jbd: Journal block numbers can ever be only 32-bit use unsigned int for them
[linux/fpc-iii.git] / net / ipv4 / ip_output.c
blob9fe5d7b81580fe8b8768b03e4fff05f9059d14c6
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 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <Alan.Cox@linux.org>
12 * Richard Underwood
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16 * Hirokazu Takahashi, <taka@valinux.co.jp>
18 * See ip_input.c for original log
20 * Fixes:
21 * Alan Cox : Missing nonblock feature in ip_build_xmit.
22 * Mike Kilburn : htons() missing in ip_build_xmit.
23 * Bradford Johnson: Fix faulty handling of some frames when
24 * no route is found.
25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
26 * (in case if packet not accepted by
27 * output firewall rules)
28 * Mike McLagan : Routing by source
29 * Alexey Kuznetsov: use new route cache
30 * Andi Kleen: Fix broken PMTU recovery and remove
31 * some redundant tests.
32 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
33 * Andi Kleen : Replace ip_reply with ip_send_reply.
34 * Andi Kleen : Split fast and slow ip_build_xmit path
35 * for decreased register pressure on x86
36 * and more readibility.
37 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
38 * silently drop skb instead of failing with -EPERM.
39 * Detlev Wengorz : Copy protocol for fragments.
40 * Hirokazu Takahashi: HW checksumming for outgoing UDP
41 * datagrams.
42 * Hirokazu Takahashi: sendfile() on UDP works now.
45 #include <asm/uaccess.h>
46 #include <asm/system.h>
47 #include <linux/module.h>
48 #include <linux/types.h>
49 #include <linux/kernel.h>
50 #include <linux/mm.h>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
55 #include <linux/socket.h>
56 #include <linux/sockios.h>
57 #include <linux/in.h>
58 #include <linux/inet.h>
59 #include <linux/netdevice.h>
60 #include <linux/etherdevice.h>
61 #include <linux/proc_fs.h>
62 #include <linux/stat.h>
63 #include <linux/init.h>
65 #include <net/snmp.h>
66 #include <net/ip.h>
67 #include <net/protocol.h>
68 #include <net/route.h>
69 #include <net/xfrm.h>
70 #include <linux/skbuff.h>
71 #include <net/sock.h>
72 #include <net/arp.h>
73 #include <net/icmp.h>
74 #include <net/checksum.h>
75 #include <net/inetpeer.h>
76 #include <linux/igmp.h>
77 #include <linux/netfilter_ipv4.h>
78 #include <linux/netfilter_bridge.h>
79 #include <linux/mroute.h>
80 #include <linux/netlink.h>
81 #include <linux/tcp.h>
83 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
85 /* Generate a checksum for an outgoing IP datagram. */
86 __inline__ void ip_send_check(struct iphdr *iph)
88 iph->check = 0;
89 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
92 int __ip_local_out(struct sk_buff *skb)
94 struct iphdr *iph = ip_hdr(skb);
96 iph->tot_len = htons(skb->len);
97 ip_send_check(iph);
98 return nf_hook(PF_INET, NF_INET_LOCAL_OUT, skb, NULL, skb_dst(skb)->dev,
99 dst_output);
102 int ip_local_out(struct sk_buff *skb)
104 int err;
106 err = __ip_local_out(skb);
107 if (likely(err == 1))
108 err = dst_output(skb);
110 return err;
112 EXPORT_SYMBOL_GPL(ip_local_out);
114 /* dev_loopback_xmit for use with netfilter. */
115 static int ip_dev_loopback_xmit(struct sk_buff *newskb)
117 skb_reset_mac_header(newskb);
118 __skb_pull(newskb, skb_network_offset(newskb));
119 newskb->pkt_type = PACKET_LOOPBACK;
120 newskb->ip_summed = CHECKSUM_UNNECESSARY;
121 WARN_ON(!skb_dst(newskb));
122 netif_rx(newskb);
123 return 0;
126 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
128 int ttl = inet->uc_ttl;
130 if (ttl < 0)
131 ttl = dst_metric(dst, RTAX_HOPLIMIT);
132 return ttl;
136 * Add an ip header to a skbuff and send it out.
