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/inet_hashtables.h>
107 #include <net/route.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <trace/events/udp.h>
111 #include <linux/static_key.h>
112 #include <trace/events/skb.h>
113 #include <net/busy_poll.h>
114 #include "udp_impl.h"
116 struct udp_table udp_table __read_mostly
;
117 EXPORT_SYMBOL(udp_table
);
119 long sysctl_udp_mem
[3] __read_mostly
;
120 EXPORT_SYMBOL(sysctl_udp_mem
);
122 int sysctl_udp_rmem_min __read_mostly
;
123 EXPORT_SYMBOL(sysctl_udp_rmem_min
);
125 int sysctl_udp_wmem_min __read_mostly
;
126 EXPORT_SYMBOL(sysctl_udp_wmem_min
);
128 atomic_long_t udp_memory_allocated
;
129 EXPORT_SYMBOL(udp_memory_allocated
);
131 #define MAX_UDP_PORTS 65536
132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
134 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
135 const struct udp_hslot
*hslot
,
136 unsigned long *bitmap
,
138 int (*saddr_comp
)(const struct sock
*sk1
,
139 const struct sock
*sk2
),
143 struct hlist_nulls_node
*node
;
144 kuid_t uid
= sock_i_uid(sk
);
146 sk_nulls_for_each(sk2
, node
, &hslot
->head
)
147 if (net_eq(sock_net(sk2
), net
) &&
149 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
150 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
151 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
152 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
153 (!sk2
->sk_reuseport
|| !sk
->sk_reuseport
||
154 !uid_eq(uid
, sock_i_uid(sk2
))) &&
155 (*saddr_comp
)(sk
, sk2
)) {
157 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
166 * Note: we still hold spinlock of primary hash chain, so no other writer
167 * can insert/delete a socket with local_port == num
169 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
170 struct udp_hslot
*hslot2
,
172 int (*saddr_comp
)(const struct sock
*sk1
,
173 const struct sock
*sk2
))
176 struct hlist_nulls_node
*node
;
177 kuid_t uid
= sock_i_uid(sk
);
180 spin_lock(&hslot2
->lock
);
181 udp_portaddr_for_each_entry(sk2
, node
, &hslot2
->head
)
182 if (net_eq(sock_net(sk2
), net
) &&
184 (udp_sk(sk2
)->udp_port_hash
== num
) &&
185 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
186 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
187 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
188 (!sk2
->sk_reuseport
|| !sk
->sk_reuseport
||
189 !uid_eq(uid
, sock_i_uid(sk2
))) &&
190 (*saddr_comp
)(sk
, sk2
)) {
194 spin_unlock(&hslot2
->lock
);
199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
201 * @sk: socket struct in question
202 * @snum: port number to look up
203 * @saddr_comp: AF-dependent comparison of bound local IP addresses
204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
207 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
208 int (*saddr_comp
)(const struct sock
*sk1
,
209 const struct sock
*sk2
),
210 unsigned int hash2_nulladdr
)
212 struct udp_hslot
*hslot
, *hslot2
;
213 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
215 struct net
*net
= sock_net(sk
);
218 int low
, high
, remaining
;
220 unsigned short first
, last
;
221 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
223 inet_get_local_port_range(net
, &low
, &high
);
224 remaining
= (high
- low
) + 1;
227 first
= (((u64
)rand
* remaining
) >> 32) + low
;
229 * force rand to be an odd multiple of UDP_HTABLE_SIZE
231 rand
= (rand
| 1) * (udptable
->mask
+ 1);
232 last
= first
+ udptable
->mask
+ 1;
234 hslot
= udp_hashslot(udptable
, net
, first
);
235 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
236 spin_lock_bh(&hslot
->lock
);
237 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
238 saddr_comp
, udptable
->log
);
242 * Iterate on all possible values of snum for this hash.
243 * Using steps of an odd multiple of UDP_HTABLE_SIZE
244 * give us randomization and full range coverage.
247 if (low
<= snum
&& snum
<= high
&&
248 !test_bit(snum
>> udptable
->log
, bitmap
) &&
249 !inet_is_reserved_local_port(snum
))
252 } while (snum
!= first
);
253 spin_unlock_bh(&hslot
->lock
);
254 } while (++first
!= last
);
257 hslot
= udp_hashslot(udptable
, net
, snum
);
258 spin_lock_bh(&hslot
->lock
);
259 if (hslot
->count
> 10) {
261 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
263 slot2
&= udptable
->mask
;
264 hash2_nulladdr
&= udptable
->mask
;
266 hslot2
= udp_hashslot2(udptable
, slot2
);
267 if (hslot
->count
< hslot2
->count
)
268 goto scan_primary_hash
;
270 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
272 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
273 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
274 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
283 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
,
288 inet_sk(sk
)->inet_num
= snum
;
289 udp_sk(sk
)->udp_port_hash
= snum
;
290 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
291 if (sk_unhashed(sk
)) {
292 sk_nulls_add_node_rcu(sk
, &hslot
->head
);
294 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
296 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
297 spin_lock(&hslot2
->lock
);
298 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
301 spin_unlock(&hslot2
->lock
);
305 spin_unlock_bh(&hslot
->lock
);
309 EXPORT_SYMBOL(udp_lib_get_port
);
311 static int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
)
313 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
315 return (!ipv6_only_sock(sk2
) &&
316 (!inet1
->inet_rcv_saddr
|| !inet2
->inet_rcv_saddr
||
317 inet1
->inet_rcv_saddr
== inet2
->inet_rcv_saddr
));
320 static unsigned int udp4_portaddr_hash(struct net
*net
, __be32 saddr
,
323 return jhash_1word((__force u32
)saddr
, net_hash_mix(net
)) ^ port
;
326 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
328 unsigned int hash2_nulladdr
=
329 udp4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
330 unsigned int hash2_partial
=
331 udp4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
333 /* precompute partial secondary hash */
334 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
335 return udp_lib_get_port(sk
, snum
, ipv4_rcv_saddr_equal
, hash2_nulladdr
);
338 static inline int compute_score(struct sock
*sk
, struct net
*net
, __be32 saddr
,
340 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
)
344 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
345 !ipv6_only_sock(sk
)) {
346 struct inet_sock
*inet
= inet_sk(sk
);
348 score
= (sk
->sk_family
== PF_INET
? 2 : 1);
349 if (inet
->inet_rcv_saddr
) {
350 if (inet
->inet_rcv_saddr
!= daddr
)
354 if (inet
->inet_daddr
) {
355 if (inet
->inet_daddr
!= saddr
)
359 if (inet
->inet_dport
) {
360 if (inet
->inet_dport
!= sport
)
364 if (sk
->sk_bound_dev_if
) {
365 if (sk
->sk_bound_dev_if
!= dif
)
374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
376 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
377 __be32 saddr
, __be16 sport
,
378 __be32 daddr
, unsigned int hnum
, int dif
)
382 if (net_eq(sock_net(sk
), net
) && !ipv6_only_sock(sk
)) {
383 struct inet_sock
*inet
= inet_sk(sk
);
385 if (inet
->inet_rcv_saddr
!= daddr
)
387 if (inet
->inet_num
!= hnum
)
390 score
= (sk
->sk_family
== PF_INET
? 2 : 1);
391 if (inet
->inet_daddr
) {
392 if (inet
->inet_daddr
!= saddr
)
396 if (inet
->inet_dport
) {
397 if (inet
->inet_dport
!= sport
)
401 if (sk
->sk_bound_dev_if
) {
402 if (sk
->sk_bound_dev_if
!= dif
)
410 static unsigned int udp_ehashfn(struct net
*net
, const __be32 laddr
,
411 const __u16 lport
, const __be32 faddr
,
414 return __inet_ehashfn(laddr
, lport
, faddr
, fport
,
415 inet_ehash_secret
+ net_hash_mix(net
));
419 /* called with read_rcu_lock() */
420 static struct sock
*udp4_lib_lookup2(struct net
*net
,
421 __be32 saddr
, __be16 sport
,
422 __be32 daddr
, unsigned int hnum
, int dif
,
423 struct udp_hslot
*hslot2
, unsigned int slot2
)
425 struct sock
*sk
, *result
;
426 struct hlist_nulls_node
*node
;
427 int score
, badness
, matches
= 0, reuseport
= 0;
433 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
434 score
= compute_score2(sk
, net
, saddr
, sport
,
436 if (score
> badness
) {
439 reuseport
= sk
->sk_reuseport
;
441 hash
= udp_ehashfn(net
, daddr
, hnum
,
445 } else if (score
== badness
&& reuseport
) {
447 if (((u64
)hash
* matches
) >> 32 == 0)
449 hash
= next_pseudo_random32(hash
);
453 * if the nulls value we got at the end of this lookup is
454 * not the expected one, we must restart lookup.
