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>
57 #include <linux/socket.h>
58 #include <linux/sockios.h>
60 #include <linux/inet.h>
61 #include <linux/netdevice.h>
62 #include <linux/etherdevice.h>
63 #include <linux/proc_fs.h>
64 #include <linux/stat.h>
65 #include <linux/init.h>
69 #include <net/protocol.h>
70 #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 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
206 /* Policy lookup after SNAT yielded a new policy */
207 if (skb
->dst
->xfrm
!= NULL
) {
208 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
209 return dst_output(skb
);
212 if (skb
->len
> dst_mtu(skb
->dst
) && !skb_is_gso(skb
))
213 return ip_fragment(skb
, ip_finish_output2
);
215 return ip_finish_output2(skb
);
218 int ip_mc_output(struct sk_buff
*skb
)
220 struct sock
*sk
= skb
->sk
;
221 struct rtable
*rt
= (struct rtable
*)skb
->dst
;
222 struct net_device
*dev
= rt
->u
.dst
.dev
;
225 * If the indicated interface is up and running, send the packet.
227 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS
);
230 skb
->protocol
= htons(ETH_P_IP
);
233 * Multicasts are looped back for other local users
236 if (rt
->rt_flags
&RTCF_MULTICAST
) {
237 if ((!sk
|| inet_sk(sk
)->mc_loop
)
238 #ifdef CONFIG_IP_MROUTE
239 /* Small optimization: do not loopback not local frames,
240 which returned after forwarding; they will be dropped
241 by ip_mr_input in any case.
242 Note, that local frames are looped back to be delivered
245 This check is duplicated in ip_mr_input at the moment.
247 && ((rt
->rt_flags
&RTCF_LOCAL
) || !(IPCB(skb
)->flags
&IPSKB_FORWARDED
))
250 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
252 NF_HOOK(PF_INET
, NF_IP_POST_ROUTING
, newskb
, NULL
,
254 ip_dev_loopback_xmit
);
257 /* Multicasts with ttl 0 must not go beyond the host */
259 if (skb
->nh
.iph
->ttl
== 0) {
265 if (rt
->rt_flags
&RTCF_BROADCAST
) {
266 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
268 NF_HOOK(PF_INET
, NF_IP_POST_ROUTING
, newskb
, NULL
,
269 newskb
->dev
, ip_dev_loopback_xmit
);
272 return NF_HOOK_COND(PF_INET
, NF_IP_POST_ROUTING
, skb
, NULL
, skb
->dev
,
274 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
277 int ip_output(struct sk_buff
*skb
)
279 struct net_device
*dev
= skb
->dst
->dev
;
281 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS
);
284 skb
->protocol
= htons(ETH_P_IP
);
286 return NF_HOOK_COND(PF_INET
, NF_IP_POST_ROUTING
, skb
, NULL
, dev
,
288 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
291 int ip_queue_xmit(struct sk_buff
*skb
, int ipfragok
)
293 struct sock
*sk
= skb
->sk
;
294 struct inet_sock
*inet
= inet_sk(sk
);
295 struct ip_options
*opt
= inet
->opt
;
299 /* Skip all of this if the packet is already routed,
300 * f.e. by something like SCTP.
