xref: /linux/net/ipv4/ip_output.c (revision 1a9239bb4253f9076b5b4b2a1a4e8d7defd77a95)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		The Internet Protocol (IP) output module.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Donald Becker, <becker@super.org>
12  *		Alan Cox, <Alan.Cox@linux.org>
13  *		Richard Underwood
14  *		Stefan Becker, <stefanb@yello.ping.de>
15  *		Jorge Cwik, <jorge@laser.satlink.net>
16  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
17  *		Hirokazu Takahashi, <taka@valinux.co.jp>
18  *
19  *	See ip_input.c for original log
20  *
21  *	Fixes:
22  *		Alan Cox	:	Missing nonblock feature in ip_build_xmit.
23  *		Mike Kilburn	:	htons() missing in ip_build_xmit.
24  *		Bradford Johnson:	Fix faulty handling of some frames when
25  *					no route is found.
26  *		Alexander Demenshin:	Missing sk/skb free in ip_queue_xmit
27  *					(in case if packet not accepted by
28  *					output firewall rules)
29  *		Mike McLagan	:	Routing by source
30  *		Alexey Kuznetsov:	use new route cache
31  *		Andi Kleen:		Fix broken PMTU recovery and remove
32  *					some redundant tests.
33  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
34  *		Andi Kleen	: 	Replace ip_reply with ip_send_reply.
35  *		Andi Kleen	:	Split fast and slow ip_build_xmit path
36  *					for decreased register pressure on x86
37  *					and more readability.
38  *		Marc Boucher	:	When call_out_firewall returns FW_QUEUE,
39  *					silently drop skb instead of failing with -EPERM.
40  *		Detlev Wengorz	:	Copy protocol for fragments.
41  *		Hirokazu Takahashi:	HW checksumming for outgoing UDP
42  *					datagrams.
43  *		Hirokazu Takahashi:	sendfile() on UDP works now.
44  */
45 
46 #include <linux/uaccess.h>
47 #include <linux/module.h>
48 #include <linux/types.h>
49 #include <linux/kernel.h>
50 #include <linux/mm.h>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
54 #include <linux/slab.h>
55 
56 #include <linux/socket.h>
57 #include <linux/sockios.h>
58 #include <linux/in.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/etherdevice.h>
62 #include <linux/proc_fs.h>
63 #include <linux/stat.h>
64 #include <linux/init.h>
65 
66 #include <net/snmp.h>
67 #include <net/ip.h>
68 #include <net/protocol.h>
69 #include <net/route.h>
70 #include <net/xfrm.h>
71 #include <linux/skbuff.h>
72 #include <net/sock.h>
73 #include <net/arp.h>
74 #include <net/icmp.h>
75 #include <net/checksum.h>
76 #include <net/gso.h>
77 #include <net/inetpeer.h>
78 #include <net/lwtunnel.h>
79 #include <net/inet_dscp.h>
80 #include <linux/bpf-cgroup.h>
81 #include <linux/igmp.h>
82 #include <linux/netfilter_ipv4.h>
83 #include <linux/netfilter_bridge.h>
84 #include <linux/netlink.h>
85 #include <linux/tcp.h>
86 
87 static int
88 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
89 	    unsigned int mtu,
90 	    int (*output)(struct net *, struct sock *, struct sk_buff *));
91 
92 /* Generate a checksum for an outgoing IP datagram. */
ip_send_check(struct iphdr * iph)93 void ip_send_check(struct iphdr *iph)
94 {
95 	iph->check = 0;
96 	iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
97 }
98 EXPORT_SYMBOL(ip_send_check);
99 
__ip_local_out(struct net * net,struct sock * sk,struct sk_buff * skb)100 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
101 {
102 	struct iphdr *iph = ip_hdr(skb);
103 
104 	IP_INC_STATS(net, IPSTATS_MIB_OUTREQUESTS);
105 
106 	iph_set_totlen(iph, skb->len);
107 	ip_send_check(iph);
108 
109 	/* if egress device is enslaved to an L3 master device pass the
110 	 * skb to its handler for processing
111 	 */
112 	skb = l3mdev_ip_out(sk, skb);
113 	if (unlikely(!skb))
114 		return 0;
115 
116 	skb->protocol = htons(ETH_P_IP);
117 
118 	return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
119 		       net, sk, skb, NULL, skb_dst(skb)->dev,
120 		       dst_output);
121 }
122 
ip_local_out(struct net * net,struct sock * sk,struct sk_buff * skb)123 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
124 {
125 	int err;
126 
127 	err = __ip_local_out(net, sk, skb);
128 	if (likely(err == 1))
129 		err = dst_output(net, sk, skb);
130 
131 	return err;
132 }
133 EXPORT_SYMBOL_GPL(ip_local_out);
134 
ip_select_ttl(const struct inet_sock * inet,const struct dst_entry * dst)135 static inline int ip_select_ttl(const struct inet_sock *inet,
136 				const struct dst_entry *dst)
137 {
138 	int ttl = READ_ONCE(inet->uc_ttl);
139 
140 	if (ttl < 0)
141 		ttl = ip4_dst_hoplimit(dst);
142 	return ttl;
143 }
144 
145 /*
146  *		Add an ip header to a skbuff and send it out.
147  *
148  */
ip_build_and_send_pkt(struct sk_buff * skb,const struct sock * sk,__be32 saddr,__be32 daddr,struct ip_options_rcu * opt,u8 tos)149 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
150 			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
151 			  u8 tos)
152 {
153 	const struct inet_sock *inet = inet_sk(sk);
154 	struct rtable *rt = skb_rtable(skb);
155 	struct net *net = sock_net(sk);
156 	struct iphdr *iph;
157 
158 	/* Build the IP header. */
159 	skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
160 	skb_reset_network_header(skb);
161 	iph = ip_hdr(skb);
162 	iph->version  = 4;
163 	iph->ihl      = 5;
164 	iph->tos      = tos;
165 	iph->ttl      = ip_select_ttl(inet, &rt->dst);
166 	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
167 	iph->saddr    = saddr;
168 	iph->protocol = sk->sk_protocol;
169 	/* Do not bother generating IPID for small packets (eg SYNACK) */
170 	if (skb->len <= IPV4_MIN_MTU || ip_dont_fragment(sk, &rt->dst)) {
171 		iph->frag_off = htons(IP_DF);
172 		iph->id = 0;
173 	} else {
174 		iph->frag_off = 0;
175 		/* TCP packets here are SYNACK with fat IPv4/TCP options.
176 		 * Avoid using the hashed IP ident generator.
