1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * The Internet Protocol (IP) output module.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Donald Becker, <becker@super.org>
12 * Alan Cox, <Alan.Cox@linux.org>
13 * Richard Underwood
14 * Stefan Becker, <stefanb@yello.ping.de>
15 * Jorge Cwik, <jorge@laser.satlink.net>
16 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
17 * Hirokazu Takahashi, <taka@valinux.co.jp>
18 *
19 * See ip_input.c for original log
20 *
21 * Fixes:
22 * Alan Cox : Missing nonblock feature in ip_build_xmit.
23 * Mike Kilburn : htons() missing in ip_build_xmit.
24 * Bradford Johnson: Fix faulty handling of some frames when
25 * no route is found.
26 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
27 * (in case if packet not accepted by
28 * output firewall rules)
29 * Mike McLagan : Routing by source
30 * Alexey Kuznetsov: use new route cache
31 * Andi Kleen: Fix broken PMTU recovery and remove
32 * some redundant tests.
33 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
34 * Andi Kleen : Replace ip_reply with ip_send_reply.
35 * Andi Kleen : Split fast and slow ip_build_xmit path
36 * for decreased register pressure on x86
37 * and more readability.
38 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
39 * silently drop skb instead of failing with -EPERM.
40 * Detlev Wengorz : Copy protocol for fragments.
41 * Hirokazu Takahashi: HW checksumming for outgoing UDP
42 * datagrams.
43 * Hirokazu Takahashi: sendfile() on UDP works now.
44 */
45
46#include <linux/uaccess.h>
47#include <linux/module.h>
48#include <linux/types.h>
49#include <linux/kernel.h>
50#include <linux/mm.h>
51#include <linux/string.h>
52#include <linux/errno.h>
53#include <linux/highmem.h>
54#include <linux/slab.h>
55
56#include <linux/socket.h>
57#include <linux/sockios.h>
58#include <linux/in.h>
59#include <linux/inet.h>
60#include <linux/netdevice.h>
61#include <linux/etherdevice.h>
62#include <linux/proc_fs.h>
63#include <linux/stat.h>
64#include <linux/init.h>
65
66#include <net/snmp.h>
67#include <net/ip.h>
68#include <net/protocol.h>
69#include <net/route.h>
70#include <net/xfrm.h>
71#include <linux/skbuff.h>
72#include <net/sock.h>
73#include <net/arp.h>
74#include <net/icmp.h>
75#include <net/checksum.h>
76#include <net/gso.h>
77#include <net/inetpeer.h>
78#include <net/inet_ecn.h>
79#include <net/lwtunnel.h>
80#include <linux/bpf-cgroup.h>
81#include <linux/igmp.h>
82#include <linux/netfilter_ipv4.h>
83#include <linux/netfilter_bridge.h>
84#include <linux/netlink.h>
85#include <linux/tcp.h>
86
87static int
88ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
89 unsigned int mtu,
90 int (*output)(struct net *, struct sock *, struct sk_buff *));
91
92/* Generate a checksum for an outgoing IP datagram. */
93void ip_send_check(struct iphdr *iph)
94{
95 iph->check = 0;
96 iph->check = ip_fast_csum(iph: (unsigned char *)iph, ihl: iph->ihl);
97}
98EXPORT_SYMBOL(ip_send_check);
99
100int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
101{
102 struct iphdr *iph = ip_hdr(skb);
103
104 IP_INC_STATS(net, IPSTATS_MIB_OUTREQUESTS);
105
106 iph_set_totlen(iph, len: skb->len);
107 ip_send_check(iph);
108
109 /* if egress device is enslaved to an L3 master device pass the
110 * skb to its handler for processing
111 */
112 skb = l3mdev_ip_out(sk, skb);
113 if (unlikely(!skb))
114 return 0;
115
116 skb->protocol = htons(ETH_P_IP);
117
118 return nf_hook(pf: NFPROTO_IPV4, hook: NF_INET_LOCAL_OUT,
119 net, sk, skb, NULL, outdev: skb_dst(skb)->dev,
120 okfn: dst_output);
121}
122
123int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
124{
125 int err;
126
127 err = __ip_local_out(net, sk, skb);
128 if (likely(err == 1))
129 err = dst_output(net, sk, skb);
130
131 return err;
132}
133EXPORT_SYMBOL_GPL(ip_local_out);
134
135static inline int ip_select_ttl(const struct inet_sock *inet,
136 const struct dst_entry *dst)
137{
138 int ttl = READ_ONCE(inet->uc_ttl);
139
140 if (ttl < 0)
141 ttl = ip4_dst_hoplimit(dst);
142 return ttl;
143}
144
145/*
146 * Add an ip header to a skbuff and send it out.
147 *
148 */
149int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
150 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
151 u8 tos)
152{
153 const struct inet_sock *inet = inet_sk(sk);
154 struct rtable *rt = skb_rtable(skb);
155 struct net *net = sock_net(sk);
156 struct iphdr *iph;
157
158 /* Build the IP header. */
159 skb_push(skb, len: sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
160 skb_reset_network_header(skb);
161 iph = ip_hdr(skb);
162 iph->version = 4;
163 iph->ihl = 5;
164 iph->tos = tos;
165 iph->ttl = ip_select_ttl(inet, dst: &rt->dst);
166 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
167 iph->saddr = saddr;
168 iph->protocol = sk->sk_protocol;
169 /* Do not bother generating IPID for small packets (eg SYNACK) */
170 if (skb->len <= IPV4_MIN_MTU || ip_dont_fragment(sk, dst: &rt->dst)) {
171 iph->frag_off = htons(IP_DF);
172 iph->id = 0;
173 } else {
174 iph->frag_off = 0;
175 /* TCP packets here are SYNACK with fat IPv4/TCP options.
176 * Avoid using the hashed IP ident generator.
