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 * Definitions for the UDP module.
9 * Version: @(#)udp.h 1.0.2 05/07/93
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
15 * Alan Cox : Turned on udp checksums. I don't want to
16 * chase 'memory corruption' bugs that aren't!
21 #include <linux/list.h>
22 #include <linux/bug.h>
23 #include <net/inet_sock.h>
27 #include <linux/ipv6.h>
28 #include <linux/seq_file.h>
29 #include <linux/poll.h>
30 #include <linux/indirect_call_wrapper.h>
33 * struct udp_skb_cb - UDP(-Lite) private variables
35 * @header: private variables used by IPv4/IPv6
36 * @cscov: checksum coverage length (UDP-Lite only)
37 * @partial_cov: if set indicates partial csum coverage
41 struct inet_skb_parm h4;
42 #if IS_ENABLED(CONFIG_IPV6)
43 struct inet6_skb_parm h6;
49 #define UDP_SKB_CB(__skb) ((struct udp_skb_cb *)((__skb)->cb))
52 * struct udp_hslot - UDP hash slot
54 * @head: head of list of sockets
55 * @count: number of sockets in 'head' list
56 * @lock: spinlock protecting changes to head/count
59 struct hlist_head head;
62 } __attribute__((aligned(2 * sizeof(long))));
65 * struct udp_table - UDP table
67 * @hash: hash table, sockets are hashed on (local port)
68 * @hash2: hash table, sockets are hashed on (local port, local address)
69 * @mask: number of slots in hash tables, minus 1
70 * @log: log2(number of slots in hash table)
73 struct udp_hslot *hash;
74 struct udp_hslot *hash2;
78 extern struct udp_table udp_table;
79 void udp_table_init(struct udp_table *, const char *);
80 static inline struct udp_hslot *udp_hashslot(struct udp_table *table,
81 struct net *net, unsigned int num)
83 return &table->hash[udp_hashfn(net, num, table->mask)];
86 * For secondary hash, net_hash_mix() is performed before calling
87 * udp_hashslot2(), this explains difference with udp_hashslot()
89 static inline struct udp_hslot *udp_hashslot2(struct udp_table *table,
92 return &table->hash2[hash & table->mask];
95 extern struct proto udp_prot;
97 extern atomic_long_t udp_memory_allocated;
98 DECLARE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
100 /* sysctl variables for udp */
101 extern long sysctl_udp_mem[3];
102 extern int sysctl_udp_rmem_min;
103 extern int sysctl_udp_wmem_min;
108 * Generic checksumming routines for UDP(-Lite) v4 and v6
110 static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb)
112 return (UDP_SKB_CB(skb)->cscov == skb->len ?
113 __skb_checksum_complete(skb) :
114 __skb_checksum_complete_head(skb, UDP_SKB_CB(skb)->cscov));
117 static inline int udp_lib_checksum_complete(struct sk_buff *skb)
119 return !skb_csum_unnecessary(skb) &&
120 __udp_lib_checksum_complete(skb);
124 * udp_csum_outgoing - compute UDPv4/v6 checksum over fragments
125 * @sk: socket we are writing to
126 * @skb: sk_buff containing the filled-in UDP header
127 * (checksum field must be zeroed out)
129 static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb)
131 __wsum csum = csum_partial(skb_transport_header(skb),
132 sizeof(struct udphdr), 0);
133 skb_queue_walk(&sk->sk_write_queue, skb) {
134 csum = csum_add(csum, skb->csum);
139 static inline __wsum udp_csum(struct sk_buff *skb)
141 __wsum csum = csum_partial(skb_transport_header(skb),
142 sizeof(struct udphdr), skb->csum);
144 for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) {
145 csum = csum_add(csum, skb->csum);
150 static inline __sum16 udp_v4_check(int len, __be32 saddr,
151 __be32 daddr, __wsum base)
153 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base);
156 void udp_set_csum(bool nocheck, struct sk_buff *skb,
157 __be32 saddr, __be32 daddr, int len);
159 static inline void udp_csum_pull_header(struct sk_buff *skb)
161 if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE)
162 skb->csum = csum_partial(skb->data, sizeof(struct udphdr),
164 skb_pull_rcsum(skb, sizeof(struct udphdr));
165 UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr);
168 typedef struct sock *(*udp_lookup_t)(const struct sk_buff *skb, __be16 sport,
171 void udp_v6_early_demux(struct sk_buff *skb);
172 INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
174 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
175 netdev_features_t features, bool is_ipv6);
177 static inline void udp_lib_init_sock(struct sock *sk)
179 struct udp_sock *up = udp_sk(sk);
181 skb_queue_head_init(&up->reader_queue);
182 up->forward_threshold = sk->sk_rcvbuf >> 2;
183 set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
186 /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
187 static inline int udp_lib_hash(struct sock *sk)
193 void udp_lib_unhash(struct sock *sk);
194 void udp_lib_rehash(struct sock *sk, u16 new_hash);
196 static inline void udp_lib_close(struct sock *sk, long timeout)
198 sk_common_release(sk);
201 int udp_lib_get_port(struct sock *sk, unsigned short snum,
202 unsigned int hash2_nulladdr);
204 u32 udp_flow_hashrnd(void);
206 static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
207 int min, int max, bool use_eth)
212 /* Use default range */
213 inet_get_local_port_range(net, &min, &max);
216 hash = skb_get_hash(skb);
217 if (unlikely(!hash)) {
219 /* Can't find a normal hash, caller has indicated an
220 * Ethernet packet so use that to compute a hash.
