ip_output.c 38.1 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

/* Generate a checksum for an outgoing IP datagram. */
87
void ip_send_check(struct iphdr *iph)
L
Linus Torvalds 已提交
88 89 90 91
{
	iph->check = 0;
	iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
}
E
Eric Dumazet 已提交
92
EXPORT_SYMBOL(ip_send_check);
L
Linus Torvalds 已提交
93

H
Herbert Xu 已提交
94 95 96 97 98 99
int __ip_local_out(struct sk_buff *skb)
{
	struct iphdr *iph = ip_hdr(skb);

	iph->tot_len = htons(skb->len);
	ip_send_check(iph);
100 101
	return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
		       skb_dst(skb)->dev, dst_output);
H
Herbert Xu 已提交
102 103 104 105 106 107 108 109 110 111 112 113 114 115
}

int ip_local_out(struct sk_buff *skb)
{
	int err;

	err = __ip_local_out(skb);
	if (likely(err == 1))
		err = dst_output(skb);

	return err;
}
EXPORT_SYMBOL_GPL(ip_local_out);

L
Linus Torvalds 已提交
116 117 118 119 120
static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
{
	int ttl = inet->uc_ttl;

	if (ttl < 0)
121
		ttl = ip4_dst_hoplimit(dst);
L
Linus Torvalds 已提交
122 123 124
	return ttl;
}

125
/*
L
Linus Torvalds 已提交
126 127 128 129
 *		Add an ip header to a skbuff and send it out.
 *
 */
int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
130
			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
L
Linus Torvalds 已提交
131 132
{
	struct inet_sock *inet = inet_sk(sk);
E
Eric Dumazet 已提交
133
	struct rtable *rt = skb_rtable(skb);
L
Linus Torvalds 已提交
134 135 136
	struct iphdr *iph;

	/* Build the IP header. */
137
	skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
138
	skb_reset_network_header(skb);
139
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
140 141 142
	iph->version  = 4;
	iph->ihl      = 5;
	iph->tos      = inet->tos;
143
	if (ip_dont_fragment(sk, &rt->dst))
L
Linus Torvalds 已提交
144 145 146
		iph->frag_off = htons(IP_DF);
	else
		iph->frag_off = 0;
147
	iph->ttl      = ip_select_ttl(inet, &rt->dst);
148 149
	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
	iph->saddr    = saddr;
L
Linus Torvalds 已提交
150
	iph->protocol = sk->sk_protocol;
151
	ip_select_ident(skb, &rt->dst, sk);
L
Linus Torvalds 已提交
152

153 154 155
	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
156 157 158
	}

	skb->priority = sk->sk_priority;
159
	skb->mark = sk->sk_mark;
L
Linus Torvalds 已提交
160 161

	/* Send it out. */
H
Herbert Xu 已提交
162
	return ip_local_out(skb);
L
Linus Torvalds 已提交
163
}
164 165
EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);

L
Linus Torvalds 已提交
166 167
static inline int ip_finish_output2(struct sk_buff *skb)
{
E
Eric Dumazet 已提交
168
	struct dst_entry *dst = skb_dst(skb);
169
	struct rtable *rt = (struct rtable *)dst;
L
Linus Torvalds 已提交
170
	struct net_device *dev = dst->dev;
171
	unsigned int hh_len = LL_RESERVED_SPACE(dev);
172
	struct neighbour *neigh;
173
	u32 nexthop;
L
Linus Torvalds 已提交
174

175 176 177 178
	if (rt->rt_type == RTN_MULTICAST) {
		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
	} else if (rt->rt_type == RTN_BROADCAST)
		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
179

L
Linus Torvalds 已提交
180
	/* Be paranoid, rather than too clever. */
181
	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
L
Linus Torvalds 已提交
182 183 184 185 186 187 188 189 190
		struct sk_buff *skb2;

		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
		if (skb2 == NULL) {
			kfree_skb(skb);
			return -ENOMEM;
		}
		if (skb->sk)
			skb_set_owner_w(skb2, skb->sk);
191
		consume_skb(skb);
L
Linus Torvalds 已提交
192 193 194
		skb = skb2;
	}

195
	rcu_read_lock_bh();
J
Julian Anastasov 已提交
196
	nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
197 198 199
	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
	if (unlikely(!neigh))
		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
200
	if (!IS_ERR(neigh)) {
201
		int res = dst_neigh_output(dst, neigh, skb);
202

203
		rcu_read_unlock_bh();
204 205
		return res;
	}
206
	rcu_read_unlock_bh();
207

208 209
	net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
			    __func__);
L
Linus Torvalds 已提交
210 211 212 213
	kfree_skb(skb);
	return -EINVAL;
}

P
Patrick McHardy 已提交
214
static int ip_finish_output(struct sk_buff *skb)
L
Linus Torvalds 已提交
215
{
216 217
#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
	/* Policy lookup after SNAT yielded a new policy */
E
Eric Dumazet 已提交
218
	if (skb_dst(skb)->xfrm != NULL) {
219 220 221
		IPCB(skb)->flags |= IPSKB_REROUTED;
		return dst_output(skb);
	}
222
#endif
J
John Heffner 已提交
223
	if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
224 225 226
		return ip_fragment(skb, ip_finish_output2);
	else
		return ip_finish_output2(skb);
L
Linus Torvalds 已提交
227 228 229 230 231
}

int ip_mc_output(struct sk_buff *skb)
{
	struct sock *sk = skb->sk;
E
Eric Dumazet 已提交
232
	struct rtable *rt = skb_rtable(skb);
233
	struct net_device *dev = rt->dst.dev;
L
Linus Torvalds 已提交
234 235 236 237

