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 | */ |
40 | struct 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 | } ; |
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 |
54 | * |
55 | * @head: head of list of sockets |
56 | * @count: number of sockets in 'head' list |
57 | * @lock: spinlock protecting changes to head/count |
58 | */ |
59 | struct udp_hslot { |
60 | struct hlist_head head; |
61 | int count; |
62 | spinlock_t lock; |
63 | } __attribute__((aligned(2 * sizeof(long)))); |
64 | |
65 | /** |
66 | * struct udp_table - UDP table |
67 | * |
68 | * @hash: hash table, sockets are hashed on (local port) |
69 | * @hash2: hash table, sockets are hashed on (local port, local address) |
70 | * @mask: number of slots in hash tables, minus 1 |
71 | * @log: log2(number of slots in hash table) |
72 | */ |
73 | struct udp_table { |
74 | struct udp_hslot *hash; |
75 | struct udp_hslot *hash2; |
76 | unsigned int mask; |
77 | unsigned int log; |
78 | }; |
79 | extern struct udp_table udp_table; |
80 | void udp_table_init(struct udp_table *, const char *); |
81 | static inline struct udp_hslot *udp_hashslot(struct udp_table *table, |
82 | struct net *net, unsigned int num) |
83 | { |
84 | return &table->hash[udp_hashfn(net, num, mask: table->mask)]; |
85 | } |
86 | /* |
87 | * For secondary hash, net_hash_mix() is performed before calling |
88 | * udp_hashslot2(), this explains difference with udp_hashslot() |
89 | */ |
90 | static inline struct udp_hslot *udp_hashslot2(struct udp_table *table, |
91 | unsigned int hash) |
92 | { |
93 | return &table->hash2[hash & table->mask]; |
94 | } |
95 | |
96 | extern struct proto udp_prot; |
97 | |
98 | extern atomic_long_t udp_memory_allocated; |
99 | DECLARE_PER_CPU(int, udp_memory_per_cpu_fw_alloc); |
100 | |
101 | /* sysctl variables for udp */ |
102 | extern long sysctl_udp_mem[3]; |
103 | extern int sysctl_udp_rmem_min; |
104 | extern int sysctl_udp_wmem_min; |
105 | |
106 | struct sk_buff; |
107 | |
108 | /* |
109 | * Generic checksumming routines for UDP(-Lite) v4 and v6 |
110 | */ |
111 | static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb) |
112 | { |
113 | return (UDP_SKB_CB(skb)->cscov == skb->len ? |
114 | __skb_checksum_complete(skb) : |
115 | __skb_checksum_complete_head(skb, UDP_SKB_CB(skb)->cscov)); |
116 | } |
117 | |
118 | static inline int udp_lib_checksum_complete(struct sk_buff *skb) |
119 | { |
120 | return !skb_csum_unnecessary(skb) && |
121 | __udp_lib_checksum_complete(skb); |
122 | } |
123 | |
124 | /** |
125 | * udp_csum_outgoing - compute UDPv4/v6 checksum over fragments |
126 | * @sk: socket we are writing to |
127 | * @skb: sk_buff containing the filled-in UDP header |
128 | * (checksum field must be zeroed out) |
129 | */ |
130 | static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb) |
131 | { |
132 | __wsum csum = csum_partial(buff: skb_transport_header(skb), |
133 | len: sizeof(struct udphdr), sum: 0); |
134 | skb_queue_walk(&sk->sk_write_queue, skb) { |
135 | csum = csum_add(csum, addend: skb->csum); |
136 | } |
137 | return csum; |
138 | } |
139 | |
140 | static inline __wsum udp_csum(struct sk_buff *skb) |
141 | { |
142 | __wsum csum = csum_partial(buff: skb_transport_header(skb), |
143 | len: sizeof(struct udphdr), sum: skb->csum); |
144 | |
145 | for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) { |
146 | csum = csum_add(csum, addend: skb->csum); |
147 | } |
148 | return csum; |
149 | } |
150 | |
151 | static inline __sum16 udp_v4_check(int len, __be32 saddr, |
152 | __be32 daddr, __wsum base) |
153 | { |
154 | return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, sum: base); |
155 | } |
156 | |
157 | void udp_set_csum(bool nocheck, struct sk_buff *skb, |
158 | __be32 saddr, __be32 daddr, int len); |
159 | |
160 | static inline void (struct sk_buff *skb) |
161 | { |
162 | if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE) |
163 | skb->csum = csum_partial(buff: skb->data, len: sizeof(struct udphdr), |
164 | sum: skb->csum); |
165 | skb_pull_rcsum(skb, len: sizeof(struct udphdr)); |
166 | UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr); |
167 | } |
168 | |
169 | typedef struct sock *(*udp_lookup_t)(const struct sk_buff *skb, __be16 sport, |
170 | __be16 dport); |
171 | |
172 | void udp_v6_early_demux(struct sk_buff *skb); |
173 | INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *)); |
174 | |
175 | struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb, |
