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/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(iph: (unsigned char *)iph, ihl: 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, len: 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(pf: NFPROTO_IPV4, hook: NF_INET_LOCAL_OUT,
121 net, sk, skb, NULL, outdev: skb_dst_dev(skb),
122 okfn: 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, len: 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, dst: &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, dst: &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, segs: 1);
184 }
185
186 if (opt && opt->opt.optlen) {
187 iph->ihl += opt->opt.optlen>>2;
188 ip_options_build(skb, opt: &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}
198EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
199
200static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
201{
202 struct dst_entry *dst = skb_dst(skb);
203 struct rtable *rt = dst_rtable(dst);
204 struct net_device *dev = dst_dev(dst);
205 unsigned int hh_len = LL_RESERVED_SPACE(dev);
206 struct neighbour *neigh;
207 bool is_v6gw = false;
208
209 if (rt->rt_type == RTN_MULTICAST) {
210 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
211 } else if (rt->rt_type == RTN_BROADCAST)
212 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
213
214 /* OUTOCTETS should be counted after fragment */
215 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
216
217 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
218 skb = skb_expand_head(skb, headroom: hh_len);
219 if (!skb)
220 return -ENOMEM;
221 }
222
223 if (lwtunnel_xmit_redirect(lwtstate: dst->lwtstate)) {
224 int res = lwtunnel_xmit(skb);
225
226 if (res != LWTUNNEL_XMIT_CONTINUE)
227 return res;
228 }
229
230 rcu_read_lock();
231 neigh = ip_neigh_for_gw(rt, skb, is_v6gw: &is_v6gw);
232 if (!IS_ERR(ptr: neigh)) {
233 int res;
234
235 sock_confirm_neigh(skb, n: neigh);
236 /* if crossing protocols, can not use the cached header */
237 res = neigh_output(n: neigh, skb, skip_cache: is_v6gw);
238 rcu_read_unlock();
239 return res;
240 }
241 rcu_read_unlock();
242
243 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
244 __func__);
245 kfree_skb_reason(skb, reason: SKB_DROP_REASON_NEIGH_CREATEFAIL);
246 return PTR_ERR(ptr: neigh);
247}
248
249static int ip_finish_output_gso(struct net *net, struct sock *sk,
250 struct sk_buff *skb, unsigned int mtu)
251{
252 struct sk_buff *segs, *nskb;
253 netdev_features_t features;
254 int ret = 0;
255
256 /* common case: seglen is <= mtu
257 */
258 if (skb_gso_validate_network_len(skb, mtu))
259 return ip_finish_output2(net, sk, skb);
260
261 /* Slowpath - GSO segment length exceeds the egress MTU.
262 *
263 * This can happen in several cases:
264 * - Forwarding of a TCP GRO skb, when DF flag is not set.
265 * - Forwarding of an skb that arrived on a virtualization interface
266 * (virtio-net/vhost/tap) with TSO/GSO size set by other network
267 * stack.
268 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
269 * interface with a smaller MTU.
270 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is
271 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an
272 * insufficient MTU.
273 */
274 features = netif_skb_features(skb);
275 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
276 segs = skb_gso_segment(skb, features: features & ~NETIF_F_GSO_MASK);
277 if (IS_ERR_OR_NULL(ptr: segs)) {
278 kfree_skb(skb);
279 return -ENOMEM;
280 }
281
282 consume_skb(skb);
283
284 skb_list_walk_safe(segs, segs, nskb) {
285 int err;
286
287 skb_mark_not_on_list(skb: segs);
288 err = ip_fragment(net, sk, skb: segs, mtu, output: ip_finish_output2);
289
290 if (err && ret == 0)
291 ret = err;
292 }
293
294 return ret;
295}
296
297static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
298{
299 unsigned int mtu;
300
301#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
302 /* Policy lookup after SNAT yielded a new policy */
303 if (skb_dst(skb)->xfrm) {
304 IPCB(skb)->flags |= IPSKB_REROUTED;
305 return dst_output(net, sk, skb);
306 }
307#endif
308 mtu = ip_skb_dst_mtu(sk, skb);
309 if (skb_is_gso(skb))
310 return ip_finish_output_gso(net, sk, skb, mtu);
311
312 if (skb->len > mtu || IPCB(skb)->frag_max_size)
313 return ip_fragment(net, sk, skb, mtu, output: ip_finish_output2);
314
315 return ip_finish_output2(net, sk, skb);
316}
317
318static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
319{
320 int ret;
321
322 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
323 switch (ret) {
324 case NET_XMIT_SUCCESS:
325 return __ip_finish_output(net, sk, skb);
326 case NET_XMIT_CN:
327 return __ip_finish_output(net, sk, skb) ? : ret;
328 default:
329 kfree_skb_reason(skb, reason: SKB_DROP_REASON_BPF_CGROUP_EGRESS);
330 return ret;
331 }
332}
333
334static int ip_mc_finish_output(struct net *net, struct sock *sk,
335 struct sk_buff *skb)
336{
337 struct rtable *new_rt;
338 bool do_cn = false;
339 int ret, err;
340
341 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
342 switch (ret) {
343 case NET_XMIT_CN:
344 do_cn = true;
345 fallthrough;
346 case NET_XMIT_SUCCESS:
347 break;
348 default:
349 kfree_skb_reason(skb, reason: SKB_DROP_REASON_BPF_CGROUP_EGRESS);
350 return ret;
351 }
352
353 /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
354 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
355 * see ipv4_pktinfo_prepare().
356 */
357 new_rt = rt_dst_clone(dev: net->loopback_dev, rt: skb_rtable(skb));
358 if (new_rt) {
359 new_rt->rt_iif = 0;
360 skb_dst_drop(skb);
361 skb_dst_set(skb, dst: &new_rt->dst);
362 }
363
364 err = dev_loopback_xmit(net, sk, newskb: skb);
365 return (do_cn && err) ? ret : err;
366}
367
368int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
369{
370 struct rtable *rt = skb_rtable(skb);
371 struct net_device *dev = rt->dst.dev;
372
373 /*
374 * If the indicated interface is up and running, send the packet.
375 */
376 skb->dev = dev;
377 skb->protocol = htons(ETH_P_IP);
378
379 /*
380 * Multicasts are looped back for other local users
381 */
382
383 if (rt->rt_flags&RTCF_MULTICAST) {
384 if (sk_mc_loop(sk)
385#ifdef CONFIG_IP_MROUTE
386 /* Small optimization: do not loopback not local frames,
387 which returned after forwarding; they will be dropped
388 by ip_mr_input in any case.
