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 /* Generate a checksum for an outgoing IP datagram. */
87 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 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
118 int ttl
= inet
->uc_ttl
;
121 ttl
= ip4_dst_hoplimit(dst
);
126 * Add an ip header to a skbuff and send it out.
129 int ip_build_and_send_pkt(struct sk_buff
*skb
, struct sock
*sk
,
130 __be32 saddr
, __be32 daddr
, struct ip_options_rcu
*opt
)
132 struct inet_sock
*inet
= inet_sk(sk
);
133 struct rtable
*rt
= skb_rtable(skb
);
136 /* Build the IP header. */
137 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->opt
.optlen
: 0));
138 skb_reset_network_header(skb
);
142 iph
->tos
= inet
->tos
;
143 if (ip_dont_fragment(sk
, &rt
->dst
))
144 iph
->frag_off
= htons(IP_DF
);
147 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
148 iph
->daddr
= (opt
&& opt
->opt
.srr
? opt
->opt
.faddr
: daddr
);
150 iph
->protocol
= sk
->sk_protocol
;
151 ip_select_ident(skb
, &rt
->dst
, sk
);
153 if (opt
&& opt
->opt
.optlen
) {
154 iph
->ihl
+= opt
->opt
.optlen
>>2;
155 ip_options_build(skb
, &opt
->opt
, daddr
, rt
, 0);
158 skb
->priority
= sk
->sk_priority
;
159 skb
->mark
= sk
->sk_mark
;
162 return ip_local_out(skb
);
164 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
166 static inline int ip_finish_output2(struct sk_buff
*skb
)
168 struct dst_entry
*dst
= skb_dst(skb
);
169 struct rtable
*rt
= (struct rtable
*)dst
;
170 struct net_device
*dev
= dst
->dev
;
171 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
172 struct neighbour
*neigh
;
175 if (rt
->rt_type
== RTN_MULTICAST
) {
176 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTMCAST
, skb
->len
);
177 } else if (rt
->rt_type
== RTN_BROADCAST
)
178 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUTBCAST
, skb
->len
);
180 /* Be paranoid, rather than too clever. */
181 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
182 struct sk_buff
*skb2
;
184 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
190 skb_set_owner_w(skb2
, skb
->sk
);
196 nexthop
= (__force u32
) rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
197 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
198 if (unlikely(!neigh
))
199 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
200 if (!IS_ERR(neigh
)) {
201 int res
= dst_neigh_output(dst
, neigh
, skb
);
203 rcu_read_unlock_bh();
206 rcu_read_unlock_bh();
208 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
214 static int ip_finish_output(struct sk_buff
*skb
)
216 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
217 /* Policy lookup after SNAT yielded a new policy */
218 if (skb_dst(skb
)->xfrm
!= NULL
) {
219 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
220 return dst_output(skb
);
223 if (skb
->len
> ip_skb_dst_mtu(skb
) && !skb_is_gso(skb
))
224 return ip_fragment(skb
, ip_finish_output2
);
226 return ip_finish_output2(skb
);
229 int ip_mc_output(struct sk_buff
*skb
)
231 struct sock
*sk
= skb
->sk
;
232 struct rtable
*rt
= skb_rtable(skb
);
233 struct net_device
*dev
= rt
->dst
.dev
;
236 * If the indicated interface is up and running, send the packet.
238 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
241 skb
->protocol
= htons(ETH_P_IP
);
244 * Multicasts are looped back for other local users
247 if (rt
->rt_flags
&RTCF_MULTICAST
) {
249 #ifdef CONFIG_IP_MROUTE
250 /* Small optimization: do not loopback not local frames,
251 which returned after forwarding; they will be dropped
252 by ip_mr_input in any case.
