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>
53 #include <linux/string.h>
54 #include <linux/errno.h>
55 #include <linux/highmem.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 __read_mostly
= 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 skb_reset_mac_header(newskb
);
99 __skb_pull(newskb
, skb_network_offset(newskb
));
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 __be32 saddr
, __be32 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. */
128 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
129 skb_reset_network_header(skb
);
133 iph
->tos
= inet
->tos
;
134 if (ip_dont_fragment(sk
, &rt
->u
.dst
))
135 iph
->frag_off
= htons(IP_DF
);
138 iph
->ttl
= ip_select_ttl(inet
, &rt
->u
.dst
);
139 iph
->daddr
= rt
->rt_dst
;
140 iph
->saddr
= rt
->rt_src
;
141 iph
->protocol
= sk
->sk_protocol
;
142 iph
->tot_len
= htons(skb
->len
);
143 ip_select_ident(iph
, &rt
->u
.dst
, sk
);
145 if (opt
&& opt
->optlen
) {
146 iph
->ihl
+= opt
->optlen
>>2;
147 ip_options_build(skb
, opt
, daddr
, rt
, 0);
151 skb
->priority
= sk
->sk_priority
;
154 return NF_HOOK(PF_INET
, NF_IP_LOCAL_OUT
, skb
, NULL
, rt
->u
.dst
.dev
,
158 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
160 static inline int ip_finish_output2(struct sk_buff
*skb
)
162 struct dst_entry
*dst
= skb
->dst
;
163 struct net_device
*dev
= dst
->dev
;
164 int hh_len
= LL_RESERVED_SPACE(dev
);
166 /* Be paranoid, rather than too clever. */
167 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->hard_header
)) {
168 struct sk_buff
*skb2
;
170 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
176 skb_set_owner_w(skb2
, skb
->sk
);
182 return neigh_hh_output(dst
->hh
, skb
);
183 else if (dst
->neighbour
)
184 return dst
->neighbour
->output(skb
);
187 printk(KERN_DEBUG
"ip_finish_output2: No header cache and no neighbour!\n");
192 static inline int ip_finish_output(struct sk_buff
*skb
)
194 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
195 /* Policy lookup after SNAT yielded a new policy */
196 if (skb
->dst
->xfrm
!= NULL
) {
197 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
198 return dst_output(skb
);
201 if (skb
->len
> dst_mtu(skb
->dst
) && !skb_is_gso(skb
))
202 return ip_fragment(skb
, ip_finish_output2
);
204 return ip_finish_output2(skb
);
207 int ip_mc_output(struct sk_buff
*skb
)
209 struct sock
*sk
= skb
->sk
;
210 struct rtable
*rt
= (struct rtable
*)skb
->dst
;
211 struct net_device
*dev
= rt
->u
.dst
.dev
;
214 * If the indicated interface is up and running, send the packet.
216 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS
);
219 skb
->protocol
= htons(ETH_P_IP
);
222 * Multicasts are looped back for other local users
225 if (rt
->rt_flags
&RTCF_MULTICAST
) {
226 if ((!sk
|| inet_sk(sk
)->mc_loop
)
227 #ifdef CONFIG_IP_MROUTE
228 /* Small optimization: do not loopback not local frames,
229 which returned after forwarding; they will be dropped
230 by ip_mr_input in any case.
