xref: /openbmc/linux/net/ipv4/ip_output.c (revision 40662333)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		The Internet Protocol (IP) output module.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Donald Becker, <becker@super.org>
12  *		Alan Cox, <Alan.Cox@linux.org>
13  *		Richard Underwood
14  *		Stefan Becker, <stefanb@yello.ping.de>
15  *		Jorge Cwik, <jorge@laser.satlink.net>
16  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
17  *		Hirokazu Takahashi, <taka@valinux.co.jp>
18  *
19  *	See ip_input.c for original log
20  *
21  *	Fixes:
22  *		Alan Cox	:	Missing nonblock feature in ip_build_xmit.
23  *		Mike Kilburn	:	htons() missing in ip_build_xmit.
24  *		Bradford Johnson:	Fix faulty handling of some frames when
25  *					no route is found.
26  *		Alexander Demenshin:	Missing sk/skb free in ip_queue_xmit
27  *					(in case if packet not accepted by
28  *					output firewall rules)
29  *		Mike McLagan	:	Routing by source
30  *		Alexey Kuznetsov:	use new route cache
31  *		Andi Kleen:		Fix broken PMTU recovery and remove
32  *					some redundant tests.
33  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
34  *		Andi Kleen	: 	Replace ip_reply with ip_send_reply.
35  *		Andi Kleen	:	Split fast and slow ip_build_xmit path
36  *					for decreased register pressure on x86
37  *					and more readability.
38  *		Marc Boucher	:	When call_out_firewall returns FW_QUEUE,
39  *					silently drop skb instead of failing with -EPERM.
40  *		Detlev Wengorz	:	Copy protocol for fragments.
41  *		Hirokazu Takahashi:	HW checksumming for outgoing UDP
42  *					datagrams.
43  *		Hirokazu Takahashi:	sendfile() on UDP works now.
44  */
45 
46 #include <linux/uaccess.h>
47 #include <linux/module.h>
48 #include <linux/types.h>
49 #include <linux/kernel.h>
50 #include <linux/mm.h>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
54 #include <linux/slab.h>
55 
56 #include <linux/socket.h>
57 #include <linux/sockios.h>
58 #include <linux/in.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/etherdevice.h>
62 #include <linux/proc_fs.h>
63 #include <linux/stat.h>
64 #include <linux/init.h>
65 
66 #include <net/snmp.h>
67 #include <net/ip.h>
68 #include <net/protocol.h>
69 #include <net/route.h>
70 #include <net/xfrm.h>
71 #include <linux/skbuff.h>
72 #include <net/sock.h>
73 #include <net/arp.h>
74 #include <net/icmp.h>
75 #include <net/checksum.h>
76 #include <net/inetpeer.h>
77 #include <net/inet_ecn.h>
78 #include <net/lwtunnel.h>
79 #include <linux/bpf-cgroup.h>
80 #include <linux/igmp.h>
81 #include <linux/netfilter_ipv4.h>
82 #include <linux/netfilter_bridge.h>
83 #include <linux/netlink.h>
84 #include <linux/tcp.h>
85 
86 static int
87 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
88 	    unsigned int mtu,
89 	    int (*output)(struct net *, struct sock *, struct sk_buff *));
90 
91 /* Generate a checksum for an outgoing IP datagram. */
92 void ip_send_check(struct iphdr *iph)
93 {
94 	iph->check = 0;
95 	iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
96 }
97 EXPORT_SYMBOL(ip_send_check);
98 
99 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
100 {
101 	struct iphdr *iph = ip_hdr(skb);
102 
103 	iph->tot_len = htons(skb->len);
104 	ip_send_check(iph);
105 
106 	/* if egress device is enslaved to an L3 master device pass the
107 	 * skb to its handler for processing
108 	 */
109 	skb = l3mdev_ip_out(sk, skb);
110 	if (unlikely(!skb))
111 		return 0;
112 
113 	skb->protocol = htons(ETH_P_IP);
114 
115 	return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
116 		       net, sk, skb, NULL, skb_dst(skb)->dev,
117 		       dst_output);
118 }
119 
120 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
121 {
122 	int err;
123 
124 	err = __ip_local_out(net, sk, skb);
125 	if (likely(err == 1))
126 		err = dst_output(net, sk, skb);
127 
128 	return err;
129 }
130 EXPORT_SYMBOL_GPL(ip_local_out);
131 
132 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
133 {
134 	int ttl = inet->uc_ttl;
135 
136 	if (ttl < 0)
137 		ttl = ip4_dst_hoplimit(dst);
138 	return ttl;
139 }
140 
141 /*
142  *		Add an ip header to a skbuff and send it out.
143  *
144  */
145 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
146 			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
147 			  u8 tos)
148 {
149 	struct inet_sock *inet = inet_sk(sk);
150 	struct rtable *rt = skb_rtable(skb);
151 	struct net *net = sock_net(sk);
152 	struct iphdr *iph;
153 
154 	/* Build the IP header. */
155 	skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
156 	skb_reset_network_header(skb);
157 	iph = ip_hdr(skb);
158 	iph->version  = 4;
159 	iph->ihl      = 5;
160 	iph->tos      = tos;
161 	iph->ttl      = ip_select_ttl(inet, &rt->dst);
162 	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
163 	iph->saddr    = saddr;
164 	iph->protocol = sk->sk_protocol;
165 	/* Do not bother generating IPID for small packets (eg SYNACK) */
166 	if (skb->len <= IPV4_MIN_MTU || ip_dont_fragment(sk, &rt->dst)) {
167 		iph->frag_off = htons(IP_DF);
168 		iph->id = 0;
169 	} else {
170 		iph->frag_off = 0;
171 		/* TCP packets here are SYNACK with fat IPv4/TCP options.
172 		 * Avoid using the hashed IP ident generator.
173 		 */
174 		if (sk->sk_protocol == IPPROTO_TCP)
175 			iph->id = (__force __be16)prandom_u32();
176 		else
177 			__ip_select_ident(net, iph, 1);
178 	}
179 
180 	if (opt && opt->opt.optlen) {
181 		iph->ihl += opt->opt.optlen>>2;
182 		ip_options_build(skb, &opt->opt, daddr, rt);
183 	}
184 
185 	skb->priority = sk->sk_priority;
186 	if (!skb->mark)
187 		skb->mark = sk->sk_mark;
188 
189 	/* Send it out. */
190 	return ip_local_out(net, skb->sk, skb);
191 }
192 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
193 
194 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
195 {
196 	struct dst_entry *dst = skb_dst(skb);
197 	struct rtable *rt = (struct rtable *)dst;
198 	struct net_device *dev = dst->dev;
199 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
200 	struct neighbour *neigh;
201 	bool is_v6gw = false;
202 
203 	if (rt->rt_type == RTN_MULTICAST) {
204 		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
205 	} else if (rt->rt_type == RTN_BROADCAST)
206 		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
207 
208 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
209 		skb = skb_expand_head(skb, hh_len);
210 		if (!skb)
211 			return -ENOMEM;
212 	}
213 
214 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
215 		int res = lwtunnel_xmit(skb);
216 
217 		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
218 			return res;
219 	}
220 
221 	rcu_read_lock_bh();
222 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
223 	if (!IS_ERR(neigh)) {
224 		int res;
225 
226 		sock_confirm_neigh(skb, neigh);
227 		/* if crossing protocols, can not use the cached header */
228 		res = neigh_output(neigh, skb, is_v6gw);
229 		rcu_read_unlock_bh();
230 		return res;
231 	}
232 	rcu_read_unlock_bh();
233 
234 	net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
235 			    __func__);
236 	kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
237 	return -EINVAL;
238 }
239 
240 static int ip_finish_output_gso(struct net *net, struct sock *sk,
241 				struct sk_buff *skb, unsigned int mtu)
242 {
243 	struct sk_buff *segs, *nskb;
244 	netdev_features_t features;
245 	int ret = 0;
246 
247 	/* common case: seglen is <= mtu
248 	 */
249 	if (skb_gso_validate_network_len(skb, mtu))
250 		return ip_finish_output2(net, sk, skb);
251 
252 	/* Slowpath -  GSO segment length exceeds the egress MTU.
