2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The Internet Protocol (IP) output module.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <Alan.Cox@linux.org>
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16 * Hirokazu Takahashi, <taka@valinux.co.jp>
18 * See ip_input.c for original log
21 * Alan Cox : Missing nonblock feature in ip_build_xmit.
22 * Mike Kilburn : htons() missing in ip_build_xmit.
23 * Bradford Johnson: Fix faulty handling of some frames when
25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
26 * (in case if packet not accepted by
27 * output firewall rules)
28 * Mike McLagan : Routing by source
29 * Alexey Kuznetsov: use new route cache
30 * Andi Kleen: Fix broken PMTU recovery and remove
31 * some redundant tests.
32 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
33 * Andi Kleen : Replace ip_reply with ip_send_reply.
34 * Andi Kleen : Split fast and slow ip_build_xmit path
35 * for decreased register pressure on x86
36 * and more readibility.
37 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
38 * silently drop skb instead of failing with -EPERM.
39 * Detlev Wengorz : Copy protocol for fragments.
40 * Hirokazu Takahashi: HW checksumming for outgoing UDP
42 * Hirokazu Takahashi: sendfile() on UDP works now.
45 #include <asm/uaccess.h>
46 #include <asm/system.h>
47 #include <linux/module.h>
48 #include <linux/types.h>
49 #include <linux/kernel.h>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
54 #include <linux/slab.h>
56 #include <linux/socket.h>
57 #include <linux/sockios.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/etherdevice.h>
62 #include <linux/proc_fs.h>
63 #include <linux/stat.h>
64 #include <linux/init.h>
68 #include <net/protocol.h>
69 #include <net/route.h>
71 #include <linux/skbuff.h>
75 #include <net/checksum.h>
76 #include <net/inetpeer.h>
77 #include <linux/igmp.h>
78 #include <linux/netfilter_ipv4.h>
79 #include <linux/netfilter_bridge.h>
80 #include <linux/mroute.h>
81 #include <linux/netlink.h>
82 #include <linux/tcp.h>
84 int sysctl_ip_default_ttl __read_mostly
= IPDEFTTL
;
86 /* Generate a checksum for an outgoing IP datagram. */
87 __inline__
void ip_send_check(struct iphdr
*iph
)
90 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
93 int __ip_local_out(struct sk_buff
*skb
)
95 struct iphdr
*iph
= ip_hdr(skb
);
97 iph
->tot_len
= htons(skb
->len
);
99 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, skb
, NULL
,
100 skb_dst(skb
)->dev
, dst_output
);
103 int ip_local_out(struct sk_buff
*skb
)
107 err
= __ip_local_out(skb
);
108 if (likely(err
== 1))
109 err
= dst_output(skb
);
113 EXPORT_SYMBOL_GPL(ip_local_out
);
115 /* dev_loopback_xmit for use with netfilter. */
116 static int ip_dev_loopback_xmit(struct sk_buff
*newskb
)
118 skb_reset_mac_header(newskb
);
119 __skb_pull(newskb
, skb_network_offset(newskb
));
120 newskb
->pkt_type
= PACKET_LOOPBACK
;
121 newskb
->ip_summed
= CHECKSUM_UNNECESSARY
;
122 WARN_ON(!skb_dst(newskb
));
127 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
129 int ttl
= inet
->uc_ttl
;
132 ttl
= dst_metric(dst
, RTAX_HOPLIMIT
);
137 * Add an ip header to a skbuff and send it out.
