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 <linux/module.h>
47 #include <linux/types.h>
48 #include <linux/kernel.h>
50 #include <linux/string.h>
51 #include <linux/errno.h>
52 #include <linux/highmem.h>
53 #include <linux/slab.h>
55 #include <linux/socket.h>
56 #include <linux/sockios.h>
58 #include <linux/inet.h>
59 #include <linux/netdevice.h>
60 #include <linux/etherdevice.h>
61 #include <linux/proc_fs.h>
62 #include <linux/stat.h>
63 #include <linux/init.h>
67 #include <net/protocol.h>
68 #include <net/route.h>
70 #include <linux/skbuff.h>
74 #include <net/checksum.h>
75 #include <net/inetpeer.h>
76 #include <linux/igmp.h>
77 #include <linux/netfilter_ipv4.h>
78 #include <linux/netfilter_bridge.h>
79 #include <linux/mroute.h>
80 #include <linux/netlink.h>
81 #include <linux/tcp.h>
83 int sysctl_ip_default_ttl __read_mostly
= IPDEFTTL
;
84 EXPORT_SYMBOL(sysctl_ip_default_ttl
);
86 static int ip_fragment(struct sock
*sk
, struct sk_buff
*skb
,
88 int (*output
)(struct sock
*, struct sk_buff
*));
90 /* Generate a checksum for an outgoing IP datagram. */
91 void ip_send_check(struct iphdr
*iph
)
94 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
96 EXPORT_SYMBOL(ip_send_check
);
98 static int __ip_local_out_sk(struct sock
*sk
, struct sk_buff
*skb
)
100 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
101 struct iphdr
*iph
= ip_hdr(skb
);
103 iph
->tot_len
= htons(skb
->len
);
105 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
,
106 net
, sk
, skb
, NULL
, skb_dst(skb
)->dev
,
110 int __ip_local_out(struct sk_buff
*skb
)
112 return __ip_local_out_sk(skb
->sk
, skb
);
115 int ip_local_out_sk(struct sock
*sk
, struct sk_buff
*skb
)
119 err
= __ip_local_out(skb
);
120 if (likely(err
== 1))
121 err
= dst_output(sk
, skb
);
125 EXPORT_SYMBOL_GPL(ip_local_out_sk
);
127 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
129 int ttl
= inet
->uc_ttl
;
132 ttl
= ip4_dst_hoplimit(dst
);
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_rcu
*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
->opt
.optlen
: 0));
149 skb_reset_network_header(skb
);
153 iph
->tos
= inet
->tos
;
154 if (ip_dont_fragment(sk
, &rt
->dst
))
155 iph
->frag_off
= htons(IP_DF
);
158 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
159 iph
->daddr
= (opt
&& opt
->opt
.srr
? opt
->opt
.faddr
: daddr
);
161 iph
->protocol
= sk
->sk_protocol
;
162 ip_select_ident(sock_net(sk
), skb
, sk
);
164 if (opt
&& opt
->opt
.optlen
) {
165 iph
->ihl
+= opt
->opt
.optlen
>>2;
166 ip_options_build(skb
, &opt
->opt
, daddr
, rt
, 0);
169 skb
->priority
= sk
->sk_priority
;
170 skb
->mark
= sk
->sk_mark
;
173 return ip_local_out(skb
);
175 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
177 static int ip_finish_output2(struct sock
*sk
, struct sk_buff
*skb
)
179 struct dst_entry
*dst
= skb_dst(skb
);
180 struct rtable
*rt
= (struct rtable
*)dst
;
181 struct net_device
*dev
= dst
->dev
;
182 struct net
*net
= dev_net(dev
);
183 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
184 struct neighbour
*neigh
;
187 if (rt
->rt_type
== RTN_MULTICAST
) {
188 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUTMCAST
, skb
->len
);
189 } else if (rt
->rt_type
== RTN_BROADCAST
)
190 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUTBCAST
, skb
->len
);
192 /* Be paranoid, rather than too clever. */
193 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
194 struct sk_buff
*skb2
;
196 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
202 skb_set_owner_w(skb2
, skb
->sk
);
208 nexthop
= (__force u32
) rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
209 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
210 if (unlikely(!neigh
))
211 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
212 if (!IS_ERR(neigh
)) {
213 int res
= dst_neigh_output(dst
, neigh
, skb
);
215 rcu_read_unlock_bh();
218 rcu_read_unlock_bh();
220 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
226 static int ip_finish_output_gso(struct sock
*sk
, struct sk_buff
*skb
,
229 netdev_features_t features
;
230 struct sk_buff
*segs
;
233 /* common case: locally created skb or seglen is <= mtu */
234 if (((IPCB(skb
)->flags
& IPSKB_FORWARDED
) == 0) ||
235 skb_gso_network_seglen(skb
) <= mtu
)
236 return ip_finish_output2(sk
, skb
);
238 /* Slowpath - GSO segment length is exceeding the dst MTU.
