2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The Internet Protocol (IP) output module.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <Alan.Cox@linux.org>
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16 * Hirokazu Takahashi, <taka@valinux.co.jp>
18 * See ip_input.c for original log
21 * Alan Cox : Missing nonblock feature in ip_build_xmit.
22 * Mike Kilburn : htons() missing in ip_build_xmit.
23 * Bradford Johnson: Fix faulty handling of some frames when
25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
26 * (in case if packet not accepted by
27 * output firewall rules)
28 * Mike McLagan : Routing by source
29 * Alexey Kuznetsov: use new route cache
30 * Andi Kleen: Fix broken PMTU recovery and remove
31 * some redundant tests.
32 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
33 * Andi Kleen : Replace ip_reply with ip_send_reply.
34 * Andi Kleen : Split fast and slow ip_build_xmit path
35 * for decreased register pressure on x86
36 * and more readibility.
37 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
38 * silently drop skb instead of failing with -EPERM.
39 * Detlev Wengorz : Copy protocol for fragments.
40 * Hirokazu Takahashi: HW checksumming for outgoing UDP
42 * Hirokazu Takahashi: sendfile() on UDP works now.
45 #include <asm/uaccess.h>
46 #include <asm/system.h>
47 #include <linux/module.h>
48 #include <linux/types.h>
49 #include <linux/kernel.h>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
54 #include <linux/slab.h>
56 #include <linux/socket.h>
57 #include <linux/sockios.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/etherdevice.h>
62 #include <linux/proc_fs.h>
63 #include <linux/stat.h>
64 #include <linux/init.h>
68 #include <net/protocol.h>
69 #include <net/route.h>
71 #include <linux/skbuff.h>
75 #include <net/checksum.h>
76 #include <net/inetpeer.h>
77 #include <linux/igmp.h>
78 #include <linux/netfilter_ipv4.h>
79 #include <linux/netfilter_bridge.h>
80 #include <linux/mroute.h>
81 #include <linux/netlink.h>
82 #include <linux/tcp.h>
87 int sysctl_ip_default_ttl __read_mostly
= IPDEFTTL
;
89 /* Generate a checksum for an outgoing IP datagram. */
90 __inline__
void ip_send_check(struct iphdr
*iph
)
93 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
95 EXPORT_SYMBOL(ip_send_check
);
97 int __ip_local_out(struct sk_buff
*skb
)
99 struct iphdr
*iph
= ip_hdr(skb
);
101 iph
->tot_len
= htons(skb
->len
);
104 /* Mark skb to identify SMB data packet */
105 if (ip_hdr(skb
)->protocol
== IPPROTO_TCP
)
106 skb
->tcpf_smb
= (tcp_hdr(skb
)->source
== htons(0x01bd));
108 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, skb
, NULL
,
109 skb_dst(skb
)->dev
, dst_output
);
112 int ip_local_out(struct sk_buff
*skb
)
116 err
= __ip_local_out(skb
);
117 if (likely(err
== 1))
118 err
= dst_output(skb
);
122 EXPORT_SYMBOL_GPL(ip_local_out
);
124 /* dev_loopback_xmit for use with netfilter. */
125 static int ip_dev_loopback_xmit(struct sk_buff
*newskb
)
127 skb_reset_mac_header(newskb
);
128 __skb_pull(newskb
, skb_network_offset(newskb
));
129 newskb
->pkt_type
= PACKET_LOOPBACK
;
130 newskb
->ip_summed
= CHECKSUM_UNNECESSARY
;
131 WARN_ON(!skb_dst(newskb
));
136 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
138 int ttl
= inet
->uc_ttl
;
141 ttl
= dst_metric(dst
, RTAX_HOPLIMIT
);
146 * Add an ip header to a skbuff and send it out.
