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 <linux/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>
93 #include <linux/inetdevice.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
117 #include <net/addrconf.h>
119 struct udp_table udp_table __read_mostly
;
120 EXPORT_SYMBOL(udp_table
);
122 long sysctl_udp_mem
[3] __read_mostly
;
123 EXPORT_SYMBOL(sysctl_udp_mem
);
125 int sysctl_udp_rmem_min __read_mostly
;
126 EXPORT_SYMBOL(sysctl_udp_rmem_min
);
128 int sysctl_udp_wmem_min __read_mostly
;
129 EXPORT_SYMBOL(sysctl_udp_wmem_min
);
131 atomic_long_t udp_memory_allocated
;
132 EXPORT_SYMBOL(udp_memory_allocated
);
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
137 /* IPCB reference means this can not be used from early demux */
138 static bool udp_lib_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
140 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
141 if (!net
->ipv4
.sysctl_udp_l3mdev_accept
&&
142 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
148 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
149 const struct udp_hslot
*hslot
,
150 unsigned long *bitmap
,
151 struct sock
*sk
, unsigned int log
)
154 kuid_t uid
= sock_i_uid(sk
);
156 sk_for_each(sk2
, &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 inet_rcv_saddr_equal(sk
, sk2
, true)) {
164 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
165 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
166 uid_eq(uid
, sock_i_uid(sk2
))) {
172 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
181 * Note: we still hold spinlock of primary hash chain, so no other writer
182 * can insert/delete a socket with local_port == num
184 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
185 struct udp_hslot
*hslot2
,
189 kuid_t uid
= sock_i_uid(sk
);
192 spin_lock(&hslot2
->lock
);
193 udp_portaddr_for_each_entry(sk2
, &hslot2
->head
) {
194 if (net_eq(sock_net(sk2
), net
) &&
196 (udp_sk(sk2
)->udp_port_hash
== num
) &&
197 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
198 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
199 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
200 inet_rcv_saddr_equal(sk
, sk2
, true)) {
201 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
202 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
203 uid_eq(uid
, sock_i_uid(sk2
))) {
211 spin_unlock(&hslot2
->lock
);
215 static int udp_reuseport_add_sock(struct sock
*sk
, struct udp_hslot
*hslot
)
217 struct net
*net
= sock_net(sk
);
218 kuid_t uid
= sock_i_uid(sk
);
221 sk_for_each(sk2
, &hslot
->head
) {
222 if (net_eq(sock_net(sk2
), net
) &&
224 sk2
->sk_family
== sk
->sk_family
&&
225 ipv6_only_sock(sk2
) == ipv6_only_sock(sk
) &&
226 (udp_sk(sk2
)->udp_port_hash
== udp_sk(sk
)->udp_port_hash
) &&
227 (sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
228 sk2
->sk_reuseport
&& uid_eq(uid
, sock_i_uid(sk2
)) &&
229 inet_rcv_saddr_equal(sk
, sk2
, false)) {
230 return reuseport_add_sock(sk
, sk2
);
234 return reuseport_alloc(sk
);
238 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
240 * @sk: socket struct in question
241 * @snum: port number to look up
242 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
245 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
246 unsigned int hash2_nulladdr
)
248 struct udp_hslot
*hslot
, *hslot2
;
249 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
251 struct net
*net
= sock_net(sk
);
254 int low
, high
, remaining
;
256 unsigned short first
, last
;
257 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
259 inet_get_local_port_range(net
, &low
, &high
);
260 remaining
= (high
- low
) + 1;
262 rand
= prandom_u32();
263 first
= reciprocal_scale(rand
, remaining
) + low
;
265 * force rand to be an odd multiple of UDP_HTABLE_SIZE
267 rand
= (rand
| 1) * (udptable
->mask
+ 1);
268 last
= first
+ udptable
->mask
+ 1;
270 hslot
= udp_hashslot(udptable
, net
, first
);
271 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
272 spin_lock_bh(&hslot
->lock
);
273 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
278 * Iterate on all possible values of snum for this hash.
279 * Using steps of an odd multiple of UDP_HTABLE_SIZE
280 * give us randomization and full range coverage.
283 if (low
<= snum
&& snum
<= high
&&
284 !test_bit(snum
>> udptable
->log
, bitmap
) &&
285 !inet_is_local_reserved_port(net
, snum
))
288 } while (snum
!= first
);
289 spin_unlock_bh(&hslot
->lock
);
291 } while (++first
!= last
);
294 hslot
= udp_hashslot(udptable
, net
, snum
);
295 spin_lock_bh(&hslot
->lock
);
296 if (hslot
->count
> 10) {
298 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
300 slot2
&= udptable
->mask
;
301 hash2_nulladdr
&= udptable
->mask
;
303 hslot2
= udp_hashslot2(udptable
, slot2
);
304 if (hslot
->count
< hslot2
->count
)
305 goto scan_primary_hash
;
307 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
, sk
);
308 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
309 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
310 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
319 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
, 0))
323 inet_sk(sk
)->inet_num
= snum
;
324 udp_sk(sk
)->udp_port_hash
= snum
;
325 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
326 if (sk_unhashed(sk
)) {
327 if (sk
->sk_reuseport
&&
328 udp_reuseport_add_sock(sk
, hslot
)) {
329 inet_sk(sk
)->inet_num
= 0;
330 udp_sk(sk
)->udp_port_hash
= 0;
331 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
335 sk_add_node_rcu(sk
, &hslot
->head
);
337 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
339 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
340 spin_lock(&hslot2
->lock
);
341 if (IS_ENABLED(CONFIG_IPV6
) && sk
->sk_reuseport
&&
342 sk
->sk_family
== AF_INET6
)
343 hlist_add_tail_rcu(&udp_sk(sk
)->udp_portaddr_node
,
346 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
349 spin_unlock(&hslot2
->lock
);
351 sock_set_flag(sk
, SOCK_RCU_FREE
);
354 spin_unlock_bh(&hslot
->lock
);
358 EXPORT_SYMBOL(udp_lib_get_port
);
360 static u32
udp4_portaddr_hash(const struct net
*net
, __be32 saddr
,
363 return jhash_1word((__force u32
)saddr
, net_hash_mix(net
)) ^ port
;
366 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
368 unsigned int hash2_nulladdr
=
369 udp4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
370 unsigned int hash2_partial
=
371 udp4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
373 /* precompute partial secondary hash */
374 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
375 return udp_lib_get_port(sk
, snum
, hash2_nulladdr
);
378 static int compute_score(struct sock
*sk
, struct net
*net
,
379 __be32 saddr
, __be16 sport
,
380 __be32 daddr
, unsigned short hnum
,
381 int dif
, int sdif
, bool exact_dif
)
384 struct inet_sock
*inet
;
386 if (!net_eq(sock_net(sk
), net
) ||
387 udp_sk(sk
)->udp_port_hash
!= hnum
||
391 score
= (sk
->sk_family
== PF_INET
) ? 2 : 1;
394 if (inet
->inet_rcv_saddr
) {
395 if (inet
->inet_rcv_saddr
!= daddr
)
400 if (inet
->inet_daddr
) {
401 if (inet
->inet_daddr
!= saddr
)
406 if (inet
->inet_dport
) {
407 if (inet
->inet_dport
!= sport
)
412 if (sk
->sk_bound_dev_if
|| exact_dif
) {
413 bool dev_match
= (sk
->sk_bound_dev_if
== dif
||
414 sk
->sk_bound_dev_if
== sdif
);
416 if (exact_dif
&& !dev_match
)
418 if (sk
->sk_bound_dev_if
&& dev_match
)
422 if (sk
->sk_incoming_cpu
== raw_smp_processor_id())
427 static u32
udp_ehashfn(const struct net
*net
, const __be32 laddr
,
428 const __u16 lport
, const __be32 faddr
,
431 static u32 udp_ehash_secret __read_mostly
;
433 net_get_random_once(&udp_ehash_secret
, sizeof(udp_ehash_secret
));
435 return __inet_ehashfn(laddr
, lport
, faddr
, fport
,
436 udp_ehash_secret
+ net_hash_mix(net
));
439 /* called with rcu_read_lock() */
440 static struct sock
*udp4_lib_lookup2(struct net
*net
,
441 __be32 saddr
, __be16 sport
,
442 __be32 daddr
, unsigned int hnum
,
443 int dif
, int sdif
, bool exact_dif
,
444 struct udp_hslot
*hslot2
,
447 struct sock
*sk
, *result
;
448 int score
, badness
, matches
= 0, reuseport
= 0;
453 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
454 score
= compute_score(sk
, net
, saddr
, sport
,
455 daddr
, hnum
, dif
, sdif
, exact_dif
);
456 if (score
> badness
) {
457 reuseport
= sk
->sk_reuseport
;
459 hash
= udp_ehashfn(net
, daddr
, hnum
,
461 result
= reuseport_select_sock(sk
, hash
, skb
,
462 sizeof(struct udphdr
));
469 } else if (score
== badness
&& reuseport
) {
471 if (reciprocal_scale(hash
, matches
) == 0)
473 hash
= next_pseudo_random32(hash
);
479 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
480 * harder than this. -DaveM
