Back to home page

OSCL-LXR

 
 

    


0001 /* SPDX-License-Identifier: GPL-2.0-or-later */
0002 /*
0003  * INET     An implementation of the TCP/IP protocol suite for the LINUX
0004  *      operating system.  INET is implemented using the  BSD Socket
0005  *      interface as the means of communication with the user level.
0006  *
0007  *      Definitions for the UDP module.
0008  *
0009  * Version: @(#)udp.h   1.0.2   05/07/93
0010  *
0011  * Authors: Ross Biro
0012  *      Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
0013  *
0014  * Fixes:
0015  *      Alan Cox    : Turned on udp checksums. I don't want to
0016  *                chase 'memory corruption' bugs that aren't!
0017  */
0018 #ifndef _UDP_H
0019 #define _UDP_H
0020 
0021 #include <linux/list.h>
0022 #include <linux/bug.h>
0023 #include <net/inet_sock.h>
0024 #include <net/sock.h>
0025 #include <net/snmp.h>
0026 #include <net/ip.h>
0027 #include <linux/ipv6.h>
0028 #include <linux/seq_file.h>
0029 #include <linux/poll.h>
0030 #include <linux/indirect_call_wrapper.h>
0031 
0032 /**
0033  *  struct udp_skb_cb  -  UDP(-Lite) private variables
0034  *
0035  *  @header:      private variables used by IPv4/IPv6
0036  *  @cscov:       checksum coverage length (UDP-Lite only)
0037  *  @partial_cov: if set indicates partial csum coverage
0038  */
0039 struct udp_skb_cb {
0040     union {
0041         struct inet_skb_parm    h4;
0042 #if IS_ENABLED(CONFIG_IPV6)
0043         struct inet6_skb_parm   h6;
0044 #endif
0045     } header;
0046     __u16       cscov;
0047     __u8        partial_cov;
0048 };
0049 #define UDP_SKB_CB(__skb)   ((struct udp_skb_cb *)((__skb)->cb))
0050 
0051 /**
0052  *  struct udp_hslot - UDP hash slot
0053  *
0054  *  @head:  head of list of sockets
0055  *  @count: number of sockets in 'head' list
0056  *  @lock:  spinlock protecting changes to head/count
0057  */
0058 struct udp_hslot {
0059     struct hlist_head   head;
0060     int         count;
0061     spinlock_t      lock;
0062 } __attribute__((aligned(2 * sizeof(long))));
0063 
0064 /**
0065  *  struct udp_table - UDP table
0066  *
0067  *  @hash:  hash table, sockets are hashed on (local port)
0068  *  @hash2: hash table, sockets are hashed on (local port, local address)
0069  *  @mask:  number of slots in hash tables, minus 1
0070  *  @log:   log2(number of slots in hash table)
0071  */
0072 struct udp_table {
0073     struct udp_hslot    *hash;
0074     struct udp_hslot    *hash2;
0075     unsigned int        mask;
0076     unsigned int        log;
0077 };
0078 extern struct udp_table udp_table;
0079 void udp_table_init(struct udp_table *, const char *);
0080 static inline struct udp_hslot *udp_hashslot(struct udp_table *table,
0081                          struct net *net, unsigned int num)
0082 {
0083     return &table->hash[udp_hashfn(net, num, table->mask)];
0084 }
0085 /*
0086  * For secondary hash, net_hash_mix() is performed before calling
0087  * udp_hashslot2(), this explains difference with udp_hashslot()
0088  */
0089 static inline struct udp_hslot *udp_hashslot2(struct udp_table *table,
0090                           unsigned int hash)
0091 {
0092     return &table->hash2[hash & table->mask];
0093 }
0094 
0095 extern struct proto udp_prot;
0096 
0097 extern atomic_long_t udp_memory_allocated;
0098 DECLARE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
0099 
0100 /* sysctl variables for udp */
0101 extern long sysctl_udp_mem[3];
0102 extern int sysctl_udp_rmem_min;
0103 extern int sysctl_udp_wmem_min;
0104 
0105 struct sk_buff;
0106 
0107 /*
0108  *  Generic checksumming routines for UDP(-Lite) v4 and v6
0109  */
0110 static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb)
0111 {
0112     return (UDP_SKB_CB(skb)->cscov == skb->len ?
