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.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <Alan.Cox@linux.org>
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
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
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
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>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
54 #include <linux/slab.h>
56 #include <linux/socket.h>
57 #include <linux/sockios.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/etherdevice.h>
62 #include <linux/proc_fs.h>
63 #include <linux/stat.h>
64 #include <linux/init.h>
68 #include <net/protocol.h>
69 #include <net/route.h>
71 #include <linux/skbuff.h>
75 #include <net/checksum.h>
76 #include <net/inetpeer.h>
77 #include <linux/igmp.h>
78 #include <linux/netfilter_ipv4.h>
79 #include <linux/netfilter_bridge.h>
80 #include <linux/mroute.h>
81 #include <linux/netlink.h>
82 #include <linux/tcp.h>
84 int sysctl_ip_default_ttl __read_mostly
= IPDEFTTL
;
85 EXPORT_SYMBOL(sysctl_ip_default_ttl
);
87 /* Generate a checksum for an outgoing IP datagram. */
88 __inline__
void ip_send_check(struct iphdr
*iph
)
91 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
93 EXPORT_SYMBOL(ip_send_check
);
95 int __ip_local_out(struct sk_buff
*skb
)
97 struct iphdr
*iph
= ip_hdr(skb
);
99 iph
->tot_len
= htons(skb
->len
);
101 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, skb
, NULL
,
102 skb_dst(skb
)->dev
, dst_output
);
105 int ip_local_out(struct sk_buff
*skb
)
109 err
= __ip_local_out(skb
);
110 if (likely(err
== 1))
111 err
= dst_output(skb
);
115 EXPORT_SYMBOL_GPL(ip_local_out
);
117 /* dev_loopback_xmit for use with netfilter. */
118 static int ip_dev_loopback_xmit(struct sk_buff
*newskb
)
120 skb_reset_mac_header(newskb
);
121 __skb_pull(newskb
, skb_network_offset(newskb
));
122 newskb
->pkt_type
= PACKET_LOOPBACK
;
123 newskb
->ip_summed
= CHECKSUM_UNNECESSARY
;
124 WARN_ON(!skb_dst(newskb
));
129 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
131 int ttl
= inet
->uc_ttl
;
134 ttl
= ip4_dst_hoplimit(dst
);
139 * Add an ip header to a skbuff and send it out.
142 int ip_build_and_send_pkt(struct sk_buff
*skb
, struct sock
*sk
,
143 __be32 saddr
, __be32 daddr
, struct ip_options_rcu
*opt
)
145 struct inet_sock
*inet
= inet_sk(sk
);
146 struct rtable
*rt
= skb_rtable(skb
);
149 /* Build the IP header. */
150 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->opt
.optlen
: 0));
151 skb_reset_network_header(skb
);
155 iph
->tos
= inet
->tos
;
156 if (ip_dont_fragment(sk
, &rt
->dst
))
157 iph
->frag_off
= htons(IP_DF
);
160 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
161 iph
->daddr
= rt
->rt_dst
;
162 iph
->saddr
= rt
->rt_src
;
163 iph
->protocol
= sk
->sk_protocol
;
164 ip_select_ident(iph
, &rt
->dst
, sk
);
166 if (opt
&& opt
->opt
.optlen
) {
167 iph
->ihl
+= opt
->opt
.optlen
>>2;
168 ip_options_build(skb
, &opt
->opt
, daddr
, rt
, 0);
171 skb
->priority
= sk
->sk_priority
;
172 skb
->mark
= sk
->sk_mark
;
175 return ip_local_out(skb
);
177 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
179 static inline int ip_finish_output2(struct sk_buff
*skb
)
181 struct dst_entry
*dst
= skb_dst(skb
);
182 struct rtable
*rt
= (struct rtable
*)dst
;
183 struct net_device
*dev
= dst
->dev
;
184 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
186 if (rt
->rt_type
== RTN_MULTICAST
) {
187 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTMCAST
, skb
->len
);
188 } else if (rt
->rt_type
== RTN_BROADCAST
)
189 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTBCAST
, skb
->len
);
191 /* Be paranoid, rather than too clever. */
192 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
193 struct sk_buff
*skb2
;
195 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
201 skb_set_owner_w(skb2
, skb
->sk
);
207 return neigh_hh_output(dst
->hh
, skb
);
208 else if (dst
->neighbour
)
209 return dst
->neighbour
->output(skb
);
212 printk(KERN_DEBUG
"ip_finish_output2: No header cache and no neighbour!\n");
217 static inline int ip_skb_dst_mtu(struct sk_buff
*skb
)
219 struct inet_sock
*inet
= skb
->sk
? inet_sk(skb
->sk
) : NULL
;
221 return (inet
&& inet
->pmtudisc
== IP_PMTUDISC_PROBE
) ?