482 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
483 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
,
484 int sdif
, struct udp_table
*udptable
, struct sk_buff
*skb
)
486 struct sock
*sk
, *result
;
487 unsigned short hnum
= ntohs(dport
);
488 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
489 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
490 bool exact_dif
= udp_lib_exact_dif_match(net
, skb
);
491 int score
, badness
, matches
= 0, reuseport
= 0;
494 if (hslot
->count
> 10) {
495 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
);
496 slot2
= hash2
& udptable
->mask
;
497 hslot2
= &udptable
->hash2
[slot2
];
498 if (hslot
->count
< hslot2
->count
)
501 result
= udp4_lib_lookup2(net
, saddr
, sport
,
502 daddr
, hnum
, dif
, sdif
,
503 exact_dif
, hslot2
, skb
);
505 unsigned int old_slot2
= slot2
;
506 hash2
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
507 slot2
= hash2
& udptable
->mask
;
508 /* avoid searching the same slot again. */
509 if (unlikely(slot2
== old_slot2
))
512 hslot2
= &udptable
->hash2
[slot2
];
513 if (hslot
->count
< hslot2
->count
)
516 result
= udp4_lib_lookup2(net
, saddr
, sport
,
517 daddr
, hnum
, dif
, sdif
,
518 exact_dif
, hslot2
, skb
);
525 sk_for_each_rcu(sk
, &hslot
->head
) {
526 score
= compute_score(sk
, net
, saddr
, sport
,
527 daddr
, hnum
, dif
, sdif
, exact_dif
);
528 if (score
> badness
) {
529 reuseport
= sk
->sk_reuseport
;
531 hash
= udp_ehashfn(net
, daddr
, hnum
,
533 result
= reuseport_select_sock(sk
, hash
, skb
,
534 sizeof(struct udphdr
));
541 } else if (score
== badness
&& reuseport
) {
543 if (reciprocal_scale(hash
, matches
) == 0)
545 hash
= next_pseudo_random32(hash
);
550 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
552 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
553 __be16 sport
, __be16 dport
,
554 struct udp_table
*udptable
)
556 const struct iphdr
*iph
= ip_hdr(skb
);
558 return __udp4_lib_lookup(dev_net(skb
->dev
), iph
->saddr
, sport
,
559 iph
->daddr
, dport
, inet_iif(skb
),
560 inet_sdif(skb
), udptable
, skb
);
563 struct sock
*udp4_lib_lookup_skb(struct sk_buff
*skb
,
564 __be16 sport
, __be16 dport
)
566 return __udp4_lib_lookup_skb(skb
, sport
, dport
, &udp_table
);
568 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb
);
570 /* Must be called under rcu_read_lock().
571 * Does increment socket refcount.
573 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
574 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
575 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
576 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
577 __be32 daddr
, __be16 dport
, int dif
)
581 sk
= __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
,
582 dif
, 0, &udp_table
, NULL
);
583 if (sk
&& !refcount_inc_not_zero(&sk
->sk_refcnt
))
587 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
590 static inline bool __udp_is_mcast_sock(struct net
*net
, struct sock
*sk
,
591 __be16 loc_port
, __be32 loc_addr
,
592 __be16 rmt_port
, __be32 rmt_addr
,
593 int dif
, int sdif
, unsigned short hnum
)
595 struct inet_sock
*inet
= inet_sk(sk
);
597 if (!net_eq(sock_net(sk
), net
) ||
598 udp_sk(sk
)->udp_port_hash
!= hnum
||
599 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
600 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
601 (inet
->inet_rcv_saddr
&& inet
->inet_rcv_saddr
!= loc_addr
) ||
602 ipv6_only_sock(sk
) ||
603 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
&&
604 sk
->sk_bound_dev_if
!= sdif
))
606 if (!ip_mc_sf_allow(sk
, loc_addr
, rmt_addr
, dif
, sdif
))
612 * This routine is called by the ICMP module when it gets some
613 * sort of error condition. If err < 0 then the socket should
614 * be closed and the error returned to the user. If err > 0
615 * it's just the icmp type << 8 | icmp code.
616 * Header points to the ip header of the error packet. We move
617 * on past this. Then (as it used to claim before adjustment)
618 * header points to the first 8 bytes of the udp header. We need
619 * to find the appropriate port.
622 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
624 struct inet_sock
*inet
;
625 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
626 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
627 const int type
= icmp_hdr(skb
)->type
;
628 const int code
= icmp_hdr(skb
)->code
;
632 struct net
*net
= dev_net(skb
->dev
);
634 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
635 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, 0,
638 __ICMP_INC_STATS(net
, ICMP_MIB_INERRORS
);
639 return; /* No socket for error */
648 case ICMP_TIME_EXCEEDED
:
651 case ICMP_SOURCE_QUENCH
:
653 case ICMP_PARAMETERPROB
:
657 case ICMP_DEST_UNREACH
:
658 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
659 ipv4_sk_update_pmtu(skb
, sk
, info
);
660 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
668 if (code
<= NR_ICMP_UNREACH
) {
669 harderr
= icmp_err_convert
[code
].fatal
;
670 err
= icmp_err_convert
[code
].errno
;
674 ipv4_sk_redirect(skb
, sk
);
679 * RFC1122: OK. Passes ICMP errors back to application, as per
682 if (!inet
->recverr
) {
683 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
686 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
689 sk
->sk_error_report(sk
);
694 void udp_err(struct sk_buff
*skb
, u32 info
)
696 __udp4_lib_err(skb
, info
, &udp_table
);
700 * Throw away all pending data and cancel the corking. Socket is locked.
702 void udp_flush_pending_frames(struct sock
*sk
)
704 struct udp_sock
*up
= udp_sk(sk
);
709 ip_flush_pending_frames(sk
);
712 EXPORT_SYMBOL(udp_flush_pending_frames
);
715 * udp4_hwcsum - handle outgoing HW checksumming
716 * @skb: sk_buff containing the filled-in UDP header
717 * (checksum field must be zeroed out)
718 * @src: source IP address
719 * @dst: destination IP address
721 void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
723 struct udphdr
*uh
= udp_hdr(skb
);
724 int offset
= skb_transport_offset(skb
);
725 int len
= skb
->len
- offset
;
729 if (!skb_has_frag_list(skb
)) {
731 * Only one fragment on the socket.
733 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
734 skb
->csum_offset
= offsetof(struct udphdr
, check
);
735 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
738 struct sk_buff
*frags
;
741 * HW-checksum won't work as there are two or more
742 * fragments on the socket so that all csums of sk_buffs
745 skb_walk_frags(skb
, frags
) {
746 csum
= csum_add(csum
, frags
->csum
);
750 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
751 skb
->ip_summed
= CHECKSUM_NONE
;
753 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
755 uh
->check
= CSUM_MANGLED_0
;
758 EXPORT_SYMBOL_GPL(udp4_hwcsum
);
760 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
761 * for the simple case like when setting the checksum for a UDP tunnel.
763 void udp_set_csum(bool nocheck
, struct sk_buff
*skb
,
764 __be32 saddr
, __be32 daddr
, int len
)
766 struct udphdr
*uh
= udp_hdr(skb
);
770 } else if (skb_is_gso(skb
)) {
771 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
772 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
774 uh
->check
= udp_v4_check(len
, saddr
, daddr
, lco_csum(skb
));
776 uh
->check
= CSUM_MANGLED_0
;
778 skb
->ip_summed
= CHECKSUM_PARTIAL
;
779 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
780 skb
->csum_offset
= offsetof(struct udphdr
, check
);
781 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
784 EXPORT_SYMBOL(udp_set_csum
);
786 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
)
788 struct sock
*sk
= skb
->sk
;
789 struct inet_sock
*inet
= inet_sk(sk
);
792 int is_udplite
= IS_UDPLITE(sk
);
793 int offset
= skb_transport_offset(skb
);
794 int len
= skb
->len
- offset
;
798 * Create a UDP header
801 uh
->source
= inet
->inet_sport
;
802 uh
->dest
= fl4
->fl4_dport
;
803 uh
->len
= htons(len
);
806 if (is_udplite
) /* UDP-Lite */
807 csum
= udplite_csum(skb
);
809 else if (sk
->sk_no_check_tx
) { /* UDP csum off */
811 skb
->ip_summed
= CHECKSUM_NONE
;
814 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
816 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
820 csum
= udp_csum(skb
);
822 /* add protocol-dependent pseudo-header */
823 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
824 sk
->sk_protocol
, csum
);
826 uh
->check
= CSUM_MANGLED_0
;
829 err
= ip_send_skb(sock_net(sk
), skb
);
831 if (err
== -ENOBUFS
&& !inet
->recverr
) {
832 UDP_INC_STATS(sock_net(sk
),
833 UDP_MIB_SNDBUFERRORS
, is_udplite
);
837 UDP_INC_STATS(sock_net(sk
),
838 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
843 * Push out all pending data as one UDP datagram. Socket is locked.