140 int ip_build_and_send_pkt(struct sk_buff
*skb
, struct sock
*sk
,
141 __be32 saddr
, __be32 daddr
, struct ip_options
*opt
)
143 struct inet_sock
*inet
= inet_sk(sk
);
144 struct rtable
*rt
= skb_rtable(skb
);
147 /* Build the IP header. */
148 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
149 skb_reset_network_header(skb
);
153 iph
->tos
= inet
->tos
;
154 if (ip_dont_fragment(sk
, &rt
->u
.dst
))
155 iph
->frag_off
= htons(IP_DF
);
158 iph
->ttl
= ip_select_ttl(inet
, &rt
->u
.dst
);
159 iph
->daddr
= rt
->rt_dst
;
160 iph
->saddr
= rt
->rt_src
;
161 iph
->protocol
= sk
->sk_protocol
;
162 ip_select_ident(iph
, &rt
->u
.dst
, sk
);
164 if (opt
&& opt
->optlen
) {
165 iph
->ihl
+= opt
->optlen
>>2;
166 ip_options_build(skb
, opt
, daddr
, rt
, 0);
169 skb
->priority
= sk
->sk_priority
;
170 skb
->mark
= sk
->sk_mark
;
173 return ip_local_out(skb
);
176 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
178 static inline int ip_finish_output2(struct sk_buff
*skb
)
180 struct dst_entry
*dst
= skb_dst(skb
);
181 struct rtable
*rt
= (struct rtable
*)dst
;
182 struct net_device
*dev
= dst
->dev
;
183 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
185 if (rt
->rt_type
== RTN_MULTICAST
) {
186 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTMCAST
, skb
->len
);
187 } else if (rt
->rt_type
== RTN_BROADCAST
)
188 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTBCAST
, skb
->len
);
190 /* Be paranoid, rather than too clever. */
191 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
192 struct sk_buff
*skb2
;
194 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
200 skb_set_owner_w(skb2
, skb
->sk
);
206 return neigh_hh_output(dst
->hh
, skb
);
207 else if (dst
->neighbour
)
208 return dst
->neighbour
->output(skb
);
211 printk(KERN_DEBUG
"ip_finish_output2: No header cache and no neighbour!\n");
216 static inline int ip_skb_dst_mtu(struct sk_buff
*skb
)
218 struct inet_sock
*inet
= skb
->sk
? inet_sk(skb
->sk
) : NULL
;
220 return (inet
&& inet
->pmtudisc
== IP_PMTUDISC_PROBE
) ?
221 skb_dst(skb
)->dev
->mtu
: dst_mtu(skb_dst(skb
));
224 static int ip_finish_output(struct sk_buff
*skb
)
226 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
227 /* Policy lookup after SNAT yielded a new policy */
228 if (skb_dst(skb
)->xfrm
!= NULL
) {
229 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
230 return dst_output(skb
);
233 if (skb
->len
> ip_skb_dst_mtu(skb
) && !skb_is_gso(skb
))
234 return ip_fragment(skb
, ip_finish_output2
);
236 return ip_finish_output2(skb
);
239 int ip_mc_output(struct sk_buff
*skb
)
241 struct sock
*sk
= skb
->sk
;
242 struct rtable
*rt
= skb_rtable(skb
);
243 struct net_device
*dev
= rt
->u
.dst
.dev
;
246 * If the indicated interface is up and running, send the packet.
248 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
251 skb
->protocol
= htons(ETH_P_IP
);
254 * Multicasts are looped back for other local users
257 if (rt
->rt_flags
&RTCF_MULTICAST
) {
259 #ifdef CONFIG_IP_MROUTE
260 /* Small optimization: do not loopback not local frames,
261 which returned after forwarding; they will be dropped
262 by ip_mr_input in any case.
263 Note, that local frames are looped back to be delivered
266 This check is duplicated in ip_mr_input at the moment.
269 ((rt
->rt_flags
& RTCF_LOCAL
) ||
270 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
273 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
275 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
276 newskb
, NULL
, newskb
->dev
,
277 ip_dev_loopback_xmit
);
280 /* Multicasts with ttl 0 must not go beyond the host */
282 if (ip_hdr(skb
)->ttl
== 0) {
288 if (rt
->rt_flags
&RTCF_BROADCAST
) {
289 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
291 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, newskb
,
292 NULL
, newskb
->dev
, ip_dev_loopback_xmit
);
295 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
,
296 skb
->dev
, ip_finish_output
,
297 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
300 int ip_output(struct sk_buff
*skb
)
302 struct net_device
*dev
= skb_dst(skb
)->dev
;
304 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
307 skb
->protocol
= htons(ETH_P_IP
);
309 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
, dev
,
311 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
314 int ip_queue_xmit(struct sk_buff
*skb
)
316 struct sock
*sk
= skb
->sk
;
317 struct inet_sock
*inet
= inet_sk(sk
);
318 struct ip_options
*opt
= inet
->opt
;
322 /* Skip all of this if the packet is already routed,
323 * f.e. by something like SCTP.