240 * This can happen in two cases:
241 * 1) TCP GRO packet, DF bit not set
242 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
243 * from host network stack.
245 features
= netif_skb_features(skb
);
246 segs
= skb_gso_segment(skb
, features
& ~NETIF_F_GSO_MASK
);
247 if (IS_ERR_OR_NULL(segs
)) {
255 struct sk_buff
*nskb
= segs
->next
;
259 err
= ip_fragment(sk
, segs
, mtu
, ip_finish_output2
);
269 static int ip_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
273 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
274 /* Policy lookup after SNAT yielded a new policy */
275 if (skb_dst(skb
)->xfrm
) {
276 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
277 return dst_output(sk
, skb
);
280 mtu
= ip_skb_dst_mtu(skb
);
282 return ip_finish_output_gso(sk
, skb
, mtu
);
284 if (skb
->len
> mtu
|| (IPCB(skb
)->flags
& IPSKB_FRAG_PMTU
))
285 return ip_fragment(sk
, skb
, mtu
, ip_finish_output2
);
287 return ip_finish_output2(sk
, skb
);
290 int ip_mc_output(struct sock
*sk
, struct sk_buff
*skb
)
292 struct rtable
*rt
= skb_rtable(skb
);
293 struct net_device
*dev
= rt
->dst
.dev
;
294 struct net
*net
= dev_net(dev
);
297 * If the indicated interface is up and running, send the packet.
299 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
302 skb
->protocol
= htons(ETH_P_IP
);
305 * Multicasts are looped back for other local users
308 if (rt
->rt_flags
&RTCF_MULTICAST
) {
310 #ifdef CONFIG_IP_MROUTE
311 /* Small optimization: do not loopback not local frames,
312 which returned after forwarding; they will be dropped
313 by ip_mr_input in any case.
314 Note, that local frames are looped back to be delivered
317 This check is duplicated in ip_mr_input at the moment.
320 ((rt
->rt_flags
& RTCF_LOCAL
) ||
321 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
324 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
326 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
327 net
, sk
, newskb
, NULL
, newskb
->dev
,
331 /* Multicasts with ttl 0 must not go beyond the host */
333 if (ip_hdr(skb
)->ttl
== 0) {
339 if (rt
->rt_flags
&RTCF_BROADCAST
) {
340 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
342 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
343 net
, sk
, newskb
, NULL
, newskb
->dev
,
347 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
348 net
, sk
, skb
, NULL
, skb
->dev
,
350 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
353 int ip_output(struct sock
*sk
, struct sk_buff
*skb
)
355 struct net_device
*dev
= skb_dst(skb
)->dev
;
356 struct net
*net
= dev_net(dev
);
358 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
361 skb
->protocol
= htons(ETH_P_IP
);
363 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
364 net
, sk
, skb
, NULL
, dev
,
366 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
370 * copy saddr and daddr, possibly using 64bit load/stores
372 * iph->saddr = fl4->saddr;
373 * iph->daddr = fl4->daddr;
375 static void ip_copy_addrs(struct iphdr
*iph
, const struct flowi4
*fl4
)
377 BUILD_BUG_ON(offsetof(typeof(*fl4
), daddr
) !=
378 offsetof(typeof(*fl4
), saddr
) + sizeof(fl4
->saddr
));
379 memcpy(&iph
->saddr
, &fl4
->saddr
,
380 sizeof(fl4
->saddr
) + sizeof(fl4
->daddr
));
383 /* Note: skb->sk can be different from sk, in case of tunnels */
384 int ip_queue_xmit(struct sock
*sk
, struct sk_buff
*skb
, struct flowi
*fl
)
386 struct inet_sock
*inet
= inet_sk(sk
);
387 struct ip_options_rcu
*inet_opt
;
393 /* Skip all of this if the packet is already routed,
394 * f.e. by something like SCTP.