	/*
	 *	If the indicated interface is up and running, send the packet.
	 */
238
	IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
L
Linus Torvalds 已提交
239 240 241 242 243 244 245 246 247

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

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

	if (rt->rt_flags&RTCF_MULTICAST) {
248
		if (sk_mc_loop(sk)
L
Linus Torvalds 已提交
249 250 251 252 253 254 255 256 257
#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.
		 */
258 259 260
		    &&
		    ((rt->rt_flags & RTCF_LOCAL) ||
		     !(IPCB(skb)->flags & IPSKB_FORWARDED))
L
Linus Torvalds 已提交
261
#endif
262
		   ) {
L
Linus Torvalds 已提交
263 264
			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
			if (newskb)
265 266
				NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
					newskb, NULL, newskb->dev,
267
					dev_loopback_xmit);
L
Linus Torvalds 已提交
268 269 270 271
		}

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

272
		if (ip_hdr(skb)->ttl == 0) {
L
Linus Torvalds 已提交
273 274 275 276 277 278 279 280
			kfree_skb(skb);
			return 0;
		}
	}

	if (rt->rt_flags&RTCF_BROADCAST) {
		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
		if (newskb)
281
			NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
282
				NULL, newskb->dev, dev_loopback_xmit);
L
Linus Torvalds 已提交
283 284
	}

285 286
	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
			    skb->dev, ip_finish_output,
287
			    !(IPCB(skb)->flags & IPSKB_REROUTED));
L
Linus Torvalds 已提交
288 289 290 291
}

int ip_output(struct sk_buff *skb)
{
E
Eric Dumazet 已提交
292
	struct net_device *dev = skb_dst(skb)->dev;
293

294
	IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
L
Linus Torvalds 已提交
295

296 297 298
	skb->dev = dev;
	skb->protocol = htons(ETH_P_IP);

299
	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
300
			    ip_finish_output,
301
			    !(IPCB(skb)->flags & IPSKB_REROUTED));
L
Linus Torvalds 已提交
302 303
}

304 305 306 307 308 309 310 311 312 313 314 315 316 317
/*
 * 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));
}

318
int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl)
L
Linus Torvalds 已提交
319
{
320
	struct sock *sk = skb->sk;
L
Linus Torvalds 已提交
321
	struct inet_sock *inet = inet_sk(sk);
322
	struct ip_options_rcu *inet_opt;
323
	struct flowi4 *fl4;
L
Linus Torvalds 已提交
324 325
	struct rtable *rt;
	struct iphdr *iph;
326
	int res;
L
Linus Torvalds 已提交
327 328 329 330

	/* Skip all of this if the packet is already routed,
	 * f.e. by something like SCTP.
	 */
331
	rcu_read_lock();
332
	inet_opt = rcu_dereference(inet->inet_opt);
333
	fl4 = &fl->u.ip4;
E
Eric Dumazet 已提交
334
	rt = skb_rtable(skb);
L
Linus Torvalds 已提交
335 336 337 338 339 340
	if (rt != NULL)
		goto packet_routed;

	/* Make sure we can route this packet. */
	rt = (struct rtable *)__sk_dst_check(sk, 0);
	if (rt == NULL) {
A
Al Viro 已提交
341
		__be32 daddr;
L
Linus Torvalds 已提交
342 343

		/* Use correct destination address if we have options. */
E
Eric Dumazet 已提交
344
		daddr = inet->inet_daddr;
345 346
		if (inet_opt && inet_opt->opt.srr)
			daddr = inet_opt->opt.faddr;
L
Linus Torvalds 已提交
347

348 349 350 351
		/* If this fails, retransmit mechanism of transport layer will
		 * keep trying until route appears or the connection times
		 * itself out.
		 */
352
		rt = ip_route_output_ports(sock_net(sk), fl4, sk,
353 354 355 356 357 358 359 360
					   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;
361
		sk_setup_caps(sk, &rt->dst);
L
Linus Torvalds 已提交
362
	}
363
	skb_dst_set_noref(skb, &rt->dst);
L
Linus Torvalds 已提交
364 365

packet_routed:
J
Julian Anastasov 已提交
366
	if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
L
Linus Torvalds 已提交
367 368 369
		goto no_route;

	/* OK, we know where to send it, allocate and build IP header. */
370
	skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
371
	skb_reset_network_header(skb);
372
	iph = ip_hdr(skb);
373
	*((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
374
	if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df)
L
Linus Torvalds 已提交
375 376 377
		iph->frag_off = htons(IP_DF);
	else
		iph->frag_off = 0;
378
	iph->ttl      = ip_select_ttl(inet, &rt->dst);
L
Linus Torvalds 已提交
379
	iph->protocol = sk->sk_protocol;
380 381
	ip_copy_addrs(iph, fl4);

L
Linus Torvalds 已提交
382 383
	/* Transport layer set skb->h.foo itself. */

384 385 386
	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
387 388
	}

389
	ip_select_ident_more(skb, &rt->dst, sk,
390
			     (skb_shinfo(skb)->gso_segs ?: 1) - 1);
L
Linus Torvalds 已提交
391 392

	skb->priority = sk->sk_priority;
393
	skb->mark = sk->sk_mark;
L
Linus Torvalds 已提交
394

395 396 397
	res = ip_local_out(skb);
	rcu_read_unlock();
	return res;
L
Linus Torvalds 已提交
398 399

no_route:
400
	rcu_read_unlock();
P
Pavel Emelyanov 已提交
401
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
L
Linus Torvalds 已提交
402 403 404
	kfree_skb(skb);
	return -EHOSTUNREACH;
}
E
Eric Dumazet 已提交
405
EXPORT_SYMBOL(ip_queue_xmit);
L
Linus Torvalds 已提交
406 407 408 409 410 411 412