222 skb_dst(skb
)->dev
->mtu
: dst_mtu(skb_dst(skb
));
225 static int ip_finish_output(struct sk_buff
*skb
)
227 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
228 /* Policy lookup after SNAT yielded a new policy */
229 if (skb_dst(skb
)->xfrm
!= NULL
) {
230 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
231 return dst_output(skb
);
234 if (skb
->len
> ip_skb_dst_mtu(skb
) && !skb_is_gso(skb
))
235 return ip_fragment(skb
, ip_finish_output2
);
237 return ip_finish_output2(skb
);
240 int ip_mc_output(struct sk_buff
*skb
)
242 struct sock
*sk
= skb
->sk
;
243 struct rtable
*rt
= skb_rtable(skb
);
244 struct net_device
*dev
= rt
->dst
.dev
;
247 * If the indicated interface is up and running, send the packet.
249 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
252 skb
->protocol
= htons(ETH_P_IP
);
255 * Multicasts are looped back for other local users
258 if (rt
->rt_flags
&RTCF_MULTICAST
) {
260 #ifdef CONFIG_IP_MROUTE
261 /* Small optimization: do not loopback not local frames,
262 which returned after forwarding; they will be dropped
263 by ip_mr_input in any case.
264 Note, that local frames are looped back to be delivered
267 This check is duplicated in ip_mr_input at the moment.
270 ((rt
->rt_flags
& RTCF_LOCAL
) ||
271 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
274 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
276 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
277 newskb
, NULL
, newskb
->dev
,
278 ip_dev_loopback_xmit
);
281 /* Multicasts with ttl 0 must not go beyond the host */
283 if (ip_hdr(skb
)->ttl
== 0) {
289 if (rt
->rt_flags
&RTCF_BROADCAST
) {
290 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
292 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, newskb
,
293 NULL
, newskb
->dev
, ip_dev_loopback_xmit
);
296 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
,
297 skb
->dev
, ip_finish_output
,
298 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
301 int ip_output(struct sk_buff
*skb
)
303 struct net_device
*dev
= skb_dst(skb
)->dev
;
305 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
308 skb
->protocol
= htons(ETH_P_IP
);
310 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
, dev
,
312 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
315 int ip_queue_xmit(struct sk_buff
*skb
)
317 struct sock
*sk
= skb
->sk
;
318 struct inet_sock
*inet
= inet_sk(sk
);
319 struct ip_options_rcu
*inet_opt
;
324 /* Skip all of this if the packet is already routed,
325 * f.e. by something like SCTP.