845 int udp_push_pending_frames(struct sock
*sk
)
847 struct udp_sock
*up
= udp_sk(sk
);
848 struct inet_sock
*inet
= inet_sk(sk
);
849 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
853 skb
= ip_finish_skb(sk
, fl4
);
857 err
= udp_send_skb(skb
, fl4
);
864 EXPORT_SYMBOL(udp_push_pending_frames
);
866 int udp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
868 struct inet_sock
*inet
= inet_sk(sk
);
869 struct udp_sock
*up
= udp_sk(sk
);
870 struct flowi4 fl4_stack
;
873 struct ipcm_cookie ipc
;
874 struct rtable
*rt
= NULL
;
877 __be32 daddr
, faddr
, saddr
;
880 int err
, is_udplite
= IS_UDPLITE(sk
);
881 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
882 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
884 struct ip_options_data opt_copy
;
893 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
901 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
903 fl4
= &inet
->cork
.fl
.u
.ip4
;
906 * There are pending frames.
907 * The socket lock must be held while it's corked.
910 if (likely(up
->pending
)) {
911 if (unlikely(up
->pending
!= AF_INET
)) {
919 ulen
+= sizeof(struct udphdr
);
922 * Get and verify the address.
925 DECLARE_SOCKADDR(struct sockaddr_in
*, usin
, msg
->msg_name
);
926 if (msg
->msg_namelen
< sizeof(*usin
))
928 if (usin
->sin_family
!= AF_INET
) {
929 if (usin
->sin_family
!= AF_UNSPEC
)
930 return -EAFNOSUPPORT
;
933 daddr
= usin
->sin_addr
.s_addr
;
934 dport
= usin
->sin_port
;
938 if (sk
->sk_state
!= TCP_ESTABLISHED
)
939 return -EDESTADDRREQ
;
940 daddr
= inet
->inet_daddr
;
941 dport
= inet
->inet_dport
;
942 /* Open fast path for connected socket.
943 Route will not be used, if at least one option is set.
948 ipc
.sockc
.tsflags
= sk
->sk_tsflags
;
949 ipc
.addr
= inet
->inet_saddr
;
950 ipc
.oif
= sk
->sk_bound_dev_if
;
952 if (msg
->msg_controllen
) {
953 err
= ip_cmsg_send(sk
, msg
, &ipc
, sk
->sk_family
== AF_INET6
);
963 struct ip_options_rcu
*inet_opt
;
966 inet_opt
= rcu_dereference(inet
->inet_opt
);
968 memcpy(&opt_copy
, inet_opt
,
969 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
970 ipc
.opt
= &opt_copy
.opt
;
976 ipc
.addr
= faddr
= daddr
;
978 sock_tx_timestamp(sk
, ipc
.sockc
.tsflags
, &ipc
.tx_flags
);
980 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
983 faddr
= ipc
.opt
->opt
.faddr
;
986 tos
= get_rttos(&ipc
, inet
);
987 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
988 (msg
->msg_flags
& MSG_DONTROUTE
) ||
989 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
994 if (ipv4_is_multicast(daddr
)) {
996 ipc
.oif
= inet
->mc_index
;
998 saddr
= inet
->mc_addr
;
1000 } else if (!ipc
.oif
)
1001 ipc
.oif
= inet
->uc_index
;
1004 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
1007 struct net
*net
= sock_net(sk
);
1008 __u8 flow_flags
= inet_sk_flowi_flags(sk
);
1012 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
1013 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
1015 faddr
, saddr
, dport
, inet
->inet_sport
,
1018 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
1019 rt
= ip_route_output_flow(net
, fl4
, sk
);
1023 if (err
== -ENETUNREACH
)
1024 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
1029 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
1030 !sock_flag(sk
, SOCK_BROADCAST
))
1033 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1036 if (msg
->msg_flags
&MSG_CONFIRM
)
1042 daddr
= ipc
.addr
= fl4
->daddr
;
1044 /* Lockless fast path for the non-corking case. */
1046 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
, ulen
,
1047 sizeof(struct udphdr
), &ipc
, &rt
,
1050 if (!IS_ERR_OR_NULL(skb
))
1051 err
= udp_send_skb(skb
, fl4
);
1056 if (unlikely(up
->pending
)) {
1057 /* The socket is already corked while preparing it. */
1058 /* ... which is an evident application bug. --ANK */
1061 net_dbg_ratelimited("socket already corked\n");
1066 * Now cork the socket to pend data.
1068 fl4
= &inet
->cork
.fl
.u
.ip4
;
1071 fl4
->fl4_dport
= dport
;
1072 fl4
->fl4_sport
= inet
->inet_sport
;
1073 up
->pending
= AF_INET
;
1077 err
= ip_append_data(sk
, fl4
, getfrag
, msg
, ulen
,
1078 sizeof(struct udphdr
), &ipc
, &rt
,
1079 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1081 udp_flush_pending_frames(sk
);
1083 err
= udp_push_pending_frames(sk
);
1084 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1095 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1096 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1097 * we don't have a good statistic (IpOutDiscards but it can be too many
1098 * things). We could add another new stat but at least for now that
1099 * seems like overkill.
1101 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1102 UDP_INC_STATS(sock_net(sk
),
1103 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1108 if (msg
->msg_flags
& MSG_PROBE
)
1109 dst_confirm_neigh(&rt
->dst
, &fl4
->daddr
);
1110 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1111 goto back_from_confirm
;
1115 EXPORT_SYMBOL(udp_sendmsg
);
1117 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1118 size_t size
, int flags
)
1120 struct inet_sock
*inet
= inet_sk(sk
);
1121 struct udp_sock
*up
= udp_sk(sk
);
1124 if (flags
& MSG_SENDPAGE_NOTLAST
)
1128 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1130 /* Call udp_sendmsg to specify destination address which
1131 * sendpage interface can't pass.
1132 * This will succeed only when the socket is connected.
1134 ret
= udp_sendmsg(sk
, &msg
, 0);
1141 if (unlikely(!up
->pending
)) {
1144 net_dbg_ratelimited("cork failed\n");
1148 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1149 page
, offset
, size
, flags
);
1150 if (ret
== -EOPNOTSUPP
) {
1152 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1156 udp_flush_pending_frames(sk
);
1161 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1162 ret
= udp_push_pending_frames(sk
);
1170 #define UDP_SKB_IS_STATELESS 0x80000000
1172 static void udp_set_dev_scratch(struct sk_buff
*skb
)
1174 struct udp_dev_scratch
*scratch
= udp_skb_scratch(skb
);
1176 BUILD_BUG_ON(sizeof(struct udp_dev_scratch
) > sizeof(long));
1177 scratch
->_tsize_state
= skb
->truesize
;
1178 #if BITS_PER_LONG == 64
1179 scratch
->len
= skb
->len
;
1180 scratch
->csum_unnecessary
= !!skb_csum_unnecessary(skb
);
1181 scratch
->is_linear
= !skb_is_nonlinear(skb
);
1183 /* all head states execept sp (dst, sk, nf) are always cleared by
1184 * udp_rcv() and we need to preserve secpath, if present, to eventually
1185 * process IP_CMSG_PASSSEC at recvmsg() time
1187 if (likely(!skb_sec_path(skb
)))
1188 scratch
->_tsize_state
|= UDP_SKB_IS_STATELESS
;
1191 static int udp_skb_truesize(struct sk_buff
*skb
)
1193 return udp_skb_scratch(skb
)->_tsize_state
& ~UDP_SKB_IS_STATELESS
;
1196 static bool udp_skb_has_head_state(struct sk_buff
*skb
)
1198 return !(udp_skb_scratch(skb
)->_tsize_state
& UDP_SKB_IS_STATELESS
);
1201 /* fully reclaim rmem/fwd memory allocated for skb */
1202 static void udp_rmem_release(struct sock
*sk
, int size
, int partial
,
1203 bool rx_queue_lock_held
)
1205 struct udp_sock
*up
= udp_sk(sk
);
1206 struct sk_buff_head
*sk_queue
;
1209 if (likely(partial
)) {
1210 up
->forward_deficit
+= size
;
1211 size
= up
->forward_deficit
;
1212 if (size
< (sk
->sk_rcvbuf
>> 2) &&
1213 !skb_queue_empty(&up
->reader_queue
))
1216 size
+= up
->forward_deficit
;
1218 up
->forward_deficit
= 0;
1220 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1221 * if the called don't held it already
1223 sk_queue
= &sk
->sk_receive_queue
;
1224 if (!rx_queue_lock_held
)
1225 spin_lock(&sk_queue
->lock
);
1228 sk
->sk_forward_alloc
+= size
;
1229 amt
= (sk
->sk_forward_alloc
- partial
) & ~(SK_MEM_QUANTUM
- 1);
1230 sk
->sk_forward_alloc
-= amt
;
1233 __sk_mem_reduce_allocated(sk
, amt
>> SK_MEM_QUANTUM_SHIFT
);
1235 atomic_sub(size
, &sk
->sk_rmem_alloc
);
1237 /* this can save us from acquiring the rx queue lock on next receive */
1238 skb_queue_splice_tail_init(sk_queue
, &up
->reader_queue
);
1240 if (!rx_queue_lock_held
)
1241 spin_unlock(&sk_queue
->lock
);
1244 /* Note: called with reader_queue.lock held.
