net/mptcp/protocol.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 /* Multipath TCP
   3  *
   4  * Copyright (c) 2017 - 2019, Intel Corporation.
   5  */
   6
   7 #ifndef __MPTCP_PROTOCOL_H
   8 #define __MPTCP_PROTOCOL_H
   9
  10 #include <linux/random.h>
  11 #include <net/tcp.h>
  12 #include <net/inet_connection_sock.h>
  13
  14 #define MPTCP_SUPPORTED_VERSION 1
  15
  16 /* MPTCP option bits */
  17 #define OPTION_MPTCP_MPC_SYN    BIT(0)
  18 #define OPTION_MPTCP_MPC_SYNACK BIT(1)
  19 #define OPTION_MPTCP_MPC_ACK    BIT(2)
  20 #define OPTION_MPTCP_MPJ_SYN    BIT(3)
  21 #define OPTION_MPTCP_MPJ_SYNACK BIT(4)
  22 #define OPTION_MPTCP_MPJ_ACK    BIT(5)
  23 #define OPTION_MPTCP_ADD_ADDR   BIT(6)
  24 #define OPTION_MPTCP_ADD_ADDR6  BIT(7)
  25 #define OPTION_MPTCP_RM_ADDR    BIT(8)
  26
  27 /* MPTCP option subtypes */
  28 #define MPTCPOPT_MP_CAPABLE     0
  29 #define MPTCPOPT_MP_JOIN        1
  30 #define MPTCPOPT_DSS            2
  31 #define MPTCPOPT_ADD_ADDR       3
  32 #define MPTCPOPT_RM_ADDR        4
  33 #define MPTCPOPT_MP_PRIO        5
  34 #define MPTCPOPT_MP_FAIL        6
  35 #define MPTCPOPT_MP_FASTCLOSE   7
  36
  37 /* MPTCP suboption lengths */
  38 #define TCPOLEN_MPTCP_MPC_SYN           4
  39 #define TCPOLEN_MPTCP_MPC_SYNACK        12
  40 #define TCPOLEN_MPTCP_MPC_ACK           20
  41 #define TCPOLEN_MPTCP_MPC_ACK_DATA      22
  42 #define TCPOLEN_MPTCP_MPJ_SYN           12
  43 #define TCPOLEN_MPTCP_MPJ_SYNACK        16
  44 #define TCPOLEN_MPTCP_MPJ_ACK           24
  45 #define TCPOLEN_MPTCP_DSS_BASE          4
  46 #define TCPOLEN_MPTCP_DSS_ACK32         4
  47 #define TCPOLEN_MPTCP_DSS_ACK64         8
  48 #define TCPOLEN_MPTCP_DSS_MAP32         10
  49 #define TCPOLEN_MPTCP_DSS_MAP64         14
  50 #define TCPOLEN_MPTCP_DSS_CHECKSUM      2
  51 #define TCPOLEN_MPTCP_ADD_ADDR          16
  52 #define TCPOLEN_MPTCP_ADD_ADDR_PORT     18
  53 #define TCPOLEN_MPTCP_ADD_ADDR_BASE     8
  54 #define TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT        10
  55 #define TCPOLEN_MPTCP_ADD_ADDR6         28
  56 #define TCPOLEN_MPTCP_ADD_ADDR6_PORT    30
  57 #define TCPOLEN_MPTCP_ADD_ADDR6_BASE    20
  58 #define TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT       22
  59 #define TCPOLEN_MPTCP_PORT_LEN          2
  60 #define TCPOLEN_MPTCP_RM_ADDR_BASE      4
  61
  62 /* MPTCP MP_JOIN flags */
  63 #define MPTCPOPT_BACKUP         BIT(0)
  64 #define MPTCPOPT_HMAC_LEN       20
  65 #define MPTCPOPT_THMAC_LEN      8
  66
  67 /* MPTCP MP_CAPABLE flags */
  68 #define MPTCP_VERSION_MASK      (0x0F)
  69 #define MPTCP_CAP_CHECKSUM_REQD BIT(7)
  70 #define MPTCP_CAP_EXTENSIBILITY BIT(6)
  71 #define MPTCP_CAP_HMAC_SHA256   BIT(0)
  72 #define MPTCP_CAP_FLAG_MASK     (0x3F)
  73
  74 /* MPTCP DSS flags */
  75 #define MPTCP_DSS_DATA_FIN      BIT(4)
  76 #define MPTCP_DSS_DSN64         BIT(3)
  77 #define MPTCP_DSS_HAS_MAP       BIT(2)
  78 #define MPTCP_DSS_ACK64         BIT(1)
  79 #define MPTCP_DSS_HAS_ACK       BIT(0)
  80 #define MPTCP_DSS_FLAG_MASK     (0x1F)
  81
  82 /* MPTCP ADD_ADDR flags */
  83 #define MPTCP_ADDR_ECHO         BIT(0)
  84 #define MPTCP_ADDR_HMAC_LEN     20
  85 #define MPTCP_ADDR_IPVERSION_4  4
  86 #define MPTCP_ADDR_IPVERSION_6  6
  87
  88 /* MPTCP socket flags */
  89 #define MPTCP_DATA_READY        0
  90 #define MPTCP_SEND_SPACE        1
  91 #define MPTCP_WORK_RTX          2
  92 #define MPTCP_WORK_EOF          3
  93
  94 static inline __be32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
  95 {
  96         return htonl((TCPOPT_MPTCP << 24) | (len << 16) | (subopt << 12) |
  97                      ((nib & 0xF) << 8) | field);
  98 }
  99
 100 #define MPTCP_PM_MAX_ADDR       4
 101
 102 struct mptcp_addr_info {
 103         sa_family_t             family;
 104         __be16                  port;
 105         u8                      id;
 106         union {
 107                 struct in_addr addr;