302 rt
= (struct rtable
*) skb
->dst
;
306 /* Make sure we can route this packet. */
307 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
311 /* Use correct destination address if we have options. */
317 struct flowi fl
= { .oif
= sk
->sk_bound_dev_if
,
320 .saddr
= inet
->saddr
,
321 .tos
= RT_CONN_FLAGS(sk
) } },
322 .proto
= sk
->sk_protocol
,
324 { .sport
= inet
->sport
,
325 .dport
= inet
->dport
} } };
327 /* If this fails, retransmit mechanism of transport layer will
328 * keep trying until route appears or the connection times
331 if (ip_route_output_flow(&rt
, &fl
, sk
, 0))
334 sk_setup_caps(sk
, &rt
->u
.dst
);
336 skb
->dst
= dst_clone(&rt
->u
.dst
);
339 if (opt
&& opt
->is_strictroute
&& rt
->rt_dst
!= rt
->rt_gateway
)
342 /* OK, we know where to send it, allocate and build IP header. */
343 iph
= (struct iphdr
*) skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
344 *((__u16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
345 iph
->tot_len
= htons(skb
->len
);
346 if (ip_dont_fragment(sk
, &rt
->u
.dst
) && !ipfragok
)
347 iph
->frag_off
= htons(IP_DF
);
350 iph
->ttl
= ip_select_ttl(inet
, &rt
->u
.dst
);
351 iph
->protocol
= sk
->sk_protocol
;
352 iph
->saddr
= rt
->rt_src
;
353 iph
->daddr
= rt
->rt_dst
;
355 /* Transport layer set skb->h.foo itself. */
357 if (opt
&& opt
->optlen
) {
358 iph
->ihl
+= opt
->optlen
>> 2;
359 ip_options_build(skb
, opt
, inet
->daddr
, rt
, 0);
362 ip_select_ident_more(iph
, &rt
->u
.dst
, sk
,
363 (skb_shinfo(skb
)->gso_segs
?: 1) - 1);
365 /* Add an IP checksum. */
368 skb
->priority
= sk
->sk_priority
;
370 return NF_HOOK(PF_INET
, NF_IP_LOCAL_OUT
, skb
, NULL
, rt
->u
.dst
.dev
,
374 IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES
);
376 return -EHOSTUNREACH
;
380 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
382 to
->pkt_type
= from
->pkt_type
;
383 to
->priority
= from
->priority
;
384 to
->protocol
= from
->protocol
;
385 dst_release(to
->dst
);
386 to
->dst
= dst_clone(from
->dst
);
389 /* Copy the flags to each fragment. */
390 IPCB(to
)->flags
= IPCB(from
)->flags
;
392 #ifdef CONFIG_NET_SCHED
393 to
->tc_index
= from
->tc_index
;
395 #ifdef CONFIG_NETFILTER
396 to
->nfmark
= from
->nfmark
;
397 /* Connection association is same as pre-frag packet */
398 nf_conntrack_put(to
->nfct
);
399 to
->nfct
= from
->nfct
;
400 nf_conntrack_get(to
->nfct
);
401 to
->nfctinfo
= from
->nfctinfo
;
402 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
403 to
->ipvs_property
= from
->ipvs_property
;
405 #ifdef CONFIG_BRIDGE_NETFILTER
406 nf_bridge_put(to
->nf_bridge
);
407 to
->nf_bridge
= from
->nf_bridge
;
408 nf_bridge_get(to
->nf_bridge
);
411 skb_copy_secmark(to
, from
);
415 * This IP datagram is too large to be sent in one piece. Break it up into
416 * smaller pieces (each of size equal to IP header plus
417 * a block of the data of the original IP data part) that will yet fit in a
418 * single device frame, and queue such a frame for sending.
421 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
426 struct net_device
*dev
;
427 struct sk_buff
*skb2
;
428 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
430 __be16 not_last_frag
;
431 struct rtable
*rt
= (struct rtable
*)skb
->dst
;
437 * Point into the IP datagram header.
442 if (unlikely((iph
->frag_off
& htons(IP_DF
)) && !skb
->local_df
)) {
443 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
444 htonl(dst_mtu(&rt
->u
.dst
)));
450 * Setup starting values.
454 mtu
= dst_mtu(&rt
->u
.dst
) - hlen
; /* Size of data space */
455 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
457 /* When frag_list is given, use it. First, check its validity:
458 * some transformers could create wrong frag_list or break existing
459 * one, it is not prohibited. In this case fall back to copying.
461 * LATER: this step can be merged to real generation of fragments,
462 * we can switch to copy when see the first bad fragment.
