Linux 2.6.19-rc6
[cris-mirror.git] / net / ipv4 / ip_output.c
blobfc195a44fc2e1800d0ba5e32fedc9019d29562f2
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The Internet Protocol (IP) output module.
8 * Version: $Id: ip_output.c,v 1.100 2002/02/01 22:01:03 davem Exp $
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Donald Becker, <becker@super.org>
13 * Alan Cox, <Alan.Cox@linux.org>
14 * Richard Underwood
15 * Stefan Becker, <stefanb@yello.ping.de>
16 * Jorge Cwik, <jorge@laser.satlink.net>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18 * Hirokazu Takahashi, <taka@valinux.co.jp>
20 * See ip_input.c for original log
22 * Fixes:
23 * Alan Cox : Missing nonblock feature in ip_build_xmit.
24 * Mike Kilburn : htons() missing in ip_build_xmit.
25 * Bradford Johnson: Fix faulty handling of some frames when
26 * no route is found.
27 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
28 * (in case if packet not accepted by
29 * output firewall rules)
30 * Mike McLagan : Routing by source
31 * Alexey Kuznetsov: use new route cache
32 * Andi Kleen: Fix broken PMTU recovery and remove
33 * some redundant tests.
34 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
35 * Andi Kleen : Replace ip_reply with ip_send_reply.
36 * Andi Kleen : Split fast and slow ip_build_xmit path
37 * for decreased register pressure on x86
38 * and more readibility.
39 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
40 * silently drop skb instead of failing with -EPERM.
41 * Detlev Wengorz : Copy protocol for fragments.
42 * Hirokazu Takahashi: HW checksumming for outgoing UDP
43 * datagrams.
44 * Hirokazu Takahashi: sendfile() on UDP works now.
47 #include <asm/uaccess.h>
48 #include <asm/system.h>
49 #include <linux/module.h>
50 #include <linux/types.h>
51 #include <linux/kernel.h>
52 #include <linux/sched.h>
53 #include <linux/mm.h>
54 #include <linux/string.h>
55 #include <linux/errno.h>
57 #include <linux/socket.h>
58 #include <linux/sockios.h>
59 #include <linux/in.h>
60 #include <linux/inet.h>
61 #include <linux/netdevice.h>
62 #include <linux/etherdevice.h>
63 #include <linux/proc_fs.h>
64 #include <linux/stat.h>
65 #include <linux/init.h>
67 #include <net/snmp.h>
68 #include <net/ip.h>
69 #include <net/protocol.h>
70 #include <net/route.h>
71 #include <net/xfrm.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <net/arp.h>
75 #include <net/icmp.h>
76 #include <net/checksum.h>
77 #include <net/inetpeer.h>
78 #include <net/checksum.h>
79 #include <linux/igmp.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/netfilter_bridge.h>
82 #include <linux/mroute.h>
83 #include <linux/netlink.h>
84 #include <linux/tcp.h>
86 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
88 /* Generate a checksum for an outgoing IP datagram. */
89 __inline__ void ip_send_check(struct iphdr *iph)
91 iph->check = 0;
92 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
95 /* dev_loopback_xmit for use with netfilter. */
96 static int ip_dev_loopback_xmit(struct sk_buff *newskb)
98 newskb->mac.raw = newskb->data;
99 __skb_pull(newskb, newskb->nh.raw - newskb->data);
100 newskb->pkt_type = PACKET_LOOPBACK;
101 newskb->ip_summed = CHECKSUM_UNNECESSARY;
102 BUG_TRAP(newskb->dst);
103 netif_rx(newskb);
104 return 0;
107 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
109 int ttl = inet->uc_ttl;
111 if (ttl < 0)
112 ttl = dst_metric(dst, RTAX_HOPLIMIT);
113 return ttl;
117 * Add an ip header to a skbuff and send it out.
