ip_output.c 39.5 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		The Internet Protocol (IP) output module.
 *
8
 * Authors:	Ross Biro
L
Linus Torvalds 已提交
9 10 11 12 13 14 15 16 17 18 19 20 21 22
 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 *		Donald Becker, <becker@super.org>
 *		Alan Cox, <Alan.Cox@linux.org>
 *		Richard Underwood
 *		Stefan Becker, <stefanb@yello.ping.de>
 *		Jorge Cwik, <jorge@laser.satlink.net>
 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 *		Hirokazu Takahashi, <taka@valinux.co.jp>
 *
 *	See ip_input.c for original log
 *
 *	Fixes:
 *		Alan Cox	:	Missing nonblock feature in ip_build_xmit.
 *		Mike Kilburn	:	htons() missing in ip_build_xmit.
23
 *		Bradford Johnson:	Fix faulty handling of some frames when
L
Linus Torvalds 已提交
24 25 26 27 28 29 30 31 32 33
 *					no route is found.
 *		Alexander Demenshin:	Missing sk/skb free in ip_queue_xmit
 *					(in case if packet not accepted by
 *					output firewall rules)
 *		Mike McLagan	:	Routing by source
 *		Alexey Kuznetsov:	use new route cache
 *		Andi Kleen:		Fix broken PMTU recovery and remove
 *					some redundant tests.
 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
 *		Andi Kleen	: 	Replace ip_reply with ip_send_reply.
34 35 36
 *		Andi Kleen	:	Split fast and slow ip_build_xmit path
 *					for decreased register pressure on x86
 *					and more readibility.
L
Linus Torvalds 已提交
37 38 39 40 41 42 43 44 45 46 47 48 49 50 51
 *		Marc Boucher	:	When call_out_firewall returns FW_QUEUE,
 *					silently drop skb instead of failing with -EPERM.
 *		Detlev Wengorz	:	Copy protocol for fragments.
 *		Hirokazu Takahashi:	HW checksumming for outgoing UDP
 *					datagrams.
 *		Hirokazu Takahashi:	sendfile() on UDP works now.
 */

#include <asm/uaccess.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/errno.h>
A
Al Viro 已提交
52
#include <linux/highmem.h>
53
#include <linux/slab.h>
L
Linus Torvalds 已提交
54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

#include <linux/socket.h>
#include <linux/sockios.h>
#include <linux/in.h>
#include <linux/inet.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/proc_fs.h>
#include <linux/stat.h>
#include <linux/init.h>

#include <net/snmp.h>
#include <net/ip.h>
#include <net/protocol.h>
#include <net/route.h>
69
#include <net/xfrm.h>
L
Linus Torvalds 已提交
70 71 72 73 74 75 76 77 78 79 80
#include <linux/skbuff.h>
#include <net/sock.h>
#include <net/arp.h>
#include <net/icmp.h>
#include <net/checksum.h>
#include <net/inetpeer.h>
#include <linux/igmp.h>
#include <linux/netfilter_ipv4.h>
#include <linux/netfilter_bridge.h>
#include <linux/mroute.h>
#include <linux/netlink.h>
81
#include <linux/tcp.h>
L
Linus Torvalds 已提交
82

83
int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
84
EXPORT_SYMBOL(sysctl_ip_default_ttl);
L
Linus Torvalds 已提交
85

86 87 88 89
static int
ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
	    unsigned int mtu,
	    int (*output)(struct net *, struct sock *, struct sk_buff *));
90

L
Linus Torvalds 已提交
91
/* Generate a checksum for an outgoing IP datagram. */
92
void ip_send_check(struct iphdr *iph)
L
Linus Torvalds 已提交
93 94 95 96
{
	iph->check = 0;
	iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
}
E
Eric Dumazet 已提交
97
EXPORT_SYMBOL(ip_send_check);
L
Linus Torvalds 已提交
98

99
int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
H
Herbert Xu 已提交
100 101 102 103 104
{
	struct iphdr *iph = ip_hdr(skb);

	iph->tot_len = htons(skb->len);
	ip_send_check(iph);
105 106
	return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
		       net, sk, skb, NULL, skb_dst(skb)->dev,
107
		       dst_output);
108 109
}

110
int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
H
Herbert Xu 已提交
111 112 113
{
	int err;

114
	err = __ip_local_out(net, sk, skb);
H
Herbert Xu 已提交
115
	if (likely(err == 1))
116
		err = dst_output(net, sk, skb);
H
Herbert Xu 已提交
117 118 119

	return err;
}
120
EXPORT_SYMBOL_GPL(ip_local_out);
H
Herbert Xu 已提交
121

L
Linus Torvalds 已提交
122 123 124 125 126
static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
{
	int ttl = inet->uc_ttl;

	if (ttl < 0)
127
		ttl = ip4_dst_hoplimit(dst);
L
Linus Torvalds 已提交
128 129 130
	return ttl;
}

131
/*
L
Linus Torvalds 已提交
132 133 134
 *		Add an ip header to a skbuff and send it out.
 *
 */
135
int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
136
			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
L
Linus Torvalds 已提交
137 138
{
	struct inet_sock *inet = inet_sk(sk);
E
Eric Dumazet 已提交
139
	struct rtable *rt = skb_rtable(skb);
140
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
141 142 143
	struct iphdr *iph;

	/* Build the IP header. */
144
	skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
145
	skb_reset_network_header(skb);
146
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
147 148 149
	iph->version  = 4;
	iph->ihl      = 5;
	iph->tos      = inet->tos;
150
	iph->ttl      = ip_select_ttl(inet, &rt->dst);
151 152
	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
	iph->saddr    = saddr;
L
Linus Torvalds 已提交
153
	iph->protocol = sk->sk_protocol;
154 155 156 157 158
	if (ip_dont_fragment(sk, &rt->dst)) {
		iph->frag_off = htons(IP_DF);
		iph->id = 0;
	} else {
		iph->frag_off = 0;
159
		__ip_select_ident(net, iph, 1);
160
	}
L
Linus Torvalds 已提交
161

162 163 164
	if (opt && opt->opt.optlen) {
		iph->ihl += opt->opt.optlen>>2;
		ip_options_build(skb, &opt->opt, daddr, rt, 0);
L
Linus Torvalds 已提交
165 166 167
	}

	skb->priority = sk->sk_priority;
168
	skb->mark = sk->sk_mark;
L
Linus Torvalds 已提交
169 170

	/* Send it out. */
171
	return ip_local_out(net, skb->sk, skb);
L
Linus Torvalds 已提交
172
}
173 174
EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);

175
static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
176
{
E
Eric Dumazet 已提交
177
	struct dst_entry *dst = skb_dst(skb);
178
	struct rtable *rt = (struct rtable *)dst;
L
Linus Torvalds 已提交
179
	struct net_device *dev = dst->dev;
180
	unsigned int hh_len = LL_RESERVED_SPACE(dev);
181
	struct neighbour *neigh;
182
	u32 nexthop;
L
Linus Torvalds 已提交
183

184
	if (rt->rt_type == RTN_MULTICAST) {
185
		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
186
	} else if (rt->rt_type == RTN_BROADCAST)
187
		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
188

L
Linus Torvalds 已提交
189
	/* Be paranoid, rather than too clever. */
190
	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
L
Linus Torvalds 已提交
191 192 193
		struct sk_buff *skb2;