177 		 */
178 		if (sk->sk_protocol == IPPROTO_TCP)
179 			iph->id = (__force __be16)get_random_u16();
180 		else
181 			__ip_select_ident(net, iph, 1);
182 	}
183 
184 	if (opt && opt->opt.optlen) {
185 		iph->ihl += opt->opt.optlen>>2;
186 		ip_options_build(skb, &opt->opt, daddr, rt);
187 	}
188 
189 	skb->priority = READ_ONCE(sk->sk_priority);
190 	if (!skb->mark)
191 		skb->mark = READ_ONCE(sk->sk_mark);
192 
193 	/* Send it out. */
194 	return ip_local_out(net, skb->sk, skb);
195 }
196 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
197 
ip_finish_output2(struct net * net,struct sock * sk,struct sk_buff * skb)198 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
199 {
200 	struct dst_entry *dst = skb_dst(skb);
201 	struct rtable *rt = dst_rtable(dst);
202 	struct net_device *dev = dst->dev;
203 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
204 	struct neighbour *neigh;
205 	bool is_v6gw = false;
206 
207 	if (rt->rt_type == RTN_MULTICAST) {
208 		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
209 	} else if (rt->rt_type == RTN_BROADCAST)
210 		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
211 
212 	/* OUTOCTETS should be counted after fragment */
213 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
214 
215 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
216 		skb = skb_expand_head(skb, hh_len);
217 		if (!skb)
218 			return -ENOMEM;
219 	}
220 
221 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
222 		int res = lwtunnel_xmit(skb);
223 
224 		if (res != LWTUNNEL_XMIT_CONTINUE)
225 			return res;
226 	}
227 
228 	rcu_read_lock();
229 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
230 	if (!IS_ERR(neigh)) {
231 		int res;
232 
233 		sock_confirm_neigh(skb, neigh);
234 		/* if crossing protocols, can not use the cached header */
235 		res = neigh_output(neigh, skb, is_v6gw);
236 		rcu_read_unlock();
237 		return res;
238 	}
239 	rcu_read_unlock();
240 
241 	net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
242 			    __func__);
243 	kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
244 	return PTR_ERR(neigh);
245 }
246 
ip_finish_output_gso(struct net * net,struct sock * sk,struct sk_buff * skb,unsigned int mtu)247 static int ip_finish_output_gso(struct net *net, struct sock *sk,
248 				struct sk_buff *skb, unsigned int mtu)
249 {
250 	struct sk_buff *segs, *nskb;
251 	netdev_features_t features;
252 	int ret = 0;
253 
254 	/* common case: seglen is <= mtu
255 	 */
256 	if (skb_gso_validate_network_len(skb, mtu))
257 		return ip_finish_output2(net, sk, skb);
258 
259 	/* Slowpath -  GSO segment length exceeds the egress MTU.
260 	 *
261 	 * This can happen in several cases:
262 	 *  - Forwarding of a TCP GRO skb, when DF flag is not set.
263 	 *  - Forwarding of an skb that arrived on a virtualization interface
264 	 *    (virtio-net/vhost/tap) with TSO/GSO size set by other network
265 	 *    stack.
266 	 *  - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
267 	 *    interface with a smaller MTU.
268 	 *  - Arriving GRO skb (or GSO skb in a virtualized environment) that is
269 	 *    bridged to a NETIF_F_TSO tunnel stacked over an interface with an
270 	 *    insufficient MTU.
271 	 */
272 	features = netif_skb_features(skb);
273 	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
274 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
275 	if (IS_ERR_OR_NULL(segs)) {
276 		kfree_skb(skb);
277 		return -ENOMEM;
278 	}
279 
280 	consume_skb(skb);
281 
282 	skb_list_walk_safe(segs, segs, nskb) {
283 		int err;
284 
285 		skb_mark_not_on_list(segs);
286 		err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
287 
288 		if (err && ret == 0)
289 			ret = err;
290 	}
291 
292 	return ret;
293 }
294 
__ip_finish_output(struct net * net,struct sock * sk,struct sk_buff * skb)295 static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
296 {
297 	unsigned int mtu;
298 
299 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
300 	/* Policy lookup after SNAT yielded a new policy */
301 	if (skb_dst(skb)->xfrm) {
302 		IPCB(skb)->flags |= IPSKB_REROUTED;
303 		return dst_output(net, sk, skb);
304 	}
305 #endif
306 	mtu = ip_skb_dst_mtu(sk, skb);
307 	if (skb_is_gso(skb))
308 		return ip_finish_output_gso(net, sk, skb, mtu);
309 
310 	if (skb->len > mtu || IPCB(skb)->frag_max_size)
311 		return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
312 
313 	return ip_finish_output2(net, sk, skb);
314 }
315 
ip_finish_output(struct net * net,struct sock * sk,struct sk_buff * skb)316 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
317 {
318 	int ret;
319 
320 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
321 	switch (ret) {
322 	case NET_XMIT_SUCCESS:
323 		return __ip_finish_output(net, sk, skb);
324 	case NET_XMIT_CN:
325 		return __ip_finish_output(net, sk, skb) ? : ret;
326 	default:
327 		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
328 		return ret;
329 	}
330 }
331 
ip_mc_finish_output(struct net * net,struct sock * sk,struct sk_buff * skb)332 static int ip_mc_finish_output(struct net *net, struct sock *sk,
333 			       struct sk_buff *skb)
334 {
335 	struct rtable *new_rt;
336 	bool do_cn = false;
337 	int ret, err;
338 
339 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
340 	switch (ret) {
341 	case NET_XMIT_CN:
342 		do_cn = true;
343 		fallthrough;
344 	case NET_XMIT_SUCCESS:
345 		break;
346 	default:
347 		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
348 		return ret;
349 	}
350 
351 	/* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
352 	 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
353 	 * see ipv4_pktinfo_prepare().
354 	 */
355 	new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
356 	if (new_rt) {
357 		new_rt->rt_iif = 0;
358 		skb_dst_drop(skb);
359 		skb_dst_set(skb, &new_rt->dst);
360 	}
361 
362 	err = dev_loopback_xmit(net, sk, skb);
363 	return (do_cn && err) ? ret : err;
364 }
365 
ip_mc_output(struct net * net,struct sock * sk,struct sk_buff * skb)366 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
367 {
368 	struct rtable *rt = skb_rtable(skb);
369 	struct net_device *dev = rt->dst.dev;
370 
371 	/*
372 	 *	If the indicated interface is up and running, send the packet.
373 	 */
374 	skb->dev = dev;
375 	skb->protocol = htons(ETH_P_IP);
376 
377 	/*
378 	 *	Multicasts are looped back for other local users
379 	 */
380 
381 	if (rt->rt_flags&RTCF_MULTICAST) {
382 		if (sk_mc_loop(sk)
383 #ifdef CONFIG_IP_MROUTE
384 		/* Small optimization: do not loopback not local frames,
385 		   which returned after forwarding; they will be  dropped
386 		   by ip_mr_input in any case.
387 		   Note, that local frames are looped back to be delivered
388 		   to local recipients.
389 
390 		   This check is duplicated in ip_mr_input at the moment.
391 		 */
392 		    &&
393 		    ((rt->rt_flags & RTCF_LOCAL) ||
394 		     !(IPCB(skb)->flags & IPSKB_FORWARDED))
395 #endif
396 		   ) {
397 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
398 			if (newskb)
399 				NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
400 					net, sk, newskb, NULL, newskb->dev,
401 					ip_mc_finish_output);
402 		}
403 
404 		/* Multicasts with ttl 0 must not go beyond the host */
405 
406 		if (ip_hdr(skb)->ttl == 0) {
407 			kfree_skb(skb);
408 			return 0;
409 		}
410 	}
411 
412 	if (rt->rt_flags&RTCF_BROADCAST) {
413 		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
414 		if (newskb)
415 			NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
416 				net, sk, newskb, NULL, newskb->dev,
417 				ip_mc_finish_output);
418 	}
419 
420 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
421 			    net, sk, skb, NULL, skb->dev,
422 			    ip_finish_output,
423 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
424 }
425 
ip_output(struct net * net,struct sock * sk,struct sk_buff * skb)426 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
427 {
428 	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
429 
430 	skb->dev = dev;
431 	skb->protocol = htons(ETH_P_IP);
432 
433 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
434 			    net, sk, skb, indev, dev,
435 			    ip_finish_output,
436 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
437 }
438 EXPORT_SYMBOL(ip_output);
439 
440 /*
441  * copy saddr and daddr, possibly using 64bit load/stores
442  * Equivalent to :
443  *   iph->saddr = fl4->saddr;
444  *   iph->daddr = fl4->daddr;
445  */
ip_copy_addrs(struct iphdr * iph,const struct flowi4 * fl4)446 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
447 {
448 	BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
449 		     offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
450 
451 	iph->saddr = fl4->saddr;
452 	iph->daddr = fl4->daddr;
453 }
454 
455 /* Note: skb->sk can be different from sk, in case of tunnels */
__ip_queue_xmit(struct sock * sk,struct sk_buff * skb,struct flowi * fl,__u8 tos)456 int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
457 		    __u8 tos)
458 {
459 	struct inet_sock *inet = inet_sk(sk);
460 	struct net *net = sock_net(sk);
461 	struct ip_options_rcu *inet_opt;
462 	struct flowi4 *fl4;
463 	struct rtable *rt;
464 	struct iphdr *iph;
465 	int res;
466 
467 	/* Skip all of this if the packet is already routed,
468 	 * f.e. by something like SCTP.