397 inet_opt
= rcu_dereference(inet
->inet_opt
);
399 rt
= skb_rtable(skb
);
403 /* Make sure we can route this packet. */
404 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
408 /* Use correct destination address if we have options. */
409 daddr
= inet
->inet_daddr
;
410 if (inet_opt
&& inet_opt
->opt
.srr
)
411 daddr
= inet_opt
->opt
.faddr
;
413 /* If this fails, retransmit mechanism of transport layer will
414 * keep trying until route appears or the connection times
417 rt
= ip_route_output_ports(sock_net(sk
), fl4
, sk
,
418 daddr
, inet
->inet_saddr
,
423 sk
->sk_bound_dev_if
);
426 sk_setup_caps(sk
, &rt
->dst
);
428 skb_dst_set_noref(skb
, &rt
->dst
);
431 if (inet_opt
&& inet_opt
->opt
.is_strictroute
&& rt
->rt_uses_gateway
)
434 /* OK, we know where to send it, allocate and build IP header. */
435 skb_push(skb
, sizeof(struct iphdr
) + (inet_opt
? inet_opt
->opt
.optlen
: 0));
436 skb_reset_network_header(skb
);
438 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
439 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->ignore_df
)
440 iph
->frag_off
= htons(IP_DF
);
443 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
444 iph
->protocol
= sk
->sk_protocol
;
445 ip_copy_addrs(iph
, fl4
);
447 /* Transport layer set skb->h.foo itself. */
449 if (inet_opt
&& inet_opt
->opt
.optlen
) {
450 iph
->ihl
+= inet_opt
->opt
.optlen
>> 2;
451 ip_options_build(skb
, &inet_opt
->opt
, inet
->inet_daddr
, rt
, 0);
454 ip_select_ident_segs(sock_net(sk
), skb
, sk
,
455 skb_shinfo(skb
)->gso_segs
?: 1);
457 /* TODO : should we use skb->sk here instead of sk ? */
458 skb
->priority
= sk
->sk_priority
;
459 skb
->mark
= sk
->sk_mark
;
461 res
= ip_local_out(skb
);
467 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
469 return -EHOSTUNREACH
;
471 EXPORT_SYMBOL(ip_queue_xmit
);
473 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
475 to
->pkt_type
= from
->pkt_type
;
476 to
->priority
= from
->priority
;
477 to
->protocol
= from
->protocol
;
479 skb_dst_copy(to
, from
);
481 to
->mark
= from
->mark
;
483 /* Copy the flags to each fragment. */
484 IPCB(to
)->flags
= IPCB(from
)->flags
;
486 #ifdef CONFIG_NET_SCHED
487 to
->tc_index
= from
->tc_index
;
490 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
491 to
->ipvs_property
= from
->ipvs_property
;
493 skb_copy_secmark(to
, from
);
496 static int ip_fragment(struct sock
*sk
, struct sk_buff
*skb
,
498 int (*output
)(struct sock
*, struct sk_buff
*))
500 struct iphdr
*iph
= ip_hdr(skb
);
502 if ((iph
->frag_off
& htons(IP_DF
)) == 0)
503 return ip_do_fragment(sk
, skb
, output
);
505 if (unlikely(!skb
->ignore_df
||
506 (IPCB(skb
)->frag_max_size
&&
507 IPCB(skb
)->frag_max_size
> mtu
))) {
508 struct net
*net
= dev_net(skb_rtable(skb
)->dst
.dev
);
510 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
511 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
517 return ip_do_fragment(sk
, skb
, output
);
521 * This IP datagram is too large to be sent in one piece. Break it up into
522 * smaller pieces (each of size equal to IP header plus
523 * a block of the data of the original IP data part) that will yet fit in a
524 * single device frame, and queue such a frame for sending.
527 int ip_do_fragment(struct sock
*sk
, struct sk_buff
*skb
,
528 int (*output
)(struct sock
*, struct sk_buff
*))
532 struct net_device
*dev
;
533 struct sk_buff
*skb2
;
534 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
536 __be16 not_last_frag
;
537 struct rtable
*rt
= skb_rtable(skb
);
545 * Point into the IP datagram header.
550 mtu
= ip_skb_dst_mtu(skb
);
551 if (IPCB(skb
)->frag_max_size
&& IPCB(skb
)->frag_max_size
< mtu
)
552 mtu
= IPCB(skb
)->frag_max_size
;
555 * Setup starting values.
559 mtu
= mtu
- hlen
; /* Size of data space */
560 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
562 /* When frag_list is given, use it. First, check its validity:
563 * some transformers could create wrong frag_list or break existing
564 * one, it is not prohibited. In this case fall back to copying.
566 * LATER: this step can be merged to real generation of fragments,
567 * we can switch to copy when see the first bad fragment.
