projects/linux·blob master · refs

1699 lines · 42.9 KiB

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, <[email protected]>
11 * Donald Becker, <[email protected]>
12 * Alan Cox, <[email protected]>
13 * Richard Underwood
14 * Stefan Becker, <[email protected]>
15 * Jorge Cwik, <[email protected]>
16 * Arnt Gulbrandsen, <[email protected]>
17 * Hirokazu Takahashi, <[email protected]>
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/flow.h>
67#include <net/snmp.h>
68#include <net/ip.h>
69#include <net/protocol.h>
70#include <net/route.h>
71#include <net/xfrm.h>
72#include <linux/skbuff.h>
73#include <net/sock.h>
74#include <net/arp.h>
75#include <net/icmp.h>
76#include <net/checksum.h>
77#include <net/gso.h>
78#include <net/inetpeer.h>
79#include <net/lwtunnel.h>
80#include <net/inet_dscp.h>
81#include <linux/bpf-cgroup.h>
82#include <linux/igmp.h>
83#include <linux/netfilter_ipv4.h>
84#include <linux/netfilter_bridge.h>
85#include <linux/netlink.h>
86#include <linux/tcp.h>
87#include <net/psp.h>
88
89static int
90ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
91 unsigned int mtu,
92 int (*output)(struct net *, struct sock *, struct sk_buff *));
93
94/* Generate a checksum for an outgoing IP datagram. */
95void ip_send_check(struct iphdr *iph)
96{
97 iph->check = 0;
98 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
99}
100EXPORT_SYMBOL(ip_send_check);
101
102int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
103{
104 struct iphdr *iph = ip_hdr(skb);
105
106 IP_INC_STATS(net, IPSTATS_MIB_OUTREQUESTS);
107
108 iph_set_totlen(iph, skb->len);
109 ip_send_check(iph);
110
111 /* if egress device is enslaved to an L3 master device pass the
112 * skb to its handler for processing
113 */
114 skb = l3mdev_ip_out(sk, skb);
115 if (unlikely(!skb))
116 return 0;
117
118 skb->protocol = htons(ETH_P_IP);
119
120 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
121 net, sk, skb, NULL, skb_dst_dev(skb),
122 dst_output);
123}
124
125int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
126{
127 int err;
128
129 err = __ip_local_out(net, sk, skb);
130 if (likely(err == 1))
131 err = dst_output(net, sk, skb);
132
133 return err;
134}
135EXPORT_SYMBOL_GPL(ip_local_out);
136
137static inline int ip_select_ttl(const struct inet_sock *inet,
138 const struct dst_entry *dst)
139{
140 int ttl = READ_ONCE(inet->uc_ttl);
141
142 if (ttl < 0)
143 ttl = ip4_dst_hoplimit(dst);
144 return ttl;
145}
146
147/*
148 * Add an ip header to a skbuff and send it out.
149 *
150 */
151int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
152 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
153 u8 tos)
154{
155 const struct inet_sock *inet = inet_sk(sk);
156 struct rtable *rt = skb_rtable(skb);
157 struct net *net = sock_net(sk);
158 struct iphdr *iph;
159
160 /* Build the IP header. */
161 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
162 skb_reset_network_header(skb);
163 iph = ip_hdr(skb);
164 iph->version = 4;
165 iph->ihl = 5;
166 iph->tos = tos;
167 iph->ttl = ip_select_ttl(inet, &rt->dst);
168 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
169 iph->saddr = saddr;
170 iph->protocol = sk->sk_protocol;
171 /* Do not bother generating IPID for small packets (eg SYNACK) */
172 if (skb->len <= IPV4_MIN_MTU || ip_dont_fragment(sk, &rt->dst)) {
173 iph->frag_off = htons(IP_DF);
174 iph->id = 0;
175 } else {
176 iph->frag_off = 0;
177 /* TCP packets here are SYNACK with fat IPv4/TCP options.
178 * Avoid using the hashed IP ident generator.
179 */
180 if (sk->sk_protocol == IPPROTO_TCP)
181 iph->id = (__force __be16)get_random_u16();
182 else
183 __ip_select_ident(net, iph, 1);
184 }
185
186 if (opt && opt->opt.optlen) {
187 iph->ihl += opt->opt.optlen>>2;
188 ip_options_build(skb, &opt->opt, daddr, rt);
189 }
190
191 skb->priority = READ_ONCE(sk->sk_priority);
192 if (!skb->mark)
193 skb->mark = READ_ONCE(sk->sk_mark);
194
195 /* Send it out. */
196 return ip_local_out(net, skb->sk, skb);
197}
198
199static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
200{
201 struct dst_entry *dst = skb_dst(skb);
202 struct rtable *rt = dst_rtable(dst);
203 struct net_device *dev = dst_dev(dst);
204 unsigned int hh_len = LL_RESERVED_SPACE(dev);
205 struct neighbour *neigh;
206 bool is_v6gw = false;
207
208 if (rt->rt_type == RTN_MULTICAST) {
209 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
210 } else if (rt->rt_type == RTN_BROADCAST)
211 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
212
213 /* OUTOCTETS should be counted after fragment */
214 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
215
216 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
217 skb = skb_expand_head(skb, hh_len);
218 if (!skb)
219 return -ENOMEM;
220 }
221
222 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
223 int res = lwtunnel_xmit(skb);
224
225 if (res != LWTUNNEL_XMIT_CONTINUE)
226 return res;
227 }
228
229 rcu_read_lock();
230 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
231 if (!IS_ERR(neigh)) {
232 int res;
233
234 sock_confirm_neigh(skb, neigh);
235 /* if crossing protocols, can not use the cached header */
236 res = neigh_output(neigh, skb, is_v6gw);
237 rcu_read_unlock();
238 return res;
239 }
240 rcu_read_unlock();
241
242 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
243 __func__);
244 kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
245 return PTR_ERR(neigh);
246}
247
248static int ip_finish_output_gso(struct net *net, struct sock *sk,
249 struct sk_buff *skb, unsigned int mtu)
250{
251 struct sk_buff *segs, *nskb;
252 netdev_features_t features;
253 int ret = 0;
254
255 /* common case: seglen is <= mtu
256 */
257 if (skb_gso_validate_network_len(skb, mtu))
258 return ip_finish_output2(net, sk, skb);
259
260 /* Slowpath - GSO segment length exceeds the egress MTU.
261 *
262 * This can happen in several cases:
263 * - Forwarding of a TCP GRO skb, when DF flag is not set.