455 * We probably met an item that was moved to another chain.
457 if (get_nulls_value(node
) != slot2
)
460 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
462 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
463 daddr
, hnum
, dif
) < badness
)) {
471 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
472 * harder than this. -DaveM
474 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
475 __be16 sport
, __be32 daddr
, __be16 dport
,
476 int dif
, struct udp_table
*udptable
)
478 struct sock
*sk
, *result
;
479 struct hlist_nulls_node
*node
;
480 unsigned short hnum
= ntohs(dport
);
481 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
482 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
483 int score
, badness
, matches
= 0, reuseport
= 0;
487 if (hslot
->count
> 10) {
488 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
);
489 slot2
= hash2
& udptable
->mask
;
490 hslot2
= &udptable
->hash2
[slot2
];
491 if (hslot
->count
< hslot2
->count
)
494 result
= udp4_lib_lookup2(net
, saddr
, sport
,
498 hash2
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
499 slot2
= hash2
& udptable
->mask
;
500 hslot2
= &udptable
->hash2
[slot2
];
501 if (hslot
->count
< hslot2
->count
)
504 result
= udp4_lib_lookup2(net
, saddr
, sport
,
505 htonl(INADDR_ANY
), hnum
, dif
,
514 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
515 score
= compute_score(sk
, net
, saddr
, hnum
, sport
,
517 if (score
> badness
) {
520 reuseport
= sk
->sk_reuseport
;
522 hash
= udp_ehashfn(net
, daddr
, hnum
,
526 } else if (score
== badness
&& reuseport
) {
528 if (((u64
)hash
* matches
) >> 32 == 0)
530 hash
= next_pseudo_random32(hash
);
534 * if the nulls value we got at the end of this lookup is
535 * not the expected one, we must restart lookup.
536 * We probably met an item that was moved to another chain.
538 if (get_nulls_value(node
) != slot
)
542 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
544 else if (unlikely(compute_score(result
, net
, saddr
, hnum
, sport
,
545 daddr
, dport
, dif
) < badness
)) {
553 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
555 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
556 __be16 sport
, __be16 dport
,
557 struct udp_table
*udptable
)
560 const struct iphdr
*iph
= ip_hdr(skb
);
562 if (unlikely(sk
= skb_steal_sock(skb
)))
565 return __udp4_lib_lookup(dev_net(skb_dst(skb
)->dev
), iph
->saddr
, sport
,
566 iph
->daddr
, dport
, inet_iif(skb
),
570 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
571 __be32 daddr
, __be16 dport
, int dif
)
573 return __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
575 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
577 static inline bool __udp_is_mcast_sock(struct net
*net
, struct sock
*sk
,
578 __be16 loc_port
, __be32 loc_addr
,
579 __be16 rmt_port
, __be32 rmt_addr
,
580 int dif
, unsigned short hnum
)
582 struct inet_sock
*inet
= inet_sk(sk
);
584 if (!net_eq(sock_net(sk
), net
) ||
585 udp_sk(sk
)->udp_port_hash
!= hnum
||
586 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
587 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
588 (inet
->inet_rcv_saddr
&& inet
->inet_rcv_saddr
!= loc_addr
) ||
589 ipv6_only_sock(sk
) ||
590 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
))
592 if (!ip_mc_sf_allow(sk
, loc_addr
, rmt_addr
, dif
))
597 static inline struct sock
*udp_v4_mcast_next(struct net
*net
, struct sock
*sk
,
598 __be16 loc_port
, __be32 loc_addr
,
599 __be16 rmt_port
, __be32 rmt_addr
,
602 struct hlist_nulls_node
*node
;
604 unsigned short hnum
= ntohs(loc_port
);
606 sk_nulls_for_each_from(s
, node
) {
607 if (__udp_is_mcast_sock(net
, s
,
619 * This routine is called by the ICMP module when it gets some
620 * sort of error condition. If err < 0 then the socket should
621 * be closed and the error returned to the user. If err > 0
622 * it's just the icmp type << 8 | icmp code.
623 * Header points to the ip header of the error packet. We move
624 * on past this. Then (as it used to claim before adjustment)
625 * header points to the first 8 bytes of the udp header. We need
626 * to find the appropriate port.
629 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
631 struct inet_sock
*inet
;
632 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
633 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
634 const int type
= icmp_hdr(skb
)->type
;
635 const int code
= icmp_hdr(skb
)->code
;
639 struct net
*net
= dev_net(skb
->dev
);
641 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
642 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, udptable
);
644 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
645 return; /* No socket for error */
654 case ICMP_TIME_EXCEEDED
:
657 case ICMP_SOURCE_QUENCH
:
659 case ICMP_PARAMETERPROB
:
663 case ICMP_DEST_UNREACH
:
664 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
665 ipv4_sk_update_pmtu(skb
, sk
, info
);
666 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
674 if (code
<= NR_ICMP_UNREACH
) {
675 harderr
= icmp_err_convert
[code
].fatal
;
676 err
= icmp_err_convert
[code
].errno
;
680 ipv4_sk_redirect(skb
, sk
);
685 * RFC1122: OK. Passes ICMP errors back to application, as per
688 if (!inet
->recverr
) {
689 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
692 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
695 sk
->sk_error_report(sk
);
700 void udp_err(struct sk_buff
*skb
, u32 info
)
702 __udp4_lib_err(skb
, info
, &udp_table
);
706 * Throw away all pending data and cancel the corking. Socket is locked.