149 int ip_build_and_send_pkt(struct sk_buff
*skb
, struct sock
*sk
,
150 __be32 saddr
, __be32 daddr
, struct ip_options
*opt
)
152 struct inet_sock
*inet
= inet_sk(sk
);
153 struct rtable
*rt
= skb_rtable(skb
);
156 /* Build the IP header. */
157 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
158 skb_reset_network_header(skb
);
162 iph
->tos
= inet
->tos
;
163 if (ip_dont_fragment(sk
, &rt
->dst
))
164 iph
->frag_off
= htons(IP_DF
);
167 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
168 iph
->daddr
= rt
->rt_dst
;
169 iph
->saddr
= rt
->rt_src
;
170 iph
->protocol
= sk
->sk_protocol
;
171 ip_select_ident(iph
, &rt
->dst
, sk
);
173 if (opt
&& opt
->optlen
) {
174 iph
->ihl
+= opt
->optlen
>>2;
175 ip_options_build(skb
, opt
, daddr
, rt
, 0);
178 skb
->priority
= sk
->sk_priority
;
179 skb
->mark
= sk
->sk_mark
;
182 return ip_local_out(skb
);
184 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
186 static inline int ip_finish_output2(struct sk_buff
*skb
)
188 struct dst_entry
*dst
= skb_dst(skb
);
189 struct rtable
*rt
= (struct rtable
*)dst
;
190 struct net_device
*dev
= dst
->dev
;
191 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
193 if (rt
->rt_type
== RTN_MULTICAST
) {
194 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTMCAST
, skb
->len
);
195 } else if (rt
->rt_type
== RTN_BROADCAST
)
196 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTBCAST
, skb
->len
);
198 /* Be paranoid, rather than too clever. */
199 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
200 struct sk_buff
*skb2
;
202 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
208 skb_set_owner_w(skb2
, skb
->sk
);
214 return neigh_hh_output(dst
->hh
, skb
);
215 else if (dst
->neighbour
)
216 return dst
->neighbour
->output(skb
);
219 printk(KERN_DEBUG
"ip_finish_output2: No header cache and no neighbour!\n");
224 static inline int ip_skb_dst_mtu(struct sk_buff
*skb
)
226 struct inet_sock
*inet
= skb
->sk
? inet_sk(skb
->sk
) : NULL
;
228 return (inet
&& inet
->pmtudisc
== IP_PMTUDISC_PROBE
) ?
229 skb_dst(skb
)->dev
->mtu
: dst_mtu(skb_dst(skb
));
232 static int ip_finish_output(struct sk_buff
*skb
)
234 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
235 /* Policy lookup after SNAT yielded a new policy */
236 if (skb_dst(skb
)->xfrm
!= NULL
) {
237 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
238 return dst_output(skb
);
241 if (skb
->len
> ip_skb_dst_mtu(skb
) && !skb_is_gso(skb
))
242 return ip_fragment(skb
, ip_finish_output2
);
244 return ip_finish_output2(skb
);
247 int ip_mc_output(struct sk_buff
*skb
)
249 struct sock
*sk
= skb
->sk
;
250 struct rtable
*rt
= skb_rtable(skb
);
251 struct net_device
*dev
= rt
->dst
.dev
;
254 * If the indicated interface is up and running, send the packet.
256 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
259 skb
->protocol
= htons(ETH_P_IP
);
262 * Multicasts are looped back for other local users
265 if (rt
->rt_flags
&RTCF_MULTICAST
) {
267 #ifdef CONFIG_IP_MROUTE
268 /* Small optimization: do not loopback not local frames,
269 which returned after forwarding; they will be dropped
270 by ip_mr_input in any case.
271 Note, that local frames are looped back to be delivered
274 This check is duplicated in ip_mr_input at the moment.
277 ((rt
->rt_flags
& RTCF_LOCAL
) ||
278 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
281 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
283 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
284 newskb
, NULL
, newskb
->dev
,
285 ip_dev_loopback_xmit
);
288 /* Multicasts with ttl 0 must not go beyond the host */
290 if (ip_hdr(skb
)->ttl
== 0) {
296 if (rt
->rt_flags
&RTCF_BROADCAST
) {
297 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
299 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, newskb
,
300 NULL
, newskb
->dev
, ip_dev_loopback_xmit
);
303 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
,
304 skb
->dev
, ip_finish_output
,
305 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
308 int ip_output(struct sk_buff
*skb
)
310 struct net_device
*dev
= skb_dst(skb
)->dev
;
312 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
315 skb
->protocol
= htons(ETH_P_IP
);
317 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
, dev
,
319 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
322 int BCMFASTPATH_HOST
ip_queue_xmit(struct sk_buff
*skb
)
324 struct sock
*sk
= skb
->sk
;
325 struct inet_sock
*inet
= inet_sk(sk
);
326 struct ip_options
*opt
= inet
->opt
;
331 /* Skip all of this if the packet is already routed,
332 * f.e. by something like SCTP.
