfo(skb)->tskey = tskey;
1019 tskey = 0;
1020
1021 /*
1022 * Find where to start putting bytes.
1023 */
1024 data = skb_put(skb, fraglen + exthdrlen);
1025 skb_set_network_header(skb, exthdrlen);
1026 skb->transport_header = (skb->network_header +
1027 fragheaderlen);
1028 data += fragheaderlen + exthdrlen;
1029
1030 if (fraggap) {
1031 skb->csum = skb_copy_and_csum_bits(
1032 skb_prev, maxfraglen,
1033 data + transhdrlen, fraggap, 0);
1034 skb_prev->csum = csum_sub(skb_prev->csum,
1035 skb->csum);
1036 data += fraggap;
1037 pskb_trim_unique(skb_prev, maxfraglen);
1038 }
1039
1040 copy = datalen - transhdrlen - fraggap;
1041 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1042 err = -EFAULT;
1043 kfree_skb(skb);
1044 goto error;
1045 }
1046
1047 offset += copy;
1048 length -= datalen - fraggap;
1049 transhdrlen = 0;
1050 exthdrlen = 0;
1051 csummode = CHECKSUM_NONE;
1052
1053 /*
1054 * Put the packet on the pending queue.
1055 */
1056 __skb_queue_tail(queue, skb);
1057 continue;
1058 }
1059
1060 if (copy > length)
1061 copy = length;
1062
1063 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1064 unsigned int off;
1065
1066 off = skb->len;
1067 if (getfrag(from, skb_put(skb, copy),
1068 offset, copy, off, skb) < 0) {
1069 __skb_trim(skb, off);
1070 err = -EFAULT;
1071 goto error;
1072 }
1073 } else {
1074 int i = skb_shinfo(skb)->nr_frags;
1075
1076 err = -ENOMEM;
1077 if (!sk_page_frag_refill(sk, pfrag))
1078 goto error;
1079
1080 if (!skb_can_coalesce(skb, i, pfrag->page,
1081 pfrag->offset)) {
1082 err = -EMSGSIZE;
1083 if (i == MAX_SKB_FRAGS)
1084 goto error;
1085
1086 __skb_fill_page_desc(skb, i, pfrag->page,
1087 pfrag->offset, 0);
1088 skb_shinfo(skb)->nr_frags = ++i;
1089 get_page(pfrag->page);
1090 }
1091 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1092 if (getfrag(from,
1093 page_address(pfrag->page) + pfrag->offset,
1094 offset, copy, skb->len, skb) < 0)
1095 goto error_efault;
1096
1097 pfrag->offset += copy;
1098 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1099 skb->len += copy;
1100 skb->data_len += copy;
1101 skb->truesize += copy;
1102 atomic_add(copy, &sk->sk_wmem_alloc);
1103 }
1104 offset += copy;
1105 length -= copy;
1106 }
1107
1108 return 0;
1109
1110 error_efault:
1111 err = -EFAULT;
1112 error:
1113 cork->length -= length;
1114 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1115 return err;
1116 }
1117
1118 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1119 struct ipcm_cookie *ipc, struct rtable **rtp)
1120 {
1121 struct ip_options_rcu *opt;
1122 struct rtable *rt;
1123
1124 /*
1125 * setup for corking.
1126 */
1127 opt = ipc->opt;
1128 if (opt) {
1129 if (!cork->opt) {
1130 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1131 sk->sk_allocation);
1132 if (unlikely(!cork->opt))
1133 return -ENOBUFS;
1134 }
1135 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1136 cork->flags |= IPCORK_OPT;
1137 cork->addr = ipc->addr;
1138 }
1139 rt = *rtp;
1140 if (unlikely(!rt))
1141 return -EFAULT;
1142 /*
1143 * We steal reference to this route, caller should not release it
1144 */
1145 *rtp = NULL;
1146 cork->fragsize = ip_sk_use_pmtu(sk) ?
1147 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
1148 cork->dst = &rt->dst;
1149 cork->length = 0;
1150 cork->ttl = ipc->ttl;
1151 cork->tos = ipc->tos;
1152 cork->priority = ipc->priority;
1153 cork->tx_flags = ipc->tx_flags;
1154
1155 return 0;
1156 }
1157
1158 /*
1159 * ip_append_data() and ip_append_page() can make one large IP datagram
1160 * from many pieces of data. Each pieces will be holded on the socket
1161 * until ip_push_pending_frames() is called. Each piece can be a page
1162 * or non-page data.
1163 *
1164 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1165 * this interface potentially.
1166 *
1167 * LATER: length must be adjusted by pad at tail, when it is required.