0113         __skb_checksum_complete(skb) :
0114         __skb_checksum_complete_head(skb, UDP_SKB_CB(skb)->cscov));
0115 }
0116 
0117 static inline int udp_lib_checksum_complete(struct sk_buff *skb)
0118 {
0119     return !skb_csum_unnecessary(skb) &&
0120         __udp_lib_checksum_complete(skb);
0121 }
0122 
0123 /**
0124  *  udp_csum_outgoing  -  compute UDPv4/v6 checksum over fragments
0125  *  @sk:    socket we are writing to
0126  *  @skb:   sk_buff containing the filled-in UDP header
0127  *          (checksum field must be zeroed out)
0128  */
0129 static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb)
0130 {
0131     __wsum csum = csum_partial(skb_transport_header(skb),
0132                    sizeof(struct udphdr), 0);
0133     skb_queue_walk(&sk->sk_write_queue, skb) {
0134         csum = csum_add(csum, skb->csum);
0135     }
0136     return csum;
0137 }
0138 
0139 static inline __wsum udp_csum(struct sk_buff *skb)
0140 {
0141     __wsum csum = csum_partial(skb_transport_header(skb),
0142                    sizeof(struct udphdr), skb->csum);
0143 
0144     for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) {
0145         csum = csum_add(csum, skb->csum);
0146     }
0147     return csum;
0148 }
0149 
0150 static inline __sum16 udp_v4_check(int len, __be32 saddr,
0151                    __be32 daddr, __wsum base)
0152 {
0153     return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base);
0154 }
0155 
0156 void udp_set_csum(bool nocheck, struct sk_buff *skb,
0157           __be32 saddr, __be32 daddr, int len);
0158 
0159 static inline void udp_csum_pull_header(struct sk_buff *skb)
0160 {
0161     if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE)
0162         skb->csum = csum_partial(skb->data, sizeof(struct udphdr),
0163                      skb->csum);
0164     skb_pull_rcsum(skb, sizeof(struct udphdr));
0165     UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr);
0166 }
0167 
0168 typedef struct sock *(*udp_lookup_t)(const struct sk_buff *skb, __be16 sport,
0169                      __be16 dport);
0170 
0171 void udp_v6_early_demux(struct sk_buff *skb);
0172 INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
0173 
0174 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
0175                   netdev_features_t features, bool is_ipv6);
0176 
0177 /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
0178 static inline int udp_lib_hash(struct sock *sk)
0179 {
0180     BUG();
0181     return 0;
0182 }
0183 
0184 void udp_lib_unhash(struct sock *sk);
0185 void udp_lib_rehash(struct sock *sk, u16 new_hash);
0186 
0187 static inline void udp_lib_close(struct sock *sk, long timeout)
0188 {
0189     sk_common_release(sk);
0190 }
0191 
0192 int udp_lib_get_port(struct sock *sk, unsigned short snum,
0193              unsigned int hash2_nulladdr);
0194 
0195 u32 udp_flow_hashrnd(void);
0196 
0197 static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
0198                        int min, int max, bool use_eth)
0199 {
0200     u32 hash;
0201 
0202     if (min >= max) {
0203         /* Use default range */
0204         inet_get_local_port_range(net, &min, &max);
0205     }
0206 
0207     hash = skb_get_hash(skb);
0208     if (unlikely(!hash)) {
0209         if (use_eth) {
0210             /* Can't find a normal hash, caller has indicated an
0211              * Ethernet packet so use that to compute a hash.
0212              */
0213             hash = jhash(skb->data, 2 * ETH_ALEN,
0214                      (__force u32) skb->protocol);
0215         } else {
0216             /* Can't derive any sort of hash for the packet, set
0217              * to some consistent random value.
0218              */
0219             hash = udp_flow_hashrnd();
0220         }
0221     }
0222 
0223     /* Since this is being sent on the wire obfuscate hash a bit
0224      * to minimize possbility that any useful information to an
0225      * attacker is leaked. Only upper 16 bits are relevant in the
0226      * computation for 16 bit port value.