120 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
121 __be32 saddr, __be32 daddr, struct ip_options *opt)
123 struct inet_sock *inet = inet_sk(sk);
124 struct rtable *rt = (struct rtable *)skb->dst;
125 struct iphdr *iph;
127 /* Build the IP header. */
128 if (opt)
129 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr) + opt->optlen);
130 else
131 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr));
133 iph->version = 4;
134 iph->ihl = 5;
135 iph->tos = inet->tos;
136 if (ip_dont_fragment(sk, &rt->u.dst))
137 iph->frag_off = htons(IP_DF);
138 else
139 iph->frag_off = 0;
140 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
141 iph->daddr = rt->rt_dst;
142 iph->saddr = rt->rt_src;
143 iph->protocol = sk->sk_protocol;
144 iph->tot_len = htons(skb->len);
145 ip_select_ident(iph, &rt->u.dst, sk);
146 skb->nh.iph = iph;
148 if (opt && opt->optlen) {
149 iph->ihl += opt->optlen>>2;
150 ip_options_build(skb, opt, daddr, rt, 0);
152 ip_send_check(iph);
154 skb->priority = sk->sk_priority;
156 /* Send it out. */
157 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
158 dst_output);
161 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
163 static inline int ip_finish_output2(struct sk_buff *skb)
165 struct dst_entry *dst = skb->dst;
166 struct hh_cache *hh = dst->hh;
167 struct net_device *dev = dst->dev;
168 int hh_len = LL_RESERVED_SPACE(dev);
170 /* Be paranoid, rather than too clever. */
171 if (unlikely(skb_headroom(skb) < hh_len && dev->hard_header)) {
172 struct sk_buff *skb2;
174 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
175 if (skb2 == NULL) {
176 kfree_skb(skb);
177 return -ENOMEM;
179 if (skb->sk)
180 skb_set_owner_w(skb2, skb->sk);
181 kfree_skb(skb);
182 skb = skb2;
185 if (hh) {
186 int hh_alen;
188 read_lock_bh(&hh->hh_lock);
189 hh_alen = HH_DATA_ALIGN(hh->hh_len);
190 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
191 read_unlock_bh(&hh->hh_lock);
192 skb_push(skb, hh->hh_len);
193 return hh->hh_output(skb);
194 } else if (dst->neighbour)
195 return dst->neighbour->output(skb);
197 if (net_ratelimit())
198 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
199 kfree_skb(skb);
200 return -EINVAL;
203 static inline int ip_finish_output(struct sk_buff *skb)
205 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
206 /* Policy lookup after SNAT yielded a new policy */
207 if (skb->dst->xfrm != NULL) {
208 IPCB(skb)->flags |= IPSKB_REROUTED;
209 return dst_output(skb);
211 #endif
212 if (skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb))
213 return ip_fragment(skb, ip_finish_output2);
214 else
215 return ip_finish_output2(skb);
218 int ip_mc_output(struct sk_buff *skb)
220 struct sock *sk = skb->sk;
221 struct rtable *rt = (struct rtable*)skb->dst;
222 struct net_device *dev = rt->u.dst.dev;
225 * If the indicated interface is up and running, send the packet.
227 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
229 skb->dev = dev;
230 skb->protocol = htons(ETH_P_IP);
233 * Multicasts are looped back for other local users
236 if (rt->rt_flags&RTCF_MULTICAST) {
237 if ((!sk || inet_sk(sk)->mc_loop)
238 #ifdef CONFIG_IP_MROUTE
239 /* Small optimization: do not loopback not local frames,
240 which returned after forwarding; they will be dropped
241 by ip_mr_input in any case.
242 Note, that local frames are looped back to be delivered
243 to local recipients.
245 This check is duplicated in ip_mr_input at the moment.
247 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
248 #endif
250 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
251 if (newskb)
252 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
253 newskb->dev,
254 ip_dev_loopback_xmit);
257 /* Multicasts with ttl 0 must not go beyond the host */
259 if (skb->nh.iph->ttl == 0) {
260 kfree_skb(skb);
261 return 0;
265 if (rt->rt_flags&RTCF_BROADCAST) {
266 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
267 if (newskb)
268 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
269 newskb->dev, ip_dev_loopback_xmit);
272 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev,
273 ip_finish_output,
274 !(IPCB(skb)->flags & IPSKB_REROUTED));
277 int ip_output(struct sk_buff *skb)
279 struct net_device *dev = skb->dst->dev;
281 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
283 skb->dev = dev;
284 skb->protocol = htons(ETH_P_IP);
286 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
287 ip_finish_output,
288 !(IPCB(skb)->flags & IPSKB_REROUTED));
291 int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
293 struct sock *sk = skb->sk;
294 struct inet_sock *inet = inet_sk(sk);
295 struct ip_options *opt = inet->opt;
296 struct rtable *rt;
297 struct iphdr *iph;
299 /* Skip all of this if the packet is already routed,
300 * f.e. by something like SCTP.
302 rt = (struct rtable *) skb->dst;
303 if (rt != NULL)
304 goto packet_routed;
306 /* Make sure we can route this packet. */
307 rt = (struct rtable *)__sk_dst_check(sk, 0);
308 if (rt == NULL) {
309 __be32 daddr;
311 /* Use correct destination address if we have options. */
312 daddr = inet->daddr;
313 if(opt && opt->srr)
314 daddr = opt->faddr;
317 struct flowi fl = { .oif = sk->sk_bound_dev_if,
318 .nl_u = { .ip4_u =
319 { .daddr = daddr,
320 .saddr = inet->saddr,
321 .tos = RT_CONN_FLAGS(sk) } },
322 .proto = sk->sk_protocol,
323 .uli_u = { .ports =
324 { .sport = inet->sport,
325 .dport = inet->dport } } };
327 /* If this fails, retransmit mechanism of transport layer will
328 * keep trying until route appears or the connection times
329 * itself out.