708 void udp_flush_pending_frames(struct sock
*sk
)
710 struct udp_sock
*up
= udp_sk(sk
);
715 ip_flush_pending_frames(sk
);
718 EXPORT_SYMBOL(udp_flush_pending_frames
);
721 * udp4_hwcsum - handle outgoing HW checksumming
722 * @skb: sk_buff containing the filled-in UDP header
723 * (checksum field must be zeroed out)
724 * @src: source IP address
725 * @dst: destination IP address
727 void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
729 struct udphdr
*uh
= udp_hdr(skb
);
730 struct sk_buff
*frags
= skb_shinfo(skb
)->frag_list
;
731 int offset
= skb_transport_offset(skb
);
732 int len
= skb
->len
- offset
;
738 * Only one fragment on the socket.
740 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
741 skb
->csum_offset
= offsetof(struct udphdr
, check
);
742 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
746 * HW-checksum won't work as there are two or more
747 * fragments on the socket so that all csums of sk_buffs
751 csum
= csum_add(csum
, frags
->csum
);
753 } while ((frags
= frags
->next
));
755 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
756 skb
->ip_summed
= CHECKSUM_NONE
;
758 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
760 uh
->check
= CSUM_MANGLED_0
;
763 EXPORT_SYMBOL_GPL(udp4_hwcsum
);
765 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
)
767 struct sock
*sk
= skb
->sk
;
768 struct inet_sock
*inet
= inet_sk(sk
);
771 int is_udplite
= IS_UDPLITE(sk
);
772 int offset
= skb_transport_offset(skb
);
773 int len
= skb
->len
- offset
;
777 * Create a UDP header
780 uh
->source
= inet
->inet_sport
;
781 uh
->dest
= fl4
->fl4_dport
;
782 uh
->len
= htons(len
);
785 if (is_udplite
) /* UDP-Lite */
786 csum
= udplite_csum(skb
);
788 else if (sk
->sk_no_check
== UDP_CSUM_NOXMIT
) { /* UDP csum disabled */
790 skb
->ip_summed
= CHECKSUM_NONE
;
793 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
795 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
799 csum
= udp_csum(skb
);
801 /* add protocol-dependent pseudo-header */
802 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
803 sk
->sk_protocol
, csum
);
805 uh
->check
= CSUM_MANGLED_0
;
808 err
= ip_send_skb(sock_net(sk
), skb
);
810 if (err
== -ENOBUFS
&& !inet
->recverr
) {
811 UDP_INC_STATS_USER(sock_net(sk
),
812 UDP_MIB_SNDBUFERRORS
, is_udplite
);
816 UDP_INC_STATS_USER(sock_net(sk
),
817 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
822 * Push out all pending data as one UDP datagram. Socket is locked.
824 int udp_push_pending_frames(struct sock
*sk
)
826 struct udp_sock
*up
= udp_sk(sk
);
827 struct inet_sock
*inet
= inet_sk(sk
);
828 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
832 skb
= ip_finish_skb(sk
, fl4
);
836 err
= udp_send_skb(skb
, fl4
);
843 EXPORT_SYMBOL(udp_push_pending_frames
);
845 int udp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
848 struct inet_sock
*inet
= inet_sk(sk
);
849 struct udp_sock
*up
= udp_sk(sk
);
850 struct flowi4 fl4_stack
;
853 struct ipcm_cookie ipc
;
854 struct rtable
*rt
= NULL
;
857 __be32 daddr
, faddr
, saddr
;
860 int err
, is_udplite
= IS_UDPLITE(sk
);
861 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
862 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
864 struct ip_options_data opt_copy
;
873 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
881 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
883 fl4
= &inet
->cork
.fl
.u
.ip4
;
886 * There are pending frames.
887 * The socket lock must be held while it's corked.
890 if (likely(up
->pending
)) {
891 if (unlikely(up
->pending
!= AF_INET
)) {
899 ulen
+= sizeof(struct udphdr
);
902 * Get and verify the address.
905 struct sockaddr_in
*usin
= (struct sockaddr_in
*)msg
->msg_name
;
906 if (msg
->msg_namelen
< sizeof(*usin
))
908 if (usin
->sin_family
!= AF_INET
) {
909 if (usin
->sin_family
!= AF_UNSPEC
)
910 return -EAFNOSUPPORT
;
913 daddr
= usin
->sin_addr
.s_addr
;
914 dport
= usin
->sin_port
;
918 if (sk
->sk_state
!= TCP_ESTABLISHED
)
919 return -EDESTADDRREQ
;
920 daddr
= inet
->inet_daddr
;
921 dport
= inet
->inet_dport
;
922 /* Open fast path for connected socket.
923 Route will not be used, if at least one option is set.
927 ipc
.addr
= inet
->inet_saddr
;
929 ipc
.oif
= sk
->sk_bound_dev_if
;
931 sock_tx_timestamp(sk
, &ipc
.tx_flags
);
933 if (msg
->msg_controllen
) {
934 err
= ip_cmsg_send(sock_net(sk
), msg
, &ipc
);
942 struct ip_options_rcu
*inet_opt
;
945 inet_opt
= rcu_dereference(inet
->inet_opt
);
947 memcpy(&opt_copy
, inet_opt
,
948 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
949 ipc
.opt
= &opt_copy
.opt
;
955 ipc
.addr
= faddr
= daddr
;
957 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
960 faddr
= ipc
.opt
->opt
.faddr
;
963 tos
= get_rttos(&ipc
, inet
);
964 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
965 (msg
->msg_flags
& MSG_DONTROUTE
) ||
966 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
971 if (ipv4_is_multicast(daddr
)) {
973 ipc
.oif
= inet
->mc_index
;
975 saddr
= inet
->mc_addr
;
978 ipc
.oif
= inet
->uc_index
;
981 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
984 struct net
*net
= sock_net(sk
);
987 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
988 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
989 inet_sk_flowi_flags(sk
)|FLOWI_FLAG_CAN_SLEEP
,
990 faddr
, saddr
, dport
, inet
->inet_sport
);
992 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
993 rt
= ip_route_output_flow(net
, fl4
, sk
);
997 if (err
== -ENETUNREACH
)
998 IP_INC_STATS_BH(net
, IPSTATS_MIB_OUTNOROUTES
);
1003 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
1004 !sock_flag(sk
, SOCK_BROADCAST
))
1007 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1010 if (msg
->msg_flags
&MSG_CONFIRM
)
1016 daddr
= ipc
.addr
= fl4
->daddr
;
1018 /* Lockless fast path for the non-corking case. */
1020 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
->msg_iov
, ulen
,
1021 sizeof(struct udphdr
), &ipc
, &rt
,
1024 if (!IS_ERR_OR_NULL(skb
))
1025 err
= udp_send_skb(skb
, fl4
);
1030 if (unlikely(up
->pending
)) {
1031 /* The socket is already corked while preparing it. */
1032 /* ... which is an evident application bug. --ANK */
1035 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("cork app bug 2\n"));
1040 * Now cork the socket to pend data.