0227      */
0228     hash ^= hash << 16;
0229 
0230     return htons((((u64) hash * (max - min)) >> 32) + min);
0231 }
0232 
0233 static inline int udp_rqueue_get(struct sock *sk)
0234 {
0235     return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
0236 }
0237 
0238 static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if,
0239                        int dif, int sdif)
0240 {
0241 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
0242     return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_udp_l3mdev_accept),
0243                  bound_dev_if, dif, sdif);
0244 #else
0245     return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
0246 #endif
0247 }
0248 
0249 /* net/ipv4/udp.c */
0250 void udp_destruct_sock(struct sock *sk);
0251 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
0252 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
0253 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
0254 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, int *off,
0255                    int *err);
0256 static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
0257                        int *err)
0258 {
0259     int off = 0;
0260 
0261     return __skb_recv_udp(sk, flags, &off, err);
0262 }
0263 
0264 int udp_v4_early_demux(struct sk_buff *skb);
0265 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
0266 int udp_get_port(struct sock *sk, unsigned short snum,
0267          int (*saddr_cmp)(const struct sock *,
0268                   const struct sock *));
0269 int udp_err(struct sk_buff *, u32);
0270 int udp_abort(struct sock *sk, int err);
0271 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
0272 int udp_push_pending_frames(struct sock *sk);
0273 void udp_flush_pending_frames(struct sock *sk);
0274 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
0275 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
0276 int udp_rcv(struct sk_buff *skb);
0277 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
0278 int udp_init_sock(struct sock *sk);
0279 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
0280 int __udp_disconnect(struct sock *sk, int flags);
0281 int udp_disconnect(struct sock *sk, int flags);
0282 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
0283 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
0284                        netdev_features_t features,
0285                        bool is_ipv6);
0286 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
0287                char __user *optval, int __user *optlen);
0288 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
0289                sockptr_t optval, unsigned int optlen,
0290                int (*push_pending_frames)(struct sock *));
0291 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
0292                  __be32 daddr, __be16 dport, int dif);
0293 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
0294                    __be32 daddr, __be16 dport, int dif, int sdif,
0295                    struct udp_table *tbl, struct sk_buff *skb);
0296 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
0297                  __be16 sport, __be16 dport);
0298 struct sock *udp6_lib_lookup(struct net *net,
0299                  const struct in6_addr *saddr, __be16 sport,
0300                  const struct in6_addr *daddr, __be16 dport,
0301                  int dif);
0302 struct sock *__udp6_lib_lookup(struct net *net,
0303                    const struct in6_addr *saddr, __be16 sport,
0304                    const struct in6_addr *daddr, __be16 dport,
0305                    int dif, int sdif, struct udp_table *tbl,
0306                    struct sk_buff *skb);
0307 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
0308                  __be16 sport, __be16 dport);
0309 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
0310 
0311 /* UDP uses skb->dev_scratch to cache as much information as possible and avoid
0312  * possibly multiple cache miss on dequeue()
0313  */
0314 struct udp_dev_scratch {
0315     /* skb->truesize and the stateless bit are embedded in a single field;
0316      * do not use a bitfield since the compiler emits better/smaller code
0317      * this way
0318      */
0319     u32 _tsize_state;
0320 
0321 #if BITS_PER_LONG == 64
0322     /* len and the bit needed to compute skb_csum_unnecessary
0323      * will be on cold cache lines at recvmsg time.
0324      * skb->len can be stored on 16 bits since the udp header has been
0325      * already validated and pulled.
0326      */
0327     u16 len;
0328     bool is_linear;
0329     bool csum_unnecessary;
0330 #endif
0331 };
0332 
0333 static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
0334 {
0335     return (struct udp_dev_scratch *)&skb->dev_scratch;
0336 }
0337 
0338 #if BITS_PER_LONG == 64
0339 static inline unsigned int udp_skb_len(struct sk_buff *skb)
0340 {
0341     return udp_skb_scratch(skb)->len;
0342 }
0343 
0344 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
0345 {
0346     return udp_skb_scratch(skb)->csum_unnecessary;
0347 }
0348 
0349 static inline bool udp_skb_is_linear(struct sk_buff *skb)
0350 {
0351     return udp_skb_scratch(skb)->is_linear;
0352 }
0353 
0354 #else
0355 static inline unsigned int udp_skb_len(struct sk_buff *skb)
0356 {
0357     return skb->len;
0358 }
0359 
0360 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
0361 {
0362     return skb_csum_unnecessary(skb);
0363 }
0364 
0365 static inline bool udp_skb_is_linear(struct sk_buff *skb)
0366 {
0367     return !skb_is_nonlinear(skb);
0368 }
0369 #endif
0370 
0371 static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
0372                   struct iov_iter *to)
0373 {
0374     int n;
0375 
0376     n = copy_to_iter(skb->data + off, len, to);
0377     if (n == len)
0378         return 0;
0379 