331 security_sk_classify_flow(sk, &fl);
332 if (ip_route_output_flow(&rt, &fl, sk, 0))
333 goto no_route;
335 sk_setup_caps(sk, &rt->u.dst);
337 skb->dst = dst_clone(&rt->u.dst);
339 packet_routed:
340 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
341 goto no_route;
343 /* OK, we know where to send it, allocate and build IP header. */
344 iph = (struct iphdr *) skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
345 *((__u16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
346 iph->tot_len = htons(skb->len);
347 if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
348 iph->frag_off = htons(IP_DF);
349 else
350 iph->frag_off = 0;
351 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
352 iph->protocol = sk->sk_protocol;
353 iph->saddr = rt->rt_src;
354 iph->daddr = rt->rt_dst;
355 skb->nh.iph = iph;
356 /* Transport layer set skb->h.foo itself. */
358 if (opt && opt->optlen) {
359 iph->ihl += opt->optlen >> 2;
360 ip_options_build(skb, opt, inet->daddr, rt, 0);
363 ip_select_ident_more(iph, &rt->u.dst, sk,
364 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
366 /* Add an IP checksum. */
367 ip_send_check(iph);
369 skb->priority = sk->sk_priority;
371 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
372 dst_output);
374 no_route:
375 IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
376 kfree_skb(skb);
377 return -EHOSTUNREACH;
381 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
383 to->pkt_type = from->pkt_type;
384 to->priority = from->priority;
385 to->protocol = from->protocol;
386 dst_release(to->dst);
387 to->dst = dst_clone(from->dst);
388 to->dev = from->dev;
390 /* Copy the flags to each fragment. */
391 IPCB(to)->flags = IPCB(from)->flags;
393 #ifdef CONFIG_NET_SCHED
394 to->tc_index = from->tc_index;
395 #endif
396 #ifdef CONFIG_NETFILTER
397 to->nfmark = from->nfmark;
398 /* Connection association is same as pre-frag packet */
399 nf_conntrack_put(to->nfct);
400 to->nfct = from->nfct;
401 nf_conntrack_get(to->nfct);
402 to->nfctinfo = from->nfctinfo;
403 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
404 to->ipvs_property = from->ipvs_property;
405 #endif
406 #ifdef CONFIG_BRIDGE_NETFILTER
407 nf_bridge_put(to->nf_bridge);
408 to->nf_bridge = from->nf_bridge;
409 nf_bridge_get(to->nf_bridge);
410 #endif
411 #endif
412 skb_copy_secmark(to, from);
416 * This IP datagram is too large to be sent in one piece. Break it up into
417 * smaller pieces (each of size equal to IP header plus
418 * a block of the data of the original IP data part) that will yet fit in a
419 * single device frame, and queue such a frame for sending.
422 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
424 struct iphdr *iph;
425 int raw = 0;
426 int ptr;
427 struct net_device *dev;
428 struct sk_buff *skb2;
429 unsigned int mtu, hlen, left, len, ll_rs, pad;
430 int offset;
431 __be16 not_last_frag;
432 struct rtable *rt = (struct rtable*)skb->dst;
433 int err = 0;
435 dev = rt->u.dst.dev;
438 * Point into the IP datagram header.
441 iph = skb->nh.iph;
443 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
444 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
445 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
446 htonl(dst_mtu(&rt->u.dst)));
447 kfree_skb(skb);
448 return -EMSGSIZE;
452 * Setup starting values.
455 hlen = iph->ihl * 4;
456 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
457 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
459 /* When frag_list is given, use it. First, check its validity:
460 * some transformers could create wrong frag_list or break existing
461 * one, it is not prohibited. In this case fall back to copying.
463 * LATER: this step can be merged to real generation of fragments,
464 * we can switch to copy when see the first bad fragment.