569 if (skb_has_frag_list(skb
)) {
570 struct sk_buff
*frag
, *frag2
;
571 int first_len
= skb_pagelen(skb
);
573 if (first_len
- hlen
> mtu
||
574 ((first_len
- hlen
) & 7) ||
575 ip_is_fragment(iph
) ||
579 skb_walk_frags(skb
, frag
) {
580 /* Correct geometry. */
581 if (frag
->len
> mtu
||
582 ((frag
->len
& 7) && frag
->next
) ||
583 skb_headroom(frag
) < hlen
)
584 goto slow_path_clean
;
586 /* Partially cloned skb? */
587 if (skb_shared(frag
))
588 goto slow_path_clean
;
593 frag
->destructor
= sock_wfree
;
595 skb
->truesize
-= frag
->truesize
;
598 /* Everything is OK. Generate! */
602 frag
= skb_shinfo(skb
)->frag_list
;
603 skb_frag_list_init(skb
);
604 skb
->data_len
= first_len
- skb_headlen(skb
);
605 skb
->len
= first_len
;
606 iph
->tot_len
= htons(first_len
);
607 iph
->frag_off
= htons(IP_MF
);
611 /* Prepare header of the next frame,
612 * before previous one went down. */
614 frag
->ip_summed
= CHECKSUM_NONE
;
615 skb_reset_transport_header(frag
);
616 __skb_push(frag
, hlen
);
617 skb_reset_network_header(frag
);
618 memcpy(skb_network_header(frag
), iph
, hlen
);
620 iph
->tot_len
= htons(frag
->len
);
621 ip_copy_metadata(frag
, skb
);
623 ip_options_fragment(frag
);
624 offset
+= skb
->len
- hlen
;
625 iph
->frag_off
= htons(offset
>>3);
627 iph
->frag_off
|= htons(IP_MF
);
628 /* Ready, complete checksum */
632 err
= output(sk
, skb
);
635 IP_INC_STATS(net
, IPSTATS_MIB_FRAGCREATES
);
645 IP_INC_STATS(net
, IPSTATS_MIB_FRAGOKS
);
654 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
658 skb_walk_frags(skb
, frag2
) {
662 frag2
->destructor
= NULL
;
663 skb
->truesize
+= frag2
->truesize
;
668 /* for offloaded checksums cleanup checksum before fragmentation */
669 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) && skb_checksum_help(skb
))
673 left
= skb
->len
- hlen
; /* Space per frame */
674 ptr
= hlen
; /* Where to start from */
676 ll_rs
= LL_RESERVED_SPACE(rt
->dst
.dev
);
679 * Fragment the datagram.
682 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
683 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
686 * Keep copying data until we run out.
691 /* IF: it doesn't fit, use 'mtu' - the data space left */
694 /* IF: we are not sending up to and including the packet end
695 then align the next start on an eight byte boundary */
700 /* Allocate buffer */
701 skb2
= alloc_skb(len
+ hlen
+ ll_rs
, GFP_ATOMIC
);
708 * Set up data on packet
711 ip_copy_metadata(skb2
, skb
);
712 skb_reserve(skb2
, ll_rs
);
713 skb_put(skb2
, len
+ hlen
);
714 skb_reset_network_header(skb2
);
715 skb2
->transport_header
= skb2
->network_header
+ hlen
;
718 * Charge the memory for the fragment to any owner
723 skb_set_owner_w(skb2
, skb
->sk
);
726 * Copy the packet header into the new buffer.
729 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
732 * Copy a block of the IP datagram.
734 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
739 * Fill in the new header fields.
742 iph
->frag_off
= htons((offset
>> 3));
744 if (IPCB(skb
)->flags
& IPSKB_FRAG_PMTU
)
745 iph
->frag_off
|= htons(IP_DF
);
747 /* ANK: dirty, but effective trick. Upgrade options only if
748 * the segment to be fragmented was THE FIRST (otherwise,
749 * options are already fixed) and make it ONCE
750 * on the initial skb, so that all the following fragments
751 * will inherit fixed options.
754 ip_options_fragment(skb
);
757 * Added AC : If we are fragmenting a fragment that's not the
758 * last fragment then keep MF on each bit
760 if (left
> 0 || not_last_frag
)
761 iph
->frag_off
|= htons(IP_MF
);
766 * Put this fragment into the sending queue.
768 iph
->tot_len
= htons(len
+ hlen
);
772 err
= output(sk
, skb2
);
776 IP_INC_STATS(net
, IPSTATS_MIB_FRAGCREATES
);
779 IP_INC_STATS(net
, IPSTATS_MIB_FRAGOKS
);
784 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
787 EXPORT_SYMBOL(ip_do_fragment
);
790 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
792 struct msghdr
*msg
= from
;
794 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
795 if (copy_from_iter(to
, len
, &msg
->msg_iter
) != len
)
799 if (csum_and_copy_from_iter(to
, len
, &csum
, &msg
->msg_iter
) != len
)
801 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
805 EXPORT_SYMBOL(ip_generic_getfrag
);
808 csum_page(struct page
*page
, int offset
, int copy
)
813 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
818 static inline int ip_ufo_append_data(struct sock
*sk
,
819 struct sk_buff_head
*queue
,
820 int getfrag(void *from
, char *to
, int offset
, int len
,
821 int odd