222 hash = jhash(skb->data, 2 * ETH_ALEN,
223 (__force u32) skb->protocol);
225 /* Can't derive any sort of hash for the packet, set
226 * to some consistent random value.
228 hash = udp_flow_hashrnd();
232 /* Since this is being sent on the wire obfuscate hash a bit
233 * to minimize possbility that any useful information to an
234 * attacker is leaked. Only upper 16 bits are relevant in the
235 * computation for 16 bit port value.
239 return htons((((u64) hash * (max - min)) >> 32) + min);
242 static inline int udp_rqueue_get(struct sock *sk)
244 return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
247 static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if,
250 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
251 return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_udp_l3mdev_accept),
252 bound_dev_if, dif, sdif);
254 return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
259 void udp_destruct_common(struct sock *sk);
260 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
261 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
262 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
263 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, int *off,
265 static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
270 return __skb_recv_udp(sk, flags, &off, err);
273 int udp_v4_early_demux(struct sk_buff *skb);
274 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
275 int udp_get_port(struct sock *sk, unsigned short snum,
276 int (*saddr_cmp)(const struct sock *,
277 const struct sock *));
278 int udp_err(struct sk_buff *, u32);
279 int udp_abort(struct sock *sk, int err);
280 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
281 int udp_push_pending_frames(struct sock *sk);
282 void udp_flush_pending_frames(struct sock *sk);
283 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
284 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
285 int udp_rcv(struct sk_buff *skb);
286 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
287 int udp_init_sock(struct sock *sk);
288 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
289 int __udp_disconnect(struct sock *sk, int flags);
290 int udp_disconnect(struct sock *sk, int flags);
291 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
292 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
293 netdev_features_t features,
295 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
296 char __user *optval, int __user *optlen);
297 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
298 sockptr_t optval, unsigned int optlen,
299 int (*push_pending_frames)(struct sock *));
300 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
301 __be32 daddr, __be16 dport, int dif);
302 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
303 __be32 daddr, __be16 dport, int dif, int sdif,
304 struct udp_table *tbl, struct sk_buff *skb);
305 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
306 __be16 sport, __be16 dport);
307 struct sock *udp6_lib_lookup(struct net *net,
308 const struct in6_addr *saddr, __be16 sport,
309 const struct in6_addr *daddr, __be16 dport,
311 struct sock *__udp6_lib_lookup(struct net *net,
312 const struct in6_addr *saddr, __be16 sport,
313 const struct in6_addr *daddr, __be16 dport,
314 int dif, int sdif, struct udp_table *tbl,
315 struct sk_buff *skb);
316 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
317 __be16 sport, __be16 dport);
318 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
320 /* UDP uses skb->dev_scratch to cache as much information as possible and avoid
321 * possibly multiple cache miss on dequeue()
323 struct udp_dev_scratch {
324 /* skb->truesize and the stateless bit are embedded in a single field;
325 * do not use a bitfield since the compiler emits better/smaller code
330 #if BITS_PER_LONG == 64
331 /* len and the bit needed to compute skb_csum_unnecessary
332 * will be on cold cache lines at recvmsg time.
333 * skb->len can be stored on 16 bits since the udp header has been
334 * already validated and pulled.
