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
);
92 EXPORT_SYMBOL(ip_send_check
);
94 int __ip_local_out(struct sk_buff
*skb
)
96 struct iphdr
*iph
= ip_hdr(skb
);
98 iph
->tot_len
= htons(skb
->len
);
100 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
, skb
, NULL
,
101 skb_dst(skb
)->dev
, dst_output
);
104 int ip_local_out(struct sk_buff
*skb
)
108 err
= __ip_local_out(skb
);
109 if (likely(err
== 1))
110 err
= dst_output(skb
);
114 EXPORT_SYMBOL_GPL(ip_local_out
);
116 /* dev_loopback_xmit for use with netfilter. */
117 static int ip_dev_loopback_xmit(struct sk_buff
*newskb
)
119 skb_reset_mac_header(newskb
);
120 __skb_pull(newskb
, skb_network_offset(newskb
));
121 newskb
->pkt_type
= PACKET_LOOPBACK
;
122 newskb
->ip_summed
= CHECKSUM_UNNECESSARY
;
123 WARN_ON(!skb_dst(newskb
));
128 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
130 int ttl
= inet
->uc_ttl
;
133 ttl
= dst_metric(dst
, RTAX_HOPLIMIT
);
138 * Add an ip header to a skbuff and send it out.
141 int ip_build_and_send_pkt(struct sk_buff
*skb
, struct sock
*sk
,
142 __be32 saddr
, __be32 daddr
, struct ip_options
*opt
)
144 struct inet_sock
*inet
= inet_sk(sk
);
145 struct rtable
*rt
= skb_rtable(skb
);
148 /* Build the IP header. */
149 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
150 skb_reset_network_header(skb
);
154 iph
->tos
= inet
->tos
;
155 if (ip_dont_fragment(sk
, &rt
->dst
))
156 iph
->frag_off
= htons(IP_DF
);
159 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
160 iph
->daddr
= rt
->rt_dst
;
161 iph
->saddr
= rt
->rt_src
;
162 iph
->protocol
= sk
->sk_protocol
;
163 ip_select_ident(iph
, &rt
->dst
, sk
);
165 if (opt
&& opt
->optlen
) {
166 iph
->ihl
+= opt
->optlen
>>2;
167 ip_options_build(skb
, opt
, daddr
, rt
, 0);
170 skb
->priority
= sk
->sk_priority
;
171 skb
->mark
= sk
->sk_mark
;
174 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
->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
;
323 /* Skip all of this if the packet is already routed,
324 * f.e. by something like SCTP.
327 rt
= skb_rtable(skb
);
331 /* Make sure we can route this packet. */
332 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
336 /* Use correct destination address if we have options. */
337 daddr
= inet
->inet_daddr
;
342 struct flowi fl
= { .oif
= sk
->sk_bound_dev_if
,
346 .saddr
= inet
->inet_saddr
,
347 .tos
= RT_CONN_FLAGS(sk
) } },
348 .proto
= sk
->sk_protocol
,
349 .flags
= inet_sk_flowi_flags(sk
),
351 { .sport
= inet
->inet_sport
,
352 .dport
= inet
->inet_dport
} } };
354 /* If this fails, retransmit mechanism of transport layer will
355 * keep trying until route appears or the connection times
358 security_sk_classify_flow(sk
, &fl
);
359 if (ip_route_output_flow(sock_net(sk
), &rt
, &fl
, sk
, 0))
362 sk_setup_caps(sk
, &rt
->dst
);
364 skb_dst_set_noref(skb
, &rt
->dst
);
367 if (opt
&& opt
->is_strictroute
&& rt
->rt_dst
!= rt
->rt_gateway
)
370 /* OK, we know where to send it, allocate and build IP header. */
371 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0));
372 skb_reset_network_header(skb
);
374 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
375 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->local_df
)
376 iph
->frag_off
= htons(IP_DF
);
379 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
380 iph
->protocol
= sk
->sk_protocol
;
381 iph
->saddr
= rt
->rt_src
;
382 iph
->daddr
= rt
->rt_dst
;
383 /* Transport layer set skb->h.foo itself. */
385 if (opt
&& opt
->optlen
) {
386 iph
->ihl
+= opt
->optlen
>> 2;
387 ip_options_build(skb
, opt
, inet
->inet_daddr
, rt
, 0);
390 ip_select_ident_more(iph
, &rt
->dst
, sk
,
391 (skb_shinfo(skb
)->gso_segs
?: 1) - 1);
393 skb
->priority
= sk
->sk_priority
;
394 skb
->mark
= sk
->sk_mark
;
396 res
= ip_local_out(skb
);
402 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
404 return -EHOSTUNREACH
;
406 EXPORT_SYMBOL(ip_queue_xmit