464 if (skb_shinfo(skb
)->frag_list
) {
465 struct sk_buff
*frag
;
466 int first_len
= skb_pagelen(skb
);
468 if (first_len
- hlen
> mtu
||
469 ((first_len
- hlen
) & 7) ||
470 (iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) ||
474 for (frag
= skb_shinfo(skb
)->frag_list
; frag
; frag
= frag
->next
) {
475 /* Correct geometry. */
476 if (frag
->len
> mtu
||
477 ((frag
->len
& 7) && frag
->next
) ||
478 skb_headroom(frag
) < hlen
)
481 /* Partially cloned skb? */
482 if (skb_shared(frag
))
489 frag
->destructor
= sock_wfree
;
490 skb
->truesize
-= frag
->truesize
;
494 /* Everything is OK. Generate! */
498 frag
= skb_shinfo(skb
)->frag_list
;
499 skb_shinfo(skb
)->frag_list
= NULL
;
500 skb
->data_len
= first_len
- skb_headlen(skb
);
501 skb
->len
= first_len
;
502 iph
->tot_len
= htons(first_len
);
503 iph
->frag_off
= htons(IP_MF
);
507 /* Prepare header of the next frame,
508 * before previous one went down. */
510 frag
->ip_summed
= CHECKSUM_NONE
;
511 frag
->h
.raw
= frag
->data
;
512 frag
->nh
.raw
= __skb_push(frag
, hlen
);
513 memcpy(frag
->nh
.raw
, iph
, hlen
);
515 iph
->tot_len
= htons(frag
->len
);
516 ip_copy_metadata(frag
, skb
);
518 ip_options_fragment(frag
);
519 offset
+= skb
->len
- hlen
;
520 iph
->frag_off
= htons(offset
>>3);
521 if (frag
->next
!= NULL
)
522 iph
->frag_off
|= htons(IP_MF
);
523 /* Ready, complete checksum */
538 IP_INC_STATS(IPSTATS_MIB_FRAGOKS
);
547 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS
);
552 left
= skb
->len
- hlen
; /* Space per frame */
553 ptr
= raw
+ hlen
; /* Where to start from */
555 #ifdef CONFIG_BRIDGE_NETFILTER
556 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
557 * we need to make room for the encapsulating header */
558 ll_rs
= LL_RESERVED_SPACE_EXTRA(rt
->u
.dst
.dev
, nf_bridge_pad(skb
));
559 mtu
-= nf_bridge_pad(skb
);
561 ll_rs
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
564 * Fragment the datagram.
567 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
568 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
571 * Keep copying data until we run out.
576 /* IF: it doesn't fit, use 'mtu' - the data space left */
579 /* IF: we are not sending upto and including the packet end
580 then align the next start on an eight byte boundary */
588 if ((skb2
= alloc_skb(len
+hlen
+ll_rs
, GFP_ATOMIC
)) == NULL
) {
589 NETDEBUG(KERN_INFO
"IP: frag: no memory for new fragment!\n");
595 * Set up data on packet
598 ip_copy_metadata(skb2
, skb
);
599 skb_reserve(skb2
, ll_rs
);
600 skb_put(skb2
, len
+ hlen
);
601 skb2
->nh
.raw
= skb2
->data
;
602 skb2
->h
.raw
= skb2
->data
+ hlen
;
605 * Charge the memory for the fragment to any owner
610 skb_set_owner_w(skb2
, skb
->sk
);
613 * Copy the packet header into the new buffer.
616 memcpy(skb2
->nh
.raw
, skb
->data
, hlen
);
619 * Copy a block of the IP datagram.
621 if (skb_copy_bits(skb
, ptr
, skb2
->h
.raw
, len
))
626 * Fill in the new header fields.
629 iph
->frag_off
= htons((offset
>> 3));
631 /* ANK: dirty, but effective trick. Upgrade options only if
632 * the segment to be fragmented was THE FIRST (otherwise,
633 * options are already fixed) and make it ONCE
634 * on the initial skb, so that all the following fragments
635 * will inherit fixed options.
638 ip_options_fragment(skb
);
641 * Added AC : If we are fragmenting a fragment that's not the
642 * last fragment then keep MF on each bit
644 if (left
> 0 || not_last_frag
)
645 iph
->frag_off
|= htons(IP_MF
);
650 * Put this fragment into the sending queue.
