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 $
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Donald Becker, <becker@super.org>
13 * Alan Cox, <Alan.Cox@linux.org>
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
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
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
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>
54 #include <linux/string.h>
55 #include <linux/errno.h>
56 #include <linux/config.h>
58 #include <linux/socket.h>
59 #include <linux/sockios.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/etherdevice.h>
64 #include <linux/proc_fs.h>
65 #include <linux/stat.h>
66 #include <linux/init.h>
70 #include <net/protocol.h>
71 #include <net/route.h>
72 #include <linux/skbuff.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 = IPDEFTTL;
88 /* Generate a checksum for an outgoing IP datagram. */
89 __inline__ void ip_send_check(struct iphdr *iph)
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);
107 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
109 int ttl = inet->uc_ttl;
112 ttl = dst_metric(dst, RTAX_HOPLIMIT);
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 u32 saddr, u32 daddr, struct ip_options *opt)
123 struct inet_sock *inet = inet_sk(sk);
124 struct rtable *rt = (struct rtable *)skb->dst;
127 /* Build the IP header. */
129 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr) + opt->optlen);
131 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr));
135 iph->tos = inet->tos;
136 if (ip_dont_fragment(sk, &rt->u.dst))
137 iph->frag_off = htons(IP_DF);
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);
148 if (opt && opt->optlen) {
149 iph->ihl += opt->optlen>>2;
150 ip_options_build(skb, opt, daddr, rt, 0);
154 skb->priority = sk->sk_priority;
157 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
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));
180 skb_set_owner_w(skb2, skb->sk);
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);
198 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
203 static inline int ip_finish_output(struct sk_buff *skb)
205 struct net_device *dev = skb->dst->dev;
208 skb->protocol = htons(ETH_P_IP);
210 return NF_HOOK(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
214 int ip_mc_output(struct sk_buff *skb)
216 struct sock *sk = skb->sk;
217 struct rtable *rt = (struct rtable*)skb->dst;
218 struct net_device *dev = rt->u.dst.dev;
221 * If the indicated interface is up and running, send the packet.
223 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
226 skb->protocol = htons(ETH_P_IP);
229 * Multicasts are looped back for other local users
232 if (rt->rt_flags&RTCF_MULTICAST) {
233 if ((!sk || inet_sk(sk)->mc_loop)
234 #ifdef CONFIG_IP_MROUTE
235 /* Small optimization: do not loopback not local frames,
236 which returned after forwarding; they will be dropped
237 by ip_mr_input in any case.
238 Note, that local frames are looped back to be delivered
241 This check is duplicated in ip_mr_input at the moment.
243 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
246 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
248 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
250 ip_dev_loopback_xmit);
253 /* Multicasts with ttl 0 must not go beyond the host */
255 if (skb->nh.iph->ttl == 0) {
261 if (rt->rt_flags&RTCF_BROADCAST) {
262 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
264 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
265 newskb->dev, ip_dev_loopback_xmit);
268 if (skb->len > dst_mtu(&rt->u.dst))
269 return ip_fragment(skb, ip_finish_output);
271 return ip_finish_output(skb);
274 int ip_output(struct sk_buff *skb)
276 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
278 if (skb->len > dst_mtu(skb->dst) &&
279 !(skb_shinfo(skb)->ufo_size || skb_shinfo(skb)->tso_size))
280 return ip_fragment(skb, ip_finish_output);
282 return ip_finish_output(skb);
285 int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
287 struct sock *sk = skb->sk;
288 struct inet_sock *inet = inet_sk(sk);
289 struct ip_options *opt = inet->opt;
293 /* Skip all of this if the packet is already routed,
294 * f.e. by something like SCTP.