139 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
140 __be32 saddr, __be32 daddr, struct ip_options *opt)
142 struct inet_sock *inet = inet_sk(sk);
143 struct rtable *rt = skb_rtable(skb);
144 struct iphdr *iph;
146 /* Build the IP header. */
147 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
148 skb_reset_network_header(skb);
149 iph = ip_hdr(skb);
150 iph->version = 4;
151 iph->ihl = 5;
152 iph->tos = inet->tos;
153 if (ip_dont_fragment(sk, &rt->u.dst))
154 iph->frag_off = htons(IP_DF);
155 else
156 iph->frag_off = 0;
157 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
158 iph->daddr = rt->rt_dst;
159 iph->saddr = rt->rt_src;
160 iph->protocol = sk->sk_protocol;
161 ip_select_ident(iph, &rt->u.dst, sk);
163 if (opt && opt->optlen) {
164 iph->ihl += opt->optlen>>2;
165 ip_options_build(skb, opt, daddr, rt, 0);
168 skb->priority = sk->sk_priority;
169 skb->mark = sk->sk_mark;
171 /* Send it out. */
172 return ip_local_out(skb);
175 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
177 static inline int ip_finish_output2(struct sk_buff *skb)
179 struct dst_entry *dst = skb_dst(skb);
180 struct rtable *rt = (struct rtable *)dst;
181 struct net_device *dev = dst->dev;
182 unsigned int hh_len = LL_RESERVED_SPACE(dev);
184 if (rt->rt_type == RTN_MULTICAST) {
185 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
186 } else if (rt->rt_type == RTN_BROADCAST)
187 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
189 /* Be paranoid, rather than too clever. */
190 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
191 struct sk_buff *skb2;
193 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
194 if (skb2 == NULL) {
195 kfree_skb(skb);
196 return -ENOMEM;
198 if (skb->sk)
199 skb_set_owner_w(skb2, skb->sk);
200 kfree_skb(skb);
201 skb = skb2;
204 if (dst->hh)
205 return neigh_hh_output(dst->hh, skb);
206 else if (dst->neighbour)
207 return dst->neighbour->output(skb);
209 if (net_ratelimit())
210 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
211 kfree_skb(skb);
212 return -EINVAL;
215 static inline int ip_skb_dst_mtu(struct sk_buff *skb)
217 struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
219 return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
220 skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb));
223 static int ip_finish_output(struct sk_buff *skb)
225 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
226 /* Policy lookup after SNAT yielded a new policy */
227 if (skb_dst(skb)->xfrm != NULL) {
228 IPCB(skb)->flags |= IPSKB_REROUTED;
229 return dst_output(skb);
231 #endif
232 if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
233 return ip_fragment(skb, ip_finish_output2);
234 else
235 return ip_finish_output2(skb);
238 int ip_mc_output(struct sk_buff *skb)
240 struct sock *sk = skb->sk;
241 struct rtable *rt = skb_rtable(skb);
242 struct net_device *dev = rt->u.dst.dev;
245 * If the indicated interface is up and running, send the packet.
247 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
249 skb->dev = dev;
250 skb->protocol = htons(ETH_P_IP);
253 * Multicasts are looped back for other local users
256 if (rt->rt_flags&RTCF_MULTICAST) {
257 if ((!sk || inet_sk(sk)->mc_loop)
258 #ifdef CONFIG_IP_MROUTE
259 /* Small optimization: do not loopback not local frames,
260 which returned after forwarding; they will be dropped
261 by ip_mr_input in any case.
262 Note, that local frames are looped back to be delivered
263 to local recipients.
265 This check is duplicated in ip_mr_input at the moment.
267 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
268 #endif
270 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
271 if (newskb)
272 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb,
273 NULL, newskb->dev,
274 ip_dev_loopback_xmit);
277 /* Multicasts with ttl 0 must not go beyond the host */
279 if (ip_hdr(skb)->ttl == 0) {
280 kfree_skb(skb);
281 return 0;
285 if (rt->rt_flags&RTCF_BROADCAST) {
286 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
287 if (newskb)
288 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb, NULL,
289 newskb->dev, ip_dev_loopback_xmit);
292 return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
293 ip_finish_output,
294 !(IPCB(skb)->flags & IPSKB_REROUTED));
297 int ip_output(struct sk_buff *skb)
299 struct net_device *dev = skb_dst(skb)->dev;
301 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
303 skb->dev = dev;
304 skb->protocol = htons(ETH_P_IP);
306 return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, dev,
307 ip_finish_output,
308 !(IPCB(skb)->flags & IPSKB_REROUTED));
311 int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
313 struct sock *sk = skb->sk;
314 struct inet_sock *inet = inet_sk(sk);
315 struct ip_options *opt = inet->opt;
316 struct rtable *rt;
317 struct iphdr *iph;
319 /* Skip all of this if the packet is already routed,
320 * f.e. by something like SCTP.
322 rt = skb_rtable(skb);
323 if (rt != NULL)
324 goto packet_routed;
326 /* Make sure we can route this packet. */
327 rt = (struct rtable *)__sk_dst_check(sk, 0);
328 if (rt == NULL) {
329 __be32 daddr;
331 /* Use correct destination address if we have options. */
332 daddr = inet->daddr;
333 if(opt && opt->srr)
334 daddr = opt->faddr;
337 struct flowi fl = { .oif = sk->sk_bound_dev_if,
338 .nl_u = { .ip4_u =
339 { .daddr = daddr,
340 .saddr = inet->saddr,
341 .tos = RT_CONN_FLAGS(sk) } },
342 .proto = sk->sk_protocol,
343 .flags = inet_sk_flowi_flags(sk),
344 .uli_u = { .ports =
345 { .sport = inet->sport,
346 .dport = inet->dport } } };
348 /* If this fails, retransmit mechanism of transport layer will
349 * keep trying until route appears or the connection times
350 * itself out.