		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
194
		if (!skb2) {
L
Linus Torvalds 已提交
195 196 197 198 199
			kfree_skb(skb);
			return -ENOMEM;
		}
		if (skb->sk)
			skb_set_owner_w(skb2, skb->sk);
200
		consume_skb(skb);
L
Linus Torvalds 已提交
201 202 203
		skb = skb2;
	}

204
	rcu_read_lock_bh();
J
Julian Anastasov 已提交
205
	nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
206 207 208
	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
	if (unlikely(!neigh))
		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
209
	if (!IS_ERR(neigh)) {
210
		int res = dst_neigh_output(dst, neigh, skb);
211

212
		rcu_read_unlock_bh();
213 214
		return res;
	}
215
	rcu_read_unlock_bh();
216

217 218
	net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
			    __func__);
L
Linus Torvalds 已提交
219 220 221 222
	kfree_skb(skb);
	return -EINVAL;
}

223 224
static int ip_finish_output_gso(struct net *net, struct sock *sk,
				struct sk_buff *skb, unsigned int mtu)
225 226 227 228 229 230 231
{
	netdev_features_t features;
	struct sk_buff *segs;
	int ret = 0;

	/* common case: locally created skb or seglen is <= mtu */
	if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) ||
232
	      skb_gso_network_seglen(skb) <= mtu)
233
		return ip_finish_output2(net, sk, skb);
234 235 236 237 238 239 240 241 242 243

	/* Slowpath -  GSO segment length is exceeding the dst MTU.
	 *
	 * This can happen in two cases:
	 * 1) TCP GRO packet, DF bit not set
	 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
	 * from host network stack.
	 */
	features = netif_skb_features(skb);
	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
244
	if (IS_ERR_OR_NULL(segs)) {
245 246 247 248 249 250 251 252 253 254 255
		kfree_skb(skb);
		return -ENOMEM;
	}

	consume_skb(skb);

	do {
		struct sk_buff *nskb = segs->next;
		int err;

		segs->next = NULL;
256
		err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
257 258 259 260 261 262 263 264 265

		if (err && ret == 0)
			ret = err;
		segs = nskb;
	} while (segs);

	return ret;
}

266
static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
267
{
268 269
	unsigned int mtu;

270 271
#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
	/* Policy lookup after SNAT yielded a new policy */
272
	if (skb_dst(skb)->xfrm) {
273
		IPCB(skb)->flags |= IPSKB_REROUTED;
274
		return dst_output(net, sk, skb);
275
	}
276
#endif
277
	mtu = ip_skb_dst_mtu(skb);
278
	if (skb_is_gso(skb))
279
		return ip_finish_output_gso(net, sk, skb, mtu);
280

281
	if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
282
		return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
283

284
	return ip_finish_output2(net, sk, skb);
L
Linus Torvalds 已提交
285 286
}

E
Eric W. Biederman 已提交
287
int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
288
{
E
Eric Dumazet 已提交
289
	struct rtable *rt = skb_rtable(skb);
290
	struct net_device *dev = rt->dst.dev;
L
Linus Torvalds 已提交
291 292 293 294

	/*
	 *	If the indicated interface is up and running, send the packet.
	 */
295
	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
L
Linus Torvalds 已提交
296 297 298 299 300 301 302 303 304

	skb->dev = dev;
	skb->protocol = htons(ETH_P_IP);

	/*
	 *	Multicasts are looped back for other local users
	 */

	if (rt->rt_flags&RTCF_MULTICAST) {
305
		if (sk_mc_loop(sk)
L
Linus Torvalds 已提交
306 307 308 309 310 311 312 313 314
#ifdef CONFIG_IP_MROUTE
		/* Small optimization: do not loopback not local frames,
		   which returned after forwarding; they will be  dropped
		   by ip_mr_input in any case.
		   Note, that local frames are looped back to be delivered
		   to local recipients.

		   This check is duplicated in ip_mr_input at the moment.
		 */
315 316 317
		    &&
		    ((rt->rt_flags & RTCF_LOCAL) ||
		     !(IPCB(skb)->flags & IPSKB_FORWARDED))
L
Linus Torvalds 已提交
318
#endif
319
		   ) {
L
Linus Torvalds 已提交
320 321
			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
			if (newskb)
322
				NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
323
					net, sk, newskb, NULL, newskb->dev,
324
					dev_loopback_xmit);
L
Linus Torvalds 已提交
325 326 327 328
		}

		/* Multicasts with ttl 0 must not go beyond the host */

329
		if (ip_hdr(skb)->ttl == 0) {
L
Linus Torvalds 已提交
330 331 332 333 334 335 336 337
			kfree_skb(skb);
			return 0;
		}
	}

	if (rt->rt_flags&RTCF_BROADCAST) {
		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
		if (newskb)
338 339 340
			NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
				net, sk, newskb, NULL, newskb->dev,
				dev_loopback_xmit);
L
Linus Torvalds 已提交
341 342
	}

343 344 345
	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
			    net, sk, skb, NULL, skb->dev,
			    ip_finish_output,
346
			    !(IPCB(skb)->flags & IPSKB_REROUTED));
L
Linus Torvalds 已提交
347 348
}

E
Eric W. Biederman 已提交
349
int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
350
{
E
Eric Dumazet 已提交
351
	struct net_device *dev = skb_dst(skb)->dev;
352

353
	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
L
Linus Torvalds 已提交
354

355 356 357
	skb->dev = dev;
	skb->protocol = htons(ETH_P_IP);

358 359
	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
			    net, sk, skb, NULL, dev,
360
			    ip_finish_output,
361
			    !(IPCB(skb)->flags & IPSKB_REROUTED));
L
Linus Torvalds 已提交
362 363
}

364 365 366 367 368 369 370 371 372 373 374 375 376 377
/*
 * copy saddr and daddr, possibly using 64bit load/stores
 * Equivalent to :
 *   iph->saddr = fl4->saddr;
 *   iph->daddr = fl4->daddr;
 */
static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
{
	BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
		     offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
	memcpy(&iph->saddr, &fl4->saddr,
	       sizeof(fl4->saddr) + sizeof(fl4->daddr));
}

378 379
/* Note: skb->sk can be different from sk, in case of tunnels */
int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
L
Linus Torvalds 已提交
380 381
{
	struct inet_sock *inet = inet_sk(sk);
382
	struct net *net = sock_net(sk);
383
	struct ip_options_rcu *inet_opt;
384
	struct flowi4 *fl4;
L
Linus Torvalds 已提交
385 386
	struct rtable *rt;
	struct iphdr *iph;
387
	int res;
L
Linus Torvalds 已提交
388 389 390 391

	/* Skip all of this if the packet is already routed,
	 * f.e. by something like SCTP.
	 */
392
	rcu_read_lock();
393
	inet_opt = rcu_dereference(inet->inet_opt);
394
	fl4 = &fl->u.ip4;
E
Eric Dumazet 已提交
395
	rt = skb_rtable(skb);
396
	if (rt)
L
Linus Torvalds 已提交
397 398 399 400
		goto packet_routed;

	/* Make sure we can route this packet. */
	rt = (struct rtable *)__sk_dst_check(sk, 0);
401
	if (!rt) {
A
Al Viro 已提交
402
		__be32 daddr;
L
Linus Torvalds 已提交
403 404

		/* Use correct destination address if we have options. */
E
Eric Dumazet 已提交
405
		daddr = inet->inet_daddr;
406 407
		if (inet_opt && inet_opt->opt.srr)
			daddr = inet_opt->opt.faddr;
L
Linus Torvalds 已提交
408