653 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES
);
655 iph
->tot_len
= htons(len
+ hlen
);
664 IP_INC_STATS(IPSTATS_MIB_FRAGOKS
);
669 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS
);
673 EXPORT_SYMBOL(ip_fragment
);
676 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
678 struct iovec
*iov
= from
;
680 if (skb
->ip_summed
== CHECKSUM_HW
) {
681 if (memcpy_fromiovecend(to
, iov
, offset
, len
) < 0)
684 unsigned int csum
= 0;
685 if (csum_partial_copy_fromiovecend(to
, iov
, offset
, len
, &csum
) < 0)
687 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
692 static inline unsigned int
693 csum_page(struct page
*page
, int offset
, int copy
)
698 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
703 static inline int ip_ufo_append_data(struct sock
*sk
,
704 int getfrag(void *from
, char *to
, int offset
, int len
,
705 int odd
, struct sk_buff
*skb
),
706 void *from
, int length
, int hh_len
, int fragheaderlen
,
707 int transhdrlen
, int mtu
,unsigned int flags
)
712 /* There is support for UDP fragmentation offload by network
713 * device, so create one single skb packet containing complete
716 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
717 skb
= sock_alloc_send_skb(sk
,
718 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
719 (flags
& MSG_DONTWAIT
), &err
);
724 /* reserve space for Hardware header */
725 skb_reserve(skb
, hh_len
);
727 /* create space for UDP/IP header */
728 skb_put(skb
,fragheaderlen
+ transhdrlen
);
730 /* initialize network header pointer */
731 skb
->nh
.raw
= skb
->data
;
733 /* initialize protocol header pointer */
734 skb
->h
.raw
= skb
->data
+ fragheaderlen
;
736 skb
->ip_summed
= CHECKSUM_HW
;
738 sk
->sk_sndmsg_off
= 0;
741 err
= skb_append_datato_frags(sk
,skb
, getfrag
, from
,
742 (length
- transhdrlen
));
744 /* specify the length of each IP datagram fragment*/
745 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
746 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
747 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
751 /* There is not enough support do UFO ,
752 * so follow normal path
759 * ip_append_data() and ip_append_page() can make one large IP datagram
760 * from many pieces of data. Each pieces will be holded on the socket
761 * until ip_push_pending_frames() is called. Each piece can be a page
764 * Not only UDP, other transport protocols - e.g. raw sockets - can use
765 * this interface potentially.
767 * LATER: length must be adjusted by pad at tail, when it is required.
769 int ip_append_data(struct sock
*sk
,
770 int getfrag(void *from
, char *to
, int offset
, int len
,
771 int odd
, struct sk_buff
*skb
),
772 void *from
, int length
, int transhdrlen
,
773 struct ipcm_cookie
*ipc
, struct rtable
*rt
,
776 struct inet_sock
*inet
= inet_sk(sk
);
779 struct ip_options
*opt
= NULL
;
786 unsigned int maxfraglen
, fragheaderlen
;
787 int csummode
= CHECKSUM_NONE
;
792 if (skb_queue_empty(&sk
->sk_write_queue
)) {
798 if (inet
->cork
.opt
== NULL
) {
799 inet
->cork
.opt
= kmalloc(sizeof(struct ip_options
) + 40, sk
->sk_allocation
);
800 if (unlikely(inet
->cork
.opt
== NULL
))
803 memcpy(inet
->cork
.opt
, opt
, sizeof(struct ip_options
)+opt
->optlen
);
804 inet
->cork
.flags
|= IPCORK_OPT
;
805 inet
->cork
.addr
= ipc
->addr
;
807 dst_hold(&rt
->u
.dst
);
808 inet
->cork
.fragsize
= mtu
= dst_mtu(rt
->u
.dst
.path
);
810 inet
->cork
.length
= 0;
811 sk
->sk_sndmsg_page
= NULL
;
812 sk
->sk_sndmsg_off
= 0;
813 if ((exthdrlen
= rt
->u
.dst
.header_len
) != 0) {
815 transhdrlen
+= exthdrlen
;
819 if (inet
->cork
.flags
& IPCORK_OPT
)
820 opt
= inet
->cork
.opt
;
824 mtu
= inet
->cork
.fragsize
;
826 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
828 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
829 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
831 if (inet
->cork
.length
+ length
> 0xFFFF - fragheaderlen
) {
832 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->dport
, mtu
-exthdrlen
);
837 * transhdrlen > 0 means that this is the first fragment and we wish
838 * it won't be fragmented in the future.
841 length
+ fragheaderlen
<= mtu
&&
842 rt
->u
.dst
.dev
->features
& NETIF_F_ALL_CSUM
&&
844 csummode
= CHECKSUM_HW
;
846 inet
->cork
.length
+= length
;
847 if (((length
> mtu
) && (sk
->sk_protocol
== IPPROTO_UDP
)) &&
848 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
850 err
= ip_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
851 fragheaderlen
, transhdrlen
, mtu
,
858 /* So, what's going on in the loop below?