231 Note, that local frames are looped back to be delivered
234 This check is duplicated in ip_mr_input at the moment.
236 && ((rt
->rt_flags
&RTCF_LOCAL
) || !(IPCB(skb
)->flags
&IPSKB_FORWARDED
))
239 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
241 NF_HOOK(PF_INET
, NF_IP_POST_ROUTING
, newskb
, NULL
,
243 ip_dev_loopback_xmit
);
246 /* Multicasts with ttl 0 must not go beyond the host */
248 if (ip_hdr(skb
)->ttl
== 0) {
254 if (rt
->rt_flags
&RTCF_BROADCAST
) {
255 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
257 NF_HOOK(PF_INET
, NF_IP_POST_ROUTING
, newskb
, NULL
,
258 newskb
->dev
, ip_dev_loopback_xmit
);
261 return NF_HOOK_COND(PF_INET
, NF_IP_POST_ROUTING
, skb
, NULL
, skb
->dev
,
263 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
266 int ip_output(struct sk_buff
*skb
)
268 struct net_device
*dev
= skb
->dst
->dev
;
270 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS
);
273 skb
->protocol
= htons(ETH_P_IP
);
275 return NF_HOOK_COND(PF_INET
, NF_IP_POST_ROUTING
, skb
, NULL
, dev
,
277 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
280 int ip_queue_xmit(struct sk_buff
*skb
, int ipfragok
)
282 struct sock
*sk
= skb
->sk
;
283 struct inet_sock
*inet
= inet_sk(sk
);
284 struct ip_options
*opt
= inet
->opt
;
288 /* Skip all of this if the packet is already routed,
289 * f.e. by something like SCTP.
291 rt
= (struct rtable
*) skb
->dst
;
295 /* Make sure we can route this packet. */
296 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
300 /* Use correct destination address if we have options. */
306 struct flowi fl
= { .oif
= sk
->sk_bound_dev_if
,
309 .saddr
= inet
->saddr
,
310 .tos
= RT_CONN_FLAGS(sk
) } },
311 .proto
= sk
->sk_protocol
,
313 { .sport
= inet
->sport
,
314 .dport
= inet
->dport
} } };
316 /* If this fails, retransmit mechanism of transport layer will
317 * keep trying until route appears or the connection times
320 security_sk_classify_flow(sk
, &fl
);
321 if (ip_route_output_flow(&rt
, &fl
, sk
, 0))
324 sk_setup_caps(sk
, &rt
->u
.dst
);
326 skb
->dst
= dst_clone(&rt
->u
.dst
);
329 if (opt
&& opt
->is_strictroute
&& rt
->rt_dst
!= rt
->rt_gateway
)
332 /* OK, we know where to send it, allocate and build IP header. */
333 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
334 skb_reset_network_header(skb
);
336 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
337 iph
->tot_len
= htons(skb
->len
);
338 if (ip_dont_fragment(sk
, &rt
->u
.dst
) && !ipfragok
)
339 iph
->frag_off
= htons(IP_DF
);
342 iph
->ttl
= ip_select_ttl(inet
, &rt
->u
.dst
);
343 iph
->protocol
= sk
->sk_protocol
;
344 iph
->saddr
= rt
->rt_src
;
345 iph
->daddr
= rt
->rt_dst
;
346 /* Transport layer set skb->h.foo itself. */
348 if (opt
&& opt
->optlen
) {
349 iph
->ihl
+= opt
->optlen
>> 2;
350 ip_options_build(skb
, opt
, inet
->daddr
, rt
, 0);
353 ip_select_ident_more(iph
, &rt
->u
.dst
, sk
,
354 (skb_shinfo(skb
)->gso_segs
?: 1) - 1);
356 /* Add an IP checksum. */
359 skb
->priority
= sk
->sk_priority
;
361 return NF_HOOK(PF_INET
, NF_IP_LOCAL_OUT
, skb
, NULL
, rt
->u
.dst
.dev
,
365 IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES
);
367 return -EHOSTUNREACH
;
371 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
373 to
->pkt_type
= from
->pkt_type
;
374 to
->priority
= from
->priority
;
375 to
->protocol
= from
->protocol
;
376 dst_release(to
->dst
);
377 to
->dst
= dst_clone(from
->dst
);
379 to
->mark
= from
->mark
;
381 /* Copy the flags to each fragment. */
382 IPCB(to
)->flags
= IPCB(from
)->flags
;
384 #ifdef CONFIG_NET_SCHED
385 to
->tc_index
= from
->tc_index
;
388 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
389 to
->ipvs_property
= from
->ipvs_property
;
391 skb_copy_secmark(to
, from
);
395 * This IP datagram is too large to be sent in one piece. Break it up into
396 * smaller pieces (each of size equal to IP header plus
397 * a block of the data of the original IP data part) that will yet fit in a
398 * single device frame, and queue such a frame for sending.
