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/memblock.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>
118 #include <net/udp_tunnel.h>
120 struct udp_table udp_table __read_mostly
;
121 EXPORT_SYMBOL(udp_table
);
123 long sysctl_udp_mem
[3] __read_mostly
;
124 EXPORT_SYMBOL(sysctl_udp_mem
);
126 atomic_long_t udp_memory_allocated
;
127 EXPORT_SYMBOL(udp_memory_allocated
);
129 #define MAX_UDP_PORTS 65536
130 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
132 /* IPCB reference means this can not be used from early demux */
133 static bool udp_lib_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
135 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
136 if (!net
->ipv4
.sysctl_udp_l3mdev_accept
&&
137 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
143 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
144 const struct udp_hslot
*hslot
,
145 unsigned long *bitmap
,
146 struct sock
*sk
, unsigned int log
)
149 kuid_t uid
= sock_i_uid(sk
);
151 sk_for_each(sk2
, &hslot
->head
) {
152 if (net_eq(sock_net(sk2
), net
) &&
154 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
155 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
156 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
157 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
158 inet_rcv_saddr_equal(sk
, sk2
, true)) {
159 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
160 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
161 uid_eq(uid
, sock_i_uid(sk2
))) {
167 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
176 * Note: we still hold spinlock of primary hash chain, so no other writer
177 * can insert/delete a socket with local_port == num
179 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
180 struct udp_hslot
*hslot2
,
184 kuid_t uid
= sock_i_uid(sk
);
187 spin_lock(&hslot2
->lock
);
188 udp_portaddr_for_each_entry(sk2
, &hslot2
->head
) {
189 if (net_eq(sock_net(sk2
), net
) &&
191 (udp_sk(sk2
)->udp_port_hash
== num
) &&
192 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
193 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
194 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
195 inet_rcv_saddr_equal(sk
, sk2
, true)) {
196 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
197 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
198 uid_eq(uid
, sock_i_uid(sk2
))) {
206 spin_unlock(&hslot2
->lock
);
210 static int udp_reuseport_add_sock(struct sock
*sk
, struct udp_hslot
*hslot
)
212 struct net
*net
= sock_net(sk
);
213 kuid_t uid
= sock_i_uid(sk
);
216 sk_for_each(sk2
, &hslot
->head
) {
217 if (net_eq(sock_net(sk2
), net
) &&
219 sk2
->sk_family
== sk
->sk_family
&&
220 ipv6_only_sock(sk2
) == ipv6_only_sock(sk
) &&
221 (udp_sk(sk2
)->udp_port_hash
== udp_sk(sk
)->udp_port_hash
) &&
222 (sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
223 sk2
->sk_reuseport
&& uid_eq(uid
, sock_i_uid(sk2
)) &&
224 inet_rcv_saddr_equal(sk
, sk2
, false)) {
225 return reuseport_add_sock(sk
, sk2
,
226 inet_rcv_saddr_any(sk
));
230 return reuseport_alloc(sk
, inet_rcv_saddr_any(sk
));
234 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
236 * @sk: socket struct in question
237 * @snum: port number to look up
238 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
241 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
242 unsigned int hash2_nulladdr
)
244 struct udp_hslot
*hslot
, *hslot2
;
245 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
247 struct net
*net
= sock_net(sk
);
250 int low
, high
, remaining
;
252 unsigned short first
, last
;
253 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
255 inet_get_local_port_range(net
, &low
, &high
);
256 remaining
= (high
- low
) + 1;
258 rand
= prandom_u32();
259 first
= reciprocal_scale(rand
, remaining
) + low
;
261 * force rand to be an odd multiple of UDP_HTABLE_SIZE
263 rand
= (rand
| 1) * (udptable
->mask
+ 1);
264 last
= first
+ udptable
->mask
+ 1;
266 hslot
= udp_hashslot(udptable
, net
, first
);
267 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
268 spin_lock_bh(&hslot
->lock
);
269 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
274 * Iterate on all possible values of snum for this hash.
275 * Using steps of an odd multiple of UDP_HTABLE_SIZE
276 * give us randomization and full range coverage.
279 if (low
<= snum
&& snum
<= high
&&
280 !test_bit(snum
>> udptable
->log
, bitmap
) &&
281 !inet_is_local_reserved_port(net
, snum
))
284 } while (snum
!= first
);
285 spin_unlock_bh(&hslot
->lock
);
287 } while (++first
!= last
);
290 hslot
= udp_hashslot(udptable
, net
, snum
);
291 spin_lock_bh(&hslot
->lock
);
292 if (hslot
->count
> 10) {
294 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
296 slot2
&= udptable
->mask
;
297 hash2_nulladdr
&= udptable
->mask
;
299 hslot2
= udp_hashslot2(udptable
, slot2
);
300 if (hslot
->count
< hslot2
->count
)
301 goto scan_primary_hash
;
303 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
, sk
);
304 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
305 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
306 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
315 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
, 0))
319 inet_sk(sk
)->inet_num
= snum
;
320 udp_sk(sk
)->udp_port_hash
= snum
;
321 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
322 if (sk_unhashed(sk
)) {
323 if (sk
->sk_reuseport
&&
324 udp_reuseport_add_sock(sk
, hslot
)) {
325 inet_sk(sk
)->inet_num
= 0;
326 udp_sk(sk
)->udp_port_hash
= 0;
327 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
331 sk_add_node_rcu(sk
, &hslot
->head
);
333 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
335 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
336 spin_lock(&hslot2
->lock
);
337 if (IS_ENABLED(CONFIG_IPV6
) && sk
->sk_reuseport
&&
338 sk
->sk_family
== AF_INET6
)
339 hlist_add_tail_rcu(&udp_sk(sk
)->udp_portaddr_node
,
342 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
345 spin_unlock(&hslot2
->lock
);
347 sock_set_flag(sk
, SOCK_RCU_FREE
);
350 spin_unlock_bh(&hslot
->lock
);
354 EXPORT_SYMBOL(udp_lib_get_port
);
356 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
358 unsigned int hash2_nulladdr
=
359 ipv4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
360 unsigned int hash2_partial
=
361 ipv4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
363 /* precompute partial secondary hash */
364 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
365 return udp_lib_get_port(sk
, snum
, hash2_nulladdr
);
368 static int compute_score(struct sock
*sk
, struct net
*net
,
369 __be32 saddr
, __be16 sport
,
370 __be32 daddr
, unsigned short hnum
,
371 int dif
, int sdif
, bool exact_dif
)
374 struct inet_sock
*inet
;
377 if (!net_eq(sock_net(sk
), net
) ||
378 udp_sk(sk
)->udp_port_hash
!= hnum
||
382 score
= (sk
->sk_family
== PF_INET
) ? 2 : 1;
385 if (inet
->inet_rcv_saddr
) {
386 if (inet
->inet_rcv_saddr
!= daddr
)
391 if (inet
->inet_daddr
) {
392 if (inet
->inet_daddr
!= saddr
)
397 if (inet
->inet_dport
) {
398 if (inet
->inet_dport
!= sport
)
403 dev_match
= udp_sk_bound_dev_eq(net
, sk
->sk_bound_dev_if
,
409 if (sk
->sk_incoming_cpu
== raw_smp_processor_id())
414 static u32
udp_ehashfn(const struct net
*net
, const __be32 laddr
,
415 const __u16 lport
, const __be32 faddr
,
418 static u32 udp_ehash_secret __read_mostly
;
420 net_get_random_once(&udp_ehash_secret
, sizeof(udp_ehash_secret
));
422 return __inet_ehashfn(laddr
, lport
, faddr
, fport
,
423 udp_ehash_secret
+ net_hash_mix(net
));
426 /* called with rcu_read_lock() */
427 static struct sock
*udp4_lib_lookup2(struct net
*net
,
428 __be32 saddr
, __be16 sport
,
429 __be32 daddr
, unsigned int hnum
,
430 int dif
, int sdif
, bool exact_dif
,
431 struct udp_hslot
*hslot2
,
434 struct sock
*sk
, *result
;
440 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
441 score
= compute_score(sk
, net
, saddr
, sport
,
442 daddr
, hnum
, dif
, sdif
, exact_dif
);
443 if (score
> badness
) {
444 if (sk
->sk_reuseport
) {
445 hash
= udp_ehashfn(net
, daddr
, hnum
,
447 result
= reuseport_select_sock(sk
, hash
, skb
,
448 sizeof(struct udphdr
));
459 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
460 * harder than this. -DaveM
462 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
463 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
,
464 int sdif
, struct udp_table
*udptable
, struct sk_buff
*skb
)
466 struct sock
*sk
, *result
;
467 unsigned short hnum
= ntohs(dport
);
468 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
469 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
470 bool exact_dif
= udp_lib_exact_dif_match(net
, skb
);
474 if (hslot
->count
> 10) {
475 hash2
= ipv4_portaddr_hash(net
, daddr
, hnum
);
476 slot2
= hash2
& udptable
->mask
;
477 hslot2
= &udptable
->hash2
[slot2
];
478 if (hslot
->count
< hslot2
->count
)
481 result
= udp4_lib_lookup2(net
, saddr
, sport
,
482 daddr
, hnum
, dif
, sdif
,
483 exact_dif
, hslot2
, skb
);
485 unsigned int old_slot2
= slot2
;
486 hash2
= ipv4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
487 slot2
= hash2
& udptable
->mask
;
488 /* avoid searching the same slot again. */
489 if (unlikely(slot2
== old_slot2
))
492 hslot2
= &udptable
->hash2
[slot2
];
493 if (hslot
->count
< hslot2
->count
)
496 result
= udp4_lib_lookup2(net
, saddr
, sport
,
497 daddr
, hnum
, dif
, sdif
,
498 exact_dif
, hslot2
, skb
);
500 if (unlikely(IS_ERR(result
)))
507 sk_for_each_rcu(sk
, &hslot
->head
) {
508 score
= compute_score(sk
, net
, saddr
, sport
,
509 daddr
, hnum
, dif
, sdif
, exact_dif
);
510 if (score
> badness
) {
511 if (sk
->sk_reuseport
) {
512 hash
= udp_ehashfn(net
, daddr
, hnum
,
514 result
= reuseport_select_sock(sk
, hash
, skb
,
515 sizeof(struct udphdr
));
516 if (unlikely(IS_ERR(result
)))
527 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
529 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
530 __be16 sport
, __be16 dport
,
531 struct udp_table
*udptable
)
533 const struct iphdr
*iph
= ip_hdr(skb
);
535 return __udp4_lib_lookup(dev_net(skb
->dev
), iph
->saddr
, sport
,
536 iph
->daddr
, dport
, inet_iif(skb
),
537 inet_sdif(skb
), udptable
, skb
);
540 struct sock
*udp4_lib_lookup_skb(struct sk_buff
*skb
,
541 __be16 sport
, __be16 dport
)
543 return __udp4_lib_lookup_skb(skb
, sport
, dport
, &udp_table
);
545 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb
);
547 /* Must be called under rcu_read_lock().