253 Note, that local frames are looped back to be delivered
256 This check is duplicated in ip_mr_input at the moment.
259 ((rt
->rt_flags
& RTCF_LOCAL
) ||
260 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
263 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
265 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
266 newskb
, NULL
, newskb
->dev
,
270 /* Multicasts with ttl 0 must not go beyond the host */
272 if (ip_hdr(skb
)->ttl
== 0) {
278 if (rt
->rt_flags
&RTCF_BROADCAST
) {
279 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
281 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, newskb
,
282 NULL
, newskb
->dev
, dev_loopback_xmit
);
285 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
,
286 skb
->dev
, ip_finish_output
,
287 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
290 int ip_output(struct sk_buff
*skb
)
292 struct net_device
*dev
= skb_dst(skb
)->dev
;
294 IP_UPD_PO_STATS(dev_net(dev
), IPSTATS_MIB_OUT
, skb
->len
);
297 skb
->protocol
= htons(ETH_P_IP
);
299 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
, skb
, NULL
, dev
,
301 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
305 * copy saddr and daddr, possibly using 64bit load/stores
307 * iph->saddr = fl4->saddr;
308 * iph->daddr = fl4->daddr;
310 static void ip_copy_addrs(struct iphdr
*iph
, const struct flowi4
*fl4
)
312 BUILD_BUG_ON(offsetof(typeof(*fl4
), daddr
) !=
313 offsetof(typeof(*fl4
), saddr
) + sizeof(fl4
->saddr
));
314 memcpy(&iph
->saddr
, &fl4
->saddr
,
315 sizeof(fl4
->saddr
) + sizeof(fl4
->daddr
));
318 int ip_queue_xmit(struct sk_buff
*skb
, struct flowi
*fl
)
320 struct sock
*sk
= skb
->sk
;
321 struct inet_sock
*inet
= inet_sk(sk
);
322 struct ip_options_rcu
*inet_opt
;
328 /* Skip all of this if the packet is already routed,
329 * f.e. by something like SCTP.
332 inet_opt
= rcu_dereference(inet
->inet_opt
);
334 rt
= skb_rtable(skb
);
338 /* Make sure we can route this packet. */
339 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
343 /* Use correct destination address if we have options. */
344 daddr
= inet
->inet_daddr
;
345 if (inet_opt
&& inet_opt
->opt
.srr
)
346 daddr
= inet_opt
->opt
.faddr
;
348 /* If this fails, retransmit mechanism of transport layer will
349 * keep trying until route appears or the connection times
352 rt
= ip_route_output_ports(sock_net(sk
), fl4
, sk
,
353 daddr
, inet
->inet_saddr
,
358 sk
->sk_bound_dev_if
);
361 sk_setup_caps(sk
, &rt
->dst
);
363 skb_dst_set_noref(skb
, &rt
->dst
);
366 if (inet_opt
&& inet_opt
->opt
.is_strictroute
&& rt
->rt_uses_gateway
)
369 /* OK, we know where to send it, allocate and build IP header. */
370 skb_push(skb
, sizeof(struct iphdr
) + (inet_opt
? inet_opt
->opt
.optlen
: 0));
371 skb_reset_network_header(skb
);
373 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
374 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->local_df
)
375 iph
->frag_off
= htons(IP_DF
);
378 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
379 iph
->protocol
= sk
->sk_protocol
;
380 ip_copy_addrs(iph
, fl4
);
382 /* Transport layer set skb->h.foo itself. */
384 if (inet_opt
&& inet_opt
->opt
.optlen
) {
385 iph
->ihl
+= inet_opt
->opt
.optlen
>> 2;
386 ip_options_build(skb
, &inet_opt
->opt
, inet
->inet_daddr
, rt
, 0);
389 ip_select_ident_more(skb
, &rt
->dst
, sk
,
390 (skb_shinfo(skb
)->gso_segs
?: 1) - 1);
392 skb
->priority
= sk
->sk_priority
;
393 skb
->mark
= sk
->sk_mark
;
395 res
= ip_local_out(skb
);
401 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
403 return -EHOSTUNREACH
;
405 EXPORT_SYMBOL(ip_queue_xmit
);
408 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
410 to
->pkt_type
= from
->pkt_type
;
411 to
->priority
= from
->priority
;
412 to
->protocol
= from
->protocol
;
414 skb_dst_copy(to
, from
);
416 to
->mark
= from
->mark
;
418 /* Copy the flags to each fragment. */
419 IPCB(to
)->flags
= IPCB(from
)->flags
;
421 #ifdef CONFIG_NET_SCHED
422 to
->tc_index
= from
->tc_index
;
425 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
426 to
->nf_trace
= from
->nf_trace
;
428 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
429 to
->ipvs_property
= from
->ipvs_property
;
431 skb_copy_secmark(to
, from
);
435 * This IP datagram is too large to be sent in one piece. Break it up into
436 * smaller pieces (each of size equal to IP header plus
437 * a block of the data of the original IP data part) that will yet fit in a
438 * single device frame, and queue such a frame for sending.