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
413
	skb_dst_drop(to);
414
	skb_dst_copy(to, from);
L
Linus Torvalds 已提交
415
	to->dev = from->dev;
T
Thomas Graf 已提交
416
	to->mark = from->mark;
L
Linus Torvalds 已提交
417 418 419 420 421 422 423

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

#ifdef CONFIG_NET_SCHED
	to->tc_index = from->tc_index;
#endif
424
	nf_copy(to, from);
425
#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
426 427
	to->nf_trace = from->nf_trace;
#endif
428 429
#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
	to->ipvs_property = from->ipvs_property;
L
Linus Torvalds 已提交
430
#endif
431
	skb_copy_secmark(to, from);
L
Linus Torvalds 已提交
432 433 434 435 436 437 438 439 440
}

/*
 *	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.
 */

441
int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
L
Linus Torvalds 已提交
442 443 444 445 446
{
	struct iphdr *iph;
	int ptr;
	struct net_device *dev;
	struct sk_buff *skb2;
447
	unsigned int mtu, hlen, left, len, ll_rs;
L
Linus Torvalds 已提交
448
	int offset;
449
	__be16 not_last_frag;
E
Eric Dumazet 已提交
450
	struct rtable *rt = skb_rtable(skb);
L
Linus Torvalds 已提交
451 452
	int err = 0;

453
	dev = rt->dst.dev;
L
Linus Torvalds 已提交
454 455 456 457 458

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

459
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
460

461 462 463
	if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) ||
		     (IPCB(skb)->frag_max_size &&
		      IPCB(skb)->frag_max_size > dst_mtu(&rt->dst)))) {
P
Pavel Emelyanov 已提交
464
		IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
465
		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
J
John Heffner 已提交
466
			  htonl(ip_skb_dst_mtu(skb)));
L
Linus Torvalds 已提交
467 468 469 470 471 472 473 474 475
		kfree_skb(skb);
		return -EMSGSIZE;
	}

	/*
	 *	Setup starting values.
	 */

	hlen = iph->ihl * 4;
476
	mtu = dst_mtu(&rt->dst) - hlen;	/* Size of data space */
477 478 479 480
#ifdef CONFIG_BRIDGE_NETFILTER
	if (skb->nf_bridge)
		mtu -= nf_bridge_mtu_reduction(skb);
#endif
H
Herbert Xu 已提交
481
	IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
L
Linus Torvalds 已提交
482 483 484 485 486 487 488 489

	/* 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.
	 */
490
	if (skb_has_frag_list(skb)) {
491
		struct sk_buff *frag, *frag2;
L
Linus Torvalds 已提交
492 493 494 495
		int first_len = skb_pagelen(skb);

		if (first_len - hlen > mtu ||
		    ((first_len - hlen) & 7) ||
496
		    ip_is_fragment(iph) ||
L
Linus Torvalds 已提交
497 498 499
		    skb_cloned(skb))
			goto slow_path;

500
		skb_walk_frags(skb, frag) {
L
Linus Torvalds 已提交
501 502 503 504
			/* Correct geometry. */
			if (frag->len > mtu ||
			    ((frag->len & 7) && frag->next) ||
			    skb_headroom(frag) < hlen)
505
				goto slow_path_clean;
L
Linus Torvalds 已提交
506 507 508

			/* Partially cloned skb? */
			if (skb_shared(frag))
509
				goto slow_path_clean;
510 511 512 513 514 515

			BUG_ON(frag->sk);
			if (skb->sk) {
				frag->sk = skb->sk;
				frag->destructor = sock_wfree;
			}
516
			skb->truesize -= frag->truesize;
L
Linus Torvalds 已提交
517 518 519 520 521 522 523
		}

		/* Everything is OK. Generate! */

		err = 0;
		offset = 0;
		frag = skb_shinfo(skb)->frag_list;
524
		skb_frag_list_init(skb);
L
Linus Torvalds 已提交
525 526 527 528 529 530 531 532 533 534 535
		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;
536
				skb_reset_transport_header(frag);
537 538
				__skb_push(frag, hlen);
				skb_reset_network_header(frag);
539
				memcpy(skb_network_header(frag), iph, hlen);
540
				iph = ip_hdr(frag);
L
Linus Torvalds 已提交
541 542 543 544 545 546 547 548 549 550 551 552 553 554
				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);
				if (frag->next != NULL)
					iph->frag_off |= htons(IP_MF);
				/* Ready, complete checksum */
				ip_send_check(iph);
			}

			err = output(skb);

555
			if (!err)
P
Pavel Emelyanov 已提交
556
				IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
L
Linus Torvalds 已提交
557 558 559 560 561 562 563 564 565
			if (err || !frag)
				break;

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

		if (err == 0) {
P
Pavel Emelyanov 已提交
566
			IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
L
Linus Torvalds 已提交
567 568 569 570 571 572 573 574
			return 0;
		}

		while (frag) {
			skb = frag->next;
			kfree_skb(frag);
			frag = skb;
		}
P
Pavel Emelyanov 已提交
575
		IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
576
		return err;
577 578 579 580 581 582 583 584 585

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
586 587 588
	}

slow_path:
589 590 591
	/* for offloaded checksums cleanup checksum before fragmentation */
	if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
		goto fail;
592
	iph = ip_hdr(skb);
593

L
Linus Torvalds 已提交
594
	left = skb->len - hlen;		/* Space per frame */
595
	ptr = hlen;		/* Where to start from */
L
Linus Torvalds 已提交
596 597