296 rt = (struct rtable *) skb->dst;
300 /* Make sure we can route this packet. */
301 rt = (struct rtable *)__sk_dst_check(sk, 0);
305 /* Use correct destination address if we have options. */
311 struct flowi fl = { .oif = sk->sk_bound_dev_if,
314 .saddr = inet->saddr,
315 .tos = RT_CONN_FLAGS(sk) } },
316 .proto = sk->sk_protocol,
318 { .sport = inet->sport,
319 .dport = inet->dport } } };
321 /* If this fails, retransmit mechanism of transport layer will
322 * keep trying until route appears or the connection times
325 if (ip_route_output_flow(&rt, &fl, sk, 0))
328 sk_setup_caps(sk, &rt->u.dst);
330 skb->dst = dst_clone(&rt->u.dst);
333 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
336 /* OK, we know where to send it, allocate and build IP header. */
337 iph = (struct iphdr *) skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
338 *((__u16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
339 iph->tot_len = htons(skb->len);
340 if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
341 iph->frag_off = htons(IP_DF);
344 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
345 iph->protocol = sk->sk_protocol;
346 iph->saddr = rt->rt_src;
347 iph->daddr = rt->rt_dst;
349 /* Transport layer set skb->h.foo itself. */
351 if (opt && opt->optlen) {
352 iph->ihl += opt->optlen >> 2;
353 ip_options_build(skb, opt, inet->daddr, rt, 0);
356 ip_select_ident_more(iph, &rt->u.dst, sk,
357 (skb_shinfo(skb)->tso_segs ?: 1) - 1);
359 /* Add an IP checksum. */
362 skb->priority = sk->sk_priority;
364 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
368 IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
370 return -EHOSTUNREACH;
374 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
376 to->pkt_type = from->pkt_type;
377 to->priority = from->priority;
378 to->protocol = from->protocol;
379 dst_release(to->dst);
380 to->dst = dst_clone(from->dst);
383 /* Copy the flags to each fragment. */
384 IPCB(to)->flags = IPCB(from)->flags;
386 #ifdef CONFIG_NET_SCHED
387 to->tc_index = from->tc_index;
389 #ifdef CONFIG_NETFILTER
390 to->nfmark = from->nfmark;
391 /* Connection association is same as pre-frag packet */
392 nf_conntrack_put(to->nfct);
393 to->nfct = from->nfct;
394 nf_conntrack_get(to->nfct);
395 to->nfctinfo = from->nfctinfo;
396 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
397 to->ipvs_property = from->ipvs_property;
399 #ifdef CONFIG_BRIDGE_NETFILTER
400 nf_bridge_put(to->nf_bridge);
401 to->nf_bridge = from->nf_bridge;
402 nf_bridge_get(to->nf_bridge);
408 * This IP datagram is too large to be sent in one piece. Break it up into
409 * smaller pieces (each of size equal to IP header plus
410 * a block of the data of the original IP data part) that will yet fit in a
411 * single device frame, and queue such a frame for sending.
414 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
419 struct net_device *dev;
420 struct sk_buff *skb2;
421 unsigned int mtu, hlen, left, len, ll_rs;
424 struct rtable *rt = (struct rtable*)skb->dst;
430 * Point into the IP datagram header.
435 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
436 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
437 htonl(dst_mtu(&rt->u.dst)));
443 * Setup starting values.
447 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
448 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
450 /* When frag_list is given, use it. First, check its validity:
451 * some transformers could create wrong frag_list or break existing
452 * one, it is not prohibited. In this case fall back to copying.
454 * LATER: this step can be merged to real generation of fragments,
455 * we can switch to copy when see the first bad fragment.
457 if (skb_shinfo(skb)->frag_list) {
458 struct sk_buff *frag;
459 int first_len = skb_pagelen(skb);
461 if (first_len - hlen > mtu ||
462 ((first_len - hlen) & 7) ||
463 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
467 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
468 /* Correct geometry. */
469 if (frag->len > mtu ||
470 ((frag->len & 7) && frag->next) ||
471 skb_headroom(frag) < hlen)
474 /* Partially cloned skb? */
475 if (skb_shared(frag))
482 frag->destructor = sock_wfree;
483 skb->truesize -= frag->truesize;
487 /* Everything is OK. Generate! */
491 frag = skb_shinfo(skb)->frag_list;
492 skb_shinfo(skb)->frag_list = NULL;
493 skb->data_len = first_len - skb_headlen(skb);
494 skb->len = first_len;
495 iph->tot_len = htons(first_len);
496 iph->frag_off = htons(IP_MF);
500 /* Prepare header of the next frame,
501 * before previous one went down. */
503 frag->ip_summed = CHECKSUM_NONE;
504 frag->h.raw = frag->data;
505 frag->nh.raw = __skb_push(frag, hlen);
506 memcpy(frag->nh.raw, iph, hlen);
508 iph->tot_len = htons(frag->len);
509 ip_copy_metadata(frag, skb);
511 ip_options_fragment(frag);
512 offset += skb->len - hlen;
513 iph->frag_off = htons(offset>>3);
514 if (frag->next != NULL)
515 iph->frag_off |= htons(IP_MF);
516 /* Ready, complete checksum */
531 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
540 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
545 left = skb->len - hlen; /* Space per frame */
546 ptr = raw + hlen; /* Where to start from */
548 #ifdef CONFIG_BRIDGE_NETFILTER
549 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
550 * we need to make room for the encapsulating header */
551 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, nf_bridge_pad(skb));
552 mtu -= nf_bridge_pad(skb);
554 ll_rs = LL_RESERVED_SPACE(rt->u.dst.dev);
557 * Fragment the datagram.
