1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The User Datagram Protocol (UDP).
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Hirokazu Takahashi, <taka@valinux.co.jp>
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72 * James Chapman : Add L2TP encapsulation type.
75 #define pr_fmt(fmt) "UDP: " fmt
77 #include <linux/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <linux/memblock.h>
80 #include <linux/highmem.h>
81 #include <linux/swap.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <trace/events/udp.h>
108 #include <linux/static_key.h>
109 #include <linux/btf_ids.h>
110 #include <trace/events/skb.h>
111 #include <net/busy_poll.h>
112 #include "udp_impl.h"
113 #include <net/sock_reuseport.h>
114 #include <net/addrconf.h>
115 #include <net/udp_tunnel.h>
116 #if IS_ENABLED(CONFIG_IPV6)
117 #include <net/ipv6_stubs.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 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
133 const struct udp_hslot
*hslot
,
134 unsigned long *bitmap
,
135 struct sock
*sk
, unsigned int log
)
138 kuid_t uid
= sock_i_uid(sk
);
140 sk_for_each(sk2
, &hslot
->head
) {
141 if (net_eq(sock_net(sk2
), net
) &&
143 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
144 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
145 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
146 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
147 inet_rcv_saddr_equal(sk
, sk2
, true)) {
148 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
149 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
150 uid_eq(uid
, sock_i_uid(sk2
))) {
156 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
165 * Note: we still hold spinlock of primary hash chain, so no other writer
166 * can insert/delete a socket with local_port == num
168 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
169 struct udp_hslot
*hslot2
,
173 kuid_t uid
= sock_i_uid(sk
);
176 spin_lock(&hslot2
->lock
);
177 udp_portaddr_for_each_entry(sk2
, &hslot2
->head
) {
178 if (net_eq(sock_net(sk2
), net
) &&
180 (udp_sk(sk2
)->udp_port_hash
== num
) &&
181 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
182 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
183 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
184 inet_rcv_saddr_equal(sk
, sk2
, true)) {
185 if (sk2
->sk_reuseport
&& sk
->sk_reuseport
&&
186 !rcu_access_pointer(sk
->sk_reuseport_cb
) &&
187 uid_eq(uid
, sock_i_uid(sk2
))) {
195 spin_unlock(&hslot2
->lock
);
199 static int udp_reuseport_add_sock(struct sock
*sk
, struct udp_hslot
*hslot
)
201 struct net
*net
= sock_net(sk
);
202 kuid_t uid
= sock_i_uid(sk
);
205 sk_for_each(sk2
, &hslot
->head
) {
206 if (net_eq(sock_net(sk2
), net
) &&
208 sk2
->sk_family
== sk
->sk_family
&&
209 ipv6_only_sock(sk2
) == ipv6_only_sock(sk
) &&
210 (udp_sk(sk2
)->udp_port_hash
== udp_sk(sk
)->udp_port_hash
) &&
211 (sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
212 sk2
->sk_reuseport
&& uid_eq(uid
, sock_i_uid(sk2
)) &&
213 inet_rcv_saddr_equal(sk
, sk2
, false)) {
214 return reuseport_add_sock(sk
, sk2
,
215 inet_rcv_saddr_any(sk
));
219 return reuseport_alloc(sk
, inet_rcv_saddr_any(sk
));
223 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
225 * @sk: socket struct in question
226 * @snum: port number to look up
227 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
230 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
231 unsigned int hash2_nulladdr
)
233 struct udp_hslot
*hslot
, *hslot2
;
234 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
236 struct net
*net
= sock_net(sk
);
239 int low
, high
, remaining
;
241 unsigned short first
, last
;
242 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
244 inet_get_local_port_range(net
, &low
, &high
);
245 remaining
= (high
- low
) + 1;
247 rand
= prandom_u32();
248 first
= reciprocal_scale(rand
, remaining
) + low
;
250 * force rand to be an odd multiple of UDP_HTABLE_SIZE
252 rand
= (rand
| 1) * (udptable
->mask
+ 1);
253 last
= first
+ udptable
->mask
+ 1;
255 hslot
= udp_hashslot(udptable
, net
, first
);
256 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
257 spin_lock_bh(&hslot
->lock
);
258 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
263 * Iterate on all possible values of snum for this hash.
264 * Using steps of an odd multiple of UDP_HTABLE_SIZE
265 * give us randomization and full range coverage.
268 if (low
<= snum
&& snum
<= high
&&
269 !test_bit(snum
>> udptable
->log
, bitmap
) &&
270 !inet_is_local_reserved_port(net
, snum
))
273 } while (snum
!= first
);
274 spin_unlock_bh(&hslot
->lock
);
276 } while (++first
!= last
);
279 hslot
= udp_hashslot(udptable
, net
, snum
);
280 spin_lock_bh(&hslot
->lock
);
281 if (hslot
->count
> 10) {
283 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
285 slot2
&= udptable
->mask
;
286 hash2_nulladdr
&= udptable
->mask
;
288 hslot2
= udp_hashslot2(udptable
, slot2
);
289 if (hslot
->count
< hslot2
->count
)
290 goto scan_primary_hash
;
292 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
, sk
);
293 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
294 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
295 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
304 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
, 0))
308 inet_sk(sk
)->inet_num
= snum
;
309 udp_sk(sk
)->udp_port_hash
= snum
;
310 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
311 if (sk_unhashed(sk
)) {
312 if (sk
->sk_reuseport
&&
313 udp_reuseport_add_sock(sk
, hslot
)) {
314 inet_sk(sk
)->inet_num
= 0;
315 udp_sk(sk
)->udp_port_hash
= 0;
316 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
320 sk_add_node_rcu(sk
, &hslot
->head
);
322 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
324 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
325 spin_lock(&hslot2
->lock
);
326 if (IS_ENABLED(CONFIG_IPV6
) && sk
->sk_reuseport
&&
327 sk
->sk_family
== AF_INET6
)
328 hlist_add_tail_rcu(&udp_sk(sk
)->udp_portaddr_node
,
331 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
334 spin_unlock(&hslot2
->lock
);
336 sock_set_flag(sk
, SOCK_RCU_FREE
);
339 spin_unlock_bh(&hslot
->lock
);
343 EXPORT_SYMBOL(udp_lib_get_port
);
345 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
347 unsigned int hash2_nulladdr
=
348 ipv4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
349 unsigned int hash2_partial
=
350 ipv4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
352 /* precompute partial secondary hash */
353 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
354 return udp_lib_get_port(sk
, snum
, hash2_nulladdr
);
357 static int compute_score(struct sock
*sk
, struct net
*net
,
358 __be32 saddr
, __be16 sport
,
359 __be32 daddr
, unsigned short hnum
,
363 struct inet_sock
*inet
;
366 if (!net_eq(sock_net(sk
), net
) ||
367 udp_sk(sk
)->udp_port_hash
!= hnum
||
371 if (sk
->sk_rcv_saddr
!= daddr
)
374 score
= (sk
->sk_family
== PF_INET
) ? 2 : 1;
377 if (inet
->inet_daddr
) {
378 if (inet
->inet_daddr
!= saddr
)
383 if (inet
->inet_dport
) {
384 if (inet
->inet_dport
!= sport
)
389 dev_match
= udp_sk_bound_dev_eq(net
, sk
->sk_bound_dev_if
,
395 if (READ_ONCE(sk
->sk_incoming_cpu
) == raw_smp_processor_id())
400 static u32
udp_ehashfn(const struct net
*net
, const __be32 laddr
,
401 const __u16 lport
, const __be32 faddr
,
404 static u32 udp_ehash_secret __read_mostly
;
406 net_get_random_once(&udp_ehash_secret
, sizeof(udp_ehash_secret
));
408 return __inet_ehashfn(laddr
, lport
, faddr
, fport
,
409 udp_ehash_secret
+ net_hash_mix(net
));
412 static struct sock
*lookup_reuseport(struct net
*net
, struct sock
*sk
,
414 __be32 saddr
, __be16 sport
,
415 __be32 daddr
, unsigned short hnum
)
417 struct sock
*reuse_sk
= NULL
;
420 if (sk
->sk_reuseport
&& sk
->sk_state
!= TCP_ESTABLISHED
) {
421 hash
= udp_ehashfn(net
, daddr
, hnum
, saddr
, sport
);
422 reuse_sk
= reuseport_select_sock(sk
, hash
, skb
,
423 sizeof(struct udphdr
));
428 /* called with rcu_read_lock() */
429 static struct sock
*udp4_lib_lookup2(struct net
*net
,
430 __be32 saddr
, __be16 sport
,
431 __be32 daddr
, unsigned int hnum
,
433 struct udp_hslot
*hslot2
,
436 struct sock
*sk
, *result
;
441 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
442 score
= compute_score(sk
, net
, saddr
, sport
,
443 daddr
, hnum
, dif
, sdif
);
444 if (score
> badness
) {
445 result
= lookup_reuseport(net
, sk
, skb
,
446 saddr
, sport
, daddr
, hnum
);
447 /* Fall back to scoring if group has connections */
448 if (result
&& !reuseport_has_conns(sk
, false))
451 result
= result
? : sk
;
458 static struct sock
*udp4_lookup_run_bpf(struct net
*net
,
459 struct udp_table
*udptable
,
461 __be32 saddr
, __be16 sport
,
462 __be32 daddr
, u16 hnum
)
464 struct sock
*sk
, *reuse_sk
;
467 if (udptable
!= &udp_table
)
468 return NULL
; /* only UDP is supported */
470 no_reuseport
= bpf_sk_lookup_run_v4(net
, IPPROTO_UDP
,
471 saddr
, sport
, daddr
, hnum
, &sk
);
472 if (no_reuseport
|| IS_ERR_OR_NULL(sk
))
475 reuse_sk
= lookup_reuseport(net
, sk
, skb
, saddr
, sport
, daddr
, hnum
);
481 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
482 * harder than this. -DaveM
484 struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
485 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
,
486 int sdif
, struct udp_table
*udptable
, struct sk_buff
*skb
)
488 unsigned short hnum
= ntohs(dport
);
489 unsigned int hash2
, slot2
;
490 struct udp_hslot
*hslot2
;
491 struct sock
*result
, *sk
;
493 hash2
= ipv4_portaddr_hash(net
, daddr
, hnum
);
494 slot2
= hash2
& udptable
->mask
;
495 hslot2
= &udptable
->hash2
[slot2
];
497 /* Lookup connected or non-wildcard socket */
498 result
= udp4_lib_lookup2(net
, saddr
, sport
,
499 daddr
, hnum
, dif
, sdif
,
501 if (!IS_ERR_OR_NULL(result
) && result
->sk_state
== TCP_ESTABLISHED
)
504 /* Lookup redirect from BPF */
505 if (static_branch_unlikely(&bpf_sk_lookup_enabled
)) {
506 sk
= udp4_lookup_run_bpf(net
, udptable
, skb
,
507 saddr
, sport
, daddr
, hnum
);
514 /* Got non-wildcard socket or error on first lookup */
518 /* Lookup wildcard sockets */
519 hash2
= ipv4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
520 slot2
= hash2
& udptable
->mask
;
521 hslot2
= &udptable
->hash2
[slot2
];
523 result
= udp4_lib_lookup2(net
, saddr
, sport
,
524 htonl(INADDR_ANY
), hnum
, dif
, sdif
,
531 EXPORT_SYMBOL_GPL(__udp4_lib_lookup
);
533 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
534 __be16 sport
, __be16 dport
,
535 struct udp_table
*udptable
)
537 const struct iphdr
*iph
= ip_hdr(skb
);
539 return __udp4_lib_lookup(dev_net(skb
->dev
), iph
->saddr
, sport
,
540 iph
->daddr
, dport
, inet_iif(skb
),
541 inet_sdif(skb
), udptable
, skb
);
544 struct sock
*udp4_lib_lookup_skb(struct sk_buff
*skb
,
545 __be16 sport
, __be16 dport
)
547 const struct iphdr
*iph
= ip_hdr(skb
);
549 return __udp4_lib_lookup(dev_net(skb
->dev
), iph
->saddr
, sport
,
550 iph
->daddr
, dport
, inet_iif(skb
),
551 inet_sdif(skb
), &udp_table
, NULL
);
553 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb
);
555 /* Must be called under rcu_read_lock().
