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 <linux/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 <net/lwtunnel.h>
77 #include <linux/bpf-cgroup.h>
78 #include <linux/igmp.h>
79 #include <linux/netfilter_ipv4.h>
80 #include <linux/netfilter_bridge.h>
81 #include <linux/netlink.h>
82 #include <linux/tcp.h>
85 ip_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
87 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*));
89 /* Generate a checksum for an outgoing IP datagram. */
90 void ip_send_check(struct iphdr
*iph
)
93 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
95 EXPORT_SYMBOL(ip_send_check
);
97 int __ip_local_out(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
99 struct iphdr
*iph
= ip_hdr(skb
);
101 iph
->tot_len
= htons(skb
->len
);
104 /* if egress device is enslaved to an L3 master device pass the
105 * skb to its handler for processing
107 skb
= l3mdev_ip_out(sk
, skb
);
111 skb
->protocol
= htons(ETH_P_IP
);
113 return nf_hook(NFPROTO_IPV4
, NF_INET_LOCAL_OUT
,
114 net
, sk
, skb
, NULL
, skb_dst(skb
)->dev
,
118 int ip_local_out(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
122 err
= __ip_local_out(net
, sk
, skb
);
123 if (likely(err
== 1))
124 err
= dst_output(net
, sk
, skb
);
128 EXPORT_SYMBOL_GPL(ip_local_out
);
130 static inline int ip_select_ttl(struct inet_sock
*inet
, struct dst_entry
*dst
)
132 int ttl
= inet
->uc_ttl
;
135 ttl
= ip4_dst_hoplimit(dst
);
140 * Add an ip header to a skbuff and send it out.
143 int ip_build_and_send_pkt(struct sk_buff
*skb
, const struct sock
*sk
,
144 __be32 saddr
, __be32 daddr
, struct ip_options_rcu
*opt
)
146 struct inet_sock
*inet
= inet_sk(sk
);
147 struct rtable
*rt
= skb_rtable(skb
);
148 struct net
*net
= sock_net(sk
);
151 /* Build the IP header. */
152 skb_push(skb
, sizeof(struct iphdr
) + (opt
? opt
->opt
.optlen
: 0));
153 skb_reset_network_header(skb
);
157 iph
->tos
= inet
->tos
;
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 if (ip_dont_fragment(sk
, &rt
->dst
)) {
163 iph
->frag_off
= htons(IP_DF
);
167 __ip_select_ident(net
, iph
, 1);
170 if (opt
&& opt
->opt
.optlen
) {
171 iph
->ihl
+= opt
->opt
.optlen
>>2;
172 ip_options_build(skb
, &opt
->opt
, daddr
, rt
, 0);
175 skb
->priority
= sk
->sk_priority
;
177 skb
->mark
= sk
->sk_mark
;
180 return ip_local_out(net
, skb
->sk
, skb
);
182 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt
);
184 static int ip_finish_output2(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
186 struct dst_entry
*dst
= skb_dst(skb
);
187 struct rtable
*rt
= (struct rtable
*)dst
;
188 struct net_device
*dev
= dst
->dev
;
189 unsigned int hh_len
= LL_RESERVED_SPACE(dev
);
190 struct neighbour
*neigh
;
193 if (rt
->rt_type
== RTN_MULTICAST
) {
194 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUTMCAST
, skb
->len
);
195 } else if (rt
->rt_type
== RTN_BROADCAST
)
196 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUTBCAST
, skb
->len
);
198 /* Be paranoid, rather than too clever. */
199 if (unlikely(skb_headroom(skb
) < hh_len
&& dev
->header_ops
)) {
200 struct sk_buff
*skb2
;
202 skb2
= skb_realloc_headroom(skb
, LL_RESERVED_SPACE(dev
));
208 skb_set_owner_w(skb2
, skb
->sk
);
213 if (lwtunnel_xmit_redirect(dst
->lwtstate
)) {
214 int res
= lwtunnel_xmit(skb
);
216 if (res
< 0 || res
== LWTUNNEL_XMIT_DONE
)
221 nexthop
= (__force u32
) rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
222 neigh
= __ipv4_neigh_lookup_noref(dev
, nexthop
);
223 if (unlikely(!neigh
))
224 neigh
= __neigh_create(&arp_tbl
, &nexthop
, dev
, false);
225 if (!IS_ERR(neigh
)) {
228 sock_confirm_neigh(skb
, neigh
);
229 res
= neigh_output(neigh
, skb
);
231 rcu_read_unlock_bh();
234 rcu_read_unlock_bh();
236 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
242 static int ip_finish_output_gso(struct net
*net
, struct sock
*sk
,
243 struct sk_buff
*skb
, unsigned int mtu
)
245 netdev_features_t features
;
246 struct sk_buff
*segs
;
249 /* common case: seglen is <= mtu
251 if (skb_gso_validate_mtu(skb
, mtu
))
252 return ip_finish_output2(net
, sk
, skb
);
254 /* Slowpath - GSO segment length exceeds the egress MTU.
256 * This can happen in several cases:
257 * - Forwarding of a TCP GRO skb, when DF flag is not set.