);
409 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
411 to
->pkt_type
= from
->pkt_type
;
412 to
->priority
= from
->priority
;
413 to
->protocol
= from
->protocol
;
415 skb_dst_copy(to
, from
);
417 to
->mark
= from
->mark
;
419 /* Copy the flags to each fragment. */
420 IPCB(to
)->flags
= IPCB(from
)->flags
;
422 #ifdef CONFIG_NET_SCHED
423 to
->tc_index
= from
->tc_index
;
426 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
427 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
428 to
->nf_trace
= from
->nf_trace
;
430 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
431 to
->ipvs_property
= from
->ipvs_property
;
433 skb_copy_secmark(to
, from
);
437 * This IP datagram is too large to be sent in one piece. Break it up into
438 * smaller pieces (each of size equal to IP header plus
439 * a block of the data of the original IP data part) that will yet fit in a
440 * single device frame, and queue such a frame for sending.
443 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
447 struct net_device
*dev
;
448 struct sk_buff
*skb2
;
449 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
451 __be16 not_last_frag
;
452 struct rtable
*rt
= skb_rtable(skb
);
458 * Point into the IP datagram header.
463 if (unlikely((iph
->frag_off
& htons(IP_DF
)) && !skb
->local_df
)) {
464 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
465 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
466 htonl(ip_skb_dst_mtu(skb
)));
472 * Setup starting values.
476 mtu
= dst_mtu(&rt
->dst
) - hlen
; /* Size of data space */
477 #ifdef CONFIG_BRIDGE_NETFILTER
479 mtu
-= nf_bridge_mtu_reduction(skb
);
481 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
483 /* When frag_list is given, use it. First, check its validity:
484 * some transformers could create wrong frag_list or break existing
485 * one, it is not prohibited. In this case fall back to copying.
487 * LATER: this step can be merged to real generation of fragments,
488 * we can switch to copy when see the first bad fragment.
490 if (skb_has_frag_list(skb
)) {
491 struct sk_buff
*frag
;
492 int first_len
= skb_pagelen(skb
);
495 if (first_len
- hlen
> mtu
||
496 ((first_len
- hlen
) & 7) ||
497 (iph
->frag_off
& htons(IP_MF
|IP_OFFSET
)) ||
501 skb_walk_frags(skb
, frag
) {
502 /* Correct geometry. */
503 if (frag
->len
> mtu
||
504 ((frag
->len
& 7) && frag
->next
) ||
505 skb_headroom(frag
) < hlen
)
508 /* Partially cloned skb? */
509 if (skb_shared(frag
))
515 frag
->destructor
= sock_wfree
;
517 truesizes
+= frag
->truesize
;
520 /* Everything is OK. Generate! */
524 frag
= skb_shinfo(skb
)->frag_list
;
525 skb_frag_list_init(skb
);
526 skb
->data_len
= first_len
- skb_headlen(skb
);
527 skb
->truesize
-= truesizes
;
528 skb
->len
= first_len
;
529 iph
->tot_len
= htons(first_len
);
530 iph
->frag_off
= htons(IP_MF
);
534 /* Prepare header of the next frame,
535 * before previous one went down. */
537 frag
->ip_summed
= CHECKSUM_NONE
;
538 skb_reset_transport_header(frag
);
539 __skb_push(frag
, hlen
);
540 skb_reset_network_header(frag
);
541 memcpy(skb_network_header(frag
), iph
, hlen
);
543 iph
->tot_len
= htons(frag
->len
);
544 ip_copy_metadata(frag
, skb
);
546 ip_options_fragment(frag
);
547 offset
+= skb
->len
- hlen
;
548 iph
->frag_off
= htons(offset
>>3);
549 if (frag
->next
!= NULL
)
550 iph
->frag_off
|= htons(IP_MF
);
551 /* Ready, complete checksum */
558 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
568 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
577 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
582 left
= skb
->len
- hlen
; /* Space per frame */
583 ptr
= hlen
; /* Where to start from */
585 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
586 * we need to make room for the encapsulating header
588 ll_rs
= LL_RESERVED_SPACE_EXTRA(rt
->dst
.dev
, nf_bridge_pad(skb
));
591 * Fragment the datagram.