352 security_sk_classify_flow(sk, &fl);
353 if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0))
354 goto no_route;
356 sk_setup_caps(sk, &rt->u.dst);
358 skb_dst_set(skb, dst_clone(&rt->u.dst));
360 packet_routed:
361 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
362 goto no_route;
364 /* OK, we know where to send it, allocate and build IP header. */
365 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
366 skb_reset_network_header(skb);
367 iph = ip_hdr(skb);
368 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
369 if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
370 iph->frag_off = htons(IP_DF);
371 else
372 iph->frag_off = 0;
373 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
374 iph->protocol = sk->sk_protocol;
375 iph->saddr = rt->rt_src;
376 iph->daddr = rt->rt_dst;
377 /* Transport layer set skb->h.foo itself. */
379 if (opt && opt->optlen) {
380 iph->ihl += opt->optlen >> 2;
381 ip_options_build(skb, opt, inet->daddr, rt, 0);
384 ip_select_ident_more(iph, &rt->u.dst, sk,
385 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
387 skb->priority = sk->sk_priority;
388 skb->mark = sk->sk_mark;
390 return ip_local_out(skb);
392 no_route:
393 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
394 kfree_skb(skb);
395 return -EHOSTUNREACH;
399 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
401 to->pkt_type = from->pkt_type;
402 to->priority = from->priority;
403 to->protocol = from->protocol;
404 skb_dst_drop(to);
405 skb_dst_set(to, dst_clone(skb_dst(from)));
406 to->dev = from->dev;
407 to->mark = from->mark;
409 /* Copy the flags to each fragment. */
410 IPCB(to)->flags = IPCB(from)->flags;
412 #ifdef CONFIG_NET_SCHED
413 to->tc_index = from->tc_index;
414 #endif
415 nf_copy(to, from);
416 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
417 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
418 to->nf_trace = from->nf_trace;
419 #endif
420 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
421 to->ipvs_property = from->ipvs_property;
422 #endif
423 skb_copy_secmark(to, from);
427 * This IP datagram is too large to be sent in one piece. Break it up into
428 * smaller pieces (each of size equal to IP header plus
429 * a block of the data of the original IP data part) that will yet fit in a
430 * single device frame, and queue such a frame for sending.
433 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
435 struct iphdr *iph;
436 int raw = 0;
437 int ptr;
438 struct net_device *dev;
439 struct sk_buff *skb2;
440 unsigned int mtu, hlen, left, len, ll_rs, pad;
441 int offset;
442 __be16 not_last_frag;
443 struct rtable *rt = skb_rtable(skb);
444 int err = 0;
446 dev = rt->u.dst.dev;
449 * Point into the IP datagram header.
452 iph = ip_hdr(skb);
454 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
455 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
456 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
457 htonl(ip_skb_dst_mtu(skb)));
458 kfree_skb(skb);
459 return -EMSGSIZE;
463 * Setup starting values.
466 hlen = iph->ihl * 4;
467 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
468 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
470 /* When frag_list is given, use it. First, check its validity:
471 * some transformers could create wrong frag_list or break existing
472 * one, it is not prohibited. In this case fall back to copying.
474 * LATER: this step can be merged to real generation of fragments,
475 * we can switch to copy when see the first bad fragment.
477 if (skb_has_frags(skb)) {
478 struct sk_buff *frag;
479 int first_len = skb_pagelen(skb);
480 int truesizes = 0;
482 if (first_len - hlen > mtu ||
483 ((first_len - hlen) & 7) ||
484 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
485 skb_cloned(skb))
486 goto slow_path;
488 skb_walk_frags(skb, frag) {
489 /* Correct geometry. */
490 if (frag->len > mtu ||
491 ((frag->len & 7) && frag->next) ||
492 skb_headroom(frag) < hlen)
493 goto slow_path;
495 /* Partially cloned skb? */
496 if (skb_shared(frag))
497 goto slow_path;
499 BUG_ON(frag->sk);
500 if (skb->sk) {
501 frag->sk = skb->sk;
502 frag->destructor = sock_wfree;
503 truesizes += frag->truesize;
507 /* Everything is OK. Generate! */
509 err = 0;
510 offset = 0;
511 frag = skb_shinfo(skb)->frag_list;
512 skb_frag_list_init(skb);
513 skb->data_len = first_len - skb_headlen(skb);
514 skb->truesize -= truesizes;
515 skb->len = first_len;
516 iph->tot_len = htons(first_len);
517 iph->frag_off = htons(IP_MF);
518 ip_send_check(iph);
520 for (;;) {
521 /* Prepare header of the next frame,
522 * before previous one went down. */
523 if (frag) {
524 frag->ip_summed = CHECKSUM_NONE;
525 skb_reset_transport_header(frag);
526 __skb_push(frag, hlen);
527 skb_reset_network_header(frag);
528 memcpy(skb_network_header(frag), iph, hlen);
529 iph = ip_hdr(frag);
530 iph->tot_len = htons(frag->len);
531 ip_copy_metadata(frag, skb);
532 if (offset == 0)
533 ip_options_fragment(frag);
534 offset += skb->len - hlen;
535 iph->frag_off = htons(offset>>3);
536 if (frag->next != NULL)
537 iph->frag_off |= htons(IP_MF);
538 /* Ready, complete checksum */
539 ip_send_check(iph);
542 err = output(skb);
544 if (!err)
545 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
546 if (err || !frag)
547 break;
549 skb = frag;
550 frag = skb->next;
551 skb->next = NULL;
554 if (err == 0) {
555 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
556 return 0;
559 while (frag) {
560 skb = frag->next;
561 kfree_skb(frag);
562 frag = skb;
564 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
565 return err;
568 slow_path:
569 left = skb->len - hlen; /* Space per frame */
570 ptr = raw + hlen; /* Where to start from */
572 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
573 * we need to make room for the encapsulating header
575 pad = nf_bridge_pad(skb);
576 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
577 mtu -= pad;
580 * Fragment the datagram.