258 * - Forwarding of an skb that arrived on a virtualization interface
259 * (virtio-net/vhost/tap) with TSO/GSO size set by other network
261 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
262 * interface with a smaller MTU.
263 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is
264 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an
267 features
= netif_skb_features(skb
);
268 BUILD_BUG_ON(sizeof(*IPCB(skb
)) > SKB_SGO_CB_OFFSET
);
269 segs
= skb_gso_segment(skb
, features
& ~NETIF_F_GSO_MASK
);
270 if (IS_ERR_OR_NULL(segs
)) {
278 struct sk_buff
*nskb
= segs
->next
;
282 err
= ip_fragment(net
, sk
, segs
, mtu
, ip_finish_output2
);
292 static int ip_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
297 ret
= BPF_CGROUP_RUN_PROG_INET_EGRESS(sk
, skb
);
303 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
304 /* Policy lookup after SNAT yielded a new policy */
305 if (skb_dst(skb
)->xfrm
) {
306 IPCB(skb
)->flags
|= IPSKB_REROUTED
;
307 return dst_output(net
, sk
, skb
);
310 mtu
= ip_skb_dst_mtu(sk
, skb
);
312 return ip_finish_output_gso(net
, sk
, skb
, mtu
);
314 if (skb
->len
> mtu
|| (IPCB(skb
)->flags
& IPSKB_FRAG_PMTU
))
315 return ip_fragment(net
, sk
, skb
, mtu
, ip_finish_output2
);
317 return ip_finish_output2(net
, sk
, skb
);
320 static int ip_mc_finish_output(struct net
*net
, struct sock
*sk
,
325 ret
= BPF_CGROUP_RUN_PROG_INET_EGRESS(sk
, skb
);
331 return dev_loopback_xmit(net
, sk
, skb
);
334 int ip_mc_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
336 struct rtable
*rt
= skb_rtable(skb
);
337 struct net_device
*dev
= rt
->dst
.dev
;
340 * If the indicated interface is up and running, send the packet.
342 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
345 skb
->protocol
= htons(ETH_P_IP
);
348 * Multicasts are looped back for other local users
351 if (rt
->rt_flags
&RTCF_MULTICAST
) {
353 #ifdef CONFIG_IP_MROUTE
354 /* Small optimization: do not loopback not local frames,
355 which returned after forwarding; they will be dropped
356 by ip_mr_input in any case.
357 Note, that local frames are looped back to be delivered
360 This check is duplicated in ip_mr_input at the moment.
363 ((rt
->rt_flags
& RTCF_LOCAL
) ||
364 !(IPCB(skb
)->flags
& IPSKB_FORWARDED
))
367 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
369 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
370 net
, sk
, newskb
, NULL
, newskb
->dev
,
371 ip_mc_finish_output
);
374 /* Multicasts with ttl 0 must not go beyond the host */
376 if (ip_hdr(skb
)->ttl
== 0) {
382 if (rt
->rt_flags
&RTCF_BROADCAST
) {
383 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
385 NF_HOOK(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
386 net
, sk
, newskb
, NULL
, newskb
->dev
,
387 ip_mc_finish_output
);
390 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
391 net
, sk
, skb
, NULL
, skb
->dev
,
393 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
396 int ip_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
398 struct net_device
*dev
= skb_dst(skb
)->dev
;
400 IP_UPD_PO_STATS(net
, IPSTATS_MIB_OUT
, skb
->len
);
403 skb
->protocol
= htons(ETH_P_IP
);
405 return NF_HOOK_COND(NFPROTO_IPV4
, NF_INET_POST_ROUTING
,
406 net
, sk
, skb
, NULL
, dev
,
408 !(IPCB(skb
)->flags
& IPSKB_REROUTED
));
412 * copy saddr and daddr, possibly using 64bit load/stores
414 * iph->saddr = fl4->saddr;
415 * iph->daddr = fl4->daddr;
417 static void ip_copy_addrs(struct iphdr
*iph
, const struct flowi4
*fl4
)
419 BUILD_BUG_ON(offsetof(typeof(*fl4
), daddr
) !=
420 offsetof(typeof(*fl4
), saddr
) + sizeof(fl4
->saddr
));
421 memcpy(&iph
->saddr
, &fl4
->saddr
,
422 sizeof(fl4
->saddr
) + sizeof(fl4
->daddr
));
425 /* Note: skb->sk can be different from sk, in case of tunnels */
426 int ip_queue_xmit(struct sock
*sk
, struct sk_buff
*skb
, struct flowi
*fl
)
428 struct inet_sock
*inet
= inet_sk(sk
);
429 struct net
*net
= sock_net(sk
);
430 struct ip_options_rcu
*inet_opt
;
436 /* Skip all of this if the packet is already routed,
437 * f.e. by something like SCTP.
