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 <linux/module.h>
47 #include <linux/types.h>
48 #include <linux/kernel.h>
50 #include <linux/string.h>
51 #include <linux/errno.h>
52 #include <linux/highmem.h>
53 #include <linux/slab.h>
55 #include <linux/socket.h>
56 #include <linux/sockios.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>
67 #include <net/protocol.h>
68 #include <net/route.h>
70 #include <linux/skbuff.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
;
84 EXPORT_SYMBOL(sysctl_ip_default_ttl
);
86 /* Generate a checksum for an outgoing IP datagram. */
87 void ip_send_check(struct iphdr
*iph
)
90 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
92 EXPORT_SYMBOL(ip_send_check
);
94 int __ip_local_out(struct sk_buff
*skb
)
96 struct iphdr
*iph
= ip_hdr(skb
);
98 iph
->tot_len
= htons(skb
->len
);
100 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, skb
, NULL
,
101 skb_dst(skb
)->dev
, dst_output
);
104 int ip_local_out(struct sk_buff
*skb
)
108 err
= __ip_local_out(skb
);
109 if (likely(err
== 1))
110 err
= dst_output(skb
);
114 EXPORT_SYMBOL_GPL(ip_local_out
);
116 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
118 int ttl
= inet
->uc_ttl
;
121 ttl
= ip4_dst_hoplimit(dst
);
126 * Add an ip header to a skbuff and send it out.
129 int ip_build_and_send_pkt(struct sk_buff
*skb
, struct sock
*sk
,
130 __be32 saddr
, __be32 daddr
, struct ip_options_rcu
*opt
)
132 struct inet_sock
*inet
= inet_sk(sk
);
133 struct rtable
*rt
= skb_rtable(skb
);
136 /* Build the IP header. */
137 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->opt
.optlen
: 0));
138 skb_reset_network_header(skb
);
142 iph
->tos
= inet
->tos
;
143 if (ip_dont_fragment(sk
, &rt
->dst
))
144 iph
->frag_off
= htons(IP_DF
);
147 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
148 iph
->daddr
= (opt
&& opt
->opt
.srr
? opt
->opt
.faddr
: daddr
);
150 iph
->protocol
= sk
->sk_protocol
;
151 ip_select_ident(skb
, &rt
->dst
, sk
);
153 if (opt
&& opt
->opt
.optlen
) {
154 iph
->ihl
+= opt
->opt
.optlen
>>2;
155 ip_options_build(skb
, &opt
->opt
, daddr
, rt
, 0);
158 skb
->priority
= sk
->sk_priority
;
159 skb
->mark
= sk
->sk_mark
;
162 return ip_local_out(skb
);
164 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
166 static inline int ip_finish_output2(struct sk_buff
*skb
)
168 struct dst_entry
*dst
= skb_dst(skb
);
169 struct rtable
*rt
= (struct rtable
*)dst
;
170 struct net_device
*dev
= dst
->dev
;
171 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
172 struct neighbour
*neigh
;
175 if (rt
->rt_type
== RTN_MULTICAST
) {
176 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTMCAST
, skb
->len
);
177 } else if (rt
->rt_type
== RTN_BROADCAST
)
178 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTBCAST
, skb
->len
);
180 /* Be paranoid, rather than too clever. */
181 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
182 struct sk_buff
*skb2
;
184 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
190 skb_set_owner_w(skb2
, skb
->sk
);
196 nexthop
= (__force u32
) rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
197 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
198 if (unlikely(!neigh
))
199 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
200 if (!IS_ERR(neigh
)) {
201 int res
= dst_neigh_output(dst
, neigh
, skb
);
203 rcu_read_unlock_bh();
206 rcu_read_unlock_bh();
208 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
214 static int ip_finish_output(struct sk_buff
*skb
)
216 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
217 /* Policy lookup after SNAT yielded a new policy */
218 if (skb_dst(skb
)->xfrm
!= NULL
) {
219 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
220 return dst_output(skb
);
223 if (skb
->len
> ip_skb_dst_mtu(skb
) && !skb_is_gso(skb
))
224 return ip_fragment(skb
, ip_finish_output2
);
226 return ip_finish_output2(skb
);
229 int ip_mc_output(struct sk_buff
*skb
)
231 struct sock
*sk
= skb
->sk
;
232 struct rtable
*rt
= skb_rtable(skb
);
233 struct net_device
*dev
= rt
->dst
.dev
;
236 * If the indicated interface is up and running, send the packet.