401 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
406 struct net_device
*dev
;
407 struct sk_buff
*skb2
;
408 unsigned int mtu
, hlen
, left
, len
, ll_rs
, pad
;
410 __be16 not_last_frag
;
411 struct rtable
*rt
= (struct rtable
*)skb
->dst
;
417 * Point into the IP datagram header.
422 if (unlikely((iph
->frag_off
& htons(IP_DF
)) && !skb
->local_df
)) {
423 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS
);
424 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
425 htonl(dst_mtu(&rt
->u
.dst
)));
431 * Setup starting values.
435 mtu
= dst_mtu(&rt
->u
.dst
) - hlen
; /* Size of data space */
436 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
438 /* When frag_list is given, use it. First, check its validity:
439 * some transformers could create wrong frag_list or break existing
440 * one, it is not prohibited. In this case fall back to copying.
442 * LATER: this step can be merged to real generation of fragments,
443 * we can switch to copy when see the first bad fragment.
445 if (skb_shinfo(skb
)->frag_list
) {
446 struct sk_buff
*frag
;
447 int first_len
= skb_pagelen(skb
);
449 if (first_len
- hlen
> mtu
||
450 ((first_len
- hlen
) & 7) ||
451 (iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) ||
455 for (frag
= skb_shinfo(skb
)->frag_list
; frag
; frag
= frag
->next
) {
456 /* Correct geometry. */
457 if (frag
->len
> mtu
||
458 ((frag
->len
& 7) && frag
->next
) ||
459 skb_headroom(frag
) < hlen
)
462 /* Partially cloned skb? */
463 if (skb_shared(frag
))
470 frag
->destructor
= sock_wfree
;
471 skb
->truesize
-= frag
->truesize
;
475 /* Everything is OK. Generate! */
479 frag
= skb_shinfo(skb
)->frag_list
;
480 skb_shinfo(skb
)->frag_list
= NULL
;
481 skb
->data_len
= first_len
- skb_headlen(skb
);
482 skb
->len
= first_len
;
483 iph
->tot_len
= htons(first_len
);
484 iph
->frag_off
= htons(IP_MF
);
488 /* Prepare header of the next frame,
489 * before previous one went down. */
491 frag
->ip_summed
= CHECKSUM_NONE
;
492 skb_reset_transport_header(frag
);
493 __skb_push(frag
, hlen
);
494 skb_reset_network_header(frag
);
495 memcpy(skb_network_header(frag
), iph
, hlen
);
497 iph
->tot_len
= htons(frag
->len
);
498 ip_copy_metadata(frag
, skb
);
500 ip_options_fragment(frag
);
501 offset
+= skb
->len
- hlen
;
502 iph
->frag_off
= htons(offset
>>3);
503 if (frag
->next
!= NULL
)
504 iph
->frag_off
|= htons(IP_MF
);
505 /* Ready, complete checksum */
512 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES
);
522 IP_INC_STATS(IPSTATS_MIB_FRAGOKS
);
531 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS
);
536 left
= skb
->len
- hlen
; /* Space per frame */
537 ptr
= raw
+ hlen
; /* Where to start from */
539 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
540 * we need to make room for the encapsulating header
542 pad
= nf_bridge_pad(skb
);
543 ll_rs
= LL_RESERVED_SPACE_EXTRA(rt
->u
.dst
.dev
, pad
);
547 * Fragment the datagram.
550 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
551 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
554 * Keep copying data until we run out.