0380     iov_iter_revert(to, n);
0381     return -EFAULT;
0382 }
0383 
0384 /*
0385  *  SNMP statistics for UDP and UDP-Lite
0386  */
0387 #define UDP_INC_STATS(net, field, is_udplite)             do { \
0388     if (is_udplite) SNMP_INC_STATS((net)->mib.udplite_statistics, field);       \
0389     else        SNMP_INC_STATS((net)->mib.udp_statistics, field);  }  while(0)
0390 #define __UDP_INC_STATS(net, field, is_udplite)           do { \
0391     if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field);         \
0392     else        __SNMP_INC_STATS((net)->mib.udp_statistics, field);    }  while(0)
0393 
0394 #define __UDP6_INC_STATS(net, field, is_udplite)        do { \
0395     if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
0396     else        __SNMP_INC_STATS((net)->mib.udp_stats_in6, field);  \
0397 } while(0)
0398 #define UDP6_INC_STATS(net, field, __lite)          do { \
0399     if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);  \
0400     else        SNMP_INC_STATS((net)->mib.udp_stats_in6, field);      \
0401 } while(0)
0402 
0403 #if IS_ENABLED(CONFIG_IPV6)
0404 #define __UDPX_MIB(sk, ipv4)                        \
0405 ({                                  \
0406     ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
0407                  sock_net(sk)->mib.udp_statistics) :    \
0408         (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 : \
0409                  sock_net(sk)->mib.udp_stats_in6);  \
0410 })
0411 #else
0412 #define __UDPX_MIB(sk, ipv4)                        \
0413 ({                                  \
0414     IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics :     \
0415              sock_net(sk)->mib.udp_statistics;      \
0416 })
0417 #endif
0418 
0419 #define __UDPX_INC_STATS(sk, field) \
0420     __SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
0421 
0422 #ifdef CONFIG_PROC_FS
0423 struct udp_seq_afinfo {
0424     sa_family_t         family;
0425     struct udp_table        *udp_table;
0426 };
0427 
0428 struct udp_iter_state {
0429     struct seq_net_private  p;
0430     int         bucket;
0431     struct udp_seq_afinfo   *bpf_seq_afinfo;
0432 };
0433 
0434 void *udp_seq_start(struct seq_file *seq, loff_t *pos);
0435 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
0436 void udp_seq_stop(struct seq_file *seq, void *v);
0437 
0438 extern const struct seq_operations udp_seq_ops;
0439 extern const struct seq_operations udp6_seq_ops;
0440 
0441 int udp4_proc_init(void);
0442 void udp4_proc_exit(void);
0443 #endif /* CONFIG_PROC_FS */
0444 
0445 int udpv4_offload_init(void);
0446 
0447 void udp_init(void);
0448 
0449 DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
0450 void udp_encap_enable(void);
0451 void udp_encap_disable(void);
0452 #if IS_ENABLED(CONFIG_IPV6)
0453 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
0454 void udpv6_encap_enable(void);
0455 #endif
0456 
0457 static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
0458                           struct sk_buff *skb, bool ipv4)
0459 {
0460     netdev_features_t features = NETIF_F_SG;
0461     struct sk_buff *segs;
0462 
0463     /* Avoid csum recalculation by skb_segment unless userspace explicitly
0464      * asks for the final checksum values
0465      */
0466     if (!inet_get_convert_csum(sk))
0467         features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
0468 
0469     /* UDP segmentation expects packets of type CHECKSUM_PARTIAL or
0470      * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial
0471      * packets in udp_gro_complete_segment. As does UDP GSO, verified by
0472      * udp_send_skb. But when those packets are looped in dev_loopback_xmit
0473      * their ip_summed CHECKSUM_NONE is changed to CHECKSUM_UNNECESSARY.
0474      * Reset in this specific case, where PARTIAL is both correct and
0475      * required.
0476      */
0477     if (skb->pkt_type == PACKET_LOOPBACK)
0478         skb->ip_summed = CHECKSUM_PARTIAL;
0479 
0480     /* the GSO CB lays after the UDP one, no need to save and restore any
0481      * CB fragment
0482      */
0483     segs = __skb_gso_segment(skb, features, false);
0484     if (IS_ERR_OR_NULL(segs)) {
0485         int segs_nr = skb_shinfo(skb)->gso_segs;
0486 
0487         atomic_add(segs_nr, &sk->sk_drops);
0488         SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
0489         kfree_skb(skb);
0490         return NULL;
0491     }
0492 
0493     consume_skb(skb);
0494     return segs;
0495 }
0496 
0497 static inline void udp_post_segment_fix_csum(struct sk_buff *skb)
0498 {
0499     /* UDP-lite can't land here - no GRO */
0500     WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov);
0501 
0502     /* UDP packets generated with UDP_SEGMENT and traversing:
0503      *
0504      * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx)
0505      *
0506      * can reach an UDP socket with CHECKSUM_NONE, because
0507      * __iptunnel_pull_header() converts CHECKSUM_PARTIAL into NONE.
0508      * SKB_GSO_UDP_L4 or SKB_GSO_FRAGLIST packets with no UDP tunnel will
0509      * have a valid checksum, as the GRO engine validates the UDP csum
0510      * before the aggregation and nobody strips such info in between.
0511      * Instead of adding another check in the tunnel fastpath, we can force
0512      * a valid csum after the segmentation.
0513      * Additionally fixup the UDP CB.
0514      */
0515     UDP_SKB_CB(skb)->cscov = skb->len;
0516     if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid)
0517         skb->csum_valid = 1;
0518 }
0519 
0520 #ifdef CONFIG_BPF_SYSCALL
0521 struct sk_psock;
0522 struct proto *udp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
0523 int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
0524 #endif
0525 
0526 #endif  /* _UDP_H */