325 rt
= skb_rtable(skb
);
329 /* Make sure we can route this packet. */
330 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
334 /* Use correct destination address if we have options. */
335 daddr
= inet
->inet_daddr
;
340 struct flowi fl
= { .oif
= sk
->sk_bound_dev_if
,
344 .saddr
= inet
->inet_saddr
,
345 .tos
= RT_CONN_FLAGS(sk
) } },
346 .proto
= sk
->sk_protocol
,
347 .flags
= inet_sk_flowi_flags(sk
),
349 { .sport
= inet
->inet_sport
,
350 .dport
= inet
->inet_dport
} } };
352 /* If this fails, retransmit mechanism of transport layer will
353 * keep trying until route appears or the connection times
356 security_sk_classify_flow(sk
, &fl
);
357 if (ip_route_output_flow(sock_net(sk
), &rt
, &fl
, sk
, 0))
360 sk_setup_caps(sk
, &rt
->u
.dst
);
362 skb_dst_set(skb
, dst_clone(&rt
->u
.dst
));
365 if (opt
&& opt
->is_strictroute
&& rt
->rt_dst
!= rt
->rt_gateway
)
368 /* OK, we know where to send it, allocate and build IP header. */
369 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
370 skb_reset_network_header(skb
);
372 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
373 if (ip_dont_fragment(sk
, &rt
->u
.dst
) && !skb
->local_df
)
374 iph
->frag_off
= htons(IP_DF
);
377 iph
->ttl
= ip_select_ttl(inet
, &rt
->u
.dst
);
378 iph
->protocol
= sk
->sk_protocol
;
379 iph
->saddr
= rt
->rt_src
;
380 iph
->daddr
= rt
->rt_dst
;
381 /* Transport layer set skb->h.foo itself. */
383 if (opt
&& opt
->optlen
) {
384 iph
->ihl
+= opt
->optlen
>> 2;
385 ip_options_build(skb
, opt
, inet
->inet_daddr
, rt
, 0);
388 ip_select_ident_more(iph
, &rt
->u
.dst
, sk
,
389 (skb_shinfo(skb
)->gso_segs
?: 1) - 1);
391 skb
->priority
= sk
->sk_priority
;
392 skb
->mark
= sk
->sk_mark
;
394 return ip_local_out(skb
);
397 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
399 return -EHOSTUNREACH
;
403 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
405 to
->pkt_type
= from
->pkt_type
;
406 to
->priority
= from
->priority
;
407 to
->protocol
= from
->protocol
;
409 skb_dst_set(to
, dst_clone(skb_dst(from
)));
411 to
->mark
= from
->mark
;
413 /* Copy the flags to each fragment. */
414 IPCB(to
)->flags
= IPCB(from
)->flags
;
416 #ifdef CONFIG_NET_SCHED
417 to
->tc_index
= from
->tc_index
;
420 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
421 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
422 to
->nf_trace
= from
->nf_trace
;
424 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
425 to
->ipvs_property
= from
->ipvs_property
;
427 skb_copy_secmark(to
, from
);
431 * This IP datagram is too large to be sent in one piece. Break it up into
432 * smaller pieces (each of size equal to IP header plus
433 * a block of the data of the original IP data part) that will yet fit in a
434 * single device frame, and queue such a frame for sending.
437 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
442 struct net_device
*dev
;
443 struct sk_buff
*skb2
;
444 unsigned int mtu
, hlen
, left
, len
, ll_rs
, pad
;
446 __be16 not_last_frag
;
447 struct rtable
*rt
= skb_rtable(skb
);
453 * Point into the IP datagram header.
458 if (unlikely((iph
->frag_off
& htons(IP_DF
)) && !skb
->local_df
)) {
459 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
460 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
461 htonl(ip_skb_dst_mtu(skb
)));
467 * Setup starting values.
471 mtu
= dst_mtu(&rt
->u
.dst
) - hlen
; /* Size of data space */
472 #ifdef CONFIG_BRIDGE_NETFILTER
474 mtu
-= nf_bridge_mtu_reduction(skb
);
476 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
478 /* When frag_list is given, use it. First, check its validity:
479 * some transformers could create wrong frag_list or break existing
480 * one, it is not prohibited. In this case fall back to copying.
482 * LATER: this step can be merged to real generation of fragments,
483 * we can switch to copy when see the first bad fragment.