238 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
241 skb
->protocol
= htons(ETH_P_IP
);
244 * Multicasts are looped back for other local users
247 if (rt
->rt_flags
&RTCF_MULTICAST
) {
249 #ifdef CONFIG_IP_MROUTE
250 /* Small optimization: do not loopback not local frames,
251 which returned after forwarding; they will be dropped
252 by ip_mr_input in any case.
253 Note, that local frames are looped back to be delivered
256 This check is duplicated in ip_mr_input at the moment.
259 ((rt
->rt_flags
& RTCF_LOCAL
) ||
260 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
263 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
265 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
266 newskb
, NULL
, newskb
->dev
,
270 /* Multicasts with ttl 0 must not go beyond the host */
272 if (ip_hdr(skb
)->ttl
== 0) {
278 if (rt
->rt_flags
&RTCF_BROADCAST
) {
279 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
281 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, newskb
,
282 NULL
, newskb
->dev
, dev_loopback_xmit
);
285 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
,
286 skb
->dev
, ip_finish_output
,
287 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
290 int ip_output(struct sk_buff
*skb
)
292 struct net_device
*dev
= skb_dst(skb
)->dev
;
294 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
297 skb
->protocol
= htons(ETH_P_IP
);
299 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
, dev
,
301 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
305 * copy saddr and daddr, possibly using 64bit load/stores
307 * iph->saddr = fl4->saddr;
308 * iph->daddr = fl4->daddr;
310 static void ip_copy_addrs(struct iphdr
*iph
, const struct flowi4
*fl4
)
312 BUILD_BUG_ON(offsetof(typeof(*fl4
), daddr
) !=
313 offsetof(typeof(*fl4
), saddr
) + sizeof(fl4
->saddr
));
314 memcpy(&iph
->saddr
, &fl4
->saddr
,
315 sizeof(fl4
->saddr
) + sizeof(fl4
->daddr
));
318 int ip_queue_xmit(struct sk_buff
*skb
, struct flowi
*fl
)
320 struct sock
*sk
= skb
->sk
;
321 struct inet_sock
*inet
= inet_sk(sk
);
322 struct ip_options_rcu
*inet_opt
;
328 /* Skip all of this if the packet is already routed,
329 * f.e. by something like SCTP.
332 inet_opt
= rcu_dereference(inet
->inet_opt
);
334 rt
= skb_rtable(skb
);
338 /* Make sure we can route this packet. */
339 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
343 /* Use correct destination address if we have options. */
344 daddr
= inet
->inet_daddr
;
345 if (inet_opt
&& inet_opt
->opt
.srr
)
346 daddr
= inet_opt
->opt
.faddr
;
348 /* If this fails, retransmit mechanism of transport layer will
349 * keep trying until route appears or the connection times
352 rt
= ip_route_output_ports(sock_net(sk
), fl4
, sk
,
353 daddr
, inet
->inet_saddr
,
358 sk
->sk_bound_dev_if
);
361 sk_setup_caps(sk
, &rt
->dst
);
363 skb_dst_set_noref(skb
, &rt
->dst
);
366 if (inet_opt
&& inet_opt
->opt
.is_strictroute
&& rt
->rt_uses_gateway
)
369 /* OK, we know where to send it, allocate and build IP header. */
370 skb_push(skb
, sizeof(struct iphdr
) + (inet_opt
? inet_opt
->opt
.optlen
: 0));
371 skb_reset_network_header(skb
);
373 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
374 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->local_df
)
375 iph
->frag_off
= htons(IP_DF
);
378 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
379 iph
->protocol
= sk
->sk_protocol
;
380 ip_copy_addrs(iph
, fl4
);
382 /* Transport layer set skb->h.foo itself. */
384 if (inet_opt
&& inet_opt
->opt
.optlen
) {
385 iph
->ihl
+= inet_opt
->opt
.optlen
>> 2;
386 ip_options_build(skb
, &inet_opt
->opt
, inet
->inet_daddr
, rt
, 0);
389 ip_select_ident_more(skb
, &rt
->dst
, sk
,
390 (skb_shinfo(skb
)->gso_segs
?: 1) - 1);
392 skb
->priority
= sk
->sk_priority
;
393 skb
->mark
= sk
->sk_mark
;
395 res
= ip_local_out(skb
);
401 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
403 return -EHOSTUNREACH
;
405 EXPORT_SYMBOL(ip_queue_xmit
);
408 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
410 to
->pkt_type
= from
->pkt_type
;
411 to
->priority
= from
->priority
;
412 to
->protocol
= from
->protocol
;
414 skb_dst_copy(to
, from
);
416 to
->mark
= from
->mark
;
418 /* Copy the flags to each fragment. */
419 IPCB(to
)->flags
= IPCB(from
)->flags
;
421 #ifdef CONFIG_NET_SCHED
422 to
->tc_index
= from
->tc_index
;
425 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
426 to
->nf_trace
= from
->nf_trace
;
428 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
429 to
->ipvs_property
= from
->ipvs_property
;
431 skb_copy_secmark(to
, from
);
435 * This IP datagram is too large to be sent in one piece. Break it up into
436 * smaller pieces (each of size equal to IP header plus
437 * a block of the data of the original IP data part) that will yet fit in a
438 * single device frame, and queue such a frame for sending.