335 rt
= skb_rtable(skb
);
339 /* Make sure we can route this packet. */
340 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
344 /* Use correct destination address if we have options. */
345 daddr
= inet
->inet_daddr
;
350 struct flowi fl
= { .oif
= sk
->sk_bound_dev_if
,
354 .saddr
= inet
->inet_saddr
,
355 .tos
= RT_CONN_FLAGS(sk
) } },
356 .proto
= sk
->sk_protocol
,
357 .flags
= inet_sk_flowi_flags(sk
),
359 { .sport
= inet
->inet_sport
,
360 .dport
= inet
->inet_dport
} } };
362 /* If this fails, retransmit mechanism of transport layer will
363 * keep trying until route appears or the connection times
366 security_sk_classify_flow(sk
, &fl
);
367 if (ip_route_output_flow(sock_net(sk
), &rt
, &fl
, sk
, 0))
370 sk_setup_caps(sk
, &rt
->dst
);
372 skb_dst_set_noref(skb
, &rt
->dst
);
375 if (opt
&& opt
->is_strictroute
&& rt
->rt_dst
!= rt
->rt_gateway
)
378 /* OK, we know where to send it, allocate and build IP header. */
379 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
380 skb_reset_network_header(skb
);
382 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
383 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->local_df
)
384 iph
->frag_off
= htons(IP_DF
);
387 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
388 iph
->protocol
= sk
->sk_protocol
;
389 iph
->saddr
= rt
->rt_src
;
390 iph
->daddr
= rt
->rt_dst
;
391 /* Transport layer set skb->h.foo itself. */
393 if (opt
&& opt
->optlen
) {
394 iph
->ihl
+= opt
->optlen
>> 2;
395 ip_options_build(skb
, opt
, inet
->inet_daddr
, rt
, 0);
398 ip_select_ident_more(iph
, &rt
->dst
, sk
,
399 (skb_shinfo(skb
)->gso_segs
?: 1) - 1);
401 skb
->priority
= sk
->sk_priority
;
402 skb
->mark
= sk
->sk_mark
;
404 res
= ip_local_out(skb
);
410 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
412 return -EHOSTUNREACH
;
414 EXPORT_SYMBOL(ip_queue_xmit
);
417 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
419 to
->pkt_type
= from
->pkt_type
;
420 to
->priority
= from
->priority
;
421 to
->protocol
= from
->protocol
;
423 skb_dst_copy(to
, from
);
425 to
->mark
= from
->mark
;
427 /* Copy the flags to each fragment. */
428 IPCB(to
)->flags
= IPCB(from
)->flags
;
430 #ifdef CONFIG_NET_SCHED
431 to
->tc_index
= from
->tc_index
;
434 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
435 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
436 to
->nf_trace
= from
->nf_trace
;
438 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
439 to
->ipvs_property
= from
->ipvs_property
;
441 skb_copy_secmark(to
, from
);
445 * This IP datagram is too large to be sent in one piece. Break it up into
446 * smaller pieces (each of size equal to IP header plus
447 * a block of the data of the original IP data part) that will yet fit in a
448 * single device frame, and queue such a frame for sending.
451 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
455 struct net_device
*dev
;
456 struct sk_buff
*skb2
;
457 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
459 __be16 not_last_frag
;
460 struct rtable
*rt
= skb_rtable(skb
);
466 * Point into the IP datagram header.
471 if (unlikely((iph
->frag_off
& htons(IP_DF
)) && !skb
->local_df
)) {
472 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
473 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
474 htonl(ip_skb_dst_mtu(skb
)));
480 * Setup starting values.
484 mtu
= dst_mtu(&rt
->dst
) - hlen
; /* Size of data space */
485 #ifdef CONFIG_BRIDGE_NETFILTER
487 mtu
-= nf_bridge_mtu_reduction(skb
);
489 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
491 /* When frag_list is given, use it. First, check its validity:
492 * some transformers could create wrong frag_list or break existing
493 * one, it is not prohibited. In this case fall back to copying.
495 * LATER: this step can be merged to real generation of fragments,
496 * we can switch to copy when see the first bad fragment.