594 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
595 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
598 * Keep copying data until we run out.
603 /* IF: it doesn't fit, use 'mtu' - the data space left */
606 /* IF: we are not sending upto and including the packet end
607 then align the next start on an eight byte boundary */
615 if ((skb2
= alloc_skb(len
+hlen
+ll_rs
, GFP_ATOMIC
)) == NULL
) {
616 NETDEBUG(KERN_INFO
"IP: frag: no memory for new fragment!\n");
622 * Set up data on packet
625 ip_copy_metadata(skb2
, skb
);
626 skb_reserve(skb2
, ll_rs
);
627 skb_put(skb2
, len
+ hlen
);
628 skb_reset_network_header(skb2
);
629 skb2
->transport_header
= skb2
->network_header
+ hlen
;
632 * Charge the memory for the fragment to any owner
637 skb_set_owner_w(skb2
, skb
->sk
);
640 * Copy the packet header into the new buffer.
643 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
646 * Copy a block of the IP datagram.
648 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
653 * Fill in the new header fields.
656 iph
->frag_off
= htons((offset
>> 3));
658 /* ANK: dirty, but effective trick. Upgrade options only if
659 * the segment to be fragmented was THE FIRST (otherwise,
660 * options are already fixed) and make it ONCE
661 * on the initial skb, so that all the following fragments
662 * will inherit fixed options.
665 ip_options_fragment(skb
);
668 * Added AC : If we are fragmenting a fragment that's not the
669 * last fragment then keep MF on each bit
671 if (left
> 0 || not_last_frag
)
672 iph
->frag_off
|= htons(IP_MF
);
677 * Put this fragment into the sending queue.
679 iph
->tot_len
= htons(len
+ hlen
);
687 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
690 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
695 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
698 EXPORT_SYMBOL(ip_fragment
);
701 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
703 struct iovec
*iov
= from
;
705 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
706 if (memcpy_fromiovecend(to
, iov
, offset
, len
) < 0)
710 if (csum_partial_copy_fromiovecend(to
, iov
, offset
, len
, &csum
) < 0)
712 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
716 EXPORT_SYMBOL(ip_generic_getfrag
);
719 csum_page(struct page
*page
, int offset
, int copy
)
724 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
729 static inline int ip_ufo_append_data(struct sock
*sk
,
730 int getfrag(void *from
, char *to
, int offset
, int len
,
731 int odd
, struct sk_buff
*skb
),
732 void *from
, int length
, int hh_len
, int fragheaderlen
,
733 int transhdrlen
, int mtu
, unsigned int flags
)
738 /* There is support for UDP fragmentation offload by network
739 * device, so create one single skb packet containing complete
742 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
743 skb
= sock_alloc_send_skb(sk
,
744 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
745 (flags
& MSG_DONTWAIT
), &err
);
750 /* reserve space for Hardware header */
751 skb_reserve(skb
, hh_len
);
753 /* create space for UDP/IP header */
754 skb_put(skb
, fragheaderlen
+ transhdrlen
);
756 /* initialize network header pointer */
757 skb_reset_network_header(skb
);
759 /* initialize protocol header pointer */
760 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
762 skb
->ip_summed
= CHECKSUM_PARTIAL
;
764 sk
->sk_sndmsg_off
= 0;
766 /* specify the length of each IP datagram fragment */
767 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
768 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
769 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
772 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
773 (length
- transhdrlen
));
777 * ip_append_data() and ip_append_page() can make one large IP datagram
778 * from many pieces of data. Each pieces will be holded on the socket
779 * until ip_push_pending_frames() is called. Each piece can be a page
782 * Not only UDP, other transport protocols - e.g. raw sockets - can use
783 * this interface potentially.