253 	 *
254 	 * This can happen in several cases:
255 	 *  - Forwarding of a TCP GRO skb, when DF flag is not set.
256 	 *  - Forwarding of an skb that arrived on a virtualization interface
257 	 *    (virtio-net/vhost/tap) with TSO/GSO size set by other network
258 	 *    stack.
259 	 *  - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
260 	 *    interface with a smaller MTU.
261 	 *  - Arriving GRO skb (or GSO skb in a virtualized environment) that is
262 	 *    bridged to a NETIF_F_TSO tunnel stacked over an interface with an
263 	 *    insufficient MTU.
264 	 */
265 	features = netif_skb_features(skb);
266 	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
267 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
268 	if (IS_ERR_OR_NULL(segs)) {
269 		kfree_skb(skb);
270 		return -ENOMEM;
271 	}
272 
273 	consume_skb(skb);
274 
275 	skb_list_walk_safe(segs, segs, nskb) {
276 		int err;
277 
278 		skb_mark_not_on_list(segs);
279 		err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
280 
281 		if (err && ret == 0)
282 			ret = err;
283 	}
284 
285 	return ret;
286 }
287 
288 static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
289 {
290 	unsigned int mtu;
291 
292 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
293 	/* Policy lookup after SNAT yielded a new policy */
294 	if (skb_dst(skb)->xfrm) {
295 		IPCB(skb)->flags |= IPSKB_REROUTED;
296 		return dst_output(net, sk, skb);
297 	}
298 #endif
299 	mtu = ip_skb_dst_mtu(sk, skb);
300 	if (skb_is_gso(skb))
301 		return ip_finish_output_gso(net, sk, skb, mtu);
302 
303 	if (skb->len > mtu || IPCB(skb)->frag_max_size)
304 		return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
305 
306 	return ip_finish_output2(net, sk, skb);
307 }
308 
309 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
310 {
311 	int ret;
312 
313 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
314 	switch (ret) {
315 	case NET_XMIT_SUCCESS:
316 		return __ip_finish_output(net, sk, skb);
317 	case NET_XMIT_CN:
318 		return __ip_finish_output(net, sk, skb) ? : ret;
319 	default:
320 		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
321 		return ret;
322 	}
323 }
324 
325 static int ip_mc_finish_output(struct net *net, struct sock *sk,
326 			       struct sk_buff *skb)
327 {
328 	struct rtable *new_rt;
329 	bool do_cn = false;
330 	int ret, err;
331 
332 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
333 	switch (ret) {
334 	case NET_XMIT_CN:
335 		do_cn = true;
336 		fallthrough;
337 	case NET_XMIT_SUCCESS:
338 		break;
339 	default:
340 		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
341 		return ret;
342 	}
343 
344 	/* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
345 	 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
346 	 * see ipv4_pktinfo_prepare().
347 	 */
348 	new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
349 	if (new_rt) {
350 		new_rt->rt_iif = 0;
351 		skb_dst_drop(skb);
352 		skb_dst_set(skb, &new_rt->dst);
353 	}
354 
355 	err = dev_loopback_xmit(net, sk, skb);
356 	return (do_cn && err) ? ret : err;
357 }
358 
359 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
360 {
361 	struct rtable *rt = skb_rtable(skb);
362 	struct net_device *dev = rt->dst.dev;
363 
364 	/*
365 	 *	If the indicated interface is up and running, send the packet.
366 	 */
367 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
368 
369 	skb->dev = dev;
370 	skb->protocol = htons(ETH_P_IP);
371 
372 	/*
373 	 *	Multicasts are looped back for other local users
374 	 */
375 
376 	if (rt->rt_flags&RTCF_MULTICAST) {
377 		if (sk_mc_loop(sk)
378 #ifdef CONFIG_IP_MROUTE
379 		/* Small optimization: do not loopback not local frames,
380 		   which returned after forwarding; they will be  dropped
381 		   by ip_mr_input in any case.
382 		   Note, that local frames are looped back to be delivered
383 		   to local recipients.
384 
385 		   This check is duplicated in ip_mr_input at the moment.
386 		 */
387 		    &&
388 		    ((rt->rt_flags & RTCF_LOCAL) ||
389 		     !(IPCB(skb)->flags & IPSKB_FORWARDED))
390 #endif
391 		   ) {
392 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
393 			if (newskb)
394 				NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
395 					net, sk, newskb, NULL, newskb->dev,
396 					ip_mc_finish_output);
397 		}
398 
399 		/* Multicasts with ttl 0 must not go beyond the host */
400 
401 		if (ip_hdr(skb)->ttl == 0) {
402 			kfree_skb(skb);
403 			return 0;
404 		}
405 	}
406 
407 	if (rt->rt_flags&RTCF_BROADCAST) {
408 		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
409 		if (newskb)
410 			NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
411 				net, sk, newskb, NULL, newskb->dev,
412 				ip_mc_finish_output);
413 	}
414 
415 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
416 			    net, sk, skb, NULL, skb->dev,
417 			    ip_finish_output,
418 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
419 }
420 
421 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
422 {
423 	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
424 
425 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
426 
427 	skb->dev = dev;
428 	skb->protocol = htons(ETH_P_IP);
429 
430 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
431 			    net, sk, skb, indev, dev,
432 			    ip_finish_output,
433 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
434 }
435 EXPORT_SYMBOL(ip_output);
436 
437 /*
438  * copy saddr and daddr, possibly using 64bit load/stores
439  * Equivalent to :
440  *   iph->saddr = fl4->saddr;
441  *   iph->daddr = fl4->daddr;
442  */
443 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
444 {
445 	BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
446 		     offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
447 
448 	iph->saddr = fl4->saddr;
449 	iph->daddr = fl4->daddr;
450 }
451 
452 /* Note: skb->sk can be different from sk, in case of tunnels */
453 int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
454 		    __u8 tos)
455 {
456 	struct inet_sock *inet = inet_sk(sk);
457 	struct net *net = sock_net(sk);
458 	struct ip_options_rcu *inet_opt;
459 	struct flowi4 *fl4;
460 	struct rtable *rt;
461 	struct iphdr *iph;
462 	int res;
463 
464 	/* Skip all of this if the packet is already routed,
465 	 * f.e. by something like SCTP.