328 inet_opt
= rcu_dereference(inet
->inet_opt
);
329 rt
= skb_rtable(skb
);
333 /* Make sure we can route this packet. */
334 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
338 /* Use correct destination address if we have options. */
339 daddr
= inet
->inet_daddr
;
340 if (inet_opt
&& inet_opt
->opt
.srr
)
341 daddr
= inet_opt
->opt
.faddr
;
343 /* If this fails, retransmit mechanism of transport layer will
344 * keep trying until route appears or the connection times
347 rt
= ip_route_output_ports(sock_net(sk
), sk
,
348 daddr
, inet
->inet_saddr
,
353 sk
->sk_bound_dev_if
);
356 sk_setup_caps(sk
, &rt
->dst
);
358 skb_dst_set_noref(skb
, &rt
->dst
);
361 if (inet_opt
&& inet_opt
->opt
.is_strictroute
&& rt
->rt_dst
!= rt
->rt_gateway
)
364 /* OK, we know where to send it, allocate and build IP header. */
365 skb_push(skb
, sizeof(struct iphdr
) + (inet_opt
? inet_opt
->opt
.optlen
: 0));
366 skb_reset_network_header(skb
);
368 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
369 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->local_df
)
370 iph
->frag_off
= htons(IP_DF
);
373 iph
->ttl
= ip_select_ttl(inet
, &rt
->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 (inet_opt
&& inet_opt
->opt
.optlen
) {
380 iph
->ihl
+= inet_opt
->opt
.optlen
>> 2;
381 ip_options_build(skb
, &inet_opt
->opt
, inet
->inet_daddr
, rt
, 0);
384 ip_select_ident_more(iph
, &rt
->dst
, sk
,
385 (skb_shinfo(skb
)->gso_segs
?: 1) - 1);
387 skb
->priority
= sk
->sk_priority
;
388 skb
->mark
= sk
->sk_mark
;
390 res
= ip_local_out(skb
);
396 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
398 return -EHOSTUNREACH
;
400 EXPORT_SYMBOL(ip_queue_xmit
);
403 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
405 to
->pkt_type
= from
->pkt_type
;
406 to
->priority
= from
->priority
;
407 to
->protocol
= from
->protocol
;
409 skb_dst_copy(to
, from
);
411 to
->mark
= from
->mark
;
413 /* Copy the flags to each fragment. */
414 IPCB(to
)->flags
= IPCB(from
)->flags
;
416 #ifdef CONFIG_NET_SCHED
417 to
->tc_index
= from
->tc_index
;
420 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
421 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
422 to
->nf_trace
= from
->nf_trace
;
424 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
425 to
->ipvs_property
= from
->ipvs_property
;
427 skb_copy_secmark(to
, from
);
431 * This IP datagram is too large to be sent in one piece. Break it up into
432 * smaller pieces (each of size equal to IP header plus
433 * a block of the data of the original IP data part) that will yet fit in a
434 * single device frame, and queue such a frame for sending.
437 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
441 struct net_device
*dev
;
442 struct sk_buff
*skb2
;
443 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
445 __be16 not_last_frag
;
446 struct rtable
*rt
= skb_rtable(skb
);
452 * Point into the IP datagram header.
457 if (unlikely((iph
->frag_off
& htons(IP_DF
)) && !skb
->local_df
)) {
458 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
459 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
460 htonl(ip_skb_dst_mtu(skb
)));
466 * Setup starting values.
470 mtu
= dst_mtu(&rt
->dst
) - hlen
; /* Size of data space */
471 #ifdef CONFIG_BRIDGE_NETFILTER
473 mtu
-= nf_bridge_mtu_reduction(skb
);
475 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
477 /* When frag_list is given, use it. First, check its validity:
478 * some transformers could create wrong frag_list or break existing
479 * one, it is not prohibited. In this case fall back to copying.
481 * LATER: this step can be merged to real generation of fragments,
482 * we can switch to copy when see the first bad fragment.
484 if (skb_has_frag_list(skb
)) {
485 struct sk_buff
*frag
, *frag2
;
486 int first_len
= skb_pagelen(skb
);
488 if (first_len
- hlen
> mtu
||
489 ((first_len
- hlen
) & 7) ||
490 (iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) ||
494 skb_walk_frags(skb
, frag
) {
495 /* Correct geometry. */
496 if (frag
->len
> mtu
||
497 ((frag
->len
& 7) && frag
->next
) ||
498 skb_headroom(frag
) < hlen
)
499 goto slow_path_clean
;
501 /* Partially cloned skb? */
502 if (skb_shared(frag
))
503 goto slow_path_clean
;
508 frag
->destructor
= sock_wfree
;
510 skb
->truesize
-= frag
->truesize
;
513 /* Everything is OK. Generate! */
517 frag
= skb_shinfo(skb
)->frag_list
;
518 skb_frag_list_init(skb
);
519 skb
->data_len
= first_len
- skb_headlen(skb
);
520 skb
->len
= first_len
;
521 iph
->tot_len
= htons(first_len
);
522 iph
->frag_off
= htons(IP_MF
);
526 /* Prepare header of the next frame,
527 * before previous one went down. */
529 frag
->ip_summed
= CHECKSUM_NONE
;
530 skb_reset_transport_header(frag
);
531 __skb_push(frag
, hlen
);
532 skb_reset_network_header(frag
);
533 memcpy(skb_network_header(frag
), iph
, hlen
);
535 iph
->tot_len
= htons(frag
->len
);
536 ip_copy_metadata(frag
, skb
);
538 ip_options_fragment(frag
);
539 offset
+= skb
->len
- hlen
;
540 iph
->frag_off
= htons(offset
>>3);
541 if (frag
->next
!= NULL
)
542 iph
->frag_off
|= htons(IP_MF
);
543 /* Ready, complete checksum */
550 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
560 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
569 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
573 skb_walk_frags(skb
, frag2
) {
577 frag2
->destructor
= NULL
;
578 skb
->truesize
+= frag2
->truesize
;
583 left
= skb
->len
- hlen
; /* Space per frame */
584 ptr
= hlen
; /* Where to start from */
586 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
587 * we need to make room for the encapsulating header
589 ll_rs
= LL_RESERVED_SPACE_EXTRA(rt
->dst
.dev
, nf_bridge_pad(skb
));
592 * Fragment the datagram.