264 * - Forwarding of an skb that arrived on a virtualization interface
265 * (virtio-net/vhost/tap) with TSO/GSO size set by other network
266 * stack.
267 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
268 * interface with a smaller MTU.
269 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is
270 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an
271 * insufficient MTU.
272 */
273 features = netif_skb_features(skb);
274 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
275 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
276 if (IS_ERR_OR_NULL(segs)) {
277 kfree_skb(skb);
278 return -ENOMEM;
279 }
280
281 consume_skb(skb);
282
283 skb_list_walk_safe(segs, segs, nskb) {
284 int err;
285
286 skb_mark_not_on_list(segs);
287 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
288
289 if (err && ret == 0)
290 ret = err;
291 }
292
293 return ret;
294}
295
296static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
297{
298 unsigned int mtu;
299
300#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
301 /* Policy lookup after SNAT yielded a new policy */
302 if (skb_dst(skb)->xfrm) {
303 IPCB(skb)->flags |= IPSKB_REROUTED;
304 return dst_output(net, sk, skb);
305 }
306#endif
307 mtu = ip_skb_dst_mtu(sk, skb);
308 if (skb_is_gso(skb))
309 return ip_finish_output_gso(net, sk, skb, mtu);
310
311 if (skb->len > mtu || IPCB(skb)->frag_max_size)
312 return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
313
314 return ip_finish_output2(net, sk, skb);
315}
316
317static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
318{
319 int ret;
320
321 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
322 switch (ret) {
323 case NET_XMIT_SUCCESS:
324 return __ip_finish_output(net, sk, skb);
325 case NET_XMIT_CN:
326 return __ip_finish_output(net, sk, skb) ? : ret;
327 default:
328 kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
329 return ret;
330 }
331}
332
333static int ip_mc_finish_output(struct net *net, struct sock *sk,
334 struct sk_buff *skb)
335{
336 struct rtable *new_rt;
337 bool do_cn = false;
338 int ret, err;
339
340 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
341 switch (ret) {
342 case NET_XMIT_CN:
343 do_cn = true;
344 fallthrough;
345 case NET_XMIT_SUCCESS:
346 break;
347 default:
348 kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
349 return ret;
350 }
351
352 /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
353 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
354 * see ipv4_pktinfo_prepare().
355 */
356 new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
357 if (new_rt) {
358 new_rt->rt_iif = 0;
359 skb_dst_drop(skb);
360 skb_dst_set(skb, &new_rt->dst);
361 }
362
363 err = dev_loopback_xmit(net, sk, skb);
364 return (do_cn && err) ? ret : err;
365}
366
367int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
368{
369 struct rtable *rt = skb_rtable(skb);
370 struct net_device *dev = rt->dst.dev;
371
372 /*
373 * If the indicated interface is up and running, send the packet.
374 */
375 skb->dev = dev;
376 skb->protocol = htons(ETH_P_IP);
377
378 /*
379 * Multicasts are looped back for other local users
380 */
381
382 if (rt->rt_flags&RTCF_MULTICAST) {
383 if (sk_mc_loop(sk)
384#ifdef CONFIG_IP_MROUTE
385 /* Small optimization: do not loopback not local frames,
386 which returned after forwarding; they will be dropped
387 by ip_mr_input in any case.
388 Note, that local frames are looped back to be delivered
389 to local recipients.
390
391 This check is duplicated in ip_mr_input at the moment.
392 */
393 &&
394 ((rt->rt_flags & RTCF_LOCAL) ||
395 !(IPCB(skb)->flags & IPSKB_FORWARDED))
396#endif
397 ) {
398 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
399 if (newskb)
400 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
401 net, sk, newskb, NULL, newskb->dev,
402 ip_mc_finish_output);
403 }
404
405 /* Multicasts with ttl 0 must not go beyond the host */
406
407 if (ip_hdr(skb)->ttl == 0) {
408 kfree_skb(skb);
409 return 0;
410 }
411 }
412
413 if (rt->rt_flags&RTCF_BROADCAST) {
414 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
415 if (newskb)
416 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
417 net, sk, newskb, NULL, newskb->dev,
418 ip_mc_finish_output);
419 }
420
421 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
422 net, sk, skb, NULL, skb->dev,
423 ip_finish_output,
424 !(IPCB(skb)->flags & IPSKB_REROUTED));
425}
426
427int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
428{
429 struct net_device *dev, *indev = skb->dev;
430 int ret_val;
431
432 rcu_read_lock();
433 dev = skb_dst_dev_rcu(skb);
434 skb->dev = dev;
435 skb->protocol = htons(ETH_P_IP);
436
437 ret_val = NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
438 net, sk, skb, indev, dev,
439 ip_finish_output,
440 !(IPCB(skb)->flags & IPSKB_REROUTED));
441 rcu_read_unlock();
442 return ret_val;
443}
444EXPORT_SYMBOL(ip_output);
445
446/*
447 * copy saddr and daddr, possibly using 64bit load/stores
448 * Equivalent to :
449 * iph->saddr = fl4->saddr;
450 * iph->daddr = fl4->daddr;
451 */
452static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
453{
454 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
455 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
456
457 iph->saddr = fl4->saddr;
458 iph->daddr = fl4->daddr;
459}
460
461/* Note: skb->sk can be different from sk, in case of tunnels */
462int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
463 __u8 tos)
464{
465 struct inet_sock *inet = inet_sk(sk);
466 struct net *net = sock_net(sk);
467 struct ip_options_rcu *inet_opt;
468 struct flowi4 *fl4;
469 struct rtable *rt;
470 struct iphdr *iph;
471 int res;
472
473 /* Skip all of this if the packet is already routed,
474 * f.e. by something like SCTP.
475 */
476 rcu_read_lock();
477 inet_opt = rcu_dereference(inet->inet_opt);
478 fl4 = &fl->u.ip4;
479 rt = skb_rtable(skb);
480 if (rt)
481 goto packet_routed;
482
483 /* Make sure we can route this packet. */
484 rt = dst_rtable(__sk_dst_check(sk, 0));
485 if (!rt) {
486 inet_sk_init_flowi4(inet, fl4);
487
488 /* sctp_v4_xmit() uses its own DSCP value */
489 fl4->flowi4_dscp = inet_dsfield_to_dscp(tos);
490
491 /* If this fails, retransmit mechanism of transport layer will
492 * keep trying until route appears or the connection times
493 * itself out.