485 if (skb_has_frags(skb
)) {
486 struct sk_buff
*frag
;
487 int first_len
= skb_pagelen(skb
);
490 if (first_len
- hlen
> mtu
||
491 ((first_len
- hlen
) & 7) ||
492 (iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) ||
496 skb_walk_frags(skb
, frag
) {
497 /* Correct geometry. */
498 if (frag
->len
> mtu
||
499 ((frag
->len
& 7) && frag
->next
) ||
500 skb_headroom(frag
) < hlen
)
503 /* Partially cloned skb? */
504 if (skb_shared(frag
))
510 frag
->destructor
= sock_wfree
;
512 truesizes
+= frag
->truesize
;
515 /* Everything is OK. Generate! */
519 frag
= skb_shinfo(skb
)->frag_list
;
520 skb_frag_list_init(skb
);
521 skb
->data_len
= first_len
- skb_headlen(skb
);
522 skb
->truesize
-= truesizes
;
523 skb
->len
= first_len
;
524 iph
->tot_len
= htons(first_len
);
525 iph
->frag_off
= htons(IP_MF
);
529 /* Prepare header of the next frame,
530 * before previous one went down. */
532 frag
->ip_summed
= CHECKSUM_NONE
;
533 skb_reset_transport_header(frag
);
534 __skb_push(frag
, hlen
);
535 skb_reset_network_header(frag
);
536 memcpy(skb_network_header(frag
), iph
, hlen
);
538 iph
->tot_len
= htons(frag
->len
);
539 ip_copy_metadata(frag
, skb
);
541 ip_options_fragment(frag
);
542 offset
+= skb
->len
- hlen
;
543 iph
->frag_off
= htons(offset
>>3);
544 if (frag
->next
!= NULL
)
545 iph
->frag_off
|= htons(IP_MF
);
546 /* Ready, complete checksum */
553 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
563 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
572 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
577 left
= skb
->len
- hlen
; /* Space per frame */
578 ptr
= raw
+ hlen
; /* Where to start from */
580 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
581 * we need to make room for the encapsulating header
583 pad
= nf_bridge_pad(skb
);
584 ll_rs
= LL_RESERVED_SPACE_EXTRA(rt
->u
.dst
.dev
, pad
);
588 * Fragment the datagram.
591 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
592 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
595 * Keep copying data until we run out.
600 /* IF: it doesn't fit, use 'mtu' - the data space left */
603 /* IF: we are not sending upto and including the packet end
604 then align the next start on an eight byte boundary */
612 if ((skb2
= alloc_skb(len
+hlen
+ll_rs
, GFP_ATOMIC
)) == NULL
) {
613 NETDEBUG(KERN_INFO
"IP: frag: no memory for new fragment!\n");
619 * Set up data on packet
622 ip_copy_metadata(skb2
, skb
);
623 skb_reserve(skb2
, ll_rs
);
624 skb_put(skb2
, len
+ hlen
);
625 skb_reset_network_header(skb2
);
626 skb2
->transport_header
= skb2
->network_header
+ hlen
;
629 * Charge the memory for the fragment to any owner
634 skb_set_owner_w(skb2
, skb
->sk
);
637 * Copy the packet header into the new buffer.
640 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
643 * Copy a block of the IP datagram.
645 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
650 * Fill in the new header fields.
653 iph
->frag_off
= htons((offset
>> 3));
655 /* ANK: dirty, but effective trick. Upgrade options only if
656 * the segment to be fragmented was THE FIRST (otherwise,
657 * options are already fixed) and make it ONCE
658 * on the initial skb, so that all the following fragments
659 * will inherit fixed options.
662 ip_options_fragment(skb
);
665 * Added AC : If we are fragmenting a fragment that's not the
666 * last fragment then keep MF on each bit
668 if (left
> 0 || not_last_frag
)
669 iph
->frag_off
|= htons(IP_MF
);
674 * Put this fragment into the sending queue.