441 int ip_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
445 struct net_device
*dev
;
446 struct sk_buff
*skb2
;
447 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
449 __be16 not_last_frag
;
450 struct rtable
*rt
= skb_rtable(skb
);
452 bool forwarding
= IPCB(skb
)->flags
& IPSKB_FORWARDED
;
457 * Point into the IP datagram header.
462 mtu
= ip_dst_mtu_maybe_forward(&rt
->dst
, forwarding
);
463 if (unlikely(((iph
->frag_off
& htons(IP_DF
)) && !skb
->local_df
) ||
464 (IPCB(skb
)->frag_max_size
&&
465 IPCB(skb
)->frag_max_size
> mtu
))) {
466 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
467 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
474 * Setup starting values.
478 mtu
= mtu
- hlen
; /* Size of data space */
479 #ifdef CONFIG_BRIDGE_NETFILTER
481 mtu
-= nf_bridge_mtu_reduction(skb
);
483 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
485 /* When frag_list is given, use it. First, check its validity:
486 * some transformers could create wrong frag_list or break existing
487 * one, it is not prohibited. In this case fall back to copying.
489 * LATER: this step can be merged to real generation of fragments,
490 * we can switch to copy when see the first bad fragment.
492 if (skb_has_frag_list(skb
)) {
493 struct sk_buff
*frag
, *frag2
;
494 int first_len
= skb_pagelen(skb
);
496 if (first_len
- hlen
> mtu
||
497 ((first_len
- hlen
) & 7) ||
498 ip_is_fragment(iph
) ||
502 skb_walk_frags(skb
, frag
) {
503 /* Correct geometry. */
504 if (frag
->len
> mtu
||
505 ((frag
->len
& 7) && frag
->next
) ||
506 skb_headroom(frag
) < hlen
)
507 goto slow_path_clean
;
509 /* Partially cloned skb? */
510 if (skb_shared(frag
))
511 goto slow_path_clean
;
516 frag
->destructor
= sock_wfree
;
518 skb
->truesize
-= frag
->truesize
;
521 /* Everything is OK. Generate! */
525 frag
= skb_shinfo(skb
)->frag_list
;
526 skb_frag_list_init(skb
);
527 skb
->data_len
= first_len
- skb_headlen(skb
);
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
);
581 skb_walk_frags(skb
, frag2
) {
585 frag2
->destructor
= NULL
;
586 skb
->truesize
+= frag2
->truesize
;
591 /* for offloaded checksums cleanup checksum before fragmentation */
592 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) && skb_checksum_help(skb
))
596 left
= skb
->len
- hlen
; /* Space per frame */
597 ptr
= hlen
; /* Where to start from */
599 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
600 * we need to make room for the encapsulating header
602 ll_rs
= LL_RESERVED_SPACE_EXTRA(rt
->dst
.dev
, nf_bridge_pad(skb
));
605 * Fragment the datagram.
608 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
609 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
612 * Keep copying data until we run out.
617 /* IF: it doesn't fit, use 'mtu' - the data space left */
620 /* IF: we are not sending up to and including the packet end
621 then align the next start on an eight byte boundary */
629 if ((skb2
= alloc_skb(len
+hlen
+ll_rs
, GFP_ATOMIC
)) == NULL
) {
630 NETDEBUG(KERN_INFO
"IP: frag: no memory for new fragment!\n");
636 * Set up data on packet
639 ip_copy_metadata(skb2
, skb
);
640 skb_reserve(skb2
, ll_rs
);
641 skb_put(skb2
, len
+ hlen
);
642 skb_reset_network_header(skb2
);
643 skb2
->transport_header
= skb2
->network_header
+ hlen
;
646 * Charge the memory for the fragment to any owner
651 skb_set_owner_w(skb2
, skb
->sk
);
654 * Copy the packet header into the new buffer.
657 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
660 * Copy a block of the IP datagram.
662 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
667 * Fill in the new header fields.
670 iph
->frag_off
= htons((offset
>> 3));
672 /* ANK: dirty, but effective trick. Upgrade options only if
673 * the segment to be fragmented was THE FIRST (otherwise,
674 * options are already fixed) and make it ONCE
675 * on the initial skb, so that all the following fragments
676 * will inherit fixed options.