556 * Does increment socket refcount.
558 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
559 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
560 __be32 daddr
, __be16 dport
, int dif
)
564 sk
= __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
,
565 dif
, 0, &udp_table
, NULL
);
566 if (sk
&& !refcount_inc_not_zero(&sk
->sk_refcnt
))
570 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
573 static inline bool __udp_is_mcast_sock(struct net
*net
, struct sock
*sk
,
574 __be16 loc_port
, __be32 loc_addr
,
575 __be16 rmt_port
, __be32 rmt_addr
,
576 int dif
, int sdif
, unsigned short hnum
)
578 struct inet_sock
*inet
= inet_sk(sk
);
580 if (!net_eq(sock_net(sk
), net
) ||
581 udp_sk(sk
)->udp_port_hash
!= hnum
||
582 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
583 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
584 (inet
->inet_rcv_saddr
&& inet
->inet_rcv_saddr
!= loc_addr
) ||
585 ipv6_only_sock(sk
) ||
586 !udp_sk_bound_dev_eq(net
, sk
->sk_bound_dev_if
, dif
, sdif
))
588 if (!ip_mc_sf_allow(sk
, loc_addr
, rmt_addr
, dif
, sdif
))
593 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key
);
594 void udp_encap_enable(void)
596 static_branch_inc(&udp_encap_needed_key
);
598 EXPORT_SYMBOL(udp_encap_enable
);
600 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
601 * through error handlers in encapsulations looking for a match.
603 static int __udp4_lib_err_encap_no_sk(struct sk_buff
*skb
, u32 info
)
607 for (i
= 0; i
< MAX_IPTUN_ENCAP_OPS
; i
++) {
608 int (*handler
)(struct sk_buff
*skb
, u32 info
);
609 const struct ip_tunnel_encap_ops
*encap
;
611 encap
= rcu_dereference(iptun_encaps
[i
]);
614 handler
= encap
->err_handler
;
615 if (handler
&& !handler(skb
, info
))
622 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
623 * reversing source and destination port: this will match tunnels that force the
624 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
625 * lwtunnels might actually break this assumption by being configured with
626 * different destination ports on endpoints, in this case we won't be able to
627 * trace ICMP messages back to them.
629 * If this doesn't match any socket, probe tunnels with arbitrary destination
630 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
631 * we've sent packets to won't necessarily match the local destination port.
633 * Then ask the tunnel implementation to match the error against a valid
636 * Return an error if we can't find a match, the socket if we need further
637 * processing, zero otherwise.
639 static struct sock
*__udp4_lib_err_encap(struct net
*net
,
640 const struct iphdr
*iph
,
642 struct udp_table
*udptable
,
643 struct sk_buff
*skb
, u32 info
)
645 int network_offset
, transport_offset
;
648 network_offset
= skb_network_offset(skb
);
649 transport_offset
= skb_transport_offset(skb
);
651 /* Network header needs to point to the outer IPv4 header inside ICMP */
652 skb_reset_network_header(skb
);
654 /* Transport header needs to point to the UDP header */
655 skb_set_transport_header(skb
, iph
->ihl
<< 2);
657 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->source
,
658 iph
->saddr
, uh
->dest
, skb
->dev
->ifindex
, 0,
661 int (*lookup
)(struct sock
*sk
, struct sk_buff
*skb
);
662 struct udp_sock
*up
= udp_sk(sk
);
664 lookup
= READ_ONCE(up
->encap_err_lookup
);
665 if (!lookup
|| lookup(sk
, skb
))
670 sk
= ERR_PTR(__udp4_lib_err_encap_no_sk(skb
, info
));
672 skb_set_transport_header(skb
, transport_offset
);
673 skb_set_network_header(skb
, network_offset
);
679 * This routine is called by the ICMP module when it gets some
680 * sort of error condition. If err < 0 then the socket should
681 * be closed and the error returned to the user. If err > 0
682 * it's just the icmp type << 8 | icmp code.
683 * Header points to the ip header of the error packet. We move
684 * on past this. Then (as it used to claim before adjustment)
685 * header points to the first 8 bytes of the udp header. We need
686 * to find the appropriate port.
689 int __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
691 struct inet_sock
*inet
;
692 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
693 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
694 const int type
= icmp_hdr(skb
)->type
;
695 const int code
= icmp_hdr(skb
)->code
;
700 struct net
*net
= dev_net(skb
->dev
);
702 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
703 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
,
704 inet_sdif(skb
), udptable
, NULL
);
706 /* No socket for error: try tunnels before discarding */
707 sk
= ERR_PTR(-ENOENT
);
708 if (static_branch_unlikely(&udp_encap_needed_key
)) {
709 sk
= __udp4_lib_err_encap(net
, iph
, uh
, udptable
, skb
,
716 __ICMP_INC_STATS(net
, ICMP_MIB_INERRORS
);
729 case ICMP_TIME_EXCEEDED
:
732 case ICMP_SOURCE_QUENCH
:
734 case ICMP_PARAMETERPROB
:
738 case ICMP_DEST_UNREACH
:
739 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
740 ipv4_sk_update_pmtu(skb
, sk
, info
);
741 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
749 if (code
<= NR_ICMP_UNREACH
) {
750 harderr
= icmp_err_convert
[code
].fatal
;
751 err
= icmp_err_convert
[code
].errno
;
755 ipv4_sk_redirect(skb
, sk
);
760 * RFC1122: OK. Passes ICMP errors back to application, as per
764 /* ...not for tunnels though: we don't have a sending socket */
767 if (!inet
->recverr
) {
768 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
771 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
774 sk
->sk_error_report(sk
);
779 int udp_err(struct sk_buff
*skb
, u32 info
)
781 return __udp4_lib_err(skb
, info
, &udp_table
);
785 * Throw away all pending data and cancel the corking. Socket is locked.
787 void udp_flush_pending_frames(struct sock
*sk
)
789 struct udp_sock
*up
= udp_sk(sk
);
794 ip_flush_pending_frames(sk
);
797 EXPORT_SYMBOL(udp_flush_pending_frames
);
800 * udp4_hwcsum - handle outgoing HW checksumming
801 * @skb: sk_buff containing the filled-in UDP header
802 * (checksum field must be zeroed out)
803 * @src: source IP address
804 * @dst: destination IP address
806 void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
808 struct udphdr
*uh
= udp_hdr(skb
);
809 int offset
= skb_transport_offset(skb
);
810 int len
= skb
->len
- offset
;
814 if (!skb_has_frag_list(skb
)) {
816 * Only one fragment on the socket.
818 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
819 skb
->csum_offset
= offsetof(struct udphdr
, check
);
820 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
823 struct sk_buff
*frags
;
826 * HW-checksum won't work as there are two or more
827 * fragments on the socket so that all csums of sk_buffs
830 skb_walk_frags(skb
, frags
) {
831 csum
= csum_add(csum
, frags
->csum
);
835 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
836 skb
->ip_summed
= CHECKSUM_NONE
;
838 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
840 uh
->check
= CSUM_MANGLED_0
;
843 EXPORT_SYMBOL_GPL(udp4_hwcsum
);
845 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
846 * for the simple case like when setting the checksum for a UDP tunnel.
848 void udp_set_csum(bool nocheck
, struct sk_buff
*skb
,
849 __be32 saddr
, __be32 daddr
, int len
)
851 struct udphdr
*uh
= udp_hdr(skb
);
855 } else if (skb_is_gso(skb
)) {
856 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
857 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
859 uh
->check
= udp_v4_check(len
, saddr
, daddr
, lco_csum(skb
));
861 uh
->check
= CSUM_MANGLED_0
;
863 skb
->ip_summed
= CHECKSUM_PARTIAL
;
864 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
865 skb
->csum_offset
= offsetof(struct udphdr
, check
);
866 uh
->check
= ~udp_v4_check(len
, saddr
, daddr
, 0);
869 EXPORT_SYMBOL(udp_set_csum
);
871 static int udp_send_skb(struct sk_buff
*skb
, struct flowi4
*fl4
,
872 struct inet_cork
*cork
)
874 struct sock
*sk
= skb
->sk
;
875 struct inet_sock
*inet
= inet_sk(sk
);
878 int is_udplite
= IS_UDPLITE(sk
);
879 int offset
= skb_transport_offset(skb
);
880 int len
= skb
->len
- offset
;
881 int datalen
= len
- sizeof(*uh
);
885 * Create a UDP header
888 uh
->source
= inet
->inet_sport
;
889 uh
->dest
= fl4
->fl4_dport
;
890 uh
->len
= htons(len
);
893 if (cork
->gso_size
) {
894 const int hlen
= skb_network_header_len(skb
) +
895 sizeof(struct udphdr
);
897 if (hlen
+ cork
->gso_size
> cork
->fragsize
) {
901 if (skb
->len
> cork
->gso_size
* UDP_MAX_SEGMENTS
) {
905 if (sk
->sk_no_check_tx
) {
909 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
|| is_udplite
||
910 dst_xfrm(skb_dst(skb
))) {
915 if (datalen
> cork
->gso_size
) {
916 skb_shinfo(skb
)->gso_size
= cork
->gso_size
;
917 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP_L4
;
918 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(datalen
,
924 if (is_udplite
) /* UDP-Lite */
925 csum
= udplite_csum(skb
);
927 else if (sk
->sk_no_check_tx
) { /* UDP csum off */
929 skb
->ip_summed
= CHECKSUM_NONE
;
932 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
935 udp4_hwcsum(skb
, fl4
->saddr
, fl4
->daddr
);
939 csum
= udp_csum(skb
);
941 /* add protocol-dependent pseudo-header */
942 uh
->check
= csum_tcpudp_magic(fl4
->saddr
, fl4
->daddr
, len
,
943 sk
->sk_protocol
, csum
);
945 uh
->check
= CSUM_MANGLED_0
;
948 err
= ip_send_skb(sock_net(sk
), skb
);
950 if (err
== -ENOBUFS
&& !inet
->recverr
) {
951 UDP_INC_STATS(sock_net(sk
),
952 UDP_MIB_SNDBUFERRORS
, is_udplite
);
956 UDP_INC_STATS(sock_net(sk
),
957 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
962 * Push out all pending data as one UDP datagram. Socket is locked.