389 Note, that local frames are looped back to be delivered
390 to local recipients.
391
392 This check is duplicated in ip_mr_input at the moment.
393 */
394 &&
395 ((rt->rt_flags & RTCF_LOCAL) ||
396 !(IPCB(skb)->flags & IPSKB_FORWARDED))
397#endif
398 ) {
399 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
400 if (newskb)
401 NF_HOOK(pf: NFPROTO_IPV4, hook: NF_INET_POST_ROUTING,
402 net, sk, skb: newskb, NULL, out: newskb->dev,
403 okfn: ip_mc_finish_output);
404 }
405
406 /* Multicasts with ttl 0 must not go beyond the host */
407
408 if (ip_hdr(skb)->ttl == 0) {
409 kfree_skb(skb);
410 return 0;
411 }
412 }
413
414 if (rt->rt_flags&RTCF_BROADCAST) {
415 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
416 if (newskb)
417 NF_HOOK(pf: NFPROTO_IPV4, hook: NF_INET_POST_ROUTING,
418 net, sk, skb: newskb, NULL, out: newskb->dev,
419 okfn: ip_mc_finish_output);
420 }
421
422 return NF_HOOK_COND(pf: NFPROTO_IPV4, hook: NF_INET_POST_ROUTING,
423 net, sk, skb, NULL, out: skb->dev,
424 okfn: ip_finish_output,
425 cond: !(IPCB(skb)->flags & IPSKB_REROUTED));
426}
427
428int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
429{
430 struct net_device *dev, *indev = skb->dev;
431 int ret_val;
432
433 rcu_read_lock();
434 dev = skb_dst_dev_rcu(skb);
435 skb->dev = dev;
436 skb->protocol = htons(ETH_P_IP);
437
438 ret_val = NF_HOOK_COND(pf: NFPROTO_IPV4, hook: NF_INET_POST_ROUTING,
439 net, sk, skb, in: indev, out: dev,
440 okfn: ip_finish_output,
441 cond: !(IPCB(skb)->flags & IPSKB_REROUTED));
442 rcu_read_unlock();
443 return ret_val;
444}
445EXPORT_SYMBOL(ip_output);
446
447/*
448 * copy saddr and daddr, possibly using 64bit load/stores
449 * Equivalent to :
450 * iph->saddr = fl4->saddr;
451 * iph->daddr = fl4->daddr;
452 */
453static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
454{
455 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
456 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
457
458 iph->saddr = fl4->saddr;
459 iph->daddr = fl4->daddr;
460}
461
462/* Note: skb->sk can be different from sk, in case of tunnels */
463int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
464 __u8 tos)
465{
466 struct inet_sock *inet = inet_sk(sk);
467 struct net *net = sock_net(sk);
468 struct ip_options_rcu *inet_opt;
469 struct flowi4 *fl4;
470 struct rtable *rt;
471 struct iphdr *iph;
472 int res;
473
474 /* Skip all of this if the packet is already routed,
475 * f.e. by something like SCTP.
476 */
477 rcu_read_lock();
478 inet_opt = rcu_dereference(inet->inet_opt);
479 fl4 = &fl->u.ip4;
480 rt = skb_rtable(skb);
481 if (rt)
482 goto packet_routed;
483
484 /* Make sure we can route this packet. */
485 rt = dst_rtable(__sk_dst_check(sk, 0));
486 if (!rt) {
487 inet_sk_init_flowi4(inet, fl4);
488
489 /* sctp_v4_xmit() uses its own DSCP value */
490 fl4->flowi4_dscp = inet_dsfield_to_dscp(dsfield: tos);
491
492 /* If this fails, retransmit mechanism of transport layer will
493 * keep trying until route appears or the connection times
494 * itself out.
495 */
496 rt = ip_route_output_flow(net, flp: fl4, sk);
497 if (IS_ERR(ptr: rt))
498 goto no_route;
499 sk_setup_caps(sk, dst: &rt->dst);
500 }
501 skb_dst_set_noref(skb, dst: &rt->dst);
502
503packet_routed:
504 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
505 goto no_route;
506
507 /* OK, we know where to send it, allocate and build IP header. */
508 skb_push(skb, len: sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
509 skb_reset_network_header(skb);
510 iph = ip_hdr(skb);
511 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
512 if (ip_dont_fragment(sk, dst: &rt->dst) && !skb->ignore_df)
513 iph->frag_off = htons(IP_DF);
514 else
515 iph->frag_off = 0;
516 iph->ttl = ip_select_ttl(inet, dst: &rt->dst);
517 iph->protocol = sk->sk_protocol;
518 ip_copy_addrs(iph, fl4);
519
520 /* Transport layer set skb->h.foo itself. */
521
522 if (inet_opt && inet_opt->opt.optlen) {
523 iph->ihl += inet_opt->opt.optlen >> 2;
524 ip_options_build(skb, opt: &inet_opt->opt, daddr: inet->inet_daddr, rt);
525 }
526
527 ip_select_ident_segs(net, skb, sk,
528 skb_shinfo(skb)->gso_segs ?: 1);
529
530 /* TODO : should we use skb->sk here instead of sk ? */
531 skb->priority = READ_ONCE(sk->sk_priority);
532 skb->mark = READ_ONCE(sk->sk_mark);
533
534 res = ip_local_out(net, sk, skb);
535 rcu_read_unlock();
536 return res;
537
538no_route:
539 rcu_read_unlock();
540 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
541 kfree_skb_reason(skb, reason: SKB_DROP_REASON_IP_OUTNOROUTES);
542 return -EHOSTUNREACH;
543}
544EXPORT_SYMBOL(__ip_queue_xmit);
545
546int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
547{
548 return __ip_queue_xmit(sk, skb, fl, READ_ONCE(inet_sk(sk)->tos));
549}
550EXPORT_SYMBOL(ip_queue_xmit);
551
552static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
553{
554 to->pkt_type = from->pkt_type;
555 to->priority = from->priority;
556 to->protocol = from->protocol;
557 to->skb_iif = from->skb_iif;
558 skb_dst_drop(skb: to);
559 skb_dst_copy(nskb: to, oskb: from);
560 to->dev = from->dev;
561 to->mark = from->mark;
562
563 skb_copy_hash(to, from);
564
565#ifdef CONFIG_NET_SCHED
566 to->tc_index = from->tc_index;
567#endif
568 nf_copy(dst: to, src: from);
569 skb_ext_copy(dst: to, src: from);
570#if IS_ENABLED(CONFIG_IP_VS)
571 to->ipvs_property = from->ipvs_property;
572#endif