469 	 */
470 	rcu_read_lock();
471 	inet_opt = rcu_dereference(inet->inet_opt);
472 	fl4 = &fl->u.ip4;
473 	rt = skb_rtable(skb);
474 	if (rt)
475 		goto packet_routed;
476 
477 	/* Make sure we can route this packet. */
478 	rt = dst_rtable(__sk_dst_check(sk, 0));
479 	if (!rt) {
480 		inet_sk_init_flowi4(inet, fl4);
481 
482 		/* sctp_v4_xmit() uses its own DSCP value */
483 		fl4->flowi4_tos = tos & INET_DSCP_MASK;
484 
485 		/* If this fails, retransmit mechanism of transport layer will
486 		 * keep trying until route appears or the connection times
487 		 * itself out.
488 		 */
489 		rt = ip_route_output_flow(net, fl4, sk);
490 		if (IS_ERR(rt))
491 			goto no_route;
492 		sk_setup_caps(sk, &rt->dst);
493 	}
494 	skb_dst_set_noref(skb, &rt->dst);
495 
496 packet_routed:
497 	if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
498 		goto no_route;
499 
500 	/* OK, we know where to send it, allocate and build IP header. */
501 	skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
502 	skb_reset_network_header(skb);
503 	iph = ip_hdr(skb);
504 	*((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
505 	if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
506 		iph->frag_off = htons(IP_DF);
507 	else
508 		iph->frag_off = 0;
509 	iph->ttl      = ip_select_ttl(inet, &rt->dst);
510 	iph->protocol = sk->sk_protocol;
511 	ip_copy_addrs(iph, fl4);
512 
513 	/* Transport layer set skb->h.foo itself. */
514 
515 	if (inet_opt && inet_opt->opt.optlen) {
516 		iph->ihl += inet_opt->opt.optlen >> 2;
517 		ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt);
518 	}
519 
520 	ip_select_ident_segs(net, skb, sk,
521 			     skb_shinfo(skb)->gso_segs ?: 1);
522 
523 	/* TODO : should we use skb->sk here instead of sk ? */
524 	skb->priority = READ_ONCE(sk->sk_priority);
525 	skb->mark = READ_ONCE(sk->sk_mark);
526 
527 	res = ip_local_out(net, sk, skb);
528 	rcu_read_unlock();
529 	return res;
530 
531 no_route:
532 	rcu_read_unlock();
533 	IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
534 	kfree_skb_reason(skb, SKB_DROP_REASON_IP_OUTNOROUTES);
535 	return -EHOSTUNREACH;
536 }
537 EXPORT_SYMBOL(__ip_queue_xmit);
538 
ip_queue_xmit(struct sock * sk,struct sk_buff * skb,struct flowi * fl)539 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
540 {
541 	return __ip_queue_xmit(sk, skb, fl, READ_ONCE(inet_sk(sk)->tos));
542 }
543 EXPORT_SYMBOL(ip_queue_xmit);
544 
ip_copy_metadata(struct sk_buff * to,struct sk_buff * from)545 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
546 {
547 	to->pkt_type = from->pkt_type;
548 	to->priority = from->priority;
549 	to->protocol = from->protocol;
550 	to->skb_iif = from->skb_iif;
551 	skb_dst_drop(to);
552 	skb_dst_copy(to, from);
553 	to->dev = from->dev;
554 	to->mark = from->mark;
555 
556 	skb_copy_hash(to, from);
557 
558 #ifdef CONFIG_NET_SCHED
559 	to->tc_index = from->tc_index;
560 #endif
561 	nf_copy(to, from);
562 	skb_ext_copy(to, from);
563 #if IS_ENABLED(CONFIG_IP_VS)
564 	to->ipvs_property = from->ipvs_property;
565 #endif
566 	skb_copy_secmark(to, from);
567 }
568 
ip_fragment(struct net * net,struct sock * sk,struct sk_buff * skb,unsigned int mtu,int (* output)(struct net *,struct sock *,struct sk_buff *))569 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
570 		       unsigned int mtu,
571 		       int (*output)(struct net *, struct sock *, struct sk_buff *))
572 {
573 	struct iphdr *iph = ip_hdr(skb);
574 
575 	if ((iph->frag_off & htons(IP_DF)) == 0)
576 		return ip_do_fragment(net, sk, skb, output);
577 
578 	if (unlikely(!skb->ignore_df ||
579 		     (IPCB(skb)->frag_max_size &&
580 		      IPCB(skb)->frag_max_size > mtu))) {
581 		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
582 		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
583 			  htonl(mtu));
584 		kfree_skb(skb);
585 		return -EMSGSIZE;
586 	}
587 
588 	return ip_do_fragment(net, sk, skb, output);
589 }
590 
ip_fraglist_init(struct sk_buff * skb,struct iphdr * iph,unsigned int hlen,struct ip_fraglist_iter * iter)591 void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
592 		      unsigned int hlen, struct ip_fraglist_iter *iter)
593 {
594 	unsigned int first_len = skb_pagelen(skb);
595 
596 	iter->frag = skb_shinfo(skb)->frag_list;
597 	skb_frag_list_init(skb);
598 
599 	iter->offset = 0;
600 	iter->iph = iph;
601 	iter->hlen = hlen;
602 
603 	skb->data_len = first_len - skb_headlen(skb);
604 	skb->len = first_len;
605 	iph->tot_len = htons(first_len);
606 	iph->frag_off = htons(IP_MF);
607 	ip_send_check(iph);
608 }
609 EXPORT_SYMBOL(ip_fraglist_init);
610 
ip_fraglist_prepare(struct sk_buff * skb,struct ip_fraglist_iter * iter)611 void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
612 {
613 	unsigned int hlen = iter->hlen;
614 	struct iphdr *iph = iter->iph;
615 	struct sk_buff *frag;
616 
617 	frag = iter->frag;
618 	frag->ip_summed = CHECKSUM_NONE;
619 	skb_reset_transport_header(frag);
620 	__skb_push(frag, hlen);
621 	skb_reset_network_header(frag);
622 	memcpy(skb_network_header(frag), iph, hlen);
623 	iter->iph = ip_hdr(frag);
624 	iph = iter->iph;
625 	iph->tot_len = htons(frag->len);
626 	ip_copy_metadata(frag, skb);
627 	iter->offset += skb->len - hlen;
628 	iph->frag_off = htons(iter->offset >> 3);
629 	if (frag->next)
630 		iph->frag_off |= htons(IP_MF);
631 	/* Ready, complete checksum */
632 	ip_send_check(iph);
633 }
634 EXPORT_SYMBOL(ip_fraglist_prepare);
635 
ip_frag_init(struct sk_buff * skb,unsigned int hlen,unsigned int ll_rs,unsigned int mtu,bool DF,struct ip_frag_state * state)636 void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
637 		  unsigned int ll_rs, unsigned int mtu, bool DF,
638 		  struct ip_frag_state *state)
639 {
640 	struct iphdr *iph = ip_hdr(skb);
641 
642 	state->DF = DF;
643 	state->hlen = hlen;
644 	state->ll_rs = ll_rs;
645 	state->mtu = mtu;
646 
647 	state->left = skb->len - hlen;	/* Space per frame */
648 	state->ptr = hlen;		/* Where to start from */
649 
650 	state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
651 	state->not_last_frag = iph->frag_off & htons(IP_MF);
652 }
653 EXPORT_SYMBOL(ip_frag_init);
654 
ip_frag_ipcb(struct sk_buff * from,struct sk_buff * to,bool first_frag)655 static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
656 			 bool first_frag)
657 {
658 	/* Copy the flags to each fragment. */
659 	IPCB(to)->flags = IPCB(from)->flags;
660 
661 	/* ANK: dirty, but effective trick. Upgrade options only if
662 	 * the segment to be fragmented was THE FIRST (otherwise,
663 	 * options are already fixed) and make it ONCE
664 	 * on the initial skb, so that all the following fragments
665 	 * will inherit fixed options.