559 /* IF: it doesn't fit, use 'mtu' - the data space left */
562 /* IF: we are not sending upto and including the packet end
563 then align the next start on an eight byte boundary */
571 if ((skb2
= alloc_skb(len
+hlen
+ll_rs
, GFP_ATOMIC
)) == NULL
) {
572 NETDEBUG(KERN_INFO
"IP: frag: no memory for new fragment!\n");
578 * Set up data on packet
581 ip_copy_metadata(skb2
, skb
);
582 skb_reserve(skb2
, ll_rs
);
583 skb_put(skb2
, len
+ hlen
);
584 skb_reset_network_header(skb2
);
585 skb2
->transport_header
= skb2
->network_header
+ hlen
;
588 * Charge the memory for the fragment to any owner
593 skb_set_owner_w(skb2
, skb
->sk
);
596 * Copy the packet header into the new buffer.
599 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
602 * Copy a block of the IP datagram.
604 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
609 * Fill in the new header fields.
612 iph
->frag_off
= htons((offset
>> 3));
614 /* ANK: dirty, but effective trick. Upgrade options only if
615 * the segment to be fragmented was THE FIRST (otherwise,
616 * options are already fixed) and make it ONCE
617 * on the initial skb, so that all the following fragments
618 * will inherit fixed options.
621 ip_options_fragment(skb
);
624 * Added AC : If we are fragmenting a fragment that's not the
625 * last fragment then keep MF on each bit
627 if (left
> 0 || not_last_frag
)
628 iph
->frag_off
|= htons(IP_MF
);
633 * Put this fragment into the sending queue.
635 iph
->tot_len
= htons(len
+ hlen
);
643 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES
);
646 IP_INC_STATS(IPSTATS_MIB_FRAGOKS
);
651 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS
);
655 EXPORT_SYMBOL(ip_fragment
);
658 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
660 struct iovec
*iov
= from
;
662 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
663 if (memcpy_fromiovecend(to
, iov
, offset
, len
) < 0)
667 if (csum_partial_copy_fromiovecend(to
, iov
, offset
, len
, &csum
) < 0)
669 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
675 csum_page(struct page
*page
, int offset
, int copy
)
680 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
685 static inline int ip_ufo_append_data(struct sock
*sk
,
686 int getfrag(void *from
, char *to
, int offset
, int len
,
687 int odd
, struct sk_buff
*skb
),
688 void *from
, int length
, int hh_len
, int fragheaderlen
,
689 int transhdrlen
, int mtu
,unsigned int flags
)
694 /* There is support for UDP fragmentation offload by network
695 * device, so create one single skb packet containing complete
698 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
699 skb
= sock_alloc_send_skb(sk
,
700 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
701 (flags
& MSG_DONTWAIT
), &err
);
706 /* reserve space for Hardware header */
707 skb_reserve(skb
, hh_len
);
709 /* create space for UDP/IP header */
710 skb_put(skb
,fragheaderlen
+ transhdrlen
);
712 /* initialize network header pointer */
713 skb_reset_network_header(skb
);
715 /* initialize protocol header pointer */
716 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
718 skb
->ip_summed
= CHECKSUM_PARTIAL
;
720 sk
->sk_sndmsg_off
= 0;
723 err
= skb_append_datato_frags(sk
,skb
, getfrag
, from
,
724 (length
- transhdrlen
));
726 /* specify the length of each IP datagram fragment*/
727 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
728 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
729 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
733 /* There is not enough support do UFO ,
734 * so follow normal path
741 * ip_append_data() and ip_append_page() can make one large IP datagram
742 * from many pieces of data. Each pieces will be holded on the socket
743 * until ip_push_pending_frames() is called. Each piece can be a page
746 * Not only UDP, other transport protocols - e.g. raw sockets - can use
747 * this interface potentially.
749 * LATER: length must be adjusted by pad at tail, when it is required.