498 if (skb_has_frags(skb
)) {
499 struct sk_buff
*frag
, *frag2
;
500 int first_len
= skb_pagelen(skb
);
502 if (first_len
- hlen
> mtu
||
503 ((first_len
- hlen
) & 7) ||
504 (iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) ||
508 skb_walk_frags(skb
, frag
) {
509 /* Correct geometry. */
510 if (frag
->len
> mtu
||
511 ((frag
->len
& 7) && frag
->next
) ||
512 skb_headroom(frag
) < hlen
)
513 goto slow_path_clean
;
515 /* Partially cloned skb? */
516 if (skb_shared(frag
))
517 goto slow_path_clean
;
522 frag
->destructor
= sock_wfree
;
524 skb
->truesize
-= frag
->truesize
;
527 /* Everything is OK. Generate! */
531 frag
= skb_shinfo(skb
)->frag_list
;
532 skb_frag_list_init(skb
);
533 skb
->data_len
= first_len
- skb_headlen(skb
);
534 skb
->len
= first_len
;
535 iph
->tot_len
= htons(first_len
);
536 iph
->frag_off
= htons(IP_MF
);
540 /* Prepare header of the next frame,
541 * before previous one went down. */
543 frag
->ip_summed
= CHECKSUM_NONE
;
544 skb_reset_transport_header(frag
);
545 __skb_push(frag
, hlen
);
546 skb_reset_network_header(frag
);
547 memcpy(skb_network_header(frag
), iph
, hlen
);
549 iph
->tot_len
= htons(frag
->len
);
550 ip_copy_metadata(frag
, skb
);
552 ip_options_fragment(frag
);
553 offset
+= skb
->len
- hlen
;
554 iph
->frag_off
= htons(offset
>>3);
555 if (frag
->next
!= NULL
)
556 iph
->frag_off
|= htons(IP_MF
);
557 /* Ready, complete checksum */
564 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
574 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
583 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
587 skb_walk_frags(skb
, frag2
) {
591 frag2
->destructor
= NULL
;
592 skb
->truesize
+= frag2
->truesize
;
597 left
= skb
->len
- hlen
; /* Space per frame */
598 ptr
= hlen
; /* Where to start from */
600 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
601 * we need to make room for the encapsulating header
603 ll_rs
= LL_RESERVED_SPACE_EXTRA(rt
->dst
.dev
, nf_bridge_pad(skb
));
606 * Fragment the datagram.
609 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
610 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
613 * Keep copying data until we run out.
618 /* IF: it doesn't fit, use 'mtu' - the data space left */
621 /* IF: we are not sending upto and including the packet end
622 then align the next start on an eight byte boundary */
630 if ((skb2
= alloc_skb(len
+hlen
+ll_rs
, GFP_ATOMIC
)) == NULL
) {
631 NETDEBUG(KERN_INFO
"IP: frag: no memory for new fragment!\n");
637 * Set up data on packet
640 ip_copy_metadata(skb2
, skb
);
641 skb_reserve(skb2
, ll_rs
);
642 skb_put(skb2
, len
+ hlen
);
643 skb_reset_network_header(skb2
);
644 skb2
->transport_header
= skb2
->network_header
+ hlen
;
647 * Charge the memory for the fragment to any owner
652 skb_set_owner_w(skb2
, skb
->sk
);
655 * Copy the packet header into the new buffer.
658 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
661 * Copy a block of the IP datagram.
663 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
668 * Fill in the new header fields.
671 iph
->frag_off
= htons((offset
>> 3));
673 /* ANK: dirty, but effective trick. Upgrade options only if
674 * the segment to be fragmented was THE FIRST (otherwise,
675 * options are already fixed) and make it ONCE
676 * on the initial skb, so that all the following fragments
677 * will inherit fixed options.
680 ip_options_fragment(skb
);
683 * Added AC : If we are fragmenting a fragment that's not the
684 * last fragment then keep MF on each bit
686 if (left
> 0 || not_last_frag
)
687 iph
->frag_off
|= htons(IP_MF
);
692 * Put this fragment into the sending queue.