1042 fl4
= &inet
->cork
.fl
.u
.ip4
;
1045 fl4
->fl4_dport
= dport
;
1046 fl4
->fl4_sport
= inet
->inet_sport
;
1047 up
->pending
= AF_INET
;
1051 err
= ip_append_data(sk
, fl4
, getfrag
, msg
->msg_iov
, ulen
,
1052 sizeof(struct udphdr
), &ipc
, &rt
,
1053 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1055 udp_flush_pending_frames(sk
);
1057 err
= udp_push_pending_frames(sk
);
1058 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1069 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1070 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1071 * we don't have a good statistic (IpOutDiscards but it can be too many
1072 * things). We could add another new stat but at least for now that
1073 * seems like overkill.
1075 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1076 UDP_INC_STATS_USER(sock_net(sk
),
1077 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1082 dst_confirm(&rt
->dst
);
1083 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1084 goto back_from_confirm
;
1088 EXPORT_SYMBOL(udp_sendmsg
);
1090 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1091 size_t size
, int flags
)
1093 struct inet_sock
*inet
= inet_sk(sk
);
1094 struct udp_sock
*up
= udp_sk(sk
);
1098 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1100 /* Call udp_sendmsg to specify destination address which
1101 * sendpage interface can't pass.
1102 * This will succeed only when the socket is connected.
1104 ret
= udp_sendmsg(NULL
, sk
, &msg
, 0);
1111 if (unlikely(!up
->pending
)) {
1114 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("udp cork app bug 3\n"));
1118 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1119 page
, offset
, size
, flags
);
1120 if (ret
== -EOPNOTSUPP
) {
1122 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1126 udp_flush_pending_frames(sk
);
1131 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1132 ret
= udp_push_pending_frames(sk
);
1142 * first_packet_length - return length of first packet in receive queue
1145 * Drops all bad checksum frames, until a valid one is found.
1146 * Returns the length of found skb, or 0 if none is found.
1148 static unsigned int first_packet_length(struct sock
*sk
)
1150 struct sk_buff_head list_kill
, *rcvq
= &sk
->sk_receive_queue
;
1151 struct sk_buff
*skb
;
1154 __skb_queue_head_init(&list_kill
);
1156 spin_lock_bh(&rcvq
->lock
);
1157 while ((skb
= skb_peek(rcvq
)) != NULL
&&
1158 udp_lib_checksum_complete(skb
)) {
1159 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1161 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1163 atomic_inc(&sk
->sk_drops
);
1164 __skb_unlink(skb
, rcvq
);
1165 __skb_queue_tail(&list_kill
, skb
);
1167 res
= skb
? skb
->len
: 0;
1168 spin_unlock_bh(&rcvq
->lock
);
1170 if (!skb_queue_empty(&list_kill
)) {
1171 bool slow
= lock_sock_fast(sk
);
1173 __skb_queue_purge(&list_kill
);
1174 sk_mem_reclaim_partial(sk
);
1175 unlock_sock_fast(sk
, slow
);
1181 * IOCTL requests applicable to the UDP protocol
1184 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1189 int amount
= sk_wmem_alloc_get(sk
);
1191 return put_user(amount
, (int __user
*)arg
);
1196 unsigned int amount
= first_packet_length(sk
);
1200 * We will only return the amount
1201 * of this packet since that is all
1202 * that will be read.
1204 amount
-= sizeof(struct udphdr
);
1206 return put_user(amount
, (int __user
*)arg
);
1210 return -ENOIOCTLCMD
;
1215 EXPORT_SYMBOL(udp_ioctl
);
1218 * This should be easy, if there is something there we
1219 * return it, otherwise we block.
1222 int udp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
1223 size_t len
, int noblock
, int flags
, int *addr_len
)
1225 struct inet_sock
*inet
= inet_sk(sk
);
1226 struct sockaddr_in
*sin
= (struct sockaddr_in
*)msg
->msg_name
;
1227 struct sk_buff
*skb
;
1228 unsigned int ulen
, copied
;
1229 int peeked
, off
= 0;
1231 int is_udplite
= IS_UDPLITE(sk
);
1235 * Check any passed addresses
1238 *addr_len
= sizeof(*sin
);
1240 if (flags
& MSG_ERRQUEUE
)
1241 return ip_recv_error(sk
, msg
, len
);
1244 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
1245 &peeked
, &off
, &err
);
1249 ulen
= skb
->len
- sizeof(struct udphdr
);
1253 else if (copied
< ulen
)
1254 msg
->msg_flags
|= MSG_TRUNC
;
1257 * If checksum is needed at all, try to do it while copying the
1258 * data. If the data is truncated, or if we only want a partial
1259 * coverage checksum (UDP-Lite), do it before the copy.
1262 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
1263 if (udp_lib_checksum_complete(skb
))
1267 if (skb_csum_unnecessary(skb
))
1268 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
1269 msg
->msg_iov
, copied
);
1271 err
= skb_copy_and_csum_datagram_iovec(skb
,
1272 sizeof(struct udphdr
),
1279 if (unlikely(err
)) {
1280 trace_kfree_skb(skb
, udp_recvmsg
);
1282 atomic_inc(&sk
->sk_drops
);
1283 UDP_INC_STATS_USER(sock_net(sk
),
1284 UDP_MIB_INERRORS
, is_udplite
);
1290 UDP_INC_STATS_USER(sock_net(sk
),
1291 UDP_MIB_INDATAGRAMS
, is_udplite
);
1293 sock_recv_ts_and_drops(msg
, sk
, skb
);
1295 /* Copy the address. */
1297 sin
->sin_family
= AF_INET
;
1298 sin
->sin_port
= udp_hdr(skb
)->source
;
1299 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1300 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1302 if (inet
->cmsg_flags
)
1303 ip_cmsg_recv(msg
, skb
);
1306 if (flags
& MSG_TRUNC
)
1310 skb_free_datagram_locked(sk
, skb
);
1315 slow
= lock_sock_fast(sk
);
1316 if (!skb_kill_datagram(sk
, skb
, flags
)) {
1317 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1318 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1320 unlock_sock_fast(sk
, slow
);
1325 /* starting over for a new packet */
1326 msg
->msg_flags
&= ~MSG_TRUNC
;
1331 int udp_disconnect(struct sock
*sk
, int flags
)
1333 struct inet_sock
*inet
= inet_sk(sk
);
1335 * 1003.1g - break association.