338 bool csum_unnecessary;
342 static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
344 return (struct udp_dev_scratch *)&skb->dev_scratch;
347 #if BITS_PER_LONG == 64
348 static inline unsigned int udp_skb_len(struct sk_buff *skb)
350 return udp_skb_scratch(skb)->len;
353 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
355 return udp_skb_scratch(skb)->csum_unnecessary;
358 static inline bool udp_skb_is_linear(struct sk_buff *skb)
360 return udp_skb_scratch(skb)->is_linear;
364 static inline unsigned int udp_skb_len(struct sk_buff *skb)
369 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
371 return skb_csum_unnecessary(skb);
374 static inline bool udp_skb_is_linear(struct sk_buff *skb)
376 return !skb_is_nonlinear(skb);
380 static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
385 n = copy_to_iter(skb->data + off, len, to);
389 iov_iter_revert(to, n);
394 * SNMP statistics for UDP and UDP-Lite
396 #define UDP_INC_STATS(net, field, is_udplite) do { \
397 if (is_udplite) SNMP_INC_STATS((net)->mib.udplite_statistics, field); \
398 else SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0)
399 #define __UDP_INC_STATS(net, field, is_udplite) do { \
400 if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field); \
401 else __SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0)
403 #define __UDP6_INC_STATS(net, field, is_udplite) do { \
404 if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
405 else __SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \
407 #define UDP6_INC_STATS(net, field, __lite) do { \
408 if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field); \
409 else SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \
412 #if IS_ENABLED(CONFIG_IPV6)
413 #define __UDPX_MIB(sk, ipv4) \
415 ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
416 sock_net(sk)->mib.udp_statistics) : \
417 (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 : \
418 sock_net(sk)->mib.udp_stats_in6); \
421 #define __UDPX_MIB(sk, ipv4) \
423 IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
424 sock_net(sk)->mib.udp_statistics; \
428 #define __UDPX_INC_STATS(sk, field) \
429 __SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
431 #ifdef CONFIG_PROC_FS
432 struct udp_seq_afinfo {
434 struct udp_table *udp_table;
437 struct udp_iter_state {
438 struct seq_net_private p;
440 struct udp_seq_afinfo *bpf_seq_afinfo;
443 void *udp_seq_start(struct seq_file *seq, loff_t *pos);
444 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
445 void udp_seq_stop(struct seq_file *seq, void *v);
447 extern const struct seq_operations udp_seq_ops;
448 extern const struct seq_operations udp6_seq_ops;
450 int udp4_proc_init(void);
451 void udp4_proc_exit(void);
452 #endif /* CONFIG_PROC_FS */
454 int udpv4_offload_init(void);
458 DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
459 void udp_encap_enable(void);
460 void udp_encap_disable(void);
461 #if IS_ENABLED(CONFIG_IPV6)
462 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
463 void udpv6_encap_enable(void);
466 static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
467 struct sk_buff *skb, bool ipv4)
469 netdev_features_t features = NETIF_F_SG;
470 struct sk_buff *segs;
472 /* Avoid csum recalculation by skb_segment unless userspace explicitly
473 * asks for the final checksum values
475 if (!inet_get_convert_csum(sk))
476 features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
478 /* UDP segmentation expects packets of type CHECKSUM_PARTIAL or
479 * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial
480 * packets in udp_gro_complete_segment. As does UDP GSO, verified by
481 * udp_send_skb. But when those packets are looped in dev_loopback_xmit
482 * their ip_summed CHECKSUM_NONE is changed to CHECKSUM_UNNECESSARY.
483 * Reset in this specific case, where PARTIAL is both correct and
486 if (skb->pkt_type == PACKET_LOOPBACK)
487 skb->ip_summed = CHECKSUM_PARTIAL;
489 /* the GSO CB lays after the UDP one, no need to save and restore any
492 segs = __skb_gso_segment(skb, features, false);
493 if (IS_ERR_OR_NULL(segs)) {
494 int segs_nr = skb_shinfo(skb)->gso_segs;
496 atomic_add(segs_nr, &sk->sk_drops);
497 SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
506 static inline void udp_post_segment_fix_csum(struct sk_buff *skb)
508 /* UDP-lite can't land here - no GRO */
509 WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov);
511 /* UDP packets generated with UDP_SEGMENT and traversing:
513 * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx)
515 * can reach an UDP socket with CHECKSUM_NONE, because
516 * __iptunnel_pull_header() converts CHECKSUM_PARTIAL into NONE.
517 * SKB_GSO_UDP_L4 or SKB_GSO_FRAGLIST packets with no UDP tunnel will
518 * have a valid checksum, as the GRO engine validates the UDP csum
519 * before the aggregation and nobody strips such info in between.
520 * Instead of adding another check in the tunnel fastpath, we can force
521 * a valid csum after the segmentation.
522 * Additionally fixup the UDP CB.
524 UDP_SKB_CB(skb)->cscov = skb->len;
525 if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid)
529 #ifdef CONFIG_BPF_SYSCALL
531 struct proto *udp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
532 int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);