694 iph
->tot_len
= htons(len
+ hlen
);
702 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
705 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
710 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
713 EXPORT_SYMBOL(ip_fragment
);
716 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
718 struct iovec
*iov
= from
;
720 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
721 if (memcpy_fromiovecend(to
, iov
, offset
, len
) < 0)
725 if (csum_partial_copy_fromiovecend(to
, iov
, offset
, len
, &csum
) < 0)
727 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
731 EXPORT_SYMBOL(ip_generic_getfrag
);
734 csum_page(struct page
*page
, int offset
, int copy
)
739 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
744 static inline int ip_ufo_append_data(struct sock
*sk
,
745 int getfrag(void *from
, char *to
, int offset
, int len
,
746 int odd
, struct sk_buff
*skb
),
747 void *from
, int length
, int hh_len
, int fragheaderlen
,
748 int transhdrlen
, int mtu
, unsigned int flags
)
753 /* There is support for UDP fragmentation offload by network
754 * device, so create one single skb packet containing complete
757 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
758 skb
= sock_alloc_send_skb(sk
,
759 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
760 (flags
& MSG_DONTWAIT
), &err
);
765 /* reserve space for Hardware header */
766 skb_reserve(skb
, hh_len
);
768 /* create space for UDP/IP header */
769 skb_put(skb
, fragheaderlen
+ transhdrlen
);
771 /* initialize network header pointer */
772 skb_reset_network_header(skb
);
774 /* initialize protocol header pointer */
775 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
777 skb
->ip_summed
= CHECKSUM_PARTIAL
;
779 sk
->sk_sndmsg_off
= 0;
781 /* specify the length of each IP datagram fragment */
782 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
783 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
784 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
787 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
788 (length
- transhdrlen
));
792 * ip_append_data() and ip_append_page() can make one large IP datagram
793 * from many pieces of data. Each pieces will be holded on the socket
794 * until ip_push_pending_frames() is called. Each piece can be a page
797 * Not only UDP, other transport protocols - e.g. raw sockets - can use
798 * this interface potentially.
800 * LATER: length must be adjusted by pad at tail, when it is required.
802 int ip_append_data(struct sock
*sk
,
803 int getfrag(void *from
, char *to
, int offset
, int len
,
804 int odd
, struct sk_buff
*skb
),
805 void *from
, int length
, int transhdrlen
,
806 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
809 struct inet_sock
*inet
= inet_sk(sk
);
812 struct ip_options
*opt
= NULL
;
819 unsigned int maxfraglen
, fragheaderlen
;
820 int csummode
= CHECKSUM_NONE
;
826 if (skb_queue_empty(&sk
->sk_write_queue
)) {
832 if (inet
->cork
.opt
== NULL
) {
833 inet
->cork
.opt
= kmalloc(sizeof(struct ip_options
) + 40, sk
->sk_allocation
);
834 if (unlikely(inet
->cork
.opt
== NULL
))
837 memcpy(inet
->cork
.opt
, opt
, sizeof(struct ip_options
)+opt
->optlen
);
838 inet
->cork
.flags
|= IPCORK_OPT
;
839 inet
->cork
.addr
= ipc
->addr
;
845 * We steal reference to this route, caller should not release it
848 inet
->cork
.fragsize
= mtu
= inet
->pmtudisc
== IP_PMTUDISC_PROBE
?
850 dst_mtu(rt
->dst
.path
);
851 inet
->cork
.dst
= &rt
->dst
;
852 inet
->cork
.length
= 0;
853 sk
->sk_sndmsg_page
= NULL
;
854 sk
->sk_sndmsg_off
= 0;
855 if ((exthdrlen
= rt
->dst
.header_len
) != 0) {
857 transhdrlen
+= exthdrlen
;
860 rt
= (struct rtable
*)inet
->cork
.dst
;
861 if (inet
->cork
.flags
& IPCORK_OPT
)
862 opt
= inet
->cork
.opt
;
866 mtu
= inet
->cork
.fragsize
;
868 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
870 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
871 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
873 if (inet
->cork
.length
+ length
> 0xFFFF - fragheaderlen
) {
874 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->inet_dport
,
880 * transhdrlen > 0 means that this is the first fragment and we wish
881 * it won't be fragmented in the future.
884 length
+ fragheaderlen
<= mtu
&&
885 rt
->dst
.dev
->features
& NETIF_F_V4_CSUM
&&
887 csummode
= CHECKSUM_PARTIAL
;
889 skb
= skb_peek_tail(&sk
->sk_write_queue
);
891 inet
->cork
.length
+= length
;
892 if (((length
> mtu
) || (skb
&& skb_is_gso(skb
))) &&
893 (sk
->sk_protocol
== IPPROTO_UDP
) &&
894 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
895 err
= ip_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
896 fragheaderlen
, transhdrlen
, mtu
,
903 /* So, what's going on in the loop below?