666 	 */
667 	if (first_frag)
668 		ip_options_fragment(from);
669 }
670 
ip_frag_next(struct sk_buff * skb,struct ip_frag_state * state)671 struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
672 {
673 	unsigned int len = state->left;
674 	struct sk_buff *skb2;
675 	struct iphdr *iph;
676 
677 	/* IF: it doesn't fit, use 'mtu' - the data space left */
678 	if (len > state->mtu)
679 		len = state->mtu;
680 	/* IF: we are not sending up to and including the packet end
681 	   then align the next start on an eight byte boundary */
682 	if (len < state->left)	{
683 		len &= ~7;
684 	}
685 
686 	/* Allocate buffer */
687 	skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
688 	if (!skb2)
689 		return ERR_PTR(-ENOMEM);
690 
691 	/*
692 	 *	Set up data on packet
693 	 */
694 
695 	ip_copy_metadata(skb2, skb);
696 	skb_reserve(skb2, state->ll_rs);
697 	skb_put(skb2, len + state->hlen);
698 	skb_reset_network_header(skb2);
699 	skb2->transport_header = skb2->network_header + state->hlen;
700 
701 	/*
702 	 *	Charge the memory for the fragment to any owner
703 	 *	it might possess
704 	 */
705 
706 	if (skb->sk)
707 		skb_set_owner_w(skb2, skb->sk);
708 
709 	/*
710 	 *	Copy the packet header into the new buffer.
711 	 */
712 
713 	skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
714 
715 	/*
716 	 *	Copy a block of the IP datagram.
717 	 */
718 	if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
719 		BUG();
720 	state->left -= len;
721 
722 	/*
723 	 *	Fill in the new header fields.
724 	 */
725 	iph = ip_hdr(skb2);
726 	iph->frag_off = htons((state->offset >> 3));
727 	if (state->DF)
728 		iph->frag_off |= htons(IP_DF);
729 
730 	/*
731 	 *	Added AC : If we are fragmenting a fragment that's not the
732 	 *		   last fragment then keep MF on each bit
733 	 */
734 	if (state->left > 0 || state->not_last_frag)
735 		iph->frag_off |= htons(IP_MF);
736 	state->ptr += len;
737 	state->offset += len;
738 
739 	iph->tot_len = htons(len + state->hlen);
740 
741 	ip_send_check(iph);
742 
743 	return skb2;
744 }
745 EXPORT_SYMBOL(ip_frag_next);
746 
747 /*
748  *	This IP datagram is too large to be sent in one piece.  Break it up into
749  *	smaller pieces (each of size equal to IP header plus
750  *	a block of the data of the original IP data part) that will yet fit in a
751  *	single device frame, and queue such a frame for sending.
752  */
753 
ip_do_fragment(struct net * net,struct sock * sk,struct sk_buff * skb,int (* output)(struct net *,struct sock *,struct sk_buff *))754 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
755 		   int (*output)(struct net *, struct sock *, struct sk_buff *))
756 {
757 	struct iphdr *iph;
758 	struct sk_buff *skb2;
759 	u8 tstamp_type = skb->tstamp_type;
760 	struct rtable *rt = skb_rtable(skb);
761 	unsigned int mtu, hlen, ll_rs;
762 	struct ip_fraglist_iter iter;
763 	ktime_t tstamp = skb->tstamp;
764 	struct ip_frag_state state;
765 	int err = 0;
766 
767 	/* for offloaded checksums cleanup checksum before fragmentation */
768 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
769 	    (err = skb_checksum_help(skb)))
770 		goto fail;
771 
772 	/*
773 	 *	Point into the IP datagram header.
774 	 */
775 
776 	iph = ip_hdr(skb);
777 
778 	mtu = ip_skb_dst_mtu(sk, skb);
779 	if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
780 		mtu = IPCB(skb)->frag_max_size;
781 
782 	/*
783 	 *	Setup starting values.
784 	 */
785 
786 	hlen = iph->ihl * 4;
787 	mtu = mtu - hlen;	/* Size of data space */
788 	IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
789 	ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
790 
791 	/* When frag_list is given, use it. First, check its validity:
792 	 * some transformers could create wrong frag_list or break existing
793 	 * one, it is not prohibited. In this case fall back to copying.
794 	 *
795 	 * LATER: this step can be merged to real generation of fragments,
796 	 * we can switch to copy when see the first bad fragment.
797 	 */
798 	if (skb_has_frag_list(skb)) {
799 		struct sk_buff *frag, *frag2;
800 		unsigned int first_len = skb_pagelen(skb);
801 
802 		if (first_len - hlen > mtu ||
803 		    ((first_len - hlen) & 7) ||
804 		    ip_is_fragment(iph) ||
805 		    skb_cloned(skb) ||
806 		    skb_headroom(skb) < ll_rs)
807 			goto slow_path;
808 
809 		skb_walk_frags(skb, frag) {
810 			/* Correct geometry. */
811 			if (frag->len > mtu ||
812 			    ((frag->len & 7) && frag->next) ||
813 			    skb_headroom(frag) < hlen + ll_rs)
814 				goto slow_path_clean;
815 
816 			/* Partially cloned skb? */
817 			if (skb_shared(frag))
818 				goto slow_path_clean;
819 
820 			BUG_ON(frag->sk);
821 			if (skb->sk) {
822 				frag->sk = skb->sk;
823 				frag->destructor = sock_wfree;
824 			}
825 			skb->truesize -= frag->truesize;
826 		}
827 
828 		/* Everything is OK. Generate! */
829 		ip_fraglist_init(skb, iph, hlen, &iter);
830 
831 		for (;;) {
832 			/* Prepare header of the next frame,
833 			 * before previous one went down. */
834 			if (iter.frag) {
835 				bool first_frag = (iter.offset == 0);
836 
837 				IPCB(iter.frag)->flags = IPCB(skb)->flags;
838 				ip_fraglist_prepare(skb, &iter);
839 				if (first_frag && IPCB(skb)->opt.optlen) {
840 					/* ipcb->opt is not populated for frags
841 					 * coming from __ip_make_skb(),
842 					 * ip_options_fragment() needs optlen
843 					 */
844 					IPCB(iter.frag)->opt.optlen =
845 						IPCB(skb)->opt.optlen;
846 					ip_options_fragment(iter.frag);
847 					ip_send_check(iter.iph);
848 				}
849 			}
850 
851 			skb_set_delivery_time(skb, tstamp, tstamp_type);
852 			err = output(net, sk, skb);
853 
854 			if (!err)
855 				IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
856 			if (err || !iter.frag)
857 				break;
858 
859 			skb = ip_fraglist_next(&iter);
860 		}
861 
862 		if (err == 0) {
863 			IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
864 			return 0;
865 		}
866 
867 		kfree_skb_list(iter.frag);
868 
869 		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
870 		return err;
871 
872 slow_path_clean:
873 		skb_walk_frags(skb, frag2) {
874 			if (frag2 == frag)
875 				break;
876 			frag2->sk = NULL;
877 			frag2->destructor = NULL;
878 			skb->truesize += frag2->truesize;
879 		}
880 	}
881 
882 slow_path:
883 	/*
884 	 *	Fragment the datagram.
