Loading...
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
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.
6 *
7 * Definitions for the UDP module.
8 *
9 * Version: @(#)udp.h 1.0.2 05/07/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 *
14 * Fixes:
15 * Alan Cox : Turned on udp checksums. I don't want to
16 * chase 'memory corruption' bugs that aren't!
17 */
18#ifndef _UDP_H
19#define _UDP_H
20
21#include <linux/list.h>
22#include <linux/bug.h>
23#include <net/inet_sock.h>
24#include <net/gso.h>
25#include <net/sock.h>
26#include <net/snmp.h>
27#include <net/ip.h>
28#include <linux/ipv6.h>
29#include <linux/seq_file.h>
30#include <linux/poll.h>
31#include <linux/indirect_call_wrapper.h>
32
33/**
34 * struct udp_skb_cb - UDP(-Lite) private variables
35 *
36 * @header: private variables used by IPv4/IPv6
37 * @cscov: checksum coverage length (UDP-Lite only)
38 * @partial_cov: if set indicates partial csum coverage
39 */
40struct udp_skb_cb {
41 union {
42 struct inet_skb_parm h4;
43#if IS_ENABLED(CONFIG_IPV6)
44 struct inet6_skb_parm h6;
45#endif
46 } header;
47 __u16 cscov;
48 __u8 partial_cov;
49};
50#define UDP_SKB_CB(__skb) ((struct udp_skb_cb *)((__skb)->cb))
51
52/**
53 * struct udp_hslot - UDP hash slot used by udp_table.hash/hash4
54 *
55 * @head: head of list of sockets
56 * @nulls_head: head of list of sockets, only used by hash4
57 * @count: number of sockets in 'head' list
58 * @lock: spinlock protecting changes to head/count
59 */
60struct udp_hslot {
61 union {
62 struct hlist_head head;
63 /* hash4 uses hlist_nulls to avoid moving wrongly onto another
64 * hlist, because rehash() can happen with lookup().
65 */
66 struct hlist_nulls_head nulls_head;
67 };
68 int count;
69 spinlock_t lock;
70} __aligned(2 * sizeof(long));
71
72/**
73 * struct udp_hslot_main - UDP hash slot used by udp_table.hash2
74 *
75 * @hslot: basic hash slot
76 * @hash4_cnt: number of sockets in hslot4 of the same
77 * (local port, local address)
78 */
79struct udp_hslot_main {
80 struct udp_hslot hslot; /* must be the first member */
81#if !IS_ENABLED(CONFIG_BASE_SMALL)
82 u32 hash4_cnt;
83#endif
84} __aligned(2 * sizeof(long));
85#define UDP_HSLOT_MAIN(__hslot) ((struct udp_hslot_main *)(__hslot))
86
87/**
88 * struct udp_table - UDP table
89 *
90 * @hash: hash table, sockets are hashed on (local port)
91 * @hash2: hash table, sockets are hashed on (local port, local address)
92 * @hash4: hash table, connected sockets are hashed on
93 * (local port, local address, remote port, remote address)
94 * @mask: number of slots in hash tables, minus 1
95 * @log: log2(number of slots in hash table)
96 */
97struct udp_table {
98 struct udp_hslot *hash;
99 struct udp_hslot_main *hash2;
100#if !IS_ENABLED(CONFIG_BASE_SMALL)
101 struct udp_hslot *hash4;
102#endif
103 unsigned int mask;
104 unsigned int log;
105};
106extern struct udp_table udp_table;
107void udp_table_init(struct udp_table *, const char *);
108static inline struct udp_hslot *udp_hashslot(struct udp_table *table,
109 const struct net *net,
110 unsigned int num)
111{
112 return &table->hash[udp_hashfn(net, num, table->mask)];
113}
114
115/*
116 * For secondary hash, net_hash_mix() is performed before calling
117 * udp_hashslot2(), this explains difference with udp_hashslot()
118 */
119static inline struct udp_hslot *udp_hashslot2(struct udp_table *table,
120 unsigned int hash)
121{
122 return &table->hash2[hash & table->mask].hslot;
123}
124
125#if IS_ENABLED(CONFIG_BASE_SMALL)
126static inline void udp_table_hash4_init(struct udp_table *table)
127{
128}
129
130static inline struct udp_hslot *udp_hashslot4(struct udp_table *table,
131 unsigned int hash)
132{
133 BUILD_BUG();
134 return NULL;
135}
136
137static inline bool udp_hashed4(const struct sock *sk)
138{
139 return false;
140}
141
142static inline unsigned int udp_hash4_slot_size(void)
143{
144 return 0;
145}
146
147static inline bool udp_has_hash4(const struct udp_hslot *hslot2)
148{
149 return false;
150}
151
152static inline void udp_hash4_inc(struct udp_hslot *hslot2)
153{
154}
155
156static inline void udp_hash4_dec(struct udp_hslot *hslot2)
157{
158}
159#else /* !CONFIG_BASE_SMALL */
160
161/* Must be called with table->hash2 initialized */
162static inline void udp_table_hash4_init(struct udp_table *table)
163{
164 table->hash4 = (void *)(table->hash2 + (table->mask + 1));
165 for (int i = 0; i <= table->mask; i++) {
166 table->hash2[i].hash4_cnt = 0;
167
168 INIT_HLIST_NULLS_HEAD(&table->hash4[i].nulls_head, i);
169 table->hash4[i].count = 0;
170 spin_lock_init(&table->hash4[i].lock);
171 }
172}
173