494 */
495 rt = ip_route_output_flow(net, fl4, sk);
496 if (IS_ERR(rt))
497 goto no_route;
498 sk_setup_caps(sk, &rt->dst);
499 }
500 skb_dst_set_noref(skb, &rt->dst);
501
502packet_routed:
503 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
504 goto no_route;
505
506 /* OK, we know where to send it, allocate and build IP header. */
507 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
508 skb_reset_network_header(skb);
509 iph = ip_hdr(skb);
510 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
511 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
512 iph->frag_off = htons(IP_DF);
513 else
514 iph->frag_off = 0;
515 iph->ttl = ip_select_ttl(inet, &rt->dst);
516 iph->protocol = sk->sk_protocol;
517 ip_copy_addrs(iph, fl4);
518
519 /* Transport layer set skb->h.foo itself. */
520
521 if (inet_opt && inet_opt->opt.optlen) {
522 iph->ihl += inet_opt->opt.optlen >> 2;
523 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt);
524 }
525
526 ip_select_ident_segs(net, skb, sk,
527 skb_shinfo(skb)->gso_segs ?: 1);
528
529 /* TODO : should we use skb->sk here instead of sk ? */
530 skb->priority = READ_ONCE(sk->sk_priority);
531 skb->mark = READ_ONCE(sk->sk_mark);
532
533 res = ip_local_out(net, sk, skb);
534 rcu_read_unlock();
535 return res;
536
537no_route:
538 rcu_read_unlock();
539 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
540 kfree_skb_reason(skb, SKB_DROP_REASON_IP_OUTNOROUTES);
541 return -EHOSTUNREACH;
542}
543EXPORT_SYMBOL(__ip_queue_xmit);
544
545int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
546{
547 return __ip_queue_xmit(sk, skb, fl, READ_ONCE(inet_sk(sk)->tos));
548}
549EXPORT_SYMBOL(ip_queue_xmit);
550
551static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
552{
553 to->pkt_type = from->pkt_type;
554 to->priority = from->priority;
555 to->protocol = from->protocol;
556 to->skb_iif = from->skb_iif;
557 skb_dst_drop(to);
558 skb_dst_copy(to, from);
559 to->dev = from->dev;
560 to->mark = from->mark;
561
562 skb_copy_hash(to, from);
563
564#ifdef CONFIG_NET_SCHED
565 to->tc_index = from->tc_index;
566#endif
567 nf_copy(to, from);
568 skb_ext_copy(to, from);
569#if IS_ENABLED(CONFIG_IP_VS)
570 to->ipvs_property = from->ipvs_property;
571#endif
572 skb_copy_secmark(to, from);
573}
574
575static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
576 unsigned int mtu,
577 int (*output)(struct net *, struct sock *, struct sk_buff *))
578{
579 struct iphdr *iph = ip_hdr(skb);
580
581 if ((iph->frag_off & htons(IP_DF)) == 0)
582 return ip_do_fragment(net, sk, skb, output);
583
584 if (unlikely(!skb->ignore_df ||
585 (IPCB(skb)->frag_max_size &&
586 IPCB(skb)->frag_max_size > mtu))) {
587 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
588 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
589 htonl(mtu));
590 kfree_skb(skb);
591 return -EMSGSIZE;
592 }
593
594 return ip_do_fragment(net, sk, skb, output);
595}
596
597void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
598 unsigned int hlen, struct ip_fraglist_iter *iter)
599{
600 unsigned int first_len = skb_pagelen(skb);
601
602 iter->frag = skb_shinfo(skb)->frag_list;
603 skb_frag_list_init(skb);
604
605 iter->offset = 0;
606 iter->iph = iph;
607 iter->hlen = hlen;
608
609 skb->data_len = first_len - skb_headlen(skb);
610 skb->len = first_len;
611 iph->tot_len = htons(first_len);
612 iph->frag_off = htons(IP_MF);
613 ip_send_check(iph);
614}
615EXPORT_SYMBOL(ip_fraglist_init);
616
617void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
618{
619 unsigned int hlen = iter->hlen;
620 struct iphdr *iph = iter->iph;
621 struct sk_buff *frag;
622
623 frag = iter->frag;
624 frag->ip_summed = CHECKSUM_NONE;
625 skb_reset_transport_header(frag);
626 __skb_push(frag, hlen);
627 skb_reset_network_header(frag);
628 memcpy(skb_network_header(frag), iph, hlen);
629 iter->iph = ip_hdr(frag);
630 iph = iter->iph;
631 iph->tot_len = htons(frag->len);
632 ip_copy_metadata(frag, skb);
633 iter->offset += skb->len - hlen;
634 iph->frag_off = htons(iter->offset >> 3);
635 if (frag->next)
636 iph->frag_off |= htons(IP_MF);
637 /* Ready, complete checksum */
638 ip_send_check(iph);
639}
640EXPORT_SYMBOL(ip_fraglist_prepare);
641
642void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
643 unsigned int ll_rs, unsigned int mtu, bool DF,
644 struct ip_frag_state *state)
645{
646 struct iphdr *iph = ip_hdr(skb);
647
648 state->DF = DF;
649 state->hlen = hlen;
650 state->ll_rs = ll_rs;
651 state->mtu = mtu;
652
653 state->left = skb->len - hlen; /* Space per frame */
654 state->ptr = hlen; /* Where to start from */
655
656 state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
657 state->not_last_frag = iph->frag_off & htons(IP_MF);
658}
659EXPORT_SYMBOL(ip_frag_init);
660
661static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
662 bool first_frag)
663{
664 /* Copy the flags to each fragment. */
665 IPCB(to)->flags = IPCB(from)->flags;
666
667 /* ANK: dirty, but effective trick. Upgrade options only if
668 * the segment to be fragmented was THE FIRST (otherwise,
669 * options are already fixed) and make it ONCE
670 * on the initial skb, so that all the following fragments
671 * will inherit fixed options.
672 */
673 if (first_frag)
674 ip_options_fragment(from);
675}
676
677struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
678{
679 unsigned int len = state->left;
680 struct sk_buff *skb2;
681 struct iphdr *iph;
682
683 /* IF: it doesn't fit, use 'mtu' - the data space left */
684 if (len > state->mtu)
685 len = state->mtu;
686 /* IF: we are not sending up to and including the packet end
687 then align the next start on an eight byte boundary */
688 if (len < state->left) {
689 len &= ~7;
690 }
691
692 /* Allocate buffer */
693 skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
694 if (!skb2)
695 return ERR_PTR(-ENOMEM);
696
697 /*
698 * Set up data on packet
699 */
700
701 ip_copy_metadata(skb2, skb);
702 skb_reserve(skb2, state->ll_rs);
703 skb_put(skb2, len + state->hlen);
704 skb_reset_network_header(skb2);
705 skb2->transport_header = skb2->network_header + state->hlen;
706
707 /*
708 * Charge the memory for the fragment to any owner
709 * it might possess
710 */
711
712 if (skb->sk)
713 skb_set_owner_w(skb2, skb->sk);
714
715 /*
716 * Copy the packet header into the new buffer.