885 	 */
886 
887 	ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
888 		     &state);
889 
890 	/*
891 	 *	Keep copying data until we run out.
892 	 */
893 
894 	while (state.left > 0) {
895 		bool first_frag = (state.offset == 0);
896 
897 		skb2 = ip_frag_next(skb, &state);
898 		if (IS_ERR(skb2)) {
899 			err = PTR_ERR(skb2);
900 			goto fail;
901 		}
902 		ip_frag_ipcb(skb, skb2, first_frag);
903 
904 		/*
905 		 *	Put this fragment into the sending queue.
906 		 */
907 		skb_set_delivery_time(skb2, tstamp, tstamp_type);
908 		err = output(net, sk, skb2);
909 		if (err)
910 			goto fail;
911 
912 		IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
913 	}
914 	consume_skb(skb);
915 	IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
916 	return err;
917 
918 fail:
919 	kfree_skb(skb);
920 	IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
921 	return err;
922 }
923 EXPORT_SYMBOL(ip_do_fragment);
924 
925 int
ip_generic_getfrag(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb)926 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
927 {
928 	struct msghdr *msg = from;
929 
930 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
931 		if (!copy_from_iter_full(to, len, &msg->msg_iter))
932 			return -EFAULT;
933 	} else {
934 		__wsum csum = 0;
935 		if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
936 			return -EFAULT;
937 		skb->csum = csum_block_add(skb->csum, csum, odd);
938 	}
939 	return 0;
940 }
941 EXPORT_SYMBOL(ip_generic_getfrag);
942 
__ip_append_data(struct sock * sk,struct flowi4 * fl4,struct sk_buff_head * queue,struct inet_cork * cork,struct page_frag * pfrag,int getfrag (void * from,char * to,int offset,int len,int odd,struct sk_buff * skb),void * from,int length,int transhdrlen,unsigned int flags)943 static int __ip_append_data(struct sock *sk,
944 			    struct flowi4 *fl4,
945 			    struct sk_buff_head *queue,
946 			    struct inet_cork *cork,
947 			    struct page_frag *pfrag,
948 			    int getfrag(void *from, char *to, int offset,
949 					int len, int odd, struct sk_buff *skb),
950 			    void *from, int length, int transhdrlen,
951 			    unsigned int flags)
952 {
953 	struct inet_sock *inet = inet_sk(sk);
954 	struct ubuf_info *uarg = NULL;
955 	struct sk_buff *skb;
956 	struct ip_options *opt = cork->opt;
957 	int hh_len;
958 	int exthdrlen;
959 	int mtu;
960 	int copy;
961 	int err;
962 	int offset = 0;
963 	bool zc = false;
964 	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
965 	int csummode = CHECKSUM_NONE;
966 	struct rtable *rt = dst_rtable(cork->dst);
967 	bool paged, hold_tskey = false, extra_uref = false;
968 	unsigned int wmem_alloc_delta = 0;
969 	u32 tskey = 0;
970 
971 	skb = skb_peek_tail(queue);
972 
973 	exthdrlen = !skb ? rt->dst.header_len : 0;
974 	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
975 	paged = !!cork->gso_size;
976 
977 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
978 
979 	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
980 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
981 	maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
982 
983 	if (cork->length + length > maxnonfragsize - fragheaderlen) {
984 		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
985 			       mtu - (opt ? opt->optlen : 0));
986 		return -EMSGSIZE;
987 	}
988 
989 	/*
990 	 * transhdrlen > 0 means that this is the first fragment and we wish
991 	 * it won't be fragmented in the future.
992 	 */
993 	if (transhdrlen &&
994 	    length + fragheaderlen <= mtu &&
995 	    rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
996 	    (!(flags & MSG_MORE) || cork->gso_size) &&
997 	    (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
998 		csummode = CHECKSUM_PARTIAL;
999 
1000 	if ((flags & MSG_ZEROCOPY) && length) {
1001 		struct msghdr *msg = from;
1002 
1003 		if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1004 			if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1005 				return -EINVAL;
1006 
1007 			/* Leave uarg NULL if can't zerocopy, callers should
1008 			 * be able to handle it.
1009 			 */
1010 			if ((rt->dst.dev->features & NETIF_F_SG) &&
1011 			    csummode == CHECKSUM_PARTIAL) {
1012 				paged = true;
1013 				zc = true;
1014 				uarg = msg->msg_ubuf;
1015 			}
1016 		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1017 			uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1018 			if (!uarg)
1019 				return -ENOBUFS;
1020 			extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1021 			if (rt->dst.dev->features & NETIF_F_SG &&
1022 			    csummode == CHECKSUM_PARTIAL) {
1023 				paged = true;
1024 				zc = true;
1025 			} else {
1026 				uarg_to_msgzc(uarg)->zerocopy = 0;
1027 				skb_zcopy_set(skb, uarg, &extra_uref);
1028 			}
1029 		}
1030 	} else if ((flags & MSG_SPLICE_PAGES) && length) {
1031 		if (inet_test_bit(HDRINCL, sk))
1032 			return -EPERM;
1033 		if (rt->dst.dev->features & NETIF_F_SG &&
1034 		    getfrag == ip_generic_getfrag)
1035 			/* We need an empty buffer to attach stuff to */
1036 			paged = true;
1037 		else
1038 			flags &= ~MSG_SPLICE_PAGES;
1039 	}
1040 
1041 	cork->length += length;
1042 
1043 	if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
1044 	    READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
1045 		if (cork->flags & IPCORK_TS_OPT_ID) {
1046 			tskey = cork->ts_opt_id;
1047 		} else {
1048 			tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1049 			hold_tskey = true;
1050 		}
1051 	}
1052 
1053 	/* So, what's going on in the loop below?
1054 	 *
1055 	 * We use calculated fragment length to generate chained skb,
1056 	 * each of segments is IP fragment ready for sending to network after
1057 	 * adding appropriate IP header.
1058 	 */
1059 
1060 	if (!skb)
1061 		goto alloc_new_skb;
1062 
1063 	while (length > 0) {
1064 		/* Check if the remaining data fits into current packet. */
1065 		copy = mtu - skb->len;
1066 		if (copy < length)
1067 			copy = maxfraglen - skb->len;
1068 		if (copy <= 0) {
1069 			char *data;
1070 			unsigned int datalen;
1071 			unsigned int fraglen;
1072 			unsigned int fraggap;
1073 			unsigned int alloclen, alloc_extra;
1074 			unsigned int pagedlen;
1075 			struct sk_buff *skb_prev;
1076 alloc_new_skb:
1077 			skb_prev = skb;
1078 			if (skb_prev)
1079 				fraggap = skb_prev->len - maxfraglen;
1080 			else
1081 				fraggap = 0;
1082 
1083 			/*
1084 			 * If remaining data exceeds the mtu,
1085 			 * we know we need more fragment(s).
1086 			 */
1087 			datalen = length + fraggap;
1088 			if (datalen > mtu - fragheaderlen)
1089 				datalen = maxfraglen - fragheaderlen;
1090 			fraglen = datalen + fragheaderlen;
1091 			pagedlen = 0;
1092 
1093 			alloc_extra = hh_len + 15;
1094 			alloc_extra += exthdrlen;
1095 
1096 			/* The last fragment gets additional space at tail.