177 */
178 if (sk->sk_protocol == IPPROTO_TCP)
179 iph->id = (__force __be16)get_random_u16();
180 else
181 __ip_select_ident(net, iph, segs: 1);
182 }
183
184 if (opt && opt->opt.optlen) {
185 iph->ihl += opt->opt.optlen>>2;
186 ip_options_build(skb, opt: &opt->opt, daddr, rt);
187 }
188
189 skb->priority = READ_ONCE(sk->sk_priority);
190 if (!skb->mark)
191 skb->mark = READ_ONCE(sk->sk_mark);
192
193 /* Send it out. */
194 return ip_local_out(net, skb->sk, skb);
195}
196EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
197
198static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
199{
200 struct dst_entry *dst = skb_dst(skb);
201 struct rtable *rt = (struct rtable *)dst;
202 struct net_device *dev = dst->dev;
203 unsigned int hh_len = LL_RESERVED_SPACE(dev);
204 struct neighbour *neigh;
205 bool is_v6gw = false;
206
207 if (rt->rt_type == RTN_MULTICAST) {
208 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
209 } else if (rt->rt_type == RTN_BROADCAST)
210 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
211
212 /* OUTOCTETS should be counted after fragment */
213 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
214
215 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
216 skb = skb_expand_head(skb, headroom: hh_len);
217 if (!skb)
218 return -ENOMEM;
219 }
220
221 if (lwtunnel_xmit_redirect(lwtstate: dst->lwtstate)) {
222 int res = lwtunnel_xmit(skb);
223
224 if (res != LWTUNNEL_XMIT_CONTINUE)
225 return res;
226 }
227
228 rcu_read_lock();
229 neigh = ip_neigh_for_gw(rt, skb, is_v6gw: &is_v6gw);
230 if (!IS_ERR(ptr: neigh)) {
231 int res;
232
233 sock_confirm_neigh(skb, n: neigh);
234 /* if crossing protocols, can not use the cached header */
235 res = neigh_output(n: neigh, skb, skip_cache: is_v6gw);
236 rcu_read_unlock();
237 return res;
238 }
239 rcu_read_unlock();
240
241 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
242 __func__);
243 kfree_skb_reason(skb, reason: SKB_DROP_REASON_NEIGH_CREATEFAIL);
244 return PTR_ERR(ptr: neigh);
245}
246
247static int ip_finish_output_gso(struct net *net, struct sock *sk,
248 struct sk_buff *skb, unsigned int mtu)
249{
250 struct sk_buff *segs, *nskb;
251 netdev_features_t features;
252 int ret = 0;
253
254 /* common case: seglen is <= mtu
255 */
256 if (skb_gso_validate_network_len(skb, mtu))
257 return ip_finish_output2(net, sk, skb);
258
259 /* Slowpath - GSO segment length exceeds the egress MTU.
260 *
261 * This can happen in several cases:
262 * - Forwarding of a TCP GRO skb, when DF flag is not set.
263 * - Forwarding of an skb that arrived on a virtualization interface
264 * (virtio-net/vhost/tap) with TSO/GSO size set by other network
265 * stack.
266 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
267 * interface with a smaller MTU.
268 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is
269 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an
270 * insufficient MTU.
271 */
272 features = netif_skb_features(skb);
273 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
274 segs = skb_gso_segment(skb, features: features & ~NETIF_F_GSO_MASK);
275 if (IS_ERR_OR_NULL(ptr: segs)) {
276 kfree_skb(skb);
277 return -ENOMEM;
278 }
279
280 consume_skb(skb);
281
282 skb_list_walk_safe(segs, segs, nskb) {
283 int err;
284
285 skb_mark_not_on_list(skb: segs);
286 err = ip_fragment(net, sk, skb: segs, mtu, output: ip_finish_output2);
287
288 if (err && ret == 0)
289 ret = err;
290 }
291
292 return ret;
293}
294
295static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
296{
297 unsigned int mtu;
298
299#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
300 /* Policy lookup after SNAT yielded a new policy */
301 if (skb_dst(skb)->xfrm) {
302 IPCB(skb)->flags |= IPSKB_REROUTED;
303 return dst_output(net, sk, skb);
304 }
305#endif
306 mtu = ip_skb_dst_mtu(sk, skb);
307 if (skb_is_gso(skb))
308 return ip_finish_output_gso(net, sk, skb, mtu);
309
310 if (skb->len > mtu || IPCB(skb)->frag_max_size)
311 return ip_fragment(net, sk, skb, mtu, output: ip_finish_output2);
312
313 return ip_finish_output2(net, sk, skb);
314}
315
316static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
317{
318 int ret;
319
320 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
321 switch (ret) {
322 case NET_XMIT_SUCCESS:
323 return __ip_finish_output(net, sk, skb);
324 case NET_XMIT_CN:
325 return __ip_finish_output(net, sk, skb) ? : ret;
326 default:
327 kfree_skb_reason(skb, reason: SKB_DROP_REASON_BPF_CGROUP_EGRESS);
328 return ret;
329 }
330}
331
332static int ip_mc_finish_output(struct net *net, struct sock *sk,
333 struct sk_buff *skb)
334{
335 struct rtable *new_rt;
336 bool do_cn = false;
337 int ret, err;
338
339 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
340 switch (ret) {
341 case NET_XMIT_CN:
342 do_cn = true;
343 fallthrough;
344 case NET_XMIT_SUCCESS:
345 break;
346 default:
347 kfree_skb_reason(skb, reason: SKB_DROP_REASON_BPF_CGROUP_EGRESS);
348 return ret;
349 }
350
351 /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
352 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
353 * see ipv4_pktinfo_prepare().
354 */
355 new_rt = rt_dst_clone(dev: net->loopback_dev, rt: skb_rtable(skb));
356 if (new_rt) {
357 new_rt->rt_iif = 0;
358 skb_dst_drop(skb);
359 skb_dst_set(skb, dst: &new_rt->dst);
360 }
361
362 err = dev_loopback_xmit(net, sk, newskb: skb);
363 return (do_cn && err) ? ret : err;
364}
365
366int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
367{
368 struct rtable *rt = skb_rtable(skb);
369 struct net_device *dev = rt->dst.dev;
370
371 /*
372 * If the indicated interface is up and running, send the packet.
373 */
374 skb->dev = dev;
375 skb->protocol = htons(ETH_P_IP);
376
377 /*
378 * Multicasts are looped back for other local users
379 */
380
381 if (rt->rt_flags&RTCF_MULTICAST) {
382 if (sk_mc_loop(sk)
383#ifdef CONFIG_IP_MROUTE
384 /* Small optimization: do not loopback not local frames,
385 which returned after forwarding; they will be dropped
386 by ip_mr_input in any case.
387 Note, that local frames are looped back to be delivered
388 to local recipients.
389
390 This check is duplicated in ip_mr_input at the moment.
391 */
392 &&
393 ((rt->rt_flags & RTCF_LOCAL) ||
394 !(IPCB(skb)->flags & IPSKB_FORWARDED))
395#endif
396 ) {
397 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
398 if (newskb)
399 NF_HOOK(pf: NFPROTO_IPV4, hook: NF_INET_POST_ROUTING,
400 net, sk, skb: newskb, NULL, out: newskb->dev,
401 okfn: ip_mc_finish_output);
402 }
403
404 /* Multicasts with ttl 0 must not go beyond the host */
405
406 if (ip_hdr(skb)->ttl == 0) {
407 kfree_skb(skb);
408 return 0;
409 }
410 }
411
412 if (rt->rt_flags&RTCF_BROADCAST) {
413 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
414 if (newskb)
415 NF_HOOK(pf: NFPROTO_IPV4, hook: NF_INET_POST_ROUTING,
416 net, sk, skb: newskb, NULL, out: newskb->dev,
417 okfn: ip_mc_finish_output);
418 }
419
420 return NF_HOOK_COND(pf: NFPROTO_IPV4, hook: NF_INET_POST_ROUTING,
421 net, sk, skb, NULL, out: skb->dev,
422 okfn: ip_finish_output,
423 cond: !(IPCB(skb)->flags & IPSKB_REROUTED));
424}
425
426int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
427{
428 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
429
430 skb->dev = dev;
431 skb->protocol = htons(ETH_P_IP);
432
433 return NF_HOOK_COND(pf: NFPROTO_IPV4, hook: NF_INET_POST_ROUTING,
434 net, sk, skb, in: indev, out: dev,
435 okfn: ip_finish_output,
436 cond: !(IPCB(skb)->flags & IPSKB_REROUTED));
437}
438EXPORT_SYMBOL(ip_output);
439
440/*
441 * copy saddr and daddr, possibly using 64bit load/stores
442 * Equivalent to :
443 * iph->saddr = fl4->saddr;
444 * iph->daddr = fl4->daddr;
445 */
446static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
447{
448 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
449 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
450
451 iph->saddr = fl4->saddr;
452 iph->daddr = fl4->daddr;
453}
454
455/* Note: skb->sk can be different from sk, in case of tunnels */
456int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
457 __u8 tos)
458{
459 struct inet_sock *inet = inet_sk(sk);
460 struct net *net = sock_net(sk);
461 struct ip_options_rcu *inet_opt;
462 struct flowi4 *fl4;
463 struct rtable *rt;
464 struct iphdr *iph;
465 int res;
466
467 /* Skip all of this if the packet is already routed,
468 * f.e. by something like SCTP.