860 * We use calculated fragment length to generate chained skb,
861 * each of segments is IP fragment ready for sending to network after
862 * adding appropriate IP header.
865 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
869 /* Check if the remaining data fits into current packet. */
870 copy
= mtu
- skb
->len
;
872 copy
= maxfraglen
- skb
->len
;
875 unsigned int datalen
;
876 unsigned int fraglen
;
877 unsigned int fraggap
;
878 unsigned int alloclen
;
879 struct sk_buff
*skb_prev
;
883 fraggap
= skb_prev
->len
- maxfraglen
;
888 * If remaining data exceeds the mtu,
889 * we know we need more fragment(s).
891 datalen
= length
+ fraggap
;
892 if (datalen
> mtu
- fragheaderlen
)
893 datalen
= maxfraglen
- fragheaderlen
;
894 fraglen
= datalen
+ fragheaderlen
;
896 if ((flags
& MSG_MORE
) &&
897 !(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
900 alloclen
= datalen
+ fragheaderlen
;
902 /* The last fragment gets additional space at tail.
903 * Note, with MSG_MORE we overallocate on fragments,
904 * because we have no idea what fragment will be
907 if (datalen
== length
+ fraggap
)
908 alloclen
+= rt
->u
.dst
.trailer_len
;
911 skb
= sock_alloc_send_skb(sk
,
912 alloclen
+ hh_len
+ 15,
913 (flags
& MSG_DONTWAIT
), &err
);
916 if (atomic_read(&sk
->sk_wmem_alloc
) <=
918 skb
= sock_wmalloc(sk
,
919 alloclen
+ hh_len
+ 15, 1,
921 if (unlikely(skb
== NULL
))
928 * Fill in the control structures
930 skb
->ip_summed
= csummode
;
932 skb_reserve(skb
, hh_len
);
935 * Find where to start putting bytes.
937 data
= skb_put(skb
, fraglen
);
938 skb
->nh
.raw
= data
+ exthdrlen
;
939 data
+= fragheaderlen
;
940 skb
->h
.raw
= data
+ 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 skb_trim(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(&sk
->sk_write_queue
, skb
);
975 if (!(rt
->u
.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
;
987 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
988 struct page
*page
= sk
->sk_sndmsg_page
;
989 int off
= sk
->sk_sndmsg_off
;
992 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
995 if (page
!= frag
->page
) {
996 if (i
== MAX_SKB_FRAGS
) {
1001 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
1002 frag
= &skb_shinfo(skb
)->frags
[i
];
1004 } else if (i
< MAX_SKB_FRAGS
) {
1005 if (copy
> PAGE_SIZE
)
1007 page
= alloc_pages(sk
->sk_allocation
, 0);
1012 sk
->sk_sndmsg_page
= page
;
1013 sk
->sk_sndmsg_off
= 0;
1015 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1016 frag
= &skb_shinfo(skb
)->frags
[i
];
1017 skb
->truesize
+= PAGE_SIZE
;
1018 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1023 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1027 sk
->sk_sndmsg_off
+= copy
;
1030 skb
->data_len
+= copy
;
1039 inet
->cork
.length
-= length
;
1040 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS
);
1044 ssize_t
ip_append_page(struct sock
*sk
, struct page
*page
,
1045 int offset
, size_t size
, int flags
)
1047 struct inet_sock
*inet
= inet_sk(sk
);
1048 struct sk_buff
*skb
;
1050 struct ip_options
*opt
= NULL
;
1055 unsigned int maxfraglen
, fragheaderlen
, fraggap
;
1060 if (flags
&MSG_PROBE
)
1063 if (skb_queue_empty(&sk
->sk_write_queue
))
1067 if (inet
->cork
.flags
& IPCORK_OPT
)
1068 opt
= inet
->cork
.opt
;
1070 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
1073 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
1074 mtu
= inet
->cork
.fragsize
;
1076 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1077 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1079 if (inet
->cork
.length
+ size
> 0xFFFF - fragheaderlen
) {
1080 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->dport
, mtu
);
1084 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1087 inet
->cork
.length
+= size
;
1088 if ((sk
->sk_protocol
== IPPROTO_UDP
) &&
1089 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
1090 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1091 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1098 if (skb_is_gso(skb
))
1102 /* Check if the remaining data fits into current packet. */
1103 len
= mtu
- skb
->len
;
1105 len
= maxfraglen
- skb
->len
;
1108 struct sk_buff
*skb_prev
;
1114 fraggap
= skb_prev
->len
- maxfraglen
;
1116 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1117 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1118 if (unlikely(!skb
)) {
1124 * Fill in the control structures
1126 skb
->ip_summed
= CHECKSUM_NONE
;
1128 skb_reserve(skb
, hh_len
);
1131 * Find where to start putting bytes.