 108 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 109                 struct in6_addr addr6;
 110 #endif
 111         };
 112 };
 113
 114 enum mptcp_pm_status {
 115         MPTCP_PM_ADD_ADDR_RECEIVED,
 116         MPTCP_PM_ESTABLISHED,
 117         MPTCP_PM_SUBFLOW_ESTABLISHED,
 118 };
 119
 120 struct mptcp_pm_data {
 121         struct mptcp_addr_info local;
 122         struct mptcp_addr_info remote;
 123
 124         spinlock_t      lock;           /*protects the whole PM data */
 125
 126         bool            addr_signal;
 127         bool            server_side;
 128         bool            work_pending;
 129         bool            accept_addr;
 130         bool            accept_subflow;
 131         u8              add_addr_signaled;
 132         u8              add_addr_accepted;
 133         u8              local_addr_used;
 134         u8              subflows;
 135         u8              add_addr_signal_max;
 136         u8              add_addr_accept_max;
 137         u8              local_addr_max;
 138         u8              subflows_max;
 139         u8              status;
 140
 141         struct          work_struct work;
 142 };
 143
 144 struct mptcp_data_frag {
 145         struct list_head list;
 146         u64 data_seq;
 147         int data_len;
 148         int offset;
 149         int overhead;
 150         struct page *page;
 151 };
 152
 153 /* MPTCP connection sock */
 154 struct mptcp_sock {
 155         /* inet_connection_sock must be the first member */
 156         struct inet_connection_sock sk;
 157         u64             local_key;
 158         u64             remote_key;
 159         u64             write_seq;
 160         u64             ack_seq;
 161         atomic64_t      snd_una;
 162         unsigned long   timer_ival;
 163         u32             token;
 164         unsigned long   flags;
 165         bool            can_ack;
 166         spinlock_t      join_list_lock;
 167         struct work_struct work;
 168         struct list_head conn_list;
 169         struct list_head rtx_queue;
 170         struct list_head join_list;
 171         struct skb_ext  *cached_ext;    /* for the next sendmsg */
 172         struct socket   *subflow; /* outgoing connect/listener/!mp_capable */
 173         struct sock     *first;
 174         struct mptcp_pm_data    pm;
 175 };
 176
 177 #define mptcp_for_each_subflow(__msk, __subflow)                        \
 178         list_for_each_entry(__subflow, &((__msk)->conn_list), node)
 179
 180 static inline struct mptcp_sock *mptcp_sk(const struct sock *sk)
 181 {
 182         return (struct mptcp_sock *)sk;
 183 }
 184
 185 static inline struct mptcp_data_frag *mptcp_rtx_tail(const struct sock *sk)
 186 {
 187         struct mptcp_sock *msk = mptcp_sk(sk);
 188
 189         if (list_empty(&msk->rtx_queue))
 190                 return NULL;
 191
 192         return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
 193 }
 194
 195 static inline struct mptcp_data_frag *mptcp_rtx_head(const struct sock *sk)
 196 {
 197         struct mptcp_sock *msk = mptcp_sk(sk);
 198
 199         if (list_empty(&msk->rtx_queue))
 200                 return NULL;
 201
 202         return list_first_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
 203 }
 204
 205 struct mptcp_subflow_request_sock {
 206         struct  tcp_request_sock sk;
 207         u16     mp_capable : 1,
 208                 mp_join : 1,
 209                 backup : 1,
 210                 remote_key_valid : 1;
 211         u8      local_id;
 212         u8      remote_id;
 213         u64     local_key;
 214         u64     remote_key;
 215         u64     idsn;
 216         u32     token;
 217         u32     ssn_offset;
 218         u64     thmac;
 219         u32     local_nonce;
 220         u32     remote_nonce;
 221 };
 222
 223 static inline struct mptcp_subflow_request_sock *
 224 mptcp_subflow_rsk(const struct request_sock *rsk)
 225 {
 226         return (struct mptcp_subflow_request_sock *)rsk;
 227 }