1338 sk
->sk_state
= TCP_CLOSE
;
1339 inet
->inet_daddr
= 0;
1340 inet
->inet_dport
= 0;
1341 sock_rps_reset_rxhash(sk
);
1342 sk
->sk_bound_dev_if
= 0;
1343 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1344 inet_reset_saddr(sk
);
1346 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1347 sk
->sk_prot
->unhash(sk
);
1348 inet
->inet_sport
= 0;
1353 EXPORT_SYMBOL(udp_disconnect
);
1355 void udp_lib_unhash(struct sock
*sk
)
1357 if (sk_hashed(sk
)) {
1358 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1359 struct udp_hslot
*hslot
, *hslot2
;
1361 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1362 udp_sk(sk
)->udp_port_hash
);
1363 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1365 spin_lock_bh(&hslot
->lock
);
1366 if (sk_nulls_del_node_init_rcu(sk
)) {
1368 inet_sk(sk
)->inet_num
= 0;
1369 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1371 spin_lock(&hslot2
->lock
);
1372 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1374 spin_unlock(&hslot2
->lock
);
1376 spin_unlock_bh(&hslot
->lock
);
1379 EXPORT_SYMBOL(udp_lib_unhash
);
1382 * inet_rcv_saddr was changed, we must rehash secondary hash
1384 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1386 if (sk_hashed(sk
)) {
1387 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1388 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1390 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1391 nhslot2
= udp_hashslot2(udptable
, newhash
);
1392 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1393 if (hslot2
!= nhslot2
) {
1394 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1395 udp_sk(sk
)->udp_port_hash
);
1396 /* we must lock primary chain too */
1397 spin_lock_bh(&hslot
->lock
);
1399 spin_lock(&hslot2
->lock
);
1400 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1402 spin_unlock(&hslot2
->lock
);
1404 spin_lock(&nhslot2
->lock
);
1405 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1408 spin_unlock(&nhslot2
->lock
);
1410 spin_unlock_bh(&hslot
->lock
);
1414 EXPORT_SYMBOL(udp_lib_rehash
);
1416 static void udp_v4_rehash(struct sock
*sk
)
1418 u16 new_hash
= udp4_portaddr_hash(sock_net(sk
),
1419 inet_sk(sk
)->inet_rcv_saddr
,
1420 inet_sk(sk
)->inet_num
);
1421 udp_lib_rehash(sk
, new_hash
);
1424 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1428 if (inet_sk(sk
)->inet_daddr
) {
1429 sock_rps_save_rxhash(sk
, skb
);
1430 sk_mark_napi_id(sk
, skb
);
1433 rc
= sock_queue_rcv_skb(sk
, skb
);
1435 int is_udplite
= IS_UDPLITE(sk
);
1437 /* Note that an ENOMEM error is charged twice */
1439 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1441 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1443 trace_udp_fail_queue_rcv_skb(rc
, sk
);
1451 static struct static_key udp_encap_needed __read_mostly
;
1452 void udp_encap_enable(void)
1454 if (!static_key_enabled(&udp_encap_needed
))
1455 static_key_slow_inc(&udp_encap_needed
);
1457 EXPORT_SYMBOL(udp_encap_enable
);
1462 * >0: "udp encap" protocol resubmission
1464 * Note that in the success and error cases, the skb is assumed to
1465 * have either been requeued or freed.
1467 int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1469 struct udp_sock
*up
= udp_sk(sk
);
1471 int is_udplite
= IS_UDPLITE(sk
);
1474 * Charge it to the socket, dropping if the queue is full.
1476 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1480 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1481 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
1484 * This is an encapsulation socket so pass the skb to
1485 * the socket's udp_encap_rcv() hook. Otherwise, just
1486 * fall through and pass this up the UDP socket.
1487 * up->encap_rcv() returns the following value:
1488 * =0 if skb was successfully passed to the encap
1489 * handler or was discarded by it.
1490 * >0 if skb should be passed on to UDP.
1491 * <0 if skb should be resubmitted as proto -N
1494 /* if we're overly short, let UDP handle it */
1495 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
1496 if (skb
->len
> sizeof(struct udphdr
) && encap_rcv
!= NULL
) {
1499 ret
= encap_rcv(sk
, skb
);
1501 UDP_INC_STATS_BH(sock_net(sk
),
1502 UDP_MIB_INDATAGRAMS
,
1508 /* FALLTHROUGH -- it's a UDP Packet */
1512 * UDP-Lite specific tests, ignored on UDP sockets
1514 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1517 * MIB statistics other than incrementing the error count are
1518 * disabled for the following two types of errors: these depend
1519 * on the application settings, not on the functioning of the
1520 * protocol stack as such.
1522 * RFC 3828 here recommends (sec 3.3): "There should also be a
1523 * way ... to ... at least let the receiving application block
1524 * delivery of packets with coverage values less than a value
1525 * provided by the application."
1527 if (up
->pcrlen
== 0) { /* full coverage was set */
1528 LIMIT_NETDEBUG(KERN_WARNING
"UDPLite: partial coverage %d while full coverage %d requested\n",
1529 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1532 /* The next case involves violating the min. coverage requested
1533 * by the receiver. This is subtle: if receiver wants x and x is
1534 * greater than the buffersize/MTU then receiver will complain
1535 * that it wants x while sender emits packets of smaller size y.
1536 * Therefore the above ...()->partial_cov statement is essential.
1538 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1539 LIMIT_NETDEBUG(KERN_WARNING
"UDPLite: coverage %d too small, need min %d\n",
1540 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1545 if (rcu_access_pointer(sk
->sk_filter
) &&
1546 udp_lib_checksum_complete(skb
))
1550 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
))
1555 ipv4_pktinfo_prepare(sk
, skb
);
1557 if (!sock_owned_by_user(sk
))
1558 rc
= __udp_queue_rcv_skb(sk
, skb
);
1559 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
1568 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1570 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1571 atomic_inc(&sk
->sk_drops
);
1577 static void flush_stack(struct sock
**stack
, unsigned int count
,
1578 struct sk_buff
*skb
, unsigned int final
)
1581 struct sk_buff
*skb1
= NULL
;
1584 for (i
= 0; i
< count
; i
++) {
1586 if (likely(skb1
== NULL
))
1587 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
1590 atomic_inc(&sk
->sk_drops
);
1591 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1593 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1597 if (skb1
&& udp_queue_rcv_skb(sk
, skb1
) <= 0)
1604 static void udp_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
1606 struct dst_entry
*dst
= skb_dst(skb
);
1609 sk
->sk_rx_dst
= dst
;
1613 * Multicasts and broadcasts go to each listener.
1615 * Note: called only from the BH handler context.
1617 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1619 __be32 saddr
, __be32 daddr
,
1620 struct udp_table
*udptable
)
1622 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
1623 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
1625 unsigned int i
, count
= 0;
1627 spin_lock(&hslot
->lock
);
1628 sk
= sk_nulls_head(&hslot
->head
);
1629 dif
= skb
->dev
->ifindex
;
1630 sk
= udp_v4_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
1632 stack
[count
++] = sk
;
1633 sk
= udp_v4_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
,
1634 daddr
, uh
->source
, saddr
, dif
);
1635 if (unlikely(count
== ARRAY_SIZE(stack
))) {
1638 flush_stack(stack
, count
, skb
, ~0);
1643 * before releasing chain lock, we must take a reference on sockets
1645 for (i
= 0; i
< count
; i
++)
1646 sock_hold(stack
[i
]);
1648 spin_unlock(&hslot
->lock
);
1651 * do the slow work with no lock held
1654 flush_stack(stack
, count
, skb
, count
- 1);
1656 for (i
= 0; i
< count
; i
++)
1664 /* Initialize UDP checksum. If exited with zero value (success),
1665 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1666 * Otherwise, csum completion requires chacksumming packet body,
1667 * including udp header and folding it to skb->csum.