573 skb_copy_secmark(to, from);
574}
575
576static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
577 unsigned int mtu,
578 int (*output)(struct net *, struct sock *, struct sk_buff *))
579{
580 struct iphdr *iph = ip_hdr(skb);
581
582 if ((iph->frag_off & htons(IP_DF)) == 0)
583 return ip_do_fragment(net, sk, skb, output);
584
585 if (unlikely(!skb->ignore_df ||
586 (IPCB(skb)->frag_max_size &&
587 IPCB(skb)->frag_max_size > mtu))) {
588 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
589 icmp_send(skb_in: skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
590 htonl(mtu));
591 kfree_skb(skb);
592 return -EMSGSIZE;
593 }
594
595 return ip_do_fragment(net, sk, skb, output);
596}
597
598void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
599 unsigned int hlen, struct ip_fraglist_iter *iter)
600{
601 unsigned int first_len = skb_pagelen(skb);
602
603 iter->frag = skb_shinfo(skb)->frag_list;
604 skb_frag_list_init(skb);
605
606 iter->offset = 0;
607 iter->iph = iph;
608 iter->hlen = hlen;
609
610 skb->data_len = first_len - skb_headlen(skb);
611 skb->len = first_len;
612 iph->tot_len = htons(first_len);
613 iph->frag_off = htons(IP_MF);
614 ip_send_check(iph);
615}
616EXPORT_SYMBOL(ip_fraglist_init);
617
618void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
619{
620 unsigned int hlen = iter->hlen;
621 struct iphdr *iph = iter->iph;
622 struct sk_buff *frag;
623
624 frag = iter->frag;
625 frag->ip_summed = CHECKSUM_NONE;
626 skb_reset_transport_header(skb: frag);
627 __skb_push(skb: frag, len: hlen);
628 skb_reset_network_header(skb: frag);
629 memcpy(to: skb_network_header(skb: frag), from: iph, len: hlen);
630 iter->iph = ip_hdr(skb: frag);
631 iph = iter->iph;
632 iph->tot_len = htons(frag->len);
633 ip_copy_metadata(to: frag, from: skb);
634 iter->offset += skb->len - hlen;
635 iph->frag_off = htons(iter->offset >> 3);
636 if (frag->next)
637 iph->frag_off |= htons(IP_MF);
638 /* Ready, complete checksum */
639 ip_send_check(iph);
640}
641EXPORT_SYMBOL(ip_fraglist_prepare);
642
643void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
644 unsigned int ll_rs, unsigned int mtu, bool DF,
645 struct ip_frag_state *state)
646{
647 struct iphdr *iph = ip_hdr(skb);
648
649 state->DF = DF;
650 state->hlen = hlen;
651 state->ll_rs = ll_rs;
652 state->mtu = mtu;
653
654 state->left = skb->len - hlen; /* Space per frame */
655 state->ptr = hlen; /* Where to start from */
656
657 state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
658 state->not_last_frag = iph->frag_off & htons(IP_MF);
659}
660EXPORT_SYMBOL(ip_frag_init);
661
662static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
663 bool first_frag)
664{
665 /* Copy the flags to each fragment. */
666 IPCB(to)->flags = IPCB(from)->flags;
667
668 /* ANK: dirty, but effective trick. Upgrade options only if
669 * the segment to be fragmented was THE FIRST (otherwise,
670 * options are already fixed) and make it ONCE
671 * on the initial skb, so that all the following fragments
672 * will inherit fixed options.
673 */
674 if (first_frag)
675 ip_options_fragment(skb: from);
676}
677
678struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
679{
680 unsigned int len = state->left;
681 struct sk_buff *skb2;
682 struct iphdr *iph;
683
684 /* IF: it doesn't fit, use 'mtu' - the data space left */
685 if (len > state->mtu)
686 len = state->mtu;
687 /* IF: we are not sending up to and including the packet end
688 then align the next start on an eight byte boundary */
689 if (len < state->left) {
690 len &= ~7;
691 }
692
693 /* Allocate buffer */
694 skb2 = alloc_skb(size: len + state->hlen + state->ll_rs, GFP_ATOMIC);
695 if (!skb2)
696 return ERR_PTR(error: -ENOMEM);
697
698 /*
699 * Set up data on packet
700 */
701
702 ip_copy_metadata(to: skb2, from: skb);
703 skb_reserve(skb: skb2, len: state->ll_rs);
704 skb_put(skb: skb2, len: len + state->hlen);
705 skb_reset_network_header(skb: skb2);
706 skb2->transport_header = skb2->network_header + state->hlen;
707
708 /*
709 * Charge the memory for the fragment to any owner
710 * it might possess
711 */
712
713 if (skb->sk)
714 skb_set_owner_w(skb: skb2, sk: skb->sk);
715
716 /*
717 * Copy the packet header into the new buffer.
718 */
719
720 skb_copy_from_linear_data(skb, to: skb_network_header(skb: skb2), len: state->hlen);
721
722 /*
723 * Copy a block of the IP datagram.
724 */
725 if (skb_copy_bits(skb, offset: state->ptr, to: skb_transport_header(skb: skb2), len))
726 BUG();
727 state->left -= len;
728
729 /*
730 * Fill in the new header fields.
731 */
732 iph = ip_hdr(skb: skb2);
733 iph->frag_off = htons((state->offset >> 3));
734 if (state->DF)
735 iph->frag_off |= htons(IP_DF);
736
737 /*
738 * Added AC : If we are fragmenting a fragment that's not the
739 * last fragment then keep MF on each bit
740 */
741 if (state->left > 0 || state->not_last_frag)
742 iph->frag_off |= htons(IP_MF);
743 state->ptr += len;
744 state->offset += len;
745
746 iph->tot_len = htons(len + state->hlen);
747
748 ip_send_check(iph);
749
750 return skb2;
751}
752EXPORT_SYMBOL(ip_frag_next);
753
754/*
755 * This IP datagram is too large to be sent in one piece. Break it up into
756 * smaller pieces (each of size equal to IP header plus
757 * a block of the data of the original IP data part) that will yet fit in a
758 * single device frame, and queue such a frame for sending.