174static inline struct udp_hslot *udp_hashslot4(struct udp_table *table,
175 unsigned int hash)
176{
177 return &table->hash4[hash & table->mask];
178}
179
180static inline bool udp_hashed4(const struct sock *sk)
181{
182 return !hlist_nulls_unhashed(&udp_sk(sk)->udp_lrpa_node);
183}
184
185static inline unsigned int udp_hash4_slot_size(void)
186{
187 return sizeof(struct udp_hslot);
188}
189
190static inline bool udp_has_hash4(const struct udp_hslot *hslot2)
191{
192 return UDP_HSLOT_MAIN(hslot2)->hash4_cnt;
193}
194
195static inline void udp_hash4_inc(struct udp_hslot *hslot2)
196{
197 UDP_HSLOT_MAIN(hslot2)->hash4_cnt++;
198}
199
200static inline void udp_hash4_dec(struct udp_hslot *hslot2)
201{
202 UDP_HSLOT_MAIN(hslot2)->hash4_cnt--;
203}
204#endif /* CONFIG_BASE_SMALL */
205
206extern struct proto udp_prot;
207
208extern atomic_long_t udp_memory_allocated;
209DECLARE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
210
211/* sysctl variables for udp */
212extern long sysctl_udp_mem[3];
213extern int sysctl_udp_rmem_min;
214extern int sysctl_udp_wmem_min;
215
216struct sk_buff;
217
218/*
219 * Generic checksumming routines for UDP(-Lite) v4 and v6
220 */
221static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb)
222{
223 return (UDP_SKB_CB(skb)->cscov == skb->len ?
224 __skb_checksum_complete(skb) :
225 __skb_checksum_complete_head(skb, UDP_SKB_CB(skb)->cscov));
226}
227
228static inline int udp_lib_checksum_complete(struct sk_buff *skb)
229{
230 return !skb_csum_unnecessary(skb) &&
231 __udp_lib_checksum_complete(skb);
232}
233
234/**
235 * udp_csum_outgoing - compute UDPv4/v6 checksum over fragments
236 * @sk: socket we are writing to
237 * @skb: sk_buff containing the filled-in UDP header
238 * (checksum field must be zeroed out)
239 */
240static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb)
241{
242 __wsum csum = csum_partial(skb_transport_header(skb),
243 sizeof(struct udphdr), 0);
244 skb_queue_walk(&sk->sk_write_queue, skb) {
245 csum = csum_add(csum, skb->csum);
246 }
247 return csum;
248}
249
250static inline __wsum udp_csum(struct sk_buff *skb)
251{
252 __wsum csum = csum_partial(skb_transport_header(skb),
253 sizeof(struct udphdr), skb->csum);
254
255 for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) {
256 csum = csum_add(csum, skb->csum);
257 }
258 return csum;
259}
260
261static inline __sum16 udp_v4_check(int len, __be32 saddr,
262 __be32 daddr, __wsum base)
263{
264 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base);
265}
266
267void udp_set_csum(bool nocheck, struct sk_buff *skb,
268 __be32 saddr, __be32 daddr, int len);
269
270static inline void udp_csum_pull_header(struct sk_buff *skb)
271{
272 if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE)
273 skb->csum = csum_partial(skb->data, sizeof(struct udphdr),
274 skb->csum);
275 skb_pull_rcsum(skb, sizeof(struct udphdr));
276 UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr);
277}
278
279typedef struct sock *(*udp_lookup_t)(const struct sk_buff *skb, __be16 sport,
280 __be16 dport);
281
282void udp_v6_early_demux(struct sk_buff *skb);
283INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
284
285struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
286 netdev_features_t features, bool is_ipv6);
287
288static inline void udp_lib_init_sock(struct sock *sk)
289{
290 struct udp_sock *up = udp_sk(sk);
291
292 skb_queue_head_init(&up->reader_queue);
293 up->forward_threshold = sk->sk_rcvbuf >> 2;
294 set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
295}
296
297/* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
298static inline int udp_lib_hash(struct sock *sk)
299{
300 BUG();
301 return 0;
302}
303
304void udp_lib_unhash(struct sock *sk);
305void udp_lib_rehash(struct sock *sk, u16 new_hash, u16 new_hash4);
306u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
307 const __be32 faddr, const __be16 fport);
308
309static inline void udp_lib_close(struct sock *sk, long timeout)
310{
311 sk_common_release(sk);
312}
313
314/* hash4 routines shared between UDPv4/6 */
315#if IS_ENABLED(CONFIG_BASE_SMALL)
316static inline void udp_lib_hash4(struct sock *sk, u16 hash)
317{
318}
319
320static inline void udp4_hash4(struct sock *sk)
321{
322}
323#else /* !CONFIG_BASE_SMALL */
324void udp_lib_hash4(struct sock *sk, u16 hash);
325void udp4_hash4(struct sock *sk);
326#endif /* CONFIG_BASE_SMALL */
327
328int udp_lib_get_port(struct sock *sk, unsigned short snum,
329 unsigned int hash2_nulladdr);
330
331u32 udp_flow_hashrnd(void);
332
333static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