441 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
445 struct net_device
*dev
;
446 struct sk_buff
*skb2
;
447 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
449 __be16 not_last_frag
;
450 struct rtable
*rt
= skb_rtable(skb
);
456 * Point into the IP datagram header.
461 if (unlikely(((iph
->frag_off
& htons(IP_DF
)) && !skb
->local_df
) ||
462 (IPCB(skb
)->frag_max_size
&&
463 IPCB(skb
)->frag_max_size
> dst_mtu(&rt
->dst
)))) {
464 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
465 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
466 htonl(ip_skb_dst_mtu(skb
)));
472 * Setup starting values.
476 mtu
= dst_mtu(&rt
->dst
) - hlen
; /* Size of data space */
477 #ifdef CONFIG_BRIDGE_NETFILTER
479 mtu
-= nf_bridge_mtu_reduction(skb
);
481 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
483 /* When frag_list is given, use it. First, check its validity:
484 * some transformers could create wrong frag_list or break existing
485 * one, it is not prohibited. In this case fall back to copying.
487 * LATER: this step can be merged to real generation of fragments,
488 * we can switch to copy when see the first bad fragment.
490 if (skb_has_frag_list(skb
)) {
491 struct sk_buff
*frag
, *frag2
;
492 int first_len
= skb_pagelen(skb
);
494 if (first_len
- hlen
> mtu
||
495 ((first_len
- hlen
) & 7) ||
496 ip_is_fragment(iph
) ||
500 skb_walk_frags(skb
, frag
) {
501 /* Correct geometry. */
502 if (frag
->len
> mtu
||
503 ((frag
->len
& 7) && frag
->next
) ||
504 skb_headroom(frag
) < hlen
)
505 goto slow_path_clean
;
507 /* Partially cloned skb? */
508 if (skb_shared(frag
))
509 goto slow_path_clean
;
514 frag
->destructor
= sock_wfree
;
516 skb
->truesize
-= frag
->truesize
;
519 /* Everything is OK. Generate! */
523 frag
= skb_shinfo(skb
)->frag_list
;
524 skb_frag_list_init(skb
);
525 skb
->data_len
= first_len
- skb_headlen(skb
);
526 skb
->len
= first_len
;
527 iph
->tot_len
= htons(first_len
);
528 iph
->frag_off
= htons(IP_MF
);
532 /* Prepare header of the next frame,
533 * before previous one went down. */
535 frag
->ip_summed
= CHECKSUM_NONE
;
536 skb_reset_transport_header(frag
);
537 __skb_push(frag
, hlen
);
538 skb_reset_network_header(frag
);
539 memcpy(skb_network_header(frag
), iph
, hlen
);
541 iph
->tot_len
= htons(frag
->len
);
542 ip_copy_metadata(frag
, skb
);
544 ip_options_fragment(frag
);
545 offset
+= skb
->len
- hlen
;
546 iph
->frag_off
= htons(offset
>>3);
547 if (frag
->next
!= NULL
)
548 iph
->frag_off
|= htons(IP_MF
);
549 /* Ready, complete checksum */
556 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
566 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
575 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
579 skb_walk_frags(skb
, frag2
) {
583 frag2
->destructor
= NULL
;
584 skb
->truesize
+= frag2
->truesize
;
589 /* for offloaded checksums cleanup checksum before fragmentation */
590 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) && skb_checksum_help(skb
))
594 left
= skb
->len
- hlen
; /* Space per frame */
595 ptr
= hlen
; /* Where to start from */
597 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
598 * we need to make room for the encapsulating header
600 ll_rs
= LL_RESERVED_SPACE_EXTRA(rt
->dst
.dev
, nf_bridge_pad(skb
));
603 * Fragment the datagram.