440 inet_opt
= rcu_dereference(inet
->inet_opt
);
442 rt
= skb_rtable(skb
);
446 /* Make sure we can route this packet. */
447 rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
451 /* Use correct destination address if we have options. */
452 daddr
= inet
->inet_daddr
;
453 if (inet_opt
&& inet_opt
->opt
.srr
)
454 daddr
= inet_opt
->opt
.faddr
;
456 /* If this fails, retransmit mechanism of transport layer will
457 * keep trying until route appears or the connection times
460 rt
= ip_route_output_ports(net
, fl4
, sk
,
461 daddr
, inet
->inet_saddr
,
466 sk
->sk_bound_dev_if
);
469 sk_setup_caps(sk
, &rt
->dst
);
471 skb_dst_set_noref(skb
, &rt
->dst
);
474 if (inet_opt
&& inet_opt
->opt
.is_strictroute
&& rt
->rt_uses_gateway
)
477 /* OK, we know where to send it, allocate and build IP header. */
478 skb_push(skb
, sizeof(struct iphdr
) + (inet_opt
? inet_opt
->opt
.optlen
: 0));
479 skb_reset_network_header(skb
);
481 *((__be16
*)iph
) = htons((4 << 12) | (5 << 8) | (inet
->tos
& 0xff));
482 if (ip_dont_fragment(sk
, &rt
->dst
) && !skb
->ignore_df
)
483 iph
->frag_off
= htons(IP_DF
);
486 iph
->ttl
= ip_select_ttl(inet
, &rt
->dst
);
487 iph
->protocol
= sk
->sk_protocol
;
488 ip_copy_addrs(iph
, fl4
);
490 /* Transport layer set skb->h.foo itself. */
492 if (inet_opt
&& inet_opt
->opt
.optlen
) {
493 iph
->ihl
+= inet_opt
->opt
.optlen
>> 2;
494 ip_options_build(skb
, &inet_opt
->opt
, inet
->inet_daddr
, rt
, 0);
497 ip_select_ident_segs(net
, skb
, sk
,
498 skb_shinfo(skb
)->gso_segs
?: 1);
500 /* TODO : should we use skb->sk here instead of sk ? */
501 skb
->priority
= sk
->sk_priority
;
502 skb
->mark
= sk
->sk_mark
;
504 res
= ip_local_out(net
, sk
, skb
);
510 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
512 return -EHOSTUNREACH
;
514 EXPORT_SYMBOL(ip_queue_xmit
);
516 static void ip_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
518 to
->pkt_type
= from
->pkt_type
;
519 to
->priority
= from
->priority
;
520 to
->protocol
= from
->protocol
;
521 to
->skb_iif
= from
->skb_iif
;
523 skb_dst_copy(to
, from
);
525 to
->mark
= from
->mark
;
527 skb_copy_hash(to
, from
);
529 /* Copy the flags to each fragment. */
530 IPCB(to
)->flags
= IPCB(from
)->flags
;
532 #ifdef CONFIG_NET_SCHED
533 to
->tc_index
= from
->tc_index
;
536 #if IS_ENABLED(CONFIG_IP_VS)
537 to
->ipvs_property
= from
->ipvs_property
;
539 skb_copy_secmark(to
, from
);
542 static int ip_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
544 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*))
546 struct iphdr
*iph
= ip_hdr(skb
);
548 if ((iph
->frag_off
& htons(IP_DF
)) == 0)
549 return ip_do_fragment(net
, sk
, skb
, output
);
551 if (unlikely(!skb
->ignore_df
||
552 (IPCB(skb
)->frag_max_size
&&
553 IPCB(skb
)->frag_max_size
> mtu
))) {
554 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
555 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
,
561 return ip_do_fragment(net
, sk
, skb
, output
);
565 * This IP datagram is too large to be sent in one piece. Break it up into
566 * smaller pieces (each of size equal to IP header plus
567 * a block of the data of the original IP data part) that will yet fit in a
568 * single device frame, and queue such a frame for sending.
571 int ip_do_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
572 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*))
576 struct sk_buff
*skb2
;
577 unsigned int mtu
, hlen
, left
, len
, ll_rs
;
579 __be16 not_last_frag
;
580 struct rtable
*rt
= skb_rtable(skb
);
583 /* for offloaded checksums cleanup checksum before fragmentation */
584 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
585 (err
= skb_checksum_help(skb
)))
589 * Point into the IP datagram header.
594 mtu
= ip_skb_dst_mtu(sk
, skb
);
595 if (IPCB(skb
)->frag_max_size
&& IPCB(skb
)->frag_max_size
< mtu
)
596 mtu
= IPCB(skb
)->frag_max_size
;
599 * Setup starting values.
603 mtu
= mtu
- hlen
; /* Size of data space */
604 IPCB(skb
)->flags
|= IPSKB_FRAG_COMPLETE
;
605 ll_rs
= LL_RESERVED_SPACE(rt
->dst
.dev
);
607 /* When frag_list is given, use it. First, check its validity:
608 * some transformers could create wrong frag_list or break existing
609 * one, it is not prohibited. In this case fall back to copying.
611 * LATER: this step can be merged to real generation of fragments,
612 * we can switch to copy when see the first bad fragment.
