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 User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #define pr_fmt(fmt) "UDP: " fmt
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/bootmem.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
94 #include <linux/errno.h>
95 #include <linux/timer.h>
97 #include <linux/inet.h>
98 #include <linux/netdevice.h>
99 #include <linux/slab.h>
100 #include <net/tcp_states.h>
101 #include <linux/skbuff.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <net/net_namespace.h>
105 #include <net/icmp.h>
106 #include <net/route.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <trace/events/udp.h>
110 #include <linux/static_key.h>
111 #include <trace/events/skb.h>
112 #include "udp_impl.h"
114 struct udp_table udp_table __read_mostly
;
115 EXPORT_SYMBOL(udp_table
);
117 long sysctl_udp_mem
[3] __read_mostly
;
118 EXPORT_SYMBOL(sysctl_udp_mem
);
120 int sysctl_udp_rmem_min __read_mostly
;
121 EXPORT_SYMBOL(sysctl_udp_rmem_min
);
123 int sysctl_udp_wmem_min __read_mostly
;
124 EXPORT_SYMBOL(sysctl_udp_wmem_min
);
127 #include <linux/kallsyms.h>
128 #include <linux/ftrace_event.h>
129 int sysctl_udp_met_port __read_mostly
= -1;
130 EXPORT_SYMBOL(sysctl_udp_met_port
);
131 int sysctl_met_is_enable __read_mostly
= -1;
132 EXPORT_SYMBOL(sysctl_met_is_enable
);
133 #ifdef CONFIG_TRACING
134 unsigned long __read_mostly udp_tracing_mark_write_addr
= 0;
138 atomic_long_t udp_memory_allocated
;
139 EXPORT_SYMBOL(udp_memory_allocated
);
141 #define MAX_UDP_PORTS 65536
142 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
144 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
145 const struct udp_hslot
*hslot
,
146 unsigned long *bitmap
,
148 int (*saddr_comp
)(const struct sock
*sk1
,
149 const struct sock
*sk2
),
153 struct hlist_nulls_node
*node
;
154 kuid_t uid
= sock_i_uid(sk
);
156 sk_nulls_for_each(sk2
, node
, &hslot
->head
)
157 if (net_eq(sock_net(sk2
), net
) &&
159 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
160 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
161 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
162 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
163 (!sk2
->sk_reuseport
|| !sk
->sk_reuseport
||
164 !uid_eq(uid
, sock_i_uid(sk2
))) &&
165 (*saddr_comp
)(sk
, sk2
)) {
167 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
176 * Note: we still hold spinlock of primary hash chain, so no other writer
177 * can insert/delete a socket with local_port == num
179 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
180 struct udp_hslot
*hslot2
,
182 int (*saddr_comp
)(const struct sock
*sk1
,
183 const struct sock
*sk2
))
186 struct hlist_nulls_node
*node
;
187 kuid_t uid
= sock_i_uid(sk
);
190 spin_lock(&hslot2
->lock
);
191 udp_portaddr_for_each_entry(sk2
, node
, &hslot2
->head
)
192 if (net_eq(sock_net(sk2
), net
) &&
194 (udp_sk(sk2
)->udp_port_hash
== num
) &&
195 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
196 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
197 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
198 (!sk2
->sk_reuseport
|| !sk
->sk_reuseport
||
199 !uid_eq(uid
, sock_i_uid(sk2
))) &&
200 (*saddr_comp
)(sk
, sk2
)) {
204 spin_unlock(&hslot2
->lock
);
209 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
211 * @sk: socket struct in question
212 * @snum: port number to look up
213 * @saddr_comp: AF-dependent comparison of bound local IP addresses
214 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
217 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
218 int (*saddr_comp
)(const struct sock
*sk1
,
219 const struct sock
*sk2
),
220 unsigned int hash2_nulladdr
)
222 struct udp_hslot
*hslot
, *hslot2
;
223 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
225 struct net
*net
= sock_net(sk
);
228 int low
, high
, remaining
;
230 unsigned short first
, last
;
231 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
233 inet_get_local_port_range(&low
, &high
);
234 remaining
= (high
- low
) + 1;
237 first
= (((u64
)rand
* remaining
) >> 32) + low
;
239 * force rand to be an odd multiple of UDP_HTABLE_SIZE
241 rand
= (rand
| 1) * (udptable
->mask
+ 1);
242 last
= first
+ udptable
->mask
+ 1;
244 hslot
= udp_hashslot(udptable
, net
, first
);
245 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
246 spin_lock_bh(&hslot
->lock
);
247 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
248 saddr_comp
, udptable
->log
);
252 * Iterate on all possible values of snum for this hash.
253 * Using steps of an odd multiple of UDP_HTABLE_SIZE
254 * give us randomization and full range coverage.
257 if (low
<= snum
&& snum
<= high
&&
258 !test_bit(snum
>> udptable
->log
, bitmap
) &&
259 !inet_is_reserved_local_port(snum
))
262 } while (snum
!= first
);
263 spin_unlock_bh(&hslot
->lock
);
264 } while (++first
!= last
);
267 hslot
= udp_hashslot(udptable
, net
, snum
);
268 spin_lock_bh(&hslot
->lock
);
269 if (hslot
->count
> 10) {
271 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
273 slot2
&= udptable
->mask
;
274 hash2_nulladdr
&= udptable
->mask
;
276 hslot2
= udp_hashslot2(udptable
, slot2
);
277 if (hslot
->count
< hslot2
->count
)
278 goto scan_primary_hash
;
280 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
282 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
283 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
284 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
293 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
,
298 inet_sk(sk
)->inet_num
= snum
;
299 udp_sk(sk
)->udp_port_hash
= snum
;
300 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
301 if (sk_unhashed(sk
)) {
302 sk_nulls_add_node_rcu(sk
, &hslot
->head
);
304 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
306 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
307 spin_lock(&hslot2
->lock
);
308 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
311 spin_unlock(&hslot2
->lock
);
315 spin_unlock_bh(&hslot
->lock
);
319 EXPORT_SYMBOL(udp_lib_get_port
);
321 static int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
)
323 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
325 return (!ipv6_only_sock(sk2
) &&
326 (!inet1
->inet_rcv_saddr
|| !inet2
->inet_rcv_saddr
||
327 inet1
->inet_rcv_saddr
== inet2
->inet_rcv_saddr
));
330 static unsigned int udp4_portaddr_hash(struct net
*net
, __be32 saddr
,
333 return jhash_1word((__force u32
)saddr
, net_hash_mix(net