548 * Does increment socket refcount.
550 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
551 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
552 __be32 daddr
, __be16 dport
, int dif
)
556 sk
= __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
,
557 dif
, 0, &udp_table
, NULL
);
558 if (sk
&& !refcount_inc_not_zero(&sk
->sk_refcnt
))
562 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
565 static inline bool __udp_is_mcast_sock(struct net
*net
, struct sock
*sk
,
566 __be16 loc_port
, __be32 loc_addr
,
567 __be16 rmt_port
, __be32 rmt_addr
,
568 int dif
, int sdif
, unsigned short hnum
)
570 struct inet_sock
*inet
= inet_sk(sk
);
572 if (!net_eq(sock_net(sk
), net
) ||
573 udp_sk(sk
)->udp_port_hash
!= hnum
||
574 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
575 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
576 (inet
->inet_rcv_saddr
&& inet
->inet_rcv_saddr
!= loc_addr
) ||
577 ipv6_only_sock(sk
) ||
578 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
&&
579 sk
->sk_bound_dev_if
!= sdif
))
581 if (!ip_mc_sf_allow(sk
, loc_addr
, rmt_addr
, dif
, sdif
))
587 * This routine is called by the ICMP module when it gets some
588 * sort of error condition. If err < 0 then the socket should
589 * be closed and the error returned to the user. If err > 0
590 * it's just the icmp type << 8 | icmp code.
591 * Header points to the ip header of the error packet. We move
592 * on past this. Then (as it used to claim before adjustment)
593 * header points to the first 8 bytes of the udp header. We need
594 * to find the appropriate port.
597 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
599 struct inet_sock
*inet
;
600 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
601 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
602 const int type
= icmp_hdr(skb
)->type
;
603 const int code
= icmp_hdr(skb
)->code
;
607 struct net
*net
= dev_net(skb
->dev
);
609 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
610 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
,
611 inet_sdif(skb
), udptable
, NULL
);
613 __ICMP_INC_STATS(net
, ICMP_MIB_INERRORS
);
614 return; /* No socket for error */
623 case ICMP_TIME_EXCEEDED
:
626 case ICMP_SOURCE_QUENCH
:
628 case ICMP_PARAMETERPROB
:
632 case ICMP_DEST_UNREACH
:
633 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
634 ipv4_sk_update_pmtu(skb
, sk
, info
);
635 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
643 if (code
<= NR_ICMP_UNREACH
) {
644 harderr
= icmp_err_convert
[code
].fatal
;
645 err
= icmp_err_convert
[code
].errno
;
649 ipv4_sk_redirect(skb
, sk
);
654 * RFC1122: OK. Passes ICMP errors back to application, as per
657 if (!inet
->recverr
) {
658 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
661 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
664 sk
->sk_error_report(sk
);
669 void udp_err(struct sk_buff
*skb
, u32 info
)
671 __udp4_lib_err(skb
, info
, &udp_table
);
675 * Throw away all pending data and cancel the corking. Socket is locked.
677 void udp_flush_pending_frames(struct sock
*sk
)
679 struct udp_sock
*up
= udp_sk(sk
);
684 ip_flush_pending_frames(sk
);
687 EXPORT_SYMBOL(udp_flush_pending_frames
);
690 * udp4_hwcsum - handle outgoing HW checksumming
691 * @skb: sk_buff containing the filled-in UDP header
692 * (checksum field must be zeroed out)
693 * @src: source IP address
694 * @dst: destination IP address
696 void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
698 struct udphdr
*uh
= udp_hdr(skb
);
699 int offset
= skb_transport_offset(skb
);
700 int len
= skb
->len
- offset
;
704 if (!skb_has_frag_list(skb
)) {
706 * Only one fragment on the socket.
708 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
709 skb
->csum_offset
= offsetof(struct udphdr
, check
);
710 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
713 struct sk_buff
*frags
;
716 * HW-checksum won't work as there are two or more
717 * fragments on the socket so that all csums of sk_buffs
720 skb_walk_frags(skb
, frags
) {
721 csum
= csum_add(csum
, frags
->csum
);
725 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
726 skb
->ip_summed
= CHECKSUM_NONE
;
728 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
730 uh
->check
= CSUM_MANGLED_0
;
733 EXPORT_SYMBOL_GPL(udp4_hwcsum
);
735 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
736 * for the simple case like when setting the checksum for a UDP tunnel.
738 void udp_set_csum(bool nocheck
, struct sk_buff
*skb
,
739 __be32 saddr
, __be32 daddr
, int len
)
741 struct udphdr
*uh
= udp_hdr(skb
);
745 } else if (skb_is_gso(skb
)) {
746 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
747 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
749 uh
->check
= udp_v4_check(len
, saddr
, daddr
, lco_csum(skb
));
751 uh
->check
= CSUM_MANGLED_0
;
753 skb
->ip_summed
= CHECKSUM_PARTIAL
;
754 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
755 skb
->csum_offset
= offsetof(struct udphdr
, check
);
756 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
759 EXPORT_SYMBOL(udp_set_csum
);
761 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
,
762 struct inet_cork
*cork
)
764 struct sock
*sk
= skb
->sk
;
765 struct inet_sock
*inet
= inet_sk(sk
);
768 int is_udplite
= IS_UDPLITE(sk
);
769 int offset
= skb_transport_offset(skb
);
770 int len
= skb
->len
- offset
;
774 * Create a UDP header
777 uh
->source
= inet
->inet_sport
;
778 uh
->dest
= fl4
->fl4_dport
;
779 uh
->len
= htons(len
);
782 if (cork
->gso_size
) {
783 const int hlen
= skb_network_header_len(skb
) +
784 sizeof(struct udphdr
);
786 if (hlen
+ cork
->gso_size
> cork
->fragsize
)
788 if (skb
->len
> cork
->gso_size
* UDP_MAX_SEGMENTS
)
790 if (sk
->sk_no_check_tx
)
792 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
|| is_udplite
||
793 dst_xfrm(skb_dst(skb
)))
796 skb_shinfo(skb
)->gso_size
= cork
->gso_size
;
797 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP_L4
;
798 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(len
- sizeof(uh
),
803 if (is_udplite
) /* UDP-Lite */
804 csum
= udplite_csum(skb
);
806 else if (sk
->sk_no_check_tx
) { /* UDP csum off */
808 skb
->ip_summed
= CHECKSUM_NONE
;
811 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
814 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
818 csum
= udp_csum(skb
);
820 /* add protocol-dependent pseudo-header */
821 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
822 sk
->sk_protocol
, csum
);
824 uh
->check
= CSUM_MANGLED_0
;
827 err
= ip_send_skb(sock_net(sk
), skb
);
829 if (err
== -ENOBUFS
&& !inet
->recverr
) {
830 UDP_INC_STATS(sock_net(sk
),
831 UDP_MIB_SNDBUFERRORS
, is_udplite
);
835 UDP_INC_STATS(sock_net(sk
),
836 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
841 * Push out all pending data as one UDP datagram. Socket is locked.