466 	 */
467 	rcu_read_lock();
468 	inet_opt = rcu_dereference(inet->inet_opt);
469 	fl4 = &fl->u.ip4;
470 	rt = skb_rtable(skb);
471 	if (rt)
472 		goto packet_routed;
473 
474 	/* Make sure we can route this packet. */
475 	rt = (struct rtable *)__sk_dst_check(sk, 0);
476 	if (!rt) {
477 		__be32 daddr;
478 
479 		/* Use correct destination address if we have options. */
480 		daddr = inet->inet_daddr;
481 		if (inet_opt && inet_opt->opt.srr)
482 			daddr = inet_opt->opt.faddr;
483 
484 		/* If this fails, retransmit mechanism of transport layer will
485 		 * keep trying until route appears or the connection times
486 		 * itself out.
487 		 */
488 		rt = ip_route_output_ports(net, fl4, sk,
489 					   daddr, inet->inet_saddr,
490 					   inet->inet_dport,
491 					   inet->inet_sport,
492 					   sk->sk_protocol,
493 					   RT_CONN_FLAGS_TOS(sk, tos),
494 					   sk->sk_bound_dev_if);
495 		if (IS_ERR(rt))
496 			goto no_route;
497 		sk_setup_caps(sk, &rt->dst);
498 	}
499 	skb_dst_set_noref(skb, &rt->dst);
500 
501 packet_routed:
502 	if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
503 		goto no_route;
504 
505 	/* OK, we know where to send it, allocate and build IP header. */
506 	skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
507 	skb_reset_network_header(skb);
508 	iph = ip_hdr(skb);
509 	*((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
510 	if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
511 		iph->frag_off = htons(IP_DF);
512 	else
513 		iph->frag_off = 0;
514 	iph->ttl      = ip_select_ttl(inet, &rt->dst);
515 	iph->protocol = sk->sk_protocol;
516 	ip_copy_addrs(iph, fl4);
517 
518 	/* Transport layer set skb->h.foo itself. */
519 
520 	if (inet_opt && inet_opt->opt.optlen) {
521 		iph->ihl += inet_opt->opt.optlen >> 2;
522 		ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt);
523 	}
524 
525 	ip_select_ident_segs(net, skb, sk,
526 			     skb_shinfo(skb)->gso_segs ?: 1);
527 
528 	/* TODO : should we use skb->sk here instead of sk ? */
529 	skb->priority = sk->sk_priority;
530 	skb->mark = sk->sk_mark;
531 
532 	res = ip_local_out(net, sk, skb);
533 	rcu_read_unlock();
534 	return res;
535 
536 no_route:
537 	rcu_read_unlock();
538 	IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
539 	kfree_skb_reason(skb, SKB_DROP_REASON_IP_OUTNOROUTES);
540 	return -EHOSTUNREACH;
541 }
542 EXPORT_SYMBOL(__ip_queue_xmit);
543 
544 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
545 {
546 	return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
547 }
548 EXPORT_SYMBOL(ip_queue_xmit);
549 
550 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
551 {
552 	to->pkt_type = from->pkt_type;
553 	to->priority = from->priority;
554 	to->protocol = from->protocol;
555 	to->skb_iif = from->skb_iif;
556 	skb_dst_drop(to);
557 	skb_dst_copy(to, from);
558 	to->dev = from->dev;
559 	to->mark = from->mark;
560 
561 	skb_copy_hash(to, from);
562 
563 #ifdef CONFIG_NET_SCHED
564 	to->tc_index = from->tc_index;
565 #endif
566 	nf_copy(to, from);
567 	skb_ext_copy(to, from);
568 #if IS_ENABLED(CONFIG_IP_VS)
569 	to->ipvs_property = from->ipvs_property;
570 #endif
571 	skb_copy_secmark(to, from);
572 }
573 
574 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
575 		       unsigned int mtu,
576 		       int (*output)(struct net *, struct sock *, struct sk_buff *))
577 {
578 	struct iphdr *iph = ip_hdr(skb);
579 
580 	if ((iph->frag_off & htons(IP_DF)) == 0)
581 		return ip_do_fragment(net, sk, skb, output);
582 
583 	if (unlikely(!skb->ignore_df ||
584 		     (IPCB(skb)->frag_max_size &&
585 		      IPCB(skb)->frag_max_size > mtu))) {
586 		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
587 		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
588 			  htonl(mtu));
589 		kfree_skb(skb);
590 		return -EMSGSIZE;
591 	}
592 
593 	return ip_do_fragment(net, sk, skb, output);
594 }
595 
596 void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
597 		      unsigned int hlen, struct ip_fraglist_iter *iter)
598 {
599 	unsigned int first_len = skb_pagelen(skb);
600 
601 	iter->frag = skb_shinfo(skb)->frag_list;
602 	skb_frag_list_init(skb);
603 
604 	iter->offset = 0;
605 	iter->iph = iph;
606 	iter->hlen = hlen;
607 
608 	skb->data_len = first_len - skb_headlen(skb);
609 	skb->len = first_len;
610 	iph->tot_len = htons(first_len);
611 	iph->frag_off = htons(IP_MF);
612 	ip_send_check(iph);
613 }
614 EXPORT_SYMBOL(ip_fraglist_init);
615 
616 void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
617 {
618 	unsigned int hlen = iter->hlen;
619 	struct iphdr *iph = iter->iph;
620 	struct sk_buff *frag;
621 
622 	frag = iter->frag;
623 	frag->ip_summed = CHECKSUM_NONE;
624 	skb_reset_transport_header(frag);
625 	__skb_push(frag, hlen);
626 	skb_reset_network_header(frag);
627 	memcpy(skb_network_header(frag), iph, hlen);
628 	iter->iph = ip_hdr(frag);
629 	iph = iter->iph;
630 	iph->tot_len = htons(frag->len);
631 	ip_copy_metadata(frag, skb);
632 	iter->offset += skb->len - hlen;
633 	iph->frag_off = htons(iter->offset >> 3);
634 	if (frag->next)
635 		iph->frag_off |= htons(IP_MF);
636 	/* Ready, complete checksum */
637 	ip_send_check(iph);
638 }
639 EXPORT_SYMBOL(ip_fraglist_prepare);
640 
641 void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
642 		  unsigned int ll_rs, unsigned int mtu, bool DF,
643 		  struct ip_frag_state *state)
644 {
645 	struct iphdr *iph = ip_hdr(skb);
646 
647 	state->DF = DF;
648 	state->hlen = hlen;
649 	state->ll_rs = ll_rs;
650 	state->mtu = mtu;
651 
652 	state->left = skb->len - hlen;	/* Space per frame */
653 	state->ptr = hlen;		/* Where to start from */
654 
655 	state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
656 	state->not_last_frag = iph->frag_off & htons(IP_MF);
657 }
658 EXPORT_SYMBOL(ip_frag_init);
659 
660 static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
661 			 bool first_frag)
662 {
663 	/* Copy the flags to each fragment. */
664 	IPCB(to)->flags = IPCB(from)->flags;
665 
666 	/* ANK: dirty, but effective trick. Upgrade options only if
667 	 * the segment to be fragmented was THE FIRST (otherwise,
668 	 * options are already fixed) and make it ONCE
669 	 * on the initial skb, so that all the following fragments
670 	 * will inherit fixed options.