717 */
718
719 skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
720
721 /*
722 * Copy a block of the IP datagram.
723 */
724 if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
725 BUG();
726 state->left -= len;
727
728 /*
729 * Fill in the new header fields.
730 */
731 iph = ip_hdr(skb2);
732 iph->frag_off = htons((state->offset >> 3));
733 if (state->DF)
734 iph->frag_off |= htons(IP_DF);
735
736 /*
737 * Added AC : If we are fragmenting a fragment that's not the
738 * last fragment then keep MF on each bit
739 */
740 if (state->left > 0 || state->not_last_frag)
741 iph->frag_off |= htons(IP_MF);
742 state->ptr += len;
743 state->offset += len;
744
745 iph->tot_len = htons(len + state->hlen);
746
747 ip_send_check(iph);
748
749 return skb2;
750}
751EXPORT_SYMBOL(ip_frag_next);
752
753/*
754 * This IP datagram is too large to be sent in one piece. Break it up into
755 * smaller pieces (each of size equal to IP header plus
756 * a block of the data of the original IP data part) that will yet fit in a
757 * single device frame, and queue such a frame for sending.
758 */
759
760int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
761 int (*output)(struct net *, struct sock *, struct sk_buff *))
762{
763 struct iphdr *iph;
764 struct sk_buff *skb2;
765 u8 tstamp_type = skb->tstamp_type;
766 struct rtable *rt = skb_rtable(skb);
767 unsigned int mtu, hlen, ll_rs;
768 struct ip_fraglist_iter iter;
769 ktime_t tstamp = skb->tstamp;
770 struct ip_frag_state state;
771 int err = 0;
772
773 /* for offloaded checksums cleanup checksum before fragmentation */
774 if (skb->ip_summed == CHECKSUM_PARTIAL) {
775 err = ip_check_csum_start(skb);
776 if (!err)
777 err = skb_checksum_help(skb);
778 if (err)
779 goto fail;
780 }
781
782 /*
783 * Point into the IP datagram header.
784 */
785
786 iph = ip_hdr(skb);
787
788 mtu = ip_skb_dst_mtu(sk, skb);
789 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
790 mtu = IPCB(skb)->frag_max_size;
791
792 /*
793 * Setup starting values.
794 */
795
796 hlen = iph->ihl * 4;
797 if (mtu < hlen + 8) {
798 err = -EMSGSIZE;
799 goto fail;
800 }
801 mtu = mtu - hlen; /* Size of data space */
802 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
803 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
804
805 /* When frag_list is given, use it. First, check its validity:
806 * some transformers could create wrong frag_list or break existing
807 * one, it is not prohibited. In this case fall back to copying.
808 *
809 * LATER: this step can be merged to real generation of fragments,
810 * we can switch to copy when see the first bad fragment.
811 */
812 if (skb_has_frag_list(skb)) {
813 struct sk_buff *frag, *frag2;
814 unsigned int first_len = skb_pagelen(skb);
815
816 if (first_len - hlen > mtu ||
817 ((first_len - hlen) & 7) ||
818 ip_is_fragment(iph) ||
819 skb_cloned(skb) ||
820 skb_headroom(skb) < ll_rs)
821 goto slow_path;
822
823 skb_walk_frags(skb, frag) {
824 /* Correct geometry. */
825 if (frag->len > mtu ||
826 ((frag->len & 7) && frag->next) ||
827 skb_headroom(frag) < hlen + ll_rs)
828 goto slow_path_clean;
829
830 /* Partially cloned skb? */
831 if (skb_shared(frag))
832 goto slow_path_clean;
833
834 BUG_ON(frag->sk);
835 if (skb->sk) {
836 frag->sk = skb->sk;
837 frag->destructor = sock_wfree;
838 }
839 skb->truesize -= frag->truesize;
840 }
841
842 /* Everything is OK. Generate! */
843 ip_fraglist_init(skb, iph, hlen, &iter);
844
845 for (;;) {
846 /* Prepare header of the next frame,
847 * before previous one went down. */
848 if (iter.frag) {
849 bool first_frag = (iter.offset == 0);
850
851 IPCB(iter.frag)->flags = IPCB(skb)->flags;
852 ip_fraglist_prepare(skb, &iter);
853 if (first_frag && IPCB(skb)->opt.optlen) {
854 /* ipcb->opt is not populated for frags
855 * coming from __ip_make_skb(),
856 * ip_options_fragment() needs optlen
857 */
858 IPCB(iter.frag)->opt.optlen =
859 IPCB(skb)->opt.optlen;
860 ip_options_fragment(iter.frag);
861 ip_send_check(iter.iph);
862 }
863 }
864
865 skb_set_delivery_time(skb, tstamp, tstamp_type);
866 err = output(net, sk, skb);
867
868 if (!err)
869 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
870 if (err || !iter.frag)
871 break;
872
873 skb = ip_fraglist_next(&iter);
874 }
875
876 if (err == 0) {
877 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
878 return 0;
879 }
880
881 kfree_skb_list(iter.frag);
882
883 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
884 return err;
885
886slow_path_clean:
887 skb_walk_frags(skb, frag2) {
888 if (frag2 == frag)
889 break;
890 frag2->sk = NULL;
891 frag2->destructor = NULL;
892 skb->truesize += frag2->truesize;
893 }
894 }
895
896slow_path:
897 /*
898 * Fragment the datagram.
899 */
900
901 ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
902 &state);
903
904 /*
905 * Keep copying data until we run out.
906 */
907
908 while (state.left > 0) {
909 bool first_frag = (state.offset == 0);
910
911 skb2 = ip_frag_next(skb, &state);
912 if (IS_ERR(skb2)) {
913 err = PTR_ERR(skb2);
914 goto fail;
915 }
916 ip_frag_ipcb(skb, skb2, first_frag);
917
918 /*
919 * Put this fragment into the sending queue.