)) ^ port
;
336 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
338 unsigned int hash2_nulladdr
=
339 udp4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
340 unsigned int hash2_partial
=
341 udp4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
343 /* precompute partial secondary hash */
344 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
345 return udp_lib_get_port(sk
, snum
, ipv4_rcv_saddr_equal
, hash2_nulladdr
);
348 static inline int compute_score(struct sock
*sk
, struct net
*net
, __be32 saddr
,
350 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
)
354 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
355 !ipv6_only_sock(sk
)) {
356 struct inet_sock
*inet
= inet_sk(sk
);
358 score
= (sk
->sk_family
== PF_INET
? 2 : 1);
359 if (inet
->inet_rcv_saddr
) {
360 if (inet
->inet_rcv_saddr
!= daddr
)
364 if (inet
->inet_daddr
) {
365 if (inet
->inet_daddr
!= saddr
)
369 if (inet
->inet_dport
) {
370 if (inet
->inet_dport
!= sport
)
374 if (sk
->sk_bound_dev_if
) {
375 if (sk
->sk_bound_dev_if
!= dif
)
384 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
386 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
387 __be32 saddr
, __be16 sport
,
388 __be32 daddr
, unsigned int hnum
, int dif
)
392 if (net_eq(sock_net(sk
), net
) && !ipv6_only_sock(sk
)) {
393 struct inet_sock
*inet
= inet_sk(sk
);
395 if (inet
->inet_rcv_saddr
!= daddr
)
397 if (inet
->inet_num
!= hnum
)
400 score
= (sk
->sk_family
== PF_INET
? 2 : 1);
401 if (inet
->inet_daddr
) {
402 if (inet
->inet_daddr
!= saddr
)
406 if (inet
->inet_dport
) {
407 if (inet
->inet_dport
!= sport
)
411 if (sk
->sk_bound_dev_if
) {
412 if (sk
->sk_bound_dev_if
!= dif
)
421 /* called with read_rcu_lock() */
422 static struct sock
*udp4_lib_lookup2(struct net
*net
,
423 __be32 saddr
, __be16 sport
,
424 __be32 daddr
, unsigned int hnum
, int dif
,
425 struct udp_hslot
*hslot2
, unsigned int slot2
)
427 struct sock
*sk
, *result
;
428 struct hlist_nulls_node
*node
;
429 int score
, badness
, matches
= 0, reuseport
= 0;
435 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
436 score
= compute_score2(sk
, net
, saddr
, sport
,
438 if (score
> badness
) {
441 reuseport
= sk
->sk_reuseport
;
443 hash
= inet_ehashfn(net
, daddr
, hnum
,
444 saddr
, htons(sport
));
447 } else if (score
== badness
&& reuseport
) {
449 if (((u64
)hash
* matches
) >> 32 == 0)
451 hash
= next_pseudo_random32(hash
);
455 * if the nulls value we got at the end of this lookup is
456 * not the expected one, we must restart lookup.
457 * We probably met an item that was moved to another chain.
459 if (get_nulls_value(node
) != slot2
)
462 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
464 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
465 daddr
, hnum
, dif
) < badness
)) {
473 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
474 * harder than this. -DaveM
476 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
477 __be16 sport
, __be32 daddr
, __be16 dport
,
478 int dif
, struct udp_table
*udptable
)
480 struct sock
*sk
, *result
;
481 struct hlist_nulls_node
*node
;
482 unsigned short hnum
= ntohs(dport
);
483 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
484 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
485 int score
, badness
, matches
= 0, reuseport
= 0;
489 if (hslot
->count
> 10) {
490 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
);
491 slot2
= hash2
& udptable
->mask
;
492 hslot2
= &udptable
->hash2
[slot2
];
493 if (hslot
->count
< hslot2
->count
)
496 result
= udp4_lib_lookup2(net
, saddr
, sport
,
500 hash2
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
501 slot2
= hash2
& udptable
->mask
;
502 hslot2
= &udptable
->hash2
[slot2
];
503 if (hslot
->count
< hslot2
->count
)
506 result
= udp4_lib_lookup2(net
, saddr
, sport
,
507 htonl(INADDR_ANY
), hnum
, dif
,
516 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
517 score
= compute_score(sk
, net
, saddr
, hnum
, sport
,
519 if (score
> badness
) {
522 reuseport
= sk
->sk_reuseport
;
524 hash
= inet_ehashfn(net
, daddr
, hnum
,
525 saddr
, htons(sport
));
528 } else if (score
== badness
&& reuseport
) {
530 if (((u64
)hash
* matches
) >> 32 == 0)
532 hash
= next_pseudo_random32(hash
);
536 * if the nulls value we got at the end of this lookup is
537 * not the expected one, we must restart lookup.
538 * We probably met an item that was moved to another chain.
540 if (get_nulls_value(node
) != slot
)
544 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
546 else if (unlikely(compute_score(result
, net
, saddr
, hnum
, sport
,
547 daddr
, dport
, dif
) < badness
)) {
555 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
557 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
558 __be16 sport
, __be16 dport
,
559 struct udp_table
*udptable
)
562 const struct iphdr
*iph
= ip_hdr(skb
);
564 if (unlikely(sk
= skb_steal_sock(skb
)))
567 return __udp4_lib_lookup(dev_net(skb_dst(skb
)->dev
), iph
->saddr
, sport
,
568 iph
->daddr
, dport
, inet_iif(skb
),
572 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
573 __be32 daddr
, __be16 dport
, int dif
)
575 return __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
577 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
579 static inline struct sock
*udp_v4_mcast_next(struct net
*net
, struct sock
*sk
,
580 __be16 loc_port
, __be32 loc_addr
,
581 __be16 rmt_port
, __be32 rmt_addr
,
584 struct hlist_nulls_node
*node
;
586 unsigned short hnum
= ntohs(loc_port
);
588 sk_nulls_for_each_from(s
, node
) {
589 struct inet_sock
*inet
= inet_sk(s
);
591 if (!net_eq(sock_net(s
), net
) ||
592 udp_sk(s
)->udp_port_hash
!= hnum
||
593 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
594 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
595 (inet
->inet_rcv_saddr
&&
596 inet
->inet_rcv_saddr
!= loc_addr
) ||
598 (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
))
600 if (!ip_mc_sf_allow(s
, loc_addr
, rmt_addr
, dif
))
610 * This routine is called by the ICMP module when it gets some
611 * sort of error condition. If err < 0 then the socket should
612 * be closed and the error returned to the user. If err > 0
613 * it's just the icmp type << 8 | icmp code.
614 * Header points to the ip header of the error packet. We move
615 * on past this. Then (as it used to claim before adjustment)
616 * header points to the first 8 bytes of the udp header. We need
617 * to find the appropriate port.