 228
 229 /* MPTCP subflow context */
 230 struct mptcp_subflow_context {
 231         struct  list_head node;/* conn_list of subflows */
 232         u64     local_key;
 233         u64     remote_key;
 234         u64     idsn;
 235         u64     map_seq;
 236         u32     snd_isn;
 237         u32     token;
 238         u32     rel_write_seq;
 239         u32     map_subflow_seq;
 240         u32     ssn_offset;
 241         u32     map_data_len;
 242         u32     request_mptcp : 1,  /* send MP_CAPABLE */
 243                 request_join : 1,   /* send MP_JOIN */
 244                 request_bkup : 1,
 245                 mp_capable : 1,     /* remote is MPTCP capable */
 246                 mp_join : 1,        /* remote is JOINing */
 247                 fully_established : 1,      /* path validated */
 248                 pm_notified : 1,    /* PM hook called for established status */
 249                 conn_finished : 1,
 250                 map_valid : 1,
 251                 mpc_map : 1,
 252                 backup : 1,
 253                 data_avail : 1,
 254                 rx_eof : 1,
 255                 data_fin_tx_enable : 1,
 256                 can_ack : 1;        /* only after processing the remote a key */
 257         u64     data_fin_tx_seq;
 258         u32     remote_nonce;
 259         u64     thmac;
 260         u32     local_nonce;
 261         u32     remote_token;
 262         u8      hmac[MPTCPOPT_HMAC_LEN];
 263         u8      local_id;
 264         u8      remote_id;
 265
 266         struct  sock *tcp_sock;     /* tcp sk backpointer */
 267         struct  sock *conn;         /* parent mptcp_sock */
 268         const   struct inet_connection_sock_af_ops *icsk_af_ops;
 269         void    (*tcp_data_ready)(struct sock *sk);
 270         void    (*tcp_state_change)(struct sock *sk);
 271         void    (*tcp_write_space)(struct sock *sk);
 272
 273         struct  rcu_head rcu;
 274 };
 275
 276 static inline struct mptcp_subflow_context *
 277 mptcp_subflow_ctx(const struct sock *sk)
 278 {
 279         struct inet_connection_sock *icsk = inet_csk(sk);
 280
 281         /* Use RCU on icsk_ulp_data only for sock diag code */
 282         return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data;
 283 }
 284
 285 static inline struct sock *
 286 mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow)
 287 {
 288         return subflow->tcp_sock;
 289 }
 290
 291 static inline u64
 292 mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow)
 293 {
 294         return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq -
 295                       subflow->ssn_offset -
 296                       subflow->map_subflow_seq;
 297 }
 298
 299 static inline u64
 300 mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow)
 301 {
 302         return subflow->map_seq + mptcp_subflow_get_map_offset(subflow);
 303 }
 304
 305 int mptcp_is_enabled(struct net *net);
 306 bool mptcp_subflow_data_available(struct sock *sk);
 307 void mptcp_subflow_init(void);
 308
 309 /* called with sk socket lock held */
 310 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
 311                             const struct mptcp_addr_info *loc,
 312                             const struct mptcp_addr_info *remote);
 313 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock);
 314
 315 static inline void mptcp_subflow_tcp_fallback(struct sock *sk,
 316                                               struct mptcp_subflow_context *ctx)
 317 {
 318         sk->sk_data_ready = ctx->tcp_data_ready;
 319         sk->sk_state_change = ctx->tcp_state_change;
 320         sk->sk_write_space = ctx->tcp_write_space;
 321
 322         inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops;
 323 }
 324
 325 extern const struct inet_connection_sock_af_ops ipv4_specific;
 326 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 327 extern const struct inet_connection_sock_af_ops ipv6_specific;
 328 #endif
 329