785 * LATER: length must be adjusted by pad at tail, when it is required.
787 int ip_append_data(struct sock
*sk
,
788 int getfrag(void *from
, char *to
, int offset
, int len
,
789 int odd
, struct sk_buff
*skb
),
790 void *from
, int length
, int transhdrlen
,
791 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
794 struct inet_sock
*inet
= inet_sk(sk
);
797 struct ip_options
*opt
= NULL
;
804 unsigned int maxfraglen
, fragheaderlen
;
805 int csummode
= CHECKSUM_NONE
;
811 if (skb_queue_empty(&sk
->sk_write_queue
)) {
817 if (inet
->cork
.opt
== NULL
) {
818 inet
->cork
.opt
= kmalloc(sizeof(struct ip_options
) + 40, sk
->sk_allocation
);
819 if (unlikely(inet
->cork
.opt
== NULL
))
822 memcpy(inet
->cork
.opt
, opt
, sizeof(struct ip_options
)+opt
->optlen
);
823 inet
->cork
.flags
|= IPCORK_OPT
;
824 inet
->cork
.addr
= ipc
->addr
;
830 * We steal reference to this route, caller should not release it
833 inet
->cork
.fragsize
= mtu
= inet
->pmtudisc
== IP_PMTUDISC_PROBE
?
835 dst_mtu(rt
->dst
.path
);
836 inet
->cork
.dst
= &rt
->dst
;
837 inet
->cork
.length
= 0;
838 sk
->sk_sndmsg_page
= NULL
;
839 sk
->sk_sndmsg_off
= 0;
840 exthdrlen
= rt
->dst
.header_len
;
842 transhdrlen
+= exthdrlen
;
844 rt
= (struct rtable
*)inet
->cork
.dst
;
845 if (inet
->cork
.flags
& IPCORK_OPT
)
846 opt
= inet
->cork
.opt
;
850 mtu
= inet
->cork
.fragsize
;
852 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
854 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
855 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
857 if (inet
->cork
.length
+ length
> 0xFFFF - fragheaderlen
) {
858 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->inet_dport
,
864 * transhdrlen > 0 means that this is the first fragment and we wish
865 * it won't be fragmented in the future.
868 length
+ fragheaderlen
<= mtu
&&
869 rt
->dst
.dev
->features
& NETIF_F_V4_CSUM
&&
871 csummode
= CHECKSUM_PARTIAL
;
873 skb
= skb_peek_tail(&sk
->sk_write_queue
);
875 inet
->cork
.length
+= length
;
876 if (((length
> mtu
) || (skb
&& skb_is_gso(skb
))) &&
877 (sk
->sk_protocol
== IPPROTO_UDP
) &&
878 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
879 err
= ip_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
880 fragheaderlen
, transhdrlen
, mtu
,
887 /* So, what's going on in the loop below?
889 * We use calculated fragment length to generate chained skb,
890 * each of segments is IP fragment ready for sending to network after
891 * adding appropriate IP header.
898 /* Check if the remaining data fits into current packet. */
899 copy
= mtu
- skb
->len
;
901 copy
= maxfraglen
- skb
->len
;
904 unsigned int datalen
;
905 unsigned int fraglen
;
906 unsigned int fraggap
;
907 unsigned int alloclen
;
908 struct sk_buff
*skb_prev
;
912 fraggap
= skb_prev
->len
- maxfraglen
;
917 * If remaining data exceeds the mtu,
918 * we know we need more fragment(s).