469 */
470 rcu_read_lock();
471 inet_opt = rcu_dereference(inet->inet_opt);
472 fl4 = &fl->u.ip4;
473 rt = skb_rtable(skb);
474 if (rt)
475 goto packet_routed;
476
477 /* Make sure we can route this packet. */
478 rt = (struct rtable *)__sk_dst_check(sk, cookie: 0);
479 if (!rt) {
480 __be32 daddr;
481
482 /* Use correct destination address if we have options. */
483 daddr = inet->inet_daddr;
484 if (inet_opt && inet_opt->opt.srr)
485 daddr = inet_opt->opt.faddr;
486
487 /* If this fails, retransmit mechanism of transport layer will
488 * keep trying until route appears or the connection times
489 * itself out.
490 */
491 rt = ip_route_output_ports(net, fl4, sk,
492 daddr, saddr: inet->inet_saddr,
493 dport: inet->inet_dport,
494 sport: inet->inet_sport,
495 proto: sk->sk_protocol,
496 RT_TOS(tos),
497 oif: sk->sk_bound_dev_if);
498 if (IS_ERR(ptr: rt))
499 goto no_route;
500 sk_setup_caps(sk, dst: &rt->dst);
501 }
502 skb_dst_set_noref(skb, dst: &rt->dst);
503
504packet_routed:
505 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
506 goto no_route;
507
508 /* OK, we know where to send it, allocate and build IP header. */
509 skb_push(skb, len: sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
510 skb_reset_network_header(skb);
511 iph = ip_hdr(skb);
512 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
513 if (ip_dont_fragment(sk, dst: &rt->dst) && !skb->ignore_df)
514 iph->frag_off = htons(IP_DF);
515 else
516 iph->frag_off = 0;
517 iph->ttl = ip_select_ttl(inet, dst: &rt->dst);
518 iph->protocol = sk->sk_protocol;
519 ip_copy_addrs(iph, fl4);
520
521 /* Transport layer set skb->h.foo itself. */
522
523 if (inet_opt && inet_opt->opt.optlen) {
524 iph->ihl += inet_opt->opt.optlen >> 2;
525 ip_options_build(skb, opt: &inet_opt->opt, daddr: inet->inet_daddr, rt);
526 }
527
528 ip_select_ident_segs(net, skb, sk,
529 skb_shinfo(skb)->gso_segs ?: 1);
530
531 /* TODO : should we use skb->sk here instead of sk ? */
532 skb->priority = READ_ONCE(sk->sk_priority);
533 skb->mark = READ_ONCE(sk->sk_mark);
534
535 res = ip_local_out(net, sk, skb);
536 rcu_read_unlock();
537 return res;
538
539no_route:
540 rcu_read_unlock();
541 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
542 kfree_skb_reason(skb, reason: SKB_DROP_REASON_IP_OUTNOROUTES);
543 return -EHOSTUNREACH;
544}
545EXPORT_SYMBOL(__ip_queue_xmit);
546
547int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
548{
549 return __ip_queue_xmit(sk, skb, fl, READ_ONCE(inet_sk(sk)->tos));
550}
551EXPORT_SYMBOL(ip_queue_xmit);
552
553static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
554{
555 to->pkt_type = from->pkt_type;
556 to->priority = from->priority;
557 to->protocol = from->protocol;
558 to->skb_iif = from->skb_iif;
559 skb_dst_drop(skb: to);
560 skb_dst_copy(nskb: to, oskb: from);
561 to->dev = from->dev;
562 to->mark = from->mark;
563
564 skb_copy_hash(to, from);
565
566#ifdef CONFIG_NET_SCHED
567 to->tc_index = from->tc_index;
568#endif
569 nf_copy(dst: to, src: from);
570 skb_ext_copy(dst: to, src: from);
571#if IS_ENABLED(CONFIG_IP_VS)
572 to->ipvs_property = from->ipvs_property;
573#endif
574 skb_copy_secmark(to, from);
575}
576
577static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
578 unsigned int mtu,
579 int (*output)(struct net *, struct sock *, struct sk_buff *))
580{
581 struct iphdr *iph = ip_hdr(skb);
582
583 if ((iph->frag_off & htons(IP_DF)) == 0)
584 return ip_do_fragment(net, sk, skb, output);
585
586 if (unlikely(!skb->ignore_df ||
587 (IPCB(skb)->frag_max_size &&
588 IPCB(skb)->frag_max_size > mtu))) {
589 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
590 icmp_send(skb_in: skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
591 htonl(mtu));
592 kfree_skb(skb);
593 return -EMSGSIZE;
594 }
595
596 return ip_do_fragment(net, sk, skb, output);
597}
598
599void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
600 unsigned int hlen, struct ip_fraglist_iter *iter)
601{
602 unsigned int first_len = skb_pagelen(skb);
603
604 iter->frag = skb_shinfo(skb)->frag_list;
605 skb_frag_list_init(skb);
606
607 iter->offset = 0;
608 iter->iph = iph;
609 iter->hlen = hlen;
610
611 skb->data_len = first_len - skb_headlen(skb);
612 skb->len = first_len;
613 iph->tot_len = htons(first_len);
614 iph->frag_off = htons(IP_MF);
615 ip_send_check(iph);
616}
617EXPORT_SYMBOL(ip_fraglist_init);
618
619void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
620{
621 unsigned int hlen = iter->hlen;
622 struct iphdr *iph = iter->iph;
623 struct sk_buff *frag;
624
625 frag = iter->frag;
626 frag->ip_summed = CHECKSUM_NONE;
627 skb_reset_transport_header(skb: frag);
628 __skb_push(skb: frag, len: hlen);
629 skb_reset_network_header(skb: frag);
630 memcpy(skb_network_header(frag), iph, hlen);
631 iter->iph = ip_hdr(skb: frag);
632 iph = iter->iph;
633 iph->tot_len = htons(frag->len);
634 ip_copy_metadata(to: frag, from: skb);
635 iter->offset += skb->len - hlen;
636 iph->frag_off = htons(iter->offset >> 3);
637 if (frag->next)
638 iph->frag_off |= htons(IP_MF);
639 /* Ready, complete checksum */
640 ip_send_check(iph);
641}
642EXPORT_SYMBOL(ip_fraglist_prepare);
643
644void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
645 unsigned int ll_rs, unsigned int mtu, bool DF,
646 struct ip_frag_state *state)
647{
648 struct iphdr *iph = ip_hdr(skb);
649
650 state->DF = DF;
651 state->hlen = hlen;
652 state->ll_rs = ll_rs;
653 state->mtu = mtu;
654
655 state->left = skb->len - hlen; /* Space per frame */
656 state->ptr = hlen; /* Where to start from */
657
658 state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
659 state->not_last_frag = iph->frag_off & htons(IP_MF);
660}
661EXPORT_SYMBOL(ip_frag_init);
662
663static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
664 bool first_frag)
665{
666 /* Copy the flags to each fragment. */
667 IPCB(to)->flags = IPCB(from)->flags;
668
669 /* ANK: dirty, but effective trick. Upgrade options only if
670 * the segment to be fragmented was THE FIRST (otherwise,
671 * options are already fixed) and make it ONCE
672 * on the initial skb, so that all the following fragments
673 * will inherit fixed options.