676 iph
->tot_len
= htons(len
+ hlen
);
684 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
687 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
692 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
696 EXPORT_SYMBOL(ip_fragment
);
699 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
701 struct iovec
*iov
= from
;
703 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
704 if (memcpy_fromiovecend(to
, iov
, offset
, len
) < 0)
708 if (csum_partial_copy_fromiovecend(to
, iov
, offset
, len
, &csum
) < 0)
710 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
716 csum_page(struct page
*page
, int offset
, int copy
)
721 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
726 static inline int ip_ufo_append_data(struct sock
*sk
,
727 int getfrag(void *from
, char *to
, int offset
, int len
,
728 int odd
, struct sk_buff
*skb
),
729 void *from
, int length
, int hh_len
, int fragheaderlen
,
730 int transhdrlen
, int mtu
, unsigned int flags
)
735 /* There is support for UDP fragmentation offload by network
736 * device, so create one single skb packet containing complete
739 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
740 skb
= sock_alloc_send_skb(sk
,
741 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
742 (flags
& MSG_DONTWAIT
), &err
);
747 /* reserve space for Hardware header */
748 skb_reserve(skb
, hh_len
);
750 /* create space for UDP/IP header */
751 skb_put(skb
, fragheaderlen
+ transhdrlen
);
753 /* initialize network header pointer */
754 skb_reset_network_header(skb
);
756 /* initialize protocol header pointer */
757 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
759 skb
->ip_summed
= CHECKSUM_PARTIAL
;
761 sk
->sk_sndmsg_off
= 0;
763 /* specify the length of each IP datagram fragment */
764 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
765 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
766 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
769 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
770 (length
- transhdrlen
));
774 * ip_append_data() and ip_append_page() can make one large IP datagram
775 * from many pieces of data. Each pieces will be holded on the socket
776 * until ip_push_pending_frames() is called. Each piece can be a page
779 * Not only UDP, other transport protocols - e.g. raw sockets - can use
780 * this interface potentially.
782 * LATER: length must be adjusted by pad at tail, when it is required.
784 int ip_append_data(struct sock
*sk
,
785 int getfrag(void *from
, char *to
, int offset
, int len
,
786 int odd
, struct sk_buff
*skb
),
787 void *from
, int length
, int transhdrlen
,
788 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
791 struct inet_sock
*inet
= inet_sk(sk
);
794 struct ip_options
*opt
= NULL
;
801 unsigned int maxfraglen
, fragheaderlen
;
802 int csummode
= CHECKSUM_NONE
;
808 if (skb_queue_empty(&sk
->sk_write_queue
)) {
814 if (inet
->cork
.opt
== NULL
) {
815 inet
->cork
.opt
= kmalloc(sizeof(struct ip_options
) + 40, sk
->sk_allocation
);
816 if (unlikely(inet
->cork
.opt
== NULL
))
819 memcpy(inet
->cork
.opt
, opt
, sizeof(struct ip_options
)+opt
->optlen
);
820 inet
->cork
.flags
|= IPCORK_OPT
;
821 inet
->cork
.addr
= ipc
->addr
;
827 * We steal reference to this route, caller should not release it
830 inet
->cork
.fragsize
= mtu
= inet
->pmtudisc
== IP_PMTUDISC_PROBE
?
832 dst_mtu(rt
->u
.dst
.path
);
833 inet
->cork
.dst
= &rt
->u
.dst
;
834 inet
->cork
.length
= 0;
835 sk
->sk_sndmsg_page
= NULL
;
836 sk
->sk_sndmsg_off
= 0;
837 if ((exthdrlen
= rt
->u
.dst
.header_len
) != 0) {
839 transhdrlen
+= exthdrlen
;
842 rt
= (struct rtable
*)inet
->cork
.dst
;
843 if (inet
->cork
.flags
& IPCORK_OPT
)
844 opt
= inet
->cork
.opt
;
848 mtu
= inet
->cork
.fragsize
;
850 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
852 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
853 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
855 if (inet
->cork
.length
+ length
> 0xFFFF - fragheaderlen
) {
856 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->inet_dport
,
862 * transhdrlen > 0 means that this is the first fragment and we wish
863 * it won't be fragmented in the future.
866 length
+ fragheaderlen
<= mtu
&&
867 rt
->u
.dst
.dev
->features
& NETIF_F_V4_CSUM
&&
869 csummode
= CHECKSUM_PARTIAL
;
871 inet
->cork
.length
+= length
;
872 if (((length
> mtu
) || !skb_queue_empty(&sk
->sk_write_queue
)) &&
873 (sk
->sk_protocol
== IPPROTO_UDP
) &&
874 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
875 err
= ip_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
876 fragheaderlen
, transhdrlen
, mtu
,
883 /* So, what's going on in the loop below?
885 * We use calculated fragment length to generate chained skb,
886 * each of segments is IP fragment ready for sending to network after
887 * adding appropriate IP header.