1245 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1246 * This avoids a cache line miss while receive_queue lock is held.
1247 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1249 void udp_skb_destructor(struct sock
*sk
, struct sk_buff
*skb
)
1251 prefetch(&skb
->data
);
1252 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, false);
1254 EXPORT_SYMBOL(udp_skb_destructor
);
1256 /* as above, but the caller held the rx queue lock, too */
1257 static void udp_skb_dtor_locked(struct sock
*sk
, struct sk_buff
*skb
)
1259 prefetch(&skb
->data
);
1260 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, true);
1263 /* Idea of busylocks is to let producers grab an extra spinlock
1264 * to relieve pressure on the receive_queue spinlock shared by consumer.
1265 * Under flood, this means that only one producer can be in line
1266 * trying to acquire the receive_queue spinlock.
1267 * These busylock can be allocated on a per cpu manner, instead of a
1268 * per socket one (that would consume a cache line per socket)
1270 static int udp_busylocks_log __read_mostly
;
1271 static spinlock_t
*udp_busylocks __read_mostly
;
1273 static spinlock_t
*busylock_acquire(void *ptr
)
1277 busy
= udp_busylocks
+ hash_ptr(ptr
, udp_busylocks_log
);
1282 static void busylock_release(spinlock_t
*busy
)
1288 int __udp_enqueue_schedule_skb(struct sock
*sk
, struct sk_buff
*skb
)
1290 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
1291 int rmem
, delta
, amt
, err
= -ENOMEM
;
1292 spinlock_t
*busy
= NULL
;
1295 /* try to avoid the costly atomic add/sub pair when the receive
1296 * queue is full; always allow at least a packet
1298 rmem
= atomic_read(&sk
->sk_rmem_alloc
);
1299 if (rmem
> sk
->sk_rcvbuf
)
1302 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1303 * having linear skbs :
1304 * - Reduce memory overhead and thus increase receive queue capacity
1305 * - Less cache line misses at copyout() time
1306 * - Less work at consume_skb() (less alien page frag freeing)
1308 if (rmem
> (sk
->sk_rcvbuf
>> 1)) {
1311 busy
= busylock_acquire(sk
);
1313 size
= skb
->truesize
;
1314 udp_set_dev_scratch(skb
);
1316 /* we drop only if the receive buf is full and the receive
1317 * queue contains some other skb
1319 rmem
= atomic_add_return(size
, &sk
->sk_rmem_alloc
);
1320 if (rmem
> (size
+ sk
->sk_rcvbuf
))
1323 spin_lock(&list
->lock
);
1324 if (size
>= sk
->sk_forward_alloc
) {
1325 amt
= sk_mem_pages(size
);
1326 delta
= amt
<< SK_MEM_QUANTUM_SHIFT
;
1327 if (!__sk_mem_raise_allocated(sk
, delta
, amt
, SK_MEM_RECV
)) {
1329 spin_unlock(&list
->lock
);
1333 sk
->sk_forward_alloc
+= delta
;
1336 sk
->sk_forward_alloc
-= size
;
1338 /* no need to setup a destructor, we will explicitly release the
1339 * forward allocated memory on dequeue
1341 sock_skb_set_dropcount(sk
, skb
);
1343 __skb_queue_tail(list
, skb
);
1344 spin_unlock(&list
->lock
);
1346 if (!sock_flag(sk
, SOCK_DEAD
))
1347 sk
->sk_data_ready(sk
);
1349 busylock_release(busy
);
1353 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1356 atomic_inc(&sk
->sk_drops
);
1357 busylock_release(busy
);
1360 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb
);
1362 void udp_destruct_sock(struct sock
*sk
)
1364 /* reclaim completely the forward allocated memory */
1365 struct udp_sock
*up
= udp_sk(sk
);
1366 unsigned int total
= 0;
1367 struct sk_buff
*skb
;
1369 skb_queue_splice_tail_init(&sk
->sk_receive_queue
, &up
->reader_queue
);
1370 while ((skb
= __skb_dequeue(&up
->reader_queue
)) != NULL
) {
1371 total
+= skb
->truesize
;
1374 udp_rmem_release(sk
, total
, 0, true);
1376 inet_sock_destruct(sk
);
1378 EXPORT_SYMBOL_GPL(udp_destruct_sock
);
1380 int udp_init_sock(struct sock
*sk
)
1382 skb_queue_head_init(&udp_sk(sk
)->reader_queue
);
1383 sk
->sk_destruct
= udp_destruct_sock
;
1386 EXPORT_SYMBOL_GPL(udp_init_sock
);
1388 void skb_consume_udp(struct sock
*sk
, struct sk_buff
*skb
, int len
)
1390 if (unlikely(READ_ONCE(sk
->sk_peek_off
) >= 0)) {
1391 bool slow
= lock_sock_fast(sk
);
1393 sk_peek_offset_bwd(sk
, len
);
1394 unlock_sock_fast(sk
, slow
);
1397 if (!skb_unref(skb
))
1400 /* In the more common cases we cleared the head states previously,
1401 * see __udp_queue_rcv_skb().
1403 if (unlikely(udp_skb_has_head_state(skb
)))
1404 skb_release_head_state(skb
);
1405 __consume_stateless_skb(skb
);
1407 EXPORT_SYMBOL_GPL(skb_consume_udp
);
1409 static struct sk_buff
*__first_packet_length(struct sock
*sk
,
1410 struct sk_buff_head
*rcvq
,
1413 struct sk_buff
*skb
;
1415 while ((skb
= skb_peek(rcvq
)) != NULL
) {
1416 if (udp_lib_checksum_complete(skb
)) {
1417 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1419 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
1421 atomic_inc(&sk
->sk_drops
);
1422 __skb_unlink(skb
, rcvq
);
1423 *total
+= skb
->truesize
;
1426 /* the csum related bits could be changed, refresh
1429 udp_set_dev_scratch(skb
);
1437 * first_packet_length - return length of first packet in receive queue
1440 * Drops all bad checksum frames, until a valid one is found.
1441 * Returns the length of found skb, or -1 if none is found.