614 if (skb_has_frag_list(skb
)) {
615 struct sk_buff
*frag
, *frag2
;
616 unsigned int first_len
= skb_pagelen(skb
);
618 if (first_len
- hlen
> mtu
||
619 ((first_len
- hlen
) & 7) ||
620 ip_is_fragment(iph
) ||
622 skb_headroom(skb
) < ll_rs
)
625 skb_walk_frags(skb
, frag
) {
626 /* Correct geometry. */
627 if (frag
->len
> mtu
||
628 ((frag
->len
& 7) && frag
->next
) ||
629 skb_headroom(frag
) < hlen
+ ll_rs
)
630 goto slow_path_clean
;
632 /* Partially cloned skb? */
633 if (skb_shared(frag
))
634 goto slow_path_clean
;
639 frag
->destructor
= sock_wfree
;
641 skb
->truesize
-= frag
->truesize
;
644 /* Everything is OK. Generate! */
648 frag
= skb_shinfo(skb
)->frag_list
;
649 skb_frag_list_init(skb
);
650 skb
->data_len
= first_len
- skb_headlen(skb
);
651 skb
->len
= first_len
;
652 iph
->tot_len
= htons(first_len
);
653 iph
->frag_off
= htons(IP_MF
);
657 /* Prepare header of the next frame,
658 * before previous one went down. */
660 frag
->ip_summed
= CHECKSUM_NONE
;
661 skb_reset_transport_header(frag
);
662 __skb_push(frag
, hlen
);
663 skb_reset_network_header(frag
);
664 memcpy(skb_network_header(frag
), iph
, hlen
);
666 iph
->tot_len
= htons(frag
->len
);
667 ip_copy_metadata(frag
, skb
);
669 ip_options_fragment(frag
);
670 offset
+= skb
->len
- hlen
;
671 iph
->frag_off
= htons(offset
>>3);
673 iph
->frag_off
|= htons(IP_MF
);
674 /* Ready, complete checksum */
678 err
= output(net
, sk
, skb
);
681 IP_INC_STATS(net
, IPSTATS_MIB_FRAGCREATES
);
691 IP_INC_STATS(net
, IPSTATS_MIB_FRAGOKS
);
700 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
704 skb_walk_frags(skb
, frag2
) {
708 frag2
->destructor
= NULL
;
709 skb
->truesize
+= frag2
->truesize
;
716 left
= skb
->len
- hlen
; /* Space per frame */
717 ptr
= hlen
; /* Where to start from */
720 * Fragment the datagram.
723 offset
= (ntohs(iph
->frag_off
) & IP_OFFSET
) << 3;
724 not_last_frag
= iph
->frag_off
& htons(IP_MF
);
727 * Keep copying data until we run out.
732 /* IF: it doesn't fit, use 'mtu' - the data space left */
735 /* IF: we are not sending up to and including the packet end
736 then align the next start on an eight byte boundary */
741 /* Allocate buffer */
742 skb2
= alloc_skb(len
+ hlen
+ ll_rs
, GFP_ATOMIC
);
749 * Set up data on packet
752 ip_copy_metadata(skb2
, skb
);
753 skb_reserve(skb2
, ll_rs
);
754 skb_put(skb2
, len
+ hlen
);
755 skb_reset_network_header(skb2
);
756 skb2
->transport_header
= skb2
->network_header
+ hlen
;
759 * Charge the memory for the fragment to any owner
764 skb_set_owner_w(skb2
, skb
->sk
);
767 * Copy the packet header into the new buffer.
770 skb_copy_from_linear_data(skb
, skb_network_header(skb2
), hlen
);
773 * Copy a block of the IP datagram.
775 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb2
), len
))
780 * Fill in the new header fields.
783 iph
->frag_off
= htons((offset
>> 3));
785 if (IPCB(skb
)->flags
& IPSKB_FRAG_PMTU
)
786 iph
->frag_off
|= htons(IP_DF
);
788 /* ANK: dirty, but effective trick. Upgrade options only if
789 * the segment to be fragmented was THE FIRST (otherwise,
790 * options are already fixed) and make it ONCE
791 * on the initial skb, so that all the following fragments
792 * will inherit fixed options.
795 ip_options_fragment(skb
);
798 * Added AC : If we are fragmenting a fragment that's not the
799 * last fragment then keep MF on each bit
801 if (left
> 0 || not_last_frag
)
802 iph
->frag_off
|= htons(IP_MF
);
807 * Put this fragment into the sending queue.