466 if (skb_shinfo(skb)->frag_list) {
467 struct sk_buff *frag;
468 int first_len = skb_pagelen(skb);
470 if (first_len - hlen > mtu ||
471 ((first_len - hlen) & 7) ||
472 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
473 skb_cloned(skb))
474 goto slow_path;
476 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
477 /* Correct geometry. */
478 if (frag->len > mtu ||
479 ((frag->len & 7) && frag->next) ||
480 skb_headroom(frag) < hlen)
481 goto slow_path;
483 /* Partially cloned skb? */
484 if (skb_shared(frag))
485 goto slow_path;
487 BUG_ON(frag->sk);
488 if (skb->sk) {
489 sock_hold(skb->sk);
490 frag->sk = skb->sk;
491 frag->destructor = sock_wfree;
492 skb->truesize -= frag->truesize;
496 /* Everything is OK. Generate! */
498 err = 0;
499 offset = 0;
500 frag = skb_shinfo(skb)->frag_list;
501 skb_shinfo(skb)->frag_list = NULL;
502 skb->data_len = first_len - skb_headlen(skb);
503 skb->len = first_len;
504 iph->tot_len = htons(first_len);
505 iph->frag_off = htons(IP_MF);
506 ip_send_check(iph);
508 for (;;) {
509 /* Prepare header of the next frame,
510 * before previous one went down. */
511 if (frag) {
512 frag->ip_summed = CHECKSUM_NONE;
513 frag->h.raw = frag->data;
514 frag->nh.raw = __skb_push(frag, hlen);
515 memcpy(frag->nh.raw, iph, hlen);
516 iph = frag->nh.iph;
517 iph->tot_len = htons(frag->len);
518 ip_copy_metadata(frag, skb);
519 if (offset == 0)
520 ip_options_fragment(frag);
521 offset += skb->len - hlen;
522 iph->frag_off = htons(offset>>3);
523 if (frag->next != NULL)
524 iph->frag_off |= htons(IP_MF);
525 /* Ready, complete checksum */
526 ip_send_check(iph);
529 err = output(skb);
531 if (!err)
532 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
533 if (err || !frag)
534 break;
536 skb = frag;
537 frag = skb->next;
538 skb->next = NULL;
541 if (err == 0) {
542 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
543 return 0;
546 while (frag) {
547 skb = frag->next;
548 kfree_skb(frag);
549 frag = skb;
551 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
552 return err;
555 slow_path:
556 left = skb->len - hlen; /* Space per frame */
557 ptr = raw + hlen; /* Where to start from */
559 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
560 * we need to make room for the encapsulating header
562 pad = nf_bridge_pad(skb);
563 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
564 mtu -= pad;
567 * Fragment the datagram.
570 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
571 not_last_frag = iph->frag_off & htons(IP_MF);
574 * Keep copying data until we run out.
577 while(left > 0) {
578 len = left;
579 /* IF: it doesn't fit, use 'mtu' - the data space left */
580 if (len > mtu)
581 len = mtu;
582 /* IF: we are not sending upto and including the packet end
583 then align the next start on an eight byte boundary */
584 if (len < left) {
585 len &= ~7;
588 * Allocate buffer.
591 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
592 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
593 err = -ENOMEM;
594 goto fail;
598 * Set up data on packet
601 ip_copy_metadata(skb2, skb);
602 skb_reserve(skb2, ll_rs);
603 skb_put(skb2, len + hlen);
604 skb2->nh.raw = skb2->data;
605 skb2->h.raw = skb2->data + hlen;
608 * Charge the memory for the fragment to any owner
609 * it might possess
612 if (skb->sk)
613 skb_set_owner_w(skb2, skb->sk);
616 * Copy the packet header into the new buffer.
619 memcpy(skb2->nh.raw, skb->data, hlen);
622 * Copy a block of the IP datagram.
624 if (skb_copy_bits(skb, ptr, skb2->h.raw, len))
625 BUG();
626 left -= len;
629 * Fill in the new header fields.
631 iph = skb2->nh.iph;
632 iph->frag_off = htons((offset >> 3));
634 /* ANK: dirty, but effective trick. Upgrade options only if
635 * the segment to be fragmented was THE FIRST (otherwise,
636 * options are already fixed) and make it ONCE
637 * on the initial skb, so that all the following fragments
638 * will inherit fixed options.
640 if (offset == 0)
641 ip_options_fragment(skb);
644 * Added AC : If we are fragmenting a fragment that's not the
645 * last fragment then keep MF on each bit
647 if (left > 0 || not_last_frag)
648 iph->frag_off |= htons(IP_MF);
649 ptr += len;
650 offset += len;
653 * Put this fragment into the sending queue.
655 iph->tot_len = htons(len + hlen);
657 ip_send_check(iph);
659 err = output(skb2);
660 if (err)
661 goto fail;
663 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
665 kfree_skb(skb);
666 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
667 return err;
669 fail:
670 kfree_skb(skb);
671 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
672 return err;
675 EXPORT_SYMBOL(ip_fragment);
678 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
680 struct iovec *iov = from;
682 if (skb->ip_summed == CHECKSUM_PARTIAL) {
683 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
684 return -EFAULT;
685 } else {
686 unsigned int csum = 0;
687 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
688 return -EFAULT;
689 skb->csum = csum_block_add(skb->csum, csum, odd);
691 return 0;
694 static inline unsigned int
695 csum_page(struct page *page, int offset, int copy)
697 char *kaddr;
698 unsigned int csum;
699 kaddr = kmap(page);
700 csum = csum_partial(kaddr + offset, copy, 0);
701 kunmap(page);
702 return csum;
705 static inline int ip_ufo_append_data(struct sock *sk,
706 int getfrag(void *from, char *to, int offset, int len,
707 int odd, struct sk_buff *skb),
708 void *from, int length, int hh_len, int fragheaderlen,
709 int transhdrlen, int mtu,unsigned int flags)
711 struct sk_buff *skb;
712 int err;
714 /* There is support for UDP fragmentation offload by network
715 * device, so create one single skb packet containing complete
716 * udp datagram
718 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
719 skb = sock_alloc_send_skb(sk,
720 hh_len + fragheaderlen + transhdrlen + 20,
721 (flags & MSG_DONTWAIT), &err);
723 if (skb == NULL)
724 return err;
726 /* reserve space for Hardware header */
727 skb_reserve(skb, hh_len);
729 /* create space for UDP/IP header */
730 skb_put(skb,fragheaderlen + transhdrlen);
732 /* initialize network header pointer */
733 skb->nh.raw = skb->data;
735 /* initialize protocol header pointer */
736 skb->h.raw = skb->data + fragheaderlen;
738 skb->ip_summed = CHECKSUM_PARTIAL;
739 skb->csum = 0;
740 sk->sk_sndmsg_off = 0;
743 err = skb_append_datato_frags(sk,skb, getfrag, from,
744 (length - transhdrlen));
745 if (!err) {
746 /* specify the length of each IP datagram fragment*/
747 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
748 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
749 __skb_queue_tail(&sk->sk_write_queue, skb);
751 return 0;
753 /* There is not enough support do UFO ,
754 * so follow normal path
756 kfree_skb(skb);
757 return err;
761 * ip_append_data() and ip_append_page() can make one large IP datagram
762 * from many pieces of data. Each pieces will be holded on the socket
763 * until ip_push_pending_frames() is called. Each piece can be a page
764 * or non-page data.