409 410 411 412
		/* If this fails, retransmit mechanism of transport layer will
		 * keep trying until route appears or the connection times
		 * itself out.
		 */
413
		rt = ip_route_output_ports(net, fl4, sk,
414 415 416 417 418 419 420 421
					   daddr, inet->inet_saddr,
					   inet->inet_dport,
					   inet->inet_sport,
					   sk->sk_protocol,
					   RT_CONN_FLAGS(sk),
					   sk->sk_bound_dev_if);
		if (IS_ERR(rt))
			goto no_route;
422
		sk_setup_caps(sk, &rt->dst);
L
Linus Torvalds 已提交
423
	}
424
	skb_dst_set_noref(skb, &rt->dst);
L
Linus Torvalds 已提交
425 426

packet_routed:
J
Julian Anastasov 已提交
427
	if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
L
Linus Torvalds 已提交
428 429 430
		goto no_route;

	/* OK, we know where to send it, allocate and build IP header. */
431
	skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
432
	skb_reset_network_header(skb);
433
	iph = ip_hdr(skb);
434
	*((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
W
WANG Cong 已提交
435
	if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
L
Linus Torvalds 已提交
436 437 438
		iph->frag_off = htons(IP_DF);
	else
		iph->frag_off = 0;
439
	iph->ttl      = ip_select_ttl(inet, &rt->dst);
L
Linus Torvalds 已提交
440
	iph->protocol = sk->sk_protocol;
441 442
	ip_copy_addrs(iph, fl4);

L
Linus Torvalds 已提交
443 444
	/* Transport layer set skb->h.foo itself. */

445 446 447
	if (inet_opt && inet_opt->opt.optlen) {
		iph->ihl += inet_opt->opt.optlen >> 2;
		ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
L
Linus Torvalds 已提交
448 449
	}

450
	ip_select_ident_segs(net, skb, sk,
451
			     skb_shinfo(skb)->gso_segs ?: 1);
L
Linus Torvalds 已提交
452

453
	/* TODO : should we use skb->sk here instead of sk ? */
L
Linus Torvalds 已提交
454
	skb->priority = sk->sk_priority;
455
	skb->mark = sk->sk_mark;
L
Linus Torvalds 已提交
456

457
	res = ip_local_out(net, sk, skb);
458 459
	rcu_read_unlock();
	return res;
L
Linus Torvalds 已提交
460 461

no_route:
462
	rcu_read_unlock();
463
	IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
L
Linus Torvalds 已提交
464 465 466
	kfree_skb(skb);
	return -EHOSTUNREACH;
}
E
Eric Dumazet 已提交
467
EXPORT_SYMBOL(ip_queue_xmit);
L
Linus Torvalds 已提交
468 469 470 471 472 473

static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
{
	to->pkt_type = from->pkt_type;
	to->priority = from->priority;
	to->protocol = from->protocol;
E
Eric Dumazet 已提交
474
	skb_dst_drop(to);
475
	skb_dst_copy(to, from);
L
Linus Torvalds 已提交
476
	to->dev = from->dev;
T
Thomas Graf 已提交
477
	to->mark = from->mark;
L
Linus Torvalds 已提交
478 479 480 481 482 483 484

	/* Copy the flags to each fragment. */
	IPCB(to)->flags = IPCB(from)->flags;

#ifdef CONFIG_NET_SCHED
	to->tc_index = from->tc_index;
#endif
485
	nf_copy(to, from);
486 487
#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
	to->ipvs_property = from->ipvs_property;
L
Linus Torvalds 已提交
488
#endif
489
	skb_copy_secmark(to, from);
L
Linus Torvalds 已提交
490 491
}

492
static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
493
		       unsigned int mtu,
494
		       int (*output)(struct net *, struct sock *, struct sk_buff *))
495 496 497
{
	struct iphdr *iph = ip_hdr(skb);

498
	if ((iph->frag_off & htons(IP_DF)) == 0)
499
		return ip_do_fragment(net, sk, skb, output);
500 501

	if (unlikely(!skb->ignore_df ||
502 503
		     (IPCB(skb)->frag_max_size &&
		      IPCB(skb)->frag_max_size > mtu))) {
504
		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
505 506 507 508 509 510
		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
			  htonl(mtu));
		kfree_skb(skb);
		return -EMSGSIZE;
	}

511
	return ip_do_fragment(net, sk, skb, output);
512 513
}

L
Linus Torvalds 已提交
514 515 516 517 518 519 520
/*
 *	This IP datagram is too large to be sent in one piece.  Break it up into
 *	smaller pieces (each of size equal to IP header plus
 *	a block of the data of the original IP data part) that will yet fit in a
 *	single device frame, and queue such a frame for sending.
 */

521 522
int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
		   int (*output)(struct net *, struct sock *, struct sk_buff *))
L
Linus Torvalds 已提交
523 524 525 526 527
{
	struct iphdr *iph;
	int ptr;
	struct net_device *dev;
	struct sk_buff *skb2;
528
	unsigned int mtu, hlen, left, len, ll_rs;
L
Linus Torvalds 已提交
529
	int offset;
530
	__be16 not_last_frag;
E
Eric Dumazet 已提交
531
	struct rtable *rt = skb_rtable(skb);
L
Linus Torvalds 已提交
532 533
	int err = 0;

534
	dev = rt->dst.dev;
L
Linus Torvalds 已提交
535 536 537 538 539

	/*
	 *	Point into the IP datagram header.
	 */

540
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
541

542
	mtu = ip_skb_dst_mtu(skb);
543 544
	if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
		mtu = IPCB(skb)->frag_max_size;
L
Linus Torvalds 已提交
545 546 547 548 549 550

	/*
	 *	Setup starting values.
	 */

	hlen = iph->ihl * 4;
551
	mtu = mtu - hlen;	/* Size of data space */
H
Herbert Xu 已提交
552
	IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
L
Linus Torvalds 已提交
553 554 555 556 557 558 559 560

	/* When frag_list is given, use it. First, check its validity:
	 * some transformers could create wrong frag_list or break existing
	 * one, it is not prohibited. In this case fall back to copying.
	 *
	 * LATER: this step can be merged to real generation of fragments,
	 * we can switch to copy when see the first bad fragment.
	 */
561
	if (skb_has_frag_list(skb)) {
562
		struct sk_buff *frag, *frag2;
L
Linus Torvalds 已提交
563 564 565 566
		int first_len = skb_pagelen(skb);

		if (first_len - hlen > mtu ||
		    ((first_len - hlen) & 7) ||
567
		    ip_is_fragment(iph) ||
L
Linus Torvalds 已提交
568 569 570
		    skb_cloned(skb))
			goto slow_path;

571
		skb_walk_frags(skb, frag) {
L
Linus Torvalds 已提交
572 573 574 575
			/* Correct geometry. */
			if (frag->len > mtu ||
			    ((frag->len & 7) && frag->next) ||
			    skb_headroom(frag) < hlen)
576
				goto slow_path_clean;
L
Linus Torvalds 已提交
577 578 579

			/* Partially cloned skb? */
			if (skb_shared(frag))
580
				goto slow_path_clean;
581 582 583 584 585 586

			BUG_ON(frag->sk);
			if (skb->sk) {
				frag->sk = skb->sk;
				frag->destructor = sock_wfree;
			}
587
			skb->truesize -= frag->truesize;
L
Linus Torvalds 已提交
588 589 590 591 592 593 594
		}