751 int ip_append_data(struct sock
*sk
,
752 int getfrag(void *from
, char *to
, int offset
, int len
,
753 int odd
, struct sk_buff
*skb
),
754 void *from
, int length
, int transhdrlen
,
755 struct ipcm_cookie
*ipc
, struct rtable
*rt
,
758 struct inet_sock
*inet
= inet_sk(sk
);
761 struct ip_options
*opt
= NULL
;
768 unsigned int maxfraglen
, fragheaderlen
;
769 int csummode
= CHECKSUM_NONE
;
774 if (skb_queue_empty(&sk
->sk_write_queue
)) {
780 if (inet
->cork
.opt
== NULL
) {
781 inet
->cork
.opt
= kmalloc(sizeof(struct ip_options
) + 40, sk
->sk_allocation
);
782 if (unlikely(inet
->cork
.opt
== NULL
))
785 memcpy(inet
->cork
.opt
, opt
, sizeof(struct ip_options
)+opt
->optlen
);
786 inet
->cork
.flags
|= IPCORK_OPT
;
787 inet
->cork
.addr
= ipc
->addr
;
789 dst_hold(&rt
->u
.dst
);
790 inet
->cork
.fragsize
= mtu
= dst_mtu(rt
->u
.dst
.path
);
792 inet
->cork
.length
= 0;
793 sk
->sk_sndmsg_page
= NULL
;
794 sk
->sk_sndmsg_off
= 0;
795 if ((exthdrlen
= rt
->u
.dst
.header_len
) != 0) {
797 transhdrlen
+= exthdrlen
;
801 if (inet
->cork
.flags
& IPCORK_OPT
)
802 opt
= inet
->cork
.opt
;
806 mtu
= inet
->cork
.fragsize
;
808 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
810 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
811 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
813 if (inet
->cork
.length
+ length
> 0xFFFF - fragheaderlen
) {
814 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->dport
, mtu
-exthdrlen
);
819 * transhdrlen > 0 means that this is the first fragment and we wish
820 * it won't be fragmented in the future.
823 length
+ fragheaderlen
<= mtu
&&
824 rt
->u
.dst
.dev
->features
& NETIF_F_ALL_CSUM
&&
826 csummode
= CHECKSUM_PARTIAL
;
828 inet
->cork
.length
+= length
;
829 if (((length
> mtu
) && (sk
->sk_protocol
== IPPROTO_UDP
)) &&
830 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
832 err
= ip_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
833 fragheaderlen
, transhdrlen
, mtu
,
840 /* So, what's going on in the loop below?
842 * We use calculated fragment length to generate chained skb,
843 * each of segments is IP fragment ready for sending to network after
844 * adding appropriate IP header.
847 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
851 /* Check if the remaining data fits into current packet. */
852 copy
= mtu
- skb
->len
;
854 copy
= maxfraglen
- skb
->len
;
857 unsigned int datalen
;
858 unsigned int fraglen
;
859 unsigned int fraggap
;
860 unsigned int alloclen
;
861 struct sk_buff
*skb_prev
;
865 fraggap
= skb_prev
->len
- maxfraglen
;
870 * If remaining data exceeds the mtu,
871 * we know we need more fragment(s).
873 datalen
= length
+ fraggap
;
874 if (datalen
> mtu
- fragheaderlen
)
875 datalen
= maxfraglen
- fragheaderlen
;
876 fraglen
= datalen
+ fragheaderlen
;
878 if ((flags
& MSG_MORE
) &&
879 !(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
882 alloclen
= datalen
+ fragheaderlen
;
884 /* The last fragment gets additional space at tail.
885 * Note, with MSG_MORE we overallocate on fragments,
886 * because we have no idea what fragment will be
889 if (datalen
== length
+ fraggap
)
890 alloclen
+= rt
->u
.dst
.trailer_len
;
893 skb
= sock_alloc_send_skb(sk
,
894 alloclen
+ hh_len
+ 15,
895 (flags
& MSG_DONTWAIT
), &err
);
898 if (atomic_read(&sk
->sk_wmem_alloc
) <=
900 skb
= sock_wmalloc(sk
,
901 alloclen
+ hh_len
+ 15, 1,
903 if (unlikely(skb
== NULL
))
910 * Fill in the control structures
912 skb
->ip_summed
= csummode
;
914 skb_reserve(skb
, hh_len
);
917 * Find where to start putting bytes.