334 int min, int max, bool use_eth)
335{
336 u32 hash;
337
338 if (min >= max) {
339 /* Use default range */
340 inet_get_local_port_range(net, &min, &max);
341 }
342
343 hash = skb_get_hash(skb);
344 if (unlikely(!hash)) {
345 if (use_eth) {
346 /* Can't find a normal hash, caller has indicated an
347 * Ethernet packet so use that to compute a hash.
348 */
349 hash = jhash(skb->data, 2 * ETH_ALEN,
350 (__force u32) skb->protocol);
351 } else {
352 /* Can't derive any sort of hash for the packet, set
353 * to some consistent random value.
354 */
355 hash = udp_flow_hashrnd();
356 }
357 }
358
359 /* Since this is being sent on the wire obfuscate hash a bit
360 * to minimize possibility that any useful information to an
361 * attacker is leaked. Only upper 16 bits are relevant in the
362 * computation for 16 bit port value.
363 */
364 hash ^= hash << 16;
365
366 return htons((((u64) hash * (max - min)) >> 32) + min);
367}
368
369static inline int udp_rqueue_get(struct sock *sk)
370{
371 return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
372}
373
374static inline bool udp_sk_bound_dev_eq(const struct net *net, int bound_dev_if,
375 int dif, int sdif)
376{
377#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
378 return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_udp_l3mdev_accept),
379 bound_dev_if, dif, sdif);
380#else
381 return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
382#endif
383}
384
385/* net/ipv4/udp.c */
386void udp_destruct_common(struct sock *sk);
387void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
388int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
389void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
390struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, int *off,
391 int *err);
392static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
393 int *err)
394{
395 int off = 0;
396
397 return __skb_recv_udp(sk, flags, &off, err);
398}
399
400int udp_v4_early_demux(struct sk_buff *skb);
401bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
402int udp_err(struct sk_buff *, u32);
403int udp_abort(struct sock *sk, int err);
404int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
405void udp_splice_eof(struct socket *sock);
406int udp_push_pending_frames(struct sock *sk);
407void udp_flush_pending_frames(struct sock *sk);
408int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
409void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
410int udp_rcv(struct sk_buff *skb);
411int udp_ioctl(struct sock *sk, int cmd, int *karg);
412int udp_init_sock(struct sock *sk);
413int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
414int __udp_disconnect(struct sock *sk, int flags);
415int udp_disconnect(struct sock *sk, int flags);
416__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
417struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
418 netdev_features_t features,
419 bool is_ipv6);
420int udp_lib_getsockopt(struct sock *sk, int level, int optname,
421 char __user *optval, int __user *optlen);
422int udp_lib_setsockopt(struct sock *sk, int level, int optname,
423 sockptr_t optval, unsigned int optlen,
424 int (*push_pending_frames)(struct sock *));
425struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport,
426 __be32 daddr, __be16 dport, int dif);
427struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr,
428 __be16 sport,
429 __be32 daddr, __be16 dport, int dif, int sdif,
430 struct udp_table *tbl, struct sk_buff *skb);
431struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
432 __be16 sport, __be16 dport);
433struct sock *udp6_lib_lookup(const struct net *net,
434 const struct in6_addr *saddr, __be16 sport,
435 const struct in6_addr *daddr, __be16 dport,
436 int dif);
437struct sock *__udp6_lib_lookup(const struct net *net,
438 const struct in6_addr *saddr, __be16 sport,
439 const struct in6_addr *daddr, __be16 dport,
440 int dif, int sdif, struct udp_table *tbl,
441 struct sk_buff *skb);
442struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
443 __be16 sport, __be16 dport);
444int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
445
446/* UDP uses skb->dev_scratch to cache as much information as possible and avoid
447 * possibly multiple cache miss on dequeue()
448 */
449struct udp_dev_scratch {
450 /* skb->truesize and the stateless bit are embedded in a single field;
451 * do not use a bitfield since the compiler emits better/smaller code
452 * this way
453 */
454 u32 _tsize_state;
455
456#if BITS_PER_LONG == 64
457 /* len and the bit needed to compute skb_csum_unnecessary
458 * will be on cold cache lines at recvmsg time.
