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 struct mptcp_options_received {
  95         u64     sndr_key;
  96         u64     rcvr_key;
  97         u64     data_ack;
  98         u64     data_seq;
  99         u32     subflow_seq;
 100         u16     data_len;
 101         u16     mp_capable : 1,
 102                 mp_join : 1,
 103                 dss : 1,
 104                 add_addr : 1,
 105                 rm_addr : 1,
 106                 family : 4,
 107                 echo : 1,
 108                 backup : 1;
 109         u32     token;
 110         u32     nonce;
 111         u64     thmac;
 112         u8      hmac[20];
 113         u8      join_id;
 114         u8      use_map:1,
 115                 dsn64:1,
 116                 data_fin:1,
 117                 use_ack:1,
 118                 ack64:1,
 119                 mpc_map:1,
 120                 __unused:2;
 121         u8      addr_id;
 122         u8      rm_id;
 123         union {
 124                 struct in_addr  addr;
 125 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 126                 struct in6_addr addr6;
 127 #endif
 128         };
 129         u64     ahmac;
 130         u16     port;
 131 };
 132
 133 static inline __be32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
 134 {
 135         return htonl((TCPOPT_MPTCP << 24) | (len << 16) | (subopt << 12) |
 136                      ((nib & 0xF) << 8) | field);
 137 }
 138
 139 #define MPTCP_PM_MAX_ADDR       4
 140
 141 struct mptcp_addr_info {
 142         sa_family_t             family;
 143         __be16                  port;
 144         u8                      id;
 145         union {
 146                 struct in_addr addr;
 147 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 148                 struct in6_addr addr6;
 149 #endif
 150         };
 151 };
 152
 153 enum mptcp_pm_status {
 154         MPTCP_PM_ADD_ADDR_RECEIVED,
 155         MPTCP_PM_ESTABLISHED,
 156         MPTCP_PM_SUBFLOW_ESTABLISHED,
 157 };
 158
 159 struct mptcp_pm_data {
 160         struct mptcp_addr_info local;
 161         struct mptcp_addr_info remote;
 162
 163         spinlock_t      lock;           /*protects the whole PM data */
 164
 165         bool            addr_signal;
 166         bool            server_side;
 167         bool            work_pending;
 168         bool            accept_addr;
 169         bool            accept_subflow;
 170         u8              add_addr_signaled;
 171         u8              add_addr_accepted;
 172         u8              local_addr_used;
 173         u8              subflows;
 174         u8              add_addr_signal_max;
 175         u8              add_addr_accept_max;
 176         u8              local_addr_max;
 177         u8              subflows_max;
 178         u8              status;
 179
 180         struct          work_struct work;
 181 };
 182
 183 struct mptcp_data_frag {
 184         struct list_head list;
 185         u64 data_seq;
 186         int data_len;
 187         int offset;
 188         int overhead;
 189         struct page *page;
 190 };
 191
 192 /* MPTCP connection sock */
 193 struct mptcp_sock {
 194         /* inet_connection_sock must be the first member */
 195         struct inet_connection_sock sk;
 196         u64             local_key;
 197         u64             remote_key;
 198         u64             write_seq;
 199         u64             ack_seq;
 200         atomic64_t      snd_una;
 201         unsigned long   timer_ival;
 202         u32             token;
 203         unsigned long   flags;
 204         bool            can_ack;
 205         spinlock_t      join_list_lock;
 206         struct work_struct work;
 207         struct list_head conn_list;
 208         struct list_head rtx_queue;
 209         struct list_head join_list;
 210         struct skb_ext  *cached_ext;    /* for the next sendmsg */
 211         struct socket   *subflow; /* outgoing connect/listener/!mp_capable */
 212         struct sock     *first;
 213         struct mptcp_pm_data    pm;
 214 };
 215
 216 #define mptcp_for_each_subflow(__msk, __subflow)                        \
 217         list_for_each_entry(__subflow, &((__msk)->conn_list), node)
 218
 219 static inline struct mptcp_sock *mptcp_sk(const struct sock *sk)
 220 {
 221         return (struct mptcp_sock *)sk;
 222 }
 223
 224 static inline struct mptcp_data_frag *mptcp_rtx_tail(const struct sock *sk)
 225 {
 226         struct mptcp_sock *msk = mptcp_sk(sk);
 227
 228         if (list_empty(&msk->rtx_queue))
 229                 return NULL;
 230
 231         return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
 232 }
 233
 234 static inline struct mptcp_data_frag *mptcp_rtx_head(const struct sock *sk)
 235 {
 236         struct mptcp_sock *msk = mptcp_sk(sk);
 237
 238         if (list_empty(&msk->rtx_queue))
 239                 return NULL;
 240
 241         return list_first_entry(&msk->rtx_queue, struct mptcp_data_frag, list);
 242 }
 243
 244 struct mptcp_subflow_request_sock {
 245         struct  tcp_request_sock sk;
 246         u16     mp_capable : 1,
 247                 mp_join : 1,
 248                 backup : 1;
 249         u8      local_id;
 250         u8      remote_id;