759 */
760
761int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
762 int (*output)(struct net *, struct sock *, struct sk_buff *))
763{
764 struct iphdr *iph;
765 struct sk_buff *skb2;
766 u8 tstamp_type = skb->tstamp_type;
767 struct rtable *rt = skb_rtable(skb);
768 unsigned int mtu, hlen, ll_rs;
769 struct ip_fraglist_iter iter;
770 ktime_t tstamp = skb->tstamp;
771 struct ip_frag_state state;
772 int err = 0;
773
774 /* for offloaded checksums cleanup checksum before fragmentation */
775 if (skb->ip_summed == CHECKSUM_PARTIAL &&
776 (err = skb_checksum_help(skb)))
777 goto fail;
778
779 /*
780 * Point into the IP datagram header.
781 */
782
783 iph = ip_hdr(skb);
784
785 mtu = ip_skb_dst_mtu(sk, skb);
786 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
787 mtu = IPCB(skb)->frag_max_size;
788
789 /*
790 * Setup starting values.
791 */
792
793 hlen = iph->ihl * 4;
794 mtu = mtu - hlen; /* Size of data space */
795 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
796 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
797
798 /* When frag_list is given, use it. First, check its validity:
799 * some transformers could create wrong frag_list or break existing
800 * one, it is not prohibited. In this case fall back to copying.
801 *
802 * LATER: this step can be merged to real generation of fragments,
803 * we can switch to copy when see the first bad fragment.
804 */
805 if (skb_has_frag_list(skb)) {
806 struct sk_buff *frag, *frag2;
807 unsigned int first_len = skb_pagelen(skb);
808
809 if (first_len - hlen > mtu ||
810 ((first_len - hlen) & 7) ||
811 ip_is_fragment(iph) ||
812 skb_cloned(skb) ||
813 skb_headroom(skb) < ll_rs)
814 goto slow_path;
815
816 skb_walk_frags(skb, frag) {
817 /* Correct geometry. */
818 if (frag->len > mtu ||
819 ((frag->len & 7) && frag->next) ||
820 skb_headroom(skb: frag) < hlen + ll_rs)
821 goto slow_path_clean;
822
823 /* Partially cloned skb? */
824 if (skb_shared(skb: frag))
825 goto slow_path_clean;
826
827 BUG_ON(frag->sk);
828 if (skb->sk) {
829 frag->sk = skb->sk;
830 frag->destructor = sock_wfree;
831 }
832 skb->truesize -= frag->truesize;
833 }
834
835 /* Everything is OK. Generate! */
836 ip_fraglist_init(skb, iph, hlen, &iter);
837
838 for (;;) {
839 /* Prepare header of the next frame,
840 * before previous one went down. */
841 if (iter.frag) {
842 bool first_frag = (iter.offset == 0);
843
844 IPCB(iter.frag)->flags = IPCB(skb)->flags;
845 ip_fraglist_prepare(skb, &iter);
846 if (first_frag && IPCB(skb)->opt.optlen) {
847 /* ipcb->opt is not populated for frags
848 * coming from __ip_make_skb(),
849 * ip_options_fragment() needs optlen
850 */
851 IPCB(iter.frag)->opt.optlen =
852 IPCB(skb)->opt.optlen;
853 ip_options_fragment(skb: iter.frag);
854 ip_send_check(iter.iph);
855 }
856 }
857
858 skb_set_delivery_time(skb, kt: tstamp, tstamp_type);
859 err = output(net, sk, skb);
860
861 if (!err)
862 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
863 if (err || !iter.frag)
864 break;
865
866 skb = ip_fraglist_next(iter: &iter);
867 }
868
869 if (err == 0) {
870 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
871 return 0;
872 }
873
874 kfree_skb_list(segs: iter.frag);
875
876 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
877 return err;
878
879slow_path_clean:
880 skb_walk_frags(skb, frag2) {
881 if (frag2 == frag)
882 break;
883 frag2->sk = NULL;
884 frag2->destructor = NULL;
885 skb->truesize += frag2->truesize;
886 }
887 }
888
889slow_path:
890 /*
891 * Fragment the datagram.
892 */
893
894 ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
895 &state);
896
897 /*
898 * Keep copying data until we run out.
899 */
900
901 while (state.left > 0) {
902 bool first_frag = (state.offset == 0);
903
904 skb2 = ip_frag_next(skb, &state);
905 if (IS_ERR(ptr: skb2)) {
906 err = PTR_ERR(ptr: skb2);
907 goto fail;
908 }
909 ip_frag_ipcb(from: skb, to: skb2, first_frag);
910
911 /*
912 * Put this fragment into the sending queue.
913 */
914 skb_set_delivery_time(skb: skb2, kt: tstamp, tstamp_type);
915 err = output(net, sk, skb2);
916 if (err)
917 goto fail;
918
919 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
920 }
921 consume_skb(skb);
922 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
923 return err;
924
925fail:
926 kfree_skb(skb);
927 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
928 return err;
929}
930EXPORT_SYMBOL(ip_do_fragment);
931
932int
933ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
934{
935 struct msghdr *msg = from;
936
937 if (skb->ip_summed == CHECKSUM_PARTIAL) {
938 if (!copy_from_iter_full(addr: to, bytes: len, i: &msg->msg_iter))
939 return -EFAULT;
940 } else {
941 __wsum csum = 0;
942 if (!csum_and_copy_from_iter_full(addr: to, bytes: len, csum: &csum, i: &msg->msg_iter))
943 return -EFAULT;
944 skb->csum = csum_block_add(csum: skb->csum, csum2: csum, offset: odd);
945 }
946 return 0;
947}
948EXPORT_SYMBOL(ip_generic_getfrag);
949
950static int __ip_append_data(struct sock *sk,
951 struct flowi4 *fl4,
952 struct sk_buff_head *queue,
953 struct inet_cork *cork,
954 struct page_frag *pfrag,
955 int getfrag(void *from, char *to, int offset,
956 int len, int odd, struct sk_buff *skb),
957 void *from, int length, int transhdrlen,
958 unsigned int flags)
959{
960 struct inet_sock *inet = inet_sk(sk);
961 struct ubuf_info *uarg = NULL;
962 struct sk_buff *skb;
963 struct ip_options *opt = cork->opt;
964 int hh_len;
965 int exthdrlen;
966 int mtu;
967 int copy;
968 int err;
969 int offset = 0;
970 bool zc = false;
971 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
972 int csummode = CHECKSUM_NONE;
973 struct rtable *rt = dst_rtable(cork->dst);
974 bool paged, hold_tskey = false, extra_uref = false;
975 unsigned int wmem_alloc_delta = 0;
976 u32 tskey = 0;
977
978 skb = skb_peek_tail(list_: queue);
979
980 exthdrlen = !skb ? rt->dst.header_len : 0;
981 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
982 paged = !!cork->gso_size;
983
984 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
985
986 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
987 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
988 maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
989
990 if (cork->length + length > maxnonfragsize - fragheaderlen) {
991 ip_local_error(sk, EMSGSIZE, daddr: fl4->daddr, dport: inet->inet_dport,
992 info: mtu - (opt ? opt->optlen : 0));
993 return -EMSGSIZE;
994 }
995
996 /*
997 * transhdrlen > 0 means that this is the first fragment and we wish
998 * it won't be fragmented in the future.