843 int udp_push_pending_frames(struct sock
*sk
)
845 struct udp_sock
*up
= udp_sk(sk
);
846 struct inet_sock
*inet
= inet_sk(sk
);
847 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
851 skb
= ip_finish_skb(sk
, fl4
);
855 err
= udp_send_skb(skb
, fl4
, &inet
->cork
.base
);
862 EXPORT_SYMBOL(udp_push_pending_frames
);
864 static int __udp_cmsg_send(struct cmsghdr
*cmsg
, u16
*gso_size
)
866 switch (cmsg
->cmsg_type
) {
868 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(__u16
)))
870 *gso_size
= *(__u16
*)CMSG_DATA(cmsg
);
877 int udp_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, u16
*gso_size
)
879 struct cmsghdr
*cmsg
;
880 bool need_ip
= false;
883 for_each_cmsghdr(cmsg
, msg
) {
884 if (!CMSG_OK(msg
, cmsg
))
887 if (cmsg
->cmsg_level
!= SOL_UDP
) {
892 err
= __udp_cmsg_send(cmsg
, gso_size
);
899 EXPORT_SYMBOL_GPL(udp_cmsg_send
);
901 int udp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
903 struct inet_sock
*inet
= inet_sk(sk
);
904 struct udp_sock
*up
= udp_sk(sk
);
905 DECLARE_SOCKADDR(struct sockaddr_in
*, usin
, msg
->msg_name
);
906 struct flowi4 fl4_stack
;
909 struct ipcm_cookie ipc
;
910 struct rtable
*rt
= NULL
;
913 __be32 daddr
, faddr
, saddr
;
916 int err
, is_udplite
= IS_UDPLITE(sk
);
917 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
918 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
920 struct ip_options_data opt_copy
;
929 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
932 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
934 fl4
= &inet
->cork
.fl
.u
.ip4
;
937 * There are pending frames.
938 * The socket lock must be held while it's corked.
941 if (likely(up
->pending
)) {
942 if (unlikely(up
->pending
!= AF_INET
)) {
950 ulen
+= sizeof(struct udphdr
);
953 * Get and verify the address.
956 if (msg
->msg_namelen
< sizeof(*usin
))
958 if (usin
->sin_family
!= AF_INET
) {
959 if (usin
->sin_family
!= AF_UNSPEC
)
960 return -EAFNOSUPPORT
;
963 daddr
= usin
->sin_addr
.s_addr
;
964 dport
= usin
->sin_port
;
968 if (sk
->sk_state
!= TCP_ESTABLISHED
)
969 return -EDESTADDRREQ
;
970 daddr
= inet
->inet_daddr
;
971 dport
= inet
->inet_dport
;
972 /* Open fast path for connected socket.
973 Route will not be used, if at least one option is set.
978 ipcm_init_sk(&ipc
, inet
);
979 ipc
.gso_size
= up
->gso_size
;
981 if (msg
->msg_controllen
) {
982 err
= udp_cmsg_send(sk
, msg
, &ipc
.gso_size
);
984 err
= ip_cmsg_send(sk
, msg
, &ipc
,
985 sk
->sk_family
== AF_INET6
);
986 if (unlikely(err
< 0)) {
995 struct ip_options_rcu
*inet_opt
;
998 inet_opt
= rcu_dereference(inet
->inet_opt
);
1000 memcpy(&opt_copy
, inet_opt
,
1001 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
1002 ipc
.opt
= &opt_copy
.opt
;
1007 if (cgroup_bpf_enabled
&& !connected
) {
1008 err
= BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk
,
1009 (struct sockaddr
*)usin
, &ipc
.addr
);
1013 if (usin
->sin_port
== 0) {
1014 /* BPF program set invalid port. Reject it. */
1018 daddr
= usin
->sin_addr
.s_addr
;
1019 dport
= usin
->sin_port
;
1024 ipc
.addr
= faddr
= daddr
;
1026 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
1031 faddr
= ipc
.opt
->opt
.faddr
;
1034 tos
= get_rttos(&ipc
, inet
);
1035 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
1036 (msg
->msg_flags
& MSG_DONTROUTE
) ||
1037 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
1042 if (ipv4_is_multicast(daddr
)) {
1043 if (!ipc
.oif
|| netif_index_is_l3_master(sock_net(sk
), ipc
.oif
))
1044 ipc
.oif
= inet
->mc_index
;
1046 saddr
= inet
->mc_addr
;
1048 } else if (!ipc
.oif
) {
1049 ipc
.oif
= inet
->uc_index
;
1050 } else if (ipv4_is_lbcast(daddr
) && inet
->uc_index
) {
1051 /* oif is set, packet is to local broadcast and
1052 * and uc_index is set. oif is most likely set
1053 * by sk_bound_dev_if. If uc_index != oif check if the
1054 * oif is an L3 master and uc_index is an L3 slave.
1055 * If so, we want to allow the send using the uc_index.
1057 if (ipc
.oif
!= inet
->uc_index
&&
1058 ipc
.oif
== l3mdev_master_ifindex_by_index(sock_net(sk
),
1060 ipc
.oif
= inet
->uc_index
;
1065 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
1068 struct net
*net
= sock_net(sk
);
1069 __u8 flow_flags
= inet_sk_flowi_flags(sk
);
1073 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
1074 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
1076 faddr
, saddr
, dport
, inet
->inet_sport
,
1079 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
1080 rt
= ip_route_output_flow(net
, fl4
, sk
);
1084 if (err
== -ENETUNREACH
)
1085 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
1090 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
1091 !sock_flag(sk
, SOCK_BROADCAST
))
1094 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1097 if (msg
->msg_flags
&MSG_CONFIRM
)
1103 daddr
= ipc
.addr
= fl4
->daddr
;
1105 /* Lockless fast path for the non-corking case. */
1107 struct inet_cork cork
;
1109 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
, ulen
,
1110 sizeof(struct udphdr
), &ipc
, &rt
,
1111 &cork
, msg
->msg_flags
);
1113 if (!IS_ERR_OR_NULL(skb
))
1114 err
= udp_send_skb(skb
, fl4
, &cork
);
1119 if (unlikely(up
->pending
)) {
1120 /* The socket is already corked while preparing it. */
1121 /* ... which is an evident application bug. --ANK */
1124 net_dbg_ratelimited("socket already corked\n");
1129 * Now cork the socket to pend data.
1131 fl4
= &inet
->cork
.fl
.u
.ip4
;
1134 fl4
->fl4_dport
= dport
;
1135 fl4
->fl4_sport
= inet
->inet_sport
;
1136 up
->pending
= AF_INET
;
1140 err
= ip_append_data(sk
, fl4
, getfrag
, msg
, ulen
,
1141 sizeof(struct udphdr
), &ipc
, &rt
,
1142 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1144 udp_flush_pending_frames(sk
);
1146 err
= udp_push_pending_frames(sk
);
1147 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1159 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1160 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1161 * we don't have a good statistic (IpOutDiscards but it can be too many
1162 * things). We could add another new stat but at least for now that
1163 * seems like overkill.
1165 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1166 UDP_INC_STATS(sock_net(sk
),
1167 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1172 if (msg
->msg_flags
& MSG_PROBE
)
1173 dst_confirm_neigh(&rt
->dst
, &fl4
->daddr
);
1174 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1175 goto back_from_confirm
;
1179 EXPORT_SYMBOL(udp_sendmsg
);
1181 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1182 size_t size
, int flags
)
1184 struct inet_sock
*inet
= inet_sk(sk
);
1185 struct udp_sock
*up
= udp_sk(sk
);
1188 if (flags
& MSG_SENDPAGE_NOTLAST
)
1192 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1194 /* Call udp_sendmsg to specify destination address which
1195 * sendpage interface can't pass.
1196 * This will succeed only when the socket is connected.