905 * We use calculated fragment length to generate chained skb,
906 * each of segments is IP fragment ready for sending to network after
907 * adding appropriate IP header.
914 /* Check if the remaining data fits into current packet. */
915 copy
= mtu
- skb
->len
;
917 copy
= maxfraglen
- skb
->len
;
920 unsigned int datalen
;
921 unsigned int fraglen
;
922 unsigned int fraggap
;
923 unsigned int alloclen
;
924 struct sk_buff
*skb_prev
;
928 fraggap
= skb_prev
->len
- maxfraglen
;
933 * If remaining data exceeds the mtu,
934 * we know we need more fragment(s).
936 datalen
= length
+ fraggap
;
937 if (datalen
> mtu
- fragheaderlen
)
938 datalen
= maxfraglen
- fragheaderlen
;
939 fraglen
= datalen
+ fragheaderlen
;
941 if ((flags
& MSG_MORE
) &&
942 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
945 alloclen
= datalen
+ fragheaderlen
;
947 /* The last fragment gets additional space at tail.
948 * Note, with MSG_MORE we overallocate on fragments,
949 * because we have no idea what fragment will be
952 if (datalen
== length
+ fraggap
)
953 alloclen
+= rt
->dst
.trailer_len
;
956 skb
= sock_alloc_send_skb(sk
,
957 alloclen
+ hh_len
+ 15,
958 (flags
& MSG_DONTWAIT
), &err
);
961 if (atomic_read(&sk
->sk_wmem_alloc
) <=
963 skb
= sock_wmalloc(sk
,
964 alloclen
+ hh_len
+ 15, 1,
966 if (unlikely(skb
== NULL
))
969 /* only the initial fragment is
977 * Fill in the control structures
979 skb
->ip_summed
= csummode
;
981 skb_reserve(skb
, hh_len
);
982 *skb_tx(skb
) = ipc
->shtx
;
985 * Find where to start putting bytes.
987 data
= skb_put(skb
, fraglen
);
988 skb_set_network_header(skb
, exthdrlen
);
989 skb
->transport_header
= (skb
->network_header
+
991 data
+= fragheaderlen
;
994 skb
->csum
= skb_copy_and_csum_bits(
995 skb_prev
, maxfraglen
,
996 data
+ transhdrlen
, fraggap
, 0);
997 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1000 pskb_trim_unique(skb_prev
, maxfraglen
);
1003 copy
= datalen
- transhdrlen
- fraggap
;
1004 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1011 length
-= datalen
- fraggap
;
1014 csummode
= CHECKSUM_NONE
;
1017 * Put the packet on the pending queue.
1019 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1026 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
)) {
1030 if (getfrag(from
, skb_put(skb
, copy
),
1031 offset
, copy
, off
, skb
) < 0) {
1032 __skb_trim(skb
, off
);
1037 int i
= skb_shinfo(skb
)->nr_frags
;
1038 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
1039 struct page
*page
= sk
->sk_sndmsg_page
;
1040 int off
= sk
->sk_sndmsg_off
;
1043 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
1046 if (page
!= frag
->page
) {
1047 if (i
== MAX_SKB_FRAGS
) {
1052 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
1053 frag
= &skb_shinfo(skb
)->frags
[i
];
1055 } else if (i
< MAX_SKB_FRAGS
) {
1056 if (copy
> PAGE_SIZE
)
1058 page
= alloc_pages(sk
->sk_allocation
, 0);
1063 sk
->sk_sndmsg_page
= page
;
1064 sk
->sk_sndmsg_off
= 0;
1066 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1067 frag
= &skb_shinfo(skb
)->frags
[i
];
1072 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1076 sk
->sk_sndmsg_off
+= copy
;
1079 skb
->data_len
+= copy
;
1080 skb
->truesize
+= copy
;
1081 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1090 inet
->cork
.length
-= length
;
1091 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1095 ssize_t
ip_append_page(struct sock
*sk
, struct page
*page
,
1096 int offset
, size_t size
, int flags
)
1098 struct inet_sock
*inet
= inet_sk(sk
);
1099 struct sk_buff
*skb
;
1101 struct ip_options
*opt
= NULL
;
1106 unsigned int maxfraglen
, fragheaderlen
, fraggap
;
1111 if (flags
&MSG_PROBE
)
1114 if (skb_queue_empty(&sk
->sk_write_queue
))
1117 rt
= (struct rtable
*)inet
->cork
.dst
;
1118 if (inet
->cork
.flags
& IPCORK_OPT
)
1119 opt
= inet
->cork
.opt
;