920 */
921 skb_set_delivery_time(skb2, tstamp, tstamp_type);
922 err = output(net, sk, skb2);
923 if (err)
924 goto fail;
925
926 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
927 }
928 consume_skb(skb);
929 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
930 return err;
931
932fail:
933 kfree_skb(skb);
934 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
935 return err;
936}
937EXPORT_SYMBOL(ip_do_fragment);
938
939int
940ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
941{
942 struct msghdr *msg = from;
943
944 if (skb->ip_summed == CHECKSUM_PARTIAL) {
945 if (!copy_from_iter_full(to, len, &msg->msg_iter))
946 return -EFAULT;
947 } else {
948 __wsum csum = 0;
949 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
950 return -EFAULT;
951 skb->csum = csum_block_add(skb->csum, csum, odd);
952 }
953 return 0;
954}
955EXPORT_SYMBOL(ip_generic_getfrag);
956
957static int __ip_append_data(struct sock *sk,
958 struct flowi4 *fl4,
959 struct sk_buff_head *queue,
960 struct inet_cork *cork,
961 struct page_frag *pfrag,
962 int getfrag(void *from, char *to, int offset,
963 int len, int odd, struct sk_buff *skb),
964 void *from, int length, int transhdrlen,
965 unsigned int flags)
966{
967 struct inet_sock *inet = inet_sk(sk);
968 struct ubuf_info *uarg = NULL;
969 struct sk_buff *skb;
970 struct ip_options *opt = cork->opt;
971 int hh_len;
972 int exthdrlen;
973 int mtu;
974 int copy;
975 int err;
976 int offset = 0;
977 bool zc = false;
978 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
979 int csummode = CHECKSUM_NONE;
980 struct rtable *rt = dst_rtable(cork->dst);
981 bool paged, hold_tskey = false, extra_uref = false;
982 unsigned int wmem_alloc_delta = 0;
983 u32 tskey = 0;
984
985 skb = skb_peek_tail(queue);
986
987 exthdrlen = !skb ? rt->dst.header_len : 0;
988 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
989 paged = !!cork->gso_size;
990
991 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
992
993 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
994 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
995 maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
996
997 if (cork->length + length > maxnonfragsize - fragheaderlen) {
998 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
999 mtu - (opt ? opt->optlen : 0));
1000 return -EMSGSIZE;
1001 }
1002
1003 /*
1004 * transhdrlen > 0 means that this is the first fragment and we wish
1005 * it won't be fragmented in the future.
1006 */
1007 if (transhdrlen &&
1008 length + fragheaderlen <= mtu &&
1009 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1010 (!(flags & MSG_MORE) || cork->gso_size) &&
1011 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1012 csummode = CHECKSUM_PARTIAL;
1013
1014 if ((flags & MSG_ZEROCOPY) && length) {
1015 struct msghdr *msg = from;
1016
1017 if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1018 if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1019 return -EINVAL;
1020
1021 /* Leave uarg NULL if can't zerocopy, callers should
1022 * be able to handle it.
1023 */
1024 if ((rt->dst.dev->features & NETIF_F_SG) &&
1025 csummode == CHECKSUM_PARTIAL) {
1026 paged = true;
1027 zc = true;
1028 uarg = msg->msg_ubuf;
1029 }
1030 } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1031 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb),
1032 false);
1033 if (!uarg)
1034 return -ENOBUFS;
1035 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1036 if (rt->dst.dev->features & NETIF_F_SG &&
1037 csummode == CHECKSUM_PARTIAL) {
1038 paged = true;
1039 zc = true;
1040 } else {
1041 uarg_to_msgzc(uarg)->zerocopy = 0;
1042 skb_zcopy_set(skb, uarg, &extra_uref);
1043 }
1044 }
1045 } else if ((flags & MSG_SPLICE_PAGES) && length) {
1046 if (inet_test_bit(HDRINCL, sk))
1047 return -EPERM;
1048 if (rt->dst.dev->features & NETIF_F_SG &&
1049 getfrag == ip_generic_getfrag)
1050 /* We need an empty buffer to attach stuff to */
1051 paged = true;
1052 else
1053 flags &= ~MSG_SPLICE_PAGES;
1054 }
1055
1056 cork->length += length;
1057
1058 if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
1059 READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
1060 if (cork->flags & IPCORK_TS_OPT_ID) {
1061 tskey = cork->ts_opt_id;
1062 } else {
1063 tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1064 hold_tskey = true;
1065 }
1066 }
1067
1068 /* So, what's going on in the loop below?
1069 *
1070 * We use calculated fragment length to generate chained skb,
1071 * each of segments is IP fragment ready for sending to network after
1072 * adding appropriate IP header.
1073 */
1074
1075 if (!skb)
1076 goto alloc_new_skb;
1077
1078 while (length > 0) {
1079 /* Check if the remaining data fits into current packet. */
1080 copy = mtu - skb->len;
1081 if (copy < length)
1082 copy = maxfraglen - skb->len;
1083 if (copy <= 0) {
1084 char *data;
1085 unsigned int datalen;
1086 unsigned int fraglen;
1087 unsigned int fraggap;
1088 unsigned int alloclen, alloc_extra;
1089 unsigned int pagedlen;
1090 struct sk_buff *skb_prev;
1091alloc_new_skb:
1092 skb_prev = skb;
1093 if (skb_prev)
1094 fraggap = skb_prev->len - maxfraglen;
1095 else
1096 fraggap = 0;
1097
1098 /*
1099 * If remaining data exceeds the mtu,
1100 * we know we need more fragment(s).
1101 */
1102 datalen = length + fraggap;
1103 if (datalen > mtu - fragheaderlen)
1104 datalen = maxfraglen - fragheaderlen;
1105 fraglen = datalen + fragheaderlen;
1106 pagedlen = 0;
1107
1108 alloc_extra = hh_len + 15;
1109 alloc_extra += exthdrlen;
1110
1111 /* The last fragment gets additional space at tail.
1112 * Note, with MSG_MORE we overallocate on fragments,
1113 * because we have no idea what fragment will be
1114 * the last.