999 */
1000 if (transhdrlen &&
1001 length + fragheaderlen <= mtu &&
1002 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1003 (!(flags & MSG_MORE) || cork->gso_size) &&
1004 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1005 csummode = CHECKSUM_PARTIAL;
1006
1007 if ((flags & MSG_ZEROCOPY) && length) {
1008 struct msghdr *msg = from;
1009
1010 if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1011 if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1012 return -EINVAL;
1013
1014 /* Leave uarg NULL if can't zerocopy, callers should
1015 * be able to handle it.
1016 */
1017 if ((rt->dst.dev->features & NETIF_F_SG) &&
1018 csummode == CHECKSUM_PARTIAL) {
1019 paged = true;
1020 zc = true;
1021 uarg = msg->msg_ubuf;
1022 }
1023 } else if (sock_flag(sk, flag: SOCK_ZEROCOPY)) {
1024 uarg = msg_zerocopy_realloc(sk, size: length, uarg: skb_zcopy(skb),
1025 devmem: false);
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 if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
1052 READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
1053 if (cork->flags & IPCORK_TS_OPT_ID) {
1054 tskey = cork->ts_opt_id;
1055 } else {
1056 tskey = atomic_inc_return(v: &sk->sk_tskey) - 1;
1057 hold_tskey = true;
1058 }
1059 }
1060
1061 /* So, what's going on in the loop below?
1062 *
1063 * We use calculated fragment length to generate chained skb,
1064 * each of segments is IP fragment ready for sending to network after
1065 * adding appropriate IP header.
1066 */
1067
1068 if (!skb)
1069 goto alloc_new_skb;
1070
1071 while (length > 0) {
1072 /* Check if the remaining data fits into current packet. */
1073 copy = mtu - skb->len;
1074 if (copy < length)
1075 copy = maxfraglen - skb->len;
1076 if (copy <= 0) {
1077 char *data;
1078 unsigned int datalen;
1079 unsigned int fraglen;
1080 unsigned int fraggap;
1081 unsigned int alloclen, alloc_extra;
1082 unsigned int pagedlen;
1083 struct sk_buff *skb_prev;
1084alloc_new_skb:
1085 skb_prev = skb;
1086 if (skb_prev)
1087 fraggap = skb_prev->len - maxfraglen;
1088 else
1089 fraggap = 0;
1090
1091 /*
1092 * If remaining data exceeds the mtu,
1093 * we know we need more fragment(s).
1094 */
1095 datalen = length + fraggap;
1096 if (datalen > mtu - fragheaderlen)
1097 datalen = maxfraglen - fragheaderlen;
1098 fraglen = datalen + fragheaderlen;
1099 pagedlen = 0;
1100
1101 alloc_extra = hh_len + 15;
1102 alloc_extra += exthdrlen;
1103
1104 /* The last fragment gets additional space at tail.
1105 * Note, with MSG_MORE we overallocate on fragments,
1106 * because we have no idea what fragment will be
1107 * the last.
1108 */
1109 if (datalen == length + fraggap)
1110 alloc_extra += rt->dst.trailer_len;
1111
1112 if ((flags & MSG_MORE) &&
1113 !(rt->dst.dev->features&NETIF_F_SG))
1114 alloclen = mtu;
1115 else if (!paged &&
1116 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1117 !(rt->dst.dev->features & NETIF_F_SG)))
1118 alloclen = fraglen;
1119 else {
1120 alloclen = fragheaderlen + transhdrlen;
1121 pagedlen = datalen - transhdrlen;
1122 }
1123
1124 alloclen += alloc_extra;
1125
1126 if (transhdrlen) {
1127 skb = sock_alloc_send_skb(sk, size: alloclen,
1128 noblock: (flags & MSG_DONTWAIT), errcode: &err);
1129 } else {
1130 skb = NULL;
1131 if (refcount_read(r: &sk->sk_wmem_alloc) + wmem_alloc_delta <=
1132 2 * sk->sk_sndbuf)
1133 skb = alloc_skb(size: alloclen,
1134 priority: sk->sk_allocation);
1135 if (unlikely(!skb))
1136 err = -ENOBUFS;
1137 }
1138 if (!skb)
1139 goto error;
1140
1141 /*
1142 * Fill in the control structures
1143 */
1144 skb->ip_summed = csummode;
1145 skb->csum = 0;
1146 skb_reserve(skb, len: hh_len);
1147
1148 /*
1149 * Find where to start putting bytes.
1150 */
1151 data = skb_put(skb, len: fraglen + exthdrlen - pagedlen);
1152 skb_set_network_header(skb, offset: exthdrlen);
1153 skb->transport_header = (skb->network_header +
1154 fragheaderlen);
1155 data += fragheaderlen + exthdrlen;
1156
1157 if (fraggap) {
1158 skb->csum = skb_copy_and_csum_bits(
1159 skb: skb_prev, offset: maxfraglen,
1160 to: data + transhdrlen, len: fraggap);
1161 skb_prev->csum = csum_sub(csum: skb_prev->csum,
1162 addend: skb->csum);
1163 data += fraggap;
1164 pskb_trim_unique(skb: skb_prev, len: maxfraglen);
1165 }
1166
1167 copy = datalen - transhdrlen - fraggap - pagedlen;
1168 /* [!] NOTE: copy will be negative if pagedlen>0
1169 * because then the equation reduces to -fraggap.