964 int udp_push_pending_frames(struct sock
*sk
)
966 struct udp_sock
*up
= udp_sk(sk
);
967 struct inet_sock
*inet
= inet_sk(sk
);
968 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
972 skb
= ip_finish_skb(sk
, fl4
);
976 err
= udp_send_skb(skb
, fl4
, &inet
->cork
.base
);
983 EXPORT_SYMBOL(udp_push_pending_frames
);
985 static int __udp_cmsg_send(struct cmsghdr
*cmsg
, u16
*gso_size
)
987 switch (cmsg
->cmsg_type
) {
989 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(__u16
)))
991 *gso_size
= *(__u16
*)CMSG_DATA(cmsg
);
998 int udp_cmsg_send(struct sock
*sk
, struct msghdr
*msg
, u16
*gso_size
)
1000 struct cmsghdr
*cmsg
;
1001 bool need_ip
= false;
1004 for_each_cmsghdr(cmsg
, msg
) {
1005 if (!CMSG_OK(msg
, cmsg
))
1008 if (cmsg
->cmsg_level
!= SOL_UDP
) {
1013 err
= __udp_cmsg_send(cmsg
, gso_size
);
1020 EXPORT_SYMBOL_GPL(udp_cmsg_send
);
1022 int udp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
1024 struct inet_sock
*inet
= inet_sk(sk
);
1025 struct udp_sock
*up
= udp_sk(sk
);
1026 DECLARE_SOCKADDR(struct sockaddr_in
*, usin
, msg
->msg_name
);
1027 struct flowi4 fl4_stack
;
1030 struct ipcm_cookie ipc
;
1031 struct rtable
*rt
= NULL
;
1034 __be32 daddr
, faddr
, saddr
;
1037 int err
, is_udplite
= IS_UDPLITE(sk
);
1038 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
1039 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
1040 struct sk_buff
*skb
;
1041 struct ip_options_data opt_copy
;
1050 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
1053 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1055 fl4
= &inet
->cork
.fl
.u
.ip4
;
1058 * There are pending frames.
1059 * The socket lock must be held while it's corked.
1062 if (likely(up
->pending
)) {
1063 if (unlikely(up
->pending
!= AF_INET
)) {
1067 goto do_append_data
;
1071 ulen
+= sizeof(struct udphdr
);
1074 * Get and verify the address.
1077 if (msg
->msg_namelen
< sizeof(*usin
))
1079 if (usin
->sin_family
!= AF_INET
) {
1080 if (usin
->sin_family
!= AF_UNSPEC
)
1081 return -EAFNOSUPPORT
;
1084 daddr
= usin
->sin_addr
.s_addr
;
1085 dport
= usin
->sin_port
;
1089 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1090 return -EDESTADDRREQ
;
1091 daddr
= inet
->inet_daddr
;
1092 dport
= inet
->inet_dport
;
1093 /* Open fast path for connected socket.
1094 Route will not be used, if at least one option is set.
1099 ipcm_init_sk(&ipc
, inet
);
1100 ipc
.gso_size
= up
->gso_size
;
1102 if (msg
->msg_controllen
) {
1103 err
= udp_cmsg_send(sk
, msg
, &ipc
.gso_size
);
1105 err
= ip_cmsg_send(sk
, msg
, &ipc
,
1106 sk
->sk_family
== AF_INET6
);
1107 if (unlikely(err
< 0)) {
1116 struct ip_options_rcu
*inet_opt
;
1119 inet_opt
= rcu_dereference(inet
->inet_opt
);
1121 memcpy(&opt_copy
, inet_opt
,
1122 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
1123 ipc
.opt
= &opt_copy
.opt
;
1128 if (cgroup_bpf_enabled
&& !connected
) {
1129 err
= BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk
,
1130 (struct sockaddr
*)usin
, &ipc
.addr
);
1134 if (usin
->sin_port
== 0) {
1135 /* BPF program set invalid port. Reject it. */
1139 daddr
= usin
->sin_addr
.s_addr
;
1140 dport
= usin
->sin_port
;
1145 ipc
.addr
= faddr
= daddr
;
1147 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
1152 faddr
= ipc
.opt
->opt
.faddr
;
1155 tos
= get_rttos(&ipc
, inet
);
1156 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
1157 (msg
->msg_flags
& MSG_DONTROUTE
) ||
1158 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
1163 if (ipv4_is_multicast(daddr
)) {
1164 if (!ipc
.oif
|| netif_index_is_l3_master(sock_net(sk
), ipc
.oif
))
1165 ipc
.oif
= inet
->mc_index
;
1167 saddr
= inet
->mc_addr
;
1169 } else if (!ipc
.oif
) {
1170 ipc
.oif
= inet
->uc_index
;
1171 } else if (ipv4_is_lbcast(daddr
) && inet
->uc_index
) {
1172 /* oif is set, packet is to local broadcast and
1173 * uc_index is set. oif is most likely set
1174 * by sk_bound_dev_if. If uc_index != oif check if the
1175 * oif is an L3 master and uc_index is an L3 slave.
1176 * If so, we want to allow the send using the uc_index.
1178 if (ipc
.oif
!= inet
->uc_index
&&
1179 ipc
.oif
== l3mdev_master_ifindex_by_index(sock_net(sk
),
1181 ipc
.oif
= inet
->uc_index
;
1186 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
1189 struct net
*net
= sock_net(sk
);
1190 __u8 flow_flags
= inet_sk_flowi_flags(sk
);
1194 flowi4_init_output(fl4
, ipc
.oif
, ipc
.sockc
.mark
, tos
,
1195 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
1197 faddr
, saddr
, dport
, inet
->inet_sport
,
1200 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
1201 rt
= ip_route_output_flow(net
, fl4
, sk
);
1205 if (err
== -ENETUNREACH
)
1206 IP_INC_STATS(net
, IPSTATS_MIB_OUTNOROUTES
);
1211 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
1212 !sock_flag(sk
, SOCK_BROADCAST
))
1215 sk_dst_set(sk
, dst_clone(&rt
->dst
));
1218 if (msg
->msg_flags
&MSG_CONFIRM
)
1224 daddr
= ipc
.addr
= fl4
->daddr
;
1226 /* Lockless fast path for the non-corking case. */
1228 struct inet_cork cork
;
1230 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
, ulen
,
1231 sizeof(struct udphdr
), &ipc
, &rt
,
1232 &cork
, msg
->msg_flags
);
1234 if (!IS_ERR_OR_NULL(skb
))
1235 err
= udp_send_skb(skb
, fl4
, &cork
);
1240 if (unlikely(up
->pending
)) {
1241 /* The socket is already corked while preparing it. */
1242 /* ... which is an evident application bug. --ANK */
1245 net_dbg_ratelimited("socket already corked\n");
1250 * Now cork the socket to pend data.
1252 fl4
= &inet
->cork
.fl
.u
.ip4
;
1255 fl4
->fl4_dport
= dport
;
1256 fl4
->fl4_sport
= inet
->inet_sport
;
1257 up
->pending
= AF_INET
;
1261 err
= ip_append_data(sk
, fl4
, getfrag
, msg
, ulen
,
1262 sizeof(struct udphdr
), &ipc
, &rt
,
1263 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1265 udp_flush_pending_frames(sk
);
1267 err
= udp_push_pending_frames(sk
);
1268 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1280 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1281 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1282 * we don't have a good statistic (IpOutDiscards but it can be too many
1283 * things). We could add another new stat but at least for now that
1284 * seems like overkill.
1286 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1287 UDP_INC_STATS(sock_net(sk
),
1288 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1293 if (msg
->msg_flags
& MSG_PROBE
)
1294 dst_confirm_neigh(&rt
->dst
, &fl4
->daddr
);
1295 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1296 goto back_from_confirm
;
1300 EXPORT_SYMBOL(udp_sendmsg
);
1302 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1303 size_t size
, int flags
)
1305 struct inet_sock
*inet
= inet_sk(sk
);
1306 struct udp_sock
*up
= udp_sk(sk
);
1309 if (flags
& MSG_SENDPAGE_NOTLAST
)
1313 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1315 /* Call udp_sendmsg to specify destination address which
1316 * sendpage interface can't pass.
1317 * This will succeed only when the socket is connected.
1319 ret
= udp_sendmsg(sk
, &msg
, 0);
1326 if (unlikely(!up
->pending
)) {
1329 net_dbg_ratelimited("cork failed\n");
1333 ret
= ip_append_page(sk
, &inet
->cork
.fl
.u
.ip4
,
1334 page
, offset
, size
, flags
);
1335 if (ret
== -EOPNOTSUPP
) {
1337 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1341 udp_flush_pending_frames(sk
);
1346 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1347 ret
= udp_push_pending_frames(sk
);
1355 #define UDP_SKB_IS_STATELESS 0x80000000
1357 /* all head states (dst, sk, nf conntrack) except skb extensions are
1358 * cleared by udp_rcv().
1360 * We need to preserve secpath, if present, to eventually process
1361 * IP_CMSG_PASSSEC at recvmsg() time.
1363 * Other extensions can be cleared.
1365 static bool udp_try_make_stateless(struct sk_buff
*skb
)
1367 if (!skb_has_extensions(skb
))
1370 if (!secpath_exists(skb
)) {
1378 static void udp_set_dev_scratch(struct sk_buff
*skb
)
1380 struct udp_dev_scratch
*scratch
= udp_skb_scratch(skb
);
1382 BUILD_BUG_ON(sizeof(struct udp_dev_scratch
) > sizeof(long));
1383 scratch
->_tsize_state
= skb
->truesize
;
1384 #if BITS_PER_LONG == 64
1385 scratch
->len
= skb
->len
;
1386 scratch
->csum_unnecessary
= !!skb_csum_unnecessary(skb
);
1387 scratch
->is_linear
= !skb_is_nonlinear(skb
);
1389 if (udp_try_make_stateless(skb
))
1390 scratch
->_tsize_state
|= UDP_SKB_IS_STATELESS
;
1393 static void udp_skb_csum_unnecessary_set(struct sk_buff
*skb
)
1395 /* We come here after udp_lib_checksum_complete() returned 0.