, struct sk_buff
*skb
),
822 void *from
, int length
, int hh_len
, int fragheaderlen
,
823 int transhdrlen
, int maxfraglen
, unsigned int flags
)
828 /* There is support for UDP fragmentation offload by network
829 * device, so create one single skb packet containing complete
832 skb
= skb_peek_tail(queue
);
834 skb
= sock_alloc_send_skb(sk
,
835 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
836 (flags
& MSG_DONTWAIT
), &err
);
841 /* reserve space for Hardware header */
842 skb_reserve(skb
, hh_len
);
844 /* create space for UDP/IP header */
845 skb_put(skb
, fragheaderlen
+ transhdrlen
);
847 /* initialize network header pointer */
848 skb_reset_network_header(skb
);
850 /* initialize protocol header pointer */
851 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
855 __skb_queue_tail(queue
, skb
);
856 } else if (skb_is_gso(skb
)) {
860 skb
->ip_summed
= CHECKSUM_PARTIAL
;
861 /* specify the length of each IP datagram fragment */
862 skb_shinfo(skb
)->gso_size
= maxfraglen
- fragheaderlen
;
863 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
866 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
867 (length
- transhdrlen
));
870 static int __ip_append_data(struct sock
*sk
,
872 struct sk_buff_head
*queue
,
873 struct inet_cork
*cork
,
874 struct page_frag
*pfrag
,
875 int getfrag(void *from
, char *to
, int offset
,
876 int len
, int odd
, struct sk_buff
*skb
),
877 void *from
, int length
, int transhdrlen
,
880 struct inet_sock
*inet
= inet_sk(sk
);
883 struct ip_options
*opt
= cork
->opt
;
890 unsigned int maxfraglen
, fragheaderlen
, maxnonfragsize
;
891 int csummode
= CHECKSUM_NONE
;
892 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
895 skb
= skb_peek_tail(queue
);
897 exthdrlen
= !skb
? rt
->dst
.header_len
: 0;
898 mtu
= cork
->fragsize
;
899 if (cork
->tx_flags
& SKBTX_ANY_SW_TSTAMP
&&
900 sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)
901 tskey
= sk
->sk_tskey
++;
903 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
905 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
906 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
907 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
909 if (cork
->length
+ length
> maxnonfragsize
- fragheaderlen
) {
910 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
911 mtu
- (opt
? opt
->optlen
: 0));
916 * transhdrlen > 0 means that this is the first fragment and we wish
917 * it won't be fragmented in the future.
920 length
+ fragheaderlen
<= mtu
&&
921 rt
->dst
.dev
->features
& NETIF_F_V4_CSUM
&&
923 csummode
= CHECKSUM_PARTIAL
;
925 cork
->length
+= length
;
926 if (((length
> mtu
) || (skb
&& skb_is_gso(skb
))) &&
927 (sk
->sk_protocol
== IPPROTO_UDP
) &&
928 (rt
->dst
.dev
->features
& NETIF_F_UFO
) && !rt
->dst
.header_len
&&
929 (sk
->sk_type
== SOCK_DGRAM
)) {
930 err
= ip_ufo_append_data(sk
, queue
, getfrag
, from
, length
,
931 hh_len
, fragheaderlen
, transhdrlen
,
938 /* So, what's going on in the loop below?
940 * We use calculated fragment length to generate chained skb,
941 * each of segments is IP fragment ready for sending to network after
942 * adding appropriate IP header.
949 /* Check if the remaining data fits into current packet. */
950 copy
= mtu
- skb
->len
;
952 copy
= maxfraglen
- skb
->len
;
955 unsigned int datalen
;
956 unsigned int fraglen
;
957 unsigned int fraggap
;
958 unsigned int alloclen
;
959 struct sk_buff
*skb_prev
;
963 fraggap
= skb_prev
->len
- maxfraglen
;
968 * If remaining data exceeds the mtu,
969 * we know we need more fragment(s).
971 datalen
= length
+ fraggap
;
972 if (datalen
> mtu
- fragheaderlen
)
973 datalen
= maxfraglen
- fragheaderlen
;
974 fraglen
= datalen
+ fragheaderlen
;
976 if ((flags
& MSG_MORE
) &&
977 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
982 alloclen
+= exthdrlen
;
984 /* The last fragment gets additional space at tail.
985 * Note, with MSG_MORE we overallocate on fragments,
986 * because we have no idea what fragment will be
989 if (datalen
== length
+ fraggap
)
990 alloclen
+= rt
->dst
.trailer_len
;
993 skb
= sock_alloc_send_skb(sk
,
994 alloclen
+ hh_len
+ 15,
995 (flags
& MSG_DONTWAIT
), &err
);
998 if (atomic_read(&sk
->sk_wmem_alloc
) <=
1000 skb
= sock_wmalloc(sk
,
1001 alloclen
+ hh_len
+ 15, 1,
1010 * Fill in the control structures
1012 skb
->ip_summed
= csummode
;
1014 skb_reserve(skb
, hh_len
);
1016 /* only the initial fragment is time stamped */
1017 skb_shinfo(skb
)->tx_flags
= cork
->tx_flags
;
1019 skb_shinfo(skb
)->tskey
= tskey
;
1023 * Find where to start putting bytes.