595 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
596 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
599 * Keep copying data until we run out.
604 /* IF: it doesn't fit, use 'mtu' - the data space left */
607 /* IF: we are not sending up to and including the packet end
608 then align the next start on an eight byte boundary */
616 if ((skb2
= alloc_skb(len
+hlen
+ll_rs
, GFP_ATOMIC
)) == NULL
) {
617 NETDEBUG(KERN_INFO
"IP: frag: no memory for new fragment!\n");
623 * Set up data on packet
626 ip_copy_metadata(skb2
, skb
);
627 skb_reserve(skb2
, ll_rs
);
628 skb_put(skb2
, len
+ hlen
);
629 skb_reset_network_header(skb2
);
630 skb2
->transport_header
= skb2
->network_header
+ hlen
;
633 * Charge the memory for the fragment to any owner
638 skb_set_owner_w(skb2
, skb
->sk
);
641 * Copy the packet header into the new buffer.
644 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
647 * Copy a block of the IP datagram.
649 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
654 * Fill in the new header fields.
657 iph
->frag_off
= htons((offset
>> 3));
659 /* ANK: dirty, but effective trick. Upgrade options only if
660 * the segment to be fragmented was THE FIRST (otherwise,
661 * options are already fixed) and make it ONCE
662 * on the initial skb, so that all the following fragments
663 * will inherit fixed options.
666 ip_options_fragment(skb
);
669 * Added AC : If we are fragmenting a fragment that's not the
670 * last fragment then keep MF on each bit
672 if (left
> 0 || not_last_frag
)
673 iph
->frag_off
|= htons(IP_MF
);
678 * Put this fragment into the sending queue.
680 iph
->tot_len
= htons(len
+ hlen
);
688 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
691 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
696 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
699 EXPORT_SYMBOL(ip_fragment
);
702 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
704 struct iovec
*iov
= from
;
706 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
707 if (memcpy_fromiovecend(to
, iov
, offset
, len
) < 0)
711 if (csum_partial_copy_fromiovecend(to
, iov
, offset
, len
, &csum
) < 0)
713 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
717 EXPORT_SYMBOL(ip_generic_getfrag
);
720 csum_page(struct page
*page
, int offset
, int copy
)
725 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
730 static inline int ip_ufo_append_data(struct sock
*sk
,
731 struct sk_buff_head
*queue
,
732 int getfrag(void *from
, char *to
, int offset
, int len
,
733 int odd
, struct sk_buff
*skb
),
734 void *from
, int length
, int hh_len
, int fragheaderlen
,
735 int transhdrlen
, int mtu
, unsigned int flags
)
740 /* There is support for UDP fragmentation offload by network
741 * device, so create one single skb packet containing complete
744 if ((skb
= skb_peek_tail(queue
)) == NULL
) {
745 skb
= sock_alloc_send_skb(sk
,
746 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
747 (flags
& MSG_DONTWAIT
), &err
);
752 /* reserve space for Hardware header */
753 skb_reserve(skb
, hh_len
);
755 /* create space for UDP/IP header */
756 skb_put(skb
, fragheaderlen
+ transhdrlen
);
758 /* initialize network header pointer */
759 skb_reset_network_header(skb
);
761 /* initialize protocol header pointer */
762 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
764 skb
->ip_summed
= CHECKSUM_PARTIAL
;
767 /* specify the length of each IP datagram fragment */