1198 ret
= udp_sendmsg(sk
, &msg
, 0);
1205 if (unlikely(!up
->pending
)) {
1208 net_dbg_ratelimited("cork failed\n");
1212 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1213 page
, offset
, size
, flags
);
1214 if (ret
== -EOPNOTSUPP
) {
1216 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1220 udp_flush_pending_frames(sk
);
1225 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1226 ret
= udp_push_pending_frames(sk
);
1234 #define UDP_SKB_IS_STATELESS 0x80000000
1236 static void udp_set_dev_scratch(struct sk_buff
*skb
)
1238 struct udp_dev_scratch
*scratch
= udp_skb_scratch(skb
);
1240 BUILD_BUG_ON(sizeof(struct udp_dev_scratch
) > sizeof(long));
1241 scratch
->_tsize_state
= skb
->truesize
;
1242 #if BITS_PER_LONG == 64
1243 scratch
->len
= skb
->len
;
1244 scratch
->csum_unnecessary
= !!skb_csum_unnecessary(skb
);
1245 scratch
->is_linear
= !skb_is_nonlinear(skb
);
1247 /* all head states execept sp (dst, sk, nf) are always cleared by
1248 * udp_rcv() and we need to preserve secpath, if present, to eventually
1249 * process IP_CMSG_PASSSEC at recvmsg() time
1251 if (likely(!skb_sec_path(skb
)))
1252 scratch
->_tsize_state
|= UDP_SKB_IS_STATELESS
;
1255 static int udp_skb_truesize(struct sk_buff
*skb
)
1257 return udp_skb_scratch(skb
)->_tsize_state
& ~UDP_SKB_IS_STATELESS
;
1260 static bool udp_skb_has_head_state(struct sk_buff
*skb
)
1262 return !(udp_skb_scratch(skb
)->_tsize_state
& UDP_SKB_IS_STATELESS
);
1265 /* fully reclaim rmem/fwd memory allocated for skb */
1266 static void udp_rmem_release(struct sock
*sk
, int size
, int partial
,
1267 bool rx_queue_lock_held
)
1269 struct udp_sock
*up
= udp_sk(sk
);
1270 struct sk_buff_head
*sk_queue
;
1273 if (likely(partial
)) {
1274 up
->forward_deficit
+= size
;
1275 size
= up
->forward_deficit
;
1276 if (size
< (sk
->sk_rcvbuf
>> 2))
1279 size
+= up
->forward_deficit
;
1281 up
->forward_deficit
= 0;
1283 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1284 * if the called don't held it already
1286 sk_queue
= &sk
->sk_receive_queue
;
1287 if (!rx_queue_lock_held
)
1288 spin_lock(&sk_queue
->lock
);
1291 sk
->sk_forward_alloc
+= size
;
1292 amt
= (sk
->sk_forward_alloc
- partial
) & ~(SK_MEM_QUANTUM
- 1);
1293 sk
->sk_forward_alloc
-= amt
;
1296 __sk_mem_reduce_allocated(sk
, amt
>> SK_MEM_QUANTUM_SHIFT
);
1298 atomic_sub(size
, &sk
->sk_rmem_alloc
);
1300 /* this can save us from acquiring the rx queue lock on next receive */
1301 skb_queue_splice_tail_init(sk_queue
, &up
->reader_queue
);
1303 if (!rx_queue_lock_held
)
1304 spin_unlock(&sk_queue
->lock
);
1307 /* Note: called with reader_queue.lock held.
1308 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1309 * This avoids a cache line miss while receive_queue lock is held.
1310 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1312 void udp_skb_destructor(struct sock
*sk
, struct sk_buff
*skb
)
1314 prefetch(&skb
->data
);
1315 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, false);
1317 EXPORT_SYMBOL(udp_skb_destructor
);
1319 /* as above, but the caller held the rx queue lock, too */
1320 static void udp_skb_dtor_locked(struct sock
*sk
, struct sk_buff
*skb
)
1322 prefetch(&skb
->data
);
1323 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, true);
1326 /* Idea of busylocks is to let producers grab an extra spinlock
1327 * to relieve pressure on the receive_queue spinlock shared by consumer.
1328 * Under flood, this means that only one producer can be in line
1329 * trying to acquire the receive_queue spinlock.
1330 * These busylock can be allocated on a per cpu manner, instead of a
1331 * per socket one (that would consume a cache line per socket)
1333 static int udp_busylocks_log __read_mostly
;
1334 static spinlock_t
*udp_busylocks __read_mostly
;
1336 static spinlock_t
*busylock_acquire(void *ptr
)
1340 busy
= udp_busylocks
+ hash_ptr(ptr
, udp_busylocks_log
);
1345 static void busylock_release(spinlock_t
*busy
)
1351 int __udp_enqueue_schedule_skb(struct sock
*sk
, struct sk_buff
*skb
)
1353 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
1354 int rmem
, delta
, amt
, err
= -ENOMEM
;
1355 spinlock_t
*busy
= NULL
;
1358 /* try to avoid the costly atomic add/sub pair when the receive
1359 * queue is full; always allow at least a packet
1361 rmem
= atomic_read(&sk
->sk_rmem_alloc
);
1362 if (rmem
> sk
->sk_rcvbuf
)
1365 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1366 * having linear skbs :
1367 * - Reduce memory overhead and thus increase receive queue capacity
1368 * - Less cache line misses at copyout() time
1369 * - Less work at consume_skb() (less alien page frag freeing)
1371 if (rmem
> (sk
->sk_rcvbuf
>> 1)) {
1374 busy
= busylock_acquire(sk
);
1376 size
= skb
->truesize
;
1377 udp_set_dev_scratch(skb
);
1379 /* we drop only if the receive buf is full and the receive
1380 * queue contains some other skb
1382 rmem
= atomic_add_return(size
, &sk
->sk_rmem_alloc
);
1383 if (rmem
> (size
+ sk
->sk_rcvbuf
))
1386 spin_lock(&list
->lock
);
1387 if (size
>= sk
->sk_forward_alloc
) {
1388 amt
= sk_mem_pages(size
);
1389 delta
= amt
<< SK_MEM_QUANTUM_SHIFT
;
1390 if (!__sk_mem_raise_allocated(sk
, delta
, amt
, SK_MEM_RECV
)) {
1392 spin_unlock(&list
->lock
);
1396 sk
->sk_forward_alloc
+= delta
;
1399 sk
->sk_forward_alloc
-= size
;
1401 /* no need to setup a destructor, we will explicitly release the
1402 * forward allocated memory on dequeue
1404 sock_skb_set_dropcount(sk
, skb
);
1406 __skb_queue_tail(list
, skb
);
1407 spin_unlock(&list
->lock
);
1409 if (!sock_flag(sk
, SOCK_DEAD
))
1410 sk
->sk_data_ready(sk
);
1412 busylock_release(busy
);
1416 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1419 atomic_inc(&sk
->sk_drops
);
1420 busylock_release(busy
);
1423 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb
);
1425 void udp_destruct_sock(struct sock
*sk
)
1427 /* reclaim completely the forward allocated memory */
1428 struct udp_sock
*up
= udp_sk(sk
);
1429 unsigned int total
= 0;
1430 struct sk_buff
*skb
;
1432 skb_queue_splice_tail_init(&sk
->sk_receive_queue
, &up
->reader_queue
);
1433 while ((skb
= __skb_dequeue(&up
->reader_queue
)) != NULL
) {
1434 total
+= skb
->truesize
;
1437 udp_rmem_release(sk
, total
, 0, true);
1439 inet_sock_destruct(sk
);
1441 EXPORT_SYMBOL_GPL(udp_destruct_sock
);
1443 int udp_init_sock(struct sock
*sk
)
1445 skb_queue_head_init(&udp_sk(sk
)->reader_queue
);
1446 sk
->sk_destruct
= udp_destruct_sock
;
1449 EXPORT_SYMBOL_GPL(udp_init_sock
);
1451 void skb_consume_udp(struct sock
*sk
, struct sk_buff
*skb
, int len
)
1453 if (unlikely(READ_ONCE(sk
->sk_peek_off
) >= 0)) {
1454 bool slow
= lock_sock_fast(sk
);
1456 sk_peek_offset_bwd(sk
, len
);
1457 unlock_sock_fast(sk
, slow
);
1460 if (!skb_unref(skb
))
1463 /* In the more common cases we cleared the head states previously,
1464 * see __udp_queue_rcv_skb().
1466 if (unlikely(udp_skb_has_head_state(skb
)))
1467 skb_release_head_state(skb
);
1468 __consume_stateless_skb(skb
);
1470 EXPORT_SYMBOL_GPL(skb_consume_udp
);
1472 static struct sk_buff
*__first_packet_length(struct sock
*sk
,
1473 struct sk_buff_head
*rcvq
,
1476 struct sk_buff
*skb
;
1478 while ((skb
= skb_peek(rcvq
)) != NULL
) {
1479 if (udp_lib_checksum_complete(skb
)) {
1480 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1482 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
1484 atomic_inc(&sk
->sk_drops
);
1485 __skb_unlink(skb
, rcvq
);
1486 *total
+= skb
->truesize
;
1489 /* the csum related bits could be changed, refresh
1492 udp_set_dev_scratch(skb
);
1500 * first_packet_length - return length of first packet in receive queue
1503 * Drops all bad checksum frames, until a valid one is found.
1504 * Returns the length of found skb, or -1 if none is found.