809 iph
->tot_len
= htons(len
+ hlen
);
813 err
= output(net
, sk
, skb2
);
817 IP_INC_STATS(net
, IPSTATS_MIB_FRAGCREATES
);
820 IP_INC_STATS(net
, IPSTATS_MIB_FRAGOKS
);
825 IP_INC_STATS(net
, IPSTATS_MIB_FRAGFAILS
);
828 EXPORT_SYMBOL(ip_do_fragment
);
831 ip_generic_getfrag(void *from
, char *to
, int offset
, int len
, int odd
, struct sk_buff
*skb
)
833 struct msghdr
*msg
= from
;
835 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
836 if (!copy_from_iter_full(to
, len
, &msg
->msg_iter
))
840 if (!csum_and_copy_from_iter_full(to
, len
, &csum
, &msg
->msg_iter
))
842 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
846 EXPORT_SYMBOL(ip_generic_getfrag
);
849 csum_page(struct page
*page
, int offset
, int copy
)
854 csum
= csum_partial(kaddr
+ offset
, copy
, 0);
859 static int __ip_append_data(struct sock
*sk
,
861 struct sk_buff_head
*queue
,
862 struct inet_cork
*cork
,
863 struct page_frag
*pfrag
,
864 int getfrag(void *from
, char *to
, int offset
,
865 int len
, int odd
, struct sk_buff
*skb
),
866 void *from
, int length
, int transhdrlen
,
869 struct inet_sock
*inet
= inet_sk(sk
);
872 struct ip_options
*opt
= cork
->opt
;
879 unsigned int maxfraglen
, fragheaderlen
, maxnonfragsize
;
880 int csummode
= CHECKSUM_NONE
;
881 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
884 skb
= skb_peek_tail(queue
);
886 exthdrlen
= !skb
? rt
->dst
.header_len
: 0;
887 mtu
= cork
->fragsize
;
888 if (cork
->tx_flags
& SKBTX_ANY_SW_TSTAMP
&&
889 sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)
890 tskey
= sk
->sk_tskey
++;
892 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
894 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
895 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
896 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
898 if (cork
->length
+ length
> maxnonfragsize
- fragheaderlen
) {
899 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
900 mtu
- (opt
? opt
->optlen
: 0));
905 * transhdrlen > 0 means that this is the first fragment and we wish
906 * it won't be fragmented in the future.
909 length
+ fragheaderlen
<= mtu
&&
910 rt
->dst
.dev
->features
& (NETIF_F_HW_CSUM
| NETIF_F_IP_CSUM
) &&
911 !(flags
& MSG_MORE
) &&
913 csummode
= CHECKSUM_PARTIAL
;
915 cork
->length
+= length
;
917 /* So, what's going on in the loop below?
919 * We use calculated fragment length to generate chained skb,
920 * each of segments is IP fragment ready for sending to network after
921 * adding appropriate IP header.
928 /* Check if the remaining data fits into current packet. */
929 copy
= mtu
- skb
->len
;
931 copy
= maxfraglen
- skb
->len
;
934 unsigned int datalen
;
935 unsigned int fraglen
;
936 unsigned int fraggap
;
937 unsigned int alloclen
;
938 struct sk_buff
*skb_prev
;
942 fraggap
= skb_prev
->len
- maxfraglen
;
947 * If remaining data exceeds the mtu,
948 * we know we need more fragment(s).
950 datalen
= length
+ fraggap
;
951 if (datalen
> mtu
- fragheaderlen
)
952 datalen
= maxfraglen
- fragheaderlen
;
953 fraglen
= datalen
+ fragheaderlen
;
955 if ((flags
& MSG_MORE
) &&
956 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
961 alloclen
+= exthdrlen
;
963 /* The last fragment gets additional space at tail.
964 * Note, with MSG_MORE we overallocate on fragments,
965 * because we have no idea what fragment will be
968 if (datalen
== length
+ fraggap
)
969 alloclen
+= rt
->dst
.trailer_len
;
972 skb
= sock_alloc_send_skb(sk
,
973 alloclen
+ hh_len
+ 15,
974 (flags
& MSG_DONTWAIT
), &err
);
977 if (refcount_read(&sk
->sk_wmem_alloc
) <=
979 skb
= sock_wmalloc(sk
,
980 alloclen
+ hh_len
+ 15, 1,
989 * Fill in the control structures
991 skb
->ip_summed
= csummode
;
993 skb_reserve(skb
, hh_len
);
995 /* only the initial fragment is time stamped */
996 skb_shinfo(skb
)->tx_flags
= cork
->tx_flags
;
998 skb_shinfo(skb
)->tskey
= tskey
;
1002 * Find where to start putting bytes.
1004 data
= skb_put(skb
, fraglen
+ exthdrlen
);
1005 skb_set_network_header(skb
, exthdrlen
);
1006 skb
->transport_header
= (skb
->network_header
+
1008 data
+= fragheaderlen
+ exthdrlen
;
1011 skb
->csum
= skb_copy_and_csum_bits(
1012 skb_prev
, maxfraglen
,
1013 data
+ transhdrlen
, fraggap
, 0);
1014 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1017 pskb_trim_unique(skb_prev
, maxfraglen
);
1020 copy
= datalen
- transhdrlen
- fraggap
;
1021 if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1028 length
-= datalen
- fraggap
;
1031 csummode
= CHECKSUM_NONE
;
1033 if ((flags
& MSG_CONFIRM
) && !skb_prev
)
1034 skb_set_dst_pending_confirm(skb
, 1);
1037 * Put the packet on the pending queue.