459 * skb->len can be stored on 16 bits since the udp header has been
460 * already validated and pulled.
461 */
462 u16 len;
463 bool is_linear;
464 bool csum_unnecessary;
465#endif
466};
467
468static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
469{
470 return (struct udp_dev_scratch *)&skb->dev_scratch;
471}
472
473#if BITS_PER_LONG == 64
474static inline unsigned int udp_skb_len(struct sk_buff *skb)
475{
476 return udp_skb_scratch(skb)->len;
477}
478
479static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
480{
481 return udp_skb_scratch(skb)->csum_unnecessary;
482}
483
484static inline bool udp_skb_is_linear(struct sk_buff *skb)
485{
486 return udp_skb_scratch(skb)->is_linear;
487}
488
489#else
490static inline unsigned int udp_skb_len(struct sk_buff *skb)
491{
492 return skb->len;
493}
494
495static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
496{
497 return skb_csum_unnecessary(skb);
498}
499
500static inline bool udp_skb_is_linear(struct sk_buff *skb)
501{
502 return !skb_is_nonlinear(skb);
503}
504#endif
505
506static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
507 struct iov_iter *to)
508{
509 return copy_to_iter_full(skb->data + off, len, to) ? 0 : -EFAULT;
510}
511
512/*
513 * SNMP statistics for UDP and UDP-Lite
514 */
515#define UDP_INC_STATS(net, field, is_udplite) do { \
516 if (is_udplite) SNMP_INC_STATS((net)->mib.udplite_statistics, field); \
517 else SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0)
518#define __UDP_INC_STATS(net, field, is_udplite) do { \
519 if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field); \
520 else __SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0)
521
522#define __UDP6_INC_STATS(net, field, is_udplite) do { \
523 if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
524 else __SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \
525} while(0)
526#define UDP6_INC_STATS(net, field, __lite) do { \
527 if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field); \
528 else SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \
529} while(0)
530
531#if IS_ENABLED(CONFIG_IPV6)
532#define __UDPX_MIB(sk, ipv4) \
533({ \
534 ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
535 sock_net(sk)->mib.udp_statistics) : \
536 (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 : \
537 sock_net(sk)->mib.udp_stats_in6); \
538})
539#else
540#define __UDPX_MIB(sk, ipv4) \
541({ \
542 IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
543 sock_net(sk)->mib.udp_statistics; \
544})
545#endif
546
547#define __UDPX_INC_STATS(sk, field) \
548 __SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
549
550#ifdef CONFIG_PROC_FS
551struct udp_seq_afinfo {
552 sa_family_t family;
553 struct udp_table *udp_table;
554};
555
556struct udp_iter_state {
557 struct seq_net_private p;
558 int bucket;
559};
560
561void *udp_seq_start(struct seq_file *seq, loff_t *pos);
562void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
563void udp_seq_stop(struct seq_file *seq, void *v);
564
565extern const struct seq_operations udp_seq_ops;
566extern const struct seq_operations udp6_seq_ops;
567
568int udp4_proc_init(void);
569void udp4_proc_exit(void);
570#endif /* CONFIG_PROC_FS */
571
572int udpv4_offload_init(void);
573
574void udp_init(void);
575
576DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
577void udp_encap_enable(void);
578void udp_encap_disable(void);
579#if IS_ENABLED(CONFIG_IPV6)
580DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
581void udpv6_encap_enable(void);
582#endif
583
584static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
585 struct sk_buff *skb, bool ipv4)
586{
587 netdev_features_t features = NETIF_F_SG;
588 struct sk_buff *segs;
589
590 /* Avoid csum recalculation by skb_segment unless userspace explicitly
591 * asks for the final checksum values
592 */
593 if (!inet_get_convert_csum(sk))
594 features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
595
596 /* UDP segmentation expects packets of type CHECKSUM_PARTIAL or
597 * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial
598 * packets in udp_gro_complete_segment. As does UDP GSO, verified by
599 * udp_send_skb. But when those packets are looped in dev_loopback_xmit
600 * their ip_summed CHECKSUM_NONE is changed to CHECKSUM_UNNECESSARY.