560 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
561 not_last_frag = iph->frag_off & htons(IP_MF);
564 * Keep copying data until we run out.
569 /* IF: it doesn't fit, use 'mtu' - the data space left */
572 /* IF: we are not sending upto and including the packet end
573 then align the next start on an eight byte boundary */
581 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
582 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
588 * Set up data on packet
591 ip_copy_metadata(skb2, skb);
592 skb_reserve(skb2, ll_rs);
593 skb_put(skb2, len + hlen);
594 skb2->nh.raw = skb2->data;
595 skb2->h.raw = skb2->data + hlen;
598 * Charge the memory for the fragment to any owner
603 skb_set_owner_w(skb2, skb->sk);
606 * Copy the packet header into the new buffer.
609 memcpy(skb2->nh.raw, skb->data, hlen);
612 * Copy a block of the IP datagram.
614 if (skb_copy_bits(skb, ptr, skb2->h.raw, len))
619 * Fill in the new header fields.
622 iph->frag_off = htons((offset >> 3));
624 /* ANK: dirty, but effective trick. Upgrade options only if
625 * the segment to be fragmented was THE FIRST (otherwise,
626 * options are already fixed) and make it ONCE
627 * on the initial skb, so that all the following fragments
628 * will inherit fixed options.
631 ip_options_fragment(skb);
634 * Added AC : If we are fragmenting a fragment that's not the
635 * last fragment then keep MF on each bit
637 if (left > 0 || not_last_frag)
638 iph->frag_off |= htons(IP_MF);
643 * Put this fragment into the sending queue.
646 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
648 iph->tot_len = htons(len + hlen);
657 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
662 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
667 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
669 struct iovec *iov = from;
671 if (skb->ip_summed == CHECKSUM_HW) {
672 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
675 unsigned int csum = 0;
676 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
678 skb->csum = csum_block_add(skb->csum, csum, odd);
683 static inline unsigned int
684 csum_page(struct page *page, int offset, int copy)
689 csum = csum_partial(kaddr + offset, copy, 0);
694 static inline int ip_ufo_append_data(struct sock *sk,
695 int getfrag(void *from, char *to, int offset, int len,
696 int odd, struct sk_buff *skb),
697 void *from, int length, int hh_len, int fragheaderlen,
698 int transhdrlen, int mtu,unsigned int flags)
703 /* There is support for UDP fragmentation offload by network
704 * device, so create one single skb packet containing complete
707 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
708 skb = sock_alloc_send_skb(sk,
709 hh_len + fragheaderlen + transhdrlen + 20,
710 (flags & MSG_DONTWAIT), &err);
715 /* reserve space for Hardware header */
716 skb_reserve(skb, hh_len);
718 /* create space for UDP/IP header */
719 skb_put(skb,fragheaderlen + transhdrlen);
721 /* initialize network header pointer */
722 skb->nh.raw = skb->data;
724 /* initialize protocol header pointer */
725 skb->h.raw = skb->data + fragheaderlen;
727 skb->ip_summed = CHECKSUM_HW;
729 sk->sk_sndmsg_off = 0;
732 err = skb_append_datato_frags(sk,skb, getfrag, from,
733 (length - transhdrlen));
735 /* specify the length of each IP datagram fragment*/
736 skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen);
737 __skb_queue_tail(&sk->sk_write_queue, skb);
741 /* There is not enough support do UFO ,
742 * so follow normal path
749 * ip_append_data() and ip_append_page() can make one large IP datagram
750 * from many pieces of data. Each pieces will be holded on the socket
751 * until ip_push_pending_frames() is called. Each piece can be a page
754 * Not only UDP, other transport protocols - e.g. raw sockets - can use
755 * this interface potentially.
757 * LATER: length must be adjusted by pad at tail, when it is required.