1170 */
1171 if (copy > 0 &&
1172 INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1173 from, data + transhdrlen, offset,
1174 copy, fraggap, skb) < 0) {
1175 err = -EFAULT;
1176 kfree_skb(skb);
1177 goto error;
1178 } else if (flags & MSG_SPLICE_PAGES) {
1179 copy = 0;
1180 }
1181
1182 offset += copy;
1183 length -= copy + transhdrlen;
1184 transhdrlen = 0;
1185 exthdrlen = 0;
1186 csummode = CHECKSUM_NONE;
1187
1188 /* only the initial fragment is time stamped */
1189 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1190 cork->tx_flags = 0;
1191 skb_shinfo(skb)->tskey = tskey;
1192 tskey = 0;
1193 skb_zcopy_set(skb, uarg, have_ref: &extra_uref);
1194
1195 if ((flags & MSG_CONFIRM) && !skb_prev)
1196 skb_set_dst_pending_confirm(skb, val: 1);
1197
1198 /*
1199 * Put the packet on the pending queue.
1200 */
1201 if (!skb->destructor) {
1202 skb->destructor = sock_wfree;
1203 skb->sk = sk;
1204 wmem_alloc_delta += skb->truesize;
1205 }
1206 __skb_queue_tail(list: queue, newsk: skb);
1207 continue;
1208 }
1209
1210 if (copy > length)
1211 copy = length;
1212
1213 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1214 skb_tailroom(skb) >= copy) {
1215 unsigned int off;
1216
1217 off = skb->len;
1218 if (INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1219 from, skb_put(skb, copy),
1220 offset, copy, off, skb) < 0) {
1221 __skb_trim(skb, len: off);
1222 err = -EFAULT;
1223 goto error;
1224 }
1225 } else if (flags & MSG_SPLICE_PAGES) {
1226 struct msghdr *msg = from;
1227
1228 err = -EIO;
1229 if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1230 goto error;
1231
1232 err = skb_splice_from_iter(skb, iter: &msg->msg_iter, maxsize: copy);
1233 if (err < 0)
1234 goto error;
1235 copy = err;
1236 wmem_alloc_delta += copy;
1237 } else if (!zc) {
1238 int i = skb_shinfo(skb)->nr_frags;
1239
1240 err = -ENOMEM;
1241 if (!sk_page_frag_refill(sk, pfrag))
1242 goto error;
1243
1244 skb_zcopy_downgrade_managed(skb);
1245 if (!skb_can_coalesce(skb, i, page: pfrag->page,
1246 off: pfrag->offset)) {
1247 err = -EMSGSIZE;
1248 if (i == MAX_SKB_FRAGS)
1249 goto error;
1250
1251 __skb_fill_page_desc(skb, i, page: pfrag->page,
1252 off: pfrag->offset, size: 0);
1253 skb_shinfo(skb)->nr_frags = ++i;
1254 get_page(page: pfrag->page);
1255 }
1256 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1257 if (INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1258 from,
1259 page_address(pfrag->page) + pfrag->offset,
1260 offset, copy, skb->len, skb) < 0)
1261 goto error_efault;
1262
1263 pfrag->offset += copy;
1264 skb_frag_size_add(frag: &skb_shinfo(skb)->frags[i - 1], delta: copy);
1265 skb_len_add(skb, delta: copy);
1266 wmem_alloc_delta += copy;
1267 } else {
1268 err = skb_zerocopy_iter_dgram(skb, msg: from, len: copy);
1269 if (err < 0)
1270 goto error;
1271 }
1272 offset += copy;
1273 length -= copy;
1274 }
1275
1276 if (wmem_alloc_delta)
1277 refcount_add(i: wmem_alloc_delta, r: &sk->sk_wmem_alloc);
1278 return 0;
1279
1280error_efault:
1281 err = -EFAULT;
1282error:
1283 net_zcopy_put_abort(uarg, have_uref: extra_uref);
1284 cork->length -= length;
1285 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1286 refcount_add(i: wmem_alloc_delta, r: &sk->sk_wmem_alloc);
1287 if (hold_tskey)
1288 atomic_dec(v: &sk->sk_tskey);
1289 return err;
1290}
1291
1292static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1293 struct ipcm_cookie *ipc, struct rtable **rtp)
1294{
1295 struct ip_options_rcu *opt;
1296 struct rtable *rt;
1297
1298 rt = *rtp;
1299 if (unlikely(!rt))
1300 return -EFAULT;
1301
1302 cork->fragsize = ip_sk_use_pmtu(sk) ?
1303 dst_mtu(dst: &rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1304
1305 if (!inetdev_valid_mtu(mtu: cork->fragsize))
1306 return -ENETUNREACH;
1307
1308 /*
1309 * setup for corking.
1310 */
1311 opt = ipc->opt;
1312 if (opt) {
1313 if (!cork->opt) {
1314 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1315 sk->sk_allocation);
1316 if (unlikely(!cork->opt))
1317 return -ENOBUFS;
1318 }
1319 memcpy(to: cork->opt, from: &opt->opt, len: sizeof(struct ip_options) + opt->opt.optlen);
1320 cork->flags |= IPCORK_OPT;
1321 cork->addr = ipc->addr;
1322 }
1323
1324 cork->gso_size = ipc->gso_size;
1325
1326 cork->dst = &rt->dst;
1327 /* We stole this route, caller should not release it. */
1328 *rtp = NULL;
1329
1330 cork->length = 0;
1331 cork->ttl = ipc->ttl;
1332 cork->tos = ipc->tos;
1333 cork->mark = ipc->sockc.mark;
1334 cork->priority = ipc->sockc.priority;
1335 cork->transmit_time = ipc->sockc.transmit_time;
1336 cork->tx_flags = 0;
1337 sock_tx_timestamp(sk, sockc: &ipc->sockc, tx_flags: &cork->tx_flags);
1338 if (ipc->sockc.tsflags & SOCKCM_FLAG_TS_OPT_ID) {
1339 cork->flags |= IPCORK_TS_OPT_ID;
1340 cork->ts_opt_id = ipc->sockc.ts_opt_id;
1341 }
1342
1343 return 0;
1344}
1345
1346/*
1347 * ip_append_data() can make one large IP datagram from many pieces of
1348 * data. Each piece will be held on the socket until
1349 * ip_push_pending_frames() is called. Each piece can be a page or
1350 * non-page data.
1351 *
1352 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1353 * this interface potentially.
1354 *
1355 * LATER: length must be adjusted by pad at tail, when it is required.