1669 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1672 const struct iphdr
*iph
;
1675 UDP_SKB_CB(skb
)->partial_cov
= 0;
1676 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1678 if (proto
== IPPROTO_UDPLITE
) {
1679 err
= udplite_checksum_init(skb
, uh
);
1685 if (uh
->check
== 0) {
1686 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1687 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1688 if (!csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1690 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1692 if (!skb_csum_unnecessary(skb
))
1693 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1694 skb
->len
, proto
, 0);
1695 /* Probably, we should checksum udp header (it should be in cache
1696 * in any case) and data in tiny packets (< rx copybreak).
1703 * All we need to do is get the socket, and then do a checksum.
1706 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
1711 unsigned short ulen
;
1712 struct rtable
*rt
= skb_rtable(skb
);
1713 __be32 saddr
, daddr
;
1714 struct net
*net
= dev_net(skb
->dev
);
1717 * Validate the packet.
1719 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1720 goto drop
; /* No space for header. */
1723 ulen
= ntohs(uh
->len
);
1724 saddr
= ip_hdr(skb
)->saddr
;
1725 daddr
= ip_hdr(skb
)->daddr
;
1727 if (ulen
> skb
->len
)
1730 if (proto
== IPPROTO_UDP
) {
1731 /* UDP validates ulen. */
1732 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1737 if (udp4_csum_init(skb
, uh
, proto
))
1744 if (unlikely(sk
->sk_rx_dst
== NULL
))
1745 udp_sk_rx_dst_set(sk
, skb
);
1747 ret
= udp_queue_rcv_skb(sk
, skb
);
1749 /* a return value > 0 means to resubmit the input, but
1750 * it wants the return to be -protocol, or 0
1756 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1757 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
1758 saddr
, daddr
, udptable
);
1760 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
1766 ret
= udp_queue_rcv_skb(sk
, skb
);
1769 /* a return value > 0 means to resubmit the input, but
1770 * it wants the return to be -protocol, or 0
1777 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1781 /* No socket. Drop packet silently, if checksum is wrong */
1782 if (udp_lib_checksum_complete(skb
))
1785 UDP_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1786 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1789 * Hmm. We got an UDP packet to a port to which we
1790 * don't wanna listen. Ignore it.
1796 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1797 proto
== IPPROTO_UDPLITE
? "Lite" : "",
1798 &saddr
, ntohs(uh
->source
),
1800 &daddr
, ntohs(uh
->dest
));
1805 * RFC1122: OK. Discards the bad packet silently (as far as
1806 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1808 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1809 proto
== IPPROTO_UDPLITE
? "Lite" : "",
1810 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
1812 UDP_INC_STATS_BH(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
1814 UDP_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1819 /* We can only early demux multicast if there is a single matching socket.
1820 * If more than one socket found returns NULL
1822 static struct sock
*__udp4_lib_mcast_demux_lookup(struct net
*net
,
1823 __be16 loc_port
, __be32 loc_addr
,
1824 __be16 rmt_port
, __be32 rmt_addr
,
1827 struct sock
*sk
, *result
;
1828 struct hlist_nulls_node
*node
;
1829 unsigned short hnum
= ntohs(loc_port
);
1830 unsigned int count
, slot
= udp_hashfn(net
, hnum
, udp_table
.mask
);
1831 struct udp_hslot
*hslot
= &udp_table
.hash
[slot
];
1837 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
1838 if (__udp_is_mcast_sock(net
, sk
,
1847 * if the nulls value we got at the end of this lookup is
1848 * not the expected one, we must restart lookup.
1849 * We probably met an item that was moved to another chain.
1851 if (get_nulls_value(node
) != slot
)
1856 unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
1858 else if (unlikely(!__udp_is_mcast_sock(net
, result
,
1870 /* For unicast we should only early demux connected sockets or we can
1871 * break forwarding setups. The chains here can be long so only check
1872 * if the first socket is an exact match and if not move on.
1874 static struct sock
*__udp4_lib_demux_lookup(struct net
*net
,
1875 __be16 loc_port
, __be32 loc_addr
,
1876 __be16 rmt_port
, __be32 rmt_addr
,
1879 struct sock
*sk
, *result
;
1880 struct hlist_nulls_node
*node
;
1881 unsigned short hnum
= ntohs(loc_port
);
1882 unsigned int hash2
= udp4_portaddr_hash(net
, loc_addr
, hnum
);
1883 unsigned int slot2
= hash2
& udp_table
.mask
;
1884 struct udp_hslot
*hslot2
= &udp_table
.hash2
[slot2
];
1885 INET_ADDR_COOKIE(acookie
, rmt_addr
, loc_addr
)
1886 const __portpair ports
= INET_COMBINED_PORTS(rmt_port
, hnum
);
1890 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
1891 if (INET_MATCH(sk
, net
, acookie
,
1892 rmt_addr
, loc_addr
, ports
, dif
))
1894 /* Only check first socket in chain */
1899 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
1901 else if (unlikely(!INET_MATCH(sk
, net
, acookie
,
1912 void udp_v4_early_demux(struct sk_buff
*skb
)
1914 const struct iphdr
*iph
= ip_hdr(skb
);
1915 const struct udphdr
*uh
= udp_hdr(skb
);
1917 struct dst_entry
*dst
;
1918 struct net
*net
= dev_net(skb
->dev
);
1919 int dif
= skb
->dev
->ifindex
;
1921 /* validate the packet */
1922 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct udphdr
)))
1925 if (skb
->pkt_type
== PACKET_BROADCAST
||
1926 skb
->pkt_type
== PACKET_MULTICAST
)
1927 sk
= __udp4_lib_mcast_demux_lookup(net
, uh
->dest
, iph
->daddr
,
1928 uh
->source
, iph
->saddr
, dif
);
1929 else if (skb
->pkt_type
== PACKET_HOST
)
1930 sk
= __udp4_lib_demux_lookup(net
, uh
->dest
, iph
->daddr
,
1931 uh
->source
, iph
->saddr
, dif
);
1939 skb
->destructor
= sock_edemux
;
1940 dst
= sk
->sk_rx_dst
;
1943 dst
= dst_check(dst
, 0);
1945 skb_dst_set_noref(skb
, dst
);
1948 int udp_rcv(struct sk_buff
*skb
)
1950 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
1953 void udp_destroy_sock(struct sock
*sk
)
1955 struct udp_sock
*up
= udp_sk(sk
);
1956 bool slow
= lock_sock_fast(sk
);
1957 udp_flush_pending_frames(sk
);
1958 unlock_sock_fast(sk
, slow
);
1959 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1960 void (*encap_destroy
)(struct sock
*sk
);
1961 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
1968 * Socket option code for UDP
1970 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
1971 char __user
*optval
, unsigned int optlen
,
1972 int (*push_pending_frames
)(struct sock
*))
1974 struct udp_sock
*up
= udp_sk(sk
);
1977 int is_udplite
= IS_UDPLITE(sk
);
1979 if (optlen
< sizeof(int))
1982 if (get_user(val
, (int __user
*)optval
))
1992 (*push_pending_frames
)(sk
);
2000 case UDP_ENCAP_ESPINUDP
:
2001 case UDP_ENCAP_ESPINUDP_NON_IKE
:
2002 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
2004 case UDP_ENCAP_L2TPINUDP
:
2005 up
->encap_type
= val
;
2015 * UDP-Lite's partial checksum coverage (RFC 3828).