766 * Not only UDP, other transport protocols - e.g. raw sockets - can use
767 * this interface potentially.
769 * LATER: length must be adjusted by pad at tail, when it is required.
771 int ip_append_data(struct sock *sk,
772 int getfrag(void *from, char *to, int offset, int len,
773 int odd, struct sk_buff *skb),
774 void *from, int length, int transhdrlen,
775 struct ipcm_cookie *ipc, struct rtable *rt,
776 unsigned int flags)
778 struct inet_sock *inet = inet_sk(sk);
779 struct sk_buff *skb;
781 struct ip_options *opt = NULL;
782 int hh_len;
783 int exthdrlen;
784 int mtu;
785 int copy;
786 int err;
787 int offset = 0;
788 unsigned int maxfraglen, fragheaderlen;
789 int csummode = CHECKSUM_NONE;
791 if (flags&MSG_PROBE)
792 return 0;
794 if (skb_queue_empty(&sk->sk_write_queue)) {
796 * setup for corking.
798 opt = ipc->opt;
799 if (opt) {
800 if (inet->cork.opt == NULL) {
801 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
802 if (unlikely(inet->cork.opt == NULL))
803 return -ENOBUFS;
805 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
806 inet->cork.flags |= IPCORK_OPT;
807 inet->cork.addr = ipc->addr;
809 dst_hold(&rt->u.dst);
810 inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path);
811 inet->cork.rt = rt;
812 inet->cork.length = 0;
813 sk->sk_sndmsg_page = NULL;
814 sk->sk_sndmsg_off = 0;
815 if ((exthdrlen = rt->u.dst.header_len) != 0) {
816 length += exthdrlen;
817 transhdrlen += exthdrlen;
819 } else {
820 rt = inet->cork.rt;
821 if (inet->cork.flags & IPCORK_OPT)
822 opt = inet->cork.opt;
824 transhdrlen = 0;
825 exthdrlen = 0;
826 mtu = inet->cork.fragsize;
828 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
830 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
831 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
833 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
834 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
835 return -EMSGSIZE;
839 * transhdrlen > 0 means that this is the first fragment and we wish
840 * it won't be fragmented in the future.
842 if (transhdrlen &&
843 length + fragheaderlen <= mtu &&
844 rt->u.dst.dev->features & NETIF_F_ALL_CSUM &&
845 !exthdrlen)
846 csummode = CHECKSUM_PARTIAL;
848 inet->cork.length += length;
849 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
850 (rt->u.dst.dev->features & NETIF_F_UFO)) {
852 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
853 fragheaderlen, transhdrlen, mtu,
854 flags);
855 if (err)
856 goto error;
857 return 0;
860 /* So, what's going on in the loop below?
862 * We use calculated fragment length to generate chained skb,
863 * each of segments is IP fragment ready for sending to network after
864 * adding appropriate IP header.
867 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
868 goto alloc_new_skb;
870 while (length > 0) {
871 /* Check if the remaining data fits into current packet. */
872 copy = mtu - skb->len;
873 if (copy < length)
874 copy = maxfraglen - skb->len;
875 if (copy <= 0) {
876 char *data;
877 unsigned int datalen;
878 unsigned int fraglen;
879 unsigned int fraggap;
880 unsigned int alloclen;
881 struct sk_buff *skb_prev;
882 alloc_new_skb:
883 skb_prev = skb;
884 if (skb_prev)
885 fraggap = skb_prev->len - maxfraglen;
886 else
887 fraggap = 0;
890 * If remaining data exceeds the mtu,
891 * we know we need more fragment(s).