759 int ip_append_data(struct sock *sk,
760 int getfrag(void *from, char *to, int offset, int len,
761 int odd, struct sk_buff *skb),
762 void *from, int length, int transhdrlen,
763 struct ipcm_cookie *ipc, struct rtable *rt,
766 struct inet_sock *inet = inet_sk(sk);
769 struct ip_options *opt = NULL;
776 unsigned int maxfraglen, fragheaderlen;
777 int csummode = CHECKSUM_NONE;
782 if (skb_queue_empty(&sk->sk_write_queue)) {
788 if (inet->cork.opt == NULL) {
789 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
790 if (unlikely(inet->cork.opt == NULL))
793 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
794 inet->cork.flags |= IPCORK_OPT;
795 inet->cork.addr = ipc->addr;
797 dst_hold(&rt->u.dst);
798 inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path);
800 inet->cork.length = 0;
801 sk->sk_sndmsg_page = NULL;
802 sk->sk_sndmsg_off = 0;
803 if ((exthdrlen = rt->u.dst.header_len) != 0) {
805 transhdrlen += exthdrlen;
809 if (inet->cork.flags & IPCORK_OPT)
810 opt = inet->cork.opt;
814 mtu = inet->cork.fragsize;
816 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
818 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
819 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
821 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
822 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
827 * transhdrlen > 0 means that this is the first fragment and we wish
828 * it won't be fragmented in the future.
831 length + fragheaderlen <= mtu &&
832 rt->u.dst.dev->features&(NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM) &&
834 csummode = CHECKSUM_HW;
836 inet->cork.length += length;
837 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
838 (rt->u.dst.dev->features & NETIF_F_UFO)) {
840 if(ip_ufo_append_data(sk, getfrag, from, length, hh_len,
841 fragheaderlen, transhdrlen, mtu, flags))
847 /* So, what's going on in the loop below?
849 * We use calculated fragment length to generate chained skb,
850 * each of segments is IP fragment ready for sending to network after
851 * adding appropriate IP header.
854 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
858 /* Check if the remaining data fits into current packet. */
859 copy = mtu - skb->len;
861 copy = maxfraglen - skb->len;
864 unsigned int datalen;
865 unsigned int fraglen;
866 unsigned int fraggap;
867 unsigned int alloclen;
868 struct sk_buff *skb_prev;
872 fraggap = skb_prev->len - maxfraglen;
877 * If remaining data exceeds the mtu,
878 * we know we need more fragment(s).
880 datalen = length + fraggap;
881 if (datalen > mtu - fragheaderlen)
882 datalen = maxfraglen - fragheaderlen;
883 fraglen = datalen + fragheaderlen;
885 if ((flags & MSG_MORE) &&
886 !(rt->u.dst.dev->features&NETIF_F_SG))
889 alloclen = datalen + fragheaderlen;
891 /* The last fragment gets additional space at tail.
892 * Note, with MSG_MORE we overallocate on fragments,
893 * because we have no idea what fragment will be
896 if (datalen == length)
897 alloclen += rt->u.dst.trailer_len;
900 skb = sock_alloc_send_skb(sk,
901 alloclen + hh_len + 15,
902 (flags & MSG_DONTWAIT), &err);
905 if (atomic_read(&sk->sk_wmem_alloc) <=
907 skb = sock_wmalloc(sk,
908 alloclen + hh_len + 15, 1,
910 if (unlikely(skb == NULL))
917 * Fill in the control structures
919 skb->ip_summed = csummode;
921 skb_reserve(skb, hh_len);
924 * Find where to start putting bytes.
926 data = skb_put(skb, fraglen);
927 skb->nh.raw = data + exthdrlen;
928 data += fragheaderlen;
929 skb->h.raw = data + exthdrlen;
932 skb->csum = skb_copy_and_csum_bits(
933 skb_prev, maxfraglen,
934 data + transhdrlen, fraggap, 0);
935 skb_prev->csum = csum_sub(skb_prev->csum,
938 skb_trim(skb_prev, maxfraglen);
941 copy = datalen - transhdrlen - fraggap;
942 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
949 length -= datalen - fraggap;
952 csummode = CHECKSUM_NONE;
955 * Put the packet on the pending queue.