1115 */
1116 if (datalen == length + fraggap)
1117 alloc_extra += rt->dst.trailer_len;
1118
1119 if ((flags & MSG_MORE) &&
1120 !(rt->dst.dev->features&NETIF_F_SG))
1121 alloclen = mtu;
1122 else if (!paged &&
1123 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1124 !(rt->dst.dev->features & NETIF_F_SG)))
1125 alloclen = fraglen;
1126 else {
1127 alloclen = fragheaderlen + transhdrlen + fraggap;
1128 pagedlen = datalen - transhdrlen - fraggap;
1129 }
1130
1131 alloclen += alloc_extra;
1132
1133 if (transhdrlen) {
1134 skb = sock_alloc_send_skb(sk, alloclen,
1135 (flags & MSG_DONTWAIT), &err);
1136 } else {
1137 skb = NULL;
1138 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1139 2 * sk->sk_sndbuf)
1140 skb = alloc_skb(alloclen,
1141 sk->sk_allocation);
1142 if (unlikely(!skb))
1143 err = -ENOBUFS;
1144 }
1145 if (!skb)
1146 goto error;
1147
1148 /*
1149 * Fill in the control structures
1150 */
1151 skb->ip_summed = csummode;
1152 skb->csum = 0;
1153 skb_reserve(skb, hh_len);
1154
1155 /*
1156 * Find where to start putting bytes.
1157 */
1158 data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1159 skb_set_network_header(skb, exthdrlen);
1160 skb->transport_header = (skb->network_header +
1161 fragheaderlen);
1162 data += fragheaderlen + exthdrlen;
1163
1164 if (fraggap) {
1165 skb->csum = skb_copy_and_csum_bits(
1166 skb_prev, maxfraglen,
1167 data + transhdrlen, fraggap);
1168 skb_prev->csum = csum_sub(skb_prev->csum,
1169 skb->csum);
1170 data += fraggap;
1171 pskb_trim_unique(skb_prev, maxfraglen);
1172 }
1173
1174 copy = datalen - transhdrlen - fraggap - pagedlen;
1175 if (copy > 0 &&
1176 INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1177 from, data + transhdrlen, offset,
1178 copy, fraggap, skb) < 0) {
1179 err = -EFAULT;
1180 kfree_skb(skb);
1181 goto error;
1182 } else if (flags & MSG_SPLICE_PAGES) {
1183 copy = 0;
1184 }
1185
1186 offset += copy;
1187 length -= copy + transhdrlen;
1188 transhdrlen = 0;
1189 exthdrlen = 0;
1190 csummode = CHECKSUM_NONE;
1191
1192 /* only the initial fragment is time stamped */
1193 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1194 cork->tx_flags = 0;
1195 skb_shinfo(skb)->tskey = tskey;
1196 tskey = 0;
1197 skb_zcopy_set(skb, uarg, &extra_uref);
1198
1199 if ((flags & MSG_CONFIRM) && !skb_prev)
1200 skb_set_dst_pending_confirm(skb, 1);
1201
1202 /*
1203 * Put the packet on the pending queue.
1204 */
1205 if (!skb->destructor) {
1206 skb->destructor = sock_wfree;
1207 skb->sk = sk;
1208 wmem_alloc_delta += skb->truesize;
1209 }
1210 __skb_queue_tail(queue, skb);
1211 continue;
1212 }
1213
1214 if (copy > length)
1215 copy = length;
1216
1217 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1218 skb_tailroom(skb) >= copy) {
1219 unsigned int off;
1220
1221 off = skb->len;
1222 if (INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1223 from, skb_put(skb, copy),
1224 offset, copy, off, skb) < 0) {
1225 __skb_trim(skb, off);
1226 err = -EFAULT;
1227 goto error;
1228 }
1229 } else if (flags & MSG_SPLICE_PAGES) {
1230 struct msghdr *msg = from;
1231
1232 err = -EIO;
1233 if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1234 goto error;
1235
1236 err = skb_splice_from_iter(skb, &msg->msg_iter, copy);
1237 if (err < 0)
1238 goto error;
1239 copy = err;
1240 if (!(flags & MSG_NO_SHARED_FRAGS))
1241 skb_shinfo(skb)->flags |= SKBFL_SHARED_FRAG;
1242 wmem_alloc_delta += copy;
1243 } else if (!zc) {
1244 int i = skb_shinfo(skb)->nr_frags;
1245
1246 err = -ENOMEM;
1247 if (!sk_page_frag_refill(sk, pfrag))
1248 goto error;
1249
1250 skb_zcopy_downgrade_managed(skb);
1251 if (!skb_can_coalesce(skb, i, pfrag->page,
1252 pfrag->offset)) {
1253 err = -EMSGSIZE;
1254 if (i == MAX_SKB_FRAGS)
1255 goto error;
1256
1257 __skb_fill_page_desc(skb, i, pfrag->page,
1258 pfrag->offset, 0);
1259 skb_shinfo(skb)->nr_frags = ++i;
1260 get_page(pfrag->page);
1261 }
1262 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1263 if (INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1264 from,
1265 page_address(pfrag->page) + pfrag->offset,
1266 offset, copy, skb->len, skb) < 0)
1267 goto error_efault;
1268
1269 pfrag->offset += copy;
1270 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1271 skb_len_add(skb, copy);
1272 wmem_alloc_delta += copy;
1273 } else {
1274 err = skb_zerocopy_iter_dgram(skb, from, copy);
1275 if (err < 0)
1276 goto error;
1277 }
1278 offset += copy;
1279 length -= copy;
1280 }
1281
1282 if (wmem_alloc_delta)
1283 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1284 return 0;
1285
1286error_efault:
1287 err = -EFAULT;
1288error:
1289 net_zcopy_put_abort(uarg, extra_uref);
1290 cork->length -= length;
1291 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1292 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1293 if (hold_tskey)
1294 atomic_dec(&sk->sk_tskey);
1295 return err;
1296}
1297
1298static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1299 struct ipcm_cookie *ipc, struct rtable **rtp)
1300{
1301 struct ip_options_rcu *opt;
1302 struct rtable *rt;
1303
1304 rt = *rtp;
1305 if (unlikely(!rt))
1306 return -EFAULT;
1307
1308 cork->fragsize = ip_sk_use_pmtu(sk) ?
1309 dst4_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1310 cork->fragsize = min(cork->fragsize, IP_MAX_MTU);
1311
1312 if (!inetdev_valid_mtu(cork->fragsize))
1313 return -ENETUNREACH;
1314
1315 /*
1316 * setup for corking.