601 * Reset in this specific case, where PARTIAL is both correct and
602 * required.
603 */
604 if (skb->pkt_type == PACKET_LOOPBACK)
605 skb->ip_summed = CHECKSUM_PARTIAL;
606
607 /* the GSO CB lays after the UDP one, no need to save and restore any
608 * CB fragment
609 */
610 segs = __skb_gso_segment(skb, features, false);
611 if (IS_ERR_OR_NULL(segs)) {
612 int segs_nr = skb_shinfo(skb)->gso_segs;
613
614 atomic_add(segs_nr, &sk->sk_drops);
615 SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
616 kfree_skb(skb);
617 return NULL;
618 }
619
620 consume_skb(skb);
621 return segs;
622}
623
624static inline void udp_post_segment_fix_csum(struct sk_buff *skb)
625{
626 /* UDP-lite can't land here - no GRO */
627 WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov);
628
629 /* UDP packets generated with UDP_SEGMENT and traversing:
630 *
631 * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx)
632 *
633 * can reach an UDP socket with CHECKSUM_NONE, because
634 * __iptunnel_pull_header() converts CHECKSUM_PARTIAL into NONE.
635 * SKB_GSO_UDP_L4 or SKB_GSO_FRAGLIST packets with no UDP tunnel will
636 * have a valid checksum, as the GRO engine validates the UDP csum
637 * before the aggregation and nobody strips such info in between.
638 * Instead of adding another check in the tunnel fastpath, we can force
639 * a valid csum after the segmentation.
640 * Additionally fixup the UDP CB.
641 */
642 UDP_SKB_CB(skb)->cscov = skb->len;
643 if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid)
644 skb->csum_valid = 1;
645}
646
647#ifdef CONFIG_BPF_SYSCALL
648struct sk_psock;
649int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
650#endif
651
652#endif /* _UDP_H */
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
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.
6 *
7 * Definitions for the UDP module.
8 *
9 * Version: @(#)udp.h 1.0.2 05/07/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 *
14 * Fixes:
15 * Alan Cox : Turned on udp checksums. I don't want to
16 * chase 'memory corruption' bugs that aren't!
17 */
18#ifndef _UDP_H
19#define _UDP_H
20
21#include <linux/list.h>
22#include <linux/bug.h>
23#include <net/inet_sock.h>
24#include <net/sock.h>
25#include <net/snmp.h>
26#include <net/ip.h>
27#include <linux/ipv6.h>
28#include <linux/seq_file.h>
29#include <linux/poll.h>
30#include <linux/indirect_call_wrapper.h>
31
32/**
33 * struct udp_skb_cb - UDP(-Lite) private variables
34 *
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
38 */
39struct udp_skb_cb {
40 union {
41 struct inet_skb_parm h4;
42#if IS_ENABLED(CONFIG_IPV6)
43 struct inet6_skb_parm h6;
44#endif
45 } header;
46 __u16 cscov;
47 __u8 partial_cov;
48};
49#define UDP_SKB_CB(__skb) ((struct udp_skb_cb *)((__skb)->cb))
50
51/**
52 * struct udp_hslot - UDP hash slot
53 *
54 * @head: head of list of sockets
55 * @count: number of sockets in 'head' list
56 * @lock: spinlock protecting changes to head/count
57 */
58struct udp_hslot {
59 struct hlist_head head;
60 int count;
61 spinlock_t lock;
62} __attribute__((aligned(2 * sizeof(long))));
63
64/**
65 * struct udp_table - UDP table
66 *
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)
71 */
72struct udp_table {
73 struct udp_hslot *hash;
74 struct udp_hslot *hash2;
75 unsigned int mask;
76 unsigned int log;
77};
78extern struct udp_table udp_table;
79void udp_table_init(struct udp_table *, const char *);
80static inline struct udp_hslot *udp_hashslot(struct udp_table *table,
81 struct net *net, unsigned int num)
82{
83 return &table->hash[udp_hashfn(net, num, table->mask)];
84}
85/*
86 * For secondary hash, net_hash_mix() is performed before calling
87 * udp_hashslot2(), this explains difference with udp_hashslot()
88 */
89static inline struct udp_hslot *udp_hashslot2(struct udp_table *table,
90 unsigned int hash)
91{
92 return &table->hash2[hash & table->mask];
93}
94
95extern struct proto udp_prot;
96
97extern atomic_long_t udp_memory_allocated;
98
99/* sysctl variables for udp */
100extern long sysctl_udp_mem[3];
101extern int sysctl_udp_rmem_min;
102extern int sysctl_udp_wmem_min;
103
104struct sk_buff;
105
106/*
107 * Generic checksumming routines for UDP(-Lite) v4 and v6
108 */
109static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb)
110{
111 return (UDP_SKB_CB(skb)->cscov == skb->len ?