176 | netdev_features_t features, bool is_ipv6); |
177 | |
178 | static inline void udp_lib_init_sock(struct sock *sk) |
179 | { |
180 | struct udp_sock *up = udp_sk(sk); |
181 | |
182 | skb_queue_head_init(list: &up->reader_queue); |
183 | up->forward_threshold = sk->sk_rcvbuf >> 2; |
184 | set_bit(SOCK_CUSTOM_SOCKOPT, addr: &sk->sk_socket->flags); |
185 | } |
186 | |
187 | /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */ |
188 | static inline int udp_lib_hash(struct sock *sk) |
189 | { |
190 | BUG(); |
191 | return 0; |
192 | } |
193 | |
194 | void udp_lib_unhash(struct sock *sk); |
195 | void udp_lib_rehash(struct sock *sk, u16 new_hash); |
196 | |
197 | static inline void udp_lib_close(struct sock *sk, long timeout) |
198 | { |
199 | sk_common_release(sk); |
200 | } |
201 | |
202 | int udp_lib_get_port(struct sock *sk, unsigned short snum, |
203 | unsigned int hash2_nulladdr); |
204 | |
205 | u32 udp_flow_hashrnd(void); |
206 | |
207 | static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb, |
208 | int min, int max, bool use_eth) |
209 | { |
210 | u32 hash; |
211 | |
212 | if (min >= max) { |
213 | /* Use default range */ |
214 | inet_get_local_port_range(net, low: &min, high: &max); |
215 | } |
216 | |
217 | hash = skb_get_hash(skb); |
218 | if (unlikely(!hash)) { |
219 | if (use_eth) { |
220 | /* Can't find a normal hash, caller has indicated an |
221 | * Ethernet packet so use that to compute a hash. |
222 | */ |
223 | hash = jhash(key: skb->data, length: 2 * ETH_ALEN, |
224 | initval: (__force u32) skb->protocol); |
225 | } else { |
226 | /* Can't derive any sort of hash for the packet, set |
227 | * to some consistent random value. |
228 | */ |
229 | hash = udp_flow_hashrnd(); |
230 | } |
231 | } |
232 | |
233 | /* Since this is being sent on the wire obfuscate hash a bit |
234 | * to minimize possbility that any useful information to an |
235 | * attacker is leaked. Only upper 16 bits are relevant in the |
236 | * computation for 16 bit port value. |
237 | */ |
238 | hash ^= hash << 16; |
239 | |
240 | return htons((((u64) hash * (max - min)) >> 32) + min); |
241 | } |
242 | |
243 | static inline int udp_rqueue_get(struct sock *sk) |
244 | { |
245 | return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit); |
246 | } |
247 | |
248 | static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if, |
249 | int dif, int sdif) |
250 | { |
251 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) |
252 | return inet_bound_dev_eq(l3mdev_accept: !!READ_ONCE(net->ipv4.sysctl_udp_l3mdev_accept), |
253 | bound_dev_if, dif, sdif); |
254 | #else |
255 | return inet_bound_dev_eq(true, bound_dev_if, dif, sdif); |
256 | #endif |
257 | } |
258 | |
259 | /* net/ipv4/udp.c */ |
260 | void udp_destruct_common(struct sock *sk); |
261 | void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len); |
262 | int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb); |
263 | void udp_skb_destructor(struct sock *sk, struct sk_buff *skb); |
264 | struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, int *off, |
265 | int *err); |
266 | static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags, |
267 | int *err) |
268 | { |
269 | int off = 0; |
270 | |
271 | return __skb_recv_udp(sk, flags, off: &off, err); |
272 | } |
273 | |
274 | int udp_v4_early_demux(struct sk_buff *skb); |
275 | bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst); |
276 | int udp_err(struct sk_buff *, u32); |
277 | int udp_abort(struct sock *sk, int err); |
278 | int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len); |
279 | void udp_splice_eof(struct socket *sock); |
280 | int udp_push_pending_frames(struct sock *sk); |
281 | void udp_flush_pending_frames(struct sock *sk); |
282 | int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size); |
283 | void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst); |
284 | int udp_rcv(struct sk_buff *skb); |
285 | int udp_ioctl(struct sock *sk, int cmd, int *karg); |
286 | int udp_init_sock(struct sock *sk); |
287 | int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); |
288 | int __udp_disconnect(struct sock *sk, int flags); |
289 | int udp_disconnect(struct sock *sk, int flags); |
290 | __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait); |
291 | struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb, |
292 | netdev_features_t features, |
293 | bool is_ipv6); |