671 	 */
672 	if (first_frag)
673 		ip_options_fragment(from);
674 }
675 
676 struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
677 {
678 	unsigned int len = state->left;
679 	struct sk_buff *skb2;
680 	struct iphdr *iph;
681 
682 	/* IF: it doesn't fit, use 'mtu' - the data space left */
683 	if (len > state->mtu)
684 		len = state->mtu;
685 	/* IF: we are not sending up to and including the packet end
686 	   then align the next start on an eight byte boundary */
687 	if (len < state->left)	{
688 		len &= ~7;
689 	}
690 
691 	/* Allocate buffer */
692 	skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
693 	if (!skb2)
694 		return ERR_PTR(-ENOMEM);
695 
696 	/*
697 	 *	Set up data on packet
698 	 */
699 
700 	ip_copy_metadata(skb2, skb);
701 	skb_reserve(skb2, state->ll_rs);
702 	skb_put(skb2, len + state->hlen);
703 	skb_reset_network_header(skb2);
704 	skb2->transport_header = skb2->network_header + state->hlen;
705 
706 	/*
707 	 *	Charge the memory for the fragment to any owner
708 	 *	it might possess
709 	 */
710 
711 	if (skb->sk)
712 		skb_set_owner_w(skb2, skb->sk);
713 
714 	/*
715 	 *	Copy the packet header into the new buffer.
716 	 */
717 
718 	skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
719 
720 	/*
721 	 *	Copy a block of the IP datagram.
722 	 */
723 	if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
724 		BUG();
725 	state->left -= len;
726 
727 	/*
728 	 *	Fill in the new header fields.
729 	 */
730 	iph = ip_hdr(skb2);
731 	iph->frag_off = htons((state->offset >> 3));
732 	if (state->DF)
733 		iph->frag_off |= htons(IP_DF);
734 
735 	/*
736 	 *	Added AC : If we are fragmenting a fragment that's not the
737 	 *		   last fragment then keep MF on each bit
738 	 */
739 	if (state->left > 0 || state->not_last_frag)
740 		iph->frag_off |= htons(IP_MF);
741 	state->ptr += len;
742 	state->offset += len;
743 
744 	iph->tot_len = htons(len + state->hlen);
745 
746 	ip_send_check(iph);
747 
748 	return skb2;
749 }
750 EXPORT_SYMBOL(ip_frag_next);
751 
752 /*
753  *	This IP datagram is too large to be sent in one piece.  Break it up into
754  *	smaller pieces (each of size equal to IP header plus
755  *	a block of the data of the original IP data part) that will yet fit in a
756  *	single device frame, and queue such a frame for sending.
757  */
758 
759 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
760 		   int (*output)(struct net *, struct sock *, struct sk_buff *))
761 {
762 	struct iphdr *iph;
763 	struct sk_buff *skb2;
764 	bool mono_delivery_time = skb->mono_delivery_time;
765 	struct rtable *rt = skb_rtable(skb);
766 	unsigned int mtu, hlen, ll_rs;
767 	struct ip_fraglist_iter iter;
768 	ktime_t tstamp = skb->tstamp;
769 	struct ip_frag_state state;
770 	int err = 0;
771 
772 	/* for offloaded checksums cleanup checksum before fragmentation */
773 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
774 	    (err = skb_checksum_help(skb)))
775 		goto fail;
776 
777 	/*
778 	 *	Point into the IP datagram header.
779 	 */
780 
781 	iph = ip_hdr(skb);
782 
783 	mtu = ip_skb_dst_mtu(sk, skb);
784 	if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
785 		mtu = IPCB(skb)->frag_max_size;
786 
787 	/*
788 	 *	Setup starting values.
789 	 */
790 
791 	hlen = iph->ihl * 4;
792 	mtu = mtu - hlen;	/* Size of data space */
793 	IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
794 	ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
795 
796 	/* When frag_list is given, use it. First, check its validity:
797 	 * some transformers could create wrong frag_list or break existing
798 	 * one, it is not prohibited. In this case fall back to copying.
799 	 *
800 	 * LATER: this step can be merged to real generation of fragments,
801 	 * we can switch to copy when see the first bad fragment.
802 	 */
803 	if (skb_has_frag_list(skb)) {
804 		struct sk_buff *frag, *frag2;
805 		unsigned int first_len = skb_pagelen(skb);
806 
807 		if (first_len - hlen > mtu ||
808 		    ((first_len - hlen) & 7) ||
809 		    ip_is_fragment(iph) ||
810 		    skb_cloned(skb) ||
811 		    skb_headroom(skb) < ll_rs)
812 			goto slow_path;
813 
814 		skb_walk_frags(skb, frag) {
815 			/* Correct geometry. */
816 			if (frag->len > mtu ||
817 			    ((frag->len & 7) && frag->next) ||
818 			    skb_headroom(frag) < hlen + ll_rs)
819 				goto slow_path_clean;
820 
821 			/* Partially cloned skb? */
822 			if (skb_shared(frag))
823 				goto slow_path_clean;
824 
825 			BUG_ON(frag->sk);
826 			if (skb->sk) {
827 				frag->sk = skb->sk;
828 				frag->destructor = sock_wfree;
829 			}
830 			skb->truesize -= frag->truesize;
831 		}
832 
833 		/* Everything is OK. Generate! */
834 		ip_fraglist_init(skb, iph, hlen, &iter);
835 
836 		for (;;) {
837 			/* Prepare header of the next frame,
838 			 * before previous one went down. */
839 			if (iter.frag) {
840 				bool first_frag = (iter.offset == 0);
841 
842 				IPCB(iter.frag)->flags = IPCB(skb)->flags;
843 				ip_fraglist_prepare(skb, &iter);
844 				if (first_frag && IPCB(skb)->opt.optlen) {
845 					/* ipcb->opt is not populated for frags
846 					 * coming from __ip_make_skb(),
847 					 * ip_options_fragment() needs optlen
848 					 */
849 					IPCB(iter.frag)->opt.optlen =
850 						IPCB(skb)->opt.optlen;
851 					ip_options_fragment(iter.frag);
852 					ip_send_check(iter.iph);
853 				}
854 			}
855 
856 			skb_set_delivery_time(skb, tstamp, mono_delivery_time);
857 			err = output(net, sk, skb);
858 
859 			if (!err)
860 				IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
861 			if (err || !iter.frag)
862 				break;
863 
864 			skb = ip_fraglist_next(&iter);
865 		}
866 
867 		if (err == 0) {
868 			IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
869 			return 0;
870 		}
871 
872 		kfree_skb_list(iter.frag);
873 
874 		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
875 		return err;
876 
877 slow_path_clean:
878 		skb_walk_frags(skb, frag2) {
879 			if (frag2 == frag)
880 				break;
881 			frag2->sk = NULL;
882 			frag2->destructor = NULL;
883 			skb->truesize += frag2->truesize;
884 		}
885 	}
886 
887 slow_path:
888 	/*
889 	 *	Fragment the datagram.
890 	 */
891 
892 	ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
893 		     &state);
894 
895 	/*
896 	 *	Keep copying data until we run out.
897 	 */
898 
899 	while (state.left > 0) {
900 		bool first_frag = (state.offset == 0);
901 
902 		skb2 = ip_frag_next(skb, &state);
903 		if (IS_ERR(skb2)) {
904 			err = PTR_ERR(skb2);
905 			goto fail;
906 		}
907 		ip_frag_ipcb(skb, skb2, first_frag);
908 
909 		/*
910 		 *	Put this fragment into the sending queue.