1506 static int first_packet_length(struct sock
*sk
)
1508 struct sk_buff_head
*rcvq
= &udp_sk(sk
)->reader_queue
;
1509 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1510 struct sk_buff
*skb
;
1514 spin_lock_bh(&rcvq
->lock
);
1515 skb
= __first_packet_length(sk
, rcvq
, &total
);
1516 if (!skb
&& !skb_queue_empty(sk_queue
)) {
1517 spin_lock(&sk_queue
->lock
);
1518 skb_queue_splice_tail_init(sk_queue
, rcvq
);
1519 spin_unlock(&sk_queue
->lock
);
1521 skb
= __first_packet_length(sk
, rcvq
, &total
);
1523 res
= skb
? skb
->len
: -1;
1525 udp_rmem_release(sk
, total
, 1, false);
1526 spin_unlock_bh(&rcvq
->lock
);
1531 * IOCTL requests applicable to the UDP protocol
1534 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1539 int amount
= sk_wmem_alloc_get(sk
);
1541 return put_user(amount
, (int __user
*)arg
);
1546 int amount
= max_t(int, 0, first_packet_length(sk
));
1548 return put_user(amount
, (int __user
*)arg
);
1552 return -ENOIOCTLCMD
;
1557 EXPORT_SYMBOL(udp_ioctl
);
1559 struct sk_buff
*__skb_recv_udp(struct sock
*sk
, unsigned int flags
,
1560 int noblock
, int *peeked
, int *off
, int *err
)
1562 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1563 struct sk_buff_head
*queue
;
1564 struct sk_buff
*last
;
1568 queue
= &udp_sk(sk
)->reader_queue
;
1569 flags
|= noblock
? MSG_DONTWAIT
: 0;
1570 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
1572 struct sk_buff
*skb
;
1574 error
= sock_error(sk
);
1581 spin_lock_bh(&queue
->lock
);
1582 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
,
1587 spin_unlock_bh(&queue
->lock
);
1591 if (skb_queue_empty(sk_queue
)) {
1592 spin_unlock_bh(&queue
->lock
);
1596 /* refill the reader queue and walk it again
1597 * keep both queues locked to avoid re-acquiring
1598 * the sk_receive_queue lock if fwd memory scheduling
1601 spin_lock(&sk_queue
->lock
);
1602 skb_queue_splice_tail_init(sk_queue
, queue
);
1604 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
,
1605 udp_skb_dtor_locked
,
1608 spin_unlock(&sk_queue
->lock
);
1609 spin_unlock_bh(&queue
->lock
);
1614 if (!sk_can_busy_loop(sk
))
1617 sk_busy_loop(sk
, flags
& MSG_DONTWAIT
);
1618 } while (!skb_queue_empty(sk_queue
));
1620 /* sk_queue is empty, reader_queue may contain peeked packets */
1622 !__skb_wait_for_more_packets(sk
, &error
, &timeo
,
1623 (struct sk_buff
*)sk_queue
));
1628 EXPORT_SYMBOL(__skb_recv_udp
);
1631 * This should be easy, if there is something there we
1632 * return it, otherwise we block.
1635 int udp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int noblock
,
1636 int flags
, int *addr_len
)
1638 struct inet_sock
*inet
= inet_sk(sk
);
1639 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, msg
->msg_name
);
1640 struct sk_buff
*skb
;
1641 unsigned int ulen
, copied
;
1642 int peeked
, peeking
, off
;
1644 int is_udplite
= IS_UDPLITE(sk
);
1645 bool checksum_valid
= false;
1647 if (flags
& MSG_ERRQUEUE
)
1648 return ip_recv_error(sk
, msg
, len
, addr_len
);
1651 peeking
= flags
& MSG_PEEK
;
1652 off
= sk_peek_offset(sk
, flags
);
1653 skb
= __skb_recv_udp(sk
, flags
, noblock
, &peeked
, &off
, &err
);
1657 ulen
= udp_skb_len(skb
);
1659 if (copied
> ulen
- off
)
1660 copied
= ulen
- off
;
1661 else if (copied
< ulen
)
1662 msg
->msg_flags
|= MSG_TRUNC
;
1665 * If checksum is needed at all, try to do it while copying the
1666 * data. If the data is truncated, or if we only want a partial
1667 * coverage checksum (UDP-Lite), do it before the copy.
1670 if (copied
< ulen
|| peeking
||
1671 (is_udplite
&& UDP_SKB_CB(skb
)->partial_cov
)) {
1672 checksum_valid
= udp_skb_csum_unnecessary(skb
) ||
1673 !__udp_lib_checksum_complete(skb
);
1674 if (!checksum_valid
)
1678 if (checksum_valid
|| udp_skb_csum_unnecessary(skb
)) {
1679 if (udp_skb_is_linear(skb
))
1680 err
= copy_linear_skb(skb
, copied
, off
, &msg
->msg_iter
);
1682 err
= skb_copy_datagram_msg(skb
, off
, msg
, copied
);
1684 err
= skb_copy_and_csum_datagram_msg(skb
, off
, msg
);
1690 if (unlikely(err
)) {
1692 atomic_inc(&sk
->sk_drops
);
1693 UDP_INC_STATS(sock_net(sk
),
1694 UDP_MIB_INERRORS
, is_udplite
);
1701 UDP_INC_STATS(sock_net(sk
),
1702 UDP_MIB_INDATAGRAMS
, is_udplite
);
1704 sock_recv_ts_and_drops(msg
, sk
, skb
);
1706 /* Copy the address. */
1708 sin
->sin_family
= AF_INET
;
1709 sin
->sin_port
= udp_hdr(skb
)->source
;
1710 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1711 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1712 *addr_len
= sizeof(*sin
);
1715 if (udp_sk(sk
)->gro_enabled
)
1716 udp_cmsg_recv(msg
, sk
, skb
);
1718 if (inet
->cmsg_flags
)
1719 ip_cmsg_recv_offset(msg
, sk
, skb
, sizeof(struct udphdr
), off
);
1722 if (flags
& MSG_TRUNC
)
1725 skb_consume_udp(sk
, skb
, peeking
? -err
: err
);
1729 if (!__sk_queue_drop_skb(sk
, &udp_sk(sk
)->reader_queue
, skb
, flags
,
1730 udp_skb_destructor
)) {
1731 UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1732 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1736 /* starting over for a new packet, but check if we need to yield */
1738 msg
->msg_flags
&= ~MSG_TRUNC
;
1742 int udp_pre_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
1744 /* This check is replicated from __ip4_datagram_connect() and
1745 * intended to prevent BPF program called below from accessing bytes
1746 * that are out of the bound specified by user in addr_len.
1748 if (addr_len
< sizeof(struct sockaddr_in
))
1751 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk
, uaddr
);
1753 EXPORT_SYMBOL(udp_pre_connect
);
1755 int __udp_disconnect(struct sock
*sk
, int flags
)
1757 struct inet_sock
*inet
= inet_sk(sk
);
1759 * 1003.1g - break association.
1762 sk
->sk_state
= TCP_CLOSE
;
1763 inet
->inet_daddr
= 0;
1764 inet
->inet_dport
= 0;
1765 sock_rps_reset_rxhash(sk
);
1766 sk
->sk_bound_dev_if
= 0;
1767 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1768 inet_reset_saddr(sk
);
1770 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1771 sk
->sk_prot
->unhash(sk
);
1772 inet
->inet_sport
= 0;
1777 EXPORT_SYMBOL(__udp_disconnect
);
1779 int udp_disconnect(struct sock
*sk
, int flags
)
1782 __udp_disconnect(sk
, flags
);
1786 EXPORT_SYMBOL(udp_disconnect
);
1788 void udp_lib_unhash(struct sock
*sk
)
1790 if (sk_hashed(sk
)) {
1791 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1792 struct udp_hslot
*hslot
, *hslot2
;
1794 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1795 udp_sk(sk
)->udp_port_hash
);
1796 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1798 spin_lock_bh(&hslot
->lock
);
1799 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1800 reuseport_detach_sock(sk
);
1801 if (sk_del_node_init_rcu(sk
)) {
1803 inet_sk(sk
)->inet_num
= 0;
1804 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1806 spin_lock(&hslot2
->lock
);
1807 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1809 spin_unlock(&hslot2
->lock
);
1811 spin_unlock_bh(&hslot
->lock
);
1814 EXPORT_SYMBOL(udp_lib_unhash
);
1817 * inet_rcv_saddr was changed, we must rehash secondary hash
1819 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1821 if (sk_hashed(sk
)) {
1822 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1823 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1825 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1826 nhslot2
= udp_hashslot2(udptable
, newhash
);
1827 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1829 if (hslot2
!= nhslot2
||
1830 rcu_access_pointer(sk
->sk_reuseport_cb
)) {
1831 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1832 udp_sk(sk
)->udp_port_hash
);
1833 /* we must lock primary chain too */
1834 spin_lock_bh(&hslot
->lock
);
1835 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1836 reuseport_detach_sock(sk
);
1838 if (hslot2
!= nhslot2
) {
1839 spin_lock(&hslot2
->lock
);
1840 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1842 spin_unlock(&hslot2
->lock
);
1844 spin_lock(&nhslot2
->lock
);
1845 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1848 spin_unlock(&nhslot2
->lock
);
1851 spin_unlock_bh(&hslot
->lock
);
1855 EXPORT_SYMBOL(udp_lib_rehash
);
1857 static void udp_v4_rehash(struct sock
*sk
)
1859 u16 new_hash
= ipv4_portaddr_hash(sock_net(sk
),
1860 inet_sk(sk
)->inet_rcv_saddr
,
1861 inet_sk(sk
)->inet_num
);
1862 udp_lib_rehash(sk
, new_hash
);
1865 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1869 if (inet_sk(sk
)->inet_daddr
) {
1870 sock_rps_save_rxhash(sk
, skb
);
1871 sk_mark_napi_id(sk
, skb
);
1872 sk_incoming_cpu_update(sk
);
1874 sk_mark_napi_id_once(sk
, skb
);
1877 rc
= __udp_enqueue_schedule_skb(sk
, skb
);
1879 int is_udplite
= IS_UDPLITE(sk
);
1881 /* Note that an ENOMEM error is charged twice */
1883 UDP_INC_STATS(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1885 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1887 trace_udp_fail_queue_rcv_skb(rc
, sk
);
1894 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key
);
1895 void udp_encap_enable(void)
1897 static_branch_enable(&udp_encap_needed_key
);
1899 EXPORT_SYMBOL(udp_encap_enable
);
1904 * >0: "udp encap" protocol resubmission
1906 * Note that in the success and error cases, the skb is assumed to
1907 * have either been requeued or freed.