294 | int udp_lib_getsockopt(struct sock *sk, int level, int optname, |
295 | char __user *optval, int __user *optlen); |
296 | int udp_lib_setsockopt(struct sock *sk, int level, int optname, |
297 | sockptr_t optval, unsigned int optlen, |
298 | int (*push_pending_frames)(struct sock *)); |
299 | struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, |
300 | __be32 daddr, __be16 dport, int dif); |
301 | struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, |
302 | __be32 daddr, __be16 dport, int dif, int sdif, |
303 | struct udp_table *tbl, struct sk_buff *skb); |
304 | struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, |
305 | __be16 sport, __be16 dport); |
306 | struct sock *udp6_lib_lookup(struct net *net, |
307 | const struct in6_addr *saddr, __be16 sport, |
308 | const struct in6_addr *daddr, __be16 dport, |
309 | int dif); |
310 | struct sock *__udp6_lib_lookup(struct net *net, |
311 | const struct in6_addr *saddr, __be16 sport, |
312 | const struct in6_addr *daddr, __be16 dport, |
313 | int dif, int sdif, struct udp_table *tbl, |
314 | struct sk_buff *skb); |
315 | struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, |
316 | __be16 sport, __be16 dport); |
317 | int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); |
318 | |
319 | /* UDP uses skb->dev_scratch to cache as much information as possible and avoid |
320 | * possibly multiple cache miss on dequeue() |
321 | */ |
322 | struct udp_dev_scratch { |
323 | /* skb->truesize and the stateless bit are embedded in a single field; |
324 | * do not use a bitfield since the compiler emits better/smaller code |
325 | * this way |
326 | */ |
327 | u32 _tsize_state; |
328 | |
329 | #if BITS_PER_LONG == 64 |
330 | /* len and the bit needed to compute skb_csum_unnecessary |
331 | * will be on cold cache lines at recvmsg time. |
332 | * skb->len can be stored on 16 bits since the udp header has been |
333 | * already validated and pulled. |
334 | */ |
335 | u16 len; |
336 | bool is_linear; |
337 | bool csum_unnecessary; |
338 | #endif |
339 | }; |
340 | |
341 | static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb) |
342 | { |
343 | return (struct udp_dev_scratch *)&skb->dev_scratch; |
344 | } |
345 | |
346 | #if BITS_PER_LONG == 64 |
347 | static inline unsigned int udp_skb_len(struct sk_buff *skb) |
348 | { |
349 | return udp_skb_scratch(skb)->len; |
350 | } |
351 | |
352 | static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb) |
353 | { |
354 | return udp_skb_scratch(skb)->csum_unnecessary; |
355 | } |
356 | |
357 | static inline bool udp_skb_is_linear(struct sk_buff *skb) |
358 | { |
359 | return udp_skb_scratch(skb)->is_linear; |
360 | } |
361 | |
362 | #else |
363 | static inline unsigned int udp_skb_len(struct sk_buff *skb) |
364 | { |
365 | return skb->len; |
366 | } |
367 | |
368 | static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb) |
369 | { |
370 | return skb_csum_unnecessary(skb); |
371 | } |
372 | |
373 | static inline bool udp_skb_is_linear(struct sk_buff *skb) |
374 | { |
375 | return !skb_is_nonlinear(skb); |
376 | } |
377 | #endif |
378 | |
379 | static inline int copy_linear_skb(struct sk_buff *skb, int len, int off, |
380 | struct iov_iter *to) |
381 | { |
382 | int n; |
383 | |
384 | n = copy_to_iter(addr: skb->data + off, bytes: len, i: to); |
385 | if (n == len) |
386 | return 0; |
387 | |
388 | iov_iter_revert(i: to, bytes: n); |
389 | return -EFAULT; |
390 | } |
391 | |
392 | /* |
393 | * SNMP statistics for UDP and UDP-Lite |
394 | */ |
395 | #define UDP_INC_STATS(net, field, is_udplite) do { \ |
396 | if (is_udplite) SNMP_INC_STATS((net)->mib.udplite_statistics, field); \ |
397 | else SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0) |
398 | #define __UDP_INC_STATS(net, field, is_udplite) do { \ |
399 | if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field); \ |
400 | else __SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0) |
401 | |
402 | #define __UDP6_INC_STATS(net, field, is_udplite) do { \ |
403 | if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\ |
404 | else __SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \ |
405 | } while(0) |
406 | #define UDP6_INC_STATS(net, field, __lite) do { \ |
407 | if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field); \ |
408 | else SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \ |
409 | } while(0) |
410 | |
411 | #if IS_ENABLED(CONFIG_IPV6) |
412 | #define __UDPX_MIB(sk, ipv4) \ |