	/* for bridged IP traffic encapsulated inside f.e. a vlan header,
598 599
	 * we need to make room for the encapsulating header
	 */
600
	ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
601

L
Linus Torvalds 已提交
602 603 604 605 606 607 608 609 610 611 612
	/*
	 *	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
613
	while (left > 0) {
L
Linus Torvalds 已提交
614 615 616 617
		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
618
		/* IF: we are not sending up to and including the packet end
L
Linus Torvalds 已提交
619 620 621 622 623 624 625 626 627
		   then align the next start on an eight byte boundary */
		if (len < left)	{
			len &= ~7;
		}
		/*
		 *	Allocate buffer.
		 */

		if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
628
			NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
L
Linus Torvalds 已提交
629 630 631 632 633 634 635 636 637 638 639
			err = -ENOMEM;
			goto fail;
		}

		/*
		 *	Set up data on packet
		 */

		ip_copy_metadata(skb2, skb);
		skb_reserve(skb2, ll_rs);
		skb_put(skb2, len + hlen);
640
		skb_reset_network_header(skb2);
641
		skb2->transport_header = skb2->network_header + hlen;
L
Linus Torvalds 已提交
642 643 644 645 646 647 648 649 650 651 652 653 654

		/*
		 *	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.
		 */

655
		skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
L
Linus Torvalds 已提交
656 657 658 659

		/*
		 *	Copy a block of the IP datagram.
		 */
660
		if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
L
Linus Torvalds 已提交
661 662 663 664 665 666
			BUG();
		left -= len;

		/*
		 *	Fill in the new header fields.
		 */
667
		iph = ip_hdr(skb2);
L
Linus Torvalds 已提交
668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697
		iph->frag_off = htons((offset >> 3));

		/* 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);

		err = output(skb2);
		if (err)
			goto fail;
698

P
Pavel Emelyanov 已提交
699
		IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
L
Linus Torvalds 已提交
700
	}
701
	consume_skb(skb);
P
Pavel Emelyanov 已提交
702
	IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
L
Linus Torvalds 已提交
703 704 705
	return err;

fail:
706
	kfree_skb(skb);
P
Pavel Emelyanov 已提交
707
	IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
708 709
	return err;
}
710 711
EXPORT_SYMBOL(ip_fragment);

L
Linus Torvalds 已提交
712 713 714 715 716
int
ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
{
	struct iovec *iov = from;

717
	if (skb->ip_summed == CHECKSUM_PARTIAL) {
L
Linus Torvalds 已提交
718 719 720
		if (memcpy_fromiovecend(to, iov, offset, len) < 0)
			return -EFAULT;
	} else {
721
		__wsum csum = 0;
L
Linus Torvalds 已提交
722 723 724 725 726 727
		if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
			return -EFAULT;
		skb->csum = csum_block_add(skb->csum, csum, odd);
	}
	return 0;
}
E
Eric Dumazet 已提交
728
EXPORT_SYMBOL(ip_generic_getfrag);
L
Linus Torvalds 已提交
729

730
static inline __wsum
L
Linus Torvalds 已提交
731 732 733
csum_page(struct page *page, int offset, int copy)
{
	char *kaddr;
734
	__wsum csum;
L
Linus Torvalds 已提交
735 736 737 738 739 740
	kaddr = kmap(page);
	csum = csum_partial(kaddr + offset, copy, 0);
	kunmap(page);
	return csum;
}

A
Adrian Bunk 已提交
741
static inline int ip_ufo_append_data(struct sock *sk,
742
			struct sk_buff_head *queue,
743 744 745
			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,
746
			int transhdrlen, int maxfraglen, unsigned int flags)
747 748 749 750 751 752 753 754
{
	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
	 */
755
	if ((skb = skb_peek_tail(queue)) == NULL) {
756 757 758 759 760 761 762 763 764 765 766
		skb = sock_alloc_send_skb(sk,
			hh_len + fragheaderlen + transhdrlen + 20,
			(flags & MSG_DONTWAIT), &err);

		if (skb == NULL)
			return err;

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

		/* create space for UDP/IP header */
767
		skb_put(skb, fragheaderlen + transhdrlen);
768 769

		/* initialize network header pointer */
770
		skb_reset_network_header(skb);
771 772

		/* initialize protocol header pointer */
773
		skb->transport_header = skb->network_header + fragheaderlen;
774 775 776

		skb->csum = 0;

777

778
		__skb_queue_tail(queue, skb);
779 780
	} else if (skb_is_gso(skb)) {
		goto append;
781
	}
782

783 784 785 786 787 788
	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:
789 790
	return skb_append_datato_frags(sk, skb, getfrag, from,
				       (length - transhdrlen));
791 792
}

793 794 795
static int __ip_append_data(struct sock *sk,
			    struct flowi4 *fl4,
			    struct sk_buff_head *queue,
796
			    struct inet_cork *cork,
797
			    struct page_frag *pfrag,
798 799 800 801
			    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
802 803 804 805
{
	struct inet_sock *inet = inet_sk(sk);
	struct sk_buff *skb;

806
	struct ip_options *opt = cork->opt;
L
Linus Torvalds 已提交
807 808 809 810 811 812
	int hh_len;
	int exthdrlen;
	int mtu;
	int copy;
	int err;
	int offset = 0;
813
	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
L
Linus Torvalds 已提交
814
	int csummode = CHECKSUM_NONE;
815
	struct rtable *rt = (struct rtable *)cork->dst;
L
Linus Torvalds 已提交
816

817 818 819
	skb = skb_peek_tail(queue);

	exthdrlen = !skb ? rt->dst.header_len : 0;
820
	mtu = cork->fragsize;
L
Linus Torvalds 已提交
821