112 __skb_checksum_complete(skb) :
113 __skb_checksum_complete_head(skb, UDP_SKB_CB(skb)->cscov));
114}
115
116static inline int udp_lib_checksum_complete(struct sk_buff *skb)
117{
118 return !skb_csum_unnecessary(skb) &&
119 __udp_lib_checksum_complete(skb);
120}
121
122/**
123 * udp_csum_outgoing - compute UDPv4/v6 checksum over fragments
124 * @sk: socket we are writing to
125 * @skb: sk_buff containing the filled-in UDP header
126 * (checksum field must be zeroed out)
127 */
128static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb)
129{
130 __wsum csum = csum_partial(skb_transport_header(skb),
131 sizeof(struct udphdr), 0);
132 skb_queue_walk(&sk->sk_write_queue, skb) {
133 csum = csum_add(csum, skb->csum);
134 }
135 return csum;
136}
137
138static inline __wsum udp_csum(struct sk_buff *skb)
139{
140 __wsum csum = csum_partial(skb_transport_header(skb),
141 sizeof(struct udphdr), skb->csum);
142
143 for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) {
144 csum = csum_add(csum, skb->csum);
145 }
146 return csum;
147}
148
149static inline __sum16 udp_v4_check(int len, __be32 saddr,
150 __be32 daddr, __wsum base)
151{
152 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base);
153}
154
155void udp_set_csum(bool nocheck, struct sk_buff *skb,
156 __be32 saddr, __be32 daddr, int len);
157
158static inline void udp_csum_pull_header(struct sk_buff *skb)
159{
160 if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE)
161 skb->csum = csum_partial(skb->data, sizeof(struct udphdr),
162 skb->csum);
163 skb_pull_rcsum(skb, sizeof(struct udphdr));
164 UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr);
165}
166
167typedef struct sock *(*udp_lookup_t)(const struct sk_buff *skb, __be16 sport,
168 __be16 dport);
169
170INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *,
171 struct sk_buff *));
172INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int));
173INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp6_gro_receive(struct list_head *,
174 struct sk_buff *));
175INDIRECT_CALLABLE_DECLARE(int udp6_gro_complete(struct sk_buff *, int));
176INDIRECT_CALLABLE_DECLARE(void udp_v6_early_demux(struct sk_buff *));
177INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
178
179struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
180 struct udphdr *uh, struct sock *sk);
181int udp_gro_complete(struct sk_buff *skb, int nhoff, udp_lookup_t lookup);
182
183struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
184 netdev_features_t features, bool is_ipv6);
185
186static inline struct udphdr *udp_gro_udphdr(struct sk_buff *skb)
187{
188 struct udphdr *uh;
189 unsigned int hlen, off;
190
191 off = skb_gro_offset(skb);
192 hlen = off + sizeof(*uh);
193 uh = skb_gro_header_fast(skb, off);
194 if (skb_gro_header_hard(skb, hlen))
195 uh = skb_gro_header_slow(skb, hlen, off);
196
197 return uh;
198}
199
200/* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
201static inline int udp_lib_hash(struct sock *sk)
202{
203 BUG();
204 return 0;
205}
206
207void udp_lib_unhash(struct sock *sk);
208void udp_lib_rehash(struct sock *sk, u16 new_hash);
209
210static inline void udp_lib_close(struct sock *sk, long timeout)
211{
212 sk_common_release(sk);
213}
214
215int udp_lib_get_port(struct sock *sk, unsigned short snum,
216 unsigned int hash2_nulladdr);
217
218u32 udp_flow_hashrnd(void);
219
220static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
221 int min, int max, bool use_eth)
222{
223 u32 hash;
224
225 if (min >= max) {
226 /* Use default range */
227 inet_get_local_port_range(net, &min, &max);
228 }
229
230 hash = skb_get_hash(skb);
231 if (unlikely(!hash)) {
232 if (use_eth) {
233 /* Can't find a normal hash, caller has indicated an
234 * Ethernet packet so use that to compute a hash.
235 */
236 hash = jhash(skb->data, 2 * ETH_ALEN,
237 (__force u32) skb->protocol);
238 } else {
239 /* Can't derive any sort of hash for the packet, set
240 * to some consistent random value.
241 */
242 hash = udp_flow_hashrnd();
243 }
244 }
245
246 /* Since this is being sent on the wire obfuscate hash a bit
247 * to minimize possbility that any useful information to an
248 * attacker is leaked. Only upper 16 bits are relevant in the
249 * computation for 16 bit port value.