768 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
769 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
770 __skb_queue_tail(queue
, skb
);
773 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
774 (length
- transhdrlen
));
777 static int __ip_append_data(struct sock
*sk
, struct sk_buff_head
*queue
,
778 struct inet_cork
*cork
,
779 int getfrag(void *from
, char *to
, int offset
,
780 int len
, int odd
, struct sk_buff
*skb
),
781 void *from
, int length
, int transhdrlen
,
784 struct inet_sock
*inet
= inet_sk(sk
);
787 struct ip_options
*opt
= cork
->opt
;
794 unsigned int maxfraglen
, fragheaderlen
;
795 int csummode
= CHECKSUM_NONE
;
796 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
798 exthdrlen
= transhdrlen
? rt
->dst
.header_len
: 0;
800 transhdrlen
+= exthdrlen
;
801 mtu
= cork
->fragsize
;
803 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
805 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
806 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
808 if (cork
->length
+ length
> 0xFFFF - fragheaderlen
) {
809 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->inet_dport
,
815 * transhdrlen > 0 means that this is the first fragment and we wish
816 * it won't be fragmented in the future.
819 length
+ fragheaderlen
<= mtu
&&
820 rt
->dst
.dev
->features
& NETIF_F_V4_CSUM
&&
822 csummode
= CHECKSUM_PARTIAL
;
824 skb
= skb_peek_tail(queue
);
826 cork
->length
+= length
;
827 if (((length
> mtu
) || (skb
&& skb_is_gso(skb
))) &&
828 (sk
->sk_protocol
== IPPROTO_UDP
) &&
829 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
830 err
= ip_ufo_append_data(sk
, queue
, getfrag
, from
, length
,
831 hh_len
, fragheaderlen
, transhdrlen
,
838 /* So, what's going on in the loop below?
840 * We use calculated fragment length to generate chained skb,
841 * each of segments is IP fragment ready for sending to network after
842 * adding appropriate IP header.
849 /* Check if the remaining data fits into current packet. */
850 copy
= mtu
- skb
->len
;
852 copy
= maxfraglen
- skb
->len
;
855 unsigned int datalen
;
856 unsigned int fraglen
;
857 unsigned int fraggap
;
858 unsigned int alloclen
;
859 struct sk_buff
*skb_prev
;
863 fraggap
= skb_prev
->len
- maxfraglen
;
868 * If remaining data exceeds the mtu,
869 * we know we need more fragment(s).
871 datalen
= length
+ fraggap
;
872 if (datalen
> mtu
- fragheaderlen
)
873 datalen
= maxfraglen
- fragheaderlen
;
874 fraglen
= datalen
+ fragheaderlen
;
876 if ((flags
& MSG_MORE
) &&
877 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
882 /* The last fragment gets additional space at tail.
883 * Note, with MSG_MORE we overallocate on fragments,
884 * because we have no idea what fragment will be
887 if (datalen
== length
+ fraggap
) {
888 alloclen
+= rt
->dst
.trailer_len
;
889 /* make sure mtu is not reached */
890 if (datalen
> mtu
- fragheaderlen
- rt
->dst
.trailer_len
)
891 datalen
-= ALIGN(rt
->dst
.trailer_len
, 8);
894 skb
= sock_alloc_send_skb(sk
,
895 alloclen
+ hh_len
+ 15,
896 (flags
& MSG_DONTWAIT
), &err
);
899 if (atomic_read(&sk
->sk_wmem_alloc
) <=
901 skb
= sock_wmalloc(sk
,
902 alloclen
+ hh_len
+ 15, 1,
904 if (unlikely(skb
== NULL
))
907 /* only the initial fragment is
915 * Fill in the control structures
917 skb
->ip_summed
= csummode
;
919 skb_reserve(skb
, hh_len
);
920 skb_shinfo(skb
)->tx_flags
= cork
->tx_flags
;
923 * Find where to start putting bytes.