620 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
622 struct inet_sock
*inet
;
623 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
624 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
625 const int type
= icmp_hdr(skb
)->type
;
626 const int code
= icmp_hdr(skb
)->code
;
630 struct net
*net
= dev_net(skb
->dev
);
632 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
633 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, udptable
);
635 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
636 return; /* No socket for error */
645 case ICMP_TIME_EXCEEDED
:
648 case ICMP_SOURCE_QUENCH
:
650 case ICMP_PARAMETERPROB
:
654 case ICMP_DEST_UNREACH
:
655 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
656 ipv4_sk_update_pmtu(skb
, sk
, info
);
657 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
665 if (code
<= NR_ICMP_UNREACH
) {
666 harderr
= icmp_err_convert
[code
].fatal
;
667 err
= icmp_err_convert
[code
].errno
;
671 ipv4_sk_redirect(skb
, sk
);
676 * RFC1122: OK. Passes ICMP errors back to application, as per
679 if (!inet
->recverr
) {
680 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
683 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
686 sk
->sk_error_report(sk
);
691 void udp_err(struct sk_buff
*skb
, u32 info
)
693 __udp4_lib_err(skb
, info
, &udp_table
);
697 * Throw away all pending data and cancel the corking. Socket is locked.
699 void udp_flush_pending_frames(struct sock
*sk
)
701 struct udp_sock
*up
= udp_sk(sk
);
706 ip_flush_pending_frames(sk
);
709 EXPORT_SYMBOL(udp_flush_pending_frames
);
712 * udp4_hwcsum - handle outgoing HW checksumming
713 * @skb: sk_buff containing the filled-in UDP header
714 * (checksum field must be zeroed out)
715 * @src: source IP address
716 * @dst: destination IP address
718 static void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
720 struct udphdr
*uh
= udp_hdr(skb
);
721 struct sk_buff
*frags
= skb_shinfo(skb
)->frag_list
;
722 int offset
= skb_transport_offset(skb
);
723 int len
= skb
->len
- offset
;
729 * Only one fragment on the socket.
731 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
732 skb
->csum_offset
= offsetof(struct udphdr
, check
);
733 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
737 * HW-checksum won't work as there are two or more
738 * fragments on the socket so that all csums of sk_buffs
742 csum
= csum_add(csum
, frags
->csum
);
744 } while ((frags
= frags
->next
));
746 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
747 skb
->ip_summed
= CHECKSUM_NONE
;
749 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
751 uh
->check
= CSUM_MANGLED_0
;
755 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
)
757 struct sock
*sk
= skb
->sk
;
758 struct inet_sock
*inet
= inet_sk(sk
);
761 int is_udplite
= IS_UDPLITE(sk
);
762 int offset
= skb_transport_offset(skb
);
763 int len
= skb
->len
- offset
;
767 * Create a UDP header
770 uh
->source
= inet
->inet_sport
;
771 uh
->dest
= fl4
->fl4_dport
;
772 uh
->len
= htons(len
);
775 if (is_udplite
) /* UDP-Lite */
776 csum
= udplite_csum(skb
);
778 else if (sk
->sk_no_check
== UDP_CSUM_NOXMIT
) { /* UDP csum disabled */
780 skb
->ip_summed
= CHECKSUM_NONE
;
783 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
785 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
789 csum
= udp_csum(skb
);
791 /* add protocol-dependent pseudo-header */
792 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
793 sk
->sk_protocol
, csum
);
795 uh
->check
= CSUM_MANGLED_0
;
798 err
= ip_send_skb(sock_net(sk
), skb
);
800 if (err
== -ENOBUFS
&& !inet
->recverr
) {
801 UDP_INC_STATS_USER(sock_net(sk
),
802 UDP_MIB_SNDBUFERRORS
, is_udplite
);
806 UDP_INC_STATS_USER(sock_net(sk
),
807 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
812 * Push out all pending data as one UDP datagram. Socket is locked.
814 int udp_push_pending_frames(struct sock
*sk
)
816 struct udp_sock
*up
= udp_sk(sk
);
817 struct inet_sock
*inet
= inet_sk(sk
);
818 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
822 skb
= ip_finish_skb(sk
, fl4
);
826 err
= udp_send_skb(skb
, fl4
);
833 EXPORT_SYMBOL(udp_push_pending_frames
);
836 void udp_event_trace_printk(const char * fmt
, int pid
, __u16 port
)
839 #ifdef CONFIG_TRACING
840 if(unlikely(0 == udp_tracing_mark_write_addr
)) {
841 udp_tracing_mark_write_addr
= kallsyms_lookup_name("tracing_mark_write");
843 event_trace_printk(udp_tracing_mark_write_addr
, fmt
, pid
, MET_SOCKET_LATENCY_NAME
, ntohs(port
));
846 EXPORT_SYMBOL(udp_event_trace_printk
);
849 int udp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
852 struct inet_sock
*inet
= inet_sk(sk
);
853 struct udp_sock
*up
= udp_sk(sk
);
854 struct flowi4 fl4_stack
;
857 struct ipcm_cookie ipc
;
858 struct rtable
*rt
= NULL
;
861 __be32 daddr
, faddr
, saddr
;
864 int err
, is_udplite
= IS_UDPLITE(sk
);
865 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
866 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
868 struct ip_options_data opt_copy
;
877 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
883 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
885 fl4
= &inet
->cork
.fl
.u
.ip4
;
888 * There are pending frames.
889 * The socket lock must be held while it's corked.
892 if (likely(up
->pending
)) {
893 if (unlikely(up
->pending
!= AF_INET
)) {
901 ulen
+= sizeof(struct udphdr
);
904 * Get and verify the address.
907 struct sockaddr_in
*usin
= (struct sockaddr_in
*)msg
->msg_name
;
908 if (msg
->msg_namelen
< sizeof(*usin
))
910 if (usin
->sin_family
!= AF_INET
) {
911 if (usin
->sin_family
!= AF_UNSPEC
)
912 return -EAFNOSUPPORT
;
915 daddr
= usin
->sin_addr
.s_addr
;
916 dport
= usin
->sin_port
;
920 if (sk
->sk_state
!= TCP_ESTABLISHED
)
921 return -EDESTADDRREQ
;
922 daddr
= inet
->inet_daddr
;
923 dport
= inet
->inet_dport
;
924 /* Open fast path for connected socket.