		/* Everything is OK. Generate! */

		err = 0;
		offset = 0;
		frag = skb_shinfo(skb)->frag_list;
595
		skb_frag_list_init(skb);
L
Linus Torvalds 已提交
596 597 598 599 600 601 602 603 604 605 606
		skb->data_len = first_len - skb_headlen(skb);
		skb->len = first_len;
		iph->tot_len = htons(first_len);
		iph->frag_off = htons(IP_MF);
		ip_send_check(iph);

		for (;;) {
			/* Prepare header of the next frame,
			 * before previous one went down. */
			if (frag) {
				frag->ip_summed = CHECKSUM_NONE;
607
				skb_reset_transport_header(frag);
608 609
				__skb_push(frag, hlen);
				skb_reset_network_header(frag);
610
				memcpy(skb_network_header(frag), iph, hlen);
611
				iph = ip_hdr(frag);
L
Linus Torvalds 已提交
612 613 614 615 616 617
				iph->tot_len = htons(frag->len);
				ip_copy_metadata(frag, skb);
				if (offset == 0)
					ip_options_fragment(frag);
				offset += skb->len - hlen;
				iph->frag_off = htons(offset>>3);
618
				if (frag->next)
L
Linus Torvalds 已提交
619 620 621 622 623
					iph->frag_off |= htons(IP_MF);
				/* Ready, complete checksum */
				ip_send_check(iph);
			}

624
			err = output(net, sk, skb);
L
Linus Torvalds 已提交
625

626
			if (!err)
627
				IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
L
Linus Torvalds 已提交
628 629 630 631 632 633 634 635 636
			if (err || !frag)
				break;

			skb = frag;
			frag = skb->next;
			skb->next = NULL;
		}

		if (err == 0) {
637
			IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
L
Linus Torvalds 已提交
638 639 640 641 642 643 644 645
			return 0;
		}

		while (frag) {
			skb = frag->next;
			kfree_skb(frag);
			frag = skb;
		}
646
		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
647
		return err;
648 649 650 651 652 653 654 655 656

slow_path_clean:
		skb_walk_frags(skb, frag2) {
			if (frag2 == frag)
				break;
			frag2->sk = NULL;
			frag2->destructor = NULL;
			skb->truesize += frag2->truesize;
		}
L
Linus Torvalds 已提交
657 658 659
	}

slow_path:
660 661 662
	/* for offloaded checksums cleanup checksum before fragmentation */
	if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
		goto fail;
663
	iph = ip_hdr(skb);
664

L
Linus Torvalds 已提交
665
	left = skb->len - hlen;		/* Space per frame */
666
	ptr = hlen;		/* Where to start from */
L
Linus Torvalds 已提交
667

668
	ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
669

L
Linus Torvalds 已提交
670 671 672 673 674 675 676 677 678 679 680
	/*
	 *	Fragment the datagram.
	 */

	offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
	not_last_frag = iph->frag_off & htons(IP_MF);

	/*
	 *	Keep copying data until we run out.
	 */

S
Stephen Hemminger 已提交
681
	while (left > 0) {
L
Linus Torvalds 已提交
682 683 684 685
		len = left;
		/* IF: it doesn't fit, use 'mtu' - the data space left */
		if (len > mtu)
			len = mtu;
L
Lucas De Marchi 已提交
686
		/* IF: we are not sending up to and including the packet end
L
Linus Torvalds 已提交
687 688 689 690 691
		   then align the next start on an eight byte boundary */
		if (len < left)	{
			len &= ~7;
		}

692 693 694
		/* Allocate buffer */
		skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
		if (!skb2) {
L
Linus Torvalds 已提交
695 696 697 698 699 700 701 702 703 704 705
			err = -ENOMEM;
			goto fail;
		}

		/*
		 *	Set up data on packet
		 */

		ip_copy_metadata(skb2, skb);
		skb_reserve(skb2, ll_rs);
		skb_put(skb2, len + hlen);
706
		skb_reset_network_header(skb2);
707
		skb2->transport_header = skb2->network_header + hlen;
L
Linus Torvalds 已提交
708 709 710 711 712 713 714 715 716 717 718 719 720

		/*
		 *	Charge the memory for the fragment to any owner
		 *	it might possess
		 */

		if (skb->sk)
			skb_set_owner_w(skb2, skb->sk);

		/*
		 *	Copy the packet header into the new buffer.
		 */

721
		skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
L
Linus Torvalds 已提交
722 723 724 725

		/*
		 *	Copy a block of the IP datagram.
		 */
726
		if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
L
Linus Torvalds 已提交
727 728 729 730 731 732
			BUG();
		left -= len;

		/*
		 *	Fill in the new header fields.
		 */
733
		iph = ip_hdr(skb2);
L
Linus Torvalds 已提交
734 735
		iph->frag_off = htons((offset >> 3));

736 737 738
		if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
			iph->frag_off |= htons(IP_DF);

L
Linus Torvalds 已提交
739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763
		/* ANK: dirty, but effective trick. Upgrade options only if
		 * the segment to be fragmented was THE FIRST (otherwise,
		 * options are already fixed) and make it ONCE
		 * on the initial skb, so that all the following fragments
		 * will inherit fixed options.
		 */
		if (offset == 0)
			ip_options_fragment(skb);

		/*
		 *	Added AC : If we are fragmenting a fragment that's not the
		 *		   last fragment then keep MF on each bit
		 */
		if (left > 0 || not_last_frag)
			iph->frag_off |= htons(IP_MF);
		ptr += len;
		offset += len;

		/*
		 *	Put this fragment into the sending queue.
		 */
		iph->tot_len = htons(len + hlen);

		ip_send_check(iph);

764
		err = output(net, sk, skb2);
L
Linus Torvalds 已提交
765 766
		if (err)
			goto fail;
767

768
		IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
L
Linus Torvalds 已提交
769
	}
770
	consume_skb(skb);
771
	IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
L
Linus Torvalds 已提交
772 773 774
	return err;

fail:
775
	kfree_skb(skb);
776
	IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
777 778
	return err;
}
779
EXPORT_SYMBOL(ip_do_fragment);
780

L
Linus Torvalds 已提交
781 782 783
int
ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
{
784
	struct msghdr *msg = from;
L
Linus Torvalds 已提交
785

786
	if (skb->ip_summed == CHECKSUM_PARTIAL) {
787
		if (copy_from_iter(to, len, &msg->msg_iter) != len)
L
Linus Torvalds 已提交
788 789
			return -EFAULT;
	} else {
790
		__wsum csum = 0;
791
		if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len)
L
Linus Torvalds 已提交
792 793 794 795 796
			return -EFAULT;
		skb->csum = csum_block_add(skb->csum, csum, odd);
	}
	return 0;
}
E
Eric Dumazet 已提交
797
EXPORT_SYMBOL(ip_generic_getfrag);
L
Linus Torvalds 已提交
798

799
static inline __wsum
L
Linus Torvalds 已提交
800 801 802
csum_page(struct page *page, int offset, int copy)
{
	char *kaddr;
803
	__wsum csum;
L
Linus Torvalds 已提交
804 805 806 807 808 809
	kaddr = kmap(page);
	csum = csum_partial(kaddr + offset, copy, 0);
	kunmap(page);
	return csum;
}