679 ip_options_fragment(skb
);
682 * Added AC : If we are fragmenting a fragment that's not the
683 * last fragment then keep MF on each bit
685 if (left
> 0 || not_last_frag
)
686 iph
->frag_off
|= htons(IP_MF
);
691 * Put this fragment into the sending queue.
693 iph
->tot_len
= htons(len
+ hlen
);
701 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGCREATES
);
704 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGOKS
);
709 IP_INC_STATS(dev_net(dev
), IPSTATS_MIB_FRAGFAILS
);
712 EXPORT_SYMBOL(ip_fragment
);
715 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
717 struct iovec
*iov
= from
;
719 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
720 if (memcpy_fromiovecend(to
, iov
, offset
, len
) < 0)
724 if (csum_partial_copy_fromiovecend(to
, iov
, offset
, len
, &csum
) < 0)
726 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
730 EXPORT_SYMBOL(ip_generic_getfrag
);
733 csum_page(struct page
*page
, int offset
, int copy
)
738 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
743 static inline int ip_ufo_append_data(struct sock
*sk
,
744 struct sk_buff_head
*queue
,
745 int getfrag(void *from
, char *to
, int offset
, int len
,
746 int odd
, struct sk_buff
*skb
),
747 void *from
, int length
, int hh_len
, int fragheaderlen
,
748 int transhdrlen
, int maxfraglen
, unsigned int flags
)
753 /* There is support for UDP fragmentation offload by network
754 * device, so create one single skb packet containing complete
757 if ((skb
= skb_peek_tail(queue
)) == NULL
) {
758 skb
= sock_alloc_send_skb(sk
,
759 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
760 (flags
& MSG_DONTWAIT
), &err
);
765 /* reserve space for Hardware header */
766 skb_reserve(skb
, hh_len
);
768 /* create space for UDP/IP header */
769 skb_put(skb
, fragheaderlen
+ transhdrlen
);
771 /* initialize network header pointer */
772 skb_reset_network_header(skb
);
774 /* initialize protocol header pointer */
775 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
780 __skb_queue_tail(queue
, skb
);
781 } else if (skb_is_gso(skb
)) {
785 skb
->ip_summed
= CHECKSUM_PARTIAL
;
786 /* specify the length of each IP datagram fragment */
787 skb_shinfo(skb
)->gso_size
= maxfraglen
- fragheaderlen
;
788 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
791 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
792 (length
- transhdrlen
));
795 static int __ip_append_data(struct sock
*sk
,
797 struct sk_buff_head
*queue
,
798 struct inet_cork
*cork
,
799 struct page_frag
*pfrag
,
800 int getfrag(void *from
, char *to
, int offset
,
801 int len
, int odd
, struct sk_buff
*skb
),
802 void *from
, int length
, int transhdrlen
,
805 struct inet_sock
*inet
= inet_sk(sk
);
808 struct ip_options
*opt
= cork
->opt
;
815 unsigned int maxfraglen
, fragheaderlen
, maxnonfragsize
;
816 int csummode
= CHECKSUM_NONE
;
817 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
819 skb
= skb_peek_tail(queue
);
821 exthdrlen
= !skb
? rt
->dst
.header_len
: 0;
822 mtu
= cork
->fragsize
;
824 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
826 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
827 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
828 maxnonfragsize
= (inet
->pmtudisc
>= IP_PMTUDISC_DO
) ?
831 if (cork
->length
+ length
> maxnonfragsize
- fragheaderlen
) {
832 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
833 mtu
- (opt
? opt
->optlen
: 0));
838 * transhdrlen > 0 means that this is the first fragment and we wish
839 * it won't be fragmented in the future.
842 length
+ fragheaderlen
<= mtu
&&
843 rt
->dst
.dev
->features
& NETIF_F_V4_CSUM
&&
845 csummode
= CHECKSUM_PARTIAL
;
847 cork
->length
+= length
;
848 if (((length
> mtu
) || (skb
&& skb_is_gso(skb
))) &&
849 (sk
->sk_protocol
== IPPROTO_UDP
) &&
850 (rt
->dst
.dev
->features
& NETIF_F_UFO
) && !rt
->dst
.header_len
) {
851 err
= ip_ufo_append_data(sk
, queue
, getfrag
, from
, length
,
852 hh_len
, fragheaderlen
, transhdrlen
,
859 /* So, what's going on in the loop below?