1443 static int first_packet_length(struct sock
*sk
)
1445 struct sk_buff_head
*rcvq
= &udp_sk(sk
)->reader_queue
;
1446 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1447 struct sk_buff
*skb
;
1451 spin_lock_bh(&rcvq
->lock
);
1452 skb
= __first_packet_length(sk
, rcvq
, &total
);
1453 if (!skb
&& !skb_queue_empty(sk_queue
)) {
1454 spin_lock(&sk_queue
->lock
);
1455 skb_queue_splice_tail_init(sk_queue
, rcvq
);
1456 spin_unlock(&sk_queue
->lock
);
1458 skb
= __first_packet_length(sk
, rcvq
, &total
);
1460 res
= skb
? skb
->len
: -1;
1462 udp_rmem_release(sk
, total
, 1, false);
1463 spin_unlock_bh(&rcvq
->lock
);
1468 * IOCTL requests applicable to the UDP protocol
1471 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1476 int amount
= sk_wmem_alloc_get(sk
);
1478 return put_user(amount
, (int __user
*)arg
);
1483 int amount
= max_t(int, 0, first_packet_length(sk
));
1485 return put_user(amount
, (int __user
*)arg
);
1489 return -ENOIOCTLCMD
;
1494 EXPORT_SYMBOL(udp_ioctl
);
1496 struct sk_buff
*__skb_recv_udp(struct sock
*sk
, unsigned int flags
,
1497 int noblock
, int *peeked
, int *off
, int *err
)
1499 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1500 struct sk_buff_head
*queue
;
1501 struct sk_buff
*last
;
1505 queue
= &udp_sk(sk
)->reader_queue
;
1506 flags
|= noblock
? MSG_DONTWAIT
: 0;
1507 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
1509 struct sk_buff
*skb
;
1511 error
= sock_error(sk
);
1518 spin_lock_bh(&queue
->lock
);
1519 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
,
1524 spin_unlock_bh(&queue
->lock
);
1528 if (skb_queue_empty(sk_queue
)) {
1529 spin_unlock_bh(&queue
->lock
);
1533 /* refill the reader queue and walk it again
1534 * keep both queues locked to avoid re-acquiring
1535 * the sk_receive_queue lock if fwd memory scheduling
1538 spin_lock(&sk_queue
->lock
);
1539 skb_queue_splice_tail_init(sk_queue
, queue
);
1541 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
,
1542 udp_skb_dtor_locked
,
1545 spin_unlock(&sk_queue
->lock
);
1546 spin_unlock_bh(&queue
->lock
);
1551 if (!sk_can_busy_loop(sk
))
1554 sk_busy_loop(sk
, flags
& MSG_DONTWAIT
);
1555 } while (!skb_queue_empty(sk_queue
));
1557 /* sk_queue is empty, reader_queue may contain peeked packets */
1559 !__skb_wait_for_more_packets(sk
, &error
, &timeo
,
1560 (struct sk_buff
*)sk_queue
));
1565 EXPORT_SYMBOL_GPL(__skb_recv_udp
);
1568 * This should be easy, if there is something there we
1569 * return it, otherwise we block.
1572 int udp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int noblock
,
1573 int flags
, int *addr_len
)
1575 struct inet_sock
*inet
= inet_sk(sk
);
1576 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, msg
->msg_name
);
1577 struct sk_buff
*skb
;
1578 unsigned int ulen
, copied
;
1579 int peeked
, peeking
, off
;
1581 int is_udplite
= IS_UDPLITE(sk
);
1582 bool checksum_valid
= false;
1584 if (flags
& MSG_ERRQUEUE
)
1585 return ip_recv_error(sk
, msg
, len
, addr_len
);
1588 peeking
= flags
& MSG_PEEK
;
1589 off
= sk_peek_offset(sk
, flags
);
1590 skb
= __skb_recv_udp(sk
, flags
, noblock
, &peeked
, &off
, &err
);
1594 ulen
= udp_skb_len(skb
);
1596 if (copied
> ulen
- off
)
1597 copied
= ulen
- off
;
1598 else if (copied
< ulen
)
1599 msg
->msg_flags
|= MSG_TRUNC
;
1602 * If checksum is needed at all, try to do it while copying the
1603 * data. If the data is truncated, or if we only want a partial
1604 * coverage checksum (UDP-Lite), do it before the copy.
1607 if (copied
< ulen
|| peeking
||
1608 (is_udplite
&& UDP_SKB_CB(skb
)->partial_cov
)) {
1609 checksum_valid
= udp_skb_csum_unnecessary(skb
) ||
1610 !__udp_lib_checksum_complete(skb
);
1611 if (!checksum_valid
)
1615 if (checksum_valid
|| udp_skb_csum_unnecessary(skb
)) {
1616 if (udp_skb_is_linear(skb
))
1617 err
= copy_linear_skb(skb
, copied
, off
, &msg
->msg_iter
);
1619 err
= skb_copy_datagram_msg(skb
, off
, msg
, copied
);
1621 err
= skb_copy_and_csum_datagram_msg(skb
, off
, msg
);
1627 if (unlikely(err
)) {
1629 atomic_inc(&sk
->sk_drops
);
1630 UDP_INC_STATS(sock_net(sk
),
1631 UDP_MIB_INERRORS
, is_udplite
);
1638 UDP_INC_STATS(sock_net(sk
),
1639 UDP_MIB_INDATAGRAMS
, is_udplite
);
1641 sock_recv_ts_and_drops(msg
, sk
, skb
);
1643 /* Copy the address. */
1645 sin
->sin_family
= AF_INET
;
1646 sin
->sin_port
= udp_hdr(skb
)->source
;
1647 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1648 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1649 *addr_len
= sizeof(*sin
);
1651 if (inet
->cmsg_flags
)
1652 ip_cmsg_recv_offset(msg
, sk
, skb
, sizeof(struct udphdr
), off
);
1655 if (flags
& MSG_TRUNC
)
1658 skb_consume_udp(sk
, skb
, peeking
? -err
: err
);
1662 if (!__sk_queue_drop_skb(sk
, &udp_sk(sk
)->reader_queue
, skb
, flags
,
1663 udp_skb_destructor
)) {
1664 UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1665 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1669 /* starting over for a new packet, but check if we need to yield */
1671 msg
->msg_flags
&= ~MSG_TRUNC
;
1675 int __udp_disconnect(struct sock
*sk
, int flags
)
1677 struct inet_sock
*inet
= inet_sk(sk
);
1679 * 1003.1g - break association.
1682 sk
->sk_state
= TCP_CLOSE
;
1683 inet
->inet_daddr
= 0;
1684 inet
->inet_dport
= 0;
1685 sock_rps_reset_rxhash(sk
);
1686 sk
->sk_bound_dev_if
= 0;
1687 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1688 inet_reset_saddr(sk
);
1690 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1691 sk
->sk_prot
->unhash(sk
);
1692 inet
->inet_sport
= 0;
1697 EXPORT_SYMBOL(__udp_disconnect
);
1699 int udp_disconnect(struct sock
*sk
, int flags
)
1702 __udp_disconnect(sk
, flags
);
1706 EXPORT_SYMBOL(udp_disconnect
);
1708 void udp_lib_unhash(struct sock
*sk
)
1710 if (sk_hashed(sk
)) {
1711 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1712 struct udp_hslot
*hslot
, *hslot2
;
1714 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1715 udp_sk(sk
)->udp_port_hash
);
1716 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1718 spin_lock_bh(&hslot
->lock
);
1719 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1720 reuseport_detach_sock(sk
);
1721 if (sk_del_node_init_rcu(sk
)) {
1723 inet_sk(sk
)->inet_num
= 0;
1724 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1726 spin_lock(&hslot2
->lock
);
1727 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1729 spin_unlock(&hslot2
->lock
);
1731 spin_unlock_bh(&hslot
->lock
);
1734 EXPORT_SYMBOL(udp_lib_unhash
);
1737 * inet_rcv_saddr was changed, we must rehash secondary hash
1739 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1741 if (sk_hashed(sk
)) {
1742 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1743 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1745 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1746 nhslot2
= udp_hashslot2(udptable
, newhash
);
1747 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1749 if (hslot2
!= nhslot2
||
1750 rcu_access_pointer(sk
->sk_reuseport_cb
)) {
1751 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1752 udp_sk(sk
)->udp_port_hash
);
1753 /* we must lock primary chain too */
1754 spin_lock_bh(&hslot
->lock
);
1755 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1756 reuseport_detach_sock(sk
);
1758 if (hslot2
!= nhslot2
) {
1759 spin_lock(&hslot2
->lock
);
1760 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1762 spin_unlock(&hslot2
->lock
);
1764 spin_lock(&nhslot2
->lock
);
1765 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1768 spin_unlock(&nhslot2
->lock
);
1771 spin_unlock_bh(&hslot
->lock
);
1775 EXPORT_SYMBOL(udp_lib_rehash
);
1777 static void udp_v4_rehash(struct sock
*sk
)
1779 u16 new_hash
= udp4_portaddr_hash(sock_net(sk
),
1780 inet_sk(sk
)->inet_rcv_saddr
,
1781 inet_sk(sk
)->inet_num
);
1782 udp_lib_rehash(sk
, new_hash
);
1785 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1789 if (inet_sk(sk
)->inet_daddr
) {
1790 sock_rps_save_rxhash(sk
, skb
);
1791 sk_mark_napi_id(sk
, skb
);
1792 sk_incoming_cpu_update(sk
);
1794 sk_mark_napi_id_once(sk
, skb
);
1797 rc
= __udp_enqueue_schedule_skb(sk
, skb
);
1799 int is_udplite
= IS_UDPLITE(sk
);
1801 /* Note that an ENOMEM error is charged twice */
1803 UDP_INC_STATS(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1805 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1807 trace_udp_fail_queue_rcv_skb(rc
, sk
);
1814 static struct static_key udp_encap_needed __read_mostly
;
1815 void udp_encap_enable(void)
1817 static_key_enable(&udp_encap_needed
);
1819 EXPORT_SYMBOL(udp_encap_enable
);
1824 * >0: "udp encap" protocol resubmission
1826 * Note that in the success and error cases, the skb is assumed to
1827 * have either been requeued or freed.