1168 */
1169 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1170 int getfrag(void *from, char *to, int offset, int len,
1171 int odd, struct sk_buff *skb),
1172 void *from, int length, int transhdrlen,
1173 struct ipcm_cookie *ipc, struct rtable **rtp,
1174 unsigned int flags)
1175 {
1176 struct inet_sock *inet = inet_sk(sk);
1177 int err;
1178
1179 if (flags&MSG_PROBE)
1180 return 0;
1181
1182 if (skb_queue_empty(&sk->sk_write_queue)) {
1183 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1184 if (err)
1185 return err;
1186 } else {
1187 transhdrlen = 0;
1188 }
1189
1190 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1191 sk_page_frag(sk), getfrag,
1192 from, length, transhdrlen, flags);
1193 }
1194
1195 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1196 int offset, size_t size, int flags)
1197 {
1198 struct inet_sock *inet = inet_sk(sk);
1199 struct sk_buff *skb;
1200 struct rtable *rt;
1201 struct ip_options *opt = NULL;
1202 struct inet_cork *cork;
1203 int hh_len;
1204 int mtu;
1205 int len;
1206 int err;
1207 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1208
1209 if (inet->hdrincl)
1210 return -EPERM;
1211
1212 if (flags&MSG_PROBE)
1213 return 0;
1214
1215 if (skb_queue_empty(&sk->sk_write_queue))
1216 return -EINVAL;
1217
1218 cork = &inet->cork.base;
1219 rt = (struct rtable *)cork->dst;
1220 if (cork->flags & IPCORK_OPT)
1221 opt = cork->opt;
1222
1223 if (!(rt->dst.dev->features&NETIF_F_SG))
1224 return -EOPNOTSUPP;
1225
1226 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1227 mtu = cork->fragsize;
1228
1229 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1230 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1231 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1232
1233 if (cork->length + size > maxnonfragsize - fragheaderlen) {
1234 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1235 mtu - (opt ? opt->optlen : 0));
1236 return -EMSGSIZE;
1237 }
1238
1239 skb = skb_peek_tail(&sk->sk_write_queue);
1240 if (!skb)
1241 return -EINVAL;
1242
1243 if ((size + skb->len > mtu) &&
1244 (sk->sk_protocol == IPPROTO_UDP) &&
1245 (rt->dst.dev->features & NETIF_F_UFO)) {
1246 if (skb->ip_summed != CHECKSUM_PARTIAL)
1247 return -EOPNOTSUPP;
1248
1249 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1250 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1251 }
1252 cork->length += size;
1253
1254 while (size > 0) {
1255 if (skb_is_gso(skb)) {
1256 len = size;
1257 } else {
1258
1259 /* Check if the remaining data fits into current packet. */
1260 len = mtu - skb->len;
1261 if (len < size)
1262 len = maxfraglen - skb->len;
1263 }
1264 if (len <= 0) {
1265 struct sk_buff *skb_prev;
1266 int alloclen;
1267
1268 skb_prev = skb;
1269 fraggap = skb_prev->len - maxfraglen;
1270
1271 alloclen = fragheaderlen + hh_len + fraggap + 15;
1272 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1273 if (unlikely(!skb)) {
1274 err = -ENOBUFS;
1275 goto error;
1276 }
1277
1278 /*
1279 * Fill in the control structures
1280 */
1281 skb->ip_summed = CHECKSUM_NONE;
1282 skb->csum = 0;
1283 skb_reserve(skb, hh_len);
1284
1285 /*
1286 * Find where to start putting bytes.
1287 */
1288 skb_put(skb, fragheaderlen + fraggap);
1289 skb_reset_network_header(skb);
1290 skb->transport_header = (skb->network_header +
1291 fragheaderlen);
1292 if (fraggap) {
1293 skb->csum = skb_copy_and_csum_bits(skb_prev,
1294 maxfraglen,
1295 skb_transport_header(skb),
1296 fraggap, 0);
1297 skb_prev->csum = csum_sub(skb_prev->csum,
1298 skb->csum);
1299 pskb_trim_unique(skb_prev, maxfraglen);
1300 }
1301
1302 /*
1303 * Put the packet on the pending queue.
1304 */
1305 __skb_queue_tail(&sk->sk_write_queue, skb);
1306 continue;
1307 }
1308
1309 if (len > size)
1310 len = size;
1311
1312 if (skb_append_pagefrags(skb, page, offset, len)) {
1313 err = -EMSGSIZE;
1314 goto error;
1315 }
1316
1317 if (skb->ip_summed == CHECKSUM_NONE) {
1318 __wsum csum;
1319 csum = csum_page(page, offset, len);
1320 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1321 }
1322
1323 skb->len += len;
1324 skb->data_len += len;
1325 skb->truesize += len;
1326 atomic_add(len, &sk->sk_wmem_alloc);
1327 offset += len;
1328 size -= len;
1329 }
1330 return 0;
1331
1332 error:
1333 cork->length -= size;
1334 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1335 return err;
1336 }
1337
1338 static void ip_cork_release(struct inet_cork *cork)
1339 {
1340 cork->flags &= ~IPCORK_OPT;
1341 kfree(cork->opt);
1342 cork->opt = NULL;
1343 dst_release(cork->dst);
1344 cork->dst = NULL;
1345 }
1346
1347 /*
1348 * Combined all pending IP fragments on the socket as one IP datagram
1349 * and push them out.