911 		 */
912 		skb_set_delivery_time(skb2, tstamp, mono_delivery_time);
913 		err = output(net, sk, skb2);
914 		if (err)
915 			goto fail;
916 
917 		IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
918 	}
919 	consume_skb(skb);
920 	IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
921 	return err;
922 
923 fail:
924 	kfree_skb(skb);
925 	IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
926 	return err;
927 }
928 EXPORT_SYMBOL(ip_do_fragment);
929 
930 int
931 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
932 {
933 	struct msghdr *msg = from;
934 
935 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
936 		if (!copy_from_iter_full(to, len, &msg->msg_iter))
937 			return -EFAULT;
938 	} else {
939 		__wsum csum = 0;
940 		if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
941 			return -EFAULT;
942 		skb->csum = csum_block_add(skb->csum, csum, odd);
943 	}
944 	return 0;
945 }
946 EXPORT_SYMBOL(ip_generic_getfrag);
947 
948 static inline __wsum
949 csum_page(struct page *page, int offset, int copy)
950 {
951 	char *kaddr;
952 	__wsum csum;
953 	kaddr = kmap(page);
954 	csum = csum_partial(kaddr + offset, copy, 0);
955 	kunmap(page);
956 	return csum;
957 }
958 
959 static int __ip_append_data(struct sock *sk,
960 			    struct flowi4 *fl4,
961 			    struct sk_buff_head *queue,
962 			    struct inet_cork *cork,
963 			    struct page_frag *pfrag,
964 			    int getfrag(void *from, char *to, int offset,
965 					int len, int odd, struct sk_buff *skb),
966 			    void *from, int length, int transhdrlen,
967 			    unsigned int flags)
968 {
969 	struct inet_sock *inet = inet_sk(sk);
970 	struct ubuf_info *uarg = NULL;
971 	struct sk_buff *skb;
972 	struct ip_options *opt = cork->opt;
973 	int hh_len;
974 	int exthdrlen;
975 	int mtu;
976 	int copy;
977 	int err;
978 	int offset = 0;
979 	bool zc = false;
980 	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
981 	int csummode = CHECKSUM_NONE;
982 	struct rtable *rt = (struct rtable *)cork->dst;
983 	unsigned int wmem_alloc_delta = 0;
984 	bool paged, extra_uref = false;
985 	u32 tskey = 0;
986 
987 	skb = skb_peek_tail(queue);
988 
989 	exthdrlen = !skb ? rt->dst.header_len : 0;
990 	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
991 	paged = !!cork->gso_size;
992 
993 	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
994 	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
995 		tskey = atomic_inc_return(&sk->sk_tskey) - 1;
996 
997 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
998 
999 	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1000 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1001 	maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
1002 
1003 	if (cork->length + length > maxnonfragsize - fragheaderlen) {
1004 		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1005 			       mtu - (opt ? opt->optlen : 0));
1006 		return -EMSGSIZE;
1007 	}
1008 
1009 	/*
1010 	 * transhdrlen > 0 means that this is the first fragment and we wish
1011 	 * it won't be fragmented in the future.
1012 	 */
1013 	if (transhdrlen &&
1014 	    length + fragheaderlen <= mtu &&
1015 	    rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1016 	    (!(flags & MSG_MORE) || cork->gso_size) &&
1017 	    (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1018 		csummode = CHECKSUM_PARTIAL;
1019 
1020 	if ((flags & MSG_ZEROCOPY) && length) {
1021 		struct msghdr *msg = from;
1022 
1023 		if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1024 			if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1025 				return -EINVAL;
1026 
1027 			/* Leave uarg NULL if can't zerocopy, callers should
1028 			 * be able to handle it.
1029 			 */
1030 			if ((rt->dst.dev->features & NETIF_F_SG) &&
1031 			    csummode == CHECKSUM_PARTIAL) {
1032 				paged = true;
1033 				zc = true;
1034 				uarg = msg->msg_ubuf;
1035 			}
1036 		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1037 			uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1038 			if (!uarg)
1039 				return -ENOBUFS;
1040 			extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1041 			if (rt->dst.dev->features & NETIF_F_SG &&
1042 			    csummode == CHECKSUM_PARTIAL) {
1043 				paged = true;
1044 				zc = true;
1045 			} else {
1046 				uarg->zerocopy = 0;
1047 				skb_zcopy_set(skb, uarg, &extra_uref);
1048 			}
1049 		}
1050 	}
1051 
1052 	cork->length += length;
1053 
1054 	/* So, what's going on in the loop below?
1055 	 *
1056 	 * We use calculated fragment length to generate chained skb,
1057 	 * each of segments is IP fragment ready for sending to network after
1058 	 * adding appropriate IP header.
1059 	 */
1060 
1061 	if (!skb)
1062 		goto alloc_new_skb;
1063 
1064 	while (length > 0) {
1065 		/* Check if the remaining data fits into current packet. */
1066 		copy = mtu - skb->len;
1067 		if (copy < length)
1068 			copy = maxfraglen - skb->len;
1069 		if (copy <= 0) {
1070 			char *data;
1071 			unsigned int datalen;
1072 			unsigned int fraglen;
1073 			unsigned int fraggap;
1074 			unsigned int alloclen, alloc_extra;
1075 			unsigned int pagedlen;
1076 			struct sk_buff *skb_prev;
1077 alloc_new_skb:
1078 			skb_prev = skb;
1079 			if (skb_prev)
1080 				fraggap = skb_prev->len - maxfraglen;
1081 			else
1082 				fraggap = 0;
1083 
1084 			/*
1085 			 * If remaining data exceeds the mtu,
1086 			 * we know we need more fragment(s).
1087 			 */
1088 			datalen = length + fraggap;
1089 			if (datalen > mtu - fragheaderlen)
1090 				datalen = maxfraglen - fragheaderlen;
1091 			fraglen = datalen + fragheaderlen;
1092 			pagedlen = 0;
1093 
1094 			alloc_extra = hh_len + 15;
1095 			alloc_extra += exthdrlen;
1096 
1097 			/* The last fragment gets additional space at tail.
1098 			 * Note, with MSG_MORE we overallocate on fragments,
1099 			 * because we have no idea what fragment will be
1100 			 * the last.
1101 			 */
1102 			if (datalen == length + fraggap)
1103 				alloc_extra += rt->dst.trailer_len;
1104 
1105 			if ((flags & MSG_MORE) &&
1106 			    !(rt->dst.dev->features&NETIF_F_SG))
1107 				alloclen = mtu;
1108 			else if (!paged &&
1109 				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1110 				  !(rt->dst.dev->features & NETIF_F_SG)))
1111 				alloclen = fraglen;
1112 			else {
1113 				alloclen = fragheaderlen + transhdrlen;
1114 				pagedlen = datalen - transhdrlen;
1115 			}
1116 
1117 			alloclen += alloc_extra;
1118 
1119 			if (transhdrlen) {
1120 				skb = sock_alloc_send_skb(sk, alloclen,
1121 						(flags & MSG_DONTWAIT), &err);
1122 			} else {
1123 				skb = NULL;
1124 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1125 				    2 * sk->sk_sndbuf)
1126 					skb = alloc_skb(alloclen,
1127 							sk->sk_allocation);
1128 				if (unlikely(!skb))
1129 					err = -ENOBUFS;
1130 			}
1131 			if (!skb)
1132 				goto error;
1133 
1134 			/*
1135 			 *	Fill in the control structures
1136 			 */
1137 			skb->ip_summed = csummode;
1138 			skb->csum = 0;
1139 			skb_reserve(skb, hh_len);
1140 
1141 			/*
1142 			 *	Find where to start putting bytes.