957 __skb_queue_tail(&sk->sk_write_queue, skb);
964 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
968 if (getfrag(from, skb_put(skb, copy),
969 offset, copy, off, skb) < 0) {
970 __skb_trim(skb, off);
975 int i = skb_shinfo(skb)->nr_frags;
976 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
977 struct page *page = sk->sk_sndmsg_page;
978 int off = sk->sk_sndmsg_off;
981 if (page && (left = PAGE_SIZE - off) > 0) {
984 if (page != frag->page) {
985 if (i == MAX_SKB_FRAGS) {
990 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
991 frag = &skb_shinfo(skb)->frags[i];
993 } else if (i < MAX_SKB_FRAGS) {
994 if (copy > PAGE_SIZE)
996 page = alloc_pages(sk->sk_allocation, 0);
1001 sk->sk_sndmsg_page = page;
1002 sk->sk_sndmsg_off = 0;
1004 skb_fill_page_desc(skb, i, page, 0, 0);
1005 frag = &skb_shinfo(skb)->frags[i];
1006 skb->truesize += PAGE_SIZE;
1007 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1012 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1016 sk->sk_sndmsg_off += copy;
1019 skb->data_len += copy;
1028 inet->cork.length -= length;
1029 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1033 ssize_t ip_append_page(struct sock *sk, struct page *page,
1034 int offset, size_t size, int flags)
1036 struct inet_sock *inet = inet_sk(sk);
1037 struct sk_buff *skb;
1039 struct ip_options *opt = NULL;
1044 unsigned int maxfraglen, fragheaderlen, fraggap;
1049 if (flags&MSG_PROBE)
1052 if (skb_queue_empty(&sk->sk_write_queue))
1056 if (inet->cork.flags & IPCORK_OPT)
1057 opt = inet->cork.opt;
1059 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1062 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1063 mtu = inet->cork.fragsize;
1065 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1066 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1068 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1069 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1073 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1076 inet->cork.length += size;
1077 if ((sk->sk_protocol == IPPROTO_UDP) &&
1078 (rt->u.dst.dev->features & NETIF_F_UFO))
1079 skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen);
1085 if (skb_shinfo(skb)->ufo_size)
1089 /* Check if the remaining data fits into current packet. */
1090 len = mtu - skb->len;
1092 len = maxfraglen - skb->len;
1095 struct sk_buff *skb_prev;
1101 fraggap = skb_prev->len - maxfraglen;
1103 alloclen = fragheaderlen + hh_len + fraggap + 15;
1104 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1105 if (unlikely(!skb)) {
1111 * Fill in the control structures
1113 skb->ip_summed = CHECKSUM_NONE;
1115 skb_reserve(skb, hh_len);
1118 * Find where to start putting bytes.
1120 data = skb_put(skb, fragheaderlen + fraggap);
1121 skb->nh.iph = iph = (struct iphdr *)data;
1122 data += fragheaderlen;
1126 skb->csum = skb_copy_and_csum_bits(
1127 skb_prev, maxfraglen,
1129 skb_prev->csum = csum_sub(skb_prev->csum,
1131 skb_trim(skb_prev, maxfraglen);
1135 * Put the packet on the pending queue.
1137 __skb_queue_tail(&sk->sk_write_queue, skb);
1141 i = skb_shinfo(skb)->nr_frags;
1144 if (skb_can_coalesce(skb, i, page, offset)) {
1145 skb_shinfo(skb)->frags[i-1].size += len;
1146 } else if (i < MAX_SKB_FRAGS) {
1148 skb_fill_page_desc(skb, i, page, offset, len);
1154 if (skb->ip_summed == CHECKSUM_NONE) {
1156 csum = csum_page(page, offset, len);
1157 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1161 skb->data_len += len;
1168 inet->cork.length -= size;
1169 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1174 * Combined all pending IP fragments on the socket as one IP datagram
1175 * and push them out.
1177 int ip_push_pending_frames(struct sock *sk)
1179 struct sk_buff *skb, *tmp_skb;
1180 struct sk_buff **tail_skb;
1181 struct inet_sock *inet = inet_sk(sk);
1182 struct ip_options *opt = NULL;
1183 struct rtable *rt = inet->cork.rt;
1189 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1191 tail_skb = &(skb_shinfo(skb)->frag_list);
1193 /* move skb->data to ip header from ext header */
1194 if (skb->data < skb->nh.raw)
1195 __skb_pull(skb, skb->nh.raw - skb->data);
1196 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1197 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1198 *tail_skb = tmp_skb;
1199 tail_skb = &(tmp_skb->next);
1200 skb->len += tmp_skb->len;
1201 skb->data_len += tmp_skb->len;
1202 skb->truesize += tmp_skb->truesize;
1203 __sock_put(tmp_skb->sk);
1204 tmp_skb->destructor = NULL;
1208 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1209 * to fragment the frame generated here. No matter, what transforms
1210 * how transforms change size of the packet, it will come out.
1212 if (inet->pmtudisc != IP_PMTUDISC_DO)
1215 /* DF bit is set when we want to see DF on outgoing frames.