583 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
584 not_last_frag = iph->frag_off & htons(IP_MF);
587 * Keep copying data until we run out.
590 while (left > 0) {
591 len = left;
592 /* IF: it doesn't fit, use 'mtu' - the data space left */
593 if (len > mtu)
594 len = mtu;
595 /* IF: we are not sending upto and including the packet end
596 then align the next start on an eight byte boundary */
597 if (len < left) {
598 len &= ~7;
601 * Allocate buffer.
604 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
605 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
606 err = -ENOMEM;
607 goto fail;
611 * Set up data on packet
614 ip_copy_metadata(skb2, skb);
615 skb_reserve(skb2, ll_rs);
616 skb_put(skb2, len + hlen);
617 skb_reset_network_header(skb2);
618 skb2->transport_header = skb2->network_header + hlen;
621 * Charge the memory for the fragment to any owner
622 * it might possess
625 if (skb->sk)
626 skb_set_owner_w(skb2, skb->sk);
629 * Copy the packet header into the new buffer.
632 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
635 * Copy a block of the IP datagram.
637 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
638 BUG();
639 left -= len;
642 * Fill in the new header fields.
644 iph = ip_hdr(skb2);
645 iph->frag_off = htons((offset >> 3));
647 /* ANK: dirty, but effective trick. Upgrade options only if
648 * the segment to be fragmented was THE FIRST (otherwise,
649 * options are already fixed) and make it ONCE
650 * on the initial skb, so that all the following fragments
651 * will inherit fixed options.
653 if (offset == 0)
654 ip_options_fragment(skb);
657 * Added AC : If we are fragmenting a fragment that's not the
658 * last fragment then keep MF on each bit
660 if (left > 0 || not_last_frag)
661 iph->frag_off |= htons(IP_MF);
662 ptr += len;
663 offset += len;
666 * Put this fragment into the sending queue.
668 iph->tot_len = htons(len + hlen);
670 ip_send_check(iph);
672 err = output(skb2);
673 if (err)
674 goto fail;
676 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
678 kfree_skb(skb);
679 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
680 return err;
682 fail:
683 kfree_skb(skb);
684 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
685 return err;
688 EXPORT_SYMBOL(ip_fragment);
691 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
693 struct iovec *iov = from;
695 if (skb->ip_summed == CHECKSUM_PARTIAL) {
696 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
697 return -EFAULT;
698 } else {
699 __wsum csum = 0;
700 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
701 return -EFAULT;
702 skb->csum = csum_block_add(skb->csum, csum, odd);
704 return 0;
707 static inline __wsum
708 csum_page(struct page *page, int offset, int copy)
710 char *kaddr;
711 __wsum csum;
712 kaddr = kmap(page);
713 csum = csum_partial(kaddr + offset, copy, 0);
714 kunmap(page);
715 return csum;
718 static inline int ip_ufo_append_data(struct sock *sk,
719 int getfrag(void *from, char *to, int offset, int len,
720 int odd, struct sk_buff *skb),
721 void *from, int length, int hh_len, int fragheaderlen,
722 int transhdrlen, int mtu, unsigned int flags)
724 struct sk_buff *skb;
725 int err;
727 /* There is support for UDP fragmentation offload by network
728 * device, so create one single skb packet containing complete
729 * udp datagram
731 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
732 skb = sock_alloc_send_skb(sk,
733 hh_len + fragheaderlen + transhdrlen + 20,
734 (flags & MSG_DONTWAIT), &err);
736 if (skb == NULL)
737 return err;
739 /* reserve space for Hardware header */
740 skb_reserve(skb, hh_len);
742 /* create space for UDP/IP header */
743 skb_put(skb, fragheaderlen + transhdrlen);
745 /* initialize network header pointer */
746 skb_reset_network_header(skb);
748 /* initialize protocol header pointer */
749 skb->transport_header = skb->network_header + fragheaderlen;
751 skb->ip_summed = CHECKSUM_PARTIAL;
752 skb->csum = 0;
753 sk->sk_sndmsg_off = 0;
755 /* specify the length of each IP datagram fragment */
756 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
757 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
758 __skb_queue_tail(&sk->sk_write_queue, skb);
761 return skb_append_datato_frags(sk, skb, getfrag, from,
762 (length - transhdrlen));
766 * ip_append_data() and ip_append_page() can make one large IP datagram
767 * from many pieces of data. Each pieces will be holded on the socket
768 * until ip_push_pending_frames() is called. Each piece can be a page
769 * or non-page data.