1039 __skb_queue_tail(queue
, skb
);
1046 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
) &&
1047 skb_tailroom(skb
) >= copy
) {
1051 if (getfrag(from
, skb_put(skb
, copy
),
1052 offset
, copy
, off
, skb
) < 0) {
1053 __skb_trim(skb
, off
);
1058 int i
= skb_shinfo(skb
)->nr_frags
;
1061 if (!sk_page_frag_refill(sk
, pfrag
))
1064 if (!skb_can_coalesce(skb
, i
, pfrag
->page
,
1067 if (i
== MAX_SKB_FRAGS
)
1070 __skb_fill_page_desc(skb
, i
, pfrag
->page
,
1072 skb_shinfo(skb
)->nr_frags
= ++i
;
1073 get_page(pfrag
->page
);
1075 copy
= min_t(int, copy
, pfrag
->size
- pfrag
->offset
);
1077 page_address(pfrag
->page
) + pfrag
->offset
,
1078 offset
, copy
, skb
->len
, skb
) < 0)
1081 pfrag
->offset
+= copy
;
1082 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], copy
);
1084 skb
->data_len
+= copy
;
1085 skb
->truesize
+= copy
;
1086 refcount_add(copy
, &sk
->sk_wmem_alloc
);
1097 cork
->length
-= length
;
1098 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1102 static int ip_setup_cork(struct sock
*sk
, struct inet_cork
*cork
,
1103 struct ipcm_cookie
*ipc
, struct rtable
**rtp
)
1105 struct ip_options_rcu
*opt
;
1109 * setup for corking.
1114 cork
->opt
= kmalloc(sizeof(struct ip_options
) + 40,
1116 if (unlikely(!cork
->opt
))
1119 memcpy(cork
->opt
, &opt
->opt
, sizeof(struct ip_options
) + opt
->opt
.optlen
);
1120 cork
->flags
|= IPCORK_OPT
;
1121 cork
->addr
= ipc
->addr
;
1127 cork
->fragsize
= ip_sk_use_pmtu(sk
) ?
1128 dst_mtu(&rt
->dst
) : READ_ONCE(rt
->dst
.dev
->mtu
);
1130 if (!inetdev_valid_mtu(cork
->fragsize
))
1131 return -ENETUNREACH
;
1133 cork
->dst
= &rt
->dst
;
1134 /* We stole this route, caller should not release it. */
1138 cork
->ttl
= ipc
->ttl
;
1139 cork
->tos
= ipc
->tos
;
1140 cork
->priority
= ipc
->priority
;
1141 cork
->tx_flags
= ipc
->tx_flags
;
1147 * ip_append_data() and ip_append_page() can make one large IP datagram
1148 * from many pieces of data. Each pieces will be holded on the socket
1149 * until ip_push_pending_frames() is called. Each piece can be a page
1152 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1153 * this interface potentially.
1155 * LATER: length must be adjusted by pad at tail, when it is required.
1157 int ip_append_data(struct sock
*sk
, struct flowi4
*fl4
,
1158 int getfrag(void *from
, char *to
, int offset
, int len
,
1159 int odd
, struct sk_buff
*skb
),
1160 void *from
, int length
, int transhdrlen
,
1161 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1164 struct inet_sock
*inet
= inet_sk(sk
);
1167 if (flags
&MSG_PROBE
)
1170 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1171 err
= ip_setup_cork(sk
, &inet
->cork
.base
, ipc
, rtp
);
1178 return __ip_append_data(sk
, fl4
, &sk
->sk_write_queue
, &inet
->cork
.base
,
1179 sk_page_frag(sk
), getfrag
,
1180 from
, length
, transhdrlen
, flags
);
1183 ssize_t
ip_append_page(struct sock
*sk
, struct flowi4
*fl4
, struct page
*page
,
1184 int offset
, size_t size
, int flags
)
1186 struct inet_sock
*inet
= inet_sk(sk
);
1187 struct sk_buff
*skb
;
1189 struct ip_options
*opt
= NULL
;
1190 struct inet_cork
*cork
;
1195 unsigned int maxfraglen
, fragheaderlen
, fraggap
, maxnonfragsize
;
1200 if (flags
&MSG_PROBE
)
1203 if (skb_queue_empty(&sk
->sk_write_queue
))
1206 cork
= &inet
->cork
.base
;
1207 rt
= (struct rtable
*)cork
->dst
;
1208 if (cork
->flags
& IPCORK_OPT
)
1211 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
))
1214 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1215 mtu
= cork
->fragsize
;
1217 fragheaderlen
= sizeof(struct iphdr
) + (opt
? opt
->optlen
: 0);
1218 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
;
1219 maxnonfragsize
= ip_sk_ignore_df(sk
) ? 0xFFFF : mtu
;
1221 if (cork
->length
+ size
> maxnonfragsize
- fragheaderlen
) {
1222 ip_local_error(sk
, EMSGSIZE
, fl4
->daddr
, inet
->inet_dport
,
1223 mtu
- (opt
? opt
->optlen
: 0));
1227 skb
= skb_peek_tail(&sk
->sk_write_queue
);
1231 cork
->length
+= size
;
1234 /* Check if the remaining data fits into current packet. */
1235 len
= mtu
- skb
->len
;
1237 len
= maxfraglen
- skb
->len
;
1240 struct sk_buff
*skb_prev
;
1244 fraggap
= skb_prev
->len
- maxfraglen
;
1246 alloclen
= fragheaderlen
+ hh_len
+ fraggap
+ 15;
1247 skb
= sock_wmalloc(sk
, alloclen
, 1, sk
->sk_allocation
);
1248 if (unlikely(!skb
)) {
1254 * Fill in the control structures
1256 skb
->ip_summed
= CHECKSUM_NONE
;
1258 skb_reserve(skb
, hh_len
);
1261 * Find where to start putting bytes.