822
	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
L
Linus Torvalds 已提交
823 824 825

	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
826 827
	maxnonfragsize = (inet->pmtudisc >= IP_PMTUDISC_DO) ?
			 mtu : 0xFFFF;
L
Linus Torvalds 已提交
828

829
	if (cork->length + length > maxnonfragsize - fragheaderlen) {
830
		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
E
Eric Dumazet 已提交
831
			       mtu-exthdrlen);
L
Linus Torvalds 已提交
832 833 834 835 836 837 838 839 840
		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 &&
841
	    rt->dst.dev->features & NETIF_F_V4_CSUM &&
L
Linus Torvalds 已提交
842
	    !exthdrlen)
843
		csummode = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
844

845
	cork->length += length;
846
	if (((length > mtu) || (skb && skb_is_gso(skb))) &&
847
	    (sk->sk_protocol == IPPROTO_UDP) &&
848
	    (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
849 850
		err = ip_ufo_append_data(sk, queue, getfrag, from, length,
					 hh_len, fragheaderlen, transhdrlen,
851
					 maxfraglen, flags);
852
		if (err)
853 854 855
			goto error;
		return 0;
	}
L
Linus Torvalds 已提交
856 857 858 859 860 861 862 863

	/* 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.
	 */

864
	if (!skb)
L
Linus Torvalds 已提交
865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894
		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;

895
			if ((flags & MSG_MORE) &&
896
			    !(rt->dst.dev->features&NETIF_F_SG))
L
Linus Torvalds 已提交
897 898
				alloclen = mtu;
			else
899
				alloclen = fraglen;
L
Linus Torvalds 已提交
900

901 902
			alloclen += exthdrlen;

L
Linus Torvalds 已提交
903 904 905 906 907
			/* 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.
			 */
908
			if (datalen == length + fraggap)
909
				alloclen += rt->dst.trailer_len;
910

L
Linus Torvalds 已提交
911
			if (transhdrlen) {
912
				skb = sock_alloc_send_skb(sk,
L
Linus Torvalds 已提交
913 914 915 916 917 918
						alloclen + hh_len + 15,
						(flags & MSG_DONTWAIT), &err);
			} else {
				skb = NULL;
				if (atomic_read(&sk->sk_wmem_alloc) <=
				    2 * sk->sk_sndbuf)
919
					skb = sock_wmalloc(sk,
L
Linus Torvalds 已提交
920 921 922 923
							   alloclen + hh_len + 15, 1,
							   sk->sk_allocation);
				if (unlikely(skb == NULL))
					err = -ENOBUFS;
924 925 926
				else
					/* only the initial fragment is
					   time stamped */
927
					cork->tx_flags = 0;
L
Linus Torvalds 已提交
928 929 930 931 932 933 934 935 936 937
			}
			if (skb == NULL)
				goto error;

			/*
			 *	Fill in the control structures
			 */
			skb->ip_summed = csummode;
			skb->csum = 0;
			skb_reserve(skb, hh_len);
938
			skb_shinfo(skb)->tx_flags = cork->tx_flags;
L
Linus Torvalds 已提交
939 940 941 942

			/*
			 *	Find where to start putting bytes.
			 */
943
			data = skb_put(skb, fraglen + exthdrlen);
944
			skb_set_network_header(skb, exthdrlen);
945 946
			skb->transport_header = (skb->network_header +
						 fragheaderlen);
947
			data += fragheaderlen + exthdrlen;
L
Linus Torvalds 已提交
948 949 950 951 952 953 954 955

			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;
956
				pskb_trim_unique(skb_prev, maxfraglen);
L
Linus Torvalds 已提交
957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974
			}

			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.
			 */
975
			__skb_queue_tail(queue, skb);
L
Linus Torvalds 已提交
976 977 978 979 980 981
			continue;
		}

		if (copy > length)
			copy = length;

982
		if (!(rt->dst.dev->features&NETIF_F_SG)) {
L
Linus Torvalds 已提交
983 984 985
			unsigned int off;

			off = skb->len;
986
			if (getfrag(from, skb_put(skb, copy),
L
Linus Torvalds 已提交
987 988 989 990 991 992 993 994
					offset, copy, off, skb) < 0) {
				__skb_trim(skb, off);
				err = -EFAULT;
				goto error;
			}
		} else {
			int i = skb_shinfo(skb)->nr_frags;

995 996
			err = -ENOMEM;
			if (!sk_page_frag_refill(sk, pfrag))
L
Linus Torvalds 已提交
997
				goto error;
998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008

			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
1009
			}
1010 1011 1012 1013 1014 1015 1016 1017
			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
1018 1019
			skb->len += copy;
			skb->data_len += copy;
1020 1021
			skb->truesize += copy;
			atomic_add(copy, &sk->sk_wmem_alloc);
L
Linus Torvalds 已提交
1022 1023 1024 1025 1026 1027 1028
		}
		offset += copy;
		length -= copy;
	}

	return 0;

1029 1030
error_efault:
	err = -EFAULT;
L
Linus Torvalds 已提交
1031
error:
1032
	cork->length -= length;
P
Pavel Emelyanov 已提交
1033
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1034
	return err;
L
Linus Torvalds 已提交
1035 1036
}

1037 1038 1039 1040
static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
			 struct ipcm_cookie *ipc, struct rtable **rtp)
{
	struct inet_sock *inet = inet_sk(sk);
1041
	struct ip_options_rcu *opt;
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
	struct rtable *rt;