771 * Not only UDP, other transport protocols - e.g. raw sockets - can use
772 * this interface potentially.
774 * LATER: length must be adjusted by pad at tail, when it is required.
776 int ip_append_data(struct sock *sk,
777 int getfrag(void *from, char *to, int offset, int len,
778 int odd, struct sk_buff *skb),
779 void *from, int length, int transhdrlen,
780 struct ipcm_cookie *ipc, struct rtable **rtp,
781 unsigned int flags)
783 struct inet_sock *inet = inet_sk(sk);
784 struct sk_buff *skb;
786 struct ip_options *opt = NULL;
787 int hh_len;
788 int exthdrlen;
789 int mtu;
790 int copy;
791 int err;
792 int offset = 0;
793 unsigned int maxfraglen, fragheaderlen;
794 int csummode = CHECKSUM_NONE;
795 struct rtable *rt;
797 if (flags&MSG_PROBE)
798 return 0;
800 if (skb_queue_empty(&sk->sk_write_queue)) {
802 * setup for corking.
804 opt = ipc->opt;
805 if (opt) {
806 if (inet->cork.opt == NULL) {
807 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
808 if (unlikely(inet->cork.opt == NULL))
809 return -ENOBUFS;
811 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
812 inet->cork.flags |= IPCORK_OPT;
813 inet->cork.addr = ipc->addr;
815 rt = *rtp;
816 if (unlikely(!rt))
817 return -EFAULT;
819 * We steal reference to this route, caller should not release it
821 *rtp = NULL;
822 inet->cork.fragsize = mtu = inet->pmtudisc == IP_PMTUDISC_PROBE ?
823 rt->u.dst.dev->mtu :
824 dst_mtu(rt->u.dst.path);
825 inet->cork.dst = &rt->u.dst;
826 inet->cork.length = 0;
827 sk->sk_sndmsg_page = NULL;
828 sk->sk_sndmsg_off = 0;
829 if ((exthdrlen = rt->u.dst.header_len) != 0) {
830 length += exthdrlen;
831 transhdrlen += exthdrlen;
833 } else {
834 rt = (struct rtable *)inet->cork.dst;
835 if (inet->cork.flags & IPCORK_OPT)
836 opt = inet->cork.opt;
838 transhdrlen = 0;
839 exthdrlen = 0;
840 mtu = inet->cork.fragsize;
842 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
844 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
845 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
847 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
848 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
849 return -EMSGSIZE;
853 * transhdrlen > 0 means that this is the first fragment and we wish
854 * it won't be fragmented in the future.
856 if (transhdrlen &&
857 length + fragheaderlen <= mtu &&
858 rt->u.dst.dev->features & NETIF_F_V4_CSUM &&
859 !exthdrlen)
860 csummode = CHECKSUM_PARTIAL;
862 inet->cork.length += length;
863 if (((length> mtu) || !skb_queue_empty(&sk->sk_write_queue)) &&
864 (sk->sk_protocol == IPPROTO_UDP) &&
865 (rt->u.dst.dev->features & NETIF_F_UFO)) {
866 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
867 fragheaderlen, transhdrlen, mtu,
868 flags);
869 if (err)
870 goto error;
871 return 0;
874 /* So, what's going on in the loop below?
876 * We use calculated fragment length to generate chained skb,
877 * each of segments is IP fragment ready for sending to network after
878 * adding appropriate IP header.
881 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
882 goto alloc_new_skb;
884 while (length > 0) {
885 /* Check if the remaining data fits into current packet. */
886 copy = mtu - skb->len;
887 if (copy < length)
888 copy = maxfraglen - skb->len;
889 if (copy <= 0) {
890 char *data;
891 unsigned int datalen;
892 unsigned int fraglen;
893 unsigned int fraggap;
894 unsigned int alloclen;
895 struct sk_buff *skb_prev;
896 alloc_new_skb:
897 skb_prev = skb;
898 if (skb_prev)
899 fraggap = skb_prev->len - maxfraglen;
900 else
901 fraggap = 0;
904 * If remaining data exceeds the mtu,
905 * we know we need more fragment(s).
907 datalen = length + fraggap;
908 if (datalen > mtu - fragheaderlen)
909 datalen = maxfraglen - fragheaderlen;
910 fraglen = datalen + fragheaderlen;
912 if ((flags & MSG_MORE) &&
913 !(rt->u.dst.dev->features&NETIF_F_SG))
914 alloclen = mtu;
915 else
916 alloclen = datalen + fragheaderlen;
918 /* The last fragment gets additional space at tail.
919 * Note, with MSG_MORE we overallocate on fragments,
920 * because we have no idea what fragment will be
921 * the last.