1396 * This means that __skb_checksum_complete() might have
1397 * set skb->csum_valid to 1.
1398 * On 64bit platforms, we can set csum_unnecessary
1399 * to true, but only if the skb is not shared.
1401 #if BITS_PER_LONG == 64
1402 if (!skb_shared(skb
))
1403 udp_skb_scratch(skb
)->csum_unnecessary
= true;
1407 static int udp_skb_truesize(struct sk_buff
*skb
)
1409 return udp_skb_scratch(skb
)->_tsize_state
& ~UDP_SKB_IS_STATELESS
;
1412 static bool udp_skb_has_head_state(struct sk_buff
*skb
)
1414 return !(udp_skb_scratch(skb
)->_tsize_state
& UDP_SKB_IS_STATELESS
);
1417 /* fully reclaim rmem/fwd memory allocated for skb */
1418 static void udp_rmem_release(struct sock
*sk
, int size
, int partial
,
1419 bool rx_queue_lock_held
)
1421 struct udp_sock
*up
= udp_sk(sk
);
1422 struct sk_buff_head
*sk_queue
;
1425 if (likely(partial
)) {
1426 up
->forward_deficit
+= size
;
1427 size
= up
->forward_deficit
;
1428 if (size
< (sk
->sk_rcvbuf
>> 2) &&
1429 !skb_queue_empty(&up
->reader_queue
))
1432 size
+= up
->forward_deficit
;
1434 up
->forward_deficit
= 0;
1436 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1437 * if the called don't held it already
1439 sk_queue
= &sk
->sk_receive_queue
;
1440 if (!rx_queue_lock_held
)
1441 spin_lock(&sk_queue
->lock
);
1444 sk
->sk_forward_alloc
+= size
;
1445 amt
= (sk
->sk_forward_alloc
- partial
) & ~(SK_MEM_QUANTUM
- 1);
1446 sk
->sk_forward_alloc
-= amt
;
1449 __sk_mem_reduce_allocated(sk
, amt
>> SK_MEM_QUANTUM_SHIFT
);
1451 atomic_sub(size
, &sk
->sk_rmem_alloc
);
1453 /* this can save us from acquiring the rx queue lock on next receive */
1454 skb_queue_splice_tail_init(sk_queue
, &up
->reader_queue
);
1456 if (!rx_queue_lock_held
)
1457 spin_unlock(&sk_queue
->lock
);
1460 /* Note: called with reader_queue.lock held.
1461 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1462 * This avoids a cache line miss while receive_queue lock is held.
1463 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1465 void udp_skb_destructor(struct sock
*sk
, struct sk_buff
*skb
)
1467 prefetch(&skb
->data
);
1468 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, false);
1470 EXPORT_SYMBOL(udp_skb_destructor
);
1472 /* as above, but the caller held the rx queue lock, too */
1473 static void udp_skb_dtor_locked(struct sock
*sk
, struct sk_buff
*skb
)
1475 prefetch(&skb
->data
);
1476 udp_rmem_release(sk
, udp_skb_truesize(skb
), 1, true);
1479 /* Idea of busylocks is to let producers grab an extra spinlock
1480 * to relieve pressure on the receive_queue spinlock shared by consumer.
1481 * Under flood, this means that only one producer can be in line
1482 * trying to acquire the receive_queue spinlock.
1483 * These busylock can be allocated on a per cpu manner, instead of a
1484 * per socket one (that would consume a cache line per socket)
1486 static int udp_busylocks_log __read_mostly
;
1487 static spinlock_t
*udp_busylocks __read_mostly
;
1489 static spinlock_t
*busylock_acquire(void *ptr
)
1493 busy
= udp_busylocks
+ hash_ptr(ptr
, udp_busylocks_log
);
1498 static void busylock_release(spinlock_t
*busy
)
1504 int __udp_enqueue_schedule_skb(struct sock
*sk
, struct sk_buff
*skb
)
1506 struct sk_buff_head
*list
= &sk
->sk_receive_queue
;
1507 int rmem
, delta
, amt
, err
= -ENOMEM
;
1508 spinlock_t
*busy
= NULL
;
1511 /* try to avoid the costly atomic add/sub pair when the receive
1512 * queue is full; always allow at least a packet
1514 rmem
= atomic_read(&sk
->sk_rmem_alloc
);
1515 if (rmem
> sk
->sk_rcvbuf
)
1518 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1519 * having linear skbs :
1520 * - Reduce memory overhead and thus increase receive queue capacity
1521 * - Less cache line misses at copyout() time
1522 * - Less work at consume_skb() (less alien page frag freeing)
1524 if (rmem
> (sk
->sk_rcvbuf
>> 1)) {
1527 busy
= busylock_acquire(sk
);
1529 size
= skb
->truesize
;
1530 udp_set_dev_scratch(skb
);
1532 /* we drop only if the receive buf is full and the receive
1533 * queue contains some other skb
1535 rmem
= atomic_add_return(size
, &sk
->sk_rmem_alloc
);
1536 if (rmem
> (size
+ (unsigned int)sk
->sk_rcvbuf
))
1539 spin_lock(&list
->lock
);
1540 if (size
>= sk
->sk_forward_alloc
) {
1541 amt
= sk_mem_pages(size
);
1542 delta
= amt
<< SK_MEM_QUANTUM_SHIFT
;
1543 if (!__sk_mem_raise_allocated(sk
, delta
, amt
, SK_MEM_RECV
)) {
1545 spin_unlock(&list
->lock
);
1549 sk
->sk_forward_alloc
+= delta
;
1552 sk
->sk_forward_alloc
-= size
;
1554 /* no need to setup a destructor, we will explicitly release the
1555 * forward allocated memory on dequeue
1557 sock_skb_set_dropcount(sk
, skb
);
1559 __skb_queue_tail(list
, skb
);
1560 spin_unlock(&list
->lock
);
1562 if (!sock_flag(sk
, SOCK_DEAD
))
1563 sk
->sk_data_ready(sk
);
1565 busylock_release(busy
);
1569 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
1572 atomic_inc(&sk
->sk_drops
);
1573 busylock_release(busy
);
1576 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb
);
1578 void udp_destruct_sock(struct sock
*sk
)
1580 /* reclaim completely the forward allocated memory */
1581 struct udp_sock
*up
= udp_sk(sk
);
1582 unsigned int total
= 0;
1583 struct sk_buff
*skb
;
1585 skb_queue_splice_tail_init(&sk
->sk_receive_queue
, &up
->reader_queue
);
1586 while ((skb
= __skb_dequeue(&up
->reader_queue
)) != NULL
) {
1587 total
+= skb
->truesize
;
1590 udp_rmem_release(sk
, total
, 0, true);
1592 inet_sock_destruct(sk
);
1594 EXPORT_SYMBOL_GPL(udp_destruct_sock
);
1596 int udp_init_sock(struct sock
*sk
)
1598 skb_queue_head_init(&udp_sk(sk
)->reader_queue
);
1599 sk
->sk_destruct
= udp_destruct_sock
;
1602 EXPORT_SYMBOL_GPL(udp_init_sock
);
1604 void skb_consume_udp(struct sock
*sk
, struct sk_buff
*skb
, int len
)
1606 if (unlikely(READ_ONCE(sk
->sk_peek_off
) >= 0)) {
1607 bool slow
= lock_sock_fast(sk
);
1609 sk_peek_offset_bwd(sk
, len
);
1610 unlock_sock_fast(sk
, slow
);
1613 if (!skb_unref(skb
))
1616 /* In the more common cases we cleared the head states previously,
1617 * see __udp_queue_rcv_skb().
1619 if (unlikely(udp_skb_has_head_state(skb
)))
1620 skb_release_head_state(skb
);
1621 __consume_stateless_skb(skb
);
1623 EXPORT_SYMBOL_GPL(skb_consume_udp
);
1625 static struct sk_buff
*__first_packet_length(struct sock
*sk
,
1626 struct sk_buff_head
*rcvq
,
1629 struct sk_buff
*skb
;
1631 while ((skb
= skb_peek(rcvq
)) != NULL
) {
1632 if (udp_lib_checksum_complete(skb
)) {
1633 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
,
1635 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
1637 atomic_inc(&sk
->sk_drops
);
1638 __skb_unlink(skb
, rcvq
);
1639 *total
+= skb
->truesize
;
1642 udp_skb_csum_unnecessary_set(skb
);
1650 * first_packet_length - return length of first packet in receive queue
1653 * Drops all bad checksum frames, until a valid one is found.
1654 * Returns the length of found skb, or -1 if none is found.
1656 static int first_packet_length(struct sock
*sk
)
1658 struct sk_buff_head
*rcvq
= &udp_sk(sk
)->reader_queue
;
1659 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1660 struct sk_buff
*skb
;
1664 spin_lock_bh(&rcvq
->lock
);
1665 skb
= __first_packet_length(sk
, rcvq
, &total
);
1666 if (!skb
&& !skb_queue_empty_lockless(sk_queue
)) {
1667 spin_lock(&sk_queue
->lock
);
1668 skb_queue_splice_tail_init(sk_queue
, rcvq
);
1669 spin_unlock(&sk_queue
->lock
);
1671 skb
= __first_packet_length(sk
, rcvq
, &total
);
1673 res
= skb
? skb
->len
: -1;
1675 udp_rmem_release(sk
, total
, 1, false);
1676 spin_unlock_bh(&rcvq
->lock
);
1681 * IOCTL requests applicable to the UDP protocol
1684 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1689 int amount
= sk_wmem_alloc_get(sk
);
1691 return put_user(amount
, (int __user
*)arg
);
1696 int amount
= max_t(int, 0, first_packet_length(sk
));
1698 return put_user(amount
, (int __user
*)arg
);
1702 return -ENOIOCTLCMD
;
1707 EXPORT_SYMBOL(udp_ioctl
);
1709 struct sk_buff
*__skb_recv_udp(struct sock
*sk
, unsigned int flags
,
1710 int noblock
, int *off
, int *err
)
1712 struct sk_buff_head
*sk_queue
= &sk
->sk_receive_queue
;
1713 struct sk_buff_head
*queue
;
1714 struct sk_buff
*last
;
1718 queue
= &udp_sk(sk
)->reader_queue
;
1719 flags
|= noblock
? MSG_DONTWAIT
: 0;
1720 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
1722 struct sk_buff
*skb
;
1724 error
= sock_error(sk
);
1730 spin_lock_bh(&queue
->lock
);
1731 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
, off
,
1734 if (!(flags
& MSG_PEEK
))
1735 udp_skb_destructor(sk
, skb
);
1736 spin_unlock_bh(&queue
->lock
);
1740 if (skb_queue_empty_lockless(sk_queue
)) {
1741 spin_unlock_bh(&queue
->lock
);
1745 /* refill the reader queue and walk it again
1746 * keep both queues locked to avoid re-acquiring
1747 * the sk_receive_queue lock if fwd memory scheduling
1750 spin_lock(&sk_queue
->lock
);
1751 skb_queue_splice_tail_init(sk_queue
, queue
);
1753 skb
= __skb_try_recv_from_queue(sk
, queue
, flags
, off
,
1755 if (skb
&& !(flags
& MSG_PEEK
))
1756 udp_skb_dtor_locked(sk
, skb
);
1757 spin_unlock(&sk_queue
->lock
);
1758 spin_unlock_bh(&queue
->lock
);
1763 if (!sk_can_busy_loop(sk
))
1766 sk_busy_loop(sk
, flags
& MSG_DONTWAIT
);
1767 } while (!skb_queue_empty_lockless(sk_queue
));
1769 /* sk_queue is empty, reader_queue may contain peeked packets */
1771 !__skb_wait_for_more_packets(sk
, &sk
->sk_receive_queue
,
1773 (struct sk_buff
*)sk_queue
));
1778 EXPORT_SYMBOL(__skb_recv_udp
);
1781 * This should be easy, if there is something there we
1782 * return it, otherwise we block.