1143 			 */
1144 			data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1145 			skb_set_network_header(skb, exthdrlen);
1146 			skb->transport_header = (skb->network_header +
1147 						 fragheaderlen);
1148 			data += fragheaderlen + exthdrlen;
1149 
1150 			if (fraggap) {
1151 				skb->csum = skb_copy_and_csum_bits(
1152 					skb_prev, maxfraglen,
1153 					data + transhdrlen, fraggap);
1154 				skb_prev->csum = csum_sub(skb_prev->csum,
1155 							  skb->csum);
1156 				data += fraggap;
1157 				pskb_trim_unique(skb_prev, maxfraglen);
1158 			}
1159 
1160 			copy = datalen - transhdrlen - fraggap - pagedlen;
1161 			if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1162 				err = -EFAULT;
1163 				kfree_skb(skb);
1164 				goto error;
1165 			}
1166 
1167 			offset += copy;
1168 			length -= copy + transhdrlen;
1169 			transhdrlen = 0;
1170 			exthdrlen = 0;
1171 			csummode = CHECKSUM_NONE;
1172 
1173 			/* only the initial fragment is time stamped */
1174 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1175 			cork->tx_flags = 0;
1176 			skb_shinfo(skb)->tskey = tskey;
1177 			tskey = 0;
1178 			skb_zcopy_set(skb, uarg, &extra_uref);
1179 
1180 			if ((flags & MSG_CONFIRM) && !skb_prev)
1181 				skb_set_dst_pending_confirm(skb, 1);
1182 
1183 			/*
1184 			 * Put the packet on the pending queue.
1185 			 */
1186 			if (!skb->destructor) {
1187 				skb->destructor = sock_wfree;
1188 				skb->sk = sk;
1189 				wmem_alloc_delta += skb->truesize;
1190 			}
1191 			__skb_queue_tail(queue, skb);
1192 			continue;
1193 		}
1194 
1195 		if (copy > length)
1196 			copy = length;
1197 
1198 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1199 		    skb_tailroom(skb) >= copy) {
1200 			unsigned int off;
1201 
1202 			off = skb->len;
1203 			if (getfrag(from, skb_put(skb, copy),
1204 					offset, copy, off, skb) < 0) {
1205 				__skb_trim(skb, off);
1206 				err = -EFAULT;
1207 				goto error;
1208 			}
1209 		} else if (!zc) {
1210 			int i = skb_shinfo(skb)->nr_frags;
1211 
1212 			err = -ENOMEM;
1213 			if (!sk_page_frag_refill(sk, pfrag))
1214 				goto error;
1215 
1216 			skb_zcopy_downgrade_managed(skb);
1217 			if (!skb_can_coalesce(skb, i, pfrag->page,
1218 					      pfrag->offset)) {
1219 				err = -EMSGSIZE;
1220 				if (i == MAX_SKB_FRAGS)
1221 					goto error;
1222 
1223 				__skb_fill_page_desc(skb, i, pfrag->page,
1224 						     pfrag->offset, 0);
1225 				skb_shinfo(skb)->nr_frags = ++i;
1226 				get_page(pfrag->page);
1227 			}
1228 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1229 			if (getfrag(from,
1230 				    page_address(pfrag->page) + pfrag->offset,
1231 				    offset, copy, skb->len, skb) < 0)
1232 				goto error_efault;
1233 
1234 			pfrag->offset += copy;
1235 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1236 			skb_len_add(skb, copy);
1237 			wmem_alloc_delta += copy;
1238 		} else {
1239 			err = skb_zerocopy_iter_dgram(skb, from, copy);
1240 			if (err < 0)
1241 				goto error;
1242 		}
1243 		offset += copy;
1244 		length -= copy;
1245 	}
1246 
1247 	if (wmem_alloc_delta)
1248 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1249 	return 0;
1250 
1251 error_efault:
1252 	err = -EFAULT;
1253 error:
1254 	net_zcopy_put_abort(uarg, extra_uref);
1255 	cork->length -= length;
1256 	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1257 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1258 	return err;
1259 }
1260 
1261 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1262 			 struct ipcm_cookie *ipc, struct rtable **rtp)
1263 {
1264 	struct ip_options_rcu *opt;
1265 	struct rtable *rt;
1266 
1267 	rt = *rtp;
1268 	if (unlikely(!rt))
1269 		return -EFAULT;
1270 
1271 	/*
1272 	 * setup for corking.
1273 	 */
1274 	opt = ipc->opt;
1275 	if (opt) {
1276 		if (!cork->opt) {
1277 			cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1278 					    sk->sk_allocation);
1279 			if (unlikely(!cork->opt))
1280 				return -ENOBUFS;
1281 		}
1282 		memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1283 		cork->flags |= IPCORK_OPT;
1284 		cork->addr = ipc->addr;
1285 	}
1286 
1287 	cork->fragsize = ip_sk_use_pmtu(sk) ?
1288 			 dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1289 
1290 	if (!inetdev_valid_mtu(cork->fragsize))
1291 		return -ENETUNREACH;
1292 
1293 	cork->gso_size = ipc->gso_size;
1294 
1295 	cork->dst = &rt->dst;
1296 	/* We stole this route, caller should not release it. */
1297 	*rtp = NULL;
1298 
1299 	cork->length = 0;
1300 	cork->ttl = ipc->ttl;
1301 	cork->tos = ipc->tos;
1302 	cork->mark = ipc->sockc.mark;
1303 	cork->priority = ipc->priority;
1304 	cork->transmit_time = ipc->sockc.transmit_time;
1305 	cork->tx_flags = 0;
1306 	sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags);
1307 
1308 	return 0;
1309 }
1310 
1311 /*
1312  *	ip_append_data() and ip_append_page() can make one large IP datagram
1313  *	from many pieces of data. Each pieces will be holded on the socket
1314  *	until ip_push_pending_frames() is called. Each piece can be a page
1315  *	or non-page data.
1316  *
1317  *	Not only UDP, other transport protocols - e.g. raw sockets - can use
1318  *	this interface potentially.
1319  *
1320  *	LATER: length must be adjusted by pad at tail, when it is required.