1025 data
= skb_put(skb
, fraglen
+ exthdrlen
);
1026 skb_set_network_header(skb
, exthdrlen
);
1027 skb
->transport_header
= (skb
->network_header
+
1029 data
+= fragheaderlen
+ exthdrlen
;
1032 skb
->csum
= skb_copy_and_csum_bits(
1033 skb_prev
, maxfraglen
,
1034 data
+ transhdrlen
, fraggap
, 0);
1035 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1038 pskb_trim_unique(skb_prev
, maxfraglen
);
1041 copy
= datalen
- transhdrlen
- fraggap
;
1042 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1049 length
-= datalen
- fraggap
;
1052 csummode
= CHECKSUM_NONE
;
1055 * Put the packet on the pending queue.
1057 __skb_queue_tail(queue
, skb
);
1064 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
)) {
1068 if (getfrag(from
, skb_put(skb
, copy
),
1069 offset
, copy
, off
, skb
) < 0) {
1070 __skb_trim(skb
, off
);
1075 int i
= skb_shinfo(skb
)->nr_frags
;
1078 if (!sk_page_frag_refill(sk
, pfrag
))
1081 if (!skb_can_coalesce(skb
, i
, pfrag
->page
,
1084 if (i
== MAX_SKB_FRAGS
)
1087 __skb_fill_page_desc(skb
, i
, pfrag
->page
,
1089 skb_shinfo(skb
)->nr_frags
= ++i
;
1090 get_page(pfrag
->page
);
1092 copy
= min_t(int, copy
, pfrag
->size
- pfrag
->offset
);
1094 page_address(pfrag
->page
) + pfrag
->offset
,
1095 offset
, copy
, skb
->len
, skb
) < 0)
1098 pfrag
->offset
+= copy
;
1099 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], copy
);
1101 skb
->data_len
+= copy
;
1102 skb
->truesize
+= copy
;
1103 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1114 cork
->length
-= length
;
1115 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1119 static int ip_setup_cork(struct sock
*sk
, struct inet_cork
*cork
,
1120 struct ipcm_cookie
*ipc
, struct rtable
**rtp
)
1122 struct ip_options_rcu
*opt
;
1126 * setup for corking.
1131 cork
->opt
= kmalloc(sizeof(struct ip_options
) + 40,
1133 if (unlikely(!cork
->opt
))
1136 memcpy(cork
->opt
, &opt
->opt
, sizeof(struct ip_options
) + opt
->opt
.optlen
);
1137 cork
->flags
|= IPCORK_OPT
;
1138 cork
->addr
= ipc
->addr
;
1144 * We steal reference to this route, caller should not release it
1147 cork
->fragsize
= ip_sk_use_pmtu(sk
) ?
1148 dst_mtu(&rt
->dst
) : rt
->dst
.dev
->mtu
;
1149 cork
->dst
= &rt
->dst
;
1151 cork
->ttl
= ipc
->ttl
;
1152 cork
->tos
= ipc
->tos
;
1153 cork
->priority
= ipc
->priority
;
1154 cork
->tx_flags
= ipc
->tx_flags
;
1160 * ip_append_data() and ip_append_page() can make one large IP datagram
1161 * from many pieces of data. Each pieces will be holded on the socket
1162 * until ip_push_pending_frames() is called. Each piece can be a page
1165 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1166 * this interface potentially.
1168 * LATER: length must be adjusted by pad at tail, when it is required.
1170 int ip_append_data(struct sock
*sk
, struct flowi4
*fl4
,
1171 int getfrag(void *from
, char *to
, int offset
, int len
,
1172 int odd
, struct sk_buff
*skb
),
1173 void *from
, int length
, int transhdrlen
,
1174 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1177 struct inet_sock
*inet
= inet_sk(sk
);
1180 if (flags
&MSG_PROBE
)
1183 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1184 err
= ip_setup_cork(sk
, &inet
->cork
.base
, ipc
, rtp
);
1191 return __ip_append_data(sk
, fl4
, &sk
->sk_write_queue
, &inet
->cork
.base
,
1192 sk_page_frag(sk
), getfrag
,
1193 from
, length
, transhdrlen
, flags
);
1196 ssize_t
ip_append_page(struct sock
*sk
, struct flowi4
*fl4
, struct page
*page
,
1197 int offset
, size_t size
, int flags
)
1199 struct inet_sock
*inet
= inet_sk(sk
);
1200 struct sk_buff
*skb
;
1202 struct ip_options
*opt
= NULL
;
1203 struct inet_cork
*cork
;
1208 unsigned int maxfraglen
, fragheaderlen
, fraggap
, maxnonfragsize
;
1213 if (flags
&MSG_PROBE
)
1216 if (skb_queue_empty(&sk
->sk_write_queue
))
1219 cork
= &inet
->cork
.base
;
1220 rt
= (struct rtable
*)cork
->dst
;
1221 if (cork
->flags
& IPCORK_OPT
)
1224 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1227 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1228 mtu
= cork
->fragsize
;
1230 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1231 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1232 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
1234 if (cork
->length
+ size
> maxnonfragsize
- fragheaderlen
) {
1235 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
1236 mtu
- (opt
? opt
->optlen
: 0));
1240 skb
= skb_peek_tail(&sk
->sk_write_queue
);
1244 cork
->length
+= size
;
1245 if ((size
+ skb
->len
> mtu
) &&
1246 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1247 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
1248 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1249 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1253 if (skb_is_gso(skb
)) {
1257 /* Check if the remaining data fits into current packet. */
1258 len
= mtu
- skb
->len
;
1260 len
= maxfraglen
- skb
->len
;
1263 struct sk_buff
*skb_prev
;
1267 fraggap
= skb_prev
->len
- maxfraglen
;
1269 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1270 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1271 if (unlikely(!skb
)) {
1277 * Fill in the control structures
1279 skb
->ip_summed
= CHECKSUM_NONE
;
1281 skb_reserve(skb
, hh_len
);
1284 * Find where to start putting bytes.