1785 int udp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int noblock
,
1786 int flags
, int *addr_len
)
1788 struct inet_sock
*inet
= inet_sk(sk
);
1789 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, msg
->msg_name
);
1790 struct sk_buff
*skb
;
1791 unsigned int ulen
, copied
;
1792 int off
, err
, peeking
= flags
& MSG_PEEK
;
1793 int is_udplite
= IS_UDPLITE(sk
);
1794 bool checksum_valid
= false;
1796 if (flags
& MSG_ERRQUEUE
)
1797 return ip_recv_error(sk
, msg
, len
, addr_len
);
1800 off
= sk_peek_offset(sk
, flags
);
1801 skb
= __skb_recv_udp(sk
, flags
, noblock
, &off
, &err
);
1805 ulen
= udp_skb_len(skb
);
1807 if (copied
> ulen
- off
)
1808 copied
= ulen
- off
;
1809 else if (copied
< ulen
)
1810 msg
->msg_flags
|= MSG_TRUNC
;
1813 * If checksum is needed at all, try to do it while copying the
1814 * data. If the data is truncated, or if we only want a partial
1815 * coverage checksum (UDP-Lite), do it before the copy.
1818 if (copied
< ulen
|| peeking
||
1819 (is_udplite
&& UDP_SKB_CB(skb
)->partial_cov
)) {
1820 checksum_valid
= udp_skb_csum_unnecessary(skb
) ||
1821 !__udp_lib_checksum_complete(skb
);
1822 if (!checksum_valid
)
1826 if (checksum_valid
|| udp_skb_csum_unnecessary(skb
)) {
1827 if (udp_skb_is_linear(skb
))
1828 err
= copy_linear_skb(skb
, copied
, off
, &msg
->msg_iter
);
1830 err
= skb_copy_datagram_msg(skb
, off
, msg
, copied
);
1832 err
= skb_copy_and_csum_datagram_msg(skb
, off
, msg
);
1838 if (unlikely(err
)) {
1840 atomic_inc(&sk
->sk_drops
);
1841 UDP_INC_STATS(sock_net(sk
),
1842 UDP_MIB_INERRORS
, is_udplite
);
1849 UDP_INC_STATS(sock_net(sk
),
1850 UDP_MIB_INDATAGRAMS
, is_udplite
);
1852 sock_recv_ts_and_drops(msg
, sk
, skb
);
1854 /* Copy the address. */
1856 sin
->sin_family
= AF_INET
;
1857 sin
->sin_port
= udp_hdr(skb
)->source
;
1858 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1859 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1860 *addr_len
= sizeof(*sin
);
1862 if (cgroup_bpf_enabled
)
1863 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk
,
1864 (struct sockaddr
*)sin
);
1867 if (udp_sk(sk
)->gro_enabled
)
1868 udp_cmsg_recv(msg
, sk
, skb
);
1870 if (inet
->cmsg_flags
)
1871 ip_cmsg_recv_offset(msg
, sk
, skb
, sizeof(struct udphdr
), off
);
1874 if (flags
& MSG_TRUNC
)
1877 skb_consume_udp(sk
, skb
, peeking
? -err
: err
);
1881 if (!__sk_queue_drop_skb(sk
, &udp_sk(sk
)->reader_queue
, skb
, flags
,
1882 udp_skb_destructor
)) {
1883 UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
1884 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1888 /* starting over for a new packet, but check if we need to yield */
1890 msg
->msg_flags
&= ~MSG_TRUNC
;
1894 int udp_pre_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
1896 /* This check is replicated from __ip4_datagram_connect() and
1897 * intended to prevent BPF program called below from accessing bytes
1898 * that are out of the bound specified by user in addr_len.
1900 if (addr_len
< sizeof(struct sockaddr_in
))
1903 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk
, uaddr
);
1905 EXPORT_SYMBOL(udp_pre_connect
);
1907 int __udp_disconnect(struct sock
*sk
, int flags
)
1909 struct inet_sock
*inet
= inet_sk(sk
);
1911 * 1003.1g - break association.
1914 sk
->sk_state
= TCP_CLOSE
;
1915 inet
->inet_daddr
= 0;
1916 inet
->inet_dport
= 0;
1917 sock_rps_reset_rxhash(sk
);
1918 sk
->sk_bound_dev_if
= 0;
1919 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
)) {
1920 inet_reset_saddr(sk
);
1921 if (sk
->sk_prot
->rehash
&&
1922 (sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
))
1923 sk
->sk_prot
->rehash(sk
);
1926 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1927 sk
->sk_prot
->unhash(sk
);
1928 inet
->inet_sport
= 0;
1933 EXPORT_SYMBOL(__udp_disconnect
);
1935 int udp_disconnect(struct sock
*sk
, int flags
)
1938 __udp_disconnect(sk
, flags
);
1942 EXPORT_SYMBOL(udp_disconnect
);
1944 void udp_lib_unhash(struct sock
*sk
)
1946 if (sk_hashed(sk
)) {
1947 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1948 struct udp_hslot
*hslot
, *hslot2
;
1950 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1951 udp_sk(sk
)->udp_port_hash
);
1952 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1954 spin_lock_bh(&hslot
->lock
);
1955 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1956 reuseport_detach_sock(sk
);
1957 if (sk_del_node_init_rcu(sk
)) {
1959 inet_sk(sk
)->inet_num
= 0;
1960 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1962 spin_lock(&hslot2
->lock
);
1963 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1965 spin_unlock(&hslot2
->lock
);
1967 spin_unlock_bh(&hslot
->lock
);
1970 EXPORT_SYMBOL(udp_lib_unhash
);
1973 * inet_rcv_saddr was changed, we must rehash secondary hash
1975 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1977 if (sk_hashed(sk
)) {
1978 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1979 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1981 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1982 nhslot2
= udp_hashslot2(udptable
, newhash
);
1983 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1985 if (hslot2
!= nhslot2
||
1986 rcu_access_pointer(sk
->sk_reuseport_cb
)) {
1987 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1988 udp_sk(sk
)->udp_port_hash
);
1989 /* we must lock primary chain too */
1990 spin_lock_bh(&hslot
->lock
);
1991 if (rcu_access_pointer(sk
->sk_reuseport_cb
))
1992 reuseport_detach_sock(sk
);
1994 if (hslot2
!= nhslot2
) {
1995 spin_lock(&hslot2
->lock
);
1996 hlist_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1998 spin_unlock(&hslot2
->lock
);
2000 spin_lock(&nhslot2
->lock
);
2001 hlist_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
2004 spin_unlock(&nhslot2
->lock
);
2007 spin_unlock_bh(&hslot
->lock
);
2011 EXPORT_SYMBOL(udp_lib_rehash
);
2013 void udp_v4_rehash(struct sock
*sk
)
2015 u16 new_hash
= ipv4_portaddr_hash(sock_net(sk
),
2016 inet_sk(sk
)->inet_rcv_saddr
,
2017 inet_sk(sk
)->inet_num
);
2018 udp_lib_rehash(sk
, new_hash
);
2021 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
2025 if (inet_sk(sk
)->inet_daddr
) {
2026 sock_rps_save_rxhash(sk
, skb
);
2027 sk_mark_napi_id(sk
, skb
);
2028 sk_incoming_cpu_update(sk
);
2030 sk_mark_napi_id_once(sk
, skb
);
2033 rc
= __udp_enqueue_schedule_skb(sk
, skb
);
2035 int is_udplite
= IS_UDPLITE(sk
);
2037 /* Note that an ENOMEM error is charged twice */
2039 UDP_INC_STATS(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
2041 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
2043 trace_udp_fail_queue_rcv_skb(rc
, sk
);
2053 * >0: "udp encap" protocol resubmission
2055 * Note that in the success and error cases, the skb is assumed to
2056 * have either been requeued or freed.
2058 static int udp_queue_rcv_one_skb(struct sock
*sk
, struct sk_buff
*skb
)
2060 struct udp_sock
*up
= udp_sk(sk
);
2061 int is_udplite
= IS_UDPLITE(sk
);
2064 * Charge it to the socket, dropping if the queue is full.
2066 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
2070 if (static_branch_unlikely(&udp_encap_needed_key
) && up
->encap_type
) {
2071 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
2074 * This is an encapsulation socket so pass the skb to
2075 * the socket's udp_encap_rcv() hook. Otherwise, just
2076 * fall through and pass this up the UDP socket.
2077 * up->encap_rcv() returns the following value:
2078 * =0 if skb was successfully passed to the encap
2079 * handler or was discarded by it.
2080 * >0 if skb should be passed on to UDP.
2081 * <0 if skb should be resubmitted as proto -N
2084 /* if we're overly short, let UDP handle it */
2085 encap_rcv
= READ_ONCE(up
->encap_rcv
);
2089 /* Verify checksum before giving to encap */
2090 if (udp_lib_checksum_complete(skb
))
2093 ret
= encap_rcv(sk
, skb
);
2095 __UDP_INC_STATS(sock_net(sk
),
2096 UDP_MIB_INDATAGRAMS
,
2102 /* FALLTHROUGH -- it's a UDP Packet */
2106 * UDP-Lite specific tests, ignored on UDP sockets
2108 if ((up
->pcflag
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
2111 * MIB statistics other than incrementing the error count are
2112 * disabled for the following two types of errors: these depend
2113 * on the application settings, not on the functioning of the
2114 * protocol stack as such.