919 data
= skb_put(skb
, fraglen
);
920 skb_set_network_header(skb
, exthdrlen
);
921 skb
->transport_header
= (skb
->network_header
+
923 data
+= fragheaderlen
;
926 skb
->csum
= skb_copy_and_csum_bits(
927 skb_prev
, maxfraglen
,
928 data
+ transhdrlen
, fraggap
, 0);
929 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
932 pskb_trim_unique(skb_prev
, maxfraglen
);
935 copy
= datalen
- transhdrlen
- fraggap
;
936 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
943 length
-= datalen
- fraggap
;
946 csummode
= CHECKSUM_NONE
;
949 * Put the packet on the pending queue.
951 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
958 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
)) {
962 if (getfrag(from
, skb_put(skb
, copy
),
963 offset
, copy
, off
, skb
) < 0) {
964 __skb_trim(skb
, off
);
969 int i
= skb_shinfo(skb
)->nr_frags
;
970 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
971 struct page
*page
= sk
->sk_sndmsg_page
;
972 int off
= sk
->sk_sndmsg_off
;
975 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
978 if (page
!= frag
->page
) {
979 if (i
== MAX_SKB_FRAGS
) {
984 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
985 frag
= &skb_shinfo(skb
)->frags
[i
];
987 } else if (i
< MAX_SKB_FRAGS
) {
988 if (copy
> PAGE_SIZE
)
990 page
= alloc_pages(sk
->sk_allocation
, 0);
995 sk
->sk_sndmsg_page
= page
;
996 sk
->sk_sndmsg_off
= 0;
998 skb_fill_page_desc(skb
, i
, page
, 0, 0);
999 frag
= &skb_shinfo(skb
)->frags
[i
];
1000 skb
->truesize
+= PAGE_SIZE
;
1001 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1006 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1010 sk
->sk_sndmsg_off
+= copy
;
1013 skb
->data_len
+= copy
;
1022 inet
->cork
.length
-= length
;
1023 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS
);
1027 ssize_t
ip_append_page(struct sock
*sk
, struct page
*page
,
1028 int offset
, size_t size
, int flags
)
1030 struct inet_sock
*inet
= inet_sk(sk
);
1031 struct sk_buff
*skb
;
1033 struct ip_options
*opt
= NULL
;
1038 unsigned int maxfraglen
, fragheaderlen
, fraggap
;
1043 if (flags
&MSG_PROBE
)
1046 if (skb_queue_empty(&sk
->sk_write_queue
))
1050 if (inet
->cork
.flags
& IPCORK_OPT
)
1051 opt
= inet
->cork
.opt
;
1053 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
1056 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
1057 mtu
= inet
->cork
.fragsize
;
1059 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1060 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1062 if (inet
->cork
.length
+ size
> 0xFFFF - fragheaderlen
) {
1063 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->dport
, mtu
);
1067 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1070 inet
->cork
.length
+= size
;
1071 if ((sk
->sk_protocol
== IPPROTO_UDP
) &&
1072 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
1073 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1074 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1081 if (skb_is_gso(skb
))
1085 /* Check if the remaining data fits into current packet. */
1086 len
= mtu
- skb
->len
;
1088 len
= maxfraglen
- skb
->len
;
1091 struct sk_buff
*skb_prev
;
1095 fraggap
= skb_prev
->len
- maxfraglen
;
1097 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1098 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1099 if (unlikely(!skb
)) {
1105 * Fill in the control structures
1107 skb
->ip_summed
= CHECKSUM_NONE
;
1109 skb_reserve(skb
, hh_len
);
1112 * Find where to start putting bytes.