920 datalen
= length
+ fraggap
;
921 if (datalen
> mtu
- fragheaderlen
)
922 datalen
= maxfraglen
- fragheaderlen
;
923 fraglen
= datalen
+ fragheaderlen
;
925 if ((flags
& MSG_MORE
) &&
926 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
929 alloclen
= datalen
+ fragheaderlen
;
931 /* The last fragment gets additional space at tail.
932 * Note, with MSG_MORE we overallocate on fragments,
933 * because we have no idea what fragment will be
936 if (datalen
== length
+ fraggap
)
937 alloclen
+= rt
->dst
.trailer_len
;
940 skb
= sock_alloc_send_skb(sk
,
941 alloclen
+ hh_len
+ 15,
942 (flags
& MSG_DONTWAIT
), &err
);
945 if (atomic_read(&sk
->sk_wmem_alloc
) <=
947 skb
= sock_wmalloc(sk
,
948 alloclen
+ hh_len
+ 15, 1,
950 if (unlikely(skb
== NULL
))
953 /* only the initial fragment is
961 * Fill in the control structures
963 skb
->ip_summed
= csummode
;
965 skb_reserve(skb
, hh_len
);
966 skb_shinfo(skb
)->tx_flags
= ipc
->tx_flags
;
969 * Find where to start putting bytes.
971 data
= skb_put(skb
, fraglen
);
972 skb_set_network_header(skb
, exthdrlen
);
973 skb
->transport_header
= (skb
->network_header
+
975 data
+= fragheaderlen
;
978 skb
->csum
= skb_copy_and_csum_bits(
979 skb_prev
, maxfraglen
,
980 data
+ transhdrlen
, fraggap
, 0);
981 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
984 pskb_trim_unique(skb_prev
, maxfraglen
);
987 copy
= datalen
- transhdrlen
- fraggap
;
988 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
995 length
-= datalen
- fraggap
;
998 csummode
= CHECKSUM_NONE
;
1001 * Put the packet on the pending queue.
1003 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1010 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
)) {
1014 if (getfrag(from
, skb_put(skb
, copy
),
1015 offset
, copy
, off
, skb
) < 0) {
1016 __skb_trim(skb
, off
);
1021 int i
= skb_shinfo(skb
)->nr_frags
;
1022 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
1023 struct page
*page
= sk
->sk_sndmsg_page
;
1024 int off
= sk
->sk_sndmsg_off
;
1027 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
1030 if (page
!= frag
->page
) {
1031 if (i
== MAX_SKB_FRAGS
) {
1036 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
1037 frag
= &skb_shinfo(skb
)->frags
[i
];
1039 } else if (i
< MAX_SKB_FRAGS
) {
1040 if (copy
> PAGE_SIZE
)
1042 page
= alloc_pages(sk
->sk_allocation
, 0);
1047 sk
->sk_sndmsg_page
= page
;
1048 sk
->sk_sndmsg_off
= 0;
1050 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1051 frag
= &skb_shinfo(skb
)->frags
[i
];
1056 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1060 sk
->sk_sndmsg_off
+= copy
;
1063 skb
->data_len
+= copy
;
1064 skb
->truesize
+= copy
;
1065 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1074 inet
->cork
.length
-= length
;
1075 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1079 ssize_t
ip_append_page(struct sock
*sk
, struct page
*page
,
1080 int offset
, size_t size
, int flags
)
1082 struct inet_sock
*inet
= inet_sk(sk
);
1083 struct sk_buff
*skb
;
1085 struct ip_options
*opt
= NULL
;
1090 unsigned int maxfraglen
, fragheaderlen
, fraggap
;
1095 if (flags
&MSG_PROBE
)
1098 if (skb_queue_empty(&sk
->sk_write_queue
))
1101 rt
= (struct rtable
*)inet
->cork
.dst
;
1102 if (inet
->cork
.flags
& IPCORK_OPT
)
1103 opt
= inet
->cork
.opt
;
1105 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1108 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1109 mtu
= inet
->cork
.fragsize
;
1111 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1112 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1114 if (inet
->cork
.length
+ size
> 0xFFFF - fragheaderlen
) {
1115 ip_local_error(sk
, EMSGSIZE
, rt
->rt_dst
, inet
->inet_dport
, mtu
);
1119 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1122 inet
->cork
.length
+= size
;
1123 if ((size
+ skb
->len
> mtu
) &&
1124 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1125 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
1126 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1127 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1134 if (skb_is_gso(skb
))
1138 /* Check if the remaining data fits into current packet. */
1139 len
= mtu
- skb
->len
;
1141 len
= maxfraglen
- skb
->len
;
1144 struct sk_buff
*skb_prev
;
1148 fraggap
= skb_prev
->len
- maxfraglen
;
1150 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1151 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1152 if (unlikely(!skb
)) {
1158 * Fill in the control structures
1160 skb
->ip_summed
= CHECKSUM_NONE
;
1162 skb_reserve(skb
, hh_len
);
1165 * Find where to start putting bytes.