2116 * RFC 3828 here recommends (sec 3.3): "There should also be a
2117 * way ... to ... at least let the receiving application block
2118 * delivery of packets with coverage values less than a value
2119 * provided by the application."
2121 if (up
->pcrlen
== 0) { /* full coverage was set */
2122 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2123 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
2126 /* The next case involves violating the min. coverage requested
2127 * by the receiver. This is subtle: if receiver wants x and x is
2128 * greater than the buffersize/MTU then receiver will complain
2129 * that it wants x while sender emits packets of smaller size y.
2130 * Therefore the above ...()->partial_cov statement is essential.
2132 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
2133 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2134 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
2139 prefetch(&sk
->sk_rmem_alloc
);
2140 if (rcu_access_pointer(sk
->sk_filter
) &&
2141 udp_lib_checksum_complete(skb
))
2144 if (sk_filter_trim_cap(sk
, skb
, sizeof(struct udphdr
)))
2147 udp_csum_pull_header(skb
);
2149 ipv4_pktinfo_prepare(sk
, skb
);
2150 return __udp_queue_rcv_skb(sk
, skb
);
2153 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
2155 __UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
2156 atomic_inc(&sk
->sk_drops
);
2161 static int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
2163 struct sk_buff
*next
, *segs
;
2166 if (likely(!udp_unexpected_gso(sk
, skb
)))
2167 return udp_queue_rcv_one_skb(sk
, skb
);
2169 BUILD_BUG_ON(sizeof(struct udp_skb_cb
) > SKB_GSO_CB_OFFSET
);
2170 __skb_push(skb
, -skb_mac_offset(skb
));
2171 segs
= udp_rcv_segment(sk
, skb
, true);
2172 skb_list_walk_safe(segs
, skb
, next
) {
2173 __skb_pull(skb
, skb_transport_offset(skb
));
2174 ret
= udp_queue_rcv_one_skb(sk
, skb
);
2176 ip_protocol_deliver_rcu(dev_net(skb
->dev
), skb
, -ret
);
2181 /* For TCP sockets, sk_rx_dst is protected by socket lock
2182 * For UDP, we use xchg() to guard against concurrent changes.
2184 bool udp_sk_rx_dst_set(struct sock
*sk
, struct dst_entry
*dst
)
2186 struct dst_entry
*old
;
2188 if (dst_hold_safe(dst
)) {
2189 old
= xchg(&sk
->sk_rx_dst
, dst
);
2195 EXPORT_SYMBOL(udp_sk_rx_dst_set
);
2198 * Multicasts and broadcasts go to each listener.
2200 * Note: called only from the BH handler context.
2202 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
2204 __be32 saddr
, __be32 daddr
,
2205 struct udp_table
*udptable
,
2208 struct sock
*sk
, *first
= NULL
;
2209 unsigned short hnum
= ntohs(uh
->dest
);
2210 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, hnum
);
2211 unsigned int hash2
= 0, hash2_any
= 0, use_hash2
= (hslot
->count
> 10);
2212 unsigned int offset
= offsetof(typeof(*sk
), sk_node
);
2213 int dif
= skb
->dev
->ifindex
;
2214 int sdif
= inet_sdif(skb
);
2215 struct hlist_node
*node
;
2216 struct sk_buff
*nskb
;
2219 hash2_any
= ipv4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
) &
2221 hash2
= ipv4_portaddr_hash(net
, daddr
, hnum
) & udptable
->mask
;
2223 hslot
= &udptable
->hash2
[hash2
];
2224 offset
= offsetof(typeof(*sk
), __sk_common
.skc_portaddr_node
);
2227 sk_for_each_entry_offset_rcu(sk
, node
, &hslot
->head
, offset
) {
2228 if (!__udp_is_mcast_sock(net
, sk
, uh
->dest
, daddr
,
2229 uh
->source
, saddr
, dif
, sdif
, hnum
))
2236 nskb
= skb_clone(skb
, GFP_ATOMIC
);
2238 if (unlikely(!nskb
)) {
2239 atomic_inc(&sk
->sk_drops
);
2240 __UDP_INC_STATS(net
, UDP_MIB_RCVBUFERRORS
,
2242 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
,
2246 if (udp_queue_rcv_skb(sk
, nskb
) > 0)
2250 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2251 if (use_hash2
&& hash2
!= hash2_any
) {
2257 if (udp_queue_rcv_skb(first
, skb
) > 0)
2261 __UDP_INC_STATS(net
, UDP_MIB_IGNOREDMULTI
,
2262 proto
== IPPROTO_UDPLITE
);
2267 /* Initialize UDP checksum. If exited with zero value (success),
2268 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2269 * Otherwise, csum completion requires checksumming packet body,
2270 * including udp header and folding it to skb->csum.
2272 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
2277 UDP_SKB_CB(skb
)->partial_cov
= 0;
2278 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
2280 if (proto
== IPPROTO_UDPLITE
) {
2281 err
= udplite_checksum_init(skb
, uh
);
2285 if (UDP_SKB_CB(skb
)->partial_cov
) {
2286 skb
->csum
= inet_compute_pseudo(skb
, proto
);
2291 /* Note, we are only interested in != 0 or == 0, thus the
2294 err
= (__force
int)skb_checksum_init_zero_check(skb
, proto
, uh
->check
,
2295 inet_compute_pseudo
);
2299 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&& !skb
->csum_valid
) {
2300 /* If SW calculated the value, we know it's bad */
2301 if (skb
->csum_complete_sw
)
2304 /* HW says the value is bad. Let's validate that.
2305 * skb->csum is no longer the full packet checksum,
2306 * so don't treat it as such.
2308 skb_checksum_complete_unset(skb
);
2314 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2315 * return code conversion for ip layer consumption
2317 static int udp_unicast_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
,
2322 if (inet_get_convert_csum(sk
) && uh
->check
&& !IS_UDPLITE(sk
))
2323 skb_checksum_try_convert(skb
, IPPROTO_UDP
, inet_compute_pseudo
);
2325 ret
= udp_queue_rcv_skb(sk
, skb
);
2327 /* a return value > 0 means to resubmit the input, but
2328 * it wants the return to be -protocol, or 0
2336 * All we need to do is get the socket, and then do a checksum.
2339 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
2344 unsigned short ulen
;
2345 struct rtable
*rt
= skb_rtable(skb
);
2346 __be32 saddr
, daddr
;
2347 struct net
*net
= dev_net(skb
->dev
);
2351 * Validate the packet.
2353 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
2354 goto drop
; /* No space for header. */
2357 ulen
= ntohs(uh
->len
);
2358 saddr
= ip_hdr(skb
)->saddr
;
2359 daddr
= ip_hdr(skb
)->daddr
;
2361 if (ulen
> skb
->len
)
2364 if (proto
== IPPROTO_UDP
) {
2365 /* UDP validates ulen. */
2366 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
2371 if (udp4_csum_init(skb
, uh
, proto
))
2374 sk
= skb_steal_sock(skb
, &refcounted
);
2376 struct dst_entry
*dst
= skb_dst(skb
);
2379 if (unlikely(sk
->sk_rx_dst
!= dst
))
2380 udp_sk_rx_dst_set(sk
, dst
);
2382 ret
= udp_unicast_rcv_skb(sk
, skb
, uh
);
2388 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
2389 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
2390 saddr
, daddr
, udptable
, proto
);
2392 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
2394 return udp_unicast_rcv_skb(sk
, skb
, uh
);
2396 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
2400 /* No socket. Drop packet silently, if checksum is wrong */
2401 if (udp_lib_checksum_complete(skb
))
2404 __UDP_INC_STATS(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
2405 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
2408 * Hmm. We got an UDP packet to a port to which we
2409 * don't wanna listen. Ignore it.
2415 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2416 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2417 &saddr
, ntohs(uh
->source
),
2419 &daddr
, ntohs(uh
->dest
));
2424 * RFC1122: OK. Discards the bad packet silently (as far as
2425 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2427 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2428 proto
== IPPROTO_UDPLITE
? "Lite" : "",
2429 &saddr
, ntohs(uh
->source
), &daddr
, ntohs(uh
->dest
),
2431 __UDP_INC_STATS(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
2433 __UDP_INC_STATS(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
2438 /* We can only early demux multicast if there is a single matching socket.
2439 * If more than one socket found returns NULL
2441 static struct sock
*__udp4_lib_mcast_demux_lookup(struct net
*net
,
2442 __be16 loc_port
, __be32 loc_addr
,
2443 __be16 rmt_port
, __be32 rmt_addr
,
2446 struct sock
*sk
, *result
;
2447 unsigned short hnum
= ntohs(loc_port
);
2448 unsigned int slot
= udp_hashfn(net
, hnum
, udp_table
.mask
);
2449 struct udp_hslot
*hslot
= &udp_table
.hash
[slot
];
2451 /* Do not bother scanning a too big list */
2452 if (hslot
->count
> 10)
2456 sk_for_each_rcu(sk
, &hslot
->head
) {
2457 if (__udp_is_mcast_sock(net
, sk
, loc_port
, loc_addr
,
2458 rmt_port
, rmt_addr
, dif
, sdif
, hnum
)) {
2468 /* For unicast we should only early demux connected sockets or we can
2469 * break forwarding setups. The chains here can be long so only check
2470 * if the first socket is an exact match and if not move on.