1114 skb_put(skb
, fragheaderlen
+ fraggap
);
1115 skb_reset_network_header(skb
);
1116 skb
->transport_header
= (skb
->network_header
+
1119 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1121 skb_transport_header(skb
),
1123 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1125 pskb_trim_unique(skb_prev
, maxfraglen
);
1129 * Put the packet on the pending queue.
1131 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1135 i
= skb_shinfo(skb
)->nr_frags
;
1138 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1139 skb_shinfo(skb
)->frags
[i
-1].size
+= len
;
1140 } else if (i
< MAX_SKB_FRAGS
) {
1142 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1148 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1150 csum
= csum_page(page
, offset
, len
);
1151 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1155 skb
->data_len
+= len
;
1162 inet
->cork
.length
-= size
;
1163 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS
);
1168 * Combined all pending IP fragments on the socket as one IP datagram
1169 * and push them out.
1171 int ip_push_pending_frames(struct sock
*sk
)
1173 struct sk_buff
*skb
, *tmp_skb
;
1174 struct sk_buff
**tail_skb
;
1175 struct inet_sock
*inet
= inet_sk(sk
);
1176 struct ip_options
*opt
= NULL
;
1177 struct rtable
*rt
= inet
->cork
.rt
;
1183 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1185 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1187 /* move skb->data to ip header from ext header */
1188 if (skb
->data
< skb_network_header(skb
))
1189 __skb_pull(skb
, skb_network_offset(skb
));
1190 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1191 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1192 *tail_skb
= tmp_skb
;
1193 tail_skb
= &(tmp_skb
->next
);
1194 skb
->len
+= tmp_skb
->len
;
1195 skb
->data_len
+= tmp_skb
->len
;
1196 skb
->truesize
+= tmp_skb
->truesize
;
1197 __sock_put(tmp_skb
->sk
);
1198 tmp_skb
->destructor
= NULL
;
1202 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1203 * to fragment the frame generated here. No matter, what transforms
1204 * how transforms change size of the packet, it will come out.
1206 if (inet
->pmtudisc
!= IP_PMTUDISC_DO
)
1209 /* DF bit is set when we want to see DF on outgoing frames.
1210 * If local_df is set too, we still allow to fragment this frame
1212 if (inet
->pmtudisc
== IP_PMTUDISC_DO
||
1213 (skb
->len
<= dst_mtu(&rt
->u
.dst
) &&
1214 ip_dont_fragment(sk
, &rt
->u
.dst
)))
1217 if (inet
->cork
.flags
& IPCORK_OPT
)
1218 opt
= inet
->cork
.opt
;
1220 if (rt
->rt_type
== RTN_MULTICAST
)
1223 ttl
= ip_select_ttl(inet
, &rt
->u
.dst
);
1225 iph
= (struct iphdr
*)skb
->data
;
1229 iph
->ihl
+= opt
->optlen
>>2;
1230 ip_options_build(skb
, opt
, inet
->cork
.addr
, rt
, 0);
1232 iph
->tos
= inet
->tos
;
1233 iph
->tot_len
= htons(skb
->len
);
1235 ip_select_ident(iph
, &rt
->u
.dst
, sk
);
1237 iph
->protocol
= sk
->sk_protocol
;
1238 iph
->saddr
= rt
->rt_src
;
1239 iph
->daddr
= rt
->rt_dst
;
1242 skb
->priority
= sk
->sk_priority
;
1243 skb
->dst
= dst_clone(&rt
->u
.dst
);
1245 /* Netfilter gets whole the not fragmented skb. */
1246 err
= NF_HOOK(PF_INET
, NF_IP_LOCAL_OUT
, skb
, NULL
,
1247 skb
->dst
->dev
, dst_output
);
1250 err
= inet
->recverr
? net_xmit_errno(err
) : 0;
1256 inet
->cork
.flags
&= ~IPCORK_OPT
;
1257 kfree(inet
->cork
.opt
);
1258 inet
->cork
.opt
= NULL
;
1259 if (inet
->cork
.rt
) {
1260 ip_rt_put(inet
->cork
.rt
);
1261 inet
->cork
.rt
= NULL
;
1266 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS
);
1271 * Throw away all pending data on the socket.