861 * We use calculated fragment length to generate chained skb,
862 * each of segments is IP fragment ready for sending to network after
863 * adding appropriate IP header.
870 /* Check if the remaining data fits into current packet. */
871 copy
= mtu
- skb
->len
;
873 copy
= maxfraglen
- skb
->len
;
876 unsigned int datalen
;
877 unsigned int fraglen
;
878 unsigned int fraggap
;
879 unsigned int alloclen
;
880 struct sk_buff
*skb_prev
;
884 fraggap
= skb_prev
->len
- maxfraglen
;
889 * If remaining data exceeds the mtu,
890 * we know we need more fragment(s).
892 datalen
= length
+ fraggap
;
893 if (datalen
> mtu
- fragheaderlen
)
894 datalen
= maxfraglen
- fragheaderlen
;
895 fraglen
= datalen
+ fragheaderlen
;
897 if ((flags
& MSG_MORE
) &&
898 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
903 alloclen
+= exthdrlen
;
905 /* The last fragment gets additional space at tail.
906 * Note, with MSG_MORE we overallocate on fragments,
907 * because we have no idea what fragment will be
910 if (datalen
== length
+ fraggap
)
911 alloclen
+= rt
->dst
.trailer_len
;
914 skb
= sock_alloc_send_skb(sk
,
915 alloclen
+ hh_len
+ 15,
916 (flags
& MSG_DONTWAIT
), &err
);
919 if (atomic_read(&sk
->sk_wmem_alloc
) <=
921 skb
= sock_wmalloc(sk
,
922 alloclen
+ hh_len
+ 15, 1,
924 if (unlikely(skb
== NULL
))
927 /* only the initial fragment is
935 * Fill in the control structures
937 skb
->ip_summed
= csummode
;
939 skb_reserve(skb
, hh_len
);
940 skb_shinfo(skb
)->tx_flags
= cork
->tx_flags
;
943 * Find where to start putting bytes.
945 data
= skb_put(skb
, fraglen
+ exthdrlen
);
946 skb_set_network_header(skb
, exthdrlen
);
947 skb
->transport_header
= (skb
->network_header
+
949 data
+= fragheaderlen
+ exthdrlen
;
952 skb
->csum
= skb_copy_and_csum_bits(
953 skb_prev
, maxfraglen
,
954 data
+ transhdrlen
, fraggap
, 0);
955 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
958 pskb_trim_unique(skb_prev
, maxfraglen
);
961 copy
= datalen
- transhdrlen
- fraggap
;
962 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
969 length
-= datalen
- fraggap
;
972 csummode
= CHECKSUM_NONE
;
975 * Put the packet on the pending queue.
977 __skb_queue_tail(queue
, skb
);
984 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
)) {
988 if (getfrag(from
, skb_put(skb
, copy
),
989 offset
, copy
, off
, skb
) < 0) {
990 __skb_trim(skb
, off
);
995 int i
= skb_shinfo(skb
)->nr_frags
;
998 if (!sk_page_frag_refill(sk
, pfrag
))
1001 if (!skb_can_coalesce(skb
, i
, pfrag
->page
,
1004 if (i
== MAX_SKB_FRAGS
)
1007 __skb_fill_page_desc(skb
, i
, pfrag
->page
,
1009 skb_shinfo(skb
)->nr_frags
= ++i
;
1010 get_page(pfrag
->page
);
1012 copy
= min_t(int, copy
, pfrag
->size
- pfrag
->offset
);
1014 page_address(pfrag
->page
) + pfrag
->offset
,
1015 offset
, copy
, skb
->len
, skb
) < 0)
1018 pfrag
->offset
+= copy
;
1019 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], copy
);
1021 skb
->data_len
+= copy
;
1022 skb
->truesize
+= copy
;
1023 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1034 cork
->length
-= length
;
1035 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1039 static int ip_setup_cork(struct sock
*sk
, struct inet_cork
*cork
,
1040 struct ipcm_cookie
*ipc
, struct rtable
**rtp
)
1042 struct ip_options_rcu
*opt
;
1046 * setup for corking.