1356 */
1357int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1358 int getfrag(void *from, char *to, int offset, int len,
1359 int odd, struct sk_buff *skb),
1360 void *from, int length, int transhdrlen,
1361 struct ipcm_cookie *ipc, struct rtable **rtp,
1362 unsigned int flags)
1363{
1364 struct inet_sock *inet = inet_sk(sk);
1365 int err;
1366
1367 if (flags&MSG_PROBE)
1368 return 0;
1369
1370 if (skb_queue_empty(list: &sk->sk_write_queue)) {
1371 err = ip_setup_cork(sk, cork: &inet->cork.base, ipc, rtp);
1372 if (err)
1373 return err;
1374 } else {
1375 transhdrlen = 0;
1376 }
1377
1378 return __ip_append_data(sk, fl4, queue: &sk->sk_write_queue, cork: &inet->cork.base,
1379 pfrag: sk_page_frag(sk), getfrag,
1380 from, length, transhdrlen, flags);
1381}
1382
1383static void ip_cork_release(struct inet_cork *cork)
1384{
1385 cork->flags &= ~IPCORK_OPT;
1386 kfree(objp: cork->opt);
1387 cork->opt = NULL;
1388 dst_release(dst: cork->dst);
1389 cork->dst = NULL;
1390}
1391
1392/*
1393 * Combined all pending IP fragments on the socket as one IP datagram
1394 * and push them out.
1395 */
1396struct sk_buff *__ip_make_skb(struct sock *sk,
1397 struct flowi4 *fl4,
1398 struct sk_buff_head *queue,
1399 struct inet_cork *cork)
1400{
1401 struct sk_buff *skb, *tmp_skb;
1402 struct sk_buff **tail_skb;
1403 struct inet_sock *inet = inet_sk(sk);
1404 struct net *net = sock_net(sk);
1405 struct ip_options *opt = NULL;
1406 struct rtable *rt = dst_rtable(cork->dst);
1407 struct iphdr *iph;
1408 u8 pmtudisc, ttl;
1409 __be16 df = 0;
1410
1411 skb = __skb_dequeue(list: queue);
1412 if (!skb)
1413 goto out;
1414 tail_skb = &(skb_shinfo(skb)->frag_list);
1415
1416 /* move skb->data to ip header from ext header */
1417 if (skb->data < skb_network_header(skb))
1418 __skb_pull(skb, len: skb_network_offset(skb));
1419 while ((tmp_skb = __skb_dequeue(list: queue)) != NULL) {
1420 __skb_pull(skb: tmp_skb, len: skb_network_header_len(skb));
1421 *tail_skb = tmp_skb;
1422 tail_skb = &(tmp_skb->next);
1423 skb->len += tmp_skb->len;
1424 skb->data_len += tmp_skb->len;
1425 skb->truesize += tmp_skb->truesize;
1426 tmp_skb->destructor = NULL;
1427 tmp_skb->sk = NULL;
1428 }
1429
1430 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1431 * to fragment the frame generated here. No matter, what transforms
1432 * how transforms change size of the packet, it will come out.
1433 */
1434 skb->ignore_df = ip_sk_ignore_df(sk);
1435
1436 /* DF bit is set when we want to see DF on outgoing frames.
1437 * If ignore_df is set too, we still allow to fragment this frame
1438 * locally. */
1439 pmtudisc = READ_ONCE(inet->pmtudisc);
1440 if (pmtudisc == IP_PMTUDISC_DO ||
1441 pmtudisc == IP_PMTUDISC_PROBE ||
1442 (skb->len <= dst_mtu(dst: &rt->dst) &&
1443 ip_dont_fragment(sk, dst: &rt->dst)))
1444 df = htons(IP_DF);
1445
1446 if (cork->flags & IPCORK_OPT)
1447 opt = cork->opt;
1448
1449 if (cork->ttl != 0)
1450 ttl = cork->ttl;
1451 else if (rt->rt_type == RTN_MULTICAST)
1452 ttl = READ_ONCE(inet->mc_ttl);
1453 else
1454 ttl = ip_select_ttl(inet, dst: &rt->dst);
1455
1456 iph = ip_hdr(skb);
1457 iph->version = 4;
1458 iph->ihl = 5;
1459 iph->tos = (cork->tos != -1) ? cork->tos : READ_ONCE(inet->tos);
1460 iph->frag_off = df;
1461 iph->ttl = ttl;
1462 iph->protocol = sk->sk_protocol;
1463 ip_copy_addrs(iph, fl4);
1464 ip_select_ident(net, skb, sk);
1465
1466 if (opt) {
1467 iph->ihl += opt->optlen >> 2;
1468 ip_options_build(skb, opt, daddr: cork->addr, rt);
1469 }
1470
1471 skb->priority = cork->priority;
1472 skb->mark = cork->mark;
1473 if (sk_is_tcp(sk))
1474 skb_set_delivery_time(skb, kt: cork->transmit_time, tstamp_type: SKB_CLOCK_MONOTONIC);
1475 else
1476 skb_set_delivery_type_by_clockid(skb, kt: cork->transmit_time, clockid: sk->sk_clockid);
1477 /*
1478 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1479 * on dst refcount
1480 */
1481 cork->dst = NULL;
1482 skb_dst_set(skb, dst: &rt->dst);
1483
1484 if (iph->protocol == IPPROTO_ICMP) {
1485 u8 icmp_type;
1486
1487 /* For such sockets, transhdrlen is zero when do ip_append_data(),
1488 * so icmphdr does not in skb linear region and can not get icmp_type
1489 * by icmp_hdr(skb)->type.
1490 */
1491 if (sk->sk_type == SOCK_RAW &&
1492 !(fl4->flowi4_flags & FLOWI_FLAG_KNOWN_NH))
1493 icmp_type = fl4->fl4_icmp_type;
1494 else
1495 icmp_type = icmp_hdr(skb)->type;
1496 icmp_out_count(net, type: icmp_type);
1497 }
1498
1499 ip_cork_release(cork);
1500out:
1501 return skb;
1502}
1503
1504int ip_send_skb(struct net *net, struct sk_buff *skb)
1505{
1506 int err;
1507
1508 err = ip_local_out(net, skb->sk, skb);
1509 if (err) {
1510 if (err > 0)
1511 err = net_xmit_errno(err);
1512 if (err)
1513 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1514 }
1515
1516 return err;
1517}
1518
1519int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1520{
1521 struct sk_buff *skb;
1522
1523 skb = ip_finish_skb(sk, fl4);
1524 if (!skb)
1525 return 0;
1526
1527 /* Netfilter gets whole the not fragmented skb. */
1528 return ip_send_skb(net: sock_net(sk), skb);
1529}
1530
1531/*
1532 * Throw away all pending data on the socket.