 330 void mptcp_proto_init(void);
 331 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 332 int mptcp_proto_v6_init(void);
 333 #endif
 334
 335 struct sock *mptcp_sk_clone(const struct sock *sk, struct request_sock *req);
 336 void mptcp_get_options(const struct sk_buff *skb,
 337                        struct tcp_options_received *opt_rx);
 338
 339 void mptcp_finish_connect(struct sock *sk);
 340 void mptcp_data_ready(struct sock *sk, struct sock *ssk);
 341 bool mptcp_finish_join(struct sock *sk);
 342 void mptcp_data_acked(struct sock *sk);
 343 void mptcp_subflow_eof(struct sock *sk);
 344
 345 int mptcp_token_new_request(struct request_sock *req);
 346 void mptcp_token_destroy_request(u32 token);
 347 int mptcp_token_new_connect(struct sock *sk);
 348 int mptcp_token_new_accept(u32 token, struct sock *conn);
 349 struct mptcp_sock *mptcp_token_get_sock(u32 token);
 350 void mptcp_token_destroy(u32 token);
 351
 352 void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn);
 353 static inline void mptcp_crypto_key_gen_sha(u64 *key, u32 *token, u64 *idsn)
 354 {
 355         /* we might consider a faster version that computes the key as a
 356          * hash of some information available in the MPTCP socket. Use
 357          * random data at the moment, as it's probably the safest option
 358          * in case multiple sockets are opened in different namespaces at
 359          * the same time.
 360          */
 361         get_random_bytes(key, sizeof(u64));
 362         mptcp_crypto_key_sha(*key, token, idsn);
 363 }
 364
 365 void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac);
 366
 367 void mptcp_pm_init(void);
 368 void mptcp_pm_data_init(struct mptcp_sock *msk);
 369 void mptcp_pm_close(struct mptcp_sock *msk);
 370 void mptcp_pm_new_connection(struct mptcp_sock *msk, int server_side);
 371 void mptcp_pm_fully_established(struct mptcp_sock *msk);
 372 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk);
 373 void mptcp_pm_connection_closed(struct mptcp_sock *msk);
 374 void mptcp_pm_subflow_established(struct mptcp_sock *msk,
 375                                   struct mptcp_subflow_context *subflow);
 376 void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id);
 377 void mptcp_pm_add_addr_received(struct mptcp_sock *msk,
 378                                 const struct mptcp_addr_info *addr);
 379
 380 int mptcp_pm_announce_addr(struct mptcp_sock *msk,
 381                            const struct mptcp_addr_info *addr);
 382 int mptcp_pm_remove_addr(struct mptcp_sock *msk, u8 local_id);
 383 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, u8 remote_id);
 384
 385 static inline bool mptcp_pm_should_signal(struct mptcp_sock *msk)
 386 {
 387         return READ_ONCE(msk->pm.addr_signal);
 388 }
 389
 390 static inline unsigned int mptcp_add_addr_len(int family)
 391 {
 392         if (family == AF_INET)
 393                 return TCPOLEN_MPTCP_ADD_ADDR;
 394         return TCPOLEN_MPTCP_ADD_ADDR6;
 395 }
 396
 397 bool mptcp_pm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
 398                           struct mptcp_addr_info *saddr);
 399 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
 400
 401 void mptcp_pm_nl_init(void);
 402 void mptcp_pm_nl_data_init(struct mptcp_sock *msk);
 403 void mptcp_pm_nl_fully_established(struct mptcp_sock *msk);
 404 void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk);
 405 void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk);
 406 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
 407
 408 static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb)
 409 {
 410         return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
 411 }
 412
 413 static inline bool before64(__u64 seq1, __u64 seq2)
 414 {
 415         return (__s64)(seq1 - seq2) < 0;
 416 }
 417
 418 #define after64(seq2, seq1)     before64(seq1, seq2)
 419
 420 void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops);
 421
 422 #endif /* __MPTCP_PROTOCOL_H */