A
Adrian Bunk 已提交
810
static inline int ip_ufo_append_data(struct sock *sk,
811
			struct sk_buff_head *queue,
812 813 814
			int getfrag(void *from, char *to, int offset, int len,
			       int odd, struct sk_buff *skb),
			void *from, int length, int hh_len, int fragheaderlen,
815
			int transhdrlen, int maxfraglen, unsigned int flags)
816 817 818 819 820 821 822 823
{
	struct sk_buff *skb;
	int err;

	/* There is support for UDP fragmentation offload by network
	 * device, so create one single skb packet containing complete
	 * udp datagram
	 */
824 825
	skb = skb_peek_tail(queue);
	if (!skb) {
826 827 828 829
		skb = sock_alloc_send_skb(sk,
			hh_len + fragheaderlen + transhdrlen + 20,
			(flags & MSG_DONTWAIT), &err);

830
		if (!skb)
831 832 833 834 835 836
			return err;

		/* reserve space for Hardware header */
		skb_reserve(skb, hh_len);

		/* create space for UDP/IP header */
837
		skb_put(skb, fragheaderlen + transhdrlen);
838 839

		/* initialize network header pointer */
840
		skb_reset_network_header(skb);
841 842

		/* initialize protocol header pointer */
843
		skb->transport_header = skb->network_header + fragheaderlen;
844 845 846

		skb->csum = 0;

847
		__skb_queue_tail(queue, skb);
848 849
	} else if (skb_is_gso(skb)) {
		goto append;
850
	}
851

852 853 854 855 856 857
	skb->ip_summed = CHECKSUM_PARTIAL;
	/* specify the length of each IP datagram fragment */
	skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
	skb_shinfo(skb)->gso_type = SKB_GSO_UDP;

append:
858 859
	return skb_append_datato_frags(sk, skb, getfrag, from,
				       (length - transhdrlen));
860 861
}

862 863 864
static int __ip_append_data(struct sock *sk,
			    struct flowi4 *fl4,
			    struct sk_buff_head *queue,
865
			    struct inet_cork *cork,
866
			    struct page_frag *pfrag,
867 868 869 870
			    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)
L
Linus Torvalds 已提交
871 872 873 874
{
	struct inet_sock *inet = inet_sk(sk);
	struct sk_buff *skb;

875
	struct ip_options *opt = cork->opt;
L
Linus Torvalds 已提交
876 877 878 879 880 881
	int hh_len;
	int exthdrlen;
	int mtu;
	int copy;
	int err;
	int offset = 0;
882
	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
L
Linus Torvalds 已提交
883
	int csummode = CHECKSUM_NONE;
884
	struct rtable *rt = (struct rtable *)cork->dst;
885
	u32 tskey = 0;
L
Linus Torvalds 已提交
886

887 888 889
	skb = skb_peek_tail(queue);

	exthdrlen = !skb ? rt->dst.header_len : 0;
890
	mtu = cork->fragsize;
891 892 893
	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
		tskey = sk->sk_tskey++;
L
Linus Torvalds 已提交
894

895
	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
L
Linus Torvalds 已提交
896 897 898

	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
W
WANG Cong 已提交
899
	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
L
Linus Torvalds 已提交
900

901
	if (cork->length + length > maxnonfragsize - fragheaderlen) {
902
		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
903
			       mtu - (opt ? opt->optlen : 0));
L
Linus Torvalds 已提交
904 905 906 907 908 909 910 911 912
		return -EMSGSIZE;
	}

	/*
	 * transhdrlen > 0 means that this is the first fragment and we wish
	 * it won't be fragmented in the future.
	 */
	if (transhdrlen &&
	    length + fragheaderlen <= mtu &&
913
	    rt->dst.dev->features & NETIF_F_V4_CSUM &&
L
Linus Torvalds 已提交
914
	    !exthdrlen)
915
		csummode = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
916

917
	cork->length += length;
918
	if (((length > mtu) || (skb && skb_is_gso(skb))) &&
919
	    (sk->sk_protocol == IPPROTO_UDP) &&
920 921
	    (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len &&
	    (sk->sk_type == SOCK_DGRAM)) {
922 923
		err = ip_ufo_append_data(sk, queue, getfrag, from, length,
					 hh_len, fragheaderlen, transhdrlen,
924
					 maxfraglen, flags);
925
		if (err)
926 927 928
			goto error;
		return 0;
	}
L
Linus Torvalds 已提交
929 930 931 932 933 934 935 936

	/* So, what's going on in the loop below?
	 *
	 * We use calculated fragment length to generate chained skb,
	 * each of segments is IP fragment ready for sending to network after
	 * adding appropriate IP header.
	 */

937
	if (!skb)
L
Linus Torvalds 已提交
938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967
		goto alloc_new_skb;

	while (length > 0) {
		/* Check if the remaining data fits into current packet. */
		copy = mtu - skb->len;
		if (copy < length)
			copy = maxfraglen - skb->len;
		if (copy <= 0) {
			char *data;
			unsigned int datalen;
			unsigned int fraglen;
			unsigned int fraggap;
			unsigned int alloclen;
			struct sk_buff *skb_prev;
alloc_new_skb:
			skb_prev = skb;
			if (skb_prev)
				fraggap = skb_prev->len - maxfraglen;
			else
				fraggap = 0;

			/*
			 * If remaining data exceeds the mtu,
			 * we know we need more fragment(s).
			 */
			datalen = length + fraggap;
			if (datalen > mtu - fragheaderlen)
				datalen = maxfraglen - fragheaderlen;
			fraglen = datalen + fragheaderlen;

968
			if ((flags & MSG_MORE) &&
969
			    !(rt->dst.dev->features&NETIF_F_SG))
L
Linus Torvalds 已提交
970 971
				alloclen = mtu;
			else
972
				alloclen = fraglen;
L
Linus Torvalds 已提交
973

974 975
			alloclen += exthdrlen;

L
Linus Torvalds 已提交
976 977 978 979 980
			/* The last fragment gets additional space at tail.
			 * Note, with MSG_MORE we overallocate on fragments,
			 * because we have no idea what fragment will be
			 * the last.
			 */
981
			if (datalen == length + fraggap)
982
				alloclen += rt->dst.trailer_len;
983

L
Linus Torvalds 已提交
984
			if (transhdrlen) {
985
				skb = sock_alloc_send_skb(sk,
L
Linus Torvalds 已提交
986 987 988 989 990 991
						alloclen + hh_len + 15,
						(flags & MSG_DONTWAIT), &err);
			} else {
				skb = NULL;
				if (atomic_read(&sk->sk_wmem_alloc) <=
				    2 * sk->sk_sndbuf)
992
					skb = sock_wmalloc(sk,
L
Linus Torvalds 已提交
993 994
							   alloclen + hh_len + 15, 1,
							   sk->sk_allocation);
995
				if (unlikely(!skb))
L
Linus Torvalds 已提交
996 997
					err = -ENOBUFS;
			}
998
			if (!skb)
L
Linus Torvalds 已提交
999 1000 1001 1002 1003 1004 1005 1006
				goto error;

			/*
			 *	Fill in the control structures
			 */
			skb->ip_summed = csummode;
			skb->csum = 0;
			skb_reserve(skb, hh_len);
1007 1008

			/* only the initial fragment is time stamped */
1009
			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1010
			cork->tx_flags = 0;
1011 1012
			skb_shinfo(skb)->tskey = tskey;
			tskey = 0;
L
Linus Torvalds 已提交
1013 1014 1015 1016