2472 static struct sock
*__udp4_lib_demux_lookup(struct net
*net
,
2473 __be16 loc_port
, __be32 loc_addr
,
2474 __be16 rmt_port
, __be32 rmt_addr
,
2477 unsigned short hnum
= ntohs(loc_port
);
2478 unsigned int hash2
= ipv4_portaddr_hash(net
, loc_addr
, hnum
);
2479 unsigned int slot2
= hash2
& udp_table
.mask
;
2480 struct udp_hslot
*hslot2
= &udp_table
.hash2
[slot2
];
2481 INET_ADDR_COOKIE(acookie
, rmt_addr
, loc_addr
);
2482 const __portpair ports
= INET_COMBINED_PORTS(rmt_port
, hnum
);
2485 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
2486 if (INET_MATCH(sk
, net
, acookie
, rmt_addr
,
2487 loc_addr
, ports
, dif
, sdif
))
2489 /* Only check first socket in chain */
2495 int udp_v4_early_demux(struct sk_buff
*skb
)
2497 struct net
*net
= dev_net(skb
->dev
);
2498 struct in_device
*in_dev
= NULL
;
2499 const struct iphdr
*iph
;
2500 const struct udphdr
*uh
;
2501 struct sock
*sk
= NULL
;
2502 struct dst_entry
*dst
;
2503 int dif
= skb
->dev
->ifindex
;
2504 int sdif
= inet_sdif(skb
);
2507 /* validate the packet */
2508 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct udphdr
)))
2514 if (skb
->pkt_type
== PACKET_MULTICAST
) {
2515 in_dev
= __in_dev_get_rcu(skb
->dev
);
2520 ours
= ip_check_mc_rcu(in_dev
, iph
->daddr
, iph
->saddr
,
2525 sk
= __udp4_lib_mcast_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2526 uh
->source
, iph
->saddr
,
2528 } else if (skb
->pkt_type
== PACKET_HOST
) {
2529 sk
= __udp4_lib_demux_lookup(net
, uh
->dest
, iph
->daddr
,
2530 uh
->source
, iph
->saddr
, dif
, sdif
);
2533 if (!sk
|| !refcount_inc_not_zero(&sk
->sk_refcnt
))
2537 skb
->destructor
= sock_efree
;
2538 dst
= READ_ONCE(sk
->sk_rx_dst
);
2541 dst
= dst_check(dst
, 0);
2545 /* set noref for now.
2546 * any place which wants to hold dst has to call
2549 skb_dst_set_noref(skb
, dst
);
2551 /* for unconnected multicast sockets we need to validate
2552 * the source on each packet
2554 if (!inet_sk(sk
)->inet_daddr
&& in_dev
)
2555 return ip_mc_validate_source(skb
, iph
->daddr
,
2556 iph
->saddr
, iph
->tos
,
2557 skb
->dev
, in_dev
, &itag
);
2562 int udp_rcv(struct sk_buff
*skb
)
2564 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
2567 void udp_destroy_sock(struct sock
*sk
)
2569 struct udp_sock
*up
= udp_sk(sk
);
2570 bool slow
= lock_sock_fast(sk
);
2571 udp_flush_pending_frames(sk
);
2572 unlock_sock_fast(sk
, slow
);
2573 if (static_branch_unlikely(&udp_encap_needed_key
)) {
2574 if (up
->encap_type
) {
2575 void (*encap_destroy
)(struct sock
*sk
);
2576 encap_destroy
= READ_ONCE(up
->encap_destroy
);
2580 if (up
->encap_enabled
)
2581 static_branch_dec(&udp_encap_needed_key
);
2586 * Socket option code for UDP
2588 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
2589 sockptr_t optval
, unsigned int optlen
,
2590 int (*push_pending_frames
)(struct sock
*))
2592 struct udp_sock
*up
= udp_sk(sk
);
2595 int is_udplite
= IS_UDPLITE(sk
);
2597 if (optlen
< sizeof(int))
2600 if (copy_from_sockptr(&val
, optval
, sizeof(val
)))
2603 valbool
= val
? 1 : 0;
2612 push_pending_frames(sk
);
2621 case UDP_ENCAP_ESPINUDP
:
2622 case UDP_ENCAP_ESPINUDP_NON_IKE
:
2623 #if IS_ENABLED(CONFIG_IPV6)
2624 if (sk
->sk_family
== AF_INET6
)
2625 up
->encap_rcv
= ipv6_stub
->xfrm6_udp_encap_rcv
;
2628 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
2631 case UDP_ENCAP_L2TPINUDP
:
2632 up
->encap_type
= val
;
2634 udp_tunnel_encap_enable(sk
->sk_socket
);
2643 case UDP_NO_CHECK6_TX
:
2644 up
->no_check6_tx
= valbool
;
2647 case UDP_NO_CHECK6_RX
:
2648 up
->no_check6_rx
= valbool
;
2652 if (val
< 0 || val
> USHRT_MAX
)
2660 udp_tunnel_encap_enable(sk
->sk_socket
);
2661 up
->gro_enabled
= valbool
;
2666 * UDP-Lite's partial checksum coverage (RFC 3828).
2668 /* The sender sets actual checksum coverage length via this option.
2669 * The case coverage > packet length is handled by send module. */
2670 case UDPLITE_SEND_CSCOV
:
2671 if (!is_udplite
) /* Disable the option on UDP sockets */
2672 return -ENOPROTOOPT
;
2673 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
2675 else if (val
> USHRT_MAX
)
2678 up
->pcflag
|= UDPLITE_SEND_CC
;
2681 /* The receiver specifies a minimum checksum coverage value. To make
2682 * sense, this should be set to at least 8 (as done below). If zero is
2683 * used, this again means full checksum coverage. */
2684 case UDPLITE_RECV_CSCOV
:
2685 if (!is_udplite
) /* Disable the option on UDP sockets */
2686 return -ENOPROTOOPT
;
2687 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
2689 else if (val
> USHRT_MAX
)
2692 up
->pcflag
|= UDPLITE_RECV_CC
;
2702 EXPORT_SYMBOL(udp_lib_setsockopt
);
2704 int udp_setsockopt(struct sock
*sk
, int level
, int optname
, sockptr_t optval
,
2705 unsigned int optlen
)
2707 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2708 return udp_lib_setsockopt(sk
, level
, optname
,
2710 udp_push_pending_frames
);
2711 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
2714 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
2715 char __user
*optval
, int __user
*optlen
)
2717 struct udp_sock
*up
= udp_sk(sk
);
2720 if (get_user(len
, optlen
))
2723 len
= min_t(unsigned int, len
, sizeof(int));
2734 val
= up
->encap_type
;
2737 case UDP_NO_CHECK6_TX
:
2738 val
= up
->no_check6_tx
;
2741 case UDP_NO_CHECK6_RX
:
2742 val
= up
->no_check6_rx
;
2749 /* The following two cannot be changed on UDP sockets, the return is
2750 * always 0 (which corresponds to the full checksum coverage of UDP). */
2751 case UDPLITE_SEND_CSCOV
:
2755 case UDPLITE_RECV_CSCOV
:
2760 return -ENOPROTOOPT
;
2763 if (put_user(len
, optlen
))
2765 if (copy_to_user(optval
, &val
, len
))
2769 EXPORT_SYMBOL(udp_lib_getsockopt
);
2771 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
2772 char __user
*optval
, int __user
*optlen
)
2774 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
2775 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
2776 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
2780 * udp_poll - wait for a UDP event.
2781 * @file: - file struct
2783 * @wait: - poll table
2785 * This is same as datagram poll, except for the special case of
2786 * blocking sockets. If application is using a blocking fd
2787 * and a packet with checksum error is in the queue;
2788 * then it could get return from select indicating data available
2789 * but then block when reading it. Add special case code
2790 * to work around these arguably broken applications.
2792 __poll_t
udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2794 __poll_t mask
= datagram_poll(file
, sock
, wait
);
2795 struct sock
*sk
= sock
->sk
;
2797 if (!skb_queue_empty_lockless(&udp_sk(sk
)->reader_queue
))
2798 mask
|= EPOLLIN
| EPOLLRDNORM
;
2800 /* Check for false positives due to checksum errors */
2801 if ((mask
& EPOLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
2802 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && first_packet_length(sk
) == -1)
2803 mask
&= ~(EPOLLIN
| EPOLLRDNORM
);
2808 EXPORT_SYMBOL(udp_poll
);
2810 int udp_abort(struct sock
*sk
, int err
)
2815 sk
->sk_error_report(sk
);
2816 __udp_disconnect(sk
, 0);
2822 EXPORT_SYMBOL_GPL(udp_abort
);
2824 struct proto udp_prot
= {
2826 .owner
= THIS_MODULE
,
2827 .close
= udp_lib_close
,
2828 .pre_connect
= udp_pre_connect
,
2829 .connect
= ip4_datagram_connect
,
2830 .disconnect
= udp_disconnect
,
2832 .init
= udp_init_sock
,
2833 .destroy
= udp_destroy_sock
,
2834 .setsockopt
= udp_setsockopt
,
2835 .getsockopt
= udp_getsockopt
,
2836 .sendmsg
= udp_sendmsg
,
2837 .recvmsg
= udp_recvmsg
,
2838 .sendpage
= udp_sendpage
,
2839 .release_cb
= ip4_datagram_release_cb
,
2840 .hash
= udp_lib_hash
,
2841 .unhash
= udp_lib_unhash
,
2842 .rehash
= udp_v4_rehash
,
2843 .get_port
= udp_v4_get_port
,
2844 .memory_allocated
= &udp_memory_allocated
,
2845 .sysctl_mem
= sysctl_udp_mem
,
2846 .sysctl_wmem_offset
= offsetof(struct net
, ipv4
.sysctl_udp_wmem_min
),
2847 .sysctl_rmem_offset
= offsetof(struct net
, ipv4
.sysctl_udp_rmem_min
),
2848 .obj_size
= sizeof(struct udp_sock
),
2849 .h
.udp_table
= &udp_table
,
2850 .diag_destroy
= udp_abort
,
2852 EXPORT_SYMBOL(udp_prot
);
2854 /* ------------------------------------------------------------------------ */
2855 #ifdef CONFIG_PROC_FS
2857 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
2860 struct udp_seq_afinfo
*afinfo
;
2861 struct udp_iter_state
*state
= seq
->private;
2862 struct net
*net
= seq_file_net(seq
);
2864 if (state
->bpf_seq_afinfo
)
2865 afinfo
= state
->bpf_seq_afinfo
;
2867 afinfo
= PDE_DATA(file_inode(seq
->file
));
2869 for (state
->bucket
= start
; state
->bucket
<= afinfo
->udp_table
->mask
;
2871 struct udp_hslot
*hslot
= &afinfo
->udp_table
->hash
[state
->bucket
];
2873 if (hlist_empty(&hslot
->head
))
2876 spin_lock_bh(&hslot
->lock
);
2877 sk_for_each(sk
, &hslot
->head
) {
2878 if (!net_eq(sock_net(sk
), net
))
2880 if (afinfo
->family
== AF_UNSPEC
||
2881 sk
->sk_family
== afinfo
->family
)