890 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
894 /* Check if the remaining data fits into current packet. */
895 copy
= mtu
- skb
->len
;
897 copy
= maxfraglen
- skb
->len
;
900 unsigned int datalen
;
901 unsigned int fraglen
;
902 unsigned int fraggap
;
903 unsigned int alloclen
;
904 struct sk_buff
*skb_prev
;
908 fraggap
= skb_prev
->len
- maxfraglen
;
913 * If remaining data exceeds the mtu,
914 * we know we need more fragment(s).
916 datalen
= length
+ fraggap
;
917 if (datalen
> mtu
- fragheaderlen
)
918 datalen
= maxfraglen
- fragheaderlen
;
919 fraglen
= datalen
+ fragheaderlen
;
921 if ((flags
& MSG_MORE
) &&
922 !(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
925 alloclen
= datalen
+ fragheaderlen
;
927 /* The last fragment gets additional space at tail.
928 * Note, with MSG_MORE we overallocate on fragments,
929 * because we have no idea what fragment will be
932 if (datalen
== length
+ fraggap
)
933 alloclen
+= rt
->u
.dst
.trailer_len
;
936 skb
= sock_alloc_send_skb(sk
,
937 alloclen
+ hh_len
+ 15,
938 (flags
& MSG_DONTWAIT
), &err
);
941 if (atomic_read(&sk
->sk_wmem_alloc
) <=
943 skb
= sock_wmalloc(sk
,
944 alloclen
+ hh_len
+ 15, 1,
946 if (unlikely(skb
== NULL
))
949 /* only the initial fragment is
957 * Fill in the control structures
959 skb
->ip_summed
= csummode
;
961 skb_reserve(skb
, hh_len
);
962 *skb_tx(skb
) = ipc
->shtx
;
965 * Find where to start putting bytes.
967 data
= skb_put(skb
, fraglen
);
968 skb_set_network_header(skb
, exthdrlen
);
969 skb
->transport_header
= (skb
->network_header
+
971 data
+= fragheaderlen
;
974 skb
->csum
= skb_copy_and_csum_bits(
975 skb_prev
, maxfraglen
,
976 data
+ transhdrlen
, fraggap
, 0);
977 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
980 pskb_trim_unique(skb_prev
, maxfraglen
);
983 copy
= datalen
- transhdrlen
- fraggap
;
984 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
991 length
-= datalen
- fraggap
;
994 csummode
= CHECKSUM_NONE
;
997 * Put the packet on the pending queue.
999 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1006 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
)) {
1010 if (getfrag(from
, skb_put(skb
, copy
),
1011 offset
, copy
, off
, skb
) < 0) {
1012 __skb_trim(skb
, off
);
1017 int i
= skb_shinfo(skb
)->nr_frags
;
1018 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
1019 struct page
*page
= sk
->sk_sndmsg_page
;
1020 int off
= sk
->sk_sndmsg_off
;
1023 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
1026 if (page
!= frag
->page
) {
1027 if (i
== MAX_SKB_FRAGS
) {
1032 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
1033 frag
= &skb_shinfo(skb
)->frags
[i
];
1035 } else if (i
< MAX_SKB_FRAGS
) {
1036 if (copy
> PAGE_SIZE
)
1038 page
= alloc_pages(sk
->sk_allocation
, 0);
1043 sk
->sk_sndmsg_page
= page
;
1044 sk
->sk_sndmsg_off
= 0;
1046 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1047 frag
= &skb_shinfo(skb
)->frags
[i
];
1052 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1056 sk
->sk_sndmsg_off
+= copy
;
1059 skb
->data_len
+= copy
;
1060 skb
->truesize
+= copy
;
1061 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1070 inet
->cork
.length
-= length
;
1071 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1075 ssize_t
ip_append_page(struct sock
*sk
, struct page
*page
,
1076 int offset
, size_t size
, int flags
)
1078 struct inet_sock
*inet
= inet_sk(sk
);
1079 struct sk_buff
*skb
;
1081 struct ip_options
*opt
= NULL
;
1086 unsigned int maxfraglen
, fragheaderlen
, fraggap
;
1091 if (flags
&MSG_PROBE
)
1094 if (skb_queue_empty(&sk
->sk_write_queue
))
1097 rt
= (struct rtable
*)inet
->cork
.dst
;
1098 if (inet
->cork
.flags
& IPCORK_OPT
)
1099 opt
= inet
->cork
.opt
;
1101 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
1104 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
1105 mtu
= inet
->cork
.fragsize
;
1107 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1108 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1110 if (inet
->cork
.length
+ size
> 0xFFFF - fragheaderlen
) {
1111 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->inet_dport
, mtu
);
1115 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1118 inet
->cork
.length
+= size
;
1119 if ((sk
->sk_protocol
== IPPROTO_UDP
) &&
1120 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
1121 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1122 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1129 if (skb_is_gso(skb
))
1133 /* Check if the remaining data fits into current packet. */
1134 len
= mtu
- skb
->len
;
1136 len
= maxfraglen
- skb
->len
;
1139 struct sk_buff
*skb_prev
;
1143 fraggap
= skb_prev
->len
- maxfraglen
;
1145 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1146 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1147 if (unlikely(!skb
)) {
1153 * Fill in the control structures
1155 skb
->ip_summed
= CHECKSUM_NONE
;
1157 skb_reserve(skb
, hh_len
);
1160 * Find where to start putting bytes.