1286 skb_put(skb
, fragheaderlen
+ fraggap
);
1287 skb_reset_network_header(skb
);
1288 skb
->transport_header
= (skb
->network_header
+
1291 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1293 skb_transport_header(skb
),
1295 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1297 pskb_trim_unique(skb_prev
, maxfraglen
);
1301 * Put the packet on the pending queue.
1303 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1310 if (skb_append_pagefrags(skb
, page
, offset
, len
)) {
1315 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1317 csum
= csum_page(page
, offset
, len
);
1318 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1322 skb
->data_len
+= len
;
1323 skb
->truesize
+= len
;
1324 atomic_add(len
, &sk
->sk_wmem_alloc
);
1331 cork
->length
-= size
;
1332 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1336 static void ip_cork_release(struct inet_cork
*cork
)
1338 cork
->flags
&= ~IPCORK_OPT
;
1341 dst_release(cork
->dst
);
1346 * Combined all pending IP fragments on the socket as one IP datagram
1347 * and push them out.
1349 struct sk_buff
*__ip_make_skb(struct sock
*sk
,
1351 struct sk_buff_head
*queue
,
1352 struct inet_cork
*cork
)
1354 struct sk_buff
*skb
, *tmp_skb
;
1355 struct sk_buff
**tail_skb
;
1356 struct inet_sock
*inet
= inet_sk(sk
);
1357 struct net
*net
= sock_net(sk
);
1358 struct ip_options
*opt
= NULL
;
1359 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
1364 skb
= __skb_dequeue(queue
);
1367 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1369 /* move skb->data to ip header from ext header */
1370 if (skb
->data
< skb_network_header(skb
))
1371 __skb_pull(skb
, skb_network_offset(skb
));
1372 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1373 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1374 *tail_skb
= tmp_skb
;
1375 tail_skb
= &(tmp_skb
->next
);
1376 skb
->len
+= tmp_skb
->len
;
1377 skb
->data_len
+= tmp_skb
->len
;
1378 skb
->truesize
+= tmp_skb
->truesize
;
1379 tmp_skb
->destructor
= NULL
;
1383 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1384 * to fragment the frame generated here. No matter, what transforms
1385 * how transforms change size of the packet, it will come out.
1387 skb
->ignore_df
= ip_sk_ignore_df(sk
);
1389 /* DF bit is set when we want to see DF on outgoing frames.
1390 * If ignore_df is set too, we still allow to fragment this frame
1392 if (inet
->pmtudisc
== IP_PMTUDISC_DO
||
1393 inet
->pmtudisc
== IP_PMTUDISC_PROBE
||
1394 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1395 ip_dont_fragment(sk
, &rt
->dst
)))
1398 if (cork
->flags
& IPCORK_OPT
)
1403 else if (rt
->rt_type
== RTN_MULTICAST
)
1406 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1411 iph
->tos
= (cork
->tos
!= -1) ? cork
->tos
: inet
->tos
;
1414 iph
->protocol
= sk
->sk_protocol
;
1415 ip_copy_addrs(iph
, fl4
);
1416 ip_select_ident(net
, skb
, sk
);
1419 iph
->ihl
+= opt
->optlen
>>2;
1420 ip_options_build(skb
, opt
, cork
->addr
, rt
, 0);
1423 skb
->priority
= (cork
->tos
!= -1) ? cork
->priority
: sk
->sk_priority
;
1424 skb
->mark
= sk
->sk_mark
;
1426 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1430 skb_dst_set(skb
, &rt
->dst
);
1432 if (iph
->protocol
== IPPROTO_ICMP
)
1433 icmp_out_count(net
, ((struct icmphdr
*)
1434 skb_transport_header(skb
))->type
);
1436 ip_cork_release(cork
);
1441 int ip_send_skb(struct net
*net
, struct sk_buff
*skb
)
1445 err
= ip_local_out(skb
);
1448 err
= net_xmit_errno(err
);
1450 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1456 int ip_push_pending_frames(struct sock
*sk
, struct flowi4
*fl4
)
1458 struct sk_buff
*skb
;
1460 skb
= ip_finish_skb(sk
, fl4
);
1464 /* Netfilter gets whole the not fragmented skb. */
1465 return ip_send_skb(sock_net(sk
), skb
);
1469 * Throw away all pending data on the socket.