1167 skb_put(skb
, fragheaderlen
+ fraggap
);
1168 skb_reset_network_header(skb
);
1169 skb
->transport_header
= (skb
->network_header
+
1172 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1174 skb_transport_header(skb
),
1176 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1178 pskb_trim_unique(skb_prev
, maxfraglen
);
1182 * Put the packet on the pending queue.
1184 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1188 i
= skb_shinfo(skb
)->nr_frags
;
1191 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1192 skb_shinfo(skb
)->frags
[i
-1].size
+= len
;
1193 } else if (i
< MAX_SKB_FRAGS
) {
1195 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1201 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1203 csum
= csum_page(page
, offset
, len
);
1204 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1208 skb
->data_len
+= len
;
1209 skb
->truesize
+= len
;
1210 atomic_add(len
, &sk
->sk_wmem_alloc
);
1217 inet
->cork
.length
-= size
;
1218 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1222 static void ip_cork_release(struct inet_sock
*inet
)
1224 inet
->cork
.flags
&= ~IPCORK_OPT
;
1225 kfree(inet
->cork
.opt
);
1226 inet
->cork
.opt
= NULL
;
1227 dst_release(inet
->cork
.dst
);
1228 inet
->cork
.dst
= NULL
;
1232 * Combined all pending IP fragments on the socket as one IP datagram
1233 * and push them out.
1235 int ip_push_pending_frames(struct sock
*sk
)
1237 struct sk_buff
*skb
, *tmp_skb
;
1238 struct sk_buff
**tail_skb
;
1239 struct inet_sock
*inet
= inet_sk(sk
);
1240 struct net
*net
= sock_net(sk
);
1241 struct ip_options
*opt
= NULL
;
1242 struct rtable
*rt
= (struct rtable
*)inet
->cork
.dst
;
1248 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1250 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1252 /* move skb->data to ip header from ext header */
1253 if (skb
->data
< skb_network_header(skb
))
1254 __skb_pull(skb
, skb_network_offset(skb
));
1255 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1256 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1257 *tail_skb
= tmp_skb
;
1258 tail_skb
= &(tmp_skb
->next
);
1259 skb
->len
+= tmp_skb
->len
;
1260 skb
->data_len
+= tmp_skb
->len
;
1261 skb
->truesize
+= tmp_skb
->truesize
;
1262 tmp_skb
->destructor
= NULL
;
1266 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1267 * to fragment the frame generated here. No matter, what transforms
1268 * how transforms change size of the packet, it will come out.
1270 if (inet
->pmtudisc
< IP_PMTUDISC_DO
)
1273 /* DF bit is set when we want to see DF on outgoing frames.