1263 skb_put(skb
, fragheaderlen
+ fraggap
);
1264 skb_reset_network_header(skb
);
1265 skb
->transport_header
= (skb
->network_header
+
1268 skb
->csum
= skb_copy_and_csum_bits(skb_prev
,
1270 skb_transport_header(skb
),
1272 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1274 pskb_trim_unique(skb_prev
, maxfraglen
);
1278 * Put the packet on the pending queue.
1280 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1287 if (skb_append_pagefrags(skb
, page
, offset
, len
)) {
1292 if (skb
->ip_summed
== CHECKSUM_NONE
) {
1294 csum
= csum_page(page
, offset
, len
);
1295 skb
->csum
= csum_block_add(skb
->csum
, csum
, skb
->len
);
1299 skb
->data_len
+= len
;
1300 skb
->truesize
+= len
;
1301 refcount_add(len
, &sk
->sk_wmem_alloc
);
1308 cork
->length
-= size
;
1309 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTDISCARDS
);
1313 static void ip_cork_release(struct inet_cork
*cork
)
1315 cork
->flags
&= ~IPCORK_OPT
;
1318 dst_release(cork
->dst
);
1323 * Combined all pending IP fragments on the socket as one IP datagram
1324 * and push them out.
1326 struct sk_buff
*__ip_make_skb(struct sock
*sk
,
1328 struct sk_buff_head
*queue
,
1329 struct inet_cork
*cork
)
1331 struct sk_buff
*skb
, *tmp_skb
;
1332 struct sk_buff
**tail_skb
;
1333 struct inet_sock
*inet
= inet_sk(sk
);
1334 struct net
*net
= sock_net(sk
);
1335 struct ip_options
*opt
= NULL
;
1336 struct rtable
*rt
= (struct rtable
*)cork
->dst
;
1341 skb
= __skb_dequeue(queue
);
1344 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1346 /* move skb->data to ip header from ext header */
1347 if (skb
->data
< skb_network_header(skb
))
1348 __skb_pull(skb
, skb_network_offset(skb
));
1349 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1350 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1351 *tail_skb
= tmp_skb
;
1352 tail_skb
= &(tmp_skb
->next
);
1353 skb
->len
+= tmp_skb
->len
;
1354 skb
->data_len
+= tmp_skb
->len
;
1355 skb
->truesize
+= tmp_skb
->truesize
;
1356 tmp_skb
->destructor
= NULL
;
1360 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1361 * to fragment the frame generated here. No matter, what transforms
1362 * how transforms change size of the packet, it will come out.
1364 skb
->ignore_df
= ip_sk_ignore_df(sk
);
1366 /* DF bit is set when we want to see DF on outgoing frames.
1367 * If ignore_df is set too, we still allow to fragment this frame
1369 if (inet
->pmtudisc
== IP_PMTUDISC_DO
||
1370 inet
->pmtudisc
== IP_PMTUDISC_PROBE
||
1371 (skb
->len
<= dst_mtu(&rt
->dst
) &&
1372 ip_dont_fragment(sk
, &rt
->dst
)))
1375 if (cork
->flags
& IPCORK_OPT
)
1380 else if (rt
->rt_type
== RTN_MULTICAST
)
1383 ttl
= ip_select_ttl(inet
, &rt
->dst
);
1388 iph
->tos
= (cork
->tos
!= -1) ? cork
->tos
: inet
->tos
;
1391 iph
->protocol
= sk
->sk_protocol
;
1392 ip_copy_addrs(iph
, fl4
);
1393 ip_select_ident(net
, skb
, sk
);
1396 iph
->ihl
+= opt
->optlen
>>2;
1397 ip_options_build(skb
, opt
, cork
->addr
, rt
, 0);
1400 skb
->priority
= (cork
->tos
!= -1) ? cork
->priority
: sk
->sk_priority
;
1401 skb
->mark
= sk
->sk_mark
;
1403 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1407 skb_dst_set(skb
, &rt
->dst
);
1409 if (iph
->protocol
== IPPROTO_ICMP
)
1410 icmp_out_count(net
, ((struct icmphdr
*)
1411 skb_transport_header(skb
))->type
);
1413 ip_cork_release(cork
);
1418 int ip_send_skb(struct net
*net
, struct sk_buff
*skb
)
1422 err
= ip_local_out(net
, skb
->sk
, skb
);
1425 err
= net_xmit_errno(err
);
1427 IP_INC_STATS(net
, IPSTATS_MIB_OUTDISCARDS
);
1433 int ip_push_pending_frames(struct sock
*sk
, struct flowi4
*fl4
)
1435 struct sk_buff
*skb
;
1437 skb
= ip_finish_skb(sk
, fl4
);
1441 /* Netfilter gets whole the not fragmented skb. */
1442 return ip_send_skb(sock_net(sk
), skb
);
1446 * Throw away all pending data on the socket.