1273 void ip_flush_pending_frames(struct sock
*sk
)
1275 struct inet_sock
*inet
= inet_sk(sk
);
1276 struct sk_buff
*skb
;
1278 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
)
1281 inet
->cork
.flags
&= ~IPCORK_OPT
;
1282 kfree(inet
->cork
.opt
);
1283 inet
->cork
.opt
= NULL
;
1284 if (inet
->cork
.rt
) {
1285 ip_rt_put(inet
->cork
.rt
);
1286 inet
->cork
.rt
= NULL
;
1292 * Fetch data from kernel space and fill in checksum if needed.
1294 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1295 int len
, int odd
, struct sk_buff
*skb
)
1299 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1300 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1305 * Generic function to send a packet as reply to another packet.
1306 * Used to send TCP resets so far. ICMP should use this function too.
1308 * Should run single threaded per socket because it uses the sock
1309 * structure to pass arguments.
1311 * LATER: switch from ip_build_xmit to ip_append_*
1313 void ip_send_reply(struct sock
*sk
, struct sk_buff
*skb
, struct ip_reply_arg
*arg
,
1316 struct inet_sock
*inet
= inet_sk(sk
);
1318 struct ip_options opt
;
1321 struct ipcm_cookie ipc
;
1323 struct rtable
*rt
= (struct rtable
*)skb
->dst
;
1325 if (ip_options_echo(&replyopts
.opt
, skb
))
1328 daddr
= ipc
.addr
= rt
->rt_src
;
1331 if (replyopts
.opt
.optlen
) {
1332 ipc
.opt
= &replyopts
.opt
;
1335 daddr
= replyopts
.opt
.faddr
;
1339 struct flowi fl
= { .nl_u
= { .ip4_u
=
1341 .saddr
= rt
->rt_spec_dst
,
1342 .tos
= RT_TOS(ip_hdr(skb
)->tos
) } },
1343 /* Not quite clean, but right. */
1345 { .sport
= tcp_hdr(skb
)->dest
,
1346 .dport
= tcp_hdr(skb
)->source
} },
1347 .proto
= sk
->sk_protocol
};
1348 security_skb_classify_flow(skb
, &fl
);
1349 if (ip_route_output_key(&rt
, &fl
))
1353 /* And let IP do all the hard work.
1355 This chunk is not reenterable, hence spinlock.
1356 Note that it uses the fact, that this function is called
1357 with locally disabled BH and that sk cannot be already spinlocked.
1360 inet
->tos
= ip_hdr(skb
)->tos
;
1361 sk
->sk_priority
= skb
->priority
;
1362 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1363 ip_append_data(sk
, ip_reply_glue_bits
, arg
->iov
->iov_base
, len
, 0,
1364 &ipc
, rt
, MSG_DONTWAIT
);
1365 if ((skb
= skb_peek(&sk
->sk_write_queue
)) != NULL
) {
1366 if (arg
->csumoffset
>= 0)
1367 *((__sum16
*)skb_transport_header(skb
) +
1368 arg
->csumoffset
) = csum_fold(csum_add(skb
->csum
,
1370 skb
->ip_summed
= CHECKSUM_NONE
;
1371 ip_push_pending_frames(sk
);
1379 void __init
ip_init(void)
1384 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1385 igmp_mc_proc_init();
1389 EXPORT_SYMBOL(ip_generic_getfrag
);
1390 EXPORT_SYMBOL(ip_queue_xmit
);
1391 EXPORT_SYMBOL(ip_send_check
);