925 data
= skb_put(skb
, fraglen
);
926 skb_set_network_header(skb
, exthdrlen
);
927 skb
->transport_header
= (skb
->network_header
+
929 data
+= fragheaderlen
;
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 pskb_trim_unique(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(queue
, skb
);
964 if (!(rt
->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
= cork
->page
;
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
, 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);
1004 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1005 frag
= &skb_shinfo(skb
)->frags
[i
];
1010 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1017 skb
->data_len
+= copy
;
1018 skb
->truesize
+= copy
;
1019 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1028 cork
->length
-= length
;
1029 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1033 static int ip_setup_cork(struct sock
*sk
, struct inet_cork
*cork
,
1034 struct ipcm_cookie
*ipc
, struct rtable
**rtp
)
1036 struct inet_sock
*inet
= inet_sk(sk
);
1037 struct ip_options_rcu
*opt
;
1041 * setup for corking.
1045 if (cork
->opt
== NULL
) {
1046 cork
->opt
= kmalloc(sizeof(struct ip_options
) + 40,
1048 if (unlikely(cork
->opt
== NULL
))
1051 memcpy(cork
->opt
, &opt
->opt
, sizeof(struct ip_options
) + opt
->opt
.optlen
);
1052 cork
->flags
|= IPCORK_OPT
;
1053 cork
->addr
= ipc
->addr
;
1059 * We steal reference to this route, caller should not release it
1062 cork
->fragsize
= inet
->pmtudisc
== IP_PMTUDISC_PROBE
?
1063 rt
->dst
.dev
->mtu
: dst_mtu(rt
->dst
.path
);
1064 cork
->dst
= &rt
->dst
;
1066 cork
->tx_flags
= ipc
->tx_flags
;
1074 * ip_append_data() and ip_append_page() can make one large IP datagram
1075 * from many pieces of data. Each pieces will be holded on the socket
1076 * until ip_push_pending_frames() is called. Each piece can be a page
1079 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1080 * this interface potentially.
1082 * LATER: length must be adjusted by pad at tail, when it is required.
1084 int ip_append_data(struct sock
*sk
,
1085 int getfrag(void *from
, char *to
, int offset
, int len
,
1086 int odd
, struct sk_buff
*skb
),
1087 void *from
, int length
, int transhdrlen
,
1088 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1091 struct inet_sock
*inet
= inet_sk(sk
);
1094 if (flags
&MSG_PROBE
)
1097 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1098 err
= ip_setup_cork(sk
, &inet
->cork
, ipc
, rtp
);
1105 return __ip_append_data(sk
, &sk
->sk_write_queue
, &inet
->cork
, getfrag
,
1106 from
, length
, transhdrlen
, flags
);
1109 ssize_t
ip_append_page(struct sock
*sk
, struct page
*page
,
1110 int offset
, size_t size
, int flags
)
1112 struct inet_sock
*inet
= inet_sk(sk
);
1113 struct sk_buff
*skb
;
1115 struct ip_options
*opt
= NULL
;
1120 unsigned int maxfraglen
, fragheaderlen
, fraggap
;
1125 if (flags
&MSG_PROBE
)
1128 if (skb_queue_empty(&sk
->sk_write_queue
))
1131 rt
= (struct rtable
*)inet
->cork
.dst
;
1132 if (inet
->cork
.flags
& IPCORK_OPT
)
1133 opt
= inet
->cork
.opt
;
1135 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1138 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1139 mtu
= inet
->cork
.fragsize
;
1141 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1142 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1144 if (inet
->cork
.length
+ size
> 0xFFFF - fragheaderlen
) {
1145 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->inet_dport
, mtu
);
1149 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1152 inet
->cork
.length
+= size
;
1153 if ((size
+ skb
->len
> mtu
) &&
1154 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1155 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
1156 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1157 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1164 if (skb_is_gso(skb
))
1168 /* Check if the remaining data fits into current packet. */
1169 len
= mtu
- skb
->len
;
1171 len
= maxfraglen
- skb
->len
;
1174 struct sk_buff
*skb_prev
;
1178 fraggap
= skb_prev
->len
- maxfraglen
;
1180 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1181 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1182 if (unlikely(!skb
)) {
1188 * Fill in the control structures
1190 skb
->ip_summed
= CHECKSUM_NONE
;
1192 skb_reserve(skb
, hh_len
);
1195 * Find where to start putting bytes.