925 Route will not be used, if at least one option is set.
929 ipc
.addr
= inet
->inet_saddr
;
931 ipc
.oif
= sk
->sk_bound_dev_if
;
935 if (unlikely((sysctl_met_is_enable
== 1) && (sysctl_udp_met_port
> 0))) {
937 if ((ntohs(inet
->inet_sport
) == sysctl_udp_met_port
) && (len
>= 4)) {
938 __u16
* seq_id
= (__u16
*)((char *)msg
->msg_iov
->iov_base
+ 2);
939 udp_event_trace_printk("S|%d|%s|%d\n", current
->pid
, *seq_id
);
946 sock_tx_timestamp(sk
, &ipc
.tx_flags
);
948 if (msg
->msg_controllen
) {
949 err
= ip_cmsg_send(sock_net(sk
), msg
, &ipc
);
957 struct ip_options_rcu
*inet_opt
;
960 inet_opt
= rcu_dereference(inet
->inet_opt
);
962 memcpy(&opt_copy
, inet_opt
,
963 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
964 ipc
.opt
= &opt_copy
.opt
;
970 ipc
.addr
= faddr
= daddr
;
972 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
975 faddr
= ipc
.opt
->opt
.faddr
;
978 tos
= RT_TOS(inet
->tos
);
979 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
980 (msg
->msg_flags
& MSG_DONTROUTE
) ||
981 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
986 if (ipv4_is_multicast(daddr
)) {
988 ipc
.oif
= inet
->mc_index
;
990 saddr
= inet
->mc_addr
;
993 ipc
.oif
= inet
->uc_index
;
996 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
999 struct net
*net
= sock_net(sk
);
1002 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
1003 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
1004 inet_sk_flowi_flags(sk
)|FLOWI_FLAG_CAN_SLEEP
,
1005 faddr
, saddr
, dport
, inet
->inet_sport
,
1008 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
1009 rt
= ip_route_output_flow(net
, fl4
, sk
);
1013 if (err
== -ENETUNREACH
)
1014 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
1019 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
1020 !sock_flag(sk
, SOCK_BROADCAST
))
1023 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1026 if (msg
->msg_flags
&MSG_CONFIRM
)
1032 daddr
= ipc
.addr
= fl4
->daddr
;
1034 /* Lockless fast path for the non-corking case. */
1036 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
->msg_iov
, ulen
,
1037 sizeof(struct udphdr
), &ipc
, &rt
,
1040 if (!IS_ERR_OR_NULL(skb
))
1041 err
= udp_send_skb(skb
, fl4
);
1046 if (unlikely(up
->pending
)) {
1047 /* The socket is already corked while preparing it. */
1048 /* ... which is an evident application bug. --ANK */
1051 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("cork app bug 2\n"));
1056 * Now cork the socket to pend data.
1058 fl4
= &inet
->cork
.fl
.u
.ip4
;
1061 fl4
->fl4_dport
= dport
;
1062 fl4
->fl4_sport
= inet
->inet_sport
;
1063 up
->pending
= AF_INET
;
1067 err
= ip_append_data(sk
, fl4
, getfrag
, msg
->msg_iov
, ulen
,
1068 sizeof(struct udphdr
), &ipc
, &rt
,
1069 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1071 udp_flush_pending_frames(sk
);
1073 err
= udp_push_pending_frames(sk
);
1074 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1085 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1086 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1087 * we don't have a good statistic (IpOutDiscards but it can be too many
1088 * things). We could add another new stat but at least for now that
1089 * seems like overkill.
1092 if (err
== -ENOBUFS
|| (sk
->sk_socket
&& test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) ) {
1093 UDP_INC_STATS_USER(sock_net(sk
),
1094 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1099 dst_confirm(&rt
->dst
);
1100 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1101 goto back_from_confirm
;
1105 EXPORT_SYMBOL(udp_sendmsg
);
1107 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1108 size_t size
, int flags
)
1110 struct inet_sock
*inet
= inet_sk(sk
);
1111 struct udp_sock
*up
= udp_sk(sk
);
1114 if (flags
& MSG_SENDPAGE_NOTLAST
)
1118 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1120 /* Call udp_sendmsg to specify destination address which
1121 * sendpage interface can't pass.
1122 * This will succeed only when the socket is connected.
1124 ret
= udp_sendmsg(NULL
, sk
, &msg
, 0);
1131 if (unlikely(!up
->pending
)) {
1134 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("udp cork app bug 3\n"));
1138 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1139 page
, offset
, size
, flags
);
1140 if (ret
== -EOPNOTSUPP
) {
1142 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1146 udp_flush_pending_frames(sk
);
1151 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1152 ret
= udp_push_pending_frames(sk
);
1162 * first_packet_length - return length of first packet in receive queue
1165 * Drops all bad checksum frames, until a valid one is found.
1166 * Returns the length of found skb, or 0 if none is found.
1168 static unsigned int first_packet_length(struct sock
*sk
)
1170 struct sk_buff_head list_kill
, *rcvq
= &sk
->sk_receive_queue
;
1171 struct sk_buff
*skb
;
1174 __skb_queue_head_init(&list_kill
);
1176 spin_lock_bh(&rcvq
->lock
);
1177 while ((skb
= skb_peek(rcvq
)) != NULL
&&
1178 udp_lib_checksum_complete(skb
)) {
1179 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1181 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1183 atomic_inc(&sk
->sk_drops
);
1184 __skb_unlink(skb
, rcvq
);
1185 __skb_queue_tail(&list_kill
, skb
);
1187 res
= skb
? skb
->len
: 0;
1188 spin_unlock_bh(&rcvq
->lock
);
1190 if (!skb_queue_empty(&list_kill
)) {
1191 bool slow
= lock_sock_fast(sk
);
1193 __skb_queue_purge(&list_kill
);
1194 sk_mem_reclaim_partial(sk
);
1195 unlock_sock_fast(sk
, slow
);
1201 * IOCTL requests applicable to the UDP protocol
1204 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1209 int amount
= sk_wmem_alloc_get(sk
);
1211 return put_user(amount
, (int __user
*)arg
);
1216 unsigned int amount
= first_packet_length(sk
);
1220 * We will only return the amount
1221 * of this packet since that is all
1222 * that will be read.
1224 amount
-= sizeof(struct udphdr
);
1226 return put_user(amount
, (int __user
*)arg
);
1230 return -ENOIOCTLCMD
;
1235 EXPORT_SYMBOL(udp_ioctl
);
1238 * This should be easy, if there is something there we
1239 * return it, otherwise we block.
1242 int udp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
1243 size_t len
, int noblock
, int flags
, int *addr_len
)
1245 struct inet_sock
*inet
= inet_sk(sk
);
1246 struct sockaddr_in
*sin
= (struct sockaddr_in
*)msg
->msg_name
;
1247 struct sk_buff
*skb
;
1248 unsigned int ulen
, copied
;
1249 int peeked
, off
= 0;
1251 int is_udplite
= IS_UDPLITE(sk
);
1252 bool checksum_valid
= false;
1255 if (flags
& MSG_ERRQUEUE
)
1256 return ip_recv_error(sk
, msg
, len
, addr_len
);
1259 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
1260 &peeked
, &off
, &err
);
1264 ulen
= skb
->len
- sizeof(struct udphdr
);
1268 else if (copied
< ulen
)
1269 msg
->msg_flags
|= MSG_TRUNC
;
1272 * If checksum is needed at all, try to do it while copying the
1273 * data. If the data is truncated, or if we only want a partial
1274 * coverage checksum (UDP-Lite), do it before the copy.