606 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
607 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
610 * Keep copying data until we run out.
615 /* IF: it doesn't fit, use 'mtu' - the data space left */
618 /* IF: we are not sending up to and including the packet end
619 then align the next start on an eight byte boundary */
627 if ((skb2
= alloc_skb(len
+hlen
+ll_rs
, GFP_ATOMIC
)) == NULL
) {
628 NETDEBUG(KERN_INFO
"IP: frag: no memory for new fragment!\n");
634 * Set up data on packet
637 ip_copy_metadata(skb2
, skb
);
638 skb_reserve(skb2
, ll_rs
);
639 skb_put(skb2
, len
+ hlen
);
640 skb_reset_network_header(skb2
);
641 skb2
->transport_header
= skb2
->network_header
+ hlen
;
644 * Charge the memory for the fragment to any owner
649 skb_set_owner_w(skb2
, skb
->sk
);
652 * Copy the packet header into the new buffer.
655 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
658 * Copy a block of the IP datagram.
660 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
665 * Fill in the new header fields.
668 iph
->frag_off
= htons((offset
>> 3));
670 /* ANK: dirty, but effective trick. Upgrade options only if
671 * the segment to be fragmented was THE FIRST (otherwise,
672 * options are already fixed) and make it ONCE
673 * on the initial skb, so that all the following fragments
674 * will inherit fixed options.
677 ip_options_fragment(skb
);
680 * Added AC : If we are fragmenting a fragment that's not the
681 * last fragment then keep MF on each bit
683 if (left
> 0 || not_last_frag
)
684 iph
->frag_off
|= htons(IP_MF
);
689 * Put this fragment into the sending queue.
691 iph
->tot_len
= htons(len
+ hlen
);
699 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
702 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
707 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
710 EXPORT_SYMBOL(ip_fragment
);
713 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
715 struct iovec
*iov
= from
;
717 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
718 if (memcpy_fromiovecend(to
, iov
, offset
, len
) < 0)
722 if (csum_partial_copy_fromiovecend(to
, iov
, offset
, len
, &csum
) < 0)
724 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
728 EXPORT_SYMBOL(ip_generic_getfrag
);
731 csum_page(struct page
*page
, int offset
, int copy
)
736 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
741 static inline int ip_ufo_append_data(struct sock
*sk
,
742 struct sk_buff_head
*queue
,
743 int getfrag(void *from
, char *to
, int offset
, int len
,
744 int odd
, struct sk_buff
*skb
),
745 void *from
, int length
, int hh_len
, int fragheaderlen
,
746 int transhdrlen
, int maxfraglen
, unsigned int flags
)
751 /* There is support for UDP fragmentation offload by network
752 * device, so create one single skb packet containing complete
755 if ((skb
= skb_peek_tail(queue
)) == NULL
) {
756 skb
= sock_alloc_send_skb(sk
,
757 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
758 (flags
& MSG_DONTWAIT
), &err
);
763 /* reserve space for Hardware header */
764 skb_reserve(skb
, hh_len
);
766 /* create space for UDP/IP header */
767 skb_put(skb
, fragheaderlen
+ transhdrlen
);
769 /* initialize network header pointer */
770 skb_reset_network_header(skb
);
772 /* initialize protocol header pointer */
773 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
775 skb
->ip_summed
= CHECKSUM_PARTIAL
;
778 /* specify the length of each IP datagram fragment */
779 skb_shinfo(skb
)->gso_size
= maxfraglen
- fragheaderlen
;
780 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
781 __skb_queue_tail(queue
, skb
);
784 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
785 (length
- transhdrlen
));
788 static int __ip_append_data(struct sock
*sk
,
790 struct sk_buff_head
*queue
,
791 struct inet_cork
*cork
,
792 struct page_frag
*pfrag
,
793 int getfrag(void *from
, char *to
, int offset
,
794 int len
, int odd
, struct sk_buff
*skb
),
795 void *from
, int length
, int transhdrlen
,
798 struct inet_sock
*inet
= inet_sk(sk
);
801 struct ip_options
*opt
= cork
->opt
;
808 unsigned int maxfraglen
, fragheaderlen
;
809 int csummode
= CHECKSUM_NONE
;
810 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
812 skb
= skb_peek_tail(queue
);
814 exthdrlen
= !skb
? rt
->dst
.header_len
: 0;
815 mtu
= cork
->fragsize
;
817 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
819 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
820 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
822 if (cork
->length
+ length
> 0xFFFF - fragheaderlen
) {
823 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
829 * transhdrlen > 0 means that this is the first fragment and we wish
830 * it won't be fragmented in the future.