1197 skb_put(skb
, fragheaderlen
+ fraggap
);
1198 skb_reset_network_header(skb
);
1199 skb
->transport_header
= (skb
->network_header
+
1202 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1204 skb_transport_header(skb
),
1206 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1208 pskb_trim_unique(skb_prev
, maxfraglen
);
1212 * Put the packet on the pending queue.
1214 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1218 i
= skb_shinfo(skb
)->nr_frags
;
1221 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1222 skb_shinfo(skb
)->frags
[i
-1].size
+= len
;
1223 } else if (i
< MAX_SKB_FRAGS
) {
1225 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1231 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1233 csum
= csum_page(page
, offset
, len
);
1234 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1238 skb
->data_len
+= len
;
1239 skb
->truesize
+= len
;
1240 atomic_add(len
, &sk
->sk_wmem_alloc
);
1247 inet
->cork
.length
-= size
;
1248 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1252 static void ip_cork_release(struct inet_cork
*cork
)
1254 cork
->flags
&= ~IPCORK_OPT
;
1257 dst_release(cork
->dst
);
1262 * Combined all pending IP fragments on the socket as one IP datagram
1263 * and push them out.
1265 struct sk_buff
*__ip_make_skb(struct sock
*sk
,
1266 struct sk_buff_head
*queue
,
1267 struct inet_cork
*cork
)
1269 struct sk_buff
*skb
, *tmp_skb
;
1270 struct sk_buff
**tail_skb
;
1271 struct inet_sock
*inet
= inet_sk(sk
);
1272 struct net
*net
= sock_net(sk
);
1273 struct ip_options
*opt
= NULL
;
1274 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
1279 if ((skb
= __skb_dequeue(queue
)) == NULL
)
1281 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1283 /* move skb->data to ip header from ext header */
1284 if (skb
->data
< skb_network_header(skb
))
1285 __skb_pull(skb
, skb_network_offset(skb
));
1286 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1287 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1288 *tail_skb
= tmp_skb
;
1289 tail_skb
= &(tmp_skb
->next
);
1290 skb
->len
+= tmp_skb
->len
;
1291 skb
->data_len
+= tmp_skb
->len
;
1292 skb
->truesize
+= tmp_skb
->truesize
;
1293 tmp_skb
->destructor
= NULL
;
1297 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1298 * to fragment the frame generated here. No matter, what transforms
1299 * how transforms change size of the packet, it will come out.
1301 if (inet
->pmtudisc
< IP_PMTUDISC_DO
)
1304 /* DF bit is set when we want to see DF on outgoing frames.
1305 * If local_df is set too, we still allow to fragment this frame
1307 if (inet
->pmtudisc
>= IP_PMTUDISC_DO
||
1308 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1309 ip_dont_fragment(sk
, &rt
->dst
)))
1312 if (cork
->flags
& IPCORK_OPT
)
1315 if (rt
->rt_type
== RTN_MULTICAST
)
1318 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1320 iph
= (struct iphdr
*)skb
->data
;
1324 iph
->ihl
+= opt
->optlen
>>2;
1325 ip_options_build(skb
, opt
, cork
->addr
, rt
, 0);
1327 iph
->tos
= inet
->tos
;
1329 ip_select_ident(iph
, &rt
->dst
, sk
);
1331 iph
->protocol
= sk
->sk_protocol
;
1332 iph
->saddr
= rt
->rt_src
;
1333 iph
->daddr
= rt
->rt_dst
;
1335 skb
->priority
= sk
->sk_priority
;
1336 skb
->mark
= sk
->sk_mark
;
1338 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1342 skb_dst_set(skb
, &rt
->dst
);
1344 if (iph
->protocol
== IPPROTO_ICMP
)
1345 icmp_out_count(net
, ((struct icmphdr
*)
1346 skb_transport_header(skb
))->type
);
1348 ip_cork_release(cork
);
1353 int ip_send_skb(struct sk_buff
*skb
)
1355 struct net
*net
= sock_net(skb
->sk
);
1358 err
= ip_local_out(skb
);
1361 err
= net_xmit_errno(err
);
1363 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1369 int ip_push_pending_frames(struct sock
*sk
)
1371 struct sk_buff
*skb
;
1373 skb
= ip_finish_skb(sk
);
1377 /* Netfilter gets whole the not fragmented skb. */
1378 return ip_send_skb(skb
);
1382 * Throw away all pending data on the socket.