1277 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
1278 checksum_valid
= !udp_lib_checksum_complete(skb
);
1279 if (!checksum_valid
)
1283 if (checksum_valid
|| skb_csum_unnecessary(skb
))
1284 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
1285 msg
->msg_iov
, copied
);
1287 err
= skb_copy_and_csum_datagram_iovec(skb
,
1288 sizeof(struct udphdr
),
1295 if (unlikely(err
)) {
1296 trace_kfree_skb(skb
, udp_recvmsg
);
1298 atomic_inc(&sk
->sk_drops
);
1299 UDP_INC_STATS_USER(sock_net(sk
),
1300 UDP_MIB_INERRORS
, is_udplite
);
1306 UDP_INC_STATS_USER(sock_net(sk
),
1307 UDP_MIB_INDATAGRAMS
, is_udplite
);
1309 sock_recv_ts_and_drops(msg
, sk
, skb
);
1311 /* Copy the address. */
1313 sin
->sin_family
= AF_INET
;
1314 sin
->sin_port
= udp_hdr(skb
)->source
;
1315 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1316 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1317 *addr_len
= sizeof(*sin
);
1319 if (inet
->cmsg_flags
)
1320 ip_cmsg_recv(msg
, skb
);
1323 if (flags
& MSG_TRUNC
)
1327 skb_free_datagram_locked(sk
, skb
);
1332 slow
= lock_sock_fast(sk
);
1333 if (!skb_kill_datagram(sk
, skb
, flags
)) {
1334 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1335 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1337 unlock_sock_fast(sk
, slow
);
1339 /* starting over for a new packet, but check if we need to yield */
1341 msg
->msg_flags
&= ~MSG_TRUNC
;
1346 int udp_disconnect(struct sock
*sk
, int flags
)
1348 struct inet_sock
*inet
= inet_sk(sk
);
1350 * 1003.1g - break association.
1353 sk
->sk_state
= TCP_CLOSE
;
1354 inet
->inet_daddr
= 0;
1355 inet
->inet_dport
= 0;
1356 sock_rps_reset_rxhash(sk
);
1357 sk
->sk_bound_dev_if
= 0;
1358 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1359 inet_reset_saddr(sk
);
1361 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1362 sk
->sk_prot
->unhash(sk
);
1363 inet
->inet_sport
= 0;
1368 EXPORT_SYMBOL(udp_disconnect
);
1370 void udp_lib_unhash(struct sock
*sk
)
1372 if (sk_hashed(sk
)) {
1373 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1374 struct udp_hslot
*hslot
, *hslot2
;
1376 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1377 udp_sk(sk
)->udp_port_hash
);
1378 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1380 spin_lock_bh(&hslot
->lock
);
1381 if (sk_nulls_del_node_init_rcu(sk
)) {
1383 inet_sk(sk
)->inet_num
= 0;
1384 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1386 spin_lock(&hslot2
->lock
);
1387 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1389 spin_unlock(&hslot2
->lock
);
1391 spin_unlock_bh(&hslot
->lock
);
1394 EXPORT_SYMBOL(udp_lib_unhash
);
1397 * inet_rcv_saddr was changed, we must rehash secondary hash
1399 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1401 if (sk_hashed(sk
)) {
1402 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1403 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1405 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1406 nhslot2
= udp_hashslot2(udptable
, newhash
);
1407 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1408 if (hslot2
!= nhslot2
) {
1409 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1410 udp_sk(sk
)->udp_port_hash
);
1411 /* we must lock primary chain too */
1412 spin_lock_bh(&hslot
->lock
);
1414 spin_lock(&hslot2
->lock
);
1415 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1417 spin_unlock(&hslot2
->lock
);
1419 spin_lock(&nhslot2
->lock
);
1420 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1423 spin_unlock(&nhslot2
->lock
);
1425 spin_unlock_bh(&hslot
->lock
);
1429 EXPORT_SYMBOL(udp_lib_rehash
);
1431 static void udp_v4_rehash(struct sock
*sk
)
1433 u16 new_hash
= udp4_portaddr_hash(sock_net(sk
),
1434 inet_sk(sk
)->inet_rcv_saddr
,
1435 inet_sk(sk
)->inet_num
);
1436 udp_lib_rehash(sk
, new_hash
);
1439 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1443 if (inet_sk(sk
)->inet_daddr
)
1444 sock_rps_save_rxhash(sk
, skb
);
1446 rc
= sock_queue_rcv_skb(sk
, skb
);
1448 int is_udplite
= IS_UDPLITE(sk
);
1450 /* Note that an ENOMEM error is charged twice */
1452 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1454 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1456 trace_udp_fail_queue_rcv_skb(rc
, sk
);
1464 static struct static_key udp_encap_needed __read_mostly
;
1465 void udp_encap_enable(void)
1467 if (!static_key_enabled(&udp_encap_needed
))
1468 static_key_slow_inc(&udp_encap_needed
);
1470 EXPORT_SYMBOL(udp_encap_enable
);
1475 * >0: "udp encap" protocol resubmission
1477 * Note that in the success and error cases, the skb is assumed to
1478 * have either been requeued or freed.
1480 int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1482 struct udp_sock
*up
= udp_sk(sk
);
1484 int is_udplite
= IS_UDPLITE(sk
);
1487 * Charge it to the socket, dropping if the queue is full.
1489 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1493 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1494 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
1497 * This is an encapsulation socket so pass the skb to
1498 * the socket's udp_encap_rcv() hook. Otherwise, just
1499 * fall through and pass this up the UDP socket.
1500 * up->encap_rcv() returns the following value:
1501 * =0 if skb was successfully passed to the encap
1502 * handler or was discarded by it.
1503 * >0 if skb should be passed on to UDP.
1504 * <0 if skb should be resubmitted as proto -N
1507 /* if we're overly short, let UDP handle it */
1508 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
1509 if (skb
->len
> sizeof(struct udphdr
) && encap_rcv
!= NULL
) {
1512 ret
= encap_rcv(sk
, skb
);
1514 UDP_INC_STATS_BH(sock_net(sk
),
1515 UDP_MIB_INDATAGRAMS
,
1521 /* FALLTHROUGH -- it's a UDP Packet */
1525 * UDP-Lite specific tests, ignored on UDP sockets
1527 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1530 * MIB statistics other than incrementing the error count are
1531 * disabled for the following two types of errors: these depend
1532 * on the application settings, not on the functioning of the
1533 * protocol stack as such.
1535 * RFC 3828 here recommends (sec 3.3): "There should also be a
1536 * way ... to ... at least let the receiving application block
1537 * delivery of packets with coverage values less than a value
1538 * provided by the application."
1540 if (up
->pcrlen
== 0) { /* full coverage was set */
1541 LIMIT_NETDEBUG(KERN_WARNING
"UDPLite: partial coverage %d while full coverage %d requested\n",
1542 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1545 /* The next case involves violating the min. coverage requested
1546 * by the receiver. This is subtle: if receiver wants x and x is
1547 * greater than the buffersize/MTU then receiver will complain
1548 * that it wants x while sender emits packets of smaller size y.
1549 * Therefore the above ...()->partial_cov statement is essential.