923 if (datalen == length + fraggap)
924 alloclen += rt->u.dst.trailer_len;
926 if (transhdrlen) {
927 skb = sock_alloc_send_skb(sk,
928 alloclen + hh_len + 15,
929 (flags & MSG_DONTWAIT), &err);
930 } else {
931 skb = NULL;
932 if (atomic_read(&sk->sk_wmem_alloc) <=
933 2 * sk->sk_sndbuf)
934 skb = sock_wmalloc(sk,
935 alloclen + hh_len + 15, 1,
936 sk->sk_allocation);
937 if (unlikely(skb == NULL))
938 err = -ENOBUFS;
939 else
940 /* only the initial fragment is
941 time stamped */
942 ipc->shtx.flags = 0;
944 if (skb == NULL)
945 goto error;
948 * Fill in the control structures
950 skb->ip_summed = csummode;
951 skb->csum = 0;
952 skb_reserve(skb, hh_len);
953 *skb_tx(skb) = ipc->shtx;
956 * Find where to start putting bytes.
958 data = skb_put(skb, fraglen);
959 skb_set_network_header(skb, exthdrlen);
960 skb->transport_header = (skb->network_header +
961 fragheaderlen);
962 data += fragheaderlen;
964 if (fraggap) {
965 skb->csum = skb_copy_and_csum_bits(
966 skb_prev, maxfraglen,
967 data + transhdrlen, fraggap, 0);
968 skb_prev->csum = csum_sub(skb_prev->csum,
969 skb->csum);
970 data += fraggap;
971 pskb_trim_unique(skb_prev, maxfraglen);
974 copy = datalen - transhdrlen - fraggap;
975 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
976 err = -EFAULT;
977 kfree_skb(skb);
978 goto error;
981 offset += copy;
982 length -= datalen - fraggap;
983 transhdrlen = 0;
984 exthdrlen = 0;
985 csummode = CHECKSUM_NONE;
988 * Put the packet on the pending queue.
990 __skb_queue_tail(&sk->sk_write_queue, skb);
991 continue;
994 if (copy > length)
995 copy = length;
997 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
998 unsigned int off;
1000 off = skb->len;
1001 if (getfrag(from, skb_put(skb, copy),
1002 offset, copy, off, skb) < 0) {
1003 __skb_trim(skb, off);
1004 err = -EFAULT;
1005 goto error;
1007 } else {
1008 int i = skb_shinfo(skb)->nr_frags;
1009 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1010 struct page *page = sk->sk_sndmsg_page;
1011 int off = sk->sk_sndmsg_off;
1012 unsigned int left;
1014 if (page && (left = PAGE_SIZE - off) > 0) {
1015 if (copy >= left)
1016 copy = left;
1017 if (page != frag->page) {
1018 if (i == MAX_SKB_FRAGS) {
1019 err = -EMSGSIZE;
1020 goto error;
1022 get_page(page);
1023 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1024 frag = &skb_shinfo(skb)->frags[i];
1026 } else if (i < MAX_SKB_FRAGS) {
1027 if (copy > PAGE_SIZE)
1028 copy = PAGE_SIZE;
1029 page = alloc_pages(sk->sk_allocation, 0);
1030 if (page == NULL) {
1031 err = -ENOMEM;
1032 goto error;
1034 sk->sk_sndmsg_page = page;
1035 sk->sk_sndmsg_off = 0;
1037 skb_fill_page_desc(skb, i, page, 0, 0);
1038 frag = &skb_shinfo(skb)->frags[i];
1039 } else {
1040 err = -EMSGSIZE;
1041 goto error;
1043 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1044 err = -EFAULT;
1045 goto error;
1047 sk->sk_sndmsg_off += copy;
1048 frag->size += copy;
1049 skb->len += copy;
1050 skb->data_len += copy;
1051 skb->truesize += copy;
1052 atomic_add(copy, &sk->sk_wmem_alloc);
1054 offset += copy;
1055 length -= copy;
1058 return 0;
1060 error:
1061 inet->cork.length -= length;
1062 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1063 return err;
1066 ssize_t ip_append_page(struct sock *sk, struct page *page,
1067 int offset, size_t size, int flags)
1069 struct inet_sock *inet = inet_sk(sk);
1070 struct sk_buff *skb;
1071 struct rtable *rt;
1072 struct ip_options *opt = NULL;
1073 int hh_len;
1074 int mtu;
1075 int len;
1076 int err;
1077 unsigned int maxfraglen, fragheaderlen, fraggap;
1079 if (inet->hdrincl)
1080 return -EPERM;
1082 if (flags&MSG_PROBE)
1083 return 0;
1085 if (skb_queue_empty(&sk->sk_write_queue))
1086 return -EINVAL;
1088 rt = (struct rtable *)inet->cork.dst;
1089 if (inet->cork.flags & IPCORK_OPT)
1090 opt = inet->cork.opt;
1092 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1093 return -EOPNOTSUPP;
1095 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1096 mtu = inet->cork.fragsize;
1098 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1099 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1101 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1102 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1103 return -EMSGSIZE;
1106 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1107 return -EINVAL;
1109 inet->cork.length += size;
1110 if ((sk->sk_protocol == IPPROTO_UDP) &&
1111 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1112 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1113 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1117 while (size > 0) {
1118 int i;
1120 if (skb_is_gso(skb))
1121 len = size;
1122 else {
1124 /* Check if the remaining data fits into current packet. */
1125 len = mtu - skb->len;
1126 if (len < size)
1127 len = maxfraglen - skb->len;
1129 if (len <= 0) {
1130 struct sk_buff *skb_prev;
1131 int alloclen;
1133 skb_prev = skb;
1134 fraggap = skb_prev->len - maxfraglen;
1136 alloclen = fragheaderlen + hh_len + fraggap + 15;
1137 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1138 if (unlikely(!skb)) {
1139 err = -ENOBUFS;
1140 goto error;
1144 * Fill in the control structures
1146 skb->ip_summed = CHECKSUM_NONE;
1147 skb->csum = 0;
1148 skb_reserve(skb, hh_len);
1151 * Find where to start putting bytes.