1829 static int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1831 struct udp_sock
*up
= udp_sk(sk
);
1832 int is_udplite
= IS_UDPLITE(sk
);
1835 * Charge it to the socket, dropping if the queue is full.
1837 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1841 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
1842 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
1845 * This is an encapsulation socket so pass the skb to
1846 * the socket's udp_encap_rcv() hook. Otherwise, just
1847 * fall through and pass this up the UDP socket.
1848 * up->encap_rcv() returns the following value:
1849 * =0 if skb was successfully passed to the encap
1850 * handler or was discarded by it.
1851 * >0 if skb should be passed on to UDP.
1852 * <0 if skb should be resubmitted as proto -N
1855 /* if we're overly short, let UDP handle it */
1856 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
1860 /* Verify checksum before giving to encap */
1861 if (udp_lib_checksum_complete(skb
))
1864 ret
= encap_rcv(sk
, skb
);
1866 __UDP_INC_STATS(sock_net(sk
),
1867 UDP_MIB_INDATAGRAMS
,
1873 /* FALLTHROUGH -- it's a UDP Packet */
1877 * UDP-Lite specific tests, ignored on UDP sockets
1879 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1882 * MIB statistics other than incrementing the error count are
1883 * disabled for the following two types of errors: these depend
1884 * on the application settings, not on the functioning of the
1885 * protocol stack as such.
1887 * RFC 3828 here recommends (sec 3.3): "There should also be a
1888 * way ... to ... at least let the receiving application block
1889 * delivery of packets with coverage values less than a value
1890 * provided by the application."
1892 if (up
->pcrlen
== 0) { /* full coverage was set */
1893 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1894 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1897 /* The next case involves violating the min. coverage requested
1898 * by the receiver. This is subtle: if receiver wants x and x is
1899 * greater than the buffersize/MTU then receiver will complain
1900 * that it wants x while sender emits packets of smaller size y.
1901 * Therefore the above ...()->partial_cov statement is essential.
1903 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1904 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1905 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1910 prefetch(&sk
->sk_rmem_alloc
);
1911 if (rcu_access_pointer(sk
->sk_filter
) &&
1912 udp_lib_checksum_complete(skb
))
1915 if (sk_filter_trim_cap(sk
, skb
, sizeof(struct udphdr
)))
1918 udp_csum_pull_header(skb
);
1920 ipv4_pktinfo_prepare(sk
, skb
);
1921 return __udp_queue_rcv_skb(sk
, skb
);
1924 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1926 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1927 atomic_inc(&sk
->sk_drops
);
1932 /* For TCP sockets, sk_rx_dst is protected by socket lock
1933 * For UDP, we use xchg() to guard against concurrent changes.
1935 bool udp_sk_rx_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
1937 struct dst_entry
*old
;
1939 if (dst_hold_safe(dst
)) {
1940 old
= xchg(&sk
->sk_rx_dst
, dst
);
1946 EXPORT_SYMBOL(udp_sk_rx_dst_set
);
1949 * Multicasts and broadcasts go to each listener.
1951 * Note: called only from the BH handler context.
1953 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1955 __be32 saddr
, __be32 daddr
,
1956 struct udp_table
*udptable
,
1959 struct sock
*sk
, *first
= NULL
;
1960 unsigned short hnum
= ntohs(uh
->dest
);
1961 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, hnum
);
1962 unsigned int hash2
= 0, hash2_any
= 0, use_hash2
= (hslot
->count
> 10);
1963 unsigned int offset
= offsetof(typeof(*sk
), sk_node
);
1964 int dif
= skb
->dev
->ifindex
;
1965 int sdif
= inet_sdif(skb
);
1966 struct hlist_node
*node
;
1967 struct sk_buff
*nskb
;
1970 hash2_any
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
) &
1972 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
) & udptable
->mask
;
1974 hslot
= &udptable
->hash2
[hash2
];
1975 offset
= offsetof(typeof(*sk
), __sk_common
.skc_portaddr_node
);
1978 sk_for_each_entry_offset_rcu(sk
, node
, &hslot
->head
, offset
) {
1979 if (!__udp_is_mcast_sock(net
, sk
, uh
->dest
, daddr
,
1980 uh
->source
, saddr
, dif
, sdif
, hnum
))
1987 nskb
= skb_clone(skb
, GFP_ATOMIC
);
1989 if (unlikely(!nskb
)) {
1990 atomic_inc(&sk
->sk_drops
);
1991 __UDP_INC_STATS(net
, UDP_MIB_RCVBUFERRORS
,
1993 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
,
1997 if (udp_queue_rcv_skb(sk
, nskb
) > 0)
2001 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2002 if (use_hash2
&& hash2
!= hash2_any
) {
2008 if (udp_queue_rcv_skb(first
, skb
) > 0)
2012 __UDP_INC_STATS(net
, UDP_MIB_IGNOREDMULTI
,
2013 proto
== IPPROTO_UDPLITE
);
2018 /* Initialize UDP checksum. If exited with zero value (success),
2019 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2020 * Otherwise, csum completion requires chacksumming packet body,
2021 * including udp header and folding it to skb->csum.
2023 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
2028 UDP_SKB_CB(skb
)->partial_cov
= 0;
2029 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
2031 if (proto
== IPPROTO_UDPLITE
) {
2032 err
= udplite_checksum_init(skb
, uh
);
2037 /* Note, we are only interested in != 0 or == 0, thus the
2040 return (__force
int)skb_checksum_init_zero_check(skb
, proto
, uh
->check
,
2041 inet_compute_pseudo
);
2045 * All we need to do is get the socket, and then do a checksum.
2048 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
2053 unsigned short ulen
;
2054 struct rtable
*rt
= skb_rtable(skb
);
2055 __be32 saddr
, daddr
;
2056 struct net
*net
= dev_net(skb
->dev
);
2059 * Validate the packet.
2061 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
2062 goto drop
; /* No space for header. */
2065 ulen
= ntohs(uh
->len
);
2066 saddr
= ip_hdr(skb
)->saddr
;
2067 daddr
= ip_hdr(skb
)->daddr
;
2069 if (ulen
> skb
->len
)
2072 if (proto
== IPPROTO_UDP
) {
2073 /* UDP validates ulen. */
2074 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
2079 if (udp4_csum_init(skb
, uh
, proto
))
2082 sk
= skb_steal_sock(skb
);
2084 struct dst_entry
*dst
= skb_dst(skb
);
2087 if (unlikely(sk
->sk_rx_dst
!= dst
))
2088 udp_sk_rx_dst_set(sk
, dst
);
2090 ret
= udp_queue_rcv_skb(sk
, skb
);
2092 /* a return value > 0 means to resubmit the input, but
2093 * it wants the return to be -protocol, or 0
2100 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
2101 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
2102 saddr
, daddr
, udptable
, proto
);
2104 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
2108 if (inet_get_convert_csum(sk
) && uh
->check
&& !IS_UDPLITE(sk
))
2109 skb_checksum_try_convert(skb
, IPPROTO_UDP
, uh
->check
,
2110 inet_compute_pseudo
);
2112 ret
= udp_queue_rcv_skb(sk
, skb
);
2114 /* a return value > 0 means to resubmit the input, but
2115 * it wants the return to be -protocol, or 0
2122 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
2126 /* No socket. Drop packet silently, if checksum is wrong */
2127 if (udp_lib_checksum_complete(skb
))
2130 __UDP_INC_STATS(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
2131 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
2134 * Hmm. We got an UDP packet to a port to which we
2135 * don't wanna listen. Ignore it.
2141 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2142 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2143 &saddr
, ntohs(uh
->source
),
2145 &daddr
, ntohs(uh
->dest
));
2150 * RFC1122: OK. Discards the bad packet silently (as far as
2151 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2153 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2154 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2155 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
2157 __UDP_INC_STATS(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
2159 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
2164 /* We can only early demux multicast if there is a single matching socket.