1317 */
1318 opt = ipc->opt;
1319 if (opt) {
1320 if (!cork->opt) {
1321 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1322 sk->sk_allocation);
1323 if (unlikely(!cork->opt))
1324 return -ENOBUFS;
1325 }
1326 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1327 cork->flags |= IPCORK_OPT;
1328 cork->addr = ipc->addr;
1329 }
1330
1331 cork->gso_size = ipc->gso_size;
1332
1333 cork->dst = &rt->dst;
1334 /* We stole this route, caller should not release it. */
1335 *rtp = NULL;
1336
1337 cork->length = 0;
1338 cork->ttl = ipc->ttl;
1339 cork->tos = ipc->tos;
1340 cork->mark = ipc->sockc.mark;
1341 cork->priority = ipc->sockc.priority;
1342 cork->transmit_time = ipc->sockc.transmit_time;
1343 cork->tx_flags = 0;
1344 sock_tx_timestamp(sk, &ipc->sockc, &cork->tx_flags);
1345 if (ipc->sockc.tsflags & SOCKCM_FLAG_TS_OPT_ID) {
1346 cork->flags |= IPCORK_TS_OPT_ID;
1347 cork->ts_opt_id = ipc->sockc.ts_opt_id;
1348 }
1349
1350 return 0;
1351}
1352
1353/*
1354 * ip_append_data() can make one large IP datagram from many pieces of
1355 * data. Each piece will be held on the socket until
1356 * ip_push_pending_frames() is called. Each piece can be a page or
1357 * non-page data.
1358 *
1359 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1360 * this interface potentially.
1361 *
1362 * LATER: length must be adjusted by pad at tail, when it is required.
1363 */
1364int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1365 int getfrag(void *from, char *to, int offset, int len,
1366 int odd, struct sk_buff *skb),
1367 void *from, int length, int transhdrlen,
1368 struct ipcm_cookie *ipc, struct rtable **rtp,
1369 unsigned int flags)
1370{
1371 struct inet_sock *inet = inet_sk(sk);
1372 int err;
1373
1374 if (flags&MSG_PROBE)
1375 return 0;
1376
1377 if (skb_queue_empty(&sk->sk_write_queue)) {
1378 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1379 if (err)
1380 return err;
1381 } else {
1382 transhdrlen = 0;
1383 }
1384
1385 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1386 sk_page_frag(sk), getfrag,
1387 from, length, transhdrlen, flags);
1388}
1389
1390static void ip_cork_release(struct inet_cork *cork)
1391{
1392 cork->flags &= ~IPCORK_OPT;
1393 kfree(cork->opt);
1394 cork->opt = NULL;
1395 dst_release(cork->dst);
1396 cork->dst = NULL;
1397}
1398
1399/*
1400 * Combined all pending IP fragments on the socket as one IP datagram
1401 * and push them out.
1402 */
1403struct sk_buff *__ip_make_skb(struct sock *sk,
1404 struct flowi4 *fl4,
1405 struct sk_buff_head *queue,
1406 struct inet_cork *cork)
1407{
1408 struct sk_buff *skb, *tmp_skb;
1409 struct sk_buff **tail_skb;
1410 struct inet_sock *inet = inet_sk(sk);
1411 struct net *net = sock_net(sk);
1412 struct ip_options *opt = NULL;
1413 struct rtable *rt = dst_rtable(cork->dst);
1414 struct iphdr *iph;
1415 u8 pmtudisc, ttl;
1416 __be16 df = 0;
1417
1418 skb = __skb_dequeue(queue);
1419 if (!skb)
1420 goto out;
1421 tail_skb = &(skb_shinfo(skb)->frag_list);
1422
1423 /* move skb->data to ip header from ext header */
1424 if (skb->data < skb_network_header(skb))
1425 __skb_pull(skb, skb_network_offset(skb));
1426 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1427 __skb_pull(tmp_skb, skb_network_header_len(skb));
1428 *tail_skb = tmp_skb;
1429 tail_skb = &(tmp_skb->next);
1430 skb->len += tmp_skb->len;
1431 skb->data_len += tmp_skb->len;
1432 skb->truesize += tmp_skb->truesize;
1433 tmp_skb->destructor = NULL;
1434 tmp_skb->sk = NULL;
1435 }
1436
1437 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1438 * to fragment the frame generated here. No matter, what transforms
1439 * how transforms change size of the packet, it will come out.
1440 */
1441 skb->ignore_df = ip_sk_ignore_df(sk);
1442
1443 /* DF bit is set when we want to see DF on outgoing frames.
1444 * If ignore_df is set too, we still allow to fragment this frame
1445 * locally. */
1446 pmtudisc = READ_ONCE(inet->pmtudisc);
1447 if (pmtudisc == IP_PMTUDISC_DO ||
1448 pmtudisc == IP_PMTUDISC_PROBE ||
1449 (skb->len <= dst4_mtu(&rt->dst) &&
1450 ip_dont_fragment(sk, &rt->dst)))
1451 df = htons(IP_DF);
1452
1453 if (cork->flags & IPCORK_OPT)
1454 opt = cork->opt;
1455
1456 if (cork->ttl != 0)
1457 ttl = cork->ttl;
1458 else if (rt->rt_type == RTN_MULTICAST)
1459 ttl = READ_ONCE(inet->mc_ttl);
1460 else
1461 ttl = ip_select_ttl(inet, &rt->dst);
1462
1463 iph = ip_hdr(skb);
1464 iph->version = 4;
1465 iph->ihl = 5;
1466 iph->tos = (cork->tos != -1) ? cork->tos : READ_ONCE(inet->tos);
1467 iph->frag_off = df;
1468 iph->ttl = ttl;
1469 iph->protocol = sk->sk_protocol;
1470 ip_copy_addrs(iph, fl4);
1471 ip_select_ident(net, skb, sk);
1472
1473 if (opt) {
1474 iph->ihl += opt->optlen >> 2;
1475 ip_options_build(skb, opt, cork->addr, rt);
1476 }
1477
1478 skb->priority = cork->priority;
1479 skb->mark = cork->mark;
1480 if (sk_is_tcp(sk))
1481 skb_set_delivery_time(skb, cork->transmit_time, SKB_CLOCK_MONOTONIC);
1482 else
1483 skb_set_delivery_type_by_clockid(skb, cork->transmit_time, sk->sk_clockid);
1484 /*
1485 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1486 * on dst refcount
1487 */
1488 cork->dst = NULL;
1489 skb_dst_set(skb, &rt->dst);
1490
1491 if (iph->protocol == IPPROTO_ICMP) {
1492 u8 icmp_type;
1493
1494 /* For such sockets, transhdrlen is zero when do ip_append_data(),
1495 * so icmphdr does not in skb linear region and can not get icmp_type
1496 * by icmp_hdr(skb)->type.