1216 * If local_df is set too, we still allow to fragment this frame
1218 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1219 (skb->len <= dst_mtu(&rt->u.dst) &&
1220 ip_dont_fragment(sk, &rt->u.dst)))
1223 if (inet->cork.flags & IPCORK_OPT)
1224 opt = inet->cork.opt;
1226 if (rt->rt_type == RTN_MULTICAST)
1229 ttl = ip_select_ttl(inet, &rt->u.dst);
1231 iph = (struct iphdr *)skb->data;
1235 iph->ihl += opt->optlen>>2;
1236 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1238 iph->tos = inet->tos;
1239 iph->tot_len = htons(skb->len);
1242 __ip_select_ident(iph, &rt->u.dst, 0);
1244 iph->id = htons(inet->id++);
1247 iph->protocol = sk->sk_protocol;
1248 iph->saddr = rt->rt_src;
1249 iph->daddr = rt->rt_dst;
1252 skb->priority = sk->sk_priority;
1253 skb->dst = dst_clone(&rt->u.dst);
1255 /* Netfilter gets whole the not fragmented skb. */
1256 err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL,
1257 skb->dst->dev, dst_output);
1260 err = inet->recverr ? net_xmit_errno(err) : 0;
1266 inet->cork.flags &= ~IPCORK_OPT;
1267 kfree(inet->cork.opt);
1268 inet->cork.opt = NULL;
1269 if (inet->cork.rt) {
1270 ip_rt_put(inet->cork.rt);
1271 inet->cork.rt = NULL;
1276 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1281 * Throw away all pending data on the socket.
1283 void ip_flush_pending_frames(struct sock *sk)
1285 struct inet_sock *inet = inet_sk(sk);
1286 struct sk_buff *skb;
1288 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1291 inet->cork.flags &= ~IPCORK_OPT;
1292 kfree(inet->cork.opt);
1293 inet->cork.opt = NULL;
1294 if (inet->cork.rt) {
1295 ip_rt_put(inet->cork.rt);
1296 inet->cork.rt = NULL;
1302 * Fetch data from kernel space and fill in checksum if needed.
1304 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1305 int len, int odd, struct sk_buff *skb)
1309 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1310 skb->csum = csum_block_add(skb->csum, csum, odd);
1315 * Generic function to send a packet as reply to another packet.
1316 * Used to send TCP resets so far. ICMP should use this function too.
1318 * Should run single threaded per socket because it uses the sock
1319 * structure to pass arguments.
1321 * LATER: switch from ip_build_xmit to ip_append_*
1323 void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1326 struct inet_sock *inet = inet_sk(sk);
1328 struct ip_options opt;
1331 struct ipcm_cookie ipc;
1333 struct rtable *rt = (struct rtable*)skb->dst;
1335 if (ip_options_echo(&replyopts.opt, skb))
1338 daddr = ipc.addr = rt->rt_src;
1341 if (replyopts.opt.optlen) {
1342 ipc.opt = &replyopts.opt;
1345 daddr = replyopts.opt.faddr;
1349 struct flowi fl = { .nl_u = { .ip4_u =
1351 .saddr = rt->rt_spec_dst,
1352 .tos = RT_TOS(skb->nh.iph->tos) } },
1353 /* Not quite clean, but right. */
1355 { .sport = skb->h.th->dest,
1356 .dport = skb->h.th->source } },
1357 .proto = sk->sk_protocol };
1358 if (ip_route_output_key(&rt, &fl))
1362 /* And let IP do all the hard work.
1364 This chunk is not reenterable, hence spinlock.
1365 Note that it uses the fact, that this function is called
1366 with locally disabled BH and that sk cannot be already spinlocked.
1369 inet->tos = skb->nh.iph->tos;
1370 sk->sk_priority = skb->priority;
1371 sk->sk_protocol = skb->nh.iph->protocol;
1372 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1373 &ipc, rt, MSG_DONTWAIT);
1374 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1375 if (arg->csumoffset >= 0)
1376 *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
1377 skb->ip_summed = CHECKSUM_NONE;
1378 ip_push_pending_frames(sk);
1386 void __init ip_init(void)
1391 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1392 igmp_mc_proc_init();
1396 EXPORT_SYMBOL(ip_fragment);
1397 EXPORT_SYMBOL(ip_generic_getfrag);
1398 EXPORT_SYMBOL(ip_queue_xmit);
1399 EXPORT_SYMBOL(ip_send_check);