1551 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1552 LIMIT_NETDEBUG(KERN_WARNING
"UDPLite: coverage %d too small, need min %d\n",
1553 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1558 if (rcu_access_pointer(sk
->sk_filter
) &&
1559 udp_lib_checksum_complete(skb
))
1563 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
))
1568 ipv4_pktinfo_prepare(skb
);
1570 if (!sock_owned_by_user(sk
))
1571 rc
= __udp_queue_rcv_skb(sk
, skb
);
1572 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
1581 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1583 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1584 atomic_inc(&sk
->sk_drops
);
1590 static void flush_stack(struct sock
**stack
, unsigned int count
,
1591 struct sk_buff
*skb
, unsigned int final
)
1594 struct sk_buff
*skb1
= NULL
;
1597 for (i
= 0; i
< count
; i
++) {
1599 if (likely(skb1
== NULL
))
1600 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
1603 atomic_inc(&sk
->sk_drops
);
1604 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1606 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1610 if (skb1
&& udp_queue_rcv_skb(sk
, skb1
) <= 0)
1618 * Multicasts and broadcasts go to each listener.
1620 * Note: called only from the BH handler context.
1622 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1624 __be32 saddr
, __be32 daddr
,
1625 struct udp_table
*udptable
)
1627 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
1628 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
1630 unsigned int i
, count
= 0;
1632 spin_lock(&hslot
->lock
);
1633 sk
= sk_nulls_head(&hslot
->head
);
1634 dif
= skb
->dev
->ifindex
;
1635 sk
= udp_v4_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
1637 stack
[count
++] = sk
;
1638 sk
= udp_v4_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
,
1639 daddr
, uh
->source
, saddr
, dif
);
1640 if (unlikely(count
== ARRAY_SIZE(stack
))) {
1643 flush_stack(stack
, count
, skb
, ~0);
1648 * before releasing chain lock, we must take a reference on sockets
1650 for (i
= 0; i
< count
; i
++)
1651 sock_hold(stack
[i
]);
1653 spin_unlock(&hslot
->lock
);
1656 * do the slow work with no lock held
1659 flush_stack(stack
, count
, skb
, count
- 1);
1661 for (i
= 0; i
< count
; i
++)
1669 /* Initialize UDP checksum. If exited with zero value (success),
1670 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1671 * Otherwise, csum completion requires chacksumming packet body,
1672 * including udp header and folding it to skb->csum.
1674 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1677 const struct iphdr
*iph
;
1680 UDP_SKB_CB(skb
)->partial_cov
= 0;
1681 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1683 if (proto
== IPPROTO_UDPLITE
) {
1684 err
= udplite_checksum_init(skb
, uh
);
1690 if (uh
->check
== 0) {
1691 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1692 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1693 if (!csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1695 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1697 if (!skb_csum_unnecessary(skb
))
1698 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1699 skb
->len
, proto
, 0);
1700 /* Probably, we should checksum udp header (it should be in cache
1701 * in any case) and data in tiny packets (< rx copybreak).
1708 * All we need to do is get the socket, and then do a checksum.
1711 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
1716 unsigned short ulen
;
1717 struct rtable
*rt
= skb_rtable(skb
);
1718 __be32 saddr
, daddr
;
1719 struct net
*net
= dev_net(skb
->dev
);
1722 * Validate the packet.
1724 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1725 goto drop
; /* No space for header. */
1728 ulen
= ntohs(uh
->len
);
1729 saddr
= ip_hdr(skb
)->saddr
;
1730 daddr
= ip_hdr(skb
)->daddr
;
1732 if (ulen
> skb
->len
)
1735 if (proto
== IPPROTO_UDP
) {
1736 /* UDP validates ulen. */
1737 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1742 if (udp4_csum_init(skb
, uh
, proto
))
1745 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1746 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
1747 saddr
, daddr
, udptable
);
1749 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
1752 int ret
= udp_queue_rcv_skb(sk
, skb
);
1755 /* a return value > 0 means to resubmit the input, but
1756 * it wants the return to be -protocol, or 0
1763 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1767 /* No socket. Drop packet silently, if checksum is wrong */
1768 if (udp_lib_checksum_complete(skb
))
1771 UDP_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1772 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1775 * Hmm. We got an UDP packet to a port to which we
1776 * don't wanna listen. Ignore it.
1782 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1783 proto
== IPPROTO_UDPLITE
? "Lite" : "",
1784 &saddr
, ntohs(uh
->source
),
1786 &daddr
, ntohs(uh
->dest
));
1791 * RFC1122: OK. Discards the bad packet silently (as far as
1792 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1794 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1795 proto
== IPPROTO_UDPLITE
? "Lite" : "",
1796 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
1798 UDP_INC_STATS_BH(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
1800 UDP_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1805 int udp_rcv(struct sk_buff
*skb
)
1807 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
1810 void udp_destroy_sock(struct sock
*sk
)
1812 struct udp_sock
*up
= udp_sk(sk
);
1813 bool slow
= lock_sock_fast(sk
);
1814 udp_flush_pending_frames(sk
);
1815 unlock_sock_fast(sk
, slow
);
1816 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1817 void (*encap_destroy
)(struct sock
*sk
);
1818 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
1825 * Socket option code for UDP
1827 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
1828 char __user
*optval
, unsigned int optlen
,
1829 int (*push_pending_frames
)(struct sock
*))
1831 struct udp_sock
*up
= udp_sk(sk
);
1834 int is_udplite
= IS_UDPLITE(sk
);
1836 if (optlen
< sizeof(int))
1839 if (get_user(val
, (int __user
*)optval
))
1849 (*push_pending_frames
)(sk
);
1857 case UDP_ENCAP_ESPINUDP
:
1858 case UDP_ENCAP_ESPINUDP_NON_IKE
:
1859 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
1861 case UDP_ENCAP_L2TPINUDP
:
1862 up
->encap_type
= val
;
1872 * UDP-Lite's partial checksum coverage (RFC 3828).
1874 /* The sender sets actual checksum coverage length via this option.
1875 * The case coverage > packet length is handled by send module. */
1876 case UDPLITE_SEND_CSCOV
:
1877 if (!is_udplite
) /* Disable the option on UDP sockets */
1878 return -ENOPROTOOPT
;
1879 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
1881 else if (val
> USHRT_MAX
)
1884 up
->pcflag
|= UDPLITE_SEND_CC
;
1887 /* The receiver specifies a minimum checksum coverage value. To make
1888 * sense, this should be set to at least 8 (as done below). If zero is
1889 * used, this again means full checksum coverage. */
1890 case UDPLITE_RECV_CSCOV
:
1891 if (!is_udplite
) /* Disable the option on UDP sockets */
1892 return -ENOPROTOOPT
;
1893 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
1895 else if (val
> USHRT_MAX
)
1898 up
->pcflag
|= UDPLITE_RECV_CC
;
1908 EXPORT_SYMBOL(udp_lib_setsockopt
);
1910 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1911 char __user
*optval
, unsigned int optlen
)
1913 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1914 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1915 udp_push_pending_frames
);
1916 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1919 #ifdef CONFIG_COMPAT
1920 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1921 char __user
*optval
, unsigned int optlen
)
1923 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1924 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1925 udp_push_pending_frames
);
1926 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1930 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
1931 char __user
*optval
, int __user
*optlen
)
1933 struct udp_sock
*up
= udp_sk(sk
);
1936 if (get_user(len
, optlen
))
1939 len
= min_t(unsigned int, len
, sizeof(int));
1950 val
= up
->encap_type
;
1953 /* The following two cannot be changed on UDP sockets, the return is
1954 * always 0 (which corresponds to the full checksum coverage of UDP). */
1955 case UDPLITE_SEND_CSCOV
:
1959 case UDPLITE_RECV_CSCOV
:
1964 return -ENOPROTOOPT
;
1967 if (put_user(len
, optlen
))
1969 if (copy_to_user(optval
, &val
, len
))
1973 EXPORT_SYMBOL(udp_lib_getsockopt
);
1975 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1976 char __user
*optval
, int __user
*optlen
)
1978 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1979 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1980 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1983 #ifdef CONFIG_COMPAT
1984 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1985 char __user
*optval
, int __user
*optlen
)
1987 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1988 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1989 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1993 * udp_poll - wait for a UDP event.