1050 if (cork
->opt
== NULL
) {
1051 cork
->opt
= kmalloc(sizeof(struct ip_options
) + 40,
1053 if (unlikely(cork
->opt
== NULL
))
1056 memcpy(cork
->opt
, &opt
->opt
, sizeof(struct ip_options
) + opt
->opt
.optlen
);
1057 cork
->flags
|= IPCORK_OPT
;
1058 cork
->addr
= ipc
->addr
;
1064 * We steal reference to this route, caller should not release it
1067 cork
->fragsize
= ip_sk_use_pmtu(sk
) ?
1068 dst_mtu(&rt
->dst
) : rt
->dst
.dev
->mtu
;
1069 cork
->dst
= &rt
->dst
;
1071 cork
->ttl
= ipc
->ttl
;
1072 cork
->tos
= ipc
->tos
;
1073 cork
->priority
= ipc
->priority
;
1074 cork
->tx_flags
= ipc
->tx_flags
;
1080 * ip_append_data() and ip_append_page() can make one large IP datagram
1081 * from many pieces of data. Each pieces will be holded on the socket
1082 * until ip_push_pending_frames() is called. Each piece can be a page
1085 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1086 * this interface potentially.
1088 * LATER: length must be adjusted by pad at tail, when it is required.
1090 int ip_append_data(struct sock
*sk
, struct flowi4
*fl4
,
1091 int getfrag(void *from
, char *to
, int offset
, int len
,
1092 int odd
, struct sk_buff
*skb
),
1093 void *from
, int length
, int transhdrlen
,
1094 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1097 struct inet_sock
*inet
= inet_sk(sk
);
1100 if (flags
&MSG_PROBE
)
1103 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1104 err
= ip_setup_cork(sk
, &inet
->cork
.base
, ipc
, rtp
);
1111 return __ip_append_data(sk
, fl4
, &sk
->sk_write_queue
, &inet
->cork
.base
,
1112 sk_page_frag(sk
), getfrag
,
1113 from
, length
, transhdrlen
, flags
);
1116 ssize_t
ip_append_page(struct sock
*sk
, struct flowi4
*fl4
, struct page
*page
,
1117 int offset
, size_t size
, int flags
)
1119 struct inet_sock
*inet
= inet_sk(sk
);
1120 struct sk_buff
*skb
;
1122 struct ip_options
*opt
= NULL
;
1123 struct inet_cork
*cork
;
1128 unsigned int maxfraglen
, fragheaderlen
, fraggap
, maxnonfragsize
;
1133 if (flags
&MSG_PROBE
)
1136 if (skb_queue_empty(&sk
->sk_write_queue
))
1139 cork
= &inet
->cork
.base
;
1140 rt
= (struct rtable
*)cork
->dst
;
1141 if (cork
->flags
& IPCORK_OPT
)
1144 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1147 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1148 mtu
= cork
->fragsize
;
1150 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1151 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1152 maxnonfragsize
= (inet
->pmtudisc
>= IP_PMTUDISC_DO
) ?
1155 if (cork
->length
+ size
> maxnonfragsize
- fragheaderlen
) {
1156 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
1157 mtu
- (opt
? opt
->optlen
: 0));
1161 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1164 cork
->length
+= size
;
1165 if ((size
+ skb
->len
> mtu
) &&
1166 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1167 (rt
->dst
.dev
->features
& NETIF_F_UFO
)) {
1168 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
;
1169 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1176 if (skb_is_gso(skb
))
1180 /* Check if the remaining data fits into current packet. */
1181 len
= mtu
- skb
->len
;
1183 len
= maxfraglen
- skb
->len
;
1186 struct sk_buff
*skb_prev
;
1190 fraggap
= skb_prev
->len
- maxfraglen
;
1192 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1193 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1194 if (unlikely(!skb
)) {
1200 * Fill in the control structures
1202 skb
->ip_summed
= CHECKSUM_NONE
;
1204 skb_reserve(skb
, hh_len
);
1207 * Find where to start putting bytes.
1209 skb_put(skb
, fragheaderlen
+ fraggap
);
1210 skb_reset_network_header(skb
);
1211 skb
->transport_header
= (skb
->network_header
+
1214 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1216 skb_transport_header(skb
),
1218 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1220 pskb_trim_unique(skb_prev
, maxfraglen
);
1224 * Put the packet on the pending queue.