1121 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1124 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1125 mtu
= inet
->cork
.fragsize
;
1127 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1128 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1130 if (inet
->cork
.length
+ size
> 0xFFFF - fragheaderlen
) {
1131 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->inet_dport
, mtu
);
1135 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1138 inet
->cork
.length
+= size
;
1139 if ((size
+ skb
->len
> mtu
) &&
1140 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1141 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
1142 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1143 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1150 if (skb_is_gso(skb
))
1154 /* Check if the remaining data fits into current packet. */
1155 len
= mtu
- skb
->len
;
1157 len
= maxfraglen
- skb
->len
;
1160 struct sk_buff
*skb_prev
;
1164 fraggap
= skb_prev
->len
- maxfraglen
;
1166 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1167 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1168 if (unlikely(!skb
)) {
1174 * Fill in the control structures
1176 skb
->ip_summed
= CHECKSUM_NONE
;
1178 skb_reserve(skb
, hh_len
);
1181 * Find where to start putting bytes.
1183 skb_put(skb
, fragheaderlen
+ fraggap
);
1184 skb_reset_network_header(skb
);
1185 skb
->transport_header
= (skb
->network_header
+
1188 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1190 skb_transport_header(skb
),
1192 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1194 pskb_trim_unique(skb_prev
, maxfraglen
);
1198 * Put the packet on the pending queue.
1200 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1204 i
= skb_shinfo(skb
)->nr_frags
;
1207 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1208 skb_shinfo(skb
)->frags
[i
-1].size
+= len
;
1209 } else if (i
< MAX_SKB_FRAGS
) {
1211 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1217 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1219 csum
= csum_page(page
, offset
, len
);
1220 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1224 skb
->data_len
+= len
;
1225 skb
->truesize
+= len
;
1226 atomic_add(len
, &sk
->sk_wmem_alloc
);
1233 inet
->cork
.length
-= size
;
1234 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1238 static void ip_cork_release(struct inet_sock
*inet
)
1240 inet
->cork
.flags
&= ~IPCORK_OPT
;
1241 kfree(inet
->cork
.opt
);
1242 inet
->cork
.opt
= NULL
;
1243 dst_release(inet
->cork
.dst
);
1244 inet
->cork
.dst
= NULL
;
1248 * Combined all pending IP fragments on the socket as one IP datagram
1249 * and push them out.
1251 int ip_push_pending_frames(struct sock
*sk
)
1253 struct sk_buff
*skb
, *tmp_skb
;
1254 struct sk_buff
**tail_skb
;
1255 struct inet_sock
*inet
= inet_sk(sk
);
1256 struct net
*net
= sock_net(sk
);
1257 struct ip_options
*opt
= NULL
;
1258 struct rtable
*rt
= (struct rtable
*)inet
->cork
.dst
;
1264 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1266 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1268 /* move skb->data to ip header from ext header */
1269 if (skb
->data
< skb_network_header(skb
))
1270 __skb_pull(skb
, skb_network_offset(skb
));
1271 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1272 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1273 *tail_skb
= tmp_skb
;
1274 tail_skb
= &(tmp_skb
->next
);
1275 skb
->len
+= tmp_skb
->len
;
1276 skb
->data_len
+= tmp_skb
->len
;
1277 skb
->truesize
+= tmp_skb
->truesize
;
1278 tmp_skb
->destructor
= NULL
;
1282 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1283 * to fragment the frame generated here. No matter, what transforms
1284 * how transforms change size of the packet, it will come out.
1286 if (inet
->pmtudisc
< IP_PMTUDISC_DO
)
1289 /* DF bit is set when we want to see DF on outgoing frames.