	/*
	 * setup for corking.
	 */
	opt = ipc->opt;
	if (opt) {
		if (cork->opt == NULL) {
			cork->opt = kmalloc(sizeof(struct ip_options) + 40,
					    sk->sk_allocation);
			if (unlikely(cork->opt == NULL))
				return -ENOBUFS;
		}
1055
		memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
		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;
	cork->fragsize = inet->pmtudisc == IP_PMTUDISC_PROBE ?
1067
			 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1068 1069
	cork->dst = &rt->dst;
	cork->length = 0;
1070 1071 1072
	cork->ttl = ipc->ttl;
	cork->tos = ipc->tos;
	cork->priority = ipc->priority;
1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088
	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.
 */
1089
int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
		   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)) {
1103
		err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1104 1105 1106 1107 1108 1109
		if (err)
			return err;
	} else {
		transhdrlen = 0;
	}

1110 1111
	return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
				sk_page_frag(sk), getfrag,
1112 1113 1114
				from, length, transhdrlen, flags);
}

1115
ssize_t	ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
L
Linus Torvalds 已提交
1116 1117 1118 1119 1120 1121
		       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;
1122
	struct inet_cork *cork;
L
Linus Torvalds 已提交
1123 1124 1125 1126
	int hh_len;
	int mtu;
	int len;
	int err;
1127
	unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
L
Linus Torvalds 已提交
1128 1129 1130 1131 1132 1133 1134 1135 1136 1137

	if (inet->hdrincl)
		return -EPERM;

	if (flags&MSG_PROBE)
		return 0;

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

1138 1139 1140 1141
	cork = &inet->cork.base;
	rt = (struct rtable *)cork->dst;
	if (cork->flags & IPCORK_OPT)
		opt = cork->opt;
L
Linus Torvalds 已提交
1142

1143
	if (!(rt->dst.dev->features&NETIF_F_SG))
L
Linus Torvalds 已提交
1144 1145
		return -EOPNOTSUPP;

1146
	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1147
	mtu = cork->fragsize;
L
Linus Torvalds 已提交
1148 1149 1150

	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1151 1152
	maxnonfragsize = (inet->pmtudisc >= IP_PMTUDISC_DO) ?
			 mtu : 0xFFFF;
L
Linus Torvalds 已提交
1153

1154
	if (cork->length + size > maxnonfragsize - fragheaderlen) {
1155
		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, mtu);
L
Linus Torvalds 已提交
1156 1157 1158 1159 1160 1161
		return -EMSGSIZE;
	}

	if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
		return -EINVAL;

1162
	cork->length += size;
1163 1164
	if ((size + skb->len > mtu) &&
	    (sk->sk_protocol == IPPROTO_UDP) &&
1165
	    (rt->dst.dev->features & NETIF_F_UFO)) {
1166
		skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
H
Herbert Xu 已提交
1167
		skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1168
	}
1169

L
Linus Torvalds 已提交
1170 1171 1172 1173

	while (size > 0) {
		int i;

H
Herbert Xu 已提交
1174
		if (skb_is_gso(skb))
1175 1176 1177 1178 1179 1180 1181 1182
			len = size;
		else {

			/* 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
1183 1184 1185 1186 1187
		if (len <= 0) {
			struct sk_buff *skb_prev;
			int alloclen;

			skb_prev = skb;
1188
			fraggap = skb_prev->len - maxfraglen;
L
Linus Torvalds 已提交
1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206

			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.
			 */
1207
			skb_put(skb, fragheaderlen + fraggap);
1208
			skb_reset_network_header(skb);
1209 1210
			skb->transport_header = (skb->network_header +
						 fragheaderlen);
L
Linus Torvalds 已提交
1211
			if (fraggap) {
1212 1213
				skb->csum = skb_copy_and_csum_bits(skb_prev,
								   maxfraglen,
1214
						    skb_transport_header(skb),
1215
								   fraggap, 0);
L
Linus Torvalds 已提交
1216 1217
				skb_prev->csum = csum_sub(skb_prev->csum,
							  skb->csum);
1218
				pskb_trim_unique(skb_prev, maxfraglen);
L
Linus Torvalds 已提交
1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231
			}

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

		i = skb_shinfo(skb)->nr_frags;
		if (len > size)
			len = size;
		if (skb_can_coalesce(skb, i, page, offset)) {
E
Eric Dumazet 已提交
1232
			skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
L
Linus Torvalds 已提交
1233 1234 1235 1236 1237 1238 1239 1240 1241
		} else if (i < MAX_SKB_FRAGS) {
			get_page(page);
			skb_fill_page_desc(skb, i, page, offset, len);
		} else {
			err = -EMSGSIZE;
			goto error;
		}

		if (skb->ip_summed == CHECKSUM_NONE) {
1242
			__wsum csum;
L
Linus Torvalds 已提交
1243 1244 1245 1246 1247 1248
			csum = csum_page(page, offset, len);
			skb->csum = csum_block_add(skb->csum, csum, skb->len);
		}

		skb->len += len;
		skb->data_len += len;
1249 1250
		skb->truesize += len;
		atomic_add(len, &sk->sk_wmem_alloc);
L
Linus Torvalds 已提交
1251 1252 1253 1254 1255 1256
		offset += len;
		size -= len;
	}
	return 0;

error:
1257
	cork->length -= size;
P
Pavel Emelyanov 已提交
1258
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
L
Linus Torvalds 已提交
1259 1260 1261
	return err;
}

1262
static void ip_cork_release(struct inet_cork *cork)
1263
{
1264 1265 1266 1267 1268
	cork->flags &= ~IPCORK_OPT;
	kfree(cork->opt);
	cork->opt = NULL;
	dst_release(cork->dst);
	cork->dst = NULL;
1269 1270
}