1226 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1230 i
= skb_shinfo(skb
)->nr_frags
;
1233 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
1234 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
-1], len
);
1235 } else if (i
< MAX_SKB_FRAGS
) {
1237 skb_fill_page_desc(skb
, i
, page
, offset
, len
);
1243 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1245 csum
= csum_page(page
, offset
, len
);
1246 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1250 skb
->data_len
+= len
;
1251 skb
->truesize
+= len
;
1252 atomic_add(len
, &sk
->sk_wmem_alloc
);
1259 cork
->length
-= size
;
1260 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1264 static void ip_cork_release(struct inet_cork
*cork
)
1266 cork
->flags
&= ~IPCORK_OPT
;
1269 dst_release(cork
->dst
);
1274 * Combined all pending IP fragments on the socket as one IP datagram
1275 * and push them out.
1277 struct sk_buff
*__ip_make_skb(struct sock
*sk
,
1279 struct sk_buff_head
*queue
,
1280 struct inet_cork
*cork
)
1282 struct sk_buff
*skb
, *tmp_skb
;
1283 struct sk_buff
**tail_skb
;
1284 struct inet_sock
*inet
= inet_sk(sk
);
1285 struct net
*net
= sock_net(sk
);
1286 struct ip_options
*opt
= NULL
;
1287 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
1292 if ((skb
= __skb_dequeue(queue
)) == NULL
)
1294 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1296 /* move skb->data to ip header from ext header */
1297 if (skb
->data
< skb_network_header(skb
))
1298 __skb_pull(skb
, skb_network_offset(skb
));
1299 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1300 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1301 *tail_skb
= tmp_skb
;
1302 tail_skb
= &(tmp_skb
->next
);
1303 skb
->len
+= tmp_skb
->len
;
1304 skb
->data_len
+= tmp_skb
->len
;
1305 skb
->truesize
+= tmp_skb
->truesize
;
1306 tmp_skb
->destructor
= NULL
;
1310 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1311 * to fragment the frame generated here. No matter, what transforms
1312 * how transforms change size of the packet, it will come out.
1314 if (inet
->pmtudisc
< IP_PMTUDISC_DO
)
1317 /* DF bit is set when we want to see DF on outgoing frames.
1318 * If local_df is set too, we still allow to fragment this frame
1320 if (inet
->pmtudisc
== IP_PMTUDISC_DO
||
1321 inet
->pmtudisc
== IP_PMTUDISC_PROBE
||
1322 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1323 ip_dont_fragment(sk
, &rt
->dst
)))
1326 if (cork
->flags
& IPCORK_OPT
)
1331 else if (rt
->rt_type
== RTN_MULTICAST
)
1334 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1339 iph
->tos
= (cork
->tos
!= -1) ? cork
->tos
: inet
->tos
;
1342 iph
->protocol
= sk
->sk_protocol
;
1343 ip_copy_addrs(iph
, fl4
);
1344 ip_select_ident(skb
, &rt
->dst
, sk
);
1347 iph
->ihl
+= opt
->optlen
>>2;
1348 ip_options_build(skb
, opt
, cork
->addr
, rt
, 0);
1351 skb
->priority
= (cork
->tos
!= -1) ? cork
->priority
: sk
->sk_priority
;
1352 skb
->mark
= sk
->sk_mark
;
1354 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1358 skb_dst_set(skb
, &rt
->dst
);
1360 if (iph
->protocol
== IPPROTO_ICMP
)
1361 icmp_out_count(net
, ((struct icmphdr
*)
1362 skb_transport_header(skb
))->type
);
1364 ip_cork_release(cork
);
1369 int ip_send_skb(struct net
*net
, struct sk_buff
*skb
)
1373 err
= ip_local_out(skb
);
1376 err
= net_xmit_errno(err
);
1378 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1384 int ip_push_pending_frames(struct sock
*sk
, struct flowi4
*fl4
)
1386 struct sk_buff
*skb
;
1388 skb
= ip_finish_skb(sk
, fl4
);
1392 /* Netfilter gets whole the not fragmented skb. */
1393 return ip_send_skb(sock_net(sk
), skb
);
1397 * Throw away all pending data on the socket.