1274 * If local_df is set too, we still allow to fragment this frame
1276 if (inet
->pmtudisc
>= IP_PMTUDISC_DO
||
1277 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1278 ip_dont_fragment(sk
, &rt
->dst
)))
1281 if (inet
->cork
.flags
& IPCORK_OPT
)
1282 opt
= inet
->cork
.opt
;
1284 if (rt
->rt_type
== RTN_MULTICAST
)
1287 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1289 iph
= (struct iphdr
*)skb
->data
;
1293 iph
->ihl
+= opt
->optlen
>>2;
1294 ip_options_build(skb
, opt
, inet
->cork
.addr
, rt
, 0);
1296 iph
->tos
= inet
->tos
;
1298 ip_select_ident(iph
, &rt
->dst
, sk
);
1300 iph
->protocol
= sk
->sk_protocol
;
1301 iph
->saddr
= rt
->rt_src
;
1302 iph
->daddr
= rt
->rt_dst
;
1304 skb
->priority
= sk
->sk_priority
;
1305 skb
->mark
= sk
->sk_mark
;
1307 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1310 inet
->cork
.dst
= NULL
;
1311 skb_dst_set(skb
, &rt
->dst
);
1313 if (iph
->protocol
== IPPROTO_ICMP
)
1314 icmp_out_count(net
, ((struct icmphdr
*)
1315 skb_transport_header(skb
))->type
);
1317 /* Netfilter gets whole the not fragmented skb. */
1318 err
= ip_local_out(skb
);
1321 err
= net_xmit_errno(err
);
1327 ip_cork_release(inet
);
1331 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1336 * Throw away all pending data on the socket.
1338 void ip_flush_pending_frames(struct sock
*sk
)
1340 struct sk_buff
*skb
;
1342 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
)
1345 ip_cork_release(inet_sk(sk
));
1350 * Fetch data from kernel space and fill in checksum if needed.
1352 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1353 int len
, int odd
, struct sk_buff
*skb
)
1357 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1358 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1363 * Generic function to send a packet as reply to another packet.
1364 * Used to send TCP resets so far. ICMP should use this function too.
1366 * Should run single threaded per socket because it uses the sock
1367 * structure to pass arguments.
1369 void ip_send_reply(struct sock
*sk
, struct sk_buff
*skb
, struct ip_reply_arg
*arg
,
1372 struct inet_sock
*inet
= inet_sk(sk
);
1374 struct ip_options opt
;
1377 struct ipcm_cookie ipc
;
1379 struct rtable
*rt
= skb_rtable(skb
);
1381 if (ip_options_echo(&replyopts
.opt
, skb
))
1384 daddr
= ipc
.addr
= rt
->rt_src
;
1388 if (replyopts
.opt
.optlen
) {
1389 ipc
.opt
= &replyopts
.opt
;
1392 daddr
= replyopts
.opt
.faddr
;
1396 struct flowi fl
= { .oif
= arg
->bound_dev_if
,
1399 .saddr
= rt
->rt_spec_dst
,
1400 .tos
= RT_TOS(ip_hdr(skb
)->tos
) } },
1401 /* Not quite clean, but right. */
1403 { .sport
= tcp_hdr(skb
)->dest
,
1404 .dport
= tcp_hdr(skb
)->source
} },
1405 .proto
= sk
->sk_protocol
,
1406 .flags
= ip_reply_arg_flowi_flags(arg
) };
1407 security_skb_classify_flow(skb
, &fl
);
1408 if (ip_route_output_key(sock_net(sk
), &rt
, &fl
))
1412 /* And let IP do all the hard work.
1414 This chunk is not reenterable, hence spinlock.
1415 Note that it uses the fact, that this function is called
1416 with locally disabled BH and that sk cannot be already spinlocked.
1419 inet
->tos
= ip_hdr(skb
)->tos
;
1420 sk
->sk_priority
= skb
->priority
;
1421 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1422 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1423 ip_append_data(sk
, ip_reply_glue_bits
, arg
->iov
->iov_base
, len
, 0,
1424 &ipc
, &rt
, MSG_DONTWAIT
);
1425 if ((skb
= skb_peek(&sk
->sk_write_queue
)) != NULL
) {
1426 if (arg
->csumoffset
>= 0)
1427 *((__sum16
*)skb_transport_header(skb
) +
1428 arg
->csumoffset
) = csum_fold(csum_add(skb
->csum
,
1430 skb
->ip_summed
= CHECKSUM_NONE
;
1431 ip_push_pending_frames(sk
);
1439 void __init
ip_init(void)
1444 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1445 igmp_mc_proc_init();