893 datalen = length + fraggap;
894 if (datalen > mtu - fragheaderlen)
895 datalen = maxfraglen - fragheaderlen;
896 fraglen = datalen + fragheaderlen;
898 if ((flags & MSG_MORE) &&
899 !(rt->u.dst.dev->features&NETIF_F_SG))
900 alloclen = mtu;
901 else
902 alloclen = datalen + fragheaderlen;
904 /* The last fragment gets additional space at tail.
905 * Note, with MSG_MORE we overallocate on fragments,
906 * because we have no idea what fragment will be
907 * the last.
909 if (datalen == length + fraggap)
910 alloclen += rt->u.dst.trailer_len;
912 if (transhdrlen) {
913 skb = sock_alloc_send_skb(sk,
914 alloclen + hh_len + 15,
915 (flags & MSG_DONTWAIT), &err);
916 } else {
917 skb = NULL;
918 if (atomic_read(&sk->sk_wmem_alloc) <=
919 2 * sk->sk_sndbuf)
920 skb = sock_wmalloc(sk,
921 alloclen + hh_len + 15, 1,
922 sk->sk_allocation);
923 if (unlikely(skb == NULL))
924 err = -ENOBUFS;
926 if (skb == NULL)
927 goto error;
930 * Fill in the control structures
932 skb->ip_summed = csummode;
933 skb->csum = 0;
934 skb_reserve(skb, hh_len);
937 * Find where to start putting bytes.
939 data = skb_put(skb, fraglen);
940 skb->nh.raw = data + exthdrlen;
941 data += fragheaderlen;
942 skb->h.raw = data + exthdrlen;
944 if (fraggap) {
945 skb->csum = skb_copy_and_csum_bits(
946 skb_prev, maxfraglen,
947 data + transhdrlen, fraggap, 0);
948 skb_prev->csum = csum_sub(skb_prev->csum,
949 skb->csum);
950 data += fraggap;
951 pskb_trim_unique(skb_prev, maxfraglen);
954 copy = datalen - transhdrlen - fraggap;
955 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
956 err = -EFAULT;
957 kfree_skb(skb);
958 goto error;
961 offset += copy;
962 length -= datalen - fraggap;
963 transhdrlen = 0;
964 exthdrlen = 0;
965 csummode = CHECKSUM_NONE;
968 * Put the packet on the pending queue.
970 __skb_queue_tail(&sk->sk_write_queue, skb);
971 continue;
974 if (copy > length)
975 copy = length;
977 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
978 unsigned int off;
980 off = skb->len;
981 if (getfrag(from, skb_put(skb, copy),
982 offset, copy, off, skb) < 0) {
983 __skb_trim(skb, off);
984 err = -EFAULT;
985 goto error;
987 } else {
988 int i = skb_shinfo(skb)->nr_frags;
989 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
990 struct page *page = sk->sk_sndmsg_page;
991 int off = sk->sk_sndmsg_off;
992 unsigned int left;
994 if (page && (left = PAGE_SIZE - off) > 0) {
995 if (copy >= left)
996 copy = left;
997 if (page != frag->page) {
998 if (i == MAX_SKB_FRAGS) {
999 err = -EMSGSIZE;
1000 goto error;
1002 get_page(page);
1003 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1004 frag = &skb_shinfo(skb)->frags[i];
1006 } else if (i < MAX_SKB_FRAGS) {
1007 if (copy > PAGE_SIZE)
1008 copy = PAGE_SIZE;
1009 page = alloc_pages(sk->sk_allocation, 0);
1010 if (page == NULL) {
1011 err = -ENOMEM;
1012 goto error;
1014 sk->sk_sndmsg_page = page;
1015 sk->sk_sndmsg_off = 0;
1017 skb_fill_page_desc(skb, i, page, 0, 0);
1018 frag = &skb_shinfo(skb)->frags[i];
1019 skb->truesize += PAGE_SIZE;
1020 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1021 } else {
1022 err = -EMSGSIZE;
1023 goto error;
1025 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1026 err = -EFAULT;
1027 goto error;
1029 sk->sk_sndmsg_off += copy;
1030 frag->size += copy;
1031 skb->len += copy;
1032 skb->data_len += copy;
1034 offset += copy;
1035 length -= copy;
1038 return 0;
1040 error:
1041 inet->cork.length -= length;
1042 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1043 return err;
1046 ssize_t ip_append_page(struct sock *sk, struct page *page,
1047 int offset, size_t size, int flags)
1049 struct inet_sock *inet = inet_sk(sk);
1050 struct sk_buff *skb;
1051 struct rtable *rt;
1052 struct ip_options *opt = NULL;
1053 int hh_len;
1054 int mtu;
1055 int len;
1056 int err;
1057 unsigned int maxfraglen, fragheaderlen, fraggap;
1059 if (inet->hdrincl)
1060 return -EPERM;
1062 if (flags&MSG_PROBE)
1063 return 0;
1065 if (skb_queue_empty(&sk->sk_write_queue))
1066 return -EINVAL;
1068 rt = inet->cork.rt;
1069 if (inet->cork.flags & IPCORK_OPT)
1070 opt = inet->cork.opt;
1072 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1073 return -EOPNOTSUPP;
1075 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1076 mtu = inet->cork.fragsize;
1078 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1079 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1081 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1082 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1083 return -EMSGSIZE;
1086 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1087 return -EINVAL;
1089 inet->cork.length += size;
1090 if ((sk->sk_protocol == IPPROTO_UDP) &&
1091 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1092 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1093 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1097 while (size > 0) {
1098 int i;
1100 if (skb_is_gso(skb))
1101 len = size;
1102 else {
1104 /* Check if the remaining data fits into current packet. */
1105 len = mtu - skb->len;
1106 if (len < size)
1107 len = maxfraglen - skb->len;
1109 if (len <= 0) {
1110 struct sk_buff *skb_prev;
1111 char *data;
1112 struct iphdr *iph;
1113 int alloclen;
1115 skb_prev = skb;
1116 fraggap = skb_prev->len - maxfraglen;
1118 alloclen = fragheaderlen + hh_len + fraggap + 15;
1119 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1120 if (unlikely(!skb)) {
1121 err = -ENOBUFS;
1122 goto error;
1126 * Fill in the control structures
1128 skb->ip_summed = CHECKSUM_NONE;
1129 skb->csum = 0;
1130 skb_reserve(skb, hh_len);
1133 * Find where to start putting bytes.