1909 static int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1911 struct udp_sock
*up
= udp_sk(sk
);
1912 int is_udplite
= IS_UDPLITE(sk
);
1915 * Charge it to the socket, dropping if the queue is full.
1917 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1921 if (static_branch_unlikely(&udp_encap_needed_key
) && up
->encap_type
) {
1922 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
1925 * This is an encapsulation socket so pass the skb to
1926 * the socket's udp_encap_rcv() hook. Otherwise, just
1927 * fall through and pass this up the UDP socket.
1928 * up->encap_rcv() returns the following value:
1929 * =0 if skb was successfully passed to the encap
1930 * handler or was discarded by it.
1931 * >0 if skb should be passed on to UDP.
1932 * <0 if skb should be resubmitted as proto -N
1935 /* if we're overly short, let UDP handle it */
1936 encap_rcv
= READ_ONCE(up
->encap_rcv
);
1940 /* Verify checksum before giving to encap */
1941 if (udp_lib_checksum_complete(skb
))
1944 ret
= encap_rcv(sk
, skb
);
1946 __UDP_INC_STATS(sock_net(sk
),
1947 UDP_MIB_INDATAGRAMS
,
1953 /* FALLTHROUGH -- it's a UDP Packet */
1957 * UDP-Lite specific tests, ignored on UDP sockets
1959 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1962 * MIB statistics other than incrementing the error count are
1963 * disabled for the following two types of errors: these depend
1964 * on the application settings, not on the functioning of the
1965 * protocol stack as such.
1967 * RFC 3828 here recommends (sec 3.3): "There should also be a
1968 * way ... to ... at least let the receiving application block
1969 * delivery of packets with coverage values less than a value
1970 * provided by the application."
1972 if (up
->pcrlen
== 0) { /* full coverage was set */
1973 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1974 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1977 /* The next case involves violating the min. coverage requested
1978 * by the receiver. This is subtle: if receiver wants x and x is
1979 * greater than the buffersize/MTU then receiver will complain
1980 * that it wants x while sender emits packets of smaller size y.
1981 * Therefore the above ...()->partial_cov statement is essential.
1983 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1984 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1985 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1990 prefetch(&sk
->sk_rmem_alloc
);
1991 if (rcu_access_pointer(sk
->sk_filter
) &&
1992 udp_lib_checksum_complete(skb
))
1995 if (sk_filter_trim_cap(sk
, skb
, sizeof(struct udphdr
)))
1998 udp_csum_pull_header(skb
);
2000 ipv4_pktinfo_prepare(sk
, skb
);
2001 return __udp_queue_rcv_skb(sk
, skb
);
2004 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
2006 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
2007 atomic_inc(&sk
->sk_drops
);
2012 /* For TCP sockets, sk_rx_dst is protected by socket lock
2013 * For UDP, we use xchg() to guard against concurrent changes.
2015 bool udp_sk_rx_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
2017 struct dst_entry
*old
;
2019 if (dst_hold_safe(dst
)) {
2020 old
= xchg(&sk
->sk_rx_dst
, dst
);
2026 EXPORT_SYMBOL(udp_sk_rx_dst_set
);
2029 * Multicasts and broadcasts go to each listener.
2031 * Note: called only from the BH handler context.
2033 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
2035 __be32 saddr
, __be32 daddr
,
2036 struct udp_table
*udptable
,
2039 struct sock
*sk
, *first
= NULL
;
2040 unsigned short hnum
= ntohs(uh
->dest
);
2041 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, hnum
);
2042 unsigned int hash2
= 0, hash2_any
= 0, use_hash2
= (hslot
->count
> 10);
2043 unsigned int offset
= offsetof(typeof(*sk
), sk_node
);
2044 int dif
= skb
->dev
->ifindex
;
2045 int sdif
= inet_sdif(skb
);
2046 struct hlist_node
*node
;
2047 struct sk_buff
*nskb
;
2050 hash2_any
= ipv4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
) &
2052 hash2
= ipv4_portaddr_hash(net
, daddr
, hnum
) & udptable
->mask
;
2054 hslot
= &udptable
->hash2
[hash2
];
2055 offset
= offsetof(typeof(*sk
), __sk_common
.skc_portaddr_node
);
2058 sk_for_each_entry_offset_rcu(sk
, node
, &hslot
->head
, offset
) {
2059 if (!__udp_is_mcast_sock(net
, sk
, uh
->dest
, daddr
,
2060 uh
->source
, saddr
, dif
, sdif
, hnum
))
2067 nskb
= skb_clone(skb
, GFP_ATOMIC
);
2069 if (unlikely(!nskb
)) {
2070 atomic_inc(&sk
->sk_drops
);
2071 __UDP_INC_STATS(net
, UDP_MIB_RCVBUFERRORS
,
2073 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
,
2077 if (udp_queue_rcv_skb(sk
, nskb
) > 0)
2081 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2082 if (use_hash2
&& hash2
!= hash2_any
) {
2088 if (udp_queue_rcv_skb(first
, skb
) > 0)
2092 __UDP_INC_STATS(net
, UDP_MIB_IGNOREDMULTI
,
2093 proto
== IPPROTO_UDPLITE
);
2098 /* Initialize UDP checksum. If exited with zero value (success),
2099 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2100 * Otherwise, csum completion requires chacksumming packet body,
2101 * including udp header and folding it to skb->csum.
2103 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
2108 UDP_SKB_CB(skb
)->partial_cov
= 0;
2109 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
2111 if (proto
== IPPROTO_UDPLITE
) {
2112 err
= udplite_checksum_init(skb
, uh
);
2116 if (UDP_SKB_CB(skb
)->partial_cov
) {
2117 skb
->csum
= inet_compute_pseudo(skb
, proto
);
2122 /* Note, we are only interested in != 0 or == 0, thus the
2125 err
= (__force
int)skb_checksum_init_zero_check(skb
, proto
, uh
->check
,
2126 inet_compute_pseudo
);
2130 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&& !skb
->csum_valid
) {
2131 /* If SW calculated the value, we know it's bad */
2132 if (skb
->csum_complete_sw
)
2135 /* HW says the value is bad. Let's validate that.
2136 * skb->csum is no longer the full packet checksum,
2137 * so don't treat it as such.
2139 skb_checksum_complete_unset(skb
);
2145 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2146 * return code conversion for ip layer consumption
2148 static int udp_unicast_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
,
2153 if (inet_get_convert_csum(sk
) && uh
->check
&& !IS_UDPLITE(sk
))
2154 skb_checksum_try_convert(skb
, IPPROTO_UDP
, uh
->check
,
2155 inet_compute_pseudo
);
2157 ret
= udp_queue_rcv_skb(sk
, skb
);
2159 /* a return value > 0 means to resubmit the input, but
2160 * it wants the return to be -protocol, or 0
2168 * All we need to do is get the socket, and then do a checksum.
2171 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
2176 unsigned short ulen
;
2177 struct rtable
*rt
= skb_rtable(skb
);
2178 __be32 saddr
, daddr
;
2179 struct net
*net
= dev_net(skb
->dev
);
2182 * Validate the packet.
2184 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
2185 goto drop
; /* No space for header. */
2188 ulen
= ntohs(uh
->len
);
2189 saddr
= ip_hdr(skb
)->saddr
;
2190 daddr
= ip_hdr(skb
)->daddr
;
2192 if (ulen
> skb
->len
)
2195 if (proto
== IPPROTO_UDP
) {
2196 /* UDP validates ulen. */
2197 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
2202 if (udp4_csum_init(skb
, uh
, proto
))
2205 sk
= skb_steal_sock(skb
);
2207 struct dst_entry
*dst
= skb_dst(skb
);
2210 if (unlikely(sk
->sk_rx_dst
!= dst
))
2211 udp_sk_rx_dst_set(sk
, dst
);
2213 ret
= udp_unicast_rcv_skb(sk
, skb
, uh
);
2218 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
2219 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
2220 saddr
, daddr
, udptable
, proto
);
2222 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
2224 return udp_unicast_rcv_skb(sk
, skb
, uh
);
2226 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
2230 /* No socket. Drop packet silently, if checksum is wrong */
2231 if (udp_lib_checksum_complete(skb
))
2234 __UDP_INC_STATS(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
2235 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
2238 * Hmm. We got an UDP packet to a port to which we
2239 * don't wanna listen. Ignore it.
2245 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2246 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2247 &saddr
, ntohs(uh
->source
),
2249 &daddr
, ntohs(uh
->dest
));
2254 * RFC1122: OK. Discards the bad packet silently (as far as
2255 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2257 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2258 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2259 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
2261 __UDP_INC_STATS(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
2263 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
2268 /* We can only early demux multicast if there is a single matching socket.