1448 static void __ip_flush_pending_frames(struct sock
*sk
,
1449 struct sk_buff_head
*queue
,
1450 struct inet_cork
*cork
)
1452 struct sk_buff
*skb
;
1454 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
)
1457 ip_cork_release(cork
);
1460 void ip_flush_pending_frames(struct sock
*sk
)
1462 __ip_flush_pending_frames(sk
, &sk
->sk_write_queue
, &inet_sk(sk
)->cork
.base
);
1465 struct sk_buff
*ip_make_skb(struct sock
*sk
,
1467 int getfrag(void *from
, char *to
, int offset
,
1468 int len
, int odd
, struct sk_buff
*skb
),
1469 void *from
, int length
, int transhdrlen
,
1470 struct ipcm_cookie
*ipc
, struct rtable
**rtp
,
1473 struct inet_cork cork
;
1474 struct sk_buff_head queue
;
1477 if (flags
& MSG_PROBE
)
1480 __skb_queue_head_init(&queue
);
1485 err
= ip_setup_cork(sk
, &cork
, ipc
, rtp
);
1487 return ERR_PTR(err
);
1489 err
= __ip_append_data(sk
, fl4
, &queue
, &cork
,
1490 ¤t
->task_frag
, getfrag
,
1491 from
, length
, transhdrlen
, flags
);
1493 __ip_flush_pending_frames(sk
, &queue
, &cork
);
1494 return ERR_PTR(err
);
1497 return __ip_make_skb(sk
, fl4
, &queue
, &cork
);
1501 * Fetch data from kernel space and fill in checksum if needed.
1503 static int ip_reply_glue_bits(void *dptr
, char *to
, int offset
,
1504 int len
, int odd
, struct sk_buff
*skb
)
1508 csum
= csum_partial_copy_nocheck(dptr
+offset
, to
, len
, 0);
1509 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
1514 * Generic function to send a packet as reply to another packet.
1515 * Used to send some TCP resets/acks so far.
1517 void ip_send_unicast_reply(struct sock
*sk
, struct sk_buff
*skb
,
1518 const struct ip_options
*sopt
,
1519 __be32 daddr
, __be32 saddr
,
1520 const struct ip_reply_arg
*arg
,
1523 struct ip_options_data replyopts
;
1524 struct ipcm_cookie ipc
;
1526 struct rtable
*rt
= skb_rtable(skb
);
1527 struct net
*net
= sock_net(sk
);
1528 struct sk_buff
*nskb
;
1532 if (__ip_options_echo(net
, &replyopts
.opt
.opt
, skb
, sopt
))
1541 if (replyopts
.opt
.opt
.optlen
) {
1542 ipc
.opt
= &replyopts
.opt
;
1544 if (replyopts
.opt
.opt
.srr
)
1545 daddr
= replyopts
.opt
.opt
.faddr
;
1548 oif
= arg
->bound_dev_if
;
1549 if (!oif
&& netif_index_is_l3_master(net
, skb
->skb_iif
))
1552 flowi4_init_output(&fl4
, oif
,
1553 IP4_REPLY_MARK(net
, skb
->mark
),
1555 RT_SCOPE_UNIVERSE
, ip_hdr(skb
)->protocol
,
1556 ip_reply_arg_flowi_flags(arg
),
1558 tcp_hdr(skb
)->source
, tcp_hdr(skb
)->dest
,
1560 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
1561 rt
= ip_route_output_key(net
, &fl4
);
1565 inet_sk(sk
)->tos
= arg
->tos
;
1567 sk
->sk_priority
= skb
->priority
;
1568 sk
->sk_protocol
= ip_hdr(skb
)->protocol
;
1569 sk
->sk_bound_dev_if
= arg
->bound_dev_if
;
1570 sk
->sk_sndbuf
= sysctl_wmem_default
;
1571 sk
->sk_mark
= fl4
.flowi4_mark
;
1572 err
= ip_append_data(sk
, &fl4
, ip_reply_glue_bits
, arg
->iov
->iov_base
,
1573 len
, 0, &ipc
, &rt
, MSG_DONTWAIT
);
1574 if (unlikely(err
)) {
1575 ip_flush_pending_frames(sk
);
1579 nskb
= skb_peek(&sk
->sk_write_queue
);
1581 if (arg
->csumoffset
>= 0)
1582 *((__sum16
*)skb_transport_header(nskb
) +
1583 arg
->csumoffset
) = csum_fold(csum_add(nskb
->csum
,
1585 nskb
->ip_summed
= CHECKSUM_NONE
;
1586 ip_push_pending_frames(sk
, &fl4
);
1592 void __init
ip_init(void)
1597 #if defined(CONFIG_IP_MULTICAST)