1321  */
1322 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1323 		   int getfrag(void *from, char *to, int offset, int len,
1324 			       int odd, struct sk_buff *skb),
1325 		   void *from, int length, int transhdrlen,
1326 		   struct ipcm_cookie *ipc, struct rtable **rtp,
1327 		   unsigned int flags)
1328 {
1329 	struct inet_sock *inet = inet_sk(sk);
1330 	int err;
1331 
1332 	if (flags&MSG_PROBE)
1333 		return 0;
1334 
1335 	if (skb_queue_empty(&sk->sk_write_queue)) {
1336 		err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1337 		if (err)
1338 			return err;
1339 	} else {
1340 		transhdrlen = 0;
1341 	}
1342 
1343 	return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1344 				sk_page_frag(sk), getfrag,
1345 				from, length, transhdrlen, flags);
1346 }
1347 
1348 ssize_t	ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1349 		       int offset, size_t size, int flags)
1350 {
1351 	struct inet_sock *inet = inet_sk(sk);
1352 	struct sk_buff *skb;
1353 	struct rtable *rt;
1354 	struct ip_options *opt = NULL;
1355 	struct inet_cork *cork;
1356 	int hh_len;
1357 	int mtu;
1358 	int len;
1359 	int err;
1360 	unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1361 
1362 	if (inet->hdrincl)
1363 		return -EPERM;
1364 
1365 	if (flags&MSG_PROBE)
1366 		return 0;
1367 
1368 	if (skb_queue_empty(&sk->sk_write_queue))
1369 		return -EINVAL;
1370 
1371 	cork = &inet->cork.base;
1372 	rt = (struct rtable *)cork->dst;
1373 	if (cork->flags & IPCORK_OPT)
1374 		opt = cork->opt;
1375 
1376 	if (!(rt->dst.dev->features & NETIF_F_SG))
1377 		return -EOPNOTSUPP;
1378 
1379 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1380 	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
1381 
1382 	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1383 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1384 	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1385 
1386 	if (cork->length + size > maxnonfragsize - fragheaderlen) {
1387 		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1388 			       mtu - (opt ? opt->optlen : 0));
1389 		return -EMSGSIZE;
1390 	}
1391 
1392 	skb = skb_peek_tail(&sk->sk_write_queue);
1393 	if (!skb)
1394 		return -EINVAL;
1395 
1396 	cork->length += size;
1397 
1398 	while (size > 0) {
1399 		/* Check if the remaining data fits into current packet. */
1400 		len = mtu - skb->len;
1401 		if (len < size)
1402 			len = maxfraglen - skb->len;
1403 
1404 		if (len <= 0) {
1405 			struct sk_buff *skb_prev;
1406 			int alloclen;
1407 
1408 			skb_prev = skb;
1409 			fraggap = skb_prev->len - maxfraglen;
1410 
1411 			alloclen = fragheaderlen + hh_len + fraggap + 15;
1412 			skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1413 			if (unlikely(!skb)) {
1414 				err = -ENOBUFS;
1415 				goto error;
1416 			}
1417 
1418 			/*
1419 			 *	Fill in the control structures
1420 			 */
1421 			skb->ip_summed = CHECKSUM_NONE;
1422 			skb->csum = 0;
1423 			skb_reserve(skb, hh_len);
1424 
1425 			/*
1426 			 *	Find where to start putting bytes.
1427 			 */
1428 			skb_put(skb, fragheaderlen + fraggap);
1429 			skb_reset_network_header(skb);
1430 			skb->transport_header = (skb->network_header +
1431 						 fragheaderlen);
1432 			if (fraggap) {
1433 				skb->csum = skb_copy_and_csum_bits(skb_prev,
1434 								   maxfraglen,
1435 						    skb_transport_header(skb),
1436 								   fraggap);
1437 				skb_prev->csum = csum_sub(skb_prev->csum,
1438 							  skb->csum);
1439 				pskb_trim_unique(skb_prev, maxfraglen);
1440 			}
1441 
1442 			/*
1443 			 * Put the packet on the pending queue.
1444 			 */
1445 			__skb_queue_tail(&sk->sk_write_queue, skb);
1446 			continue;
1447 		}
1448 
1449 		if (len > size)
1450 			len = size;
1451 
1452 		if (skb_append_pagefrags(skb, page, offset, len)) {
1453 			err = -EMSGSIZE;
1454 			goto error;
1455 		}
1456 
1457 		if (skb->ip_summed == CHECKSUM_NONE) {
1458 			__wsum csum;
1459 			csum = csum_page(page, offset, len);
1460 			skb->csum = csum_block_add(skb->csum, csum, skb->len);
1461 		}
1462 
1463 		skb_len_add(skb, len);
1464 		refcount_add(len, &sk->sk_wmem_alloc);
1465 		offset += len;
1466 		size -= len;
1467 	}
1468 	return 0;
1469 
1470 error:
1471 	cork->length -= size;
1472 	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1473 	return err;
1474 }
1475 
1476 static void ip_cork_release(struct inet_cork *cork)
1477 {
1478 	cork->flags &= ~IPCORK_OPT;
1479 	kfree(cork->opt);
1480 	cork->opt = NULL;
1481 	dst_release(cork->dst);
1482 	cork->dst = NULL;
1483 }
1484 
1485 /*
1486  *	Combined all pending IP fragments on the socket as one IP datagram
1487  *	and push them out.
1488  */
1489 struct sk_buff *__ip_make_skb(struct sock *sk,
1490 			      struct flowi4 *fl4,
1491 			      struct sk_buff_head *queue,
1492 			      struct inet_cork *cork)
1493 {
1494 	struct sk_buff *skb, *tmp_skb;
1495 	struct sk_buff **tail_skb;
1496 	struct inet_sock *inet = inet_sk(sk);
1497 	struct net *net = sock_net(sk);
1498 	struct ip_options *opt = NULL;
1499 	struct rtable *rt = (struct rtable *)cork->dst;
1500 	struct iphdr *iph;
1501 	__be16 df = 0;
1502 	__u8 ttl;
1503 
1504 	skb = __skb_dequeue(queue);
1505 	if (!skb)
1506 		goto out;
1507 	tail_skb = &(skb_shinfo(skb)->frag_list);
1508 
1509 	/* move skb->data to ip header from ext header */
1510 	if (skb->data < skb_network_header(skb))
1511 		__skb_pull(skb, skb_network_offset(skb));
1512 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1513 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1514 		*tail_skb = tmp_skb;
1515 		tail_skb = &(tmp_skb->next);
1516 		skb->len += tmp_skb->len;
1517 		skb->data_len += tmp_skb->len;
1518 		skb->truesize += tmp_skb->truesize;
1519 		tmp_skb->destructor = NULL;
1520 		tmp_skb->sk = NULL;
1521 	}
1522 
1523 	/* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1524 	 * to fragment the frame generated here. No matter, what transforms
1525 	 * how transforms change size of the packet, it will come out.
1526 	 */
1527 	skb->ignore_df = ip_sk_ignore_df(sk);
1528 
1529 	/* DF bit is set when we want to see DF on outgoing frames.