1135 data = skb_put(skb, fragheaderlen + fraggap);
1136 skb->nh.iph = iph = (struct iphdr *)data;
1137 data += fragheaderlen;
1138 skb->h.raw = data;
1140 if (fraggap) {
1141 skb->csum = skb_copy_and_csum_bits(
1142 skb_prev, maxfraglen,
1143 data, fraggap, 0);
1144 skb_prev->csum = csum_sub(skb_prev->csum,
1145 skb->csum);
1146 pskb_trim_unique(skb_prev, maxfraglen);
1150 * Put the packet on the pending queue.
1152 __skb_queue_tail(&sk->sk_write_queue, skb);
1153 continue;
1156 i = skb_shinfo(skb)->nr_frags;
1157 if (len > size)
1158 len = size;
1159 if (skb_can_coalesce(skb, i, page, offset)) {
1160 skb_shinfo(skb)->frags[i-1].size += len;
1161 } else if (i < MAX_SKB_FRAGS) {
1162 get_page(page);
1163 skb_fill_page_desc(skb, i, page, offset, len);
1164 } else {
1165 err = -EMSGSIZE;
1166 goto error;
1169 if (skb->ip_summed == CHECKSUM_NONE) {
1170 unsigned int csum;
1171 csum = csum_page(page, offset, len);
1172 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1175 skb->len += len;
1176 skb->data_len += len;
1177 offset += len;
1178 size -= len;
1180 return 0;
1182 error:
1183 inet->cork.length -= size;
1184 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1185 return err;
1189 * Combined all pending IP fragments on the socket as one IP datagram
1190 * and push them out.
1192 int ip_push_pending_frames(struct sock *sk)
1194 struct sk_buff *skb, *tmp_skb;
1195 struct sk_buff **tail_skb;
1196 struct inet_sock *inet = inet_sk(sk);
1197 struct ip_options *opt = NULL;
1198 struct rtable *rt = inet->cork.rt;
1199 struct iphdr *iph;
1200 __be16 df = 0;
1201 __u8 ttl;
1202 int err = 0;
1204 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1205 goto out;
1206 tail_skb = &(skb_shinfo(skb)->frag_list);
1208 /* move skb->data to ip header from ext header */
1209 if (skb->data < skb->nh.raw)
1210 __skb_pull(skb, skb->nh.raw - skb->data);
1211 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1212 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1213 *tail_skb = tmp_skb;
1214 tail_skb = &(tmp_skb->next);
1215 skb->len += tmp_skb->len;
1216 skb->data_len += tmp_skb->len;
1217 skb->truesize += tmp_skb->truesize;
1218 __sock_put(tmp_skb->sk);
1219 tmp_skb->destructor = NULL;
1220 tmp_skb->sk = NULL;
1223 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1224 * to fragment the frame generated here. No matter, what transforms
1225 * how transforms change size of the packet, it will come out.
1227 if (inet->pmtudisc != IP_PMTUDISC_DO)
1228 skb->local_df = 1;
1230 /* DF bit is set when we want to see DF on outgoing frames.