1133 data
= skb_put(skb
, fragheaderlen
+ fraggap
);
1134 skb
->nh
.iph
= iph
= (struct iphdr
*)data
;
1135 data
+= fragheaderlen
;
1139 skb
->csum
= skb_copy_and_csum_bits(
1140 skb_prev
, maxfraglen
,
1142 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1144 skb_trim(skb_prev
, maxfraglen
);
1148 * Put the packet on the pending queue.
1150 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1154 i
= skb_shinfo(skb
)->nr_frags
;
1157 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1158 skb_shinfo(skb
)->frags
[i
-1].size
+= len
;
1159 } else if (i
< MAX_SKB_FRAGS
) {
1161 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1167 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1169 csum
= csum_page(page
, offset
, len
);
1170 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1174 skb
->data_len
+= len
;
1181 inet
->cork
.length
-= size
;
1182 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS
);
1187 * Combined all pending IP fragments on the socket as one IP datagram
1188 * and push them out.
1190 int ip_push_pending_frames(struct sock
*sk
)
1192 struct sk_buff
*skb
, *tmp_skb
;
1193 struct sk_buff
**tail_skb
;
1194 struct inet_sock
*inet
= inet_sk(sk
);
1195 struct ip_options
*opt
= NULL
;
1196 struct rtable
*rt
= inet
->cork
.rt
;
1202 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1204 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1206 /* move skb->data to ip header from ext header */
1207 if (skb
->data
< skb
->nh
.raw
)
1208 __skb_pull(skb
, skb
->nh
.raw
- skb
->data
);
1209 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1210 __skb_pull(tmp_skb
, skb
->h
.raw
- skb
->nh
.raw
);
1211 *tail_skb
= tmp_skb
;
1212 tail_skb
= &(tmp_skb
->next
);
1213 skb
->len
+= tmp_skb
->len
;
1214 skb
->data_len
+= tmp_skb
->len
;
1215 skb
->truesize
+= tmp_skb
->truesize
;
1216 __sock_put(tmp_skb
->sk
);
1217 tmp_skb
->destructor
= NULL
;
1221 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1222 * to fragment the frame generated here. No matter, what transforms
1223 * how transforms change size of the packet, it will come out.
1225 if (inet
->pmtudisc
!= IP_PMTUDISC_DO
)
1228 /* DF bit is set when we want to see DF on outgoing frames.
1229 * If local_df is set too, we still allow to fragment this frame
1231 if (inet
->pmtudisc
== IP_PMTUDISC_DO
||
1232 (skb
->len
<= dst_mtu(&rt
->u
.dst
) &&
1233 ip_dont_fragment(sk
, &rt
->u
.dst
)))
1236 if (inet
->cork
.flags
& IPCORK_OPT
)
1237 opt
= inet
->cork
.opt
;
1239 if (rt
->rt_type
== RTN_MULTICAST
)
1242 ttl
= ip_select_ttl(inet
, &rt
->u
.dst
);
1244 iph
= (struct iphdr
*)skb
->data
;
1248 iph
->ihl
+= opt
->optlen
>>2;
1249 ip_options_build(skb
, opt
, inet
->cork
.addr
, rt
, 0);
1251 iph
->tos
= inet
->tos
;
1252 iph
->tot_len
= htons(skb
->len
);
1254 ip_select_ident(iph
, &rt
->u
.dst
, sk
);
1256 iph
->protocol
= sk
->sk_protocol
;
1257 iph
->saddr
= rt
->rt_src
;
1258 iph
->daddr
= rt
->rt_dst
;
1261 skb
->priority
= sk
->sk_priority
;
1262 skb
->dst
= dst_clone(&rt
->u
.dst
);
1264 /* Netfilter gets whole the not fragmented skb. */
1265 err
= NF_HOOK(PF_INET
, NF_IP_LOCAL_OUT
, skb
, NULL
,
1266 skb
->dst
->dev
, dst_output
);
1269 err
= inet
->recverr
? net_xmit_errno(err
) : 0;
1275 inet
->cork
.flags
&= ~IPCORK_OPT
;
1276 kfree(inet
->cork
.opt
);
1277 inet
->cork
.opt
= NULL
;
1278 if (inet
->cork
.rt
) {
1279 ip_rt_put(inet
->cork
.rt
);
1280 inet
->cork
.rt
= NULL
;
1285 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS
);
1290 * Throw away all pending data on the socket.