1097 			 * Note, with MSG_MORE we overallocate on fragments,
1098 			 * because we have no idea what fragment will be
1099 			 * the last.
1100 			 */
1101 			if (datalen == length + fraggap)
1102 				alloc_extra += rt->dst.trailer_len;
1103 
1104 			if ((flags & MSG_MORE) &&
1105 			    !(rt->dst.dev->features&NETIF_F_SG))
1106 				alloclen = mtu;
1107 			else if (!paged &&
1108 				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1109 				  !(rt->dst.dev->features & NETIF_F_SG)))
1110 				alloclen = fraglen;
1111 			else {
1112 				alloclen = fragheaderlen + transhdrlen;
1113 				pagedlen = datalen - transhdrlen;
1114 			}
1115 
1116 			alloclen += alloc_extra;
1117 
1118 			if (transhdrlen) {
1119 				skb = sock_alloc_send_skb(sk, alloclen,
1120 						(flags & MSG_DONTWAIT), &err);
1121 			} else {
1122 				skb = NULL;
1123 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1124 				    2 * sk->sk_sndbuf)
1125 					skb = alloc_skb(alloclen,
1126 							sk->sk_allocation);
1127 				if (unlikely(!skb))
1128 					err = -ENOBUFS;
1129 			}
1130 			if (!skb)
1131 				goto error;
1132 
1133 			/*
1134 			 *	Fill in the control structures
1135 			 */
1136 			skb->ip_summed = csummode;
1137 			skb->csum = 0;
1138 			skb_reserve(skb, hh_len);
1139 
1140 			/*
1141 			 *	Find where to start putting bytes.
1142 			 */
1143 			data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1144 			skb_set_network_header(skb, exthdrlen);
1145 			skb->transport_header = (skb->network_header +
1146 						 fragheaderlen);
1147 			data += fragheaderlen + exthdrlen;
1148 
1149 			if (fraggap) {
1150 				skb->csum = skb_copy_and_csum_bits(
1151 					skb_prev, maxfraglen,
1152 					data + transhdrlen, fraggap);
1153 				skb_prev->csum = csum_sub(skb_prev->csum,
1154 							  skb->csum);
1155 				data += fraggap;
1156 				pskb_trim_unique(skb_prev, maxfraglen);
1157 			}
1158 
1159 			copy = datalen - transhdrlen - fraggap - pagedlen;
1160 			/* [!] NOTE: copy will be negative if pagedlen>0
1161 			 * because then the equation reduces to -fraggap.
1162 			 */
1163 			if (copy > 0 &&
1164 			    INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1165 					    from, data + transhdrlen, offset,
1166 					    copy, fraggap, skb) < 0) {
1167 				err = -EFAULT;
1168 				kfree_skb(skb);
1169 				goto error;
1170 			} else if (flags & MSG_SPLICE_PAGES) {
1171 				copy = 0;
1172 			}
1173 
1174 			offset += copy;
1175 			length -= copy + transhdrlen;
1176 			transhdrlen = 0;
1177 			exthdrlen = 0;
1178 			csummode = CHECKSUM_NONE;
1179 
1180 			/* only the initial fragment is time stamped */
1181 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1182 			cork->tx_flags = 0;
1183 			skb_shinfo(skb)->tskey = tskey;
1184 			tskey = 0;
1185 			skb_zcopy_set(skb, uarg, &extra_uref);
1186 
1187 			if ((flags & MSG_CONFIRM) && !skb_prev)
1188 				skb_set_dst_pending_confirm(skb, 1);
1189 
1190 			/*
1191 			 * Put the packet on the pending queue.
1192 			 */
1193 			if (!skb->destructor) {
1194 				skb->destructor = sock_wfree;
1195 				skb->sk = sk;
1196 				wmem_alloc_delta += skb->truesize;
1197 			}
1198 			__skb_queue_tail(queue, skb);
1199 			continue;
1200 		}
1201 
1202 		if (copy > length)
1203 			copy = length;
1204 
1205 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1206 		    skb_tailroom(skb) >= copy) {
1207 			unsigned int off;
1208 
1209 			off = skb->len;
1210 			if (INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1211 					    from, skb_put(skb, copy),
1212 					    offset, copy, off, skb) < 0) {
1213 				__skb_trim(skb, off);
1214 				err = -EFAULT;
1215 				goto error;
1216 			}
1217 		} else if (flags & MSG_SPLICE_PAGES) {
1218 			struct msghdr *msg = from;
1219 
1220 			err = -EIO;
1221 			if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1222 				goto error;
1223 
1224 			err = skb_splice_from_iter(skb, &msg->msg_iter, copy,
1225 						   sk->sk_allocation);
1226 			if (err < 0)
1227 				goto error;
1228 			copy = err;
1229 			wmem_alloc_delta += copy;
1230 		} else if (!zc) {
1231 			int i = skb_shinfo(skb)->nr_frags;
1232 
1233 			err = -ENOMEM;
1234 			if (!sk_page_frag_refill(sk, pfrag))
1235 				goto error;
1236 
1237 			skb_zcopy_downgrade_managed(skb);
1238 			if (!skb_can_coalesce(skb, i, pfrag->page,
1239 					      pfrag->offset)) {
1240 				err = -EMSGSIZE;
1241 				if (i == MAX_SKB_FRAGS)
1242 					goto error;
1243 
1244 				__skb_fill_page_desc(skb, i, pfrag->page,
1245 						     pfrag->offset, 0);
1246 				skb_shinfo(skb)->nr_frags = ++i;
1247 				get_page(pfrag->page);
1248 			}
1249 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1250 			if (INDIRECT_CALL_1(getfrag, ip_generic_getfrag,
1251 				    from,
1252 				    page_address(pfrag->page) + pfrag->offset,
1253 				    offset, copy, skb->len, skb) < 0)
1254 				goto error_efault;
1255 
1256 			pfrag->offset += copy;
1257 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1258 			skb_len_add(skb, copy);
1259 			wmem_alloc_delta += copy;
1260 		} else {
1261 			err = skb_zerocopy_iter_dgram(skb, from, copy);
1262 			if (err < 0)
1263 				goto error;
1264 		}
1265 		offset += copy;
1266 		length -= copy;
1267 	}
1268 
1269 	if (wmem_alloc_delta)
1270 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1271 	return 0;
1272 
1273 error_efault:
1274 	err = -EFAULT;
1275 error:
1276 	net_zcopy_put_abort(uarg, extra_uref);
1277 	cork->length -= length;
1278 	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1279 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1280 	if (hold_tskey)
1281 		atomic_dec(&sk->sk_tskey);
1282 	return err;
1283 }
1284 
ip_setup_cork(struct sock * sk,struct inet_cork * cork,struct ipcm_cookie * ipc,struct rtable ** rtp)1285 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1286 			 struct ipcm_cookie *ipc, struct rtable **rtp)
1287 {
1288 	struct ip_options_rcu *opt;
1289 	struct rtable *rt;
1290 
1291 	rt = *rtp;
1292 	if (unlikely(!rt))
1293 		return -EFAULT;
1294 
1295 	cork->fragsize = ip_sk_use_pmtu(sk) ?
1296 			 dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1297 
1298 	if (!inetdev_valid_mtu(cork->fragsize))
1299 		return -ENETUNREACH;
1300 
1301 	/*
1302 	 * setup for corking.