250 */
251 hash ^= hash << 16;
252
253 return htons((((u64) hash * (max - min)) >> 32) + min);
254}
255
256static inline int udp_rqueue_get(struct sock *sk)
257{
258 return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
259}
260
261static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if,
262 int dif, int sdif)
263{
264#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
265 return inet_bound_dev_eq(!!net->ipv4.sysctl_udp_l3mdev_accept,
266 bound_dev_if, dif, sdif);
267#else
268 return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
269#endif
270}
271
272/* net/ipv4/udp.c */
273void udp_destruct_sock(struct sock *sk);
274void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
275int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
276void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
277struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
278 int noblock, int *off, int *err);
279static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
280 int noblock, int *err)
281{
282 int off = 0;
283
284 return __skb_recv_udp(sk, flags, noblock, &off, err);
285}
286
287int udp_v4_early_demux(struct sk_buff *skb);
288bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
289int udp_get_port(struct sock *sk, unsigned short snum,
290 int (*saddr_cmp)(const struct sock *,
291 const struct sock *));
292int udp_err(struct sk_buff *, u32);
293int udp_abort(struct sock *sk, int err);
294int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
295int udp_push_pending_frames(struct sock *sk);
296void udp_flush_pending_frames(struct sock *sk);
297int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
298void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
299int udp_rcv(struct sk_buff *skb);
300int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
301int udp_init_sock(struct sock *sk);
302int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
303int __udp_disconnect(struct sock *sk, int flags);
304int udp_disconnect(struct sock *sk, int flags);
305__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
306struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
307 netdev_features_t features,
308 bool is_ipv6);
309int udp_lib_getsockopt(struct sock *sk, int level, int optname,
310 char __user *optval, int __user *optlen);
311int udp_lib_setsockopt(struct sock *sk, int level, int optname,
312 sockptr_t optval, unsigned int optlen,
313 int (*push_pending_frames)(struct sock *));
314struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
315 __be32 daddr, __be16 dport, int dif);
316struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
317 __be32 daddr, __be16 dport, int dif, int sdif,
318 struct udp_table *tbl, struct sk_buff *skb);
319struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
320 __be16 sport, __be16 dport);
321struct sock *udp6_lib_lookup(struct net *net,
322 const struct in6_addr *saddr, __be16 sport,
323 const struct in6_addr *daddr, __be16 dport,
324 int dif);
325struct sock *__udp6_lib_lookup(struct net *net,
326 const struct in6_addr *saddr, __be16 sport,
327 const struct in6_addr *daddr, __be16 dport,
328 int dif, int sdif, struct udp_table *tbl,
329 struct sk_buff *skb);
330struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
331 __be16 sport, __be16 dport);
332int udp_read_sock(struct sock *sk, read_descriptor_t *desc,
333 sk_read_actor_t recv_actor);
334
335/* UDP uses skb->dev_scratch to cache as much information as possible and avoid
336 * possibly multiple cache miss on dequeue()
337 */
338struct udp_dev_scratch {
339 /* skb->truesize and the stateless bit are embedded in a single field;
340 * do not use a bitfield since the compiler emits better/smaller code
341 * this way
342 */
343 u32 _tsize_state;
344
345#if BITS_PER_LONG == 64
346 /* len and the bit needed to compute skb_csum_unnecessary
347 * will be on cold cache lines at recvmsg time.
348 * skb->len can be stored on 16 bits since the udp header has been
349 * already validated and pulled.