833 length
+ fragheaderlen
<= mtu
&&
834 rt
->dst
.dev
->features
& NETIF_F_V4_CSUM
&&
836 csummode
= CHECKSUM_PARTIAL
;
838 cork
->length
+= length
;
839 if (((length
> mtu
) || (skb
&& skb_is_gso(skb
))) &&
840 (sk
->sk_protocol
== IPPROTO_UDP
) &&
841 (rt
->dst
.dev
->features
& NETIF_F_UFO
) && !rt
->dst
.header_len
) {
842 err
= ip_ufo_append_data(sk
, queue
, getfrag
, from
, length
,
843 hh_len
, fragheaderlen
, transhdrlen
,
850 /* So, what's going on in the loop below?
852 * We use calculated fragment length to generate chained skb,
853 * each of segments is IP fragment ready for sending to network after
854 * adding appropriate IP header.
861 /* Check if the remaining data fits into current packet. */
862 copy
= mtu
- skb
->len
;
864 copy
= maxfraglen
- skb
->len
;
867 unsigned int datalen
;
868 unsigned int fraglen
;
869 unsigned int fraggap
;
870 unsigned int alloclen
;
871 struct sk_buff
*skb_prev
;
875 fraggap
= skb_prev
->len
- maxfraglen
;
880 * If remaining data exceeds the mtu,
881 * we know we need more fragment(s).
883 datalen
= length
+ fraggap
;
884 if (datalen
> mtu
- fragheaderlen
)
885 datalen
= maxfraglen
- fragheaderlen
;
886 fraglen
= datalen
+ fragheaderlen
;
888 if ((flags
& MSG_MORE
) &&
889 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
894 alloclen
+= exthdrlen
;
896 /* The last fragment gets additional space at tail.
897 * Note, with MSG_MORE we overallocate on fragments,
898 * because we have no idea what fragment will be
901 if (datalen
== length
+ fraggap
)
902 alloclen
+= rt
->dst
.trailer_len
;
905 skb
= sock_alloc_send_skb(sk
,
906 alloclen
+ hh_len
+ 15,
907 (flags
& MSG_DONTWAIT
), &err
);
910 if (atomic_read(&sk
->sk_wmem_alloc
) <=
912 skb
= sock_wmalloc(sk
,
913 alloclen
+ hh_len
+ 15, 1,
915 if (unlikely(skb
== NULL
))
918 /* only the initial fragment is
926 * Fill in the control structures
928 skb
->ip_summed
= csummode
;
930 skb_reserve(skb
, hh_len
);
931 skb_shinfo(skb
)->tx_flags
= cork
->tx_flags
;
934 * Find where to start putting bytes.
936 data
= skb_put(skb
, fraglen
+ exthdrlen
);
937 skb_set_network_header(skb
, exthdrlen
);
938 skb
->transport_header
= (skb
->network_header
+
940 data
+= fragheaderlen
+ exthdrlen
;
943 skb
->csum
= skb_copy_and_csum_bits(
944 skb_prev
, maxfraglen
,
945 data
+ transhdrlen
, fraggap
, 0);
946 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
949 pskb_trim_unique(skb_prev
, maxfraglen
);
952 copy
= datalen
- transhdrlen
- fraggap
;
953 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
960 length
-= datalen
- fraggap
;
963 csummode
= CHECKSUM_NONE
;
966 * Put the packet on the pending queue.