1153 skb_put(skb, fragheaderlen + fraggap);
1154 skb_reset_network_header(skb);
1155 skb->transport_header = (skb->network_header +
1156 fragheaderlen);
1157 if (fraggap) {
1158 skb->csum = skb_copy_and_csum_bits(skb_prev,
1159 maxfraglen,
1160 skb_transport_header(skb),
1161 fraggap, 0);
1162 skb_prev->csum = csum_sub(skb_prev->csum,
1163 skb->csum);
1164 pskb_trim_unique(skb_prev, maxfraglen);
1168 * Put the packet on the pending queue.
1170 __skb_queue_tail(&sk->sk_write_queue, skb);
1171 continue;
1174 i = skb_shinfo(skb)->nr_frags;
1175 if (len > size)
1176 len = size;
1177 if (skb_can_coalesce(skb, i, page, offset)) {
1178 skb_shinfo(skb)->frags[i-1].size += len;
1179 } else if (i < MAX_SKB_FRAGS) {
1180 get_page(page);
1181 skb_fill_page_desc(skb, i, page, offset, len);
1182 } else {
1183 err = -EMSGSIZE;
1184 goto error;
1187 if (skb->ip_summed == CHECKSUM_NONE) {
1188 __wsum csum;
1189 csum = csum_page(page, offset, len);
1190 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1193 skb->len += len;
1194 skb->data_len += len;
1195 skb->truesize += len;
1196 atomic_add(len, &sk->sk_wmem_alloc);
1197 offset += len;
1198 size -= len;
1200 return 0;
1202 error:
1203 inet->cork.length -= size;
1204 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1205 return err;
1208 static void ip_cork_release(struct inet_sock *inet)
1210 inet->cork.flags &= ~IPCORK_OPT;
1211 kfree(inet->cork.opt);
1212 inet->cork.opt = NULL;
1213 dst_release(inet->cork.dst);
1214 inet->cork.dst = NULL;
1218 * Combined all pending IP fragments on the socket as one IP datagram
1219 * and push them out.
1221 int ip_push_pending_frames(struct sock *sk)
1223 struct sk_buff *skb, *tmp_skb;
1224 struct sk_buff **tail_skb;
1225 struct inet_sock *inet = inet_sk(sk);
1226 struct net *net = sock_net(sk);
1227 struct ip_options *opt = NULL;
1228 struct rtable *rt = (struct rtable *)inet->cork.dst;
1229 struct iphdr *iph;
1230 __be16 df = 0;
1231 __u8 ttl;
1232 int err = 0;
1234 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1235 goto out;
1236 tail_skb = &(skb_shinfo(skb)->frag_list);
1238 /* move skb->data to ip header from ext header */
1239 if (skb->data < skb_network_header(skb))
1240 __skb_pull(skb, skb_network_offset(skb));
1241 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1242 __skb_pull(tmp_skb, skb_network_header_len(skb));
1243 *tail_skb = tmp_skb;
1244 tail_skb = &(tmp_skb->next);
1245 skb->len += tmp_skb->len;
1246 skb->data_len += tmp_skb->len;
1247 skb->truesize += tmp_skb->truesize;
1248 tmp_skb->destructor = NULL;
1249 tmp_skb->sk = NULL;
1252 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1253 * to fragment the frame generated here. No matter, what transforms
1254 * how transforms change size of the packet, it will come out.
1256 if (inet->pmtudisc < IP_PMTUDISC_DO)
1257 skb->local_df = 1;
1259 /* DF bit is set when we want to see DF on outgoing frames.