1497 */
1498 if (sk->sk_type == SOCK_RAW &&
1499 !(fl4->flowi4_flags & FLOWI_FLAG_KNOWN_NH))
1500 icmp_type = fl4->fl4_icmp_type;
1501 else
1502 icmp_type = icmp_hdr(skb)->type;
1503 icmp_out_count(net, icmp_type);
1504 }
1505
1506 ip_cork_release(cork);
1507out:
1508 return skb;
1509}
1510
1511int ip_send_skb(struct net *net, struct sk_buff *skb)
1512{
1513 int err;
1514
1515 err = ip_local_out(net, skb->sk, skb);
1516 if (err) {
1517 if (err > 0)
1518 err = net_xmit_errno(err);
1519 if (err)
1520 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1521 }
1522
1523 return err;
1524}
1525
1526int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1527{
1528 struct sk_buff *skb;
1529
1530 skb = ip_finish_skb(sk, fl4);
1531 if (!skb)
1532 return 0;
1533
1534 /* Netfilter gets whole the not fragmented skb. */
1535 return ip_send_skb(sock_net(sk), skb);
1536}
1537
1538/*
1539 * Throw away all pending data on the socket.
1540 */
1541static void __ip_flush_pending_frames(struct sock *sk,
1542 struct sk_buff_head *queue,
1543 struct inet_cork *cork)
1544{
1545 struct sk_buff *skb;
1546
1547 while ((skb = __skb_dequeue_tail(queue)) != NULL)
1548 kfree_skb(skb);
1549
1550 ip_cork_release(cork);
1551}
1552
1553void ip_flush_pending_frames(struct sock *sk)
1554{
1555 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1556}
1557
1558struct sk_buff *ip_make_skb(struct sock *sk,
1559 struct flowi4 *fl4,
1560 int getfrag(void *from, char *to, int offset,
1561 int len, int odd, struct sk_buff *skb),
1562 void *from, int length, int transhdrlen,
1563 struct ipcm_cookie *ipc, struct rtable **rtp,
1564 struct inet_cork *cork, unsigned int flags)
1565{
1566 struct sk_buff_head queue;
1567 int err;
1568
1569 if (flags & MSG_PROBE)
1570 return NULL;
1571
1572 __skb_queue_head_init(&queue);
1573
1574 cork->flags = 0;
1575 cork->addr = 0;
1576 cork->opt = NULL;
1577 err = ip_setup_cork(sk, cork, ipc, rtp);
1578 if (err)
1579 return ERR_PTR(err);
1580
1581 err = __ip_append_data(sk, fl4, &queue, cork,
1582 &current->task_frag, getfrag,
1583 from, length, transhdrlen, flags);
1584 if (err) {
1585 __ip_flush_pending_frames(sk, &queue, cork);
1586 return ERR_PTR(err);
1587 }
1588
1589 return __ip_make_skb(sk, fl4, &queue, cork);
1590}
1591
1592/*
1593 * Fetch data from kernel space and fill in checksum if needed.
1594 */
1595static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1596 int len, int odd, struct sk_buff *skb)
1597{
1598 __wsum csum;
1599
1600 csum = csum_partial_copy_nocheck(dptr+offset, to, len);
1601 skb->csum = csum_block_add(skb->csum, csum, odd);
1602 return 0;
1603}
1604
1605/*
1606 * Generic function to send a packet as reply to another packet.
1607 * Used to send some TCP resets/acks so far.
1608 */
1609void ip_send_unicast_reply(struct sock *sk, const struct sock *orig_sk,
1610 struct sk_buff *skb,
1611 const struct ip_options *sopt,
1612 __be32 daddr, __be32 saddr,
1613 const struct ip_reply_arg *arg,
1614 unsigned int len, u64 transmit_time, u32 txhash)
1615{
1616 DEFINE_RAW_FLEX(struct ip_options_rcu, replyopts, opt.__data,
1617 IP_OPTIONS_DATA_FIXED_SIZE);
1618 struct ipcm_cookie ipc;
1619 struct flowi4 fl4;
1620 struct rtable *rt = skb_rtable(skb);
1621 struct net *net = sock_net(sk);
1622 struct sk_buff *nskb;
1623 int err;
1624 int oif;
1625
1626 if (__ip_options_echo(net, &replyopts->opt, skb, sopt))
1627 return;
1628
1629 ipcm_init(&ipc);
1630 ipc.addr = daddr;
1631 ipc.sockc.transmit_time = transmit_time;
1632
1633 if (replyopts->opt.optlen) {
1634 ipc.opt = replyopts;
1635
1636 if (replyopts->opt.srr)
1637 daddr = replyopts->opt.faddr;
1638 }
1639
1640 oif = arg->bound_dev_if;
1641 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1642 oif = skb->skb_iif;
1643
1644 flowi4_init_output(&fl4, oif,
1645 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1646 arg->tos & INET_DSCP_MASK,
1647 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1648 ip_reply_arg_flowi_flags(arg),
1649 daddr, saddr,
1650 tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1651 arg->uid);
1652 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
1653 rt = ip_route_output_flow(net, &fl4, sk);
1654 if (IS_ERR(rt))
1655 return;
1656
1657 inet_sk(sk)->tos = arg->tos;
1658
1659 sk->sk_protocol = ip_hdr(skb)->protocol;
1660 sk->sk_bound_dev_if = arg->bound_dev_if;
1661 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1662 ipc.sockc.mark = fl4.flowi4_mark;
1663 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1664 len, 0, &ipc, &rt, MSG_DONTWAIT);
1665 if (unlikely(err)) {
1666 ip_flush_pending_frames(sk);
1667 goto out;
1668 }
1669
1670 nskb = skb_peek(&sk->sk_write_queue);
1671 if (nskb) {
1672 if (arg->csumoffset >= 0)
1673 *((__sum16 *)skb_transport_header(nskb) +
1674 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1675 arg->csum));
1676 nskb->ip_summed = CHECKSUM_NONE;
1677 if (orig_sk) {
1678 skb_set_owner_edemux(nskb, (struct sock *)orig_sk);
1679 psp_reply_set_decrypted(orig_sk, nskb);
1680 }
1681 if (transmit_time)
1682 nskb->tstamp_type = SKB_CLOCK_MONOTONIC;
1683 if (txhash)
1684 skb_set_hash(nskb, txhash, PKT_HASH_TYPE_L4);
1685 ip_push_pending_frames(sk, &fl4);
1686 }
1687out:
1688 ip_rt_put(rt);
1689}
1690
1691void __init ip_init(void)
1692{
1693 ip_rt_init();
1694 inet_initpeers();
1695
1696#if defined(CONFIG_IP_MULTICAST)
1697 igmp_mc_init();
1698#endif
1699}