2165 * If more than one socket found returns NULL
2167 static struct sock
*__udp4_lib_mcast_demux_lookup(struct net
*net
,
2168 __be16 loc_port
, __be32 loc_addr
,
2169 __be16 rmt_port
, __be32 rmt_addr
,
2172 struct sock
*sk
, *result
;
2173 unsigned short hnum
= ntohs(loc_port
);
2174 unsigned int slot
= udp_hashfn(net
, hnum
, udp_table
.mask
);
2175 struct udp_hslot
*hslot
= &udp_table
.hash
[slot
];
2177 /* Do not bother scanning a too big list */
2178 if (hslot
->count
> 10)
2182 sk_for_each_rcu(sk
, &hslot
->head
) {
2183 if (__udp_is_mcast_sock(net
, sk
, loc_port
, loc_addr
,
2184 rmt_port
, rmt_addr
, dif
, sdif
, hnum
)) {
2194 /* For unicast we should only early demux connected sockets or we can
2195 * break forwarding setups. The chains here can be long so only check
2196 * if the first socket is an exact match and if not move on.
2198 static struct sock
*__udp4_lib_demux_lookup(struct net
*net
,
2199 __be16 loc_port
, __be32 loc_addr
,
2200 __be16 rmt_port
, __be32 rmt_addr
,
2203 unsigned short hnum
= ntohs(loc_port
);
2204 unsigned int hash2
= udp4_portaddr_hash(net
, loc_addr
, hnum
);
2205 unsigned int slot2
= hash2
& udp_table
.mask
;
2206 struct udp_hslot
*hslot2
= &udp_table
.hash2
[slot2
];
2207 INET_ADDR_COOKIE(acookie
, rmt_addr
, loc_addr
);
2208 const __portpair ports
= INET_COMBINED_PORTS(rmt_port
, hnum
);
2211 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
2212 if (INET_MATCH(sk
, net
, acookie
, rmt_addr
,
2213 loc_addr
, ports
, dif
, sdif
))
2215 /* Only check first socket in chain */
2221 int udp_v4_early_demux(struct sk_buff
*skb
)
2223 struct net
*net
= dev_net(skb
->dev
);
2224 struct in_device
*in_dev
= NULL
;
2225 const struct iphdr
*iph
;
2226 const struct udphdr
*uh
;
2227 struct sock
*sk
= NULL
;
2228 struct dst_entry
*dst
;
2229 int dif
= skb
->dev
->ifindex
;
2230 int sdif
= inet_sdif(skb
);
2233 /* validate the packet */
2234 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct udphdr
)))
2240 if (skb
->pkt_type
== PACKET_MULTICAST
) {
2241 in_dev
= __in_dev_get_rcu(skb
->dev
);
2246 ours
= ip_check_mc_rcu(in_dev
, iph
->daddr
, iph
->saddr
,
2251 sk
= __udp4_lib_mcast_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2252 uh
->source
, iph
->saddr
,
2254 } else if (skb
->pkt_type
== PACKET_HOST
) {
2255 sk
= __udp4_lib_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2256 uh
->source
, iph
->saddr
, dif
, sdif
);
2259 if (!sk
|| !refcount_inc_not_zero(&sk
->sk_refcnt
))
2263 skb
->destructor
= sock_efree
;
2264 dst
= READ_ONCE(sk
->sk_rx_dst
);
2267 dst
= dst_check(dst
, 0);
2271 /* set noref for now.
2272 * any place which wants to hold dst has to call
2275 skb_dst_set_noref(skb
, dst
);
2277 /* for unconnected multicast sockets we need to validate
2278 * the source on each packet
2280 if (!inet_sk(sk
)->inet_daddr
&& in_dev
)
2281 return ip_mc_validate_source(skb
, iph
->daddr
,
2282 iph
->saddr
, iph
->tos
,
2283 skb
->dev
, in_dev
, &itag
);
2288 int udp_rcv(struct sk_buff
*skb
)
2290 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
2293 void udp_destroy_sock(struct sock
*sk
)
2295 struct udp_sock
*up
= udp_sk(sk
);
2296 bool slow
= lock_sock_fast(sk
);
2297 udp_flush_pending_frames(sk
);
2298 unlock_sock_fast(sk
, slow
);
2299 if (static_key_false(&udp_encap_needed
) && up
->encap_type
) {
2300 void (*encap_destroy
)(struct sock
*sk
);
2301 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
2308 * Socket option code for UDP
2310 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
2311 char __user
*optval
, unsigned int optlen
,
2312 int (*push_pending_frames
)(struct sock
*))
2314 struct udp_sock
*up
= udp_sk(sk
);
2317 int is_udplite
= IS_UDPLITE(sk
);
2319 if (optlen
< sizeof(int))
2322 if (get_user(val
, (int __user
*)optval
))
2325 valbool
= val
? 1 : 0;
2334 push_pending_frames(sk
);
2342 case UDP_ENCAP_ESPINUDP
:
2343 case UDP_ENCAP_ESPINUDP_NON_IKE
:
2344 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
2346 case UDP_ENCAP_L2TPINUDP
:
2347 up
->encap_type
= val
;
2356 case UDP_NO_CHECK6_TX
:
2357 up
->no_check6_tx
= valbool
;
2360 case UDP_NO_CHECK6_RX
:
2361 up
->no_check6_rx
= valbool
;
2365 * UDP-Lite's partial checksum coverage (RFC 3828).
2367 /* The sender sets actual checksum coverage length via this option.
2368 * The case coverage > packet length is handled by send module. */
2369 case UDPLITE_SEND_CSCOV
:
2370 if (!is_udplite
) /* Disable the option on UDP sockets */
2371 return -ENOPROTOOPT
;
2372 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
2374 else if (val
> USHRT_MAX
)
2377 up
->pcflag
|= UDPLITE_SEND_CC
;
2380 /* The receiver specifies a minimum checksum coverage value. To make
2381 * sense, this should be set to at least 8 (as done below). If zero is
2382 * used, this again means full checksum coverage. */
2383 case UDPLITE_RECV_CSCOV
:
2384 if (!is_udplite
) /* Disable the option on UDP sockets */
2385 return -ENOPROTOOPT
;
2386 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
2388 else if (val
> USHRT_MAX
)
2391 up
->pcflag
|= UDPLITE_RECV_CC
;
2401 EXPORT_SYMBOL(udp_lib_setsockopt
);
2403 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2404 char __user
*optval
, unsigned int optlen
)
2406 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2407 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2408 udp_push_pending_frames
);
2409 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2412 #ifdef CONFIG_COMPAT
2413 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2414 char __user
*optval
, unsigned int optlen
)
2416 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2417 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2418 udp_push_pending_frames
);
2419 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2423 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
2424 char __user
*optval
, int __user
*optlen
)
2426 struct udp_sock
*up
= udp_sk(sk
);
2429 if (get_user(len
, optlen
))
2432 len
= min_t(unsigned int, len
, sizeof(int));
2443 val
= up
->encap_type
;
2446 case UDP_NO_CHECK6_TX
:
2447 val
= up
->no_check6_tx
;
2450 case UDP_NO_CHECK6_RX
:
2451 val
= up
->no_check6_rx
;
2454 /* The following two cannot be changed on UDP sockets, the return is
2455 * always 0 (which corresponds to the full checksum coverage of UDP). */
2456 case UDPLITE_SEND_CSCOV
:
2460 case UDPLITE_RECV_CSCOV
:
2465 return -ENOPROTOOPT
;
2468 if (put_user(len
, optlen
))
2470 if (copy_to_user(optval
, &val
, len
))
2474 EXPORT_SYMBOL(udp_lib_getsockopt
);
2476 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2477 char __user
*optval
, int __user
*optlen
)
2479 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2480 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2481 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2484 #ifdef CONFIG_COMPAT
2485 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2486 char __user
*optval
, int __user
*optlen
)
2488 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2489 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2490 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2494 * udp_poll - wait for a UDP event.
2495 * @file - file struct
2497 * @wait - poll table
2499 * This is same as datagram poll, except for the special case of
2500 * blocking sockets. If application is using a blocking fd
2501 * and a packet with checksum error is in the queue;
2502 * then it could get return from select indicating data available
2503 * but then block when reading it. Add special case code
2504 * to work around these arguably broken applications.
