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