674 */
675 if (first_frag)
676 ip_options_fragment(skb: from);
677}
678
679struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
680{
681 unsigned int len = state->left;
682 struct sk_buff *skb2;
683 struct iphdr *iph;
684
685 /* IF: it doesn't fit, use 'mtu' - the data space left */
686 if (len > state->mtu)
687 len = state->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 < state->left) {
691 len &= ~7;
692 }
693
694 /* Allocate buffer */
695 skb2 = alloc_skb(size: len + state->hlen + state->ll_rs, GFP_ATOMIC);
696 if (!skb2)
697 return ERR_PTR(error: -ENOMEM);
698
699 /*
700 * Set up data on packet
701 */
702
703 ip_copy_metadata(to: skb2, from: skb);
704 skb_reserve(skb: skb2, len: state->ll_rs);
705 skb_put(skb: skb2, len: len + state->hlen);
706 skb_reset_network_header(skb: skb2);
707 skb2->transport_header = skb2->network_header + state->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(skb: skb2, sk: skb->sk);
716
717 /*
718 * Copy the packet header into the new buffer.
719 */
720
721 skb_copy_from_linear_data(skb, to: skb_network_header(skb: skb2), len: state->hlen);
722
723 /*
724 * Copy a block of the IP datagram.
725 */
726 if (skb_copy_bits(skb, offset: state->ptr, to: skb_transport_header(skb: skb2), len))
727 BUG();
728 state->left -= len;
729
730 /*
731 * Fill in the new header fields.
732 */
733 iph = ip_hdr(skb: skb2);
734 iph->frag_off = htons((state->offset >> 3));
735 if (state->DF)
736 iph->frag_off |= htons(IP_DF);
737
738 /*
739 * Added AC : If we are fragmenting a fragment that's not the
740 * last fragment then keep MF on each bit
741 */
742 if (state->left > 0 || state->not_last_frag)
743 iph->frag_off |= htons(IP_MF);
744 state->ptr += len;
745 state->offset += len;
746
747 iph->tot_len = htons(len + state->hlen);
748
749 ip_send_check(iph);
750
751 return skb2;
752}
753EXPORT_SYMBOL(ip_frag_next);
754
755/*
756 * This IP datagram is too large to be sent in one piece. Break it up into
757 * smaller pieces (each of size equal to IP header plus
758 * a block of the data of the original IP data part) that will yet fit in a
759 * single device frame, and queue such a frame for sending.
760 */
761
762int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
763 int (*output)(struct net *, struct sock *, struct sk_buff *))
764{
765 struct iphdr *iph;
766 struct sk_buff *skb2;
767 bool mono_delivery_time = skb->mono_delivery_time;
768 struct rtable *rt = skb_rtable(skb);
769 unsigned int mtu, hlen, ll_rs;
770 struct ip_fraglist_iter iter;
771 ktime_t tstamp = skb->tstamp;
772 struct ip_frag_state state;
773 int err = 0;
774
775 /* for offloaded checksums cleanup checksum before fragmentation */
776 if (skb->ip_summed == CHECKSUM_PARTIAL &&
777 (err = skb_checksum_help(skb)))
778 goto fail;
779
780 /*
781 * Point into the IP datagram header.
782 */
783
784 iph = ip_hdr(skb);
785
786 mtu = ip_skb_dst_mtu(sk, skb);
787 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
788 mtu = IPCB(skb)->frag_max_size;
789
790 /*
791 * Setup starting values.
792 */
793
794 hlen = iph->ihl * 4;
795 mtu = mtu - hlen; /* Size of data space */
796 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
797 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
798
799 /* When frag_list is given, use it. First, check its validity:
800 * some transformers could create wrong frag_list or break existing
801 * one, it is not prohibited. In this case fall back to copying.
802 *
803 * LATER: this step can be merged to real generation of fragments,
804 * we can switch to copy when see the first bad fragment.
805 */
806 if (skb_has_frag_list(skb)) {
807 struct sk_buff *frag, *frag2;
808 unsigned int first_len = skb_pagelen(skb);
809
810 if (first_len - hlen > mtu ||
811 ((first_len - hlen) & 7) ||
812 ip_is_fragment(iph) ||
813 skb_cloned(skb) ||
814 skb_headroom(skb) < ll_rs)
815 goto slow_path;
816
817 skb_walk_frags(skb, frag) {
818 /* Correct geometry. */
819 if (frag->len > mtu ||
820 ((frag->len & 7) && frag->next) ||
821 skb_headroom(skb: frag) < hlen + ll_rs)
822 goto slow_path_clean;
823
824 /* Partially cloned skb? */
825 if (skb_shared(skb: frag))
826 goto slow_path_clean;
827
828 BUG_ON(frag->sk);
829 if (skb->sk) {
830 frag->sk = skb->sk;
831 frag->destructor = sock_wfree;
832 }
833 skb->truesize -= frag->truesize;
834 }
835
836 /* Everything is OK. Generate! */
837 ip_fraglist_init(skb, iph, hlen, &iter);
838
839 for (;;) {
840 /* Prepare header of the next frame,
841 * before previous one went down. */
842 if (iter.frag) {
843 bool first_frag = (iter.offset == 0);
844
845 IPCB(iter.frag)->flags = IPCB(skb)->flags;
846 ip_fraglist_prepare(skb, &iter);
847 if (first_frag && IPCB(skb)->opt.optlen) {
848 /* ipcb->opt is not populated for frags
849 * coming from __ip_make_skb(),
850 * ip_options_fragment() needs optlen
851 */
852 IPCB(iter.frag)->opt.optlen =
853 IPCB(skb)->opt.optlen;
854 ip_options_fragment(skb: iter.frag);
855 ip_send_check(iter.iph);
856 }
857 }
858
859 skb_set_delivery_time(skb, kt: tstamp, mono: mono_delivery_time);
860 err = output(net, sk, skb);
861
862 if (!err)
863 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
864 if (err || !iter.frag)
865 break;
866
867 skb = ip_fraglist_next(iter: &iter);
868 }
869
870 if (err == 0) {
871 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
872 return 0;
873 }
874
875 kfree_skb_list(segs: iter.frag);
876
877 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
878 return err;
879
880slow_path_clean:
881 skb_walk_frags(skb, frag2) {
882 if (frag2 == frag)
883 break;
884 frag2->sk = NULL;
885 frag2->destructor = NULL;
886 skb->truesize += frag2->truesize;
887 }
888 }
889
890slow_path:
891 /*
892 * Fragment the datagram.