350 */
351 u16 len;
352 bool is_linear;
353 bool csum_unnecessary;
354#endif
355};
356
357static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
358{
359 return (struct udp_dev_scratch *)&skb->dev_scratch;
360}
361
362#if BITS_PER_LONG == 64
363static inline unsigned int udp_skb_len(struct sk_buff *skb)
364{
365 return udp_skb_scratch(skb)->len;
366}
367
368static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
369{
370 return udp_skb_scratch(skb)->csum_unnecessary;
371}
372
373static inline bool udp_skb_is_linear(struct sk_buff *skb)
374{
375 return udp_skb_scratch(skb)->is_linear;
376}
377
378#else
379static inline unsigned int udp_skb_len(struct sk_buff *skb)
380{
381 return skb->len;
382}
383
384static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
385{
386 return skb_csum_unnecessary(skb);
387}
388
389static inline bool udp_skb_is_linear(struct sk_buff *skb)
390{
391 return !skb_is_nonlinear(skb);
392}
393#endif
394
395static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
396 struct iov_iter *to)
397{
398 int n;
399
400 n = copy_to_iter(skb->data + off, len, to);
401 if (n == len)
402 return 0;
403
404 iov_iter_revert(to, n);
405 return -EFAULT;
406}
407
408/*
409 * SNMP statistics for UDP and UDP-Lite
410 */
411#define UDP_INC_STATS(net, field, is_udplite) do { \
412 if (is_udplite) SNMP_INC_STATS((net)->mib.udplite_statistics, field); \
413 else SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0)
414#define __UDP_INC_STATS(net, field, is_udplite) do { \
415 if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field); \
416 else __SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0)
417
418#define __UDP6_INC_STATS(net, field, is_udplite) do { \
419 if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
420 else __SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \
421} while(0)
422#define UDP6_INC_STATS(net, field, __lite) do { \
423 if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field); \
424 else SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \
425} while(0)
426
427#if IS_ENABLED(CONFIG_IPV6)
428#define __UDPX_MIB(sk, ipv4) \
429({ \
430 ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
431 sock_net(sk)->mib.udp_statistics) : \
432 (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 : \
433 sock_net(sk)->mib.udp_stats_in6); \
434})
435#else
436#define __UDPX_MIB(sk, ipv4) \
437({ \
438 IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
439 sock_net(sk)->mib.udp_statistics; \
440})
441#endif
442
443#define __UDPX_INC_STATS(sk, field) \
444 __SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
445
446#ifdef CONFIG_PROC_FS
447struct udp_seq_afinfo {
448 sa_family_t family;
449 struct udp_table *udp_table;
450};
451
452struct udp_iter_state {
453 struct seq_net_private p;
454 int bucket;
455 struct udp_seq_afinfo *bpf_seq_afinfo;
456};
457
458void *udp_seq_start(struct seq_file *seq, loff_t *pos);
459void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
460void udp_seq_stop(struct seq_file *seq, void *v);
461
462extern const struct seq_operations udp_seq_ops;
463extern const struct seq_operations udp6_seq_ops;
464
465int udp4_proc_init(void);
466void udp4_proc_exit(void);
467#endif /* CONFIG_PROC_FS */
468
469int udpv4_offload_init(void);
470
471void udp_init(void);
472
473DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
474void udp_encap_enable(void);
475void udp_encap_disable(void);
476#if IS_ENABLED(CONFIG_IPV6)
477DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
478void udpv6_encap_enable(void);
479#endif
480
481static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
482 struct sk_buff *skb, bool ipv4)
483{
484 netdev_features_t features = NETIF_F_SG;
485 struct sk_buff *segs;
486
487 /* Avoid csum recalculation by skb_segment unless userspace explicitly
488 * asks for the final checksum values
489 */
490 if (!inet_get_convert_csum(sk))
491 features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
492
493 /* UDP segmentation expects packets of type CHECKSUM_PARTIAL or
494 * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial
495 * packets in udp_gro_complete_segment. As does UDP GSO, verified by
496 * udp_send_skb. But when those packets are looped in dev_loopback_xmit
497 * their ip_summed is set to CHECKSUM_UNNECESSARY. Reset in this
498 * specific case, where PARTIAL is both correct and required.
499 */
500 if (skb->pkt_type == PACKET_LOOPBACK)
501 skb->ip_summed = CHECKSUM_PARTIAL;
502
503 /* the GSO CB lays after the UDP one, no need to save and restore any
504 * CB fragment
505 */
506 segs = __skb_gso_segment(skb, features, false);
507 if (IS_ERR_OR_NULL(segs)) {
508 int segs_nr = skb_shinfo(skb)->gso_segs;
509
510 atomic_add(segs_nr, &sk->sk_drops);
511 SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
512 kfree_skb(skb);
513 return NULL;
514 }
515
516 consume_skb(skb);
517 return segs;
518}
519
520static inline void udp_post_segment_fix_csum(struct sk_buff *skb)
521{
522 /* UDP-lite can't land here - no GRO */
523 WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov);
524
525 /* UDP packets generated with UDP_SEGMENT and traversing:
526 *
527 * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx)
528 *
529 * can reach an UDP socket with CHECKSUM_NONE, because
530 * __iptunnel_pull_header() converts CHECKSUM_PARTIAL into NONE.
531 * SKB_GSO_UDP_L4 or SKB_GSO_FRAGLIST packets with no UDP tunnel will
532 * have a valid checksum, as the GRO engine validates the UDP csum
533 * before the aggregation and nobody strips such info in between.
534 * Instead of adding another check in the tunnel fastpath, we can force
535 * a valid csum after the segmentation.
536 * Additionally fixup the UDP CB.
537 */
538 UDP_SKB_CB(skb)->cscov = skb->len;
539 if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid)
540 skb->csum_valid = 1;
541}
542
543#ifdef CONFIG_BPF_SYSCALL
544struct sk_psock;
545struct proto *udp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
546int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
547#endif
548
549#endif /* _UDP_H */