2884 spin_unlock_bh(&hslot
->lock
);
2891 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
2893 struct udp_seq_afinfo
*afinfo
;
2894 struct udp_iter_state
*state
= seq
->private;
2895 struct net
*net
= seq_file_net(seq
);
2897 if (state
->bpf_seq_afinfo
)
2898 afinfo
= state
->bpf_seq_afinfo
;
2900 afinfo
= PDE_DATA(file_inode(seq
->file
));
2904 } while (sk
&& (!net_eq(sock_net(sk
), net
) ||
2905 (afinfo
->family
!= AF_UNSPEC
&&
2906 sk
->sk_family
!= afinfo
->family
)));
2909 if (state
->bucket
<= afinfo
->udp_table
->mask
)
2910 spin_unlock_bh(&afinfo
->udp_table
->hash
[state
->bucket
].lock
);
2911 return udp_get_first(seq
, state
->bucket
+ 1);
2916 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
2918 struct sock
*sk
= udp_get_first(seq
, 0);
2921 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2923 return pos
? NULL
: sk
;
2926 void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2928 struct udp_iter_state
*state
= seq
->private;
2929 state
->bucket
= MAX_UDP_PORTS
;
2931 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2933 EXPORT_SYMBOL(udp_seq_start
);
2935 void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2939 if (v
== SEQ_START_TOKEN
)
2940 sk
= udp_get_idx(seq
, 0);
2942 sk
= udp_get_next(seq
, v
);
2947 EXPORT_SYMBOL(udp_seq_next
);
2949 void udp_seq_stop(struct seq_file
*seq
, void *v
)
2951 struct udp_seq_afinfo
*afinfo
;
2952 struct udp_iter_state
*state
= seq
->private;
2954 if (state
->bpf_seq_afinfo
)
2955 afinfo
= state
->bpf_seq_afinfo
;
2957 afinfo
= PDE_DATA(file_inode(seq
->file
));
2959 if (state
->bucket
<= afinfo
->udp_table
->mask
)
2960 spin_unlock_bh(&afinfo
->udp_table
->hash
[state
->bucket
].lock
);
2962 EXPORT_SYMBOL(udp_seq_stop
);
2964 /* ------------------------------------------------------------------------ */
2965 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2968 struct inet_sock
*inet
= inet_sk(sp
);
2969 __be32 dest
= inet
->inet_daddr
;
2970 __be32 src
= inet
->inet_rcv_saddr
;
2971 __u16 destp
= ntohs(inet
->inet_dport
);
2972 __u16 srcp
= ntohs(inet
->inet_sport
);
2974 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2975 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2976 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2977 sk_wmem_alloc_get(sp
),
2980 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sp
)),
2982 refcount_read(&sp
->sk_refcnt
), sp
,
2983 atomic_read(&sp
->sk_drops
));
2986 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2988 seq_setwidth(seq
, 127);
2989 if (v
== SEQ_START_TOKEN
)
2990 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2991 "rx_queue tr tm->when retrnsmt uid timeout "
2992 "inode ref pointer drops");
2994 struct udp_iter_state
*state
= seq
->private;
2996 udp4_format_sock(v
, seq
, state
->bucket
);
3002 #ifdef CONFIG_BPF_SYSCALL
3003 struct bpf_iter__udp
{
3004 __bpf_md_ptr(struct bpf_iter_meta
*, meta
);
3005 __bpf_md_ptr(struct udp_sock
*, udp_sk
);
3006 uid_t uid
__aligned(8);
3007 int bucket
__aligned(8);
3010 static int udp_prog_seq_show(struct bpf_prog
*prog
, struct bpf_iter_meta
*meta
,
3011 struct udp_sock
*udp_sk
, uid_t uid
, int bucket
)
3013 struct bpf_iter__udp ctx
;
3015 meta
->seq_num
--; /* skip SEQ_START_TOKEN */
3017 ctx
.udp_sk
= udp_sk
;
3019 ctx
.bucket
= bucket
;
3020 return bpf_iter_run_prog(prog
, &ctx
);
3023 static int bpf_iter_udp_seq_show(struct seq_file
*seq
, void *v
)
3025 struct udp_iter_state
*state
= seq
->private;
3026 struct bpf_iter_meta meta
;
3027 struct bpf_prog
*prog
;
3028 struct sock
*sk
= v
;
3031 if (v
== SEQ_START_TOKEN
)
3034 uid
= from_kuid_munged(seq_user_ns(seq
), sock_i_uid(sk
));
3036 prog
= bpf_iter_get_info(&meta
, false);
3037 return udp_prog_seq_show(prog
, &meta
, v
, uid
, state
->bucket
);
3040 static void bpf_iter_udp_seq_stop(struct seq_file
*seq
, void *v
)
3042 struct bpf_iter_meta meta
;
3043 struct bpf_prog
*prog
;
3047 prog
= bpf_iter_get_info(&meta
, true);
3049 (void)udp_prog_seq_show(prog
, &meta
, v
, 0, 0);
3052 udp_seq_stop(seq
, v
);
3055 static const struct seq_operations bpf_iter_udp_seq_ops
= {
3056 .start
= udp_seq_start
,
3057 .next
= udp_seq_next
,
3058 .stop
= bpf_iter_udp_seq_stop
,
3059 .show
= bpf_iter_udp_seq_show
,
3063 const struct seq_operations udp_seq_ops
= {
3064 .start
= udp_seq_start
,
3065 .next
= udp_seq_next
,
3066 .stop
= udp_seq_stop
,
3067 .show
= udp4_seq_show
,
3069 EXPORT_SYMBOL(udp_seq_ops
);
3071 static struct udp_seq_afinfo udp4_seq_afinfo
= {
3073 .udp_table
= &udp_table
,
3076 static int __net_init
udp4_proc_init_net(struct net
*net
)
3078 if (!proc_create_net_data("udp", 0444, net
->proc_net
, &udp_seq_ops
,
3079 sizeof(struct udp_iter_state
), &udp4_seq_afinfo
))
3084 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
3086 remove_proc_entry("udp", net
->proc_net
);
3089 static struct pernet_operations udp4_net_ops
= {
3090 .init
= udp4_proc_init_net
,
3091 .exit
= udp4_proc_exit_net
,
3094 int __init
udp4_proc_init(void)
3096 return register_pernet_subsys(&udp4_net_ops
);
3099 void udp4_proc_exit(void)
3101 unregister_pernet_subsys(&udp4_net_ops
);
3103 #endif /* CONFIG_PROC_FS */
3105 static __initdata
unsigned long uhash_entries
;
3106 static int __init
set_uhash_entries(char *str
)
3113 ret
= kstrtoul(str
, 0, &uhash_entries
);
3117 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
3118 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
3121 __setup("uhash_entries=", set_uhash_entries
);
3123 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
3127 table
->hash
= alloc_large_system_hash(name
,
3128 2 * sizeof(struct udp_hslot
),
3130 21, /* one slot per 2 MB */
3134 UDP_HTABLE_SIZE_MIN
,
3137 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
3138 for (i
= 0; i
<= table
->mask
; i
++) {
3139 INIT_HLIST_HEAD(&table
->hash
[i
].head
);
3140 table
->hash
[i
].count
= 0;
3141 spin_lock_init(&table
->hash
[i
].lock
);
3143 for (i
= 0; i
<= table
->mask
; i
++) {
3144 INIT_HLIST_HEAD(&table
->hash2
[i
].head
);
3145 table
->hash2
[i
].count
= 0;
3146 spin_lock_init(&table
->hash2
[i
].lock
);
3150 u32
udp_flow_hashrnd(void)
3152 static u32 hashrnd __read_mostly
;
3154 net_get_random_once(&hashrnd
, sizeof(hashrnd
));
3158 EXPORT_SYMBOL(udp_flow_hashrnd
);
3160 static void __udp_sysctl_init(struct net
*net
)
3162 net
->ipv4
.sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
3163 net
->ipv4
.sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
3165 #ifdef CONFIG_NET_L3_MASTER_DEV
3166 net
->ipv4
.sysctl_udp_l3mdev_accept
= 0;
3170 static int __net_init
udp_sysctl_init(struct net
*net
)
3172 __udp_sysctl_init(net
);
3176 static struct pernet_operations __net_initdata udp_sysctl_ops
= {
3177 .init
= udp_sysctl_init
,
3180 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3181 DEFINE_BPF_ITER_FUNC(udp
, struct bpf_iter_meta
*meta
,
3182 struct udp_sock
*udp_sk
, uid_t uid
, int bucket
)
3184 static int bpf_iter_init_udp(void *priv_data
, struct bpf_iter_aux_info
*aux
)
3186 struct udp_iter_state
*st
= priv_data
;
3187 struct udp_seq_afinfo
*afinfo
;
3190 afinfo
= kmalloc(sizeof(*afinfo
), GFP_USER
| __GFP_NOWARN
);
3194 afinfo
->family
= AF_UNSPEC
;
3195 afinfo
->udp_table
= &udp_table
;
3196 st
->bpf_seq_afinfo
= afinfo
;
3197 ret
= bpf_iter_init_seq_net(priv_data
, aux
);
3203 static void bpf_iter_fini_udp(void *priv_data
)
3205 struct udp_iter_state
*st
= priv_data
;
3207 kfree(st
->bpf_seq_afinfo
);
3208 bpf_iter_fini_seq_net(priv_data
);
3211 static const struct bpf_iter_seq_info udp_seq_info
= {
3212 .seq_ops
= &bpf_iter_udp_seq_ops
,
3213 .init_seq_private
= bpf_iter_init_udp
,
3214 .fini_seq_private
= bpf_iter_fini_udp
,
3215 .seq_priv_size
= sizeof(struct udp_iter_state
),
3218 static struct bpf_iter_reg udp_reg_info
= {
3220 .ctx_arg_info_size
= 1,
3222 { offsetof(struct bpf_iter__udp
, udp_sk
),
3223 PTR_TO_BTF_ID_OR_NULL
},
3225 .seq_info
= &udp_seq_info
,
3228 static void __init
bpf_iter_register(void)
3230 udp_reg_info
.ctx_arg_info
[0].btf_id
= btf_sock_ids
[BTF_SOCK_TYPE_UDP
];
3231 if (bpf_iter_reg_target(&udp_reg_info
))
3232 pr_warn("Warning: could not register bpf iterator udp\n");
3236 void __init
udp_init(void)
3238 unsigned long limit
;
3241 udp_table_init(&udp_table
, "UDP");
3242 limit
= nr_free_buffer_pages() / 8;
3243 limit
= max(limit
, 128UL);
3244 sysctl_udp_mem
[0] = limit
/ 4 * 3;
3245 sysctl_udp_mem
[1] = limit
;
3246 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
3248 __udp_sysctl_init(&init_net
);
3250 /* 16 spinlocks per cpu */
3251 udp_busylocks_log
= ilog2(nr_cpu_ids
) + 4;
3252 udp_busylocks
= kmalloc(sizeof(spinlock_t
) << udp_busylocks_log
,
3255 panic("UDP: failed to alloc udp_busylocks\n");
3256 for (i
= 0; i
< (1U << udp_busylocks_log
); i
++)
3257 spin_lock_init(udp_busylocks
+ i
);
3259 if (register_pernet_subsys(&udp_sysctl_ops
))
3260 panic("UDP: failed to init sysctl parameters.\n");
3262 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3263 bpf_iter_register();