1260 * If local_df is set too, we still allow to fragment this frame
1261 * locally. */
1262 if (inet->pmtudisc >= IP_PMTUDISC_DO ||
1263 (skb->len <= dst_mtu(&rt->u.dst) &&
1264 ip_dont_fragment(sk, &rt->u.dst)))
1265 df = htons(IP_DF);
1267 if (inet->cork.flags & IPCORK_OPT)
1268 opt = inet->cork.opt;
1270 if (rt->rt_type == RTN_MULTICAST)
1271 ttl = inet->mc_ttl;
1272 else
1273 ttl = ip_select_ttl(inet, &rt->u.dst);
1275 iph = (struct iphdr *)skb->data;
1276 iph->version = 4;
1277 iph->ihl = 5;
1278 if (opt) {
1279 iph->ihl += opt->optlen>>2;
1280 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1282 iph->tos = inet->tos;
1283 iph->frag_off = df;
1284 ip_select_ident(iph, &rt->u.dst, sk);
1285 iph->ttl = ttl;
1286 iph->protocol = sk->sk_protocol;
1287 iph->saddr = rt->rt_src;
1288 iph->daddr = rt->rt_dst;
1290 skb->priority = sk->sk_priority;
1291 skb->mark = sk->sk_mark;
1293 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1294 * on dst refcount
1296 inet->cork.dst = NULL;
1297 skb_dst_set(skb, &rt->u.dst);
1299 if (iph->protocol == IPPROTO_ICMP)
1300 icmp_out_count(net, ((struct icmphdr *)
1301 skb_transport_header(skb))->type);
1303 /* Netfilter gets whole the not fragmented skb. */
1304 err = ip_local_out(skb);
1305 if (err) {
1306 if (err > 0)
1307 err = net_xmit_errno(err);
1308 if (err)
1309 goto error;
1312 out:
1313 ip_cork_release(inet);
1314 return err;
1316 error:
1317 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1318 goto out;
1322 * Throw away all pending data on the socket.
1324 void ip_flush_pending_frames(struct sock *sk)
1326 struct sk_buff *skb;
1328 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1329 kfree_skb(skb);
1331 ip_cork_release(inet_sk(sk));
1336 * Fetch data from kernel space and fill in checksum if needed.
1338 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1339 int len, int odd, struct sk_buff *skb)
1341 __wsum csum;
1343 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1344 skb->csum = csum_block_add(skb->csum, csum, odd);
1345 return 0;
1349 * Generic function to send a packet as reply to another packet.
1350 * Used to send TCP resets so far. ICMP should use this function too.
1352 * Should run single threaded per socket because it uses the sock
1353 * structure to pass arguments.
1355 void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1356 unsigned int len)
1358 struct inet_sock *inet = inet_sk(sk);
1359 struct {
1360 struct ip_options opt;
1361 char data[40];
1362 } replyopts;
1363 struct ipcm_cookie ipc;
1364 __be32 daddr;
1365 struct rtable *rt = skb_rtable(skb);
1367 if (ip_options_echo(&replyopts.opt, skb))
1368 return;
1370 daddr = ipc.addr = rt->rt_src;
1371 ipc.opt = NULL;
1372 ipc.shtx.flags = 0;
1374 if (replyopts.opt.optlen) {
1375 ipc.opt = &replyopts.opt;
1377 if (ipc.opt->srr)
1378 daddr = replyopts.opt.faddr;
1382 struct flowi fl = { .oif = arg->bound_dev_if,
1383 .nl_u = { .ip4_u =
1384 { .daddr = daddr,
1385 .saddr = rt->rt_spec_dst,
1386 .tos = RT_TOS(ip_hdr(skb)->tos) } },
1387 /* Not quite clean, but right. */
1388 .uli_u = { .ports =
1389 { .sport = tcp_hdr(skb)->dest,
1390 .dport = tcp_hdr(skb)->source } },
1391 .proto = sk->sk_protocol,
1392 .flags = ip_reply_arg_flowi_flags(arg) };
1393 security_skb_classify_flow(skb, &fl);
1394 if (ip_route_output_key(sock_net(sk), &rt, &fl))
1395 return;
1398 /* And let IP do all the hard work.
1400 This chunk is not reenterable, hence spinlock.
1401 Note that it uses the fact, that this function is called
1402 with locally disabled BH and that sk cannot be already spinlocked.
1404 bh_lock_sock(sk);
1405 inet->tos = ip_hdr(skb)->tos;
1406 sk->sk_priority = skb->priority;
1407 sk->sk_protocol = ip_hdr(skb)->protocol;
1408 sk->sk_bound_dev_if = arg->bound_dev_if;
1409 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1410 &ipc, &rt, MSG_DONTWAIT);
1411 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1412 if (arg->csumoffset >= 0)
1413 *((__sum16 *)skb_transport_header(skb) +
1414 arg->csumoffset) = csum_fold(csum_add(skb->csum,
1415 arg->csum));
1416 skb->ip_summed = CHECKSUM_NONE;
1417 ip_push_pending_frames(sk);
1420 bh_unlock_sock(sk);
1422 ip_rt_put(rt);
1425 void __init ip_init(void)
1427 ip_rt_init();
1428 inet_initpeers();
1430 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1431 igmp_mc_proc_init();
1432 #endif
1435 EXPORT_SYMBOL(ip_generic_getfrag);
1436 EXPORT_SYMBOL(ip_queue_xmit);
1437 EXPORT_SYMBOL(ip_send_check);