893 */
894
895 ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
896 &state);
897
898 /*
899 * Keep copying data until we run out.
900 */
901
902 while (state.left > 0) {
903 bool first_frag = (state.offset == 0);
904
905 skb2 = ip_frag_next(skb, &state);
906 if (IS_ERR(ptr: skb2)) {
907 err = PTR_ERR(ptr: skb2);
908 goto fail;
909 }
910 ip_frag_ipcb(from: skb, to: skb2, first_frag);
911
912 /*
913 * Put this fragment into the sending queue.
914 */
915 skb_set_delivery_time(skb: skb2, kt: tstamp, mono: mono_delivery_time);
916 err = output(net, sk, skb2);
917 if (err)
918 goto fail;
919
920 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
921 }
922 consume_skb(skb);
923 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
924 return err;
925
926fail:
927 kfree_skb(skb);
928 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
929 return err;
930}
931EXPORT_SYMBOL(ip_do_fragment);
932
933int
934ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
935{
936 struct msghdr *msg = from;
937
938 if (skb->ip_summed == CHECKSUM_PARTIAL) {
939 if (!copy_from_iter_full(addr: to, bytes: len, i: &msg->msg_iter))
940 return -EFAULT;
941 } else {
942 __wsum csum = 0;
943 if (!csum_and_copy_from_iter_full(addr: to, bytes: len, csum: &csum, i: &msg->msg_iter))
944 return -EFAULT;
945 skb->csum = csum_block_add(csum: skb->csum, csum2: csum, offset: odd);
946 }
947 return 0;
948}
949EXPORT_SYMBOL(ip_generic_getfrag);
950
951static int __ip_append_data(struct sock *sk,
952 struct flowi4 *fl4,
953 struct sk_buff_head *queue,
954 struct inet_cork *cork,
955 struct page_frag *pfrag,
956 int getfrag(void *from, char *to, int offset,
957 int len, int odd, struct sk_buff *skb),
958 void *from, int length, int transhdrlen,
959 unsigned int flags)
960{
961 struct inet_sock *inet = inet_sk(sk);
962 struct ubuf_info *uarg = NULL;
963 struct sk_buff *skb;
964 struct ip_options *opt = cork->opt;
965 int hh_len;
966 int exthdrlen;
967 int mtu;
968 int copy;
969 int err;
970 int offset = 0;
971 bool zc = false;
972 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
973 int csummode = CHECKSUM_NONE;
974 struct rtable *rt = (struct rtable *)cork->dst;
975 bool paged, hold_tskey, extra_uref = false;
976 unsigned int wmem_alloc_delta = 0;
977 u32 tskey = 0;
978
979 skb = skb_peek_tail(list_: queue);
980
981 exthdrlen = !skb ? rt->dst.header_len : 0;
982 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
983 paged = !!cork->gso_size;
984
985 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
986
987 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
988 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
989 maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
990
991 if (cork->length + length > maxnonfragsize - fragheaderlen) {
992 ip_local_error(sk, EMSGSIZE, daddr: fl4->daddr, dport: inet->inet_dport,
993 info: mtu - (opt ? opt->optlen : 0));
994 return -EMSGSIZE;
995 }
996
997 /*
998 * transhdrlen > 0 means that this is the first fragment and we wish
999 * it won't be fragmented in the future.
1000 */
1001 if (transhdrlen &&
1002 length + fragheaderlen <= mtu &&
1003 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1004 (!(flags & MSG_MORE) || cork->gso_size) &&
1005 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1006 csummode = CHECKSUM_PARTIAL;
1007
1008 if ((flags & MSG_ZEROCOPY) && length) {
1009 struct msghdr *msg = from;
1010
1011 if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1012 if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1013 return -EINVAL;
1014
1015 /* Leave uarg NULL if can't zerocopy, callers should
1016 * be able to handle it.
1017 */
1018 if ((rt->dst.dev->features & NETIF_F_SG) &&
1019 csummode == CHECKSUM_PARTIAL) {
1020 paged = true;
1021 zc = true;
1022 uarg = msg->msg_ubuf;
1023 }
1024 } else if (sock_flag(sk, flag: SOCK_ZEROCOPY)) {
1025 uarg = msg_zerocopy_realloc(sk, size: length, uarg: skb_zcopy(skb));
1026 if (!uarg)
1027 return -ENOBUFS;
1028 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1029 if (rt->dst.dev->features & NETIF_F_SG &&
1030 csummode == CHECKSUM_PARTIAL) {
1031 paged = true;
1032 zc = true;
1033 } else {
1034 uarg_to_msgzc(uarg)->zerocopy = 0;
1035 skb_zcopy_set(skb, uarg, have_ref: &extra_uref);
1036 }
1037 }
1038 } else if ((flags & MSG_SPLICE_PAGES) && length) {
1039 if (inet_test_bit(HDRINCL, sk))
1040 return -EPERM;
1041 if (rt->dst.dev->features & NETIF_F_SG &&
1042 getfrag == ip_generic_getfrag)
1043 /* We need an empty buffer to attach stuff to */
1044 paged = true;
1045 else
1046 flags &= ~MSG_SPLICE_PAGES;
1047 }
1048
1049 cork->length += length;
1050
1051 hold_tskey = cork->tx_flags & SKBTX_ANY_TSTAMP &&
1052 READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID;
1053 if (hold_tskey)
1054 tskey = atomic_inc_return(v: &sk->sk_tskey) - 1;
1055
1056 /* So, what's going on in the loop below?
1057 *
1058 * We use calculated fragment length to generate chained skb,
1059 * each of segments is IP fragment ready for sending to network after
1060 * adding appropriate IP header.
1061 */
1062
1063 if (!skb)
1064 goto alloc_new_skb;
1065
1066 while (length > 0) {
1067 /* Check if the remaining data fits into current packet. */
1068 copy = mtu - skb->len;
1069 if (copy < length)
1070 copy = maxfraglen - skb->len;
1071 if (copy <= 0) {
1072 char *data;
1073 unsigned int datalen;
1074 unsigned int fraglen;
1075 unsigned int fraggap;
1076 unsigned int alloclen, alloc_extra;
1077 unsigned int pagedlen;
1078 struct sk_buff *skb_prev;
1079alloc_new_skb:
1080 skb_prev = skb;
1081 if (skb_prev)
1082 fraggap = skb_prev->len - maxfraglen;
1083 else
1084 fraggap = 0;
1085
1086 /*
1087 * If remaining data exceeds the mtu,
1088 * we know we need more fragment(s).