			/*
			 *	Find where to start putting bytes.
			 */
1017
			data = skb_put(skb, fraglen + exthdrlen);
1018
			skb_set_network_header(skb, exthdrlen);
1019 1020
			skb->transport_header = (skb->network_header +
						 fragheaderlen);
1021
			data += fragheaderlen + exthdrlen;
L
Linus Torvalds 已提交
1022 1023 1024 1025 1026 1027 1028 1029

			if (fraggap) {
				skb->csum = skb_copy_and_csum_bits(
					skb_prev, maxfraglen,
					data + transhdrlen, fraggap, 0);
				skb_prev->csum = csum_sub(skb_prev->csum,
							  skb->csum);
				data += fraggap;
1030
				pskb_trim_unique(skb_prev, maxfraglen);
L
Linus Torvalds 已提交
1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048
			}

			copy = datalen - transhdrlen - fraggap;
			if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
				err = -EFAULT;
				kfree_skb(skb);
				goto error;
			}

			offset += copy;
			length -= datalen - fraggap;
			transhdrlen = 0;
			exthdrlen = 0;
			csummode = CHECKSUM_NONE;

			/*
			 * Put the packet on the pending queue.
			 */
1049
			__skb_queue_tail(queue, skb);
L
Linus Torvalds 已提交
1050 1051 1052 1053 1054 1055
			continue;
		}

		if (copy > length)
			copy = length;

1056
		if (!(rt->dst.dev->features&NETIF_F_SG)) {
L
Linus Torvalds 已提交
1057 1058 1059
			unsigned int off;

			off = skb->len;
1060
			if (getfrag(from, skb_put(skb, copy),
L
Linus Torvalds 已提交
1061 1062 1063 1064 1065 1066 1067 1068
					offset, copy, off, skb) < 0) {
				__skb_trim(skb, off);
				err = -EFAULT;
				goto error;
			}
		} else {
			int i = skb_shinfo(skb)->nr_frags;

1069 1070
			err = -ENOMEM;
			if (!sk_page_frag_refill(sk, pfrag))
L
Linus Torvalds 已提交
1071
				goto error;
1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082

			if (!skb_can_coalesce(skb, i, pfrag->page,
					      pfrag->offset)) {
				err = -EMSGSIZE;
				if (i == MAX_SKB_FRAGS)
					goto error;

				__skb_fill_page_desc(skb, i, pfrag->page,
						     pfrag->offset, 0);
				skb_shinfo(skb)->nr_frags = ++i;
				get_page(pfrag->page);
L
Linus Torvalds 已提交
1083
			}
1084 1085 1086 1087 1088 1089 1090 1091
			copy = min_t(int, copy, pfrag->size - pfrag->offset);
			if (getfrag(from,
				    page_address(pfrag->page) + pfrag->offset,
				    offset, copy, skb->len, skb) < 0)
				goto error_efault;

			pfrag->offset += copy;
			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
L
Linus Torvalds 已提交
1092 1093
			skb->len += copy;
			skb->data_len += copy;
1094 1095
			skb->truesize += copy;
			atomic_add(copy, &sk->sk_wmem_alloc);
L
Linus Torvalds 已提交
1096 1097 1098 1099 1100 1101 1102
		}
		offset += copy;
		length -= copy;
	}

	return 0;

1103 1104
error_efault:
	err = -EFAULT;
L
Linus Torvalds 已提交
1105
error:
1106
	cork->length -= length;
P
Pavel Emelyanov 已提交
1107
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1108
	return err;
L
Linus Torvalds 已提交
1109 1110
}

1111 1112 1113
static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
			 struct ipcm_cookie *ipc, struct rtable **rtp)
{
1114
	struct ip_options_rcu *opt;
1115 1116 1117 1118 1119 1120 1121
	struct rtable *rt;

	/*
	 * setup for corking.
	 */
	opt = ipc->opt;
	if (opt) {
1122
		if (!cork->opt) {
1123 1124
			cork->opt = kmalloc(sizeof(struct ip_options) + 40,
					    sk->sk_allocation);
1125
			if (unlikely(!cork->opt))
1126 1127
				return -ENOBUFS;
		}
1128
		memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1129 1130 1131 1132 1133 1134 1135 1136 1137 1138
		cork->flags |= IPCORK_OPT;
		cork->addr = ipc->addr;
	}
	rt = *rtp;
	if (unlikely(!rt))
		return -EFAULT;
	/*
	 * We steal reference to this route, caller should not release it
	 */
	*rtp = NULL;
1139 1140
	cork->fragsize = ip_sk_use_pmtu(sk) ?
			 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
1141 1142
	cork->dst = &rt->dst;
	cork->length = 0;
1143 1144 1145
	cork->ttl = ipc->ttl;
	cork->tos = ipc->tos;
	cork->priority = ipc->priority;
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161
	cork->tx_flags = ipc->tx_flags;

	return 0;
}

/*
 *	ip_append_data() and ip_append_page() can make one large IP datagram
 *	from many pieces of data. Each pieces will be holded on the socket
 *	until ip_push_pending_frames() is called. Each piece can be a page
 *	or non-page data.
 *
 *	Not only UDP, other transport protocols - e.g. raw sockets - can use
 *	this interface potentially.
 *
 *	LATER: length must be adjusted by pad at tail, when it is required.
 */
1162
int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175
		   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)
{
	struct inet_sock *inet = inet_sk(sk);
	int err;

	if (flags&MSG_PROBE)
		return 0;

	if (skb_queue_empty(&sk->sk_write_queue)) {
1176
		err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1177 1178 1179 1180 1181 1182
		if (err)
			return err;
	} else {
		transhdrlen = 0;
	}

1183 1184
	return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
				sk_page_frag(sk), getfrag,
1185 1186 1187
				from, length, transhdrlen, flags);
}

1188
ssize_t	ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
L
Linus Torvalds 已提交
1189 1190 1191 1192 1193 1194
		       int offset, size_t size, int flags)
{
	struct inet_sock *inet = inet_sk(sk);
	struct sk_buff *skb;
	struct rtable *rt;
	struct ip_options *opt = NULL;
1195
	struct inet_cork *cork;
L
Linus Torvalds 已提交
1196 1197 1198 1199
	int hh_len;
	int mtu;
	int len;
	int err;
1200
	unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
L
Linus Torvalds 已提交
1201 1202 1203 1204 1205 1206 1207 1208 1209 1210

	if (inet->hdrincl)
		return -EPERM;

	if (flags&MSG_PROBE)
		return 0;

	if (skb_queue_empty(&sk->sk_write_queue))
		return -EINVAL;

1211 1212 1213 1214
	cork = &inet->cork.base;
	rt = (struct rtable *)cork->dst;
	if (cork->flags & IPCORK_OPT)
		opt = cork->opt;
L
Linus Torvalds 已提交
1215

1216
	if (!(rt->dst.dev->features&NETIF_F_SG))
L
Linus Torvalds 已提交
1217 1218
		return -EOPNOTSUPP;

1219
	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1220
	mtu = cork->fragsize;
L
Linus Torvalds 已提交
1221 1222 1223

	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
W
WANG Cong 已提交
1224
	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
L
Linus Torvalds 已提交
1225

1226
	if (cork->length + size > maxnonfragsize - fragheaderlen) {
1227 1228
		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
			       mtu - (opt ? opt->optlen : 0));
L
Linus Torvalds 已提交
1229 1230 1231
		return -EMSGSIZE;
	}

1232 1233
	skb = skb_peek_tail(&sk->sk_write_queue);
	if (!skb)
L
Linus Torvalds 已提交
1234 1235
		return -EINVAL;