1350 */
1351 struct sk_buff *__ip_make_skb(struct sock *sk,
1352 struct flowi4 *fl4,
1353 struct sk_buff_head *queue,
1354 struct inet_cork *cork)
1355 {
1356 struct sk_buff *skb, *tmp_skb;
1357 struct sk_buff **tail_skb;
1358 struct inet_sock *inet = inet_sk(sk);
1359 struct net *net = sock_net(sk);
1360 struct ip_options *opt = NULL;
1361 struct rtable *rt = (struct rtable *)cork->dst;
1362 struct iphdr *iph;
1363 __be16 df = 0;
1364 __u8 ttl;
1365
1366 skb = __skb_dequeue(queue);
1367 if (!skb)
1368 goto out;
1369 tail_skb = &(skb_shinfo(skb)->frag_list);
1370
1371 /* move skb->data to ip header from ext header */
1372 if (skb->data < skb_network_header(skb))
1373 __skb_pull(skb, skb_network_offset(skb));
1374 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1375 __skb_pull(tmp_skb, skb_network_header_len(skb));
1376 *tail_skb = tmp_skb;
1377 tail_skb = &(tmp_skb->next);
1378 skb->len += tmp_skb->len;
1379 skb->data_len += tmp_skb->len;
1380 skb->truesize += tmp_skb->truesize;
1381 tmp_skb->destructor = NULL;
1382 tmp_skb->sk = NULL;
1383 }
1384
1385 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1386 * to fragment the frame generated here. No matter, what transforms
1387 * how transforms change size of the packet, it will come out.
1388 */
1389 skb->ignore_df = ip_sk_ignore_df(sk);
1390
1391 /* DF bit is set when we want to see DF on outgoing frames.
1392 * If ignore_df is set too, we still allow to fragment this frame
1393 * locally. */
1394 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1395 inet->pmtudisc == IP_PMTUDISC_PROBE ||
1396 (skb->len <= dst_mtu(&rt->dst) &&
1397 ip_dont_fragment(sk, &rt->dst)))
1398 df = htons(IP_DF);
1399
1400 if (cork->flags & IPCORK_OPT)
1401 opt = cork->opt;
1402
1403 if (cork->ttl != 0)
1404 ttl = cork->ttl;
1405 else if (rt->rt_type == RTN_MULTICAST)
1406 ttl = inet->mc_ttl;
1407 else
1408 ttl = ip_select_ttl(inet, &rt->dst);
1409
1410 iph = ip_hdr(skb);
1411 iph->version = 4;
1412 iph->ihl = 5;
1413 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1414 iph->frag_off = df;
1415 iph->ttl = ttl;
1416 iph->protocol = sk->sk_protocol;
1417 ip_copy_addrs(iph, fl4);
1418 ip_select_ident(net, skb, sk);
1419
1420 if (opt) {
1421 iph->ihl += opt->optlen>>2;
1422 ip_options_build(skb, opt, cork->addr, rt, 0);
1423 }
1424
1425 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1426 skb->mark = sk->sk_mark;
1427 /*
1428 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1429 * on dst refcount
1430 */
1431 cork->dst = NULL;
1432 skb_dst_set(skb, &rt->dst);
1433
1434 if (iph->protocol == IPPROTO_ICMP)
1435 icmp_out_count(net, ((struct icmphdr *)
1436 skb_transport_header(skb))->type);
1437
1438 ip_cork_release(cork);
1439 out:
1440 return skb;
1441 }
1442
1443 int ip_send_skb(struct net *net, struct sk_buff *skb)
1444 {
1445 int err;
1446
1447 err = ip_local_out(net, skb->sk, skb);
1448 if (err) {
1449 if (err > 0)
1450 err = net_xmit_errno(err);
1451 if (err)
1452 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1453 }
1454
1455 return err;
1456 }
1457
1458 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1459 {
1460 struct sk_buff *skb;
1461
1462 skb = ip_finish_skb(sk, fl4);
1463 if (!skb)
1464 return 0;
1465
1466 /* Netfilter gets whole the not fragmented skb. */
1467 return ip_send_skb(sock_net(sk), skb);
1468 }
1469
1470 /*
1471 * Throw away all pending data on the socket.