2017 /* The sender sets actual checksum coverage length via this option.
2018 * The case coverage > packet length is handled by send module. */
2019 case UDPLITE_SEND_CSCOV
:
2020 if (!is_udplite
) /* Disable the option on UDP sockets */
2021 return -ENOPROTOOPT
;
2022 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
2024 else if (val
> USHRT_MAX
)
2027 up
->pcflag
|= UDPLITE_SEND_CC
;
2030 /* The receiver specifies a minimum checksum coverage value. To make
2031 * sense, this should be set to at least 8 (as done below). If zero is
2032 * used, this again means full checksum coverage. */
2033 case UDPLITE_RECV_CSCOV
:
2034 if (!is_udplite
) /* Disable the option on UDP sockets */
2035 return -ENOPROTOOPT
;
2036 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
2038 else if (val
> USHRT_MAX
)
2041 up
->pcflag
|= UDPLITE_RECV_CC
;
2051 EXPORT_SYMBOL(udp_lib_setsockopt
);
2053 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2054 char __user
*optval
, unsigned int optlen
)
2056 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2057 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2058 udp_push_pending_frames
);
2059 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2062 #ifdef CONFIG_COMPAT
2063 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2064 char __user
*optval
, unsigned int optlen
)
2066 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2067 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2068 udp_push_pending_frames
);
2069 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2073 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
2074 char __user
*optval
, int __user
*optlen
)
2076 struct udp_sock
*up
= udp_sk(sk
);
2079 if (get_user(len
, optlen
))
2082 len
= min_t(unsigned int, len
, sizeof(int));
2093 val
= up
->encap_type
;
2096 /* The following two cannot be changed on UDP sockets, the return is
2097 * always 0 (which corresponds to the full checksum coverage of UDP). */
2098 case UDPLITE_SEND_CSCOV
:
2102 case UDPLITE_RECV_CSCOV
:
2107 return -ENOPROTOOPT
;
2110 if (put_user(len
, optlen
))
2112 if (copy_to_user(optval
, &val
, len
))
2116 EXPORT_SYMBOL(udp_lib_getsockopt
);
2118 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2119 char __user
*optval
, int __user
*optlen
)
2121 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2122 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2123 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2126 #ifdef CONFIG_COMPAT
2127 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2128 char __user
*optval
, int __user
*optlen
)
2130 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2131 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2132 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2136 * udp_poll - wait for a UDP event.
2137 * @file - file struct
2139 * @wait - poll table
2141 * This is same as datagram poll, except for the special case of
2142 * blocking sockets. If application is using a blocking fd
2143 * and a packet with checksum error is in the queue;
2144 * then it could get return from select indicating data available
2145 * but then block when reading it. Add special case code
2146 * to work around these arguably broken applications.
2148 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2150 unsigned int mask
= datagram_poll(file
, sock
, wait
);
2151 struct sock
*sk
= sock
->sk
;
2153 sock_rps_record_flow(sk
);
2155 /* Check for false positives due to checksum errors */
2156 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2157 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && !first_packet_length(sk
))
2158 mask
&= ~(POLLIN
| POLLRDNORM
);
2163 EXPORT_SYMBOL(udp_poll
);
2165 struct proto udp_prot
= {
2167 .owner
= THIS_MODULE
,
2168 .close
= udp_lib_close
,
2169 .connect
= ip4_datagram_connect
,
2170 .disconnect
= udp_disconnect
,
2172 .destroy
= udp_destroy_sock
,
2173 .setsockopt
= udp_setsockopt
,
2174 .getsockopt
= udp_getsockopt
,
2175 .sendmsg
= udp_sendmsg
,
2176 .recvmsg
= udp_recvmsg
,
2177 .sendpage
= udp_sendpage
,
2178 .backlog_rcv
= __udp_queue_rcv_skb
,
2179 .release_cb
= ip4_datagram_release_cb
,
2180 .hash
= udp_lib_hash
,
2181 .unhash
= udp_lib_unhash
,
2182 .rehash
= udp_v4_rehash
,
2183 .get_port
= udp_v4_get_port
,
2184 .memory_allocated
= &udp_memory_allocated
,
2185 .sysctl_mem
= sysctl_udp_mem
,
2186 .sysctl_wmem
= &sysctl_udp_wmem_min
,
2187 .sysctl_rmem
= &sysctl_udp_rmem_min
,
2188 .obj_size
= sizeof(struct udp_sock
),
2189 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2190 .h
.udp_table
= &udp_table
,
2191 #ifdef CONFIG_COMPAT
2192 .compat_setsockopt
= compat_udp_setsockopt
,
2193 .compat_getsockopt
= compat_udp_getsockopt
,
2195 .clear_sk
= sk_prot_clear_portaddr_nulls
,
2197 EXPORT_SYMBOL(udp_prot
);
2199 /* ------------------------------------------------------------------------ */
2200 #ifdef CONFIG_PROC_FS
2202 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2205 struct udp_iter_state
*state
= seq
->private;
2206 struct net
*net
= seq_file_net(seq
);
2208 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
2210 struct hlist_nulls_node
*node
;
2211 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
2213 if (hlist_nulls_empty(&hslot
->head
))
2216 spin_lock_bh(&hslot
->lock
);
2217 sk_nulls_for_each(sk
, node
, &hslot
->head
) {
2218 if (!net_eq(sock_net(sk
), net
))
2220 if (sk
->sk_family
== state
->family
)
2223 spin_unlock_bh(&hslot
->lock
);
2230 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2232 struct udp_iter_state
*state
= seq
->private;
2233 struct net
*net
= seq_file_net(seq
);
2236 sk
= sk_nulls_next(sk
);
2237 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
2240 if (state
->bucket
<= state
->udp_table
->mask
)
2241 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2242 return udp_get_first(seq
, state
->bucket
+ 1);
2247 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2249 struct sock
*sk
= udp_get_first(seq
, 0);
2252 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2254 return pos
? NULL
: sk
;
2257 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2259 struct udp_iter_state
*state
= seq
->private;
2260 state
->bucket
= MAX_UDP_PORTS
;
2262 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2265 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2269 if (v
== SEQ_START_TOKEN
)
2270 sk
= udp_get_idx(seq
, 0);
2272 sk
= udp_get_next(seq
, v
);
2278 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
2280 struct udp_iter_state
*state
= seq
->private;
2282 if (state
->bucket
<= state
->udp_table
->mask
)
2283 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2286 int udp_seq_open(struct inode