1292 void ip_flush_pending_frames(struct sock
*sk
)
1294 struct inet_sock
*inet
= inet_sk(sk
);
1295 struct sk_buff
*skb
;
1297 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
)
1300 inet
->cork
.flags
&= ~IPCORK_OPT
;
1301 kfree(inet
->cork
.opt
);
1302 inet
->cork
.opt
= NULL
;
1303 if (inet
->cork
.rt
) {
1304 ip_rt_put(inet
->cork
.rt
);
1305 inet
->cork
.rt
= NULL
;
1311 * Fetch data from kernel space and fill in checksum if needed.
1313 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1314 int len
, int odd
, struct sk_buff
*skb
)
1318 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1319 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1324 * Generic function to send a packet as reply to another packet.
1325 * Used to send TCP resets so far. ICMP should use this function too.
1327 * Should run single threaded per socket because it uses the sock
1328 * structure to pass arguments.
1330 * LATER: switch from ip_build_xmit to ip_append_*
1332 void ip_send_reply(struct sock
*sk
, struct sk_buff
*skb
, struct ip_reply_arg
*arg
,
1335 struct inet_sock
*inet
= inet_sk(sk
);
1337 struct ip_options opt
;
1340 struct ipcm_cookie ipc
;
1342 struct rtable
*rt
= (struct rtable
*)skb
->dst
;
1344 if (ip_options_echo(&replyopts
.opt
, skb
))
1347 daddr
= ipc
.addr
= rt
->rt_src
;
1350 if (replyopts
.opt
.optlen
) {
1351 ipc
.opt
= &replyopts
.opt
;
1354 daddr
= replyopts
.opt
.faddr
;
1358 struct flowi fl
= { .nl_u
= { .ip4_u
=
1360 .saddr
= rt
->rt_spec_dst
,
1361 .tos
= RT_TOS(skb
->nh
.iph
->tos
) } },
1362 /* Not quite clean, but right. */
1364 { .sport
= skb
->h
.th
->dest
,
1365 .dport
= skb
->h
.th
->source
} },
1366 .proto
= sk
->sk_protocol
};
1367 if (ip_route_output_key(&rt
, &fl
))
1371 /* And let IP do all the hard work.
1373 This chunk is not reenterable, hence spinlock.
1374 Note that it uses the fact, that this function is called
1375 with locally disabled BH and that sk cannot be already spinlocked.
1378 inet
->tos
= skb
->nh
.iph
->tos
;
1379 sk
->sk_priority
= skb
->priority
;
1380 sk
->sk_protocol
= skb
->nh
.iph
->protocol
;
1381 ip_append_data(sk
, ip_reply_glue_bits
, arg
->iov
->iov_base
, len
, 0,
1382 &ipc
, rt
, MSG_DONTWAIT
);
1383 if ((skb
= skb_peek(&sk
->sk_write_queue
)) != NULL
) {
1384 if (arg
->csumoffset
>= 0)
1385 *((u16
*)skb
->h
.raw
+ arg
->csumoffset
) = csum_fold(csum_add(skb
->csum
, arg
->csum
));
1386 skb
->ip_summed
= CHECKSUM_NONE
;
1387 ip_push_pending_frames(sk
);
1395 void __init
ip_init(void)
1400 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1401 igmp_mc_proc_init();
1405 EXPORT_SYMBOL(ip_generic_getfrag
);
1406 EXPORT_SYMBOL(ip_queue_xmit
);
1407 EXPORT_SYMBOL(ip_send_check
);