968 __skb_queue_tail(queue
, skb
);
975 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
)) {
979 if (getfrag(from
, skb_put(skb
, copy
),
980 offset
, copy
, off
, skb
) < 0) {
981 __skb_trim(skb
, off
);
986 int i
= skb_shinfo(skb
)->nr_frags
;
989 if (!sk_page_frag_refill(sk
, pfrag
))
992 if (!skb_can_coalesce(skb
, i
, pfrag
->page
,
995 if (i
== MAX_SKB_FRAGS
)
998 __skb_fill_page_desc(skb
, i
, pfrag
->page
,
1000 skb_shinfo(skb
)->nr_frags
= ++i
;
1001 get_page(pfrag
->page
);
1003 copy
= min_t(int, copy
, pfrag
->size
- pfrag
->offset
);
1005 page_address(pfrag
->page
) + pfrag
->offset
,
1006 offset
, copy
, skb
->len
, skb
) < 0)
1009 pfrag
->offset
+= copy
;
1010 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], copy
);
1012 skb
->data_len
+= copy
;
1013 skb
->truesize
+= copy
;
1014 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1025 cork
->length
-= length
;
1026 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1030 static int ip_setup_cork(struct sock
*sk
, struct inet_cork
*cork
,
1031 struct ipcm_cookie
*ipc
, struct rtable
**rtp
)
1033 struct inet_sock
*inet
= inet_sk(sk
);
1034 struct ip_options_rcu
*opt
;
1038 * setup for corking.
1042 if (cork
->opt
== NULL
) {
1043 cork
->opt
= kmalloc(sizeof(struct ip_options
) + 40,
1045 if (unlikely(cork
->opt
== NULL
))
1048 memcpy(cork
->opt
, &opt
->opt
, sizeof(struct ip_options
) + opt
->opt
.optlen
);
1049 cork
->flags
|= IPCORK_OPT
;
1050 cork
->addr
= ipc
->addr
;
1056 * We steal reference to this route, caller should not release it
1059 cork
->fragsize
= inet
->pmtudisc
== IP_PMTUDISC_PROBE
?
1060 rt
->dst
.dev
->mtu
: dst_mtu(&rt
->dst
);
1061 cork
->dst
= &rt
->dst
;
1063 cork
->tx_flags
= ipc
->tx_flags
;
1069 * ip_append_data() and ip_append_page() can make one large IP datagram
1070 * from many pieces of data. Each pieces will be holded on the socket
1071 * until ip_push_pending_frames() is called. Each piece can be a page
1074 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1075 * this interface potentially.
1077 * LATER: length must be adjusted by pad at tail, when it is required.
1079 int ip_append_data(struct sock
*sk
, struct flowi4
*fl4
,
1080 int getfrag(void *from
, char *to
, int offset
, int len
,
1081 int odd
, struct sk_buff
*skb
),
1082 void *from
, int length
, int transhdrlen
,
1083 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1086 struct inet_sock
*inet
= inet_sk(sk
);
1089 if (flags
&MSG_PROBE
)
1092 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1093 err
= ip_setup_cork(sk
, &inet
->cork
.base
, ipc
, rtp
);
1100 return __ip_append_data(sk
, fl4
, &sk
->sk_write_queue
, &inet
->cork
.base
,
1101 sk_page_frag(sk
), getfrag
,
1102 from
, length
, transhdrlen
, flags
);
1105 ssize_t
ip_append_page(struct sock
*sk
, struct flowi4
*fl4
, struct page
*page
,
1106 int offset
, size_t size
, int flags
)
1108 struct inet_sock
*inet
= inet_sk(sk
);
1109 struct sk_buff
*skb
;
1111 struct ip_options
*opt
= NULL
;
1112 struct inet_cork
*cork
;
1117 unsigned int maxfraglen
, fragheaderlen
, fraggap
;
1122 if (flags
&MSG_PROBE
)
1125 if (skb_queue_empty(&sk
->sk_write_queue
))
1128 cork
= &inet
->cork
.base
;
1129 rt
= (struct rtable
*)cork
->dst
;
1130 if (cork
->flags
& IPCORK_OPT
)
1133 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1136 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1137 mtu
= cork
->fragsize
;
1139 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1140 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1142 if (cork
->length
+ size
> 0xFFFF - fragheaderlen
) {
1143 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
, mtu
);
1147 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1150 cork
->length
+= size
;
1151 if ((size
+ skb
->len
> mtu
) &&
1152 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1153 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
1154 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1155 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1162 if (skb_is_gso(skb
))
1166 /* Check if the remaining data fits into current packet. */
1167 len
= mtu
- skb
->len
;
1169 len
= maxfraglen
- skb
->len
;
1172 struct sk_buff
*skb_prev
;
1176 fraggap
= skb_prev
->len
- maxfraglen
;
1178 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1179 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1180 if (unlikely(!skb
)) {
1186 * Fill in the control structures
1188 skb
->ip_summed
= CHECKSUM_NONE
;
1190 skb_reserve(skb
, hh_len
);
1193 * Find where to start putting bytes.