1089 */
1090 datalen = length + fraggap;
1091 if (datalen > mtu - fragheaderlen)
1092 datalen = maxfraglen - fragheaderlen;
1093 fraglen = datalen + fragheaderlen;
1094 pagedlen = 0;
1095
1096 alloc_extra = hh_len + 15;
1097 alloc_extra += exthdrlen;
1098
1099 /* The last fragment gets additional space at tail.
1100 * Note, with MSG_MORE we overallocate on fragments,
1101 * because we have no idea what fragment will be
1102 * the last.
1103 */
1104 if (datalen == length + fraggap)
1105 alloc_extra += rt->dst.trailer_len;
1106
1107 if ((flags & MSG_MORE) &&
1108 !(rt->dst.dev->features&NETIF_F_SG))
1109 alloclen = mtu;
1110 else if (!paged &&
1111 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1112 !(rt->dst.dev->features & NETIF_F_SG)))
1113 alloclen = fraglen;
1114 else {
1115 alloclen = fragheaderlen + transhdrlen;
1116 pagedlen = datalen - transhdrlen;
1117 }
1118
1119 alloclen += alloc_extra;
1120
1121 if (transhdrlen) {
1122 skb = sock_alloc_send_skb(sk, size: alloclen,
1123 noblock: (flags & MSG_DONTWAIT), errcode: &err);
1124 } else {
1125 skb = NULL;
1126 if (refcount_read(r: &sk->sk_wmem_alloc) + wmem_alloc_delta <=
1127 2 * sk->sk_sndbuf)
1128 skb = alloc_skb(size: alloclen,
1129 priority: sk->sk_allocation);
1130 if (unlikely(!skb))
1131 err = -ENOBUFS;
1132 }
1133 if (!skb)
1134 goto error;
1135
1136 /*
1137 * Fill in the control structures
1138 */
1139 skb->ip_summed = csummode;
1140 skb->csum = 0;
1141 skb_reserve(skb, len: hh_len);
1142
1143 /*
1144 * Find where to start putting bytes.
1145 */
1146 data = skb_put(skb, len: fraglen + exthdrlen - pagedlen);
1147 skb_set_network_header(skb, offset: exthdrlen);
1148 skb->transport_header = (skb->network_header +
1149 fragheaderlen);
1150 data += fragheaderlen + exthdrlen;
1151
1152 if (fraggap) {
1153 skb->csum = skb_copy_and_csum_bits(
1154 skb: skb_prev, offset: maxfraglen,
1155 to: data + transhdrlen, len: fraggap);
1156 skb_prev->csum = csum_sub(csum: skb_prev->csum,
1157 addend: skb->csum);
1158 data += fraggap;
1159 pskb_trim_unique(skb: skb_prev, len: maxfraglen);
1160 }
1161
1162 copy = datalen - transhdrlen - fraggap - pagedlen;
1163 /* [!] NOTE: copy will be negative if pagedlen>0
1164 * because then the equation reduces to -fraggap.
1165 */
1166 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1167 err = -EFAULT;
1168 kfree_skb(skb);
1169 goto error;
1170 } else if (flags & MSG_SPLICE_PAGES) {
1171 copy = 0;
1172 }
1173
1174 offset += copy;
1175 length -= copy + transhdrlen;
1176 transhdrlen = 0;
1177 exthdrlen = 0;
1178 csummode = CHECKSUM_NONE;
1179
1180 /* only the initial fragment is time stamped */
1181 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1182 cork->tx_flags = 0;
1183 skb_shinfo(skb)->tskey = tskey;
1184 tskey = 0;
1185 skb_zcopy_set(skb, uarg, have_ref: &extra_uref);
1186
1187 if ((flags & MSG_CONFIRM) && !skb_prev)
1188 skb_set_dst_pending_confirm(skb, val: 1);
1189
1190 /*
1191 * Put the packet on the pending queue.
1192 */
1193 if (!skb->destructor) {
1194 skb->destructor = sock_wfree;
1195 skb->sk = sk;
1196 wmem_alloc_delta += skb->truesize;
1197 }
1198 __skb_queue_tail(list: queue, newsk: skb);
1199 continue;
1200 }
1201
1202 if (copy > length)
1203 copy = length;
1204
1205 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1206 skb_tailroom(skb) >= copy) {
1207 unsigned int off;
1208
1209 off = skb->len;
1210 if (getfrag(from, skb_put(skb, len: copy),
1211 offset, copy, off, skb) < 0) {
1212 __skb_trim(skb, len: off);
1213 err = -EFAULT;
1214 goto error;
1215 }
1216 } else if (flags & MSG_SPLICE_PAGES) {
1217 struct msghdr *msg = from;
1218
1219 err = -EIO;
1220 if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1221 goto error;
1222
1223 err = skb_splice_from_iter(skb, iter: &msg->msg_iter, maxsize: copy,
1224 gfp: sk->sk_allocation);
1225 if (err < 0)
1226 goto error;
1227 copy = err;
1228 wmem_alloc_delta += copy;
1229 } else if (!zc) {
1230 int i = skb_shinfo(skb)->nr_frags;
1231
1232 err = -ENOMEM;
1233 if (!sk_page_frag_refill(sk, pfrag))
1234 goto error;
1235
1236 skb_zcopy_downgrade_managed(skb);
1237 if (!skb_can_coalesce(skb, i, page: pfrag->page,
1238 off: pfrag->offset)) {
1239 err = -EMSGSIZE;
1240 if (i == MAX_SKB_FRAGS)
1241 goto error;
1242
1243 __skb_fill_page_desc(skb, i, page: pfrag->page,
1244 off: pfrag->offset, size: 0);
1245 skb_shinfo(skb)->nr_frags = ++i;
1246 get_page(page: pfrag->page);
1247 }
1248 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1249 if (getfrag(from,
1250 page_address(pfrag->page) + pfrag->offset,
1251 offset, copy, skb->len, skb) < 0)
1252 goto error_efault;
1253
1254 pfrag->offset += copy;
1255 skb_frag_size_add(frag: &skb_shinfo(skb)->frags[i - 1], delta: copy);
1256 skb_len_add(skb, delta: copy);
1257 wmem_alloc_delta += copy;
1258 } else {
1259 err = skb_zerocopy_iter_dgram(skb, msg: from, len: copy);
1260 if (err < 0)
1261 goto error;
1262 }
1263 offset += copy;
1264 length -= copy;
1265 }
1266
1267 if (wmem_alloc_delta)
1268 refcount_add(i: wmem_alloc_delta, r: &sk->sk_wmem_alloc);
1269 return 0;
1270
1271error_efault:
1272 err = -EFAULT;
1273error:
1274 net_zcopy_put_abort(uarg, have_uref: extra_uref);
1275 cork->length -= length;
1276 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1277 refcount_add(i: wmem_alloc_delta, r: &sk->sk_wmem_alloc);
1278 if (hold_tskey)
1279 atomic_dec(v: &sk->sk_tskey);
1280 return err;
1281}
1282
1283static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1284 struct ipcm_cookie *ipc, struct rtable **rtp)
1285{
1286 struct ip_options_rcu *opt;
1287 struct rtable *rt;
1288
1289 rt = *rtp;
1290 if (unlikely(!rt))
1291 return -EFAULT;
1292
1293 cork->fragsize = ip_sk_use_pmtu(sk) ?