*inode
, struct file
*file
)
2288 struct udp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2289 struct udp_iter_state
*s
;
2292 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2293 sizeof(struct udp_iter_state
));
2297 s
= ((struct seq_file
*)file
->private_data
)->private;
2298 s
->family
= afinfo
->family
;
2299 s
->udp_table
= afinfo
->udp_table
;
2302 EXPORT_SYMBOL(udp_seq_open
);
2304 /* ------------------------------------------------------------------------ */
2305 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2307 struct proc_dir_entry
*p
;
2310 afinfo
->seq_ops
.start
= udp_seq_start
;
2311 afinfo
->seq_ops
.next
= udp_seq_next
;
2312 afinfo
->seq_ops
.stop
= udp_seq_stop
;
2314 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2315 afinfo
->seq_fops
, afinfo
);
2320 EXPORT_SYMBOL(udp_proc_register
);
2322 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2324 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2326 EXPORT_SYMBOL(udp_proc_unregister
);
2328 /* ------------------------------------------------------------------------ */
2329 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2330 int bucket
, int *len
)
2332 struct inet_sock
*inet
= inet_sk(sp
);
2333 __be32 dest
= inet
->inet_daddr
;
2334 __be32 src
= inet
->inet_rcv_saddr
;
2335 __u16 destp
= ntohs(inet
->inet_dport
);
2336 __u16 srcp
= ntohs(inet
->inet_sport
);
2338 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2339 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d%n",
2340 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2341 sk_wmem_alloc_get(sp
),
2342 sk_rmem_alloc_get(sp
),
2344 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2346 atomic_read(&sp
->sk_refcnt
), sp
,
2347 atomic_read(&sp
->sk_drops
), len
);
2350 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2352 if (v
== SEQ_START_TOKEN
)
2353 seq_printf(seq
, "%-127s\n",
2354 " sl local_address rem_address st tx_queue "
2355 "rx_queue tr tm->when retrnsmt uid timeout "
2356 "inode ref pointer drops");
2358 struct udp_iter_state
*state
= seq
->private;
2361 udp4_format_sock(v
, seq
, state
->bucket
, &len
);
2362 seq_printf(seq
, "%*s\n", 127 - len
, "");
2367 static const struct file_operations udp_afinfo_seq_fops
= {
2368 .owner
= THIS_MODULE
,
2369 .open
= udp_seq_open
,
2371 .llseek
= seq_lseek
,
2372 .release
= seq_release_net
2375 /* ------------------------------------------------------------------------ */
2376 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2379 .udp_table
= &udp_table
,
2380 .seq_fops
= &udp_afinfo_seq_fops
,
2382 .show
= udp4_seq_show
,
2386 static int __net_init
udp4_proc_init_net(struct net
*net
)
2388 return udp_proc_register(net
, &udp4_seq_afinfo
);
2391 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2393 udp_proc_unregister(net
, &udp4_seq_afinfo
);
2396 static struct pernet_operations udp4_net_ops
= {
2397 .init
= udp4_proc_init_net
,
2398 .exit
= udp4_proc_exit_net
,
2401 int __init
udp4_proc_init(void)
2403 return register_pernet_subsys(&udp4_net_ops
);
2406 void udp4_proc_exit(void)
2408 unregister_pernet_subsys(&udp4_net_ops
);
2410 #endif /* CONFIG_PROC_FS */
2412 static __initdata
unsigned long uhash_entries
;
2413 static int __init
set_uhash_entries(char *str
)
2420 ret
= kstrtoul(str
, 0, &uhash_entries
);
2424 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2425 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2428 __setup("uhash_entries=", set_uhash_entries
);
2430 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2434 table
->hash
= alloc_large_system_hash(name
,
2435 2 * sizeof(struct udp_hslot
),
2437 21, /* one slot per 2 MB */
2441 UDP_HTABLE_SIZE_MIN
,
2444 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2445 for (i
= 0; i
<= table
->mask
; i
++) {
2446 INIT_HLIST_NULLS_HEAD(&table
->hash
[i
].head
, i
);
2447 table
->hash
[i
].count
= 0;
2448 spin_lock_init(&table
->hash
[i
].lock
);
2450 for (i
= 0; i
<= table
->mask
; i
++) {
2451 INIT_HLIST_NULLS_HEAD(&table
->hash2
[i
].head
, i
);
2452 table
->hash2
[i
].count
= 0;
2453 spin_lock_init(&table
->hash2
[i
].lock
);
2457 void __init
udp_init(void)
2459 unsigned long limit
;
2461 udp_table_init(&udp_table
, "UDP");
2462 limit
= nr_free_buffer_pages() / 8;
2463 limit
= max(limit
, 128UL);
2464 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2465 sysctl_udp_mem
[1] = limit
;
2466 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2468 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2469 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
2472 struct sk_buff
*skb_udp_tunnel_segment(struct sk_buff
*skb
,
2473 netdev_features_t features
)
2475 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
2476 int mac_len
= skb
->mac_len
;
2477 int tnl_hlen
= skb_inner_mac_header(skb
) - skb_transport_header(skb
);
2478 __be16 protocol
= skb
->protocol
;
2479 netdev_features_t enc_features
;
2482 if (unlikely(!pskb_may_pull(skb
, tnl_hlen
)))
2485 skb
->encapsulation
= 0;
2486 __skb_pull(skb
, tnl_hlen
);
2487 skb_reset_mac_header(skb
);
2488 skb_set_network_header(skb
, skb_inner_network_offset(skb
));
2489 skb
->mac_len
= skb_inner_network_offset(skb
);
2490 skb
->protocol
= htons(ETH_P_TEB
);
2492 /* segment inner packet. */
2493 enc_features
= skb
->dev
->hw_enc_features
& netif_skb_features(skb
);
2494 segs
= skb_mac_gso_segment(skb
, enc_features
);
2495 if (!segs
|| IS_ERR(segs
))
2498 outer_hlen
= skb_tnl_header_len(skb
);
2502 int udp_offset
= outer_hlen
- tnl_hlen
;
2504 skb_reset_inner_headers(skb
);
2505 skb
->encapsulation
= 1;
2507 skb
->mac_len
= mac_len
;
2509 skb_push(skb
, outer_hlen
);
2510 skb_reset_mac_header(skb
);
2511 skb_set_network_header(skb
, mac_len
);
2512 skb_set_transport_header(skb
, udp_offset
);
2514 uh
->len
= htons(skb
->len
- udp_offset
);
2516 /* csum segment if tunnel sets skb with csum. */
2517 if (protocol
== htons(ETH_P_IP
) && unlikely(uh
->check
)) {
2518 struct iphdr
*iph
= ip_hdr(skb
);
2520 uh
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
2521 skb
->len
- udp_offset
,
2523 uh
->check
= csum_fold(skb_checksum(skb
, udp_offset
,
2524 skb
->len
- udp_offset
, 0));
2526 uh
->check
= CSUM_MANGLED_0
;
2528 } else if (protocol
== htons(ETH_P_IPV6
)) {
2529 struct ipv6hdr
*ipv6h
= ipv6_hdr(skb
);
2530 u32 len
= skb
->len
- udp_offset
;
2532 uh
->check
= ~csum_ipv6_magic(&ipv6h
->saddr
, &ipv6h
->daddr
,
2533 len
, IPPROTO_UDP
, 0);
2534 uh
->check
= csum_fold(skb_checksum(skb
, udp_offset
, len
, 0));
2536 uh
->check
= CSUM_MANGLED_0
;
2537 skb
->ip_summed
= CHECKSUM_NONE
;
2540 skb
->protocol
= protocol
;
2541 } while ((skb
= skb
->next
));