1533 */
1534static void __ip_flush_pending_frames(struct sock *sk,
1535 struct sk_buff_head *queue,
1536 struct inet_cork *cork)
1537{
1538 struct sk_buff *skb;
1539
1540 while ((skb = __skb_dequeue_tail(list: queue)) != NULL)
1541 kfree_skb(skb);
1542
1543 ip_cork_release(cork);
1544}
1545
1546void ip_flush_pending_frames(struct sock *sk)
1547{
1548 __ip_flush_pending_frames(sk, queue: &sk->sk_write_queue, cork: &inet_sk(sk)->cork.base);
1549}
1550
1551struct sk_buff *ip_make_skb(struct sock *sk,
1552 struct flowi4 *fl4,
1553 int getfrag(void *from, char *to, int offset,
1554 int len, int odd, struct sk_buff *skb),
1555 void *from, int length, int transhdrlen,
1556 struct ipcm_cookie *ipc, struct rtable **rtp,
1557 struct inet_cork *cork, unsigned int flags)
1558{
1559 struct sk_buff_head queue;
1560 int err;
1561
1562 if (flags & MSG_PROBE)
1563 return NULL;
1564
1565 __skb_queue_head_init(list: &queue);
1566
1567 cork->flags = 0;
1568 cork->addr = 0;
1569 cork->opt = NULL;
1570 err = ip_setup_cork(sk, cork, ipc, rtp);
1571 if (err)
1572 return ERR_PTR(error: err);
1573
1574 err = __ip_append_data(sk, fl4, queue: &queue, cork,
1575 pfrag: &current->task_frag, getfrag,
1576 from, length, transhdrlen, flags);
1577 if (err) {
1578 __ip_flush_pending_frames(sk, queue: &queue, cork);
1579 return ERR_PTR(error: err);
1580 }
1581
1582 return __ip_make_skb(sk, fl4, queue: &queue, cork);
1583}
1584
1585/*
1586 * Fetch data from kernel space and fill in checksum if needed.
1587 */
1588static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1589 int len, int odd, struct sk_buff *skb)
1590{
1591 __wsum csum;
1592
1593 csum = csum_partial_copy_nocheck(src: dptr+offset, dst: to, len);
1594 skb->csum = csum_block_add(csum: skb->csum, csum2: csum, offset: odd);
1595 return 0;
1596}
1597
1598/*
1599 * Generic function to send a packet as reply to another packet.
1600 * Used to send some TCP resets/acks so far.
1601 */
1602void ip_send_unicast_reply(struct sock *sk, const struct sock *orig_sk,
1603 struct sk_buff *skb,
1604 const struct ip_options *sopt,
1605 __be32 daddr, __be32 saddr,
1606 const struct ip_reply_arg *arg,
1607 unsigned int len, u64 transmit_time, u32 txhash)
1608{
1609 struct ip_options_data replyopts;
1610 struct ipcm_cookie ipc;
1611 struct flowi4 fl4;
1612 struct rtable *rt = skb_rtable(skb);
1613 struct net *net = sock_net(sk);
1614 struct sk_buff *nskb;
1615 int err;
1616 int oif;
1617
1618 if (__ip_options_echo(net, dopt: &replyopts.opt.opt, skb, sopt))
1619 return;
1620
1621 ipcm_init(ipcm: &ipc);
1622 ipc.addr = daddr;
1623 ipc.sockc.transmit_time = transmit_time;
1624
1625 if (replyopts.opt.opt.optlen) {
1626 ipc.opt = &replyopts.opt;
1627
1628 if (replyopts.opt.opt.srr)
1629 daddr = replyopts.opt.opt.faddr;
1630 }
1631
1632 oif = arg->bound_dev_if;
1633 if (!oif && netif_index_is_l3_master(net, ifindex: skb->skb_iif))
1634 oif = skb->skb_iif;
1635
1636 flowi4_init_output(fl4: &fl4, oif,
1637 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1638 tos: arg->tos & INET_DSCP_MASK,
1639 scope: RT_SCOPE_UNIVERSE, proto: ip_hdr(skb)->protocol,
1640 flags: ip_reply_arg_flowi_flags(arg),
1641 daddr, saddr,
1642 dport: tcp_hdr(skb)->source, sport: tcp_hdr(skb)->dest,
1643 uid: arg->uid);
1644 security_skb_classify_flow(skb, flic: flowi4_to_flowi_common(fl4: &fl4));
1645 rt = ip_route_output_flow(net, flp: &fl4, sk);
1646 if (IS_ERR(ptr: rt))
1647 return;
1648
1649 inet_sk(sk)->tos = arg->tos;
1650
1651 sk->sk_protocol = ip_hdr(skb)->protocol;
1652 sk->sk_bound_dev_if = arg->bound_dev_if;
1653 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1654 ipc.sockc.mark = fl4.flowi4_mark;
1655 err = ip_append_data(sk, fl4: &fl4, getfrag: ip_reply_glue_bits, from: arg->iov->iov_base,
1656 length: len, transhdrlen: 0, ipc: &ipc, rtp: &rt, MSG_DONTWAIT);
1657 if (unlikely(err)) {
1658 ip_flush_pending_frames(sk);
1659 goto out;
1660 }
1661
1662 nskb = skb_peek(list_: &sk->sk_write_queue);
1663 if (nskb) {
1664 if (arg->csumoffset >= 0)
1665 *((__sum16 *)skb_transport_header(skb: nskb) +
1666 arg->csumoffset) = csum_fold(sum: csum_add(csum: nskb->csum,
1667 addend: arg->csum));
1668 nskb->ip_summed = CHECKSUM_NONE;
1669 if (orig_sk) {
1670 skb_set_owner_edemux(skb: nskb, sk: (struct sock *)orig_sk);
1671 psp_reply_set_decrypted(sk: orig_sk, skb: nskb);
1672 }
1673 if (transmit_time)
1674 nskb->tstamp_type = SKB_CLOCK_MONOTONIC;
1675 if (txhash)
1676 skb_set_hash(skb: nskb, hash: txhash, type: PKT_HASH_TYPE_L4);
1677 ip_push_pending_frames(sk, fl4: &fl4);
1678 }
1679out:
1680 ip_rt_put(rt);
1681}
1682
1683void __init ip_init(void)
1684{
1685 ip_rt_init();
1686 inet_initpeers();
1687
1688#if defined(CONFIG_IP_MULTICAST)
1689 igmp_mc_init();
1690#endif
1691}
1692