1294 dst_mtu(dst: &rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1295
1296 if (!inetdev_valid_mtu(mtu: cork->fragsize))
1297 return -ENETUNREACH;
1298
1299 /*
1300 * setup for corking.
1301 */
1302 opt = ipc->opt;
1303 if (opt) {
1304 if (!cork->opt) {
1305 cork->opt = kmalloc(size: sizeof(struct ip_options) + 40,
1306 flags: sk->sk_allocation);
1307 if (unlikely(!cork->opt))
1308 return -ENOBUFS;
1309 }
1310 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1311 cork->flags |= IPCORK_OPT;
1312 cork->addr = ipc->addr;
1313 }
1314
1315 cork->gso_size = ipc->gso_size;
1316
1317 cork->dst = &rt->dst;
1318 /* We stole this route, caller should not release it. */
1319 *rtp = NULL;
1320
1321 cork->length = 0;
1322 cork->ttl = ipc->ttl;
1323 cork->tos = ipc->tos;
1324 cork->mark = ipc->sockc.mark;
1325 cork->priority = ipc->priority;
1326 cork->transmit_time = ipc->sockc.transmit_time;
1327 cork->tx_flags = 0;
1328 sock_tx_timestamp(sk, tsflags: ipc->sockc.tsflags, tx_flags: &cork->tx_flags);
1329
1330 return 0;
1331}
1332
1333/*
1334 * ip_append_data() can make one large IP datagram from many pieces of
1335 * data. Each piece will be held on the socket until
1336 * ip_push_pending_frames() is called. Each piece can be a page or
1337 * non-page data.
1338 *
1339 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1340 * this interface potentially.
1341 *
1342 * LATER: length must be adjusted by pad at tail, when it is required.
1343 */
1344int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1345 int getfrag(void *from, char *to, int offset, int len,
1346 int odd, struct sk_buff *skb),
1347 void *from, int length, int transhdrlen,
1348 struct ipcm_cookie *ipc, struct rtable **rtp,
1349 unsigned int flags)
1350{
1351 struct inet_sock *inet = inet_sk(sk);
1352 int err;
1353
1354 if (flags&MSG_PROBE)
1355 return 0;
1356
1357 if (skb_queue_empty(list: &sk->sk_write_queue)) {
1358 err = ip_setup_cork(sk, cork: &inet->cork.base, ipc, rtp);
1359 if (err)
1360 return err;
1361 } else {
1362 transhdrlen = 0;
1363 }
1364
1365 return __ip_append_data(sk, fl4, queue: &sk->sk_write_queue, cork: &inet->cork.base,
1366 pfrag: sk_page_frag(sk), getfrag,
1367 from, length, transhdrlen, flags);
1368}
1369
1370static void ip_cork_release(struct inet_cork *cork)
1371{
1372 cork->flags &= ~IPCORK_OPT;
1373 kfree(objp: cork->opt);
1374 cork->opt = NULL;
1375 dst_release(dst: cork->dst);
1376 cork->dst = NULL;
1377}
1378
1379/*
1380 * Combined all pending IP fragments on the socket as one IP datagram
1381 * and push them out.
1382 */
1383struct sk_buff *__ip_make_skb(struct sock *sk,
1384 struct flowi4 *fl4,
1385 struct sk_buff_head *queue,
1386 struct inet_cork *cork)
1387{
1388 struct sk_buff *skb, *tmp_skb;
1389 struct sk_buff **tail_skb;
1390 struct inet_sock *inet = inet_sk(sk);
1391 struct net *net = sock_net(sk);
1392 struct ip_options *opt = NULL;
1393 struct rtable *rt = (struct rtable *)cork->dst;
1394 struct iphdr *iph;
1395 u8 pmtudisc, ttl;
1396 __be16 df = 0;
1397
1398 skb = __skb_dequeue(list: queue);
1399 if (!skb)
1400 goto out;
1401 tail_skb = &(skb_shinfo(skb)->frag_list);
1402
1403 /* move skb->data to ip header from ext header */
1404 if (skb->data < skb_network_header(skb))
1405 __skb_pull(skb, len: skb_network_offset(skb));
1406 while ((tmp_skb = __skb_dequeue(list: queue)) != NULL) {
1407 __skb_pull(skb: tmp_skb, len: skb_network_header_len(skb));
1408 *tail_skb = tmp_skb;
1409 tail_skb = &(tmp_skb->next);
1410 skb->len += tmp_skb->len;
1411 skb->data_len += tmp_skb->len;
1412 skb->truesize += tmp_skb->truesize;
1413 tmp_skb->destructor = NULL;
1414 tmp_skb->sk = NULL;
1415 }
1416
1417 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1418 * to fragment the frame generated here. No matter, what transforms
1419 * how transforms change size of the packet, it will come out.
1420 */
1421 skb->ignore_df = ip_sk_ignore_df(sk);
1422
1423 /* DF bit is set when we want to see DF on outgoing frames.
1424 * If ignore_df is set too, we still allow to fragment this frame
1425 * locally. */
1426 pmtudisc = READ_ONCE(inet->pmtudisc);
1427 if (pmtudisc == IP_PMTUDISC_DO ||
1428 pmtudisc == IP_PMTUDISC_PROBE ||
1429 (skb->len <= dst_mtu(dst: &rt->dst) &&
1430 ip_dont_fragment(sk, dst: &rt->dst)))
1431 df = htons(IP_DF);
1432
1433 if (cork->flags & IPCORK_OPT)
1434 opt = cork->opt;
1435
1436 if (cork->ttl != 0)
1437 ttl = cork->ttl;
1438 else if (rt->rt_type == RTN_MULTICAST)
1439 ttl = READ_ONCE(inet->mc_ttl);
1440 else
1441 ttl = ip_select_ttl(inet, dst: &rt->dst);
1442
1443 iph = ip_hdr(skb);
1444 iph->version = 4;
1445 iph->ihl = 5;
1446 iph->tos = (cork->tos != -1) ? cork->tos : READ_ONCE(inet->tos);
1447 iph->frag_off = df;
1448 iph->ttl = ttl;
1449 iph->protocol = sk->sk_protocol;
1450 ip_copy_addrs(iph, fl4);
1451 ip_select_ident(net, skb, sk);
1452
1453 if (opt) {
1454 iph->ihl += opt->optlen >> 2;
1455 ip_options_build(skb, opt, daddr: cork->addr, rt);
1456 }
1457
1458 skb->priority = (cork->tos != -1) ? cork->priority: READ_ONCE(sk->sk_priority);
1459 skb->mark = cork->mark;
1460 skb->tstamp = cork->transmit_time;
1461 /*
1462 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1463 * on dst refcount
1464 */
1465 cork->dst = NULL;
1466 skb_dst_set(skb, dst: &rt->dst);
1467
1468 if (iph->protocol == IPPROTO_ICMP) {
1469 u8 icmp_type;
1470
1471 /* For such sockets, transhdrlen is zero when do ip_append_data(),
1472 * so icmphdr does not in skb linear region and can not get icmp_type
1473 * by icmp_hdr(skb)->type.