1530 	 * If ignore_df is set too, we still allow to fragment this frame
1531 	 * locally. */
1532 	if (inet->pmtudisc == IP_PMTUDISC_DO ||
1533 	    inet->pmtudisc == IP_PMTUDISC_PROBE ||
1534 	    (skb->len <= dst_mtu(&rt->dst) &&
1535 	     ip_dont_fragment(sk, &rt->dst)))
1536 		df = htons(IP_DF);
1537 
1538 	if (cork->flags & IPCORK_OPT)
1539 		opt = cork->opt;
1540 
1541 	if (cork->ttl != 0)
1542 		ttl = cork->ttl;
1543 	else if (rt->rt_type == RTN_MULTICAST)
1544 		ttl = inet->mc_ttl;
1545 	else
1546 		ttl = ip_select_ttl(inet, &rt->dst);
1547 
1548 	iph = ip_hdr(skb);
1549 	iph->version = 4;
1550 	iph->ihl = 5;
1551 	iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1552 	iph->frag_off = df;
1553 	iph->ttl = ttl;
1554 	iph->protocol = sk->sk_protocol;
1555 	ip_copy_addrs(iph, fl4);
1556 	ip_select_ident(net, skb, sk);
1557 
1558 	if (opt) {
1559 		iph->ihl += opt->optlen >> 2;
1560 		ip_options_build(skb, opt, cork->addr, rt);
1561 	}
1562 
1563 	skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1564 	skb->mark = cork->mark;
1565 	skb->tstamp = cork->transmit_time;
1566 	/*
1567 	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1568 	 * on dst refcount
1569 	 */
1570 	cork->dst = NULL;
1571 	skb_dst_set(skb, &rt->dst);
1572 
1573 	if (iph->protocol == IPPROTO_ICMP)
1574 		icmp_out_count(net, ((struct icmphdr *)
1575 			skb_transport_header(skb))->type);
1576 
1577 	ip_cork_release(cork);
1578 out:
1579 	return skb;
1580 }
1581 
1582 int ip_send_skb(struct net *net, struct sk_buff *skb)
1583 {
1584 	int err;
1585 
1586 	err = ip_local_out(net, skb->sk, skb);
1587 	if (err) {
1588 		if (err > 0)
1589 			err = net_xmit_errno(err);
1590 		if (err)
1591 			IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1592 	}
1593 
1594 	return err;
1595 }
1596 
1597 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1598 {
1599 	struct sk_buff *skb;
1600 
1601 	skb = ip_finish_skb(sk, fl4);
1602 	if (!skb)
1603 		return 0;
1604 
1605 	/* Netfilter gets whole the not fragmented skb. */
1606 	return ip_send_skb(sock_net(sk), skb);
1607 }
1608 
1609 /*
1610  *	Throw away all pending data on the socket.
1611  */
1612 static void __ip_flush_pending_frames(struct sock *sk,
1613 				      struct sk_buff_head *queue,
1614 				      struct inet_cork *cork)
1615 {
1616 	struct sk_buff *skb;
1617 
1618 	while ((skb = __skb_dequeue_tail(queue)) != NULL)
1619 		kfree_skb(skb);
1620 
1621 	ip_cork_release(cork);
1622 }
1623 
1624 void ip_flush_pending_frames(struct sock *sk)
1625 {
1626 	__ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1627 }
1628 
1629 struct sk_buff *ip_make_skb(struct sock *sk,
1630 			    struct flowi4 *fl4,
1631 			    int getfrag(void *from, char *to, int offset,
1632 					int len, int odd, struct sk_buff *skb),
1633 			    void *from, int length, int transhdrlen,
1634 			    struct ipcm_cookie *ipc, struct rtable **rtp,
1635 			    struct inet_cork *cork, unsigned int flags)
1636 {
1637 	struct sk_buff_head queue;
1638 	int err;
1639 
1640 	if (flags & MSG_PROBE)
1641 		return NULL;
1642 
1643 	__skb_queue_head_init(&queue);
1644 
1645 	cork->flags = 0;
1646 	cork->addr = 0;
1647 	cork->opt = NULL;
1648 	err = ip_setup_cork(sk, cork, ipc, rtp);
1649 	if (err)
1650 		return ERR_PTR(err);
1651 
1652 	err = __ip_append_data(sk, fl4, &queue, cork,
1653 			       &current->task_frag, getfrag,
1654 			       from, length, transhdrlen, flags);
1655 	if (err) {
1656 		__ip_flush_pending_frames(sk, &queue, cork);
1657 		return ERR_PTR(err);
1658 	}
1659 
1660 	return __ip_make_skb(sk, fl4, &queue, cork);
1661 }
1662 
1663 /*
1664  *	Fetch data from kernel space and fill in checksum if needed.
1665  */
1666 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1667 			      int len, int odd, struct sk_buff *skb)
1668 {
1669 	__wsum csum;
1670 
1671 	csum = csum_partial_copy_nocheck(dptr+offset, to, len);
1672 	skb->csum = csum_block_add(skb->csum, csum, odd);
1673 	return 0;
1674 }
1675 
1676 /*
1677  *	Generic function to send a packet as reply to another packet.
1678  *	Used to send some TCP resets/acks so far.
1679  */
1680 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1681 			   const struct ip_options *sopt,
1682 			   __be32 daddr, __be32 saddr,
1683 			   const struct ip_reply_arg *arg,
1684 			   unsigned int len, u64 transmit_time)
1685 {
1686 	struct ip_options_data replyopts;
1687 	struct ipcm_cookie ipc;
1688 	struct flowi4 fl4;
1689 	struct rtable *rt = skb_rtable(skb);
1690 	struct net *net = sock_net(sk);
1691 	struct sk_buff *nskb;
1692 	int err;
1693 	int oif;
1694 
1695 	if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
1696 		return;
1697 
1698 	ipcm_init(&ipc);
1699 	ipc.addr = daddr;
1700 	ipc.sockc.transmit_time = transmit_time;
1701 
1702 	if (replyopts.opt.opt.optlen) {
1703 		ipc.opt = &replyopts.opt;
1704 
1705 		if (replyopts.opt.opt.srr)
1706 			daddr = replyopts.opt.opt.faddr;
1707 	}
1708 
1709 	oif = arg->bound_dev_if;
1710 	if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1711 		oif = skb->skb_iif;
1712 
1713 	flowi4_init_output(&fl4, oif,
1714 			   IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1715 			   RT_TOS(arg->tos),
1716 			   RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1717 			   ip_reply_arg_flowi_flags(arg),
1718 			   daddr, saddr,
1719 			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1720 			   arg->uid);
1721 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
1722 	rt = ip_route_output_flow(net, &fl4, sk);
1723 	if (IS_ERR(rt))
1724 		return;
1725 
1726 	inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK;
1727 
1728 	sk->sk_protocol = ip_hdr(skb)->protocol;
1729 	sk->sk_bound_dev_if = arg->bound_dev_if;
1730 	sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1731 	ipc.sockc.mark = fl4.flowi4_mark;
1732 	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1733 			     len, 0, &ipc, &rt, MSG_DONTWAIT);
1734 	if (unlikely(err)) {
1735 		ip_flush_pending_frames(sk);
1736 		goto out;
1737 	}
1738 
1739 	nskb = skb_peek(&sk->sk_write_queue);
1740 	if (nskb) {
1741 		if (arg->csumoffset >= 0)
1742 			*((__sum16 *)skb_transport_header(nskb) +
1743 			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1744 								arg->csum));
1745 		nskb->ip_summed = CHECKSUM_NONE;
1746 		nskb->mono_delivery_time = !!transmit_time;
1747 		ip_push_pending_frames(sk, &fl4);
1748 	}
1749 out:
1750 	ip_rt_put(rt);
1751 }
1752 
1753 void __init ip_init(void)
1754 {
1755 	ip_rt_init();
1756 	inet_initpeers();
1757 
1758 #if defined(CONFIG_IP_MULTICAST)
1759 	igmp_mc_init();
1760 #endif
1761 }
1762