1471 static void __ip_flush_pending_frames(struct sock
*sk
,
1472 struct sk_buff_head
*queue
,
1473 struct inet_cork
*cork
)
1475 struct sk_buff
*skb
;
1477 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
)
1480 ip_cork_release(cork
);
1483 void ip_flush_pending_frames(struct sock
*sk
)
1485 __ip_flush_pending_frames(sk
, &sk
->sk_write_queue
, &inet_sk(sk
)->cork
.base
);
1488 struct sk_buff
*ip_make_skb(struct sock
*sk
,
1490 int getfrag(void *from
, char *to
, int offset
,
1491 int len
, int odd
, struct sk_buff
*skb
),
1492 void *from
, int length
, int transhdrlen
,
1493 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1496 struct inet_cork cork
;
1497 struct sk_buff_head queue
;
1500 if (flags
& MSG_PROBE
)
1503 __skb_queue_head_init(&queue
);
1508 err
= ip_setup_cork(sk
, &cork
, ipc
, rtp
);
1510 return ERR_PTR(err
);
1512 err
= __ip_append_data(sk
, fl4
, &queue
, &cork
,
1513 ¤t
->task_frag
, getfrag
,
1514 from
, length
, transhdrlen
, flags
);
1516 __ip_flush_pending_frames(sk
, &queue
, &cork
);
1517 return ERR_PTR(err
);
1520 return __ip_make_skb(sk
, fl4
, &queue
, &cork
);
1524 * Fetch data from kernel space and fill in checksum if needed.
1526 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1527 int len
, int odd
, struct sk_buff
*skb
)
1531 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1532 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1537 * Generic function to send a packet as reply to another packet.
1538 * Used to send some TCP resets/acks so far.
1540 void ip_send_unicast_reply(struct sock
*sk
, struct sk_buff
*skb
,
1541 const struct ip_options
*sopt
,
1542 __be32 daddr
, __be32 saddr
,
1543 const struct ip_reply_arg
*arg
,
1546 struct ip_options_data replyopts
;
1547 struct ipcm_cookie ipc
;
1549 struct rtable
*rt
= skb_rtable(skb
);
1550 struct net
*net
= sock_net(sk
);
1551 struct sk_buff
*nskb
;
1555 if (__ip_options_echo(&replyopts
.opt
.opt
, skb
, sopt
))
1564 if (replyopts
.opt
.opt
.optlen
) {
1565 ipc
.opt
= &replyopts
.opt
;
1567 if (replyopts
.opt
.opt
.srr
)
1568 daddr
= replyopts
.opt
.opt
.faddr
;
1571 oif
= arg
->bound_dev_if
;
1572 if (!oif
&& netif_index_is_vrf(net
, skb
->skb_iif
))
1575 flowi4_init_output(&fl4
, oif
,
1576 IP4_REPLY_MARK(net
, skb
->mark
),
1578 RT_SCOPE_UNIVERSE
, ip_hdr(skb
)->protocol
,
1579 ip_reply_arg_flowi_flags(arg
),
1581 tcp_hdr(skb
)->source
, tcp_hdr(skb
)->dest
);
1582 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
1583 rt
= ip_route_output_key(net
, &fl4
);
1587 inet_sk(sk
)->tos
= arg
->tos
;
1589 sk
->sk_priority
= skb
->priority
;
1590 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1591 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1592 sk
->sk_sndbuf
= sysctl_wmem_default
;
1593 err
= ip_append_data(sk
, &fl4
, ip_reply_glue_bits
, arg
->iov
->iov_base
,
1594 len
, 0, &ipc
, &rt
, MSG_DONTWAIT
);
1595 if (unlikely(err
)) {
1596 ip_flush_pending_frames(sk
);
1600 nskb
= skb_peek(&sk
->sk_write_queue
);
1602 if (arg
->csumoffset
>= 0)
1603 *((__sum16
*)skb_transport_header(nskb
) +
1604 arg
->csumoffset
) = csum_fold(csum_add(nskb
->csum
,
1606 nskb
->ip_summed
= CHECKSUM_NONE
;
1607 skb_set_queue_mapping(nskb
, skb_get_queue_mapping(skb
));
1608 ip_push_pending_frames(sk
, &fl4
);
1614 void __init
ip_init(void)
1619 #if defined(CONFIG_IP_MULTICAST)