1994 * @file - file struct
1996 * @wait - poll table
1998 * This is same as datagram poll, except for the special case of
1999 * blocking sockets. If application is using a blocking fd
2000 * and a packet with checksum error is in the queue;
2001 * then it could get return from select indicating data available
2002 * but then block when reading it. Add special case code
2003 * to work around these arguably broken applications.
2005 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2007 unsigned int mask
= datagram_poll(file
, sock
, wait
);
2008 struct sock
*sk
= sock
->sk
;
2010 /* Check for false positives due to checksum errors */
2011 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2012 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && !first_packet_length(sk
))
2013 mask
&= ~(POLLIN
| POLLRDNORM
);
2018 EXPORT_SYMBOL(udp_poll
);
2020 struct proto udp_prot
= {
2022 .owner
= THIS_MODULE
,
2023 .close
= udp_lib_close
,
2024 .connect
= ip4_datagram_connect
,
2025 .disconnect
= udp_disconnect
,
2027 .destroy
= udp_destroy_sock
,
2028 .setsockopt
= udp_setsockopt
,
2029 .getsockopt
= udp_getsockopt
,
2030 .sendmsg
= udp_sendmsg
,
2031 .recvmsg
= udp_recvmsg
,
2032 .sendpage
= udp_sendpage
,
2033 .backlog_rcv
= __udp_queue_rcv_skb
,
2034 .release_cb
= ip4_datagram_release_cb
,
2035 .hash
= udp_lib_hash
,
2036 .unhash
= udp_lib_unhash
,
2037 .rehash
= udp_v4_rehash
,
2038 .get_port
= udp_v4_get_port
,
2039 .memory_allocated
= &udp_memory_allocated
,
2040 .sysctl_mem
= sysctl_udp_mem
,
2041 .sysctl_wmem
= &sysctl_udp_wmem_min
,
2042 .sysctl_rmem
= &sysctl_udp_rmem_min
,
2043 .obj_size
= sizeof(struct udp_sock
),
2044 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2045 .h
.udp_table
= &udp_table
,
2046 #ifdef CONFIG_COMPAT
2047 .compat_setsockopt
= compat_udp_setsockopt
,
2048 .compat_getsockopt
= compat_udp_getsockopt
,
2050 .clear_sk
= sk_prot_clear_portaddr_nulls
,
2052 EXPORT_SYMBOL(udp_prot
);
2054 /* ------------------------------------------------------------------------ */
2055 #ifdef CONFIG_PROC_FS
2057 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2060 struct udp_iter_state
*state
= seq
->private;
2061 struct net
*net
= seq_file_net(seq
);
2063 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
2065 struct hlist_nulls_node
*node
;
2066 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
2068 if (hlist_nulls_empty(&hslot
->head
))
2071 spin_lock_bh(&hslot
->lock
);
2072 sk_nulls_for_each(sk
, node
, &hslot
->head
) {
2073 if (!net_eq(sock_net(sk
), net
))
2075 if (sk
->sk_family
== state
->family
)
2078 spin_unlock_bh(&hslot
->lock
);
2085 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2087 struct udp_iter_state
*state
= seq
->private;
2088 struct net
*net
= seq_file_net(seq
);
2091 sk
= sk_nulls_next(sk
);
2092 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
2095 if (state
->bucket
<= state
->udp_table
->mask
)
2096 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2097 return udp_get_first(seq
, state
->bucket
+ 1);
2102 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2104 struct sock
*sk
= udp_get_first(seq
, 0);
2107 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2109 return pos
? NULL
: sk
;
2112 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2114 struct udp_iter_state
*state
= seq
->private;
2115 state
->bucket
= MAX_UDP_PORTS
;
2117 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2120 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2124 if (v
== SEQ_START_TOKEN
)
2125 sk
= udp_get_idx(seq
, 0);
2127 sk
= udp_get_next(seq
, v
);
2133 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
2135 struct udp_iter_state
*state
= seq
->private;
2137 if (state
->bucket
<= state
->udp_table
->mask
)
2138 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2141 int udp_seq_open(struct inode
*inode
, struct file
*file
)
2143 struct udp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2144 struct udp_iter_state
*s
;
2147 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2148 sizeof(struct udp_iter_state
));
2152 s
= ((struct seq_file
*)file
->private_data
)->private;
2153 s
->family
= afinfo
->family
;
2154 s
->udp_table
= afinfo
->udp_table
;
2157 EXPORT_SYMBOL(udp_seq_open
);
2159 /* ------------------------------------------------------------------------ */
2160 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2162 struct proc_dir_entry
*p
;
2165 afinfo
->seq_ops
.start
= udp_seq_start
;
2166 afinfo
->seq_ops
.next
= udp_seq_next
;
2167 afinfo
->seq_ops
.stop
= udp_seq_stop
;
2169 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2170 afinfo
->seq_fops
, afinfo
);
2175 EXPORT_SYMBOL(udp_proc_register
);
2177 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2179 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2181 EXPORT_SYMBOL(udp_proc_unregister
);
2183 /* ------------------------------------------------------------------------ */
2184 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2185 int bucket
, int *len
)
2187 struct inet_sock
*inet
= inet_sk(sp
);
2188 __be32 dest
= inet
->inet_daddr
;
2189 __be32 src
= inet
->inet_rcv_saddr
;
2190 __u16 destp
= ntohs(inet
->inet_dport
);
2191 __u16 srcp
= ntohs(inet
->inet_sport
);
2193 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2194 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2195 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2196 sk_wmem_alloc_get(sp
),
2197 sk_rmem_alloc_get(sp
),
2199 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2201 atomic_read(&sp
->sk_refcnt
), sp
,
2202 atomic_read(&sp
->sk_drops
), len
);
2205 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2207 if (v
== SEQ_START_TOKEN
)
2208 seq_printf(seq
, "%-127s\n",
2209 " sl local_address rem_address st tx_queue "
2210 "rx_queue tr tm->when retrnsmt uid timeout "
2211 "inode ref pointer drops");