1399 static void __ip_flush_pending_frames(struct sock
*sk
,
1400 struct sk_buff_head
*queue
,
1401 struct inet_cork
*cork
)
1403 struct sk_buff
*skb
;
1405 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
)
1408 ip_cork_release(cork
);
1411 void ip_flush_pending_frames(struct sock
*sk
)
1413 __ip_flush_pending_frames(sk
, &sk
->sk_write_queue
, &inet_sk(sk
)->cork
.base
);
1416 struct sk_buff
*ip_make_skb(struct sock
*sk
,
1418 int getfrag(void *from
, char *to
, int offset
,
1419 int len
, int odd
, struct sk_buff
*skb
),
1420 void *from
, int length
, int transhdrlen
,
1421 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1424 struct inet_cork cork
;
1425 struct sk_buff_head queue
;
1428 if (flags
& MSG_PROBE
)
1431 __skb_queue_head_init(&queue
);
1436 err
= ip_setup_cork(sk
, &cork
, ipc
, rtp
);
1438 return ERR_PTR(err
);
1440 err
= __ip_append_data(sk
, fl4
, &queue
, &cork
,
1441 ¤t
->task_frag
, getfrag
,
1442 from
, length
, transhdrlen
, flags
);
1444 __ip_flush_pending_frames(sk
, &queue
, &cork
);
1445 return ERR_PTR(err
);
1448 return __ip_make_skb(sk
, fl4
, &queue
, &cork
);
1452 * Fetch data from kernel space and fill in checksum if needed.
1454 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1455 int len
, int odd
, struct sk_buff
*skb
)
1459 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1460 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1465 * Generic function to send a packet as reply to another packet.
1466 * Used to send some TCP resets/acks so far.
1468 * Use a fake percpu inet socket to avoid false sharing and contention.
1470 static DEFINE_PER_CPU(struct inet_sock
, unicast_sock
) = {
1473 .skc_refcnt
= ATOMIC_INIT(1),
1475 .sk_wmem_alloc
= ATOMIC_INIT(1),
1476 .sk_allocation
= GFP_ATOMIC
,
1477 .sk_flags
= (1UL << SOCK_USE_WRITE_QUEUE
),
1479 .pmtudisc
= IP_PMTUDISC_WANT
,
1483 void ip_send_unicast_reply(struct net
*net
, struct sk_buff
*skb
, __be32 daddr
,
1484 __be32 saddr
, const struct ip_reply_arg
*arg
,
1487 struct ip_options_data replyopts
;
1488 struct ipcm_cookie ipc
;
1490 struct rtable
*rt
= skb_rtable(skb
);
1491 struct sk_buff
*nskb
;
1493 struct inet_sock
*inet
;
1495 if (ip_options_echo(&replyopts
.opt
.opt
, skb
))
1504 if (replyopts
.opt
.opt
.optlen
) {
1505 ipc
.opt
= &replyopts
.opt
;
1507 if (replyopts
.opt
.opt
.srr
)
1508 daddr
= replyopts
.opt
.opt
.faddr
;
1511 flowi4_init_output(&fl4
, arg
->bound_dev_if
, 0,
1513 RT_SCOPE_UNIVERSE
, ip_hdr(skb
)->protocol
,
1514 ip_reply_arg_flowi_flags(arg
),
1516 tcp_hdr(skb
)->source
, tcp_hdr(skb
)->dest
);
1517 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
1518 rt
= ip_route_output_key(net
, &fl4
);
1522 inet
= &get_cpu_var(unicast_sock
);
1524 inet
->tos
= arg
->tos
;
1526 sk
->sk_priority
= skb
->priority
;
1527 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1528 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1529 sock_net_set(sk
, net
);
1530 __skb_queue_head_init(&sk
->sk_write_queue
);
1531 sk
->sk_sndbuf
= sysctl_wmem_default
;
1532 ip_append_data(sk
, &fl4
, ip_reply_glue_bits
, arg
->iov
->iov_base
, len
, 0,
1533 &ipc
, &rt
, MSG_DONTWAIT
);
1534 nskb
= skb_peek(&sk
->sk_write_queue
);
1536 if (arg
->csumoffset
>= 0)
1537 *((__sum16
*)skb_transport_header(nskb
) +
1538 arg
->csumoffset
) = csum_fold(csum_add(nskb
->csum
,
1540 nskb
->ip_summed
= CHECKSUM_NONE
;
1542 skb_set_queue_mapping(nskb
, skb_get_queue_mapping(skb
));
1543 ip_push_pending_frames(sk
, &fl4
);
1546 put_cpu_var(unicast_sock
);
1551 void __init
ip_init(void)
1556 #if defined(CONFIG_IP_MULTICAST)