2506 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2508 unsigned int mask
= datagram_poll(file
, sock
, wait
);
2509 struct sock
*sk
= sock
->sk
;
2511 if (!skb_queue_empty(&udp_sk(sk
)->reader_queue
))
2512 mask
|= POLLIN
| POLLRDNORM
;
2514 sock_rps_record_flow(sk
);
2516 /* Check for false positives due to checksum errors */
2517 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2518 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && first_packet_length(sk
) == -1)
2519 mask
&= ~(POLLIN
| POLLRDNORM
);
2524 EXPORT_SYMBOL(udp_poll
);
2526 int udp_abort(struct sock
*sk
, int err
)
2531 sk
->sk_error_report(sk
);
2532 __udp_disconnect(sk
, 0);
2538 EXPORT_SYMBOL_GPL(udp_abort
);
2540 struct proto udp_prot
= {
2542 .owner
= THIS_MODULE
,
2543 .close
= udp_lib_close
,
2544 .connect
= ip4_datagram_connect
,
2545 .disconnect
= udp_disconnect
,
2547 .init
= udp_init_sock
,
2548 .destroy
= udp_destroy_sock
,
2549 .setsockopt
= udp_setsockopt
,
2550 .getsockopt
= udp_getsockopt
,
2551 .sendmsg
= udp_sendmsg
,
2552 .recvmsg
= udp_recvmsg
,
2553 .sendpage
= udp_sendpage
,
2554 .release_cb
= ip4_datagram_release_cb
,
2555 .hash
= udp_lib_hash
,
2556 .unhash
= udp_lib_unhash
,
2557 .rehash
= udp_v4_rehash
,
2558 .get_port
= udp_v4_get_port
,
2559 .memory_allocated
= &udp_memory_allocated
,
2560 .sysctl_mem
= sysctl_udp_mem
,
2561 .sysctl_wmem
= &sysctl_udp_wmem_min
,
2562 .sysctl_rmem
= &sysctl_udp_rmem_min
,
2563 .obj_size
= sizeof(struct udp_sock
),
2564 .h
.udp_table
= &udp_table
,
2565 #ifdef CONFIG_COMPAT
2566 .compat_setsockopt
= compat_udp_setsockopt
,
2567 .compat_getsockopt
= compat_udp_getsockopt
,
2569 .diag_destroy
= udp_abort
,
2571 EXPORT_SYMBOL(udp_prot
);
2573 /* ------------------------------------------------------------------------ */
2574 #ifdef CONFIG_PROC_FS
2576 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2579 struct udp_iter_state
*state
= seq
->private;
2580 struct net
*net
= seq_file_net(seq
);
2582 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
2584 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
2586 if (hlist_empty(&hslot
->head
))
2589 spin_lock_bh(&hslot
->lock
);
2590 sk_for_each(sk
, &hslot
->head
) {
2591 if (!net_eq(sock_net(sk
), net
))
2593 if (sk
->sk_family
== state
->family
)
2596 spin_unlock_bh(&hslot
->lock
);
2603 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2605 struct udp_iter_state
*state
= seq
->private;
2606 struct net
*net
= seq_file_net(seq
);
2610 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
2613 if (state
->bucket
<= state
->udp_table
->mask
)
2614 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2615 return udp_get_first(seq
, state
->bucket
+ 1);
2620 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2622 struct sock
*sk
= udp_get_first(seq
, 0);
2625 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2627 return pos
? NULL
: sk
;
2630 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2632 struct udp_iter_state
*state
= seq
->private;
2633 state
->bucket
= MAX_UDP_PORTS
;
2635 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2638 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2642 if (v
== SEQ_START_TOKEN
)
2643 sk
= udp_get_idx(seq
, 0);
2645 sk
= udp_get_next(seq
, v
);
2651 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
2653 struct udp_iter_state
*state
= seq
->private;
2655 if (state
->bucket
<= state
->udp_table
->mask
)
2656 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2659 int udp_seq_open(struct inode
*inode
, struct file
*file
)
2661 struct udp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2662 struct udp_iter_state
*s
;
2665 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2666 sizeof(struct udp_iter_state
));
2670 s
= ((struct seq_file
*)file
->private_data
)->private;
2671 s
->family
= afinfo
->family
;
2672 s
->udp_table
= afinfo
->udp_table
;
2675 EXPORT_SYMBOL(udp_seq_open
);
2677 /* ------------------------------------------------------------------------ */
2678 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2680 struct proc_dir_entry
*p
;
2683 afinfo
->seq_ops
.start
= udp_seq_start
;
2684 afinfo
->seq_ops
.next
= udp_seq_next
;
2685 afinfo
->seq_ops
.stop
= udp_seq_stop
;
2687 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2688 afinfo
->seq_fops
, afinfo
);
2693 EXPORT_SYMBOL(udp_proc_register
);
2695 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2697 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2699 EXPORT_SYMBOL(udp_proc_unregister
);
2701 /* ------------------------------------------------------------------------ */
2702 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2705 struct inet_sock
*inet
= inet_sk(sp
);
2706 __be32 dest
= inet
->inet_daddr
;
2707 __be32 src
= inet
->inet_rcv_saddr
;
2708 __u16 destp
= ntohs(inet
->inet_dport
);
2709 __u16 srcp
= ntohs(inet
->inet_sport
);
2711 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2712 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2713 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2714 sk_wmem_alloc_get(sp
),
2715 sk_rmem_alloc_get(sp
),
2717 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2719 refcount_read(&sp
->sk_refcnt
), sp
,
2720 atomic_read(&sp
->sk_drops
));
2723 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2725 seq_setwidth(seq
, 127);
2726 if (v
== SEQ_START_TOKEN
)
2727 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2728 "rx_queue tr tm->when retrnsmt uid timeout "
2729 "inode ref pointer drops");
2731 struct udp_iter_state
*state
= seq
->private;
2733 udp4_format_sock(v
, seq
, state
->bucket
);
2739 static const struct file_operations udp_afinfo_seq_fops
= {
2740 .owner
= THIS_MODULE
,
2741 .open
= udp_seq_open
,
2743 .llseek
= seq_lseek
,
2744 .release
= seq_release_net
2747 /* ------------------------------------------------------------------------ */
2748 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2751 .udp_table
= &udp_table
,
2752 .seq_fops
= &udp_afinfo_seq_fops
,
2754 .show
= udp4_seq_show
,
2758 static int __net_init
udp4_proc_init_net(struct net
*net
)
2760 return udp_proc_register(net
, &udp4_seq_afinfo
);
2763 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2765 udp_proc_unregister(net
, &udp4_seq_afinfo
);
2768 static struct pernet_operations udp4_net_ops
= {
2769 .init
= udp4_proc_init_net
,
2770 .exit
= udp4_proc_exit_net
,
2773 int __init
udp4_proc_init(void)
2775 return register_pernet_subsys(&udp4_net_ops
);
2778 void udp4_proc_exit(void)
2780 unregister_pernet_subsys(&udp4_net_ops
);
2782 #endif /* CONFIG_PROC_FS */
2784 static __initdata
unsigned long uhash_entries
;
2785 static int __init
set_uhash_entries(char *str
)
2792 ret
= kstrtoul(str
, 0, &uhash_entries
);
2796 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2797 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2800 __setup("uhash_entries=", set_uhash_entries
);
2802 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2806 table
->hash
= alloc_large_system_hash(name
,
2807 2 * sizeof(struct udp_hslot
),
2809 21, /* one slot per 2 MB */
2813 UDP_HTABLE_SIZE_MIN
,
2816 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2817 for (i
= 0; i
<= table
->mask
; i
++) {
2818 INIT_HLIST_HEAD(&table
->hash
[i
].head
);
2819 table
->hash
[i
].count
= 0;
2820 spin_lock_init(&table
->hash
[i
].lock
);
2822 for (i
= 0; i
<= table
->mask
; i
++) {
2823 INIT_HLIST_HEAD(&table
->hash2
[i
].head
);
2824 table
->hash2
[i
].count
= 0;
2825 spin_lock_init(&table
->hash2
[i
].lock
);
2829 u32
udp_flow_hashrnd(void)
2831 static u32 hashrnd __read_mostly
;
2833 net_get_random_once(&hashrnd
, sizeof(hashrnd
));
2837 EXPORT_SYMBOL(udp_flow_hashrnd
);
2839 void __init
udp_init(void)
2841 unsigned long limit
;
2844 udp_table_init(&udp_table
, "UDP");
2845 limit
= nr_free_buffer_pages() / 8;
2846 limit
= max(limit
, 128UL);
2847 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2848 sysctl_udp_mem
[1] = limit
;
2849 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2851 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2852 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
2854 /* 16 spinlocks per cpu */
2855 udp_busylocks_log
= ilog2(nr_cpu_ids
) + 4;
2856 udp_busylocks
= kmalloc(sizeof(spinlock_t
) << udp_busylocks_log
,
2859 panic("UDP: failed to alloc udp_busylocks\n");
2860 for (i
= 0; i
< (1U << udp_busylocks_log
); i
++)
2861 spin_lock_init(udp_busylocks
+ i
);