1236
	cork->length += size;
1237 1238
	if ((size + skb->len > mtu) &&
	    (sk->sk_protocol == IPPROTO_UDP) &&
1239
	    (rt->dst.dev->features & NETIF_F_UFO)) {
1240
		skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
H
Herbert Xu 已提交
1241
		skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1242
	}
1243

L
Linus Torvalds 已提交
1244
	while (size > 0) {
1245
		if (skb_is_gso(skb)) {
1246
			len = size;
1247
		} else {
1248 1249 1250 1251 1252 1253

			/* Check if the remaining data fits into current packet. */
			len = mtu - skb->len;
			if (len < size)
				len = maxfraglen - skb->len;
		}
L
Linus Torvalds 已提交
1254 1255 1256 1257 1258
		if (len <= 0) {
			struct sk_buff *skb_prev;
			int alloclen;

			skb_prev = skb;
1259
			fraggap = skb_prev->len - maxfraglen;
L
Linus Torvalds 已提交
1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277

			alloclen = fragheaderlen + hh_len + fraggap + 15;
			skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
			if (unlikely(!skb)) {
				err = -ENOBUFS;
				goto error;
			}

			/*
			 *	Fill in the control structures
			 */
			skb->ip_summed = CHECKSUM_NONE;
			skb->csum = 0;
			skb_reserve(skb, hh_len);

			/*
			 *	Find where to start putting bytes.
			 */
1278
			skb_put(skb, fragheaderlen + fraggap);
1279
			skb_reset_network_header(skb);
1280 1281
			skb->transport_header = (skb->network_header +
						 fragheaderlen);
L
Linus Torvalds 已提交
1282
			if (fraggap) {
1283 1284
				skb->csum = skb_copy_and_csum_bits(skb_prev,
								   maxfraglen,
1285
						    skb_transport_header(skb),
1286
								   fraggap, 0);
L
Linus Torvalds 已提交
1287 1288
				skb_prev->csum = csum_sub(skb_prev->csum,
							  skb->csum);
1289
				pskb_trim_unique(skb_prev, maxfraglen);
L
Linus Torvalds 已提交
1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300
			}

			/*
			 * Put the packet on the pending queue.
			 */
			__skb_queue_tail(&sk->sk_write_queue, skb);
			continue;
		}

		if (len > size)
			len = size;
1301 1302

		if (skb_append_pagefrags(skb, page, offset, len)) {
L
Linus Torvalds 已提交
1303 1304 1305 1306 1307
			err = -EMSGSIZE;
			goto error;
		}

		if (skb->ip_summed == CHECKSUM_NONE) {
1308
			__wsum csum;
L
Linus Torvalds 已提交
1309 1310 1311 1312 1313 1314
			csum = csum_page(page, offset, len);
			skb->csum = csum_block_add(skb->csum, csum, skb->len);
		}

		skb->len += len;
		skb->data_len += len;
1315 1316
		skb->truesize += len;
		atomic_add(len, &sk->sk_wmem_alloc);
L
Linus Torvalds 已提交
1317 1318 1319 1320 1321 1322
		offset += len;
		size -= len;
	}
	return 0;

error:
1323
	cork->length -= size;
P
Pavel Emelyanov 已提交
1324
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
L
Linus Torvalds 已提交
1325 1326 1327
	return err;
}

1328
static void ip_cork_release(struct inet_cork *cork)
1329
{
1330 1331 1332 1333 1334
	cork->flags &= ~IPCORK_OPT;
	kfree(cork->opt);
	cork->opt = NULL;
	dst_release(cork->dst);
	cork->dst = NULL;
1335 1336
}

L
Linus Torvalds 已提交
1337 1338 1339 1340
/*
 *	Combined all pending IP fragments on the socket as one IP datagram
 *	and push them out.
 */
1341
struct sk_buff *__ip_make_skb(struct sock *sk,
1342
			      struct flowi4 *fl4,
1343 1344
			      struct sk_buff_head *queue,
			      struct inet_cork *cork)
L
Linus Torvalds 已提交
1345 1346 1347 1348
{
	struct sk_buff *skb, *tmp_skb;
	struct sk_buff **tail_skb;
	struct inet_sock *inet = inet_sk(sk);
1349
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
1350
	struct ip_options *opt = NULL;
1351
	struct rtable *rt = (struct rtable *)cork->dst;
L
Linus Torvalds 已提交
1352
	struct iphdr *iph;
1353
	__be16 df = 0;
L
Linus Torvalds 已提交
1354 1355
	__u8 ttl;

1356 1357
	skb = __skb_dequeue(queue);
	if (!skb)
L
Linus Torvalds 已提交
1358 1359 1360 1361
		goto out;
	tail_skb = &(skb_shinfo(skb)->frag_list);

	/* move skb->data to ip header from ext header */
1362
	if (skb->data < skb_network_header(skb))
1363
		__skb_pull(skb, skb_network_offset(skb));
1364
	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1365
		__skb_pull(tmp_skb, skb_network_header_len(skb));
L
Linus Torvalds 已提交
1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378
		*tail_skb = tmp_skb;
		tail_skb = &(tmp_skb->next);
		skb->len += tmp_skb->len;
		skb->data_len += tmp_skb->len;
		skb->truesize += tmp_skb->truesize;
		tmp_skb->destructor = NULL;
		tmp_skb->sk = NULL;
	}

	/* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
	 * to fragment the frame generated here. No matter, what transforms
	 * how transforms change size of the packet, it will come out.
	 */
W
WANG Cong 已提交
1379
	skb->ignore_df = ip_sk_ignore_df(sk);
L
Linus Torvalds 已提交
1380 1381

	/* DF bit is set when we want to see DF on outgoing frames.
W
WANG Cong 已提交
1382
	 * If ignore_df is set too, we still allow to fragment this frame
L
Linus Torvalds 已提交
1383
	 * locally. */
1384 1385
	if (inet->pmtudisc == IP_PMTUDISC_DO ||
	    inet->pmtudisc == IP_PMTUDISC_PROBE ||
1386 1387
	    (skb->len <= dst_mtu(&rt->dst) &&
	     ip_dont_fragment(sk, &rt->dst)))
L
Linus Torvalds 已提交
1388 1389
		df = htons(IP_DF);

1390 1391
	if (cork->flags & IPCORK_OPT)
		opt = cork->opt;
L
Linus Torvalds 已提交
1392

1393 1394 1395
	if (cork->ttl != 0)
		ttl = cork->ttl;
	else if (rt->rt_type == RTN_MULTICAST)
L
Linus Torvalds 已提交
1396 1397
		ttl = inet->mc_ttl;
	else
1398
		ttl = ip_select_ttl(inet, &rt->dst);
L
Linus Torvalds 已提交
1399

1400
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
1401 1402
	iph->version = 4;
	iph->ihl = 5;
1403
	iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
L
Linus Torvalds 已提交
1404 1405 1406
	iph->frag_off = df;
	iph->ttl = ttl;
	iph->protocol = sk->sk_protocol;
1407
	ip_copy_addrs(iph, fl4);
1408
	ip_select_ident(net, skb, sk);
L
Linus Torvalds 已提交
1409

1410 1411 1412 1413 1414
	if (opt) {
		iph->ihl += opt->optlen>>2;
		ip_options_build(skb, opt, cork->addr, rt, 0);
	}