1162 skb_put(skb
, fragheaderlen
+ fraggap
);
1163 skb_reset_network_header(skb
);
1164 skb
->transport_header
= (skb
->network_header
+
1167 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1169 skb_transport_header(skb
),
1171 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1173 pskb_trim_unique(skb_prev
, maxfraglen
);
1177 * Put the packet on the pending queue.
1179 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1183 i
= skb_shinfo(skb
)->nr_frags
;
1186 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1187 skb_shinfo(skb
)->frags
[i
-1].size
+= len
;
1188 } else if (i
< MAX_SKB_FRAGS
) {
1190 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1196 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1198 csum
= csum_page(page
, offset
, len
);
1199 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1203 skb
->data_len
+= len
;
1204 skb
->truesize
+= len
;
1205 atomic_add(len
, &sk
->sk_wmem_alloc
);
1212 inet
->cork
.length
-= size
;
1213 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1217 static void ip_cork_release(struct inet_sock
*inet
)
1219 inet
->cork
.flags
&= ~IPCORK_OPT
;
1220 kfree(inet
->cork
.opt
);
1221 inet
->cork
.opt
= NULL
;
1222 dst_release(inet
->cork
.dst
);
1223 inet
->cork
.dst
= NULL
;
1227 * Combined all pending IP fragments on the socket as one IP datagram
1228 * and push them out.
1230 int ip_push_pending_frames(struct sock
*sk
)
1232 struct sk_buff
*skb
, *tmp_skb
;
1233 struct sk_buff
**tail_skb
;
1234 struct inet_sock
*inet
= inet_sk(sk
);
1235 struct net
*net
= sock_net(sk
);
1236 struct ip_options
*opt
= NULL
;
1237 struct rtable
*rt
= (struct rtable
*)inet
->cork
.dst
;
1243 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1245 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1247 /* move skb->data to ip header from ext header */
1248 if (skb
->data
< skb_network_header(skb
))
1249 __skb_pull(skb
, skb_network_offset(skb
));
1250 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1251 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1252 *tail_skb
= tmp_skb
;
1253 tail_skb
= &(tmp_skb
->next
);
1254 skb
->len
+= tmp_skb
->len
;
1255 skb
->data_len
+= tmp_skb
->len
;
1256 skb
->truesize
+= tmp_skb
->truesize
;
1257 tmp_skb
->destructor
= NULL
;
1261 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1262 * to fragment the frame generated here. No matter, what transforms
1263 * how transforms change size of the packet, it will come out.
1265 if (inet
->pmtudisc
< IP_PMTUDISC_DO
)
1268 /* DF bit is set when we want to see DF on outgoing frames.