413 | ({ \ |
414 | ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \ |
415 | sock_net(sk)->mib.udp_statistics) : \ |
416 | (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 : \ |
417 | sock_net(sk)->mib.udp_stats_in6); \ |
418 | }) |
419 | #else |
420 | #define __UDPX_MIB(sk, ipv4) \ |
421 | ({ \ |
422 | IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \ |
423 | sock_net(sk)->mib.udp_statistics; \ |
424 | }) |
425 | #endif |
426 | |
427 | #define __UDPX_INC_STATS(sk, field) \ |
428 | __SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field) |
429 | |
430 | #ifdef CONFIG_PROC_FS |
431 | struct udp_seq_afinfo { |
432 | sa_family_t family; |
433 | struct udp_table *udp_table; |
434 | }; |
435 | |
436 | struct udp_iter_state { |
437 | struct seq_net_private p; |
438 | int bucket; |
439 | }; |
440 | |
441 | void *udp_seq_start(struct seq_file *seq, loff_t *pos); |
442 | void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos); |
443 | void udp_seq_stop(struct seq_file *seq, void *v); |
444 | |
445 | extern const struct seq_operations udp_seq_ops; |
446 | extern const struct seq_operations udp6_seq_ops; |
447 | |
448 | int udp4_proc_init(void); |
449 | void udp4_proc_exit(void); |
450 | #endif /* CONFIG_PROC_FS */ |
451 | |
452 | int udpv4_offload_init(void); |
453 | |
454 | void udp_init(void); |
455 | |
456 | DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key); |
457 | void udp_encap_enable(void); |
458 | void udp_encap_disable(void); |
459 | #if IS_ENABLED(CONFIG_IPV6) |
460 | DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key); |
461 | void udpv6_encap_enable(void); |
462 | #endif |
463 | |
464 | static inline struct sk_buff *udp_rcv_segment(struct sock *sk, |
465 | struct sk_buff *skb, bool ipv4) |
466 | { |
467 | netdev_features_t features = NETIF_F_SG; |
468 | struct sk_buff *segs; |
469 | |
470 | /* Avoid csum recalculation by skb_segment unless userspace explicitly |
471 | * asks for the final checksum values |
472 | */ |
473 | if (!inet_get_convert_csum(sk)) |
474 | features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; |
475 | |
476 | /* UDP segmentation expects packets of type CHECKSUM_PARTIAL or |
477 | * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial |
478 | * packets in udp_gro_complete_segment. As does UDP GSO, verified by |
479 | * udp_send_skb. But when those packets are looped in dev_loopback_xmit |
480 | * their ip_summed CHECKSUM_NONE is changed to CHECKSUM_UNNECESSARY. |
481 | * Reset in this specific case, where PARTIAL is both correct and |
482 | * required. |
483 | */ |
484 | if (skb->pkt_type == PACKET_LOOPBACK) |
485 | skb->ip_summed = CHECKSUM_PARTIAL; |
486 | |
487 | /* the GSO CB lays after the UDP one, no need to save and restore any |
488 | * CB fragment |
489 | */ |
490 | segs = __skb_gso_segment(skb, features, tx_path: false); |
491 | if (IS_ERR_OR_NULL(ptr: segs)) { |
492 | int segs_nr = skb_shinfo(skb)->gso_segs; |
493 | |
494 | atomic_add(i: segs_nr, v: &sk->sk_drops); |
495 | SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr); |
496 | kfree_skb(skb); |
497 | return NULL; |
498 | } |
499 | |
500 | consume_skb(skb); |
501 | return segs; |
502 | } |
503 | |
504 | static inline void udp_post_segment_fix_csum(struct sk_buff *skb) |
505 | { |
506 | /* UDP-lite can't land here - no GRO */ |
507 | WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov); |
508 | |
509 | /* UDP packets generated with UDP_SEGMENT and traversing: |
510 | * |
511 | * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx) |
512 | * |
513 | * can reach an UDP socket with CHECKSUM_NONE, because |
514 | * __iptunnel_pull_header() converts CHECKSUM_PARTIAL into NONE. |
515 | * SKB_GSO_UDP_L4 or SKB_GSO_FRAGLIST packets with no UDP tunnel will |
516 | * have a valid checksum, as the GRO engine validates the UDP csum |
517 | * before the aggregation and nobody strips such info in between. |
518 | * Instead of adding another check in the tunnel fastpath, we can force |
519 | * a valid csum after the segmentation. |
520 | * Additionally fixup the UDP CB. |
521 | */ |
522 | UDP_SKB_CB(skb)->cscov = skb->len; |
523 | if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid) |
524 | skb->csum_valid = 1; |
525 | } |
526 | |
527 | #ifdef CONFIG_BPF_SYSCALL |
528 | struct sk_psock; |
529 | int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); |
530 | #endif |
531 | |
532 | #endif /* _UDP_H */ |
533 | |