2269 * If more than one socket found returns NULL
2271 static struct sock
*__udp4_lib_mcast_demux_lookup(struct net
*net
,
2272 __be16 loc_port
, __be32 loc_addr
,
2273 __be16 rmt_port
, __be32 rmt_addr
,
2276 struct sock
*sk
, *result
;
2277 unsigned short hnum
= ntohs(loc_port
);
2278 unsigned int slot
= udp_hashfn(net
, hnum
, udp_table
.mask
);
2279 struct udp_hslot
*hslot
= &udp_table
.hash
[slot
];
2281 /* Do not bother scanning a too big list */
2282 if (hslot
->count
> 10)
2286 sk_for_each_rcu(sk
, &hslot
->head
) {
2287 if (__udp_is_mcast_sock(net
, sk
, loc_port
, loc_addr
,
2288 rmt_port
, rmt_addr
, dif
, sdif
, hnum
)) {
2298 /* For unicast we should only early demux connected sockets or we can
2299 * break forwarding setups. The chains here can be long so only check
2300 * if the first socket is an exact match and if not move on.
2302 static struct sock
*__udp4_lib_demux_lookup(struct net
*net
,
2303 __be16 loc_port
, __be32 loc_addr
,
2304 __be16 rmt_port
, __be32 rmt_addr
,
2307 unsigned short hnum
= ntohs(loc_port
);
2308 unsigned int hash2
= ipv4_portaddr_hash(net
, loc_addr
, hnum
);
2309 unsigned int slot2
= hash2
& udp_table
.mask
;
2310 struct udp_hslot
*hslot2
= &udp_table
.hash2
[slot2
];
2311 INET_ADDR_COOKIE(acookie
, rmt_addr
, loc_addr
);
2312 const __portpair ports
= INET_COMBINED_PORTS(rmt_port
, hnum
);
2315 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
2316 if (INET_MATCH(sk
, net
, acookie
, rmt_addr
,
2317 loc_addr
, ports
, dif
, sdif
))
2319 /* Only check first socket in chain */
2325 int udp_v4_early_demux(struct sk_buff
*skb
)
2327 struct net
*net
= dev_net(skb
->dev
);
2328 struct in_device
*in_dev
= NULL
;
2329 const struct iphdr
*iph
;
2330 const struct udphdr
*uh
;
2331 struct sock
*sk
= NULL
;
2332 struct dst_entry
*dst
;
2333 int dif
= skb
->dev
->ifindex
;
2334 int sdif
= inet_sdif(skb
);
2337 /* validate the packet */
2338 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct udphdr
)))
2344 if (skb
->pkt_type
== PACKET_MULTICAST
) {
2345 in_dev
= __in_dev_get_rcu(skb
->dev
);
2350 ours
= ip_check_mc_rcu(in_dev
, iph
->daddr
, iph
->saddr
,
2355 sk
= __udp4_lib_mcast_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2356 uh
->source
, iph
->saddr
,
2358 } else if (skb
->pkt_type
== PACKET_HOST
) {
2359 sk
= __udp4_lib_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2360 uh
->source
, iph
->saddr
, dif
, sdif
);
2363 if (!sk
|| !refcount_inc_not_zero(&sk
->sk_refcnt
))
2367 skb
->destructor
= sock_efree
;
2368 dst
= READ_ONCE(sk
->sk_rx_dst
);
2371 dst
= dst_check(dst
, 0);
2375 /* set noref for now.
2376 * any place which wants to hold dst has to call
2379 skb_dst_set_noref(skb
, dst
);
2381 /* for unconnected multicast sockets we need to validate
2382 * the source on each packet
2384 if (!inet_sk(sk
)->inet_daddr
&& in_dev
)
2385 return ip_mc_validate_source(skb
, iph
->daddr
,
2386 iph
->saddr
, iph
->tos
,
2387 skb
->dev
, in_dev
, &itag
);
2392 int udp_rcv(struct sk_buff
*skb
)
2394 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
2397 void udp_destroy_sock(struct sock
*sk
)
2399 struct udp_sock
*up
= udp_sk(sk
);
2400 bool slow
= lock_sock_fast(sk
);
2401 udp_flush_pending_frames(sk
);
2402 unlock_sock_fast(sk
, slow
);
2403 if (static_branch_unlikely(&udp_encap_needed_key
)) {
2404 if (up
->encap_type
) {
2405 void (*encap_destroy
)(struct sock
*sk
);
2406 encap_destroy
= READ_ONCE(up
->encap_destroy
);
2410 if (up
->encap_enabled
)
2411 static_branch_disable(&udp_encap_needed_key
);
2416 * Socket option code for UDP
2418 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
2419 char __user
*optval
, unsigned int optlen
,
2420 int (*push_pending_frames
)(struct sock
*))
2422 struct udp_sock
*up
= udp_sk(sk
);
2425 int is_udplite
= IS_UDPLITE(sk
);
2427 if (optlen
< sizeof(int))
2430 if (get_user(val
, (int __user
*)optval
))
2433 valbool
= val
? 1 : 0;
2442 push_pending_frames(sk
);
2450 case UDP_ENCAP_ESPINUDP
:
2451 case UDP_ENCAP_ESPINUDP_NON_IKE
:
2452 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
2454 case UDP_ENCAP_L2TPINUDP
:
2455 up
->encap_type
= val
;
2457 udp_tunnel_encap_enable(sk
->sk_socket
);
2466 case UDP_NO_CHECK6_TX
:
2467 up
->no_check6_tx
= valbool
;
2470 case UDP_NO_CHECK6_RX
:
2471 up
->no_check6_rx
= valbool
;
2475 if (val
< 0 || val
> USHRT_MAX
)
2483 udp_tunnel_encap_enable(sk
->sk_socket
);
2484 up
->gro_enabled
= valbool
;
2489 * UDP-Lite's partial checksum coverage (RFC 3828).
2491 /* The sender sets actual checksum coverage length via this option.
2492 * The case coverage > packet length is handled by send module. */
2493 case UDPLITE_SEND_CSCOV
:
2494 if (!is_udplite
) /* Disable the option on UDP sockets */
2495 return -ENOPROTOOPT
;
2496 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
2498 else if (val
> USHRT_MAX
)
2501 up
->pcflag
|= UDPLITE_SEND_CC
;
2504 /* The receiver specifies a minimum checksum coverage value. To make
2505 * sense, this should be set to at least 8 (as done below). If zero is
2506 * used, this again means full checksum coverage. */
2507 case UDPLITE_RECV_CSCOV
:
2508 if (!is_udplite
) /* Disable the option on UDP sockets */
2509 return -ENOPROTOOPT
;
2510 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
2512 else if (val
> USHRT_MAX
)
2515 up
->pcflag
|= UDPLITE_RECV_CC
;
2525 EXPORT_SYMBOL(udp_lib_setsockopt
);
2527 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2528 char __user
*optval
, unsigned int optlen
)
2530 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2531 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2532 udp_push_pending_frames
);
2533 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2536 #ifdef CONFIG_COMPAT
2537 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
2538 char __user
*optval
, unsigned int optlen
)
2540 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2541 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
2542 udp_push_pending_frames
);
2543 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2547 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
2548 char __user
*optval
, int __user
*optlen
)
2550 struct udp_sock
*up
= udp_sk(sk
);
2553 if (get_user(len
, optlen
))
2556 len
= min_t(unsigned int, len
, sizeof(int));
2567 val
= up
->encap_type
;
2570 case UDP_NO_CHECK6_TX
:
2571 val
= up
->no_check6_tx
;
2574 case UDP_NO_CHECK6_RX
:
2575 val
= up
->no_check6_rx
;
2582 /* The following two cannot be changed on UDP sockets, the return is
2583 * always 0 (which corresponds to the full checksum coverage of UDP). */
2584 case UDPLITE_SEND_CSCOV
:
2588 case UDPLITE_RECV_CSCOV
:
2593 return -ENOPROTOOPT
;
2596 if (put_user(len
, optlen
))
2598 if (copy_to_user(optval
, &val
, len
))
2602 EXPORT_SYMBOL(udp_lib_getsockopt
);
2604 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2605 char __user
*optval
, int __user
*optlen
)
2607 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2608 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2609 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2612 #ifdef CONFIG_COMPAT
2613 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2614 char __user
*optval
, int __user
*optlen
)
2616 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2617 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2618 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2622 * udp_poll - wait for a UDP event.
2623 * @file - file struct
2625 * @wait - poll table
2627 * This is same as datagram poll, except for the special case of
2628 * blocking sockets. If application is using a blocking fd
2629 * and a packet with checksum error is in the queue;
2630 * then it could get return from select indicating data available
2631 * but then block when reading it. Add special case code
2632 * to work around these arguably broken applications.