1474 */
1475 if (sk->sk_type == SOCK_RAW &&
1476 !inet_test_bit(HDRINCL, sk))
1477 icmp_type = fl4->fl4_icmp_type;
1478 else
1479 icmp_type = icmp_hdr(skb)->type;
1480 icmp_out_count(net, type: icmp_type);
1481 }
1482
1483 ip_cork_release(cork);
1484out:
1485 return skb;
1486}
1487
1488int ip_send_skb(struct net *net, struct sk_buff *skb)
1489{
1490 int err;
1491
1492 err = ip_local_out(net, skb->sk, skb);
1493 if (err) {
1494 if (err > 0)
1495 err = net_xmit_errno(err);
1496 if (err)
1497 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1498 }
1499
1500 return err;
1501}
1502
1503int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1504{
1505 struct sk_buff *skb;
1506
1507 skb = ip_finish_skb(sk, fl4);
1508 if (!skb)
1509 return 0;
1510
1511 /* Netfilter gets whole the not fragmented skb. */
1512 return ip_send_skb(net: sock_net(sk), skb);
1513}
1514
1515/*
1516 * Throw away all pending data on the socket.
1517 */
1518static void __ip_flush_pending_frames(struct sock *sk,
1519 struct sk_buff_head *queue,
1520 struct inet_cork *cork)
1521{
1522 struct sk_buff *skb;
1523
1524 while ((skb = __skb_dequeue_tail(list: queue)) != NULL)
1525 kfree_skb(skb);
1526
1527 ip_cork_release(cork);
1528}
1529
1530void ip_flush_pending_frames(struct sock *sk)
1531{
1532 __ip_flush_pending_frames(sk, queue: &sk->sk_write_queue, cork: &inet_sk(sk)->cork.base);
1533}
1534
1535struct sk_buff *ip_make_skb(struct sock *sk,
1536 struct flowi4 *fl4,
1537 int getfrag(void *from, char *to, int offset,
1538 int len, int odd, struct sk_buff *skb),
1539 void *from, int length, int transhdrlen,
1540 struct ipcm_cookie *ipc, struct rtable **rtp,
1541 struct inet_cork *cork, unsigned int flags)
1542{
1543 struct sk_buff_head queue;
1544 int err;
1545
1546 if (flags & MSG_PROBE)
1547 return NULL;
1548
1549 __skb_queue_head_init(list: &queue);
1550
1551 cork->flags = 0;
1552 cork->addr = 0;
1553 cork->opt = NULL;
1554 err = ip_setup_cork(sk, cork, ipc, rtp);
1555 if (err)
1556 return ERR_PTR(error: err);
1557
1558 err = __ip_append_data(sk, fl4, queue: &queue, cork,
1559 pfrag: &current->task_frag, getfrag,
1560 from, length, transhdrlen, flags);
1561 if (err) {
1562 __ip_flush_pending_frames(sk, queue: &queue, cork);
1563 return ERR_PTR(error: err);
1564 }
1565
1566 return __ip_make_skb(sk, fl4, queue: &queue, cork);
1567}
1568
1569/*
1570 * Fetch data from kernel space and fill in checksum if needed.
1571 */
1572static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1573 int len, int odd, struct sk_buff *skb)
1574{
1575 __wsum csum;
1576
1577 csum = csum_partial_copy_nocheck(src: dptr+offset, dst: to, len);
1578 skb->csum = csum_block_add(csum: skb->csum, csum2: csum, offset: odd);
1579 return 0;
1580}
1581
1582/*
1583 * Generic function to send a packet as reply to another packet.
1584 * Used to send some TCP resets/acks so far.
1585 */
1586void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1587 const struct ip_options *sopt,
1588 __be32 daddr, __be32 saddr,
1589 const struct ip_reply_arg *arg,
1590 unsigned int len, u64 transmit_time, u32 txhash)
1591{
1592 struct ip_options_data replyopts;
1593 struct ipcm_cookie ipc;
1594 struct flowi4 fl4;
1595 struct rtable *rt = skb_rtable(skb);
1596 struct net *net = sock_net(sk);
1597 struct sk_buff *nskb;
1598 int err;
1599 int oif;
1600
1601 if (__ip_options_echo(net, dopt: &replyopts.opt.opt, skb, sopt))
1602 return;
1603
1604 ipcm_init(ipcm: &ipc);
1605 ipc.addr = daddr;
1606 ipc.sockc.transmit_time = transmit_time;
1607
1608 if (replyopts.opt.opt.optlen) {
1609 ipc.opt = &replyopts.opt;
1610
1611 if (replyopts.opt.opt.srr)
1612 daddr = replyopts.opt.opt.faddr;
1613 }
1614
1615 oif = arg->bound_dev_if;
1616 if (!oif && netif_index_is_l3_master(net, ifindex: skb->skb_iif))
1617 oif = skb->skb_iif;
1618
1619 flowi4_init_output(fl4: &fl4, oif,
1620 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1621 RT_TOS(arg->tos),
1622 scope: RT_SCOPE_UNIVERSE, proto: ip_hdr(skb)->protocol,
1623 flags: ip_reply_arg_flowi_flags(arg),
1624 daddr, saddr,
1625 dport: tcp_hdr(skb)->source, sport: tcp_hdr(skb)->dest,
1626 uid: arg->uid);
1627 security_skb_classify_flow(skb, flic: flowi4_to_flowi_common(fl4: &fl4));
1628 rt = ip_route_output_flow(net, flp: &fl4, sk);
1629 if (IS_ERR(ptr: rt))
1630 return;
1631
1632 inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK;
1633
1634 sk->sk_protocol = ip_hdr(skb)->protocol;
1635 sk->sk_bound_dev_if = arg->bound_dev_if;
1636 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1637 ipc.sockc.mark = fl4.flowi4_mark;
1638 err = ip_append_data(sk, fl4: &fl4, getfrag: ip_reply_glue_bits, from: arg->iov->iov_base,
1639 length: len, transhdrlen: 0, ipc: &ipc, rtp: &rt, MSG_DONTWAIT);
1640 if (unlikely(err)) {
1641 ip_flush_pending_frames(sk);
1642 goto out;
1643 }
1644
1645 nskb = skb_peek(list_: &sk->sk_write_queue);
1646 if (nskb) {
1647 if (arg->csumoffset >= 0)
1648 *((__sum16 *)skb_transport_header(skb: nskb) +
1649 arg->csumoffset) = csum_fold(sum: csum_add(csum: nskb->csum,
1650 addend: arg->csum));
1651 nskb->ip_summed = CHECKSUM_NONE;
1652 nskb->mono_delivery_time = !!transmit_time;
1653 if (txhash)
1654 skb_set_hash(skb: nskb, hash: txhash, type: PKT_HASH_TYPE_L4);
1655 ip_push_pending_frames(sk, fl4: &fl4);
1656 }
1657out:
1658 ip_rt_put(rt);
1659}
1660
1661void __init ip_init(void)
1662{
1663 ip_rt_init();
1664 inet_initpeers();
1665
1666#if defined(CONFIG_IP_MULTICAST)
1667 igmp_mc_init();
1668#endif
1669}
1670

source code of linux/net/ipv4/ip_output.c