1231 * If local_df is set too, we still allow to fragment this frame
1232 * locally. */
1233 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1234 (skb->len <= dst_mtu(&rt->u.dst) &&
1235 ip_dont_fragment(sk, &rt->u.dst)))
1236 df = htons(IP_DF);
1238 if (inet->cork.flags & IPCORK_OPT)
1239 opt = inet->cork.opt;
1241 if (rt->rt_type == RTN_MULTICAST)
1242 ttl = inet->mc_ttl;
1243 else
1244 ttl = ip_select_ttl(inet, &rt->u.dst);
1246 iph = (struct iphdr *)skb->data;
1247 iph->version = 4;
1248 iph->ihl = 5;
1249 if (opt) {
1250 iph->ihl += opt->optlen>>2;
1251 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1253 iph->tos = inet->tos;
1254 iph->tot_len = htons(skb->len);
1255 iph->frag_off = df;
1256 ip_select_ident(iph, &rt->u.dst, sk);
1257 iph->ttl = ttl;
1258 iph->protocol = sk->sk_protocol;
1259 iph->saddr = rt->rt_src;
1260 iph->daddr = rt->rt_dst;
1261 ip_send_check(iph);
1263 skb->priority = sk->sk_priority;
1264 skb->dst = dst_clone(&rt->u.dst);
1266 /* Netfilter gets whole the not fragmented skb. */
1267 err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL,
1268 skb->dst->dev, dst_output);
1269 if (err) {
1270 if (err > 0)
1271 err = inet->recverr ? net_xmit_errno(err) : 0;
1272 if (err)
1273 goto error;
1276 out:
1277 inet->cork.flags &= ~IPCORK_OPT;
1278 kfree(inet->cork.opt);
1279 inet->cork.opt = NULL;
1280 if (inet->cork.rt) {
1281 ip_rt_put(inet->cork.rt);
1282 inet->cork.rt = NULL;
1284 return err;
1286 error:
1287 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1288 goto out;
1292 * Throw away all pending data on the socket.
1294 void ip_flush_pending_frames(struct sock *sk)
1296 struct inet_sock *inet = inet_sk(sk);
1297 struct sk_buff *skb;
1299 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1300 kfree_skb(skb);
1302 inet->cork.flags &= ~IPCORK_OPT;
1303 kfree(inet->cork.opt);
1304 inet->cork.opt = NULL;
1305 if (inet->cork.rt) {
1306 ip_rt_put(inet->cork.rt);
1307 inet->cork.rt = NULL;
1313 * Fetch data from kernel space and fill in checksum if needed.
1315 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1316 int len, int odd, struct sk_buff *skb)
1318 unsigned int csum;
1320 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1321 skb->csum = csum_block_add(skb->csum, csum, odd);
1322 return 0;
1326 * Generic function to send a packet as reply to another packet.
1327 * Used to send TCP resets so far. ICMP should use this function too.
1329 * Should run single threaded per socket because it uses the sock
1330 * structure to pass arguments.
1332 * LATER: switch from ip_build_xmit to ip_append_*
1334 void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1335 unsigned int len)
1337 struct inet_sock *inet = inet_sk(sk);
1338 struct {
1339 struct ip_options opt;
1340 char data[40];
1341 } replyopts;
1342 struct ipcm_cookie ipc;
1343 __be32 daddr;
1344 struct rtable *rt = (struct rtable*)skb->dst;
1346 if (ip_options_echo(&replyopts.opt, skb))
1347 return;
1349 daddr = ipc.addr = rt->rt_src;
1350 ipc.opt = NULL;
1352 if (replyopts.opt.optlen) {
1353 ipc.opt = &replyopts.opt;
1355 if (ipc.opt->srr)
1356 daddr = replyopts.opt.faddr;
1360 struct flowi fl = { .nl_u = { .ip4_u =
1361 { .daddr = daddr,
1362 .saddr = rt->rt_spec_dst,
1363 .tos = RT_TOS(skb->nh.iph->tos) } },
1364 /* Not quite clean, but right. */
1365 .uli_u = { .ports =
1366 { .sport = skb->h.th->dest,
1367 .dport = skb->h.th->source } },
1368 .proto = sk->sk_protocol };
1369 security_skb_classify_flow(skb, &fl);
1370 if (ip_route_output_key(&rt, &fl))
1371 return;
1374 /* And let IP do all the hard work.
1376 This chunk is not reenterable, hence spinlock.
1377 Note that it uses the fact, that this function is called
1378 with locally disabled BH and that sk cannot be already spinlocked.
1380 bh_lock_sock(sk);
1381 inet->tos = skb->nh.iph->tos;
1382 sk->sk_priority = skb->priority;
1383 sk->sk_protocol = skb->nh.iph->protocol;
1384 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1385 &ipc, rt, MSG_DONTWAIT);
1386 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1387 if (arg->csumoffset >= 0)
1388 *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
1389 skb->ip_summed = CHECKSUM_NONE;
1390 ip_push_pending_frames(sk);
1393 bh_unlock_sock(sk);
1395 ip_rt_put(rt);
1398 void __init ip_init(void)
1400 ip_rt_init();
1401 inet_initpeers();
1403 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1404 igmp_mc_proc_init();
1405 #endif
1408 EXPORT_SYMBOL(ip_generic_getfrag);
1409 EXPORT_SYMBOL(ip_queue_xmit);
1410 EXPORT_SYMBOL(ip_send_check);