1384 static void __ip_flush_pending_frames(struct sock
*sk
,
1385 struct sk_buff_head
*queue
,
1386 struct inet_cork
*cork
)
1388 struct sk_buff
*skb
;
1390 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
)
1393 ip_cork_release(cork
);
1396 void ip_flush_pending_frames(struct sock
*sk
)
1398 __ip_flush_pending_frames(sk
, &sk
->sk_write_queue
, &inet_sk(sk
)->cork
);
1401 struct sk_buff
*ip_make_skb(struct sock
*sk
,
1402 int getfrag(void *from
, char *to
, int offset
,
1403 int len
, int odd
, struct sk_buff
*skb
),
1404 void *from
, int length
, int transhdrlen
,
1405 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1408 struct inet_cork cork
= {};
1409 struct sk_buff_head queue
;
1412 if (flags
& MSG_PROBE
)
1415 __skb_queue_head_init(&queue
);
1417 err
= ip_setup_cork(sk
, &cork
, ipc
, rtp
);
1419 return ERR_PTR(err
);
1421 err
= __ip_append_data(sk
, &queue
, &cork
, getfrag
,
1422 from
, length
, transhdrlen
, flags
);
1424 __ip_flush_pending_frames(sk
, &queue
, &cork
);
1425 return ERR_PTR(err
);
1428 return __ip_make_skb(sk
, &queue
, &cork
);
1432 * Fetch data from kernel space and fill in checksum if needed.
1434 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1435 int len
, int odd
, struct sk_buff
*skb
)
1439 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1440 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1445 * Generic function to send a packet as reply to another packet.
1446 * Used to send TCP resets so far. ICMP should use this function too.
1448 * Should run single threaded per socket because it uses the sock
1449 * structure to pass arguments.
1451 void ip_send_reply(struct sock
*sk
, struct sk_buff
*skb
, struct ip_reply_arg
*arg
,
1454 struct inet_sock
*inet
= inet_sk(sk
);
1455 struct ip_options_data replyopts
;
1456 struct ipcm_cookie ipc
;
1458 struct rtable
*rt
= skb_rtable(skb
);
1460 if (ip_options_echo(&replyopts
.opt
.opt
, skb
))
1463 daddr
= ipc
.addr
= rt
->rt_src
;
1467 if (replyopts
.opt
.opt
.optlen
) {
1468 ipc
.opt
= &replyopts
.opt
;
1470 if (replyopts
.opt
.opt
.srr
)
1471 daddr
= replyopts
.opt
.opt
.faddr
;
1477 flowi4_init_output(&fl4
, arg
->bound_dev_if
, 0,
1478 RT_TOS(ip_hdr(skb
)->tos
),
1479 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
1480 ip_reply_arg_flowi_flags(arg
),
1481 daddr
, rt
->rt_spec_dst
,
1482 tcp_hdr(skb
)->source
, tcp_hdr(skb
)->dest
);
1483 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
1484 rt
= ip_route_output_key(sock_net(sk
), &fl4
);
1489 /* And let IP do all the hard work.
1491 This chunk is not reenterable, hence spinlock.
1492 Note that it uses the fact, that this function is called
1493 with locally disabled BH and that sk cannot be already spinlocked.
1496 inet
->tos
= ip_hdr(skb
)->tos
;
1497 sk
->sk_priority
= skb
->priority
;
1498 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1499 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1500 ip_append_data(sk
, ip_reply_glue_bits
, arg
->iov
->iov_base
, len
, 0,
1501 &ipc
, &rt
, MSG_DONTWAIT
);
1502 if ((skb
= skb_peek(&sk
->sk_write_queue
)) != NULL
) {
1503 if (arg
->csumoffset
>= 0)
1504 *((__sum16
*)skb_transport_header(skb
) +
1505 arg
->csumoffset
) = csum_fold(csum_add(skb
->csum
,
1507 skb
->ip_summed
= CHECKSUM_NONE
;
1508 ip_push_pending_frames(sk
);
1516 void __init
ip_init(void)
1521 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1522 igmp_mc_proc_init();