1472 */
1473 static void __ip_flush_pending_frames(struct sock *sk,
1474 struct sk_buff_head *queue,
1475 struct inet_cork *cork)
1476 {
1477 struct sk_buff *skb;
1478
1479 while ((skb = __skb_dequeue_tail(queue)) != NULL)
1480 kfree_skb(skb);
1481
1482 ip_cork_release(cork);
1483 }
1484
1485 void ip_flush_pending_frames(struct sock *sk)
1486 {
1487 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1488 }
1489
1490 struct sk_buff *ip_make_skb(struct sock *sk,
1491 struct flowi4 *fl4,
1492 int getfrag(void *from, char *to, int offset,
1493 int len, int odd, struct sk_buff *skb),
1494 void *from, int length, int transhdrlen,
1495 struct ipcm_cookie *ipc, struct rtable **rtp,
1496 unsigned int flags)
1497 {
1498 struct inet_cork cork;
1499 struct sk_buff_head queue;
1500 int err;
1501
1502 if (flags & MSG_PROBE)
1503 return NULL;
1504
1505 __skb_queue_head_init(&queue);
1506
1507 cork.flags = 0;
1508 cork.addr = 0;
1509 cork.opt = NULL;
1510 err = ip_setup_cork(sk, &cork, ipc, rtp);
1511 if (err)
1512 return ERR_PTR(err);
1513
1514 err = __ip_append_data(sk, fl4, &queue, &cork,
1515 &current->task_frag, getfrag,
1516 from, length, transhdrlen, flags);
1517 if (err) {
1518 __ip_flush_pending_frames(sk, &queue, &cork);
1519 return ERR_PTR(err);
1520 }
1521
1522 return __ip_make_skb(sk, fl4, &queue, &cork);
1523 }
1524
1525 /*
1526 * Fetch data from kernel space and fill in checksum if needed.
1527 */
1528 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1529 int len, int odd, struct sk_buff *skb)
1530 {
1531 __wsum csum;
1532
1533 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1534 skb->csum = csum_block_add(skb->csum, csum, odd);
1535 return 0;
1536 }
1537
1538 /*
1539 * Generic function to send a packet as reply to another packet.
1540 * Used to send some TCP resets/acks so far.
1541 */
1542 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1543 const struct ip_options *sopt,
1544 __be32 daddr, __be32 saddr,
1545 const struct ip_reply_arg *arg,
1546 unsigned int len)
1547 {
1548 struct ip_options_data replyopts;
1549 struct ipcm_cookie ipc;
1550 struct flowi4 fl4;
1551 struct rtable *rt = skb_rtable(skb);
1552 struct net *net = sock_net(sk);
1553 struct sk_buff *nskb;
1554 int err;
1555 int oif;
1556
1557 if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
1558 return;
1559
1560 ipc.addr = daddr;
1561 ipc.opt = NULL;
1562 ipc.tx_flags = 0;
1563 ipc.ttl = 0;
1564 ipc.tos = -1;
1565
1566 if (replyopts.opt.opt.optlen) {
1567 ipc.opt = &replyopts.opt;
1568
1569 if (replyopts.opt.opt.srr)
1570 daddr = replyopts.opt.opt.faddr;
1571 }
1572
1573 oif = arg->bound_dev_if;
1574 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1575 oif = skb->skb_iif;
1576
1577 flowi4_init_output(&fl4, oif,
1578 IP4_REPLY_MARK(net, skb->mark),
1579 RT_TOS(arg->tos),
1580 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1581 ip_reply_arg_flowi_flags(arg),
1582 daddr, saddr,
1583 tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1584 arg->uid);
1585 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1586 rt = ip_route_output_key(net, &fl4);
1587 if (IS_ERR(rt))
1588 return;
1589
1590 inet_sk(sk)->tos = arg->tos;
1591
1592 sk->sk_priority = skb->priority;
1593 sk->sk_protocol = ip_hdr(skb)->protocol;
1594 sk->sk_bound_dev_if = arg->bound_dev_if;
1595 sk->sk_sndbuf = sysctl_wmem_default;
1596 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1597 len, 0, &ipc, &rt, MSG_DONTWAIT);
1598 if (unlikely(err)) {
1599 ip_flush_pending_frames(sk);
1600 goto out;
1601 }
1602
1603 nskb = skb_peek(&sk->sk_write_queue);
1604 if (nskb) {
1605 if (arg->csumoffset >= 0)
1606 *((__sum16 *)skb_transport_header(nskb) +
1607 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1608 arg->csum));
1609 nskb->ip_summed = CHECKSUM_NONE;
1610 ip_push_pending_frames(sk, &fl4);
1611 }
1612 out:
1613 ip_rt_put(rt);
1614 }
1615
1616 void __init ip_init(void)
1617 {
1618 ip_rt_init();
1619 inet_initpeers();
1620
1621 #if defined(CONFIG_IP_MULTICAST)
1622 igmp_mc_init();
1623 #endif
1624 }