1415
	skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1416
	skb->mark = sk->sk_mark;
1417 1418 1419 1420
	/*
	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
	 * on dst refcount
	 */
1421
	cork->dst = NULL;
1422
	skb_dst_set(skb, &rt->dst);
L
Linus Torvalds 已提交
1423

1424
	if (iph->protocol == IPPROTO_ICMP)
1425
		icmp_out_count(net, ((struct icmphdr *)
1426 1427
			skb_transport_header(skb))->type);

1428 1429 1430 1431 1432
	ip_cork_release(cork);
out:
	return skb;
}

E
Eric Dumazet 已提交
1433
int ip_send_skb(struct net *net, struct sk_buff *skb)
1434 1435 1436
{
	int err;

1437
	err = ip_local_out(net, skb->sk, skb);
L
Linus Torvalds 已提交
1438 1439
	if (err) {
		if (err > 0)
E
Eric Dumazet 已提交
1440
			err = net_xmit_errno(err);
L
Linus Torvalds 已提交
1441
		if (err)
1442
			IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
L
Linus Torvalds 已提交
1443 1444 1445 1446 1447
	}

	return err;
}

1448
int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1449
{
1450 1451
	struct sk_buff *skb;

1452
	skb = ip_finish_skb(sk, fl4);
1453 1454 1455 1456
	if (!skb)
		return 0;

	/* Netfilter gets whole the not fragmented skb. */
E
Eric Dumazet 已提交
1457
	return ip_send_skb(sock_net(sk), skb);
1458 1459
}

L
Linus Torvalds 已提交
1460 1461 1462
/*
 *	Throw away all pending data on the socket.
 */
1463 1464 1465
static void __ip_flush_pending_frames(struct sock *sk,
				      struct sk_buff_head *queue,
				      struct inet_cork *cork)
L
Linus Torvalds 已提交
1466 1467 1468
{
	struct sk_buff *skb;

1469
	while ((skb = __skb_dequeue_tail(queue)) != NULL)
L
Linus Torvalds 已提交
1470 1471
		kfree_skb(skb);

1472 1473 1474 1475 1476
	ip_cork_release(cork);
}

void ip_flush_pending_frames(struct sock *sk)
{
1477
	__ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
L
Linus Torvalds 已提交
1478 1479
}

1480
struct sk_buff *ip_make_skb(struct sock *sk,
1481
			    struct flowi4 *fl4,
1482 1483 1484 1485 1486 1487
			    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)
{
1488
	struct inet_cork cork;
1489 1490 1491 1492 1493 1494 1495 1496
	struct sk_buff_head queue;
	int err;

	if (flags & MSG_PROBE)
		return NULL;

	__skb_queue_head_init(&queue);

1497 1498
	cork.flags = 0;
	cork.addr = 0;
1499
	cork.opt = NULL;
1500 1501 1502 1503
	err = ip_setup_cork(sk, &cork, ipc, rtp);
	if (err)
		return ERR_PTR(err);

1504 1505
	err = __ip_append_data(sk, fl4, &queue, &cork,
			       &current->task_frag, getfrag,
1506 1507 1508 1509 1510 1511
			       from, length, transhdrlen, flags);
	if (err) {
		__ip_flush_pending_frames(sk, &queue, &cork);
		return ERR_PTR(err);
	}

1512
	return __ip_make_skb(sk, fl4, &queue, &cork);
1513
}
L
Linus Torvalds 已提交
1514 1515 1516 1517

/*
 *	Fetch data from kernel space and fill in checksum if needed.
 */
1518
static int ip_reply_glue_bits(void *dptr, char *to, int offset,
L
Linus Torvalds 已提交
1519 1520
			      int len, int odd, struct sk_buff *skb)
{
1521
	__wsum csum;
L
Linus Torvalds 已提交
1522 1523 1524

	csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
	skb->csum = csum_block_add(skb->csum, csum, odd);
1525
	return 0;
L
Linus Torvalds 已提交
1526 1527
}

1528
/*
L
Linus Torvalds 已提交
1529
 *	Generic function to send a packet as reply to another packet.
1530
 *	Used to send some TCP resets/acks so far.
L
Linus Torvalds 已提交
1531
 */
1532
void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1533 1534 1535
			   const struct ip_options *sopt,
			   __be32 daddr, __be32 saddr,
			   const struct ip_reply_arg *arg,
1536
			   unsigned int len)
L
Linus Torvalds 已提交
1537
{
1538
	struct ip_options_data replyopts;
L
Linus Torvalds 已提交
1539
	struct ipcm_cookie ipc;
1540
	struct flowi4 fl4;
E
Eric Dumazet 已提交
1541
	struct rtable *rt = skb_rtable(skb);
1542
	struct net *net = sock_net(sk);
1543
	struct sk_buff *nskb;
1544
	int err;
1545
	int oif;
L
Linus Torvalds 已提交
1546

1547
	if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
L
Linus Torvalds 已提交
1548 1549
		return;

1550
	ipc.addr = daddr;
L
Linus Torvalds 已提交
1551
	ipc.opt = NULL;
1552
	ipc.tx_flags = 0;
1553 1554
	ipc.ttl = 0;
	ipc.tos = -1;
L
Linus Torvalds 已提交
1555

1556
	if (replyopts.opt.opt.optlen) {
L
Linus Torvalds 已提交
1557 1558
		ipc.opt = &replyopts.opt;

1559 1560
		if (replyopts.opt.opt.srr)
			daddr = replyopts.opt.opt.faddr;
L
Linus Torvalds 已提交
1561 1562
	}

1563
	oif = arg->bound_dev_if;
1564
	if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1565 1566 1567
		oif = skb->skb_iif;

	flowi4_init_output(&fl4, oif,
1568
			   IP4_REPLY_MARK(net, skb->mark),
1569
			   RT_TOS(arg->tos),
1570
			   RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1571
			   ip_reply_arg_flowi_flags(arg),
1572
			   daddr, saddr,
1573 1574
			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1575
	rt = ip_route_output_key(net, &fl4);
1576 1577
	if (IS_ERR(rt))
		return;
L
Linus Torvalds 已提交
1578

1579
	inet_sk(sk)->tos = arg->tos;
L
Linus Torvalds 已提交
1580 1581

	sk->sk_priority = skb->priority;
1582
	sk->sk_protocol = ip_hdr(skb)->protocol;
1583
	sk->sk_bound_dev_if = arg->bound_dev_if;
1584
	sk->sk_sndbuf = sysctl_wmem_default;
1585 1586 1587 1588 1589 1590 1591
	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
			     len, 0, &ipc, &rt, MSG_DONTWAIT);
	if (unlikely(err)) {
		ip_flush_pending_frames(sk);
		goto out;
	}

1592 1593
	nskb = skb_peek(&sk->sk_write_queue);
	if (nskb) {
L
Linus Torvalds 已提交
1594
		if (arg->csumoffset >= 0)
1595 1596
			*((__sum16 *)skb_transport_header(nskb) +
			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1597
								arg->csum));
1598
		nskb->ip_summed = CHECKSUM_NONE;
1599
		ip_push_pending_frames(sk, &fl4);
L
Linus Torvalds 已提交
1600
	}
1601
out:
L
Linus Torvalds 已提交
1602 1603 1604 1605 1606 1607 1608 1609
	ip_rt_put(rt);
}

void __init ip_init(void)
{
	ip_rt_init();
	inet_initpeers();

1610 1611
#if defined(CONFIG_IP_MULTICAST)
	igmp_mc_init();
L
Linus Torvalds 已提交
1612 1613
#endif
}