1195 skb_put(skb
, fragheaderlen
+ fraggap
);
1196 skb_reset_network_header(skb
);
1197 skb
->transport_header
= (skb
->network_header
+
1200 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1202 skb_transport_header(skb
),
1204 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1206 pskb_trim_unique(skb_prev
, maxfraglen
);
1210 * Put the packet on the pending queue.
1212 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1216 i
= skb_shinfo(skb
)->nr_frags
;
1219 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1220 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
-1], len
);
1221 } else if (i
< MAX_SKB_FRAGS
) {
1223 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1229 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1231 csum
= csum_page(page
, offset
, len
);
1232 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1236 skb
->data_len
+= len
;
1237 skb
->truesize
+= len
;
1238 atomic_add(len
, &sk
->sk_wmem_alloc
);
1245 cork
->length
-= size
;
1246 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1250 static void ip_cork_release(struct inet_cork
*cork
)
1252 cork
->flags
&= ~IPCORK_OPT
;
1255 dst_release(cork
->dst
);
1260 * Combined all pending IP fragments on the socket as one IP datagram
1261 * and push them out.
1263 struct sk_buff
*__ip_make_skb(struct sock
*sk
,
1265 struct sk_buff_head
*queue
,
1266 struct inet_cork
*cork
)
1268 struct sk_buff
*skb
, *tmp_skb
;
1269 struct sk_buff
**tail_skb
;
1270 struct inet_sock
*inet
= inet_sk(sk
);
1271 struct net
*net
= sock_net(sk
);
1272 struct ip_options
*opt
= NULL
;
1273 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
1278 if ((skb
= __skb_dequeue(queue
)) == NULL
)
1280 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1282 /* move skb->data to ip header from ext header */
1283 if (skb
->data
< skb_network_header(skb
))
1284 __skb_pull(skb
, skb_network_offset(skb
));
1285 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1286 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1287 *tail_skb
= tmp_skb
;
1288 tail_skb
= &(tmp_skb
->next
);
1289 skb
->len
+= tmp_skb
->len
;
1290 skb
->data_len
+= tmp_skb
->len
;
1291 skb
->truesize
+= tmp_skb
->truesize
;
1292 tmp_skb
->destructor
= NULL
;
1296 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1297 * to fragment the frame generated here. No matter, what transforms
1298 * how transforms change size of the packet, it will come out.
1300 if (inet
->pmtudisc
< IP_PMTUDISC_DO
)
1303 /* DF bit is set when we want to see DF on outgoing frames.
1304 * If local_df is set too, we still allow to fragment this frame
1306 if (inet
->pmtudisc
>= IP_PMTUDISC_DO
||
1307 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1308 ip_dont_fragment(sk
, &rt
->dst
)))
1311 if (cork
->flags
& IPCORK_OPT
)
1314 if (rt
->rt_type
== RTN_MULTICAST
)
1317 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1322 iph
->tos
= inet
->tos
;
1325 iph
->protocol
= sk
->sk_protocol
;
1326 ip_copy_addrs(iph
, fl4
);
1327 ip_select_ident(skb
, &rt
->dst
, sk
);
1330 iph
->ihl
+= opt
->optlen
>>2;
1331 ip_options_build(skb
, opt
, cork
->addr
, rt
, 0);
1334 skb
->priority
= sk
->sk_priority
;
1335 skb
->mark
= sk
->sk_mark
;
1337 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1341 skb_dst_set(skb
, &rt
->dst
);
1343 if (iph
->protocol
== IPPROTO_ICMP
)
1344 icmp_out_count(net
, ((struct icmphdr
*)
1345 skb_transport_header(skb
))->type
);
1347 ip_cork_release(cork
);
1352 int ip_send_skb(struct net
*net
, struct sk_buff
*skb
)
1356 err
= ip_local_out(skb
);
1359 err
= net_xmit_errno(err
);
1361 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1367 int ip_push_pending_frames(struct sock
*sk
, struct flowi4
*fl4
)
1369 struct sk_buff
*skb
;
1371 skb
= ip_finish_skb(sk
, fl4
);
1375 /* Netfilter gets whole the not fragmented skb. */
1376 return ip_send_skb(sock_net(sk
), skb
);
1380 * Throw away all pending data on the socket.