2634 __poll_t
udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2636 __poll_t mask
= datagram_poll(file
, sock
, wait
);
2637 struct sock
*sk
= sock
->sk
;
2639 if (!skb_queue_empty(&udp_sk(sk
)->reader_queue
))
2640 mask
|= EPOLLIN
| EPOLLRDNORM
;
2642 /* Check for false positives due to checksum errors */
2643 if ((mask
& EPOLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2644 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && first_packet_length(sk
) == -1)
2645 mask
&= ~(EPOLLIN
| EPOLLRDNORM
);
2650 EXPORT_SYMBOL(udp_poll
);
2652 int udp_abort(struct sock
*sk
, int err
)
2657 sk
->sk_error_report(sk
);
2658 __udp_disconnect(sk
, 0);
2664 EXPORT_SYMBOL_GPL(udp_abort
);
2666 struct proto udp_prot
= {
2668 .owner
= THIS_MODULE
,
2669 .close
= udp_lib_close
,
2670 .pre_connect
= udp_pre_connect
,
2671 .connect
= ip4_datagram_connect
,
2672 .disconnect
= udp_disconnect
,
2674 .init
= udp_init_sock
,
2675 .destroy
= udp_destroy_sock
,
2676 .setsockopt
= udp_setsockopt
,
2677 .getsockopt
= udp_getsockopt
,
2678 .sendmsg
= udp_sendmsg
,
2679 .recvmsg
= udp_recvmsg
,
2680 .sendpage
= udp_sendpage
,
2681 .release_cb
= ip4_datagram_release_cb
,
2682 .hash
= udp_lib_hash
,
2683 .unhash
= udp_lib_unhash
,
2684 .rehash
= udp_v4_rehash
,
2685 .get_port
= udp_v4_get_port
,
2686 .memory_allocated
= &udp_memory_allocated
,
2687 .sysctl_mem
= sysctl_udp_mem
,
2688 .sysctl_wmem_offset
= offsetof(struct net
, ipv4
.sysctl_udp_wmem_min
),
2689 .sysctl_rmem_offset
= offsetof(struct net
, ipv4
.sysctl_udp_rmem_min
),
2690 .obj_size
= sizeof(struct udp_sock
),
2691 .h
.udp_table
= &udp_table
,
2692 #ifdef CONFIG_COMPAT
2693 .compat_setsockopt
= compat_udp_setsockopt
,
2694 .compat_getsockopt
= compat_udp_getsockopt
,
2696 .diag_destroy
= udp_abort
,
2698 EXPORT_SYMBOL(udp_prot
);
2700 /* ------------------------------------------------------------------------ */
2701 #ifdef CONFIG_PROC_FS
2703 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2706 struct udp_seq_afinfo
*afinfo
= PDE_DATA(file_inode(seq
->file
));
2707 struct udp_iter_state
*state
= seq
->private;
2708 struct net
*net
= seq_file_net(seq
);
2710 for (state
->bucket
= start
; state
->bucket
<= afinfo
->udp_table
->mask
;
2712 struct udp_hslot
*hslot
= &afinfo
->udp_table
->hash
[state
->bucket
];
2714 if (hlist_empty(&hslot
->head
))
2717 spin_lock_bh(&hslot
->lock
);
2718 sk_for_each(sk
, &hslot
->head
) {
2719 if (!net_eq(sock_net(sk
), net
))
2721 if (sk
->sk_family
== afinfo
->family
)
2724 spin_unlock_bh(&hslot
->lock
);
2731 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2733 struct udp_seq_afinfo
*afinfo
= PDE_DATA(file_inode(seq
->file
));
2734 struct udp_iter_state
*state
= seq
->private;
2735 struct net
*net
= seq_file_net(seq
);
2739 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= afinfo
->family
));
2742 if (state
->bucket
<= afinfo
->udp_table
->mask
)
2743 spin_unlock_bh(&afinfo
->udp_table
->hash
[state
->bucket
].lock
);
2744 return udp_get_first(seq
, state
->bucket
+ 1);
2749 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2751 struct sock
*sk
= udp_get_first(seq
, 0);
2754 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2756 return pos
? NULL
: sk
;
2759 void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2761 struct udp_iter_state
*state
= seq
->private;
2762 state
->bucket
= MAX_UDP_PORTS
;
2764 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2766 EXPORT_SYMBOL(udp_seq_start
);
2768 void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2772 if (v
== SEQ_START_TOKEN
)
2773 sk
= udp_get_idx(seq
, 0);
2775 sk
= udp_get_next(seq
, v
);
2780 EXPORT_SYMBOL(udp_seq_next
);
2782 void udp_seq_stop(struct seq_file
*seq
, void *v
)
2784 struct udp_seq_afinfo
*afinfo
= PDE_DATA(file_inode(seq
->file
));
2785 struct udp_iter_state
*state
= seq
->private;
2787 if (state
->bucket
<= afinfo
->udp_table
->mask
)
2788 spin_unlock_bh(&afinfo
->udp_table
->hash
[state
->bucket
].lock
);
2790 EXPORT_SYMBOL(udp_seq_stop
);
2792 /* ------------------------------------------------------------------------ */
2793 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2796 struct inet_sock
*inet
= inet_sk(sp
);
2797 __be32 dest
= inet
->inet_daddr
;
2798 __be32 src
= inet
->inet_rcv_saddr
;
2799 __u16 destp
= ntohs(inet
->inet_dport
);
2800 __u16 srcp
= ntohs(inet
->inet_sport
);
2802 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2803 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2804 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2805 sk_wmem_alloc_get(sp
),
2808 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2810 refcount_read(&sp
->sk_refcnt
), sp
,
2811 atomic_read(&sp
->sk_drops
));
2814 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2816 seq_setwidth(seq
, 127);
2817 if (v
== SEQ_START_TOKEN
)
2818 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2819 "rx_queue tr tm->when retrnsmt uid timeout "
2820 "inode ref pointer drops");
2822 struct udp_iter_state
*state
= seq
->private;
2824 udp4_format_sock(v
, seq
, state
->bucket
);
2830 const struct seq_operations udp_seq_ops
= {
2831 .start
= udp_seq_start
,
2832 .next
= udp_seq_next
,
2833 .stop
= udp_seq_stop
,
2834 .show
= udp4_seq_show
,
2836 EXPORT_SYMBOL(udp_seq_ops
);
2838 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2840 .udp_table
= &udp_table
,
2843 static int __net_init
udp4_proc_init_net(struct net
*net
)
2845 if (!proc_create_net_data("udp", 0444, net
->proc_net
, &udp_seq_ops
,
2846 sizeof(struct udp_iter_state
), &udp4_seq_afinfo
))
2851 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2853 remove_proc_entry("udp", net
->proc_net
);
2856 static struct pernet_operations udp4_net_ops
= {
2857 .init
= udp4_proc_init_net
,
2858 .exit
= udp4_proc_exit_net
,
2861 int __init
udp4_proc_init(void)
2863 return register_pernet_subsys(&udp4_net_ops
);
2866 void udp4_proc_exit(void)
2868 unregister_pernet_subsys(&udp4_net_ops
);
2870 #endif /* CONFIG_PROC_FS */
2872 static __initdata
unsigned long uhash_entries
;
2873 static int __init
set_uhash_entries(char *str
)
2880 ret
= kstrtoul(str
, 0, &uhash_entries
);
2884 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2885 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2888 __setup("uhash_entries=", set_uhash_entries
);
2890 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2894 table
->hash
= alloc_large_system_hash(name
,
2895 2 * sizeof(struct udp_hslot
),
2897 21, /* one slot per 2 MB */
2901 UDP_HTABLE_SIZE_MIN
,
2904 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2905 for (i
= 0; i
<= table
->mask
; i
++) {
2906 INIT_HLIST_HEAD(&table
->hash
[i
].head
);
2907 table
->hash
[i
].count
= 0;
2908 spin_lock_init(&table
->hash
[i
].lock
);
2910 for (i
= 0; i
<= table
->mask
; i
++) {
2911 INIT_HLIST_HEAD(&table
->hash2
[i
].head
);
2912 table
->hash2
[i
].count
= 0;
2913 spin_lock_init(&table
->hash2
[i
].lock
);
2917 u32
udp_flow_hashrnd(void)
2919 static u32 hashrnd __read_mostly
;
2921 net_get_random_once(&hashrnd
, sizeof(hashrnd
));
2925 EXPORT_SYMBOL(udp_flow_hashrnd
);
2927 static void __udp_sysctl_init(struct net
*net
)
2929 net
->ipv4
.sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2930 net
->ipv4
.sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
2932 #ifdef CONFIG_NET_L3_MASTER_DEV
2933 net
->ipv4
.sysctl_udp_l3mdev_accept
= 0;
2937 static int __net_init
udp_sysctl_init(struct net
*net
)
2939 __udp_sysctl_init(net
);
2943 static struct pernet_operations __net_initdata udp_sysctl_ops
= {
2944 .init
= udp_sysctl_init
,
2947 void __init
udp_init(void)
2949 unsigned long limit
;
2952 udp_table_init(&udp_table
, "UDP");
2953 limit
= nr_free_buffer_pages() / 8;
2954 limit
= max(limit
, 128UL);
2955 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2956 sysctl_udp_mem
[1] = limit
;
2957 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2959 __udp_sysctl_init(&init_net
);
2961 /* 16 spinlocks per cpu */
2962 udp_busylocks_log
= ilog2(nr_cpu_ids
) + 4;
2963 udp_busylocks
= kmalloc(sizeof(spinlock_t
) << udp_busylocks_log
,
2966 panic("UDP: failed to alloc udp_busylocks\n");
2967 for (i
= 0; i
< (1U << udp_busylocks_log
); i
++)
2968 spin_lock_init(udp_busylocks
+ i
);
2970 if (register_pernet_subsys(&udp_sysctl_ops
))
2971 panic("UDP: failed to init sysctl parameters.\n");