1290 * If local_df is set too, we still allow to fragment this frame
1292 if (inet
->pmtudisc
>= IP_PMTUDISC_DO
||
1293 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1294 ip_dont_fragment(sk
, &rt
->dst
)))
1297 if (inet
->cork
.flags
& IPCORK_OPT
)
1298 opt
= inet
->cork
.opt
;
1300 if (rt
->rt_type
== RTN_MULTICAST
)
1303 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1305 iph
= (struct iphdr
*)skb
->data
;
1309 iph
->ihl
+= opt
->optlen
>>2;
1310 ip_options_build(skb
, opt
, inet
->cork
.addr
, rt
, 0);
1312 iph
->tos
= inet
->tos
;
1314 ip_select_ident(iph
, &rt
->dst
, sk
);
1316 iph
->protocol
= sk
->sk_protocol
;
1317 iph
->saddr
= rt
->rt_src
;
1318 iph
->daddr
= rt
->rt_dst
;
1320 skb
->priority
= sk
->sk_priority
;
1321 skb
->mark
= sk
->sk_mark
;
1323 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1326 inet
->cork
.dst
= NULL
;
1327 skb_dst_set(skb
, &rt
->dst
);
1329 if (iph
->protocol
== IPPROTO_ICMP
)
1330 icmp_out_count(net
, ((struct icmphdr
*)
1331 skb_transport_header(skb
))->type
);
1333 /* Netfilter gets whole the not fragmented skb. */
1334 err
= ip_local_out(skb
);
1337 err
= net_xmit_errno(err
);
1343 ip_cork_release(inet
);
1347 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1352 * Throw away all pending data on the socket.
1354 void ip_flush_pending_frames(struct sock
*sk
)
1356 struct sk_buff
*skb
;
1358 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
)
1361 ip_cork_release(inet_sk(sk
));
1366 * Fetch data from kernel space and fill in checksum if needed.
1368 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1369 int len
, int odd
, struct sk_buff
*skb
)
1373 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1374 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1379 * Generic function to send a packet as reply to another packet.
1380 * Used to send TCP resets so far. ICMP should use this function too.
1382 * Should run single threaded per socket because it uses the sock
1383 * structure to pass arguments.
1385 void ip_send_reply(struct sock
*sk
, struct sk_buff
*skb
, struct ip_reply_arg
*arg
,
1388 struct inet_sock
*inet
= inet_sk(sk
);
1390 struct ip_options opt
;
1393 struct ipcm_cookie ipc
;
1395 struct rtable
*rt
= skb_rtable(skb
);
1397 if (ip_options_echo(&replyopts
.opt
, skb
))
1400 daddr
= ipc
.addr
= rt
->rt_src
;
1404 if (replyopts
.opt
.optlen
) {
1405 ipc
.opt
= &replyopts
.opt
;
1408 daddr
= replyopts
.opt
.faddr
;
1412 struct flowi fl
= { .oif
= arg
->bound_dev_if
,
1415 .saddr
= rt
->rt_spec_dst
,
1416 .tos
= RT_TOS(ip_hdr(skb
)->tos
) } },
1417 /* Not quite clean, but right. */
1419 { .sport
= tcp_hdr(skb
)->dest
,
1420 .dport
= tcp_hdr(skb
)->source
} },
1421 .proto
= sk
->sk_protocol
,
1422 .flags
= ip_reply_arg_flowi_flags(arg
) };
1423 security_skb_classify_flow(skb
, &fl
);
1424 if (ip_route_output_key(sock_net(sk
), &rt
, &fl
))
1428 /* And let IP do all the hard work.
1430 This chunk is not reenterable, hence spinlock.
1431 Note that it uses the fact, that this function is called
1432 with locally disabled BH and that sk cannot be already spinlocked.
1435 inet
->tos
= ip_hdr(skb
)->tos
;
1436 sk
->sk_priority
= skb
->priority
;
1437 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1438 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1439 ip_append_data(sk
, ip_reply_glue_bits
, arg
->iov
->iov_base
, len
, 0,
1440 &ipc
, &rt
, MSG_DONTWAIT
);
1441 if ((skb
= skb_peek(&sk
->sk_write_queue
)) != NULL
) {
1442 if (arg
->csumoffset
>= 0)
1443 *((__sum16
*)skb_transport_header(skb
) +
1444 arg
->csumoffset
) = csum_fold(csum_add(skb
->csum
,
1446 skb
->ip_summed
= CHECKSUM_NONE
;
1447 ip_push_pending_frames(sk
);
1455 void __init
ip_init(void)
1460 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1461 igmp_mc_proc_init();