1382 static void __ip_flush_pending_frames(struct sock
*sk
,
1383 struct sk_buff_head
*queue
,
1384 struct inet_cork
*cork
)
1386 struct sk_buff
*skb
;
1388 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
)
1391 ip_cork_release(cork
);
1394 void ip_flush_pending_frames(struct sock
*sk
)
1396 __ip_flush_pending_frames(sk
, &sk
->sk_write_queue
, &inet_sk(sk
)->cork
.base
);
1399 struct sk_buff
*ip_make_skb(struct sock
*sk
,
1401 int getfrag(void *from
, char *to
, int offset
,
1402 int len
, int odd
, struct sk_buff
*skb
),
1403 void *from
, int length
, int transhdrlen
,
1404 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1407 struct inet_cork cork
;
1408 struct sk_buff_head queue
;
1411 if (flags
& MSG_PROBE
)
1414 __skb_queue_head_init(&queue
);
1419 err
= ip_setup_cork(sk
, &cork
, ipc
, rtp
);
1421 return ERR_PTR(err
);
1423 err
= __ip_append_data(sk
, fl4
, &queue
, &cork
,
1424 ¤t
->task_frag
, getfrag
,
1425 from
, length
, transhdrlen
, flags
);
1427 __ip_flush_pending_frames(sk
, &queue
, &cork
);
1428 return ERR_PTR(err
);
1431 return __ip_make_skb(sk
, fl4
, &queue
, &cork
);
1435 * Fetch data from kernel space and fill in checksum if needed.
1437 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1438 int len
, int odd
, struct sk_buff
*skb
)
1442 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1443 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1448 * Generic function to send a packet as reply to another packet.
1449 * Used to send some TCP resets/acks so far.
1451 * Use a fake percpu inet socket to avoid false sharing and contention.
1453 static DEFINE_PER_CPU(struct inet_sock
, unicast_sock
) = {
1456 .skc_refcnt
= ATOMIC_INIT(1),
1458 .sk_wmem_alloc
= ATOMIC_INIT(1),
1459 .sk_allocation
= GFP_ATOMIC
,
1460 .sk_flags
= (1UL << SOCK_USE_WRITE_QUEUE
),
1462 .pmtudisc
= IP_PMTUDISC_WANT
,
1466 void ip_send_unicast_reply(struct net
*net
, struct sk_buff
*skb
, __be32 daddr
,
1467 __be32 saddr
, const struct ip_reply_arg
*arg
,
1470 struct ip_options_data replyopts
;
1471 struct ipcm_cookie ipc
;
1473 struct rtable
*rt
= skb_rtable(skb
);
1474 struct sk_buff
*nskb
;
1476 struct inet_sock
*inet
;
1478 if (ip_options_echo(&replyopts
.opt
.opt
, skb
))
1485 if (replyopts
.opt
.opt
.optlen
) {
1486 ipc
.opt
= &replyopts
.opt
;
1488 if (replyopts
.opt
.opt
.srr
)
1489 daddr
= replyopts
.opt
.opt
.faddr
;
1492 flowi4_init_output(&fl4
, arg
->bound_dev_if
, 0,
1494 RT_SCOPE_UNIVERSE
, ip_hdr(skb
)->protocol
,
1495 ip_reply_arg_flowi_flags(arg
),
1497 tcp_hdr(skb
)->source
, tcp_hdr(skb
)->dest
);
1498 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
1499 rt
= ip_route_output_key(net
, &fl4
);
1503 inet
= &get_cpu_var(unicast_sock
);
1505 inet
->tos
= arg
->tos
;
1507 sk
->sk_priority
= skb
->priority
;
1508 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1509 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1510 sock_net_set(sk
, net
);
1511 __skb_queue_head_init(&sk
->sk_write_queue
);
1512 sk
->sk_sndbuf
= sysctl_wmem_default
;
1513 ip_append_data(sk
, &fl4
, ip_reply_glue_bits
, arg
->iov
->iov_base
, len
, 0,
1514 &ipc
, &rt
, MSG_DONTWAIT
);
1515 nskb
= skb_peek(&sk
->sk_write_queue
);
1517 if (arg
->csumoffset
>= 0)
1518 *((__sum16
*)skb_transport_header(nskb
) +
1519 arg
->csumoffset
) = csum_fold(csum_add(nskb
->csum
,
1521 nskb
->ip_summed
= CHECKSUM_NONE
;
1523 skb_set_queue_mapping(nskb
, skb_get_queue_mapping(skb
));
1524 ip_push_pending_frames(sk
, &fl4
);
1527 put_cpu_var(unicast_sock
);
1532 void __init
ip_init(void)
1537 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1538 igmp_mc_proc_init();