1303 	 */
1304 	opt = ipc->opt;
1305 	if (opt) {
1306 		if (!cork->opt) {
1307 			cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1308 					    sk->sk_allocation);
1309 			if (unlikely(!cork->opt))
1310 				return -ENOBUFS;
1311 		}
1312 		memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1313 		cork->flags |= IPCORK_OPT;
1314 		cork->addr = ipc->addr;
1315 	}
1316 
1317 	cork->gso_size = ipc->gso_size;
1318 
1319 	cork->dst = &rt->dst;
1320 	/* We stole this route, caller should not release it. */
1321 	*rtp = NULL;
1322 
1323 	cork->length = 0;
1324 	cork->ttl = ipc->ttl;
1325 	cork->tos = ipc->tos;
1326 	cork->mark = ipc->sockc.mark;
1327 	cork->priority = ipc->sockc.priority;
1328 	cork->transmit_time = ipc->sockc.transmit_time;
1329 	cork->tx_flags = 0;
1330 	sock_tx_timestamp(sk, &ipc->sockc, &cork->tx_flags);
1331 	if (ipc->sockc.tsflags & SOCKCM_FLAG_TS_OPT_ID) {
1332 		cork->flags |= IPCORK_TS_OPT_ID;
1333 		cork->ts_opt_id = ipc->sockc.ts_opt_id;
1334 	}
1335 
1336 	return 0;
1337 }
1338 
1339 /*
1340  *	ip_append_data() can make one large IP datagram from many pieces of
1341  *	data.  Each piece will be held on the socket until
1342  *	ip_push_pending_frames() is called. Each piece can be a page or
1343  *	non-page data.
1344  *
1345  *	Not only UDP, other transport protocols - e.g. raw sockets - can use
1346  *	this interface potentially.
1347  *
1348  *	LATER: length must be adjusted by pad at tail, when it is required.
1349  */
ip_append_data(struct sock * sk,struct flowi4 * fl4,int getfrag (void * from,char * to,int offset,int len,int odd,struct sk_buff * skb),void * from,int length,int transhdrlen,struct ipcm_cookie * ipc,struct rtable ** rtp,unsigned int flags)1350 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1351 		   int getfrag(void *from, char *to, int offset, int len,
1352 			       int odd, struct sk_buff *skb),
1353 		   void *from, int length, int transhdrlen,
1354 		   struct ipcm_cookie *ipc, struct rtable **rtp,
1355 		   unsigned int flags)
1356 {
1357 	struct inet_sock *inet = inet_sk(sk);
1358 	int err;
1359 
1360 	if (flags&MSG_PROBE)
1361 		return 0;
1362 
1363 	if (skb_queue_empty(&sk->sk_write_queue)) {
1364 		err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1365 		if (err)
1366 			return err;
1367 	} else {
1368 		transhdrlen = 0;
1369 	}
1370 
1371 	return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1372 				sk_page_frag(sk), getfrag,
1373 				from, length, transhdrlen, flags);
1374 }
1375 
ip_cork_release(struct inet_cork * cork)1376 static void ip_cork_release(struct inet_cork *cork)
1377 {
1378 	cork->flags &= ~IPCORK_OPT;
1379 	kfree(cork->opt);
1380 	cork->opt = NULL;
1381 	dst_release(cork->dst);
1382 	cork->dst = NULL;
1383 }
1384 
1385 /*
1386  *	Combined all pending IP fragments on the socket as one IP datagram
1387  *	and push them out.
1388  */
__ip_make_skb(struct sock * sk,struct flowi4 * fl4,struct sk_buff_head * queue,struct inet_cork * cork)1389 struct sk_buff *__ip_make_skb(struct sock *sk,
1390 			      struct flowi4 *fl4,
1391 			      struct sk_buff_head *queue,
1392 			      struct inet_cork *cork)
1393 {
1394 	struct sk_buff *skb, *tmp_skb;
1395 	struct sk_buff **tail_skb;
1396 	struct inet_sock *inet = inet_sk(sk);
1397 	struct net *net = sock_net(sk);
1398 	struct ip_options *opt = NULL;
1399 	struct rtable *rt = dst_rtable(cork->dst);
1400 	struct iphdr *iph;
1401 	u8 pmtudisc, ttl;
1402 	__be16 df = 0;
1403 
1404 	skb = __skb_dequeue(queue);
1405 	if (!skb)
1406 		goto out;
1407 	tail_skb = &(skb_shinfo(skb)->frag_list);
1408 
1409 	/* move skb->data to ip header from ext header */
1410 	if (skb->data < skb_network_header(skb))
1411 		__skb_pull(skb, skb_network_offset(skb));
1412 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1413 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1414 		*tail_skb = tmp_skb;
1415 		tail_skb = &(tmp_skb->next);
1416 		skb->len += tmp_skb->len;
1417 		skb->data_len += tmp_skb->len;
1418 		skb->truesize += tmp_skb->truesize;
1419 		tmp_skb->destructor = NULL;
1420 		tmp_skb->sk = NULL;
1421 	}
1422 
1423 	/* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1424 	 * to fragment the frame generated here. No matter, what transforms
1425 	 * how transforms change size of the packet, it will come out.
1426 	 */
1427 	skb->ignore_df = ip_sk_ignore_df(sk);
1428 
1429 	/* DF bit is set when we want to see DF on outgoing frames.
1430 	 * If ignore_df is set too, we still allow to fragment this frame
1431 	 * locally. */
1432 	pmtudisc = READ_ONCE(inet->pmtudisc);
1433 	if (pmtudisc == IP_PMTUDISC_DO ||
1434 	    pmtudisc == IP_PMTUDISC_PROBE ||
1435 	    (skb->len <= dst_mtu(&rt->dst) &&
1436 	     ip_dont_fragment(sk, &rt->dst)))
1437 		df = htons(IP_DF);
1438 
1439 	if (cork->flags & IPCORK_OPT)
1440 		opt = cork->opt;
1441 
1442 	if (cork->ttl != 0)
1443 		ttl = cork->ttl;
1444 	else if (rt->rt_type == RTN_MULTICAST)
1445 		ttl = READ_ONCE(inet->mc_ttl);
1446 	else
1447 		ttl = ip_select_ttl(inet, &rt->dst);
1448 
1449 	iph = ip_hdr(skb);
1450 	iph->version = 4;
1451 	iph->ihl = 5;
1452 	iph->tos = (cork->tos != -1) ? cork->tos : READ_ONCE(inet->tos);
1453 	iph->frag_off = df;
1454 	iph->ttl = ttl;
1455 	iph->protocol = sk->sk_protocol;
1456 	ip_copy_addrs(iph, fl4);
1457 	ip_select_ident(net, skb, sk);
1458 
1459 	if (opt) {
1460 		iph->ihl += opt->optlen >> 2;
1461 		ip_options_build(skb, opt, cork->addr, rt);
1462 	}
1463 
1464 	skb->priority = cork->priority;
1465 	skb->mark = cork->mark;
1466 	if (sk_is_tcp(sk))
1467 		skb_set_delivery_time(skb, cork->transmit_time, SKB_CLOCK_MONOTONIC);
1468 	else
1469 		skb_set_delivery_type_by_clockid(skb, cork->transmit_time, sk->sk_clockid);
1470 	/*
1471 	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1472 	 * on dst refcount
1473 	 */
1474 	cork->dst = NULL;
1475 	skb_dst_set(skb, &rt->dst);
1476 
1477 	if (iph->protocol == IPPROTO_ICMP) {
1478 		u8 icmp_type;
1479 
1480 		/* For such sockets, transhdrlen is zero when do ip_append_data(),
1481 		 * so icmphdr does not in skb linear region and can not get icmp_type
1482 		 * by icmp_hdr(skb)->type.