L
Linus Torvalds 已提交
1271 1272 1273 1274
/*
 *	Combined all pending IP fragments on the socket as one IP datagram
 *	and push them out.
 */
1275
struct sk_buff *__ip_make_skb(struct sock *sk,
1276
			      struct flowi4 *fl4,
1277 1278
			      struct sk_buff_head *queue,
			      struct inet_cork *cork)
L
Linus Torvalds 已提交
1279 1280 1281 1282
{
	struct sk_buff *skb, *tmp_skb;
	struct sk_buff **tail_skb;
	struct inet_sock *inet = inet_sk(sk);
1283
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
1284
	struct ip_options *opt = NULL;
1285
	struct rtable *rt = (struct rtable *)cork->dst;
L
Linus Torvalds 已提交
1286
	struct iphdr *iph;
1287
	__be16 df = 0;
L
Linus Torvalds 已提交
1288 1289
	__u8 ttl;

1290
	if ((skb = __skb_dequeue(queue)) == NULL)
L
Linus Torvalds 已提交
1291 1292 1293 1294
		goto out;
	tail_skb = &(skb_shinfo(skb)->frag_list);

	/* move skb->data to ip header from ext header */
1295
	if (skb->data < skb_network_header(skb))
1296
		__skb_pull(skb, skb_network_offset(skb));
1297
	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1298
		__skb_pull(tmp_skb, skb_network_header_len(skb));
L
Linus Torvalds 已提交
1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311
		*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.
	 */
J
John Heffner 已提交
1312
	if (inet->pmtudisc < IP_PMTUDISC_DO)
L
Linus Torvalds 已提交
1313 1314 1315 1316 1317
		skb->local_df = 1;

	/* DF bit is set when we want to see DF on outgoing frames.
	 * If local_df is set too, we still allow to fragment this frame
	 * locally. */
J
John Heffner 已提交
1318
	if (inet->pmtudisc >= IP_PMTUDISC_DO ||
1319 1320
	    (skb->len <= dst_mtu(&rt->dst) &&
	     ip_dont_fragment(sk, &rt->dst)))
L
Linus Torvalds 已提交
1321 1322
		df = htons(IP_DF);

1323 1324
	if (cork->flags & IPCORK_OPT)
		opt = cork->opt;
L
Linus Torvalds 已提交
1325

1326 1327 1328
	if (cork->ttl != 0)
		ttl = cork->ttl;
	else if (rt->rt_type == RTN_MULTICAST)
L
Linus Torvalds 已提交
1329 1330
		ttl = inet->mc_ttl;
	else
1331
		ttl = ip_select_ttl(inet, &rt->dst);
L
Linus Torvalds 已提交
1332

1333
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
1334 1335
	iph->version = 4;
	iph->ihl = 5;
1336
	iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
L
Linus Torvalds 已提交
1337 1338 1339
	iph->frag_off = df;
	iph->ttl = ttl;
	iph->protocol = sk->sk_protocol;
1340
	ip_copy_addrs(iph, fl4);
1341
	ip_select_ident(skb, &rt->dst, sk);
L
Linus Torvalds 已提交
1342

1343 1344 1345 1346 1347
	if (opt) {
		iph->ihl += opt->optlen>>2;
		ip_options_build(skb, opt, cork->addr, rt, 0);
	}

1348
	skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1349
	skb->mark = sk->sk_mark;
1350 1351 1352 1353
	/*
	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
	 * on dst refcount
	 */
1354
	cork->dst = NULL;
1355
	skb_dst_set(skb, &rt->dst);
L
Linus Torvalds 已提交
1356

1357
	if (iph->protocol == IPPROTO_ICMP)
1358
		icmp_out_count(net, ((struct icmphdr *)
1359 1360
			skb_transport_header(skb))->type);

1361 1362 1363 1364 1365
	ip_cork_release(cork);
out:
	return skb;
}

E
Eric Dumazet 已提交
1366
int ip_send_skb(struct net *net, struct sk_buff *skb)
1367 1368 1369
{
	int err;

H
Herbert Xu 已提交
1370
	err = ip_local_out(skb);
L
Linus Torvalds 已提交
1371 1372
	if (err) {
		if (err > 0)
E
Eric Dumazet 已提交
1373
			err = net_xmit_errno(err);
L
Linus Torvalds 已提交
1374
		if (err)
1375
			IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
L
Linus Torvalds 已提交
1376 1377 1378 1379 1380
	}

	return err;
}

1381
int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1382
{
1383 1384
	struct sk_buff *skb;

1385
	skb = ip_finish_skb(sk, fl4);
1386 1387 1388 1389
	if (!skb)
		return 0;

	/* Netfilter gets whole the not fragmented skb. */
E
Eric Dumazet 已提交
1390
	return ip_send_skb(sock_net(sk), skb);
1391 1392
}

L
Linus Torvalds 已提交
1393 1394 1395
/*
 *	Throw away all pending data on the socket.
 */
1396 1397 1398
static void __ip_flush_pending_frames(struct sock *sk,
				      struct sk_buff_head *queue,
				      struct inet_cork *cork)
L
Linus Torvalds 已提交
1399 1400 1401
{
	struct sk_buff *skb;

1402
	while ((skb = __skb_dequeue_tail(queue)) != NULL)
L
Linus Torvalds 已提交
1403 1404
		kfree_skb(skb);

1405 1406 1407 1408 1409
	ip_cork_release(cork);
}

void ip_flush_pending_frames(struct sock *sk)
{
1410
	__ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
L
Linus Torvalds 已提交
1411 1412
}