2213 struct udp_iter_state
*state
= seq
->private;
2216 udp4_format_sock(v
, seq
, state
->bucket
, &len
);
2217 seq_printf(seq
, "%*s\n", 127 - len
, "");
2222 static const struct file_operations udp_afinfo_seq_fops
= {
2223 .owner
= THIS_MODULE
,
2224 .open
= udp_seq_open
,
2226 .llseek
= seq_lseek
,
2227 .release
= seq_release_net
2230 /* ------------------------------------------------------------------------ */
2231 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2234 .udp_table
= &udp_table
,
2235 .seq_fops
= &udp_afinfo_seq_fops
,
2237 .show
= udp4_seq_show
,
2241 static int __net_init
udp4_proc_init_net(struct net
*net
)
2243 return udp_proc_register(net
, &udp4_seq_afinfo
);
2246 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2248 udp_proc_unregister(net
, &udp4_seq_afinfo
);
2251 static struct pernet_operations udp4_net_ops
= {
2252 .init
= udp4_proc_init_net
,
2253 .exit
= udp4_proc_exit_net
,
2256 int __init
udp4_proc_init(void)
2258 return register_pernet_subsys(&udp4_net_ops
);
2261 void udp4_proc_exit(void)
2263 unregister_pernet_subsys(&udp4_net_ops
);
2265 #endif /* CONFIG_PROC_FS */
2267 static __initdata
unsigned long uhash_entries
;
2268 static int __init
set_uhash_entries(char *str
)
2275 ret
= kstrtoul(str
, 0, &uhash_entries
);
2279 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2280 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2283 __setup("uhash_entries=", set_uhash_entries
);
2285 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2289 table
->hash
= alloc_large_system_hash(name
,
2290 2 * sizeof(struct udp_hslot
),
2292 21, /* one slot per 2 MB */
2296 UDP_HTABLE_SIZE_MIN
,
2299 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2300 for (i
= 0; i
<= table
->mask
; i
++) {
2301 INIT_HLIST_NULLS_HEAD(&table
->hash
[i
].head
, i
);
2302 table
->hash
[i
].count
= 0;
2303 spin_lock_init(&table
->hash
[i
].lock
);
2305 for (i
= 0; i
<= table
->mask
; i
++) {
2306 INIT_HLIST_NULLS_HEAD(&table
->hash2
[i
].head
, i
);
2307 table
->hash2
[i
].count
= 0;
2308 spin_lock_init(&table
->hash2
[i
].lock
);
2312 void __init
udp_init(void)
2314 unsigned long limit
;
2316 udp_table_init(&udp_table
, "UDP");
2317 limit
= nr_free_buffer_pages() / 8;
2318 limit
= max(limit
, 128UL);
2319 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2320 sysctl_udp_mem
[1] = limit
;
2321 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2323 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2324 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
2327 int udp4_ufo_send_check(struct sk_buff
*skb
)
2329 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
2332 if (likely(!skb
->encapsulation
)) {
2333 const struct iphdr
*iph
;
2339 uh
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
2341 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
2342 skb
->csum_offset
= offsetof(struct udphdr
, check
);
2343 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2348 static struct sk_buff
*skb_udp_tunnel_segment(struct sk_buff
*skb
,
2349 netdev_features_t features
)
2351 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
2352 int mac_len
= skb
->mac_len
;
2353 int tnl_hlen
= skb_inner_mac_header(skb
) - skb_transport_header(skb
);
2354 __be16 protocol
= skb
->protocol
;
2355 netdev_features_t enc_features
;
2358 if (unlikely(!pskb_may_pull(skb
, tnl_hlen
)))
2361 skb
->encapsulation
= 0;
2362 __skb_pull(skb
, tnl_hlen
);
2363 skb_reset_mac_header(skb
);
2364 skb_set_network_header(skb
, skb_inner_network_offset(skb
));
2365 skb
->mac_len
= skb_inner_network_offset(skb
);
2366 skb
->protocol
= htons(ETH_P_TEB
);
2368 /* segment inner packet. */
2369 enc_features
= skb
->dev
->hw_enc_features
& netif_skb_features(skb
);
2370 segs
= skb_mac_gso_segment(skb
, enc_features
);
2371 if (!segs
|| IS_ERR(segs
))
2374 outer_hlen
= skb_tnl_header_len(skb
);
2378 int udp_offset
= outer_hlen
- tnl_hlen
;
2380 skb
->mac_len
= mac_len
;
2382 skb_push(skb
, outer_hlen
);
2383 skb_reset_mac_header(skb
);
2384 skb_set_network_header(skb
, mac_len
);
2385 skb_set_transport_header(skb
, udp_offset
);
2387 uh
->len
= htons(skb
->len
- udp_offset
);
2389 /* csum segment if tunnel sets skb with csum. */
2390 if (unlikely(uh
->check
)) {
2391 struct iphdr
*iph
= ip_hdr(skb
);
2393 uh
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
2394 skb
->len
- udp_offset
,
2396 uh
->check
= csum_fold(skb_checksum(skb
, udp_offset
,
2397 skb
->len
- udp_offset
, 0));
2399 uh
->check
= CSUM_MANGLED_0
;
2402 skb
->ip_summed
= CHECKSUM_NONE
;
2403 skb
->protocol
= protocol
;
2404 } while ((skb
= skb
->next
));
2409 struct sk_buff
*udp4_ufo_fragment(struct sk_buff
*skb
,
2410 netdev_features_t features
)
2412 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
2414 mss
= skb_shinfo(skb
)->gso_size
;
2415 if (unlikely(skb
->len
<= mss
))
2418 if (skb_gso_ok(skb
, features
| NETIF_F_GSO_ROBUST
)) {
2419 /* Packet is from an untrusted source, reset gso_segs. */
2420 int type
= skb_shinfo(skb
)->gso_type
;
2422 if (unlikely(type
& ~(SKB_GSO_UDP
| SKB_GSO_DODGY
|
2423 SKB_GSO_UDP_TUNNEL
|
2425 !(type
& (SKB_GSO_UDP
))))
2428 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(skb
->len
, mss
);
2434 /* Fragment the skb. IP headers of the fragments are updated in
2435 * inet_gso_segment()
2437 if (skb
->encapsulation
&& skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP_TUNNEL
)
2438 segs
= skb_udp_tunnel_segment(skb
, features
);
2443 /* Do software UFO. Complete and fill in the UDP checksum as
2444 * HW cannot do checksum of UDP packets sent as multiple
2447 offset
= skb_checksum_start_offset(skb
);
2448 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2449 offset
+= skb
->csum_offset
;
2450 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2451 skb
->ip_summed
= CHECKSUM_NONE
;
2453 segs
= skb_segment(skb
, features
);