1269 * If local_df is set too, we still allow to fragment this frame
1271 if (inet
->pmtudisc
>= IP_PMTUDISC_DO
||
1272 (skb
->len
<= dst_mtu(&rt
->u
.dst
) &&
1273 ip_dont_fragment(sk
, &rt
->u
.dst
)))
1276 if (inet
->cork
.flags
& IPCORK_OPT
)
1277 opt
= inet
->cork
.opt
;
1279 if (rt
->rt_type
== RTN_MULTICAST
)
1282 ttl
= ip_select_ttl(inet
, &rt
->u
.dst
);
1284 iph
= (struct iphdr
*)skb
->data
;
1288 iph
->ihl
+= opt
->optlen
>>2;
1289 ip_options_build(skb
, opt
, inet
->cork
.addr
, rt
, 0);
1291 iph
->tos
= inet
->tos
;
1293 ip_select_ident(iph
, &rt
->u
.dst
, sk
);
1295 iph
->protocol
= sk
->sk_protocol
;
1296 iph
->saddr
= rt
->rt_src
;
1297 iph
->daddr
= rt
->rt_dst
;
1299 skb
->priority
= sk
->sk_priority
;
1300 skb
->mark
= sk
->sk_mark
;
1302 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1305 inet
->cork
.dst
= NULL
;
1306 skb_dst_set(skb
, &rt
->u
.dst
);
1308 if (iph
->protocol
== IPPROTO_ICMP
)
1309 icmp_out_count(net
, ((struct icmphdr
*)
1310 skb_transport_header(skb
))->type
);
1312 /* Netfilter gets whole the not fragmented skb. */
1313 err
= ip_local_out(skb
);
1316 err
= net_xmit_errno(err
);
1322 ip_cork_release(inet
);
1326 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1331 * Throw away all pending data on the socket.
1333 void ip_flush_pending_frames(struct sock
*sk
)
1335 struct sk_buff
*skb
;
1337 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
)
1340 ip_cork_release(inet_sk(sk
));
1345 * Fetch data from kernel space and fill in checksum if needed.
1347 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1348 int len
, int odd
, struct sk_buff
*skb
)
1352 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1353 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1358 * Generic function to send a packet as reply to another packet.
1359 * Used to send TCP resets so far. ICMP should use this function too.
1361 * Should run single threaded per socket because it uses the sock
1362 * structure to pass arguments.
1364 void ip_send_reply(struct sock
*sk
, struct sk_buff
*skb
, struct ip_reply_arg
*arg
,
1367 struct inet_sock
*inet
= inet_sk(sk
);
1369 struct ip_options opt
;
1372 struct ipcm_cookie ipc
;
1374 struct rtable
*rt
= skb_rtable(skb
);
1376 if (ip_options_echo(&replyopts
.opt
, skb
))
1379 daddr
= ipc
.addr
= rt
->rt_src
;
1383 if (replyopts
.opt
.optlen
) {
1384 ipc
.opt
= &replyopts
.opt
;
1387 daddr
= replyopts
.opt
.faddr
;
1391 struct flowi fl
= { .oif
= arg
->bound_dev_if
,
1394 .saddr
= rt
->rt_spec_dst
,
1395 .tos
= RT_TOS(ip_hdr(skb
)->tos
) } },
1396 /* Not quite clean, but right. */
1398 { .sport
= tcp_hdr(skb
)->dest
,
1399 .dport
= tcp_hdr(skb
)->source
} },
1400 .proto
= sk
->sk_protocol
,
1401 .flags
= ip_reply_arg_flowi_flags(arg
) };
1402 security_skb_classify_flow(skb
, &fl
);
1403 if (ip_route_output_key(sock_net(sk
), &rt
, &fl
))
1407 /* And let IP do all the hard work.
1409 This chunk is not reenterable, hence spinlock.
1410 Note that it uses the fact, that this function is called
1411 with locally disabled BH and that sk cannot be already spinlocked.
1414 inet
->tos
= ip_hdr(skb
)->tos
;
1415 sk
->sk_priority
= skb
->priority
;
1416 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1417 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1418 ip_append_data(sk
, ip_reply_glue_bits
, arg
->iov
->iov_base
, len
, 0,
1419 &ipc
, &rt
, MSG_DONTWAIT
);
1420 if ((skb
= skb_peek(&sk
->sk_write_queue
)) != NULL
) {
1421 if (arg
->csumoffset
>= 0)
1422 *((__sum16
*)skb_transport_header(skb
) +
1423 arg
->csumoffset
) = csum_fold(csum_add(skb
->csum
,
1425 skb
->ip_summed
= CHECKSUM_NONE
;
1426 ip_push_pending_frames(sk
);
1434 void __init
ip_init(void)
1439 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1440 igmp_mc_proc_init();
1444 EXPORT_SYMBOL(ip_generic_getfrag
);
1445 EXPORT_SYMBOL(ip_queue_xmit
);
1446 EXPORT_SYMBOL(ip_send_check
);