1413
struct sk_buff *ip_make_skb(struct sock *sk,
1414
			    struct flowi4 *fl4,
1415 1416 1417 1418 1419 1420
			    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)
{
1421
	struct inet_cork cork;
1422 1423 1424 1425 1426 1427 1428 1429
	struct sk_buff_head queue;
	int err;

	if (flags & MSG_PROBE)
		return NULL;

	__skb_queue_head_init(&queue);

1430 1431
	cork.flags = 0;
	cork.addr = 0;
1432
	cork.opt = NULL;
1433 1434 1435 1436
	err = ip_setup_cork(sk, &cork, ipc, rtp);
	if (err)
		return ERR_PTR(err);

1437 1438
	err = __ip_append_data(sk, fl4, &queue, &cork,
			       &current->task_frag, getfrag,
1439 1440 1441 1442 1443 1444
			       from, length, transhdrlen, flags);
	if (err) {
		__ip_flush_pending_frames(sk, &queue, &cork);
		return ERR_PTR(err);
	}

1445
	return __ip_make_skb(sk, fl4, &queue, &cork);
1446
}
L
Linus Torvalds 已提交
1447 1448 1449 1450

/*
 *	Fetch data from kernel space and fill in checksum if needed.
 */
1451
static int ip_reply_glue_bits(void *dptr, char *to, int offset,
L
Linus Torvalds 已提交
1452 1453
			      int len, int odd, struct sk_buff *skb)
{
1454
	__wsum csum;
L
Linus Torvalds 已提交
1455 1456 1457

	csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
	skb->csum = csum_block_add(skb->csum, csum, odd);
1458
	return 0;
L
Linus Torvalds 已提交
1459 1460
}

1461
/*
L
Linus Torvalds 已提交
1462
 *	Generic function to send a packet as reply to another packet.
1463
 *	Used to send some TCP resets/acks so far.
L
Linus Torvalds 已提交
1464
 *
1465
 *	Use a fake percpu inet socket to avoid false sharing and contention.
L
Linus Torvalds 已提交
1466
 */
1467 1468 1469 1470 1471 1472 1473 1474 1475
static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
	.sk = {
		.__sk_common = {
			.skc_refcnt = ATOMIC_INIT(1),
		},
		.sk_wmem_alloc	= ATOMIC_INIT(1),
		.sk_allocation	= GFP_ATOMIC,
		.sk_flags	= (1UL << SOCK_USE_WRITE_QUEUE),
	},
1476 1477
	.pmtudisc	= IP_PMTUDISC_WANT,
	.uc_ttl		= -1,
1478 1479 1480
};

void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
1481 1482
			   __be32 saddr, const struct ip_reply_arg *arg,
			   unsigned int len)
L
Linus Torvalds 已提交
1483
{
1484
	struct ip_options_data replyopts;
L
Linus Torvalds 已提交
1485
	struct ipcm_cookie ipc;
1486
	struct flowi4 fl4;
E
Eric Dumazet 已提交
1487
	struct rtable *rt = skb_rtable(skb);
1488 1489 1490
	struct sk_buff *nskb;
	struct sock *sk;
	struct inet_sock *inet;
L
Linus Torvalds 已提交
1491

1492
	if (ip_options_echo(&replyopts.opt.opt, skb))
L
Linus Torvalds 已提交
1493 1494
		return;

1495
	ipc.addr = daddr;
L
Linus Torvalds 已提交
1496
	ipc.opt = NULL;
1497
	ipc.tx_flags = 0;
1498 1499
	ipc.ttl = 0;
	ipc.tos = -1;
L
Linus Torvalds 已提交
1500

1501
	if (replyopts.opt.opt.optlen) {
L
Linus Torvalds 已提交
1502 1503
		ipc.opt = &replyopts.opt;

1504 1505
		if (replyopts.opt.opt.srr)
			daddr = replyopts.opt.opt.faddr;
L
Linus Torvalds 已提交
1506 1507
	}

1508
	flowi4_init_output(&fl4, arg->bound_dev_if, 0,
1509
			   RT_TOS(arg->tos),
1510
			   RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1511
			   ip_reply_arg_flowi_flags(arg),
1512
			   daddr, saddr,
1513 1514
			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1515
	rt = ip_route_output_key(net, &fl4);
1516 1517
	if (IS_ERR(rt))
		return;
L
Linus Torvalds 已提交
1518

1519
	inet = &get_cpu_var(unicast_sock);
L
Linus Torvalds 已提交
1520

1521
	inet->tos = arg->tos;
1522
	sk = &inet->sk;
L
Linus Torvalds 已提交
1523
	sk->sk_priority = skb->priority;
1524
	sk->sk_protocol = ip_hdr(skb)->protocol;
1525
	sk->sk_bound_dev_if = arg->bound_dev_if;
1526 1527 1528
	sock_net_set(sk, net);
	__skb_queue_head_init(&sk->sk_write_queue);
	sk->sk_sndbuf = sysctl_wmem_default;
1529
	ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1530
		       &ipc, &rt, MSG_DONTWAIT);
1531 1532
	nskb = skb_peek(&sk->sk_write_queue);
	if (nskb) {
L
Linus Torvalds 已提交
1533
		if (arg->csumoffset >= 0)
1534 1535
			*((__sum16 *)skb_transport_header(nskb) +
			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1536
								arg->csum));
1537
		nskb->ip_summed = CHECKSUM_NONE;
1538
		skb_orphan(nskb);
1539
		skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
1540
		ip_push_pending_frames(sk, &fl4);
L
Linus Torvalds 已提交
1541 1542
	}

1543
	put_cpu_var(unicast_sock);
L
Linus Torvalds 已提交
1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556

	ip_rt_put(rt);
}

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

#if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
	igmp_mc_proc_init();
#endif
}