1483 		 */
1484 		if (sk->sk_type == SOCK_RAW &&
1485 		    !(fl4->flowi4_flags & FLOWI_FLAG_KNOWN_NH))
1486 			icmp_type = fl4->fl4_icmp_type;
1487 		else
1488 			icmp_type = icmp_hdr(skb)->type;
1489 		icmp_out_count(net, icmp_type);
1490 	}
1491 
1492 	ip_cork_release(cork);
1493 out:
1494 	return skb;
1495 }
1496 
ip_send_skb(struct net * net,struct sk_buff * skb)1497 int ip_send_skb(struct net *net, struct sk_buff *skb)
1498 {
1499 	int err;
1500 
1501 	err = ip_local_out(net, skb->sk, skb);
1502 	if (err) {
1503 		if (err > 0)
1504 			err = net_xmit_errno(err);
1505 		if (err)
1506 			IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1507 	}
1508 
1509 	return err;
1510 }
1511 
ip_push_pending_frames(struct sock * sk,struct flowi4 * fl4)1512 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1513 {
1514 	struct sk_buff *skb;
1515 
1516 	skb = ip_finish_skb(sk, fl4);
1517 	if (!skb)
1518 		return 0;
1519 
1520 	/* Netfilter gets whole the not fragmented skb. */
1521 	return ip_send_skb(sock_net(sk), skb);
1522 }
1523 
1524 /*
1525  *	Throw away all pending data on the socket.
1526  */
__ip_flush_pending_frames(struct sock * sk,struct sk_buff_head * queue,struct inet_cork * cork)1527 static void __ip_flush_pending_frames(struct sock *sk,
1528 				      struct sk_buff_head *queue,
1529 				      struct inet_cork *cork)
1530 {
1531 	struct sk_buff *skb;
1532 
1533 	while ((skb = __skb_dequeue_tail(queue)) != NULL)
1534 		kfree_skb(skb);
1535 
1536 	ip_cork_release(cork);
1537 }
1538 
ip_flush_pending_frames(struct sock * sk)1539 void ip_flush_pending_frames(struct sock *sk)
1540 {
1541 	__ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1542 }
1543 
ip_make_skb(struct sock * sk,struct flowi4 * fl4,int getfrag (void * from,char * to,int offset,int len,int odd,struct sk_buff * skb),void * from,int length,int transhdrlen,struct ipcm_cookie * ipc,struct rtable ** rtp,struct inet_cork * cork,unsigned int flags)1544 struct sk_buff *ip_make_skb(struct sock *sk,
1545 			    struct flowi4 *fl4,
1546 			    int getfrag(void *from, char *to, int offset,
1547 					int len, int odd, struct sk_buff *skb),
1548 			    void *from, int length, int transhdrlen,
1549 			    struct ipcm_cookie *ipc, struct rtable **rtp,
1550 			    struct inet_cork *cork, unsigned int flags)
1551 {
1552 	struct sk_buff_head queue;
1553 	int err;
1554 
1555 	if (flags & MSG_PROBE)
1556 		return NULL;
1557 
1558 	__skb_queue_head_init(&queue);
1559 
1560 	cork->flags = 0;
1561 	cork->addr = 0;
1562 	cork->opt = NULL;
1563 	err = ip_setup_cork(sk, cork, ipc, rtp);
1564 	if (err)
1565 		return ERR_PTR(err);
1566 
1567 	err = __ip_append_data(sk, fl4, &queue, cork,
1568 			       &current->task_frag, getfrag,
1569 			       from, length, transhdrlen, flags);
1570 	if (err) {
1571 		__ip_flush_pending_frames(sk, &queue, cork);
1572 		return ERR_PTR(err);
1573 	}
1574 
1575 	return __ip_make_skb(sk, fl4, &queue, cork);
1576 }
1577 
1578 /*
1579  *	Fetch data from kernel space and fill in checksum if needed.
1580  */
ip_reply_glue_bits(void * dptr,char * to,int offset,int len,int odd,struct sk_buff * skb)1581 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1582 			      int len, int odd, struct sk_buff *skb)
1583 {
1584 	__wsum csum;
1585 
1586 	csum = csum_partial_copy_nocheck(dptr+offset, to, len);
1587 	skb->csum = csum_block_add(skb->csum, csum, odd);
1588 	return 0;
1589 }
1590 
1591 /*
1592  *	Generic function to send a packet as reply to another packet.
1593  *	Used to send some TCP resets/acks so far.
1594  */
ip_send_unicast_reply(struct sock * sk,const struct sock * orig_sk,struct sk_buff * skb,const struct ip_options * sopt,__be32 daddr,__be32 saddr,const struct ip_reply_arg * arg,unsigned int len,u64 transmit_time,u32 txhash)1595 void ip_send_unicast_reply(struct sock *sk, const struct sock *orig_sk,
1596 			   struct sk_buff *skb,
1597 			   const struct ip_options *sopt,
1598 			   __be32 daddr, __be32 saddr,
1599 			   const struct ip_reply_arg *arg,
1600 			   unsigned int len, u64 transmit_time, u32 txhash)
1601 {
1602 	struct ip_options_data replyopts;
1603 	struct ipcm_cookie ipc;
1604 	struct flowi4 fl4;
1605 	struct rtable *rt = skb_rtable(skb);
1606 	struct net *net = sock_net(sk);
1607 	struct sk_buff *nskb;
1608 	int err;
1609 	int oif;
1610 
1611 	if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
1612 		return;
1613 
1614 	ipcm_init(&ipc);
1615 	ipc.addr = daddr;
1616 	ipc.sockc.transmit_time = transmit_time;
1617 
1618 	if (replyopts.opt.opt.optlen) {
1619 		ipc.opt = &replyopts.opt;
1620 
1621 		if (replyopts.opt.opt.srr)
1622 			daddr = replyopts.opt.opt.faddr;
1623 	}
1624 
1625 	oif = arg->bound_dev_if;
1626 	if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1627 		oif = skb->skb_iif;
1628 
1629 	flowi4_init_output(&fl4, oif,
1630 			   IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1631 			   arg->tos & INET_DSCP_MASK,
1632 			   RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1633 			   ip_reply_arg_flowi_flags(arg),
1634 			   daddr, saddr,
1635 			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1636 			   arg->uid);
1637 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
1638 	rt = ip_route_output_flow(net, &fl4, sk);
1639 	if (IS_ERR(rt))
1640 		return;
1641 
1642 	inet_sk(sk)->tos = arg->tos;
1643 
1644 	sk->sk_protocol = ip_hdr(skb)->protocol;
1645 	sk->sk_bound_dev_if = arg->bound_dev_if;
1646 	sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1647 	ipc.sockc.mark = fl4.flowi4_mark;
1648 	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1649 			     len, 0, &ipc, &rt, MSG_DONTWAIT);
1650 	if (unlikely(err)) {
1651 		ip_flush_pending_frames(sk);
1652 		goto out;
1653 	}
1654 
1655 	nskb = skb_peek(&sk->sk_write_queue);
1656 	if (nskb) {
1657 		if (arg->csumoffset >= 0)
1658 			*((__sum16 *)skb_transport_header(nskb) +
1659 			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1660 								arg->csum));
1661 		nskb->ip_summed = CHECKSUM_NONE;
1662 		if (orig_sk)
1663 			skb_set_owner_edemux(nskb, (struct sock *)orig_sk);
1664 		if (transmit_time)
1665 			nskb->tstamp_type = SKB_CLOCK_MONOTONIC;
1666 		if (txhash)
1667 			skb_set_hash(nskb, txhash, PKT_HASH_TYPE_L4);
1668 		ip_push_pending_frames(sk, &fl4);
1669 	}
1670 out:
1671 	ip_rt_put(rt);
1672 }
1673 
ip_init(void)1674 void __init ip_init(void)
1675 {
1676 	ip_rt_init();
1677 	inet_initpeers();
1678 
1679 #if defined(CONFIG_IP_MULTICAST)
1680 	igmp_mc_init();
1681 #endif
1682 }
1683