 251         u64     local_key;
 252         u64     idsn;
 253         u32     token;
 254         u32     ssn_offset;
 255         u64     thmac;
 256         u32     local_nonce;
 257         u32     remote_nonce;
 258 };
 259
 260 static inline struct mptcp_subflow_request_sock *
 261 mptcp_subflow_rsk(const struct request_sock *rsk)
 262 {
 263         return (struct mptcp_subflow_request_sock *)rsk;
 264 }
 265
 266 /* MPTCP subflow context */
 267 struct mptcp_subflow_context {
 268         struct  list_head node;/* conn_list of subflows */
 269         u64     local_key;
 270         u64     remote_key;
 271         u64     idsn;
 272         u64     map_seq;
 273         u32     snd_isn;
 274         u32     token;
 275         u32     rel_write_seq;
 276         u32     map_subflow_seq;
 277         u32     ssn_offset;
 278         u32     map_data_len;
 279         u32     request_mptcp : 1,  /* send MP_CAPABLE */
 280                 request_join : 1,   /* send MP_JOIN */
 281                 request_bkup : 1,
 282                 mp_capable : 1,     /* remote is MPTCP capable */
 283                 mp_join : 1,        /* remote is JOINing */
 284                 fully_established : 1,      /* path validated */
 285                 pm_notified : 1,    /* PM hook called for established status */
 286                 conn_finished : 1,
 287                 map_valid : 1,
 288                 mpc_map : 1,
 289                 backup : 1,
 290                 data_avail : 1,
 291                 rx_eof : 1,
 292                 data_fin_tx_enable : 1,
 293                 can_ack : 1;        /* only after processing the remote a key */
 294         u64     data_fin_tx_seq;
 295         u32     remote_nonce;
 296         u64     thmac;
 297         u32     local_nonce;
 298         u32     remote_token;
 299         u8      hmac[MPTCPOPT_HMAC_LEN];
 300         u8      local_id;
 301         u8      remote_id;
 302
 303         struct  sock *tcp_sock;     /* tcp sk backpointer */
 304         struct  sock *conn;         /* parent mptcp_sock */
 305         const   struct inet_connection_sock_af_ops *icsk_af_ops;
 306         void    (*tcp_data_ready)(struct sock *sk);
 307         void    (*tcp_state_change)(struct sock *sk);
 308         void    (*tcp_write_space)(struct sock *sk);
 309
 310         struct  rcu_head rcu;
 311 };
 312
 313 static inline struct mptcp_subflow_context *
 314 mptcp_subflow_ctx(const struct sock *sk)
 315 {
 316         struct inet_connection_sock *icsk = inet_csk(sk);
 317
 318         /* Use RCU on icsk_ulp_data only for sock diag code */
 319         return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data;
 320 }
 321
 322 static inline struct sock *
 323 mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow)
 324 {
 325         return subflow->tcp_sock;
 326 }
 327
 328 static inline u64
 329 mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow)
 330 {
 331         return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq -
 332                       subflow->ssn_offset -
 333                       subflow->map_subflow_seq;
 334 }
 335
 336 static inline u64
 337 mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow)
 338 {
 339         return subflow->map_seq + mptcp_subflow_get_map_offset(subflow);
 340 }
 341
 342 int mptcp_is_enabled(struct net *net);
 343 bool mptcp_subflow_data_available(struct sock *sk);
 344 void mptcp_subflow_init(void);
 345
 346 /* called with sk socket lock held */
 347 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
 348                             const struct mptcp_addr_info *loc,
 349                             const struct mptcp_addr_info *remote);
 350 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock);
 351
 352 static inline void mptcp_subflow_tcp_fallback(struct sock *sk,
 353                                               struct mptcp_subflow_context *ctx)
 354 {
 355         sk->sk_data_ready = ctx->tcp_data_ready;
 356         sk->sk_state_change = ctx->tcp_state_change;
 357         sk->sk_write_space = ctx->tcp_write_space;
 358
 359         inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops;
 360 }
 361
 362 extern const struct inet_connection_sock_af_ops ipv4_specific;
 363 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 364 extern const struct inet_connection_sock_af_ops ipv6_specific;
 365 #endif
 366
 367 void mptcp_proto_init(void);
 368 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 369 int mptcp_proto_v6_init(void);
 370 #endif
 371
 372 struct sock *mptcp_sk_clone(const struct sock *sk,
 373                             const struct mptcp_options_received *mp_opt,
 374                             struct request_sock *req);
 375 void mptcp_get_options(const struct sk_buff *skb,
 376                        struct mptcp_options_received *mp_opt);
 377
 378 void mptcp_finish_connect(struct sock *sk);
 379 void mptcp_data_ready(struct sock *sk, struct sock *ssk);
 380 bool mptcp_finish_join(struct sock *sk);
 381 void mptcp_data_acked(struct sock *sk);
 382 void mptcp_subflow_eof(struct sock *sk);
 383
 384 int mptcp_token_new_request(struct request_sock *req);
 385 void mptcp_token_destroy_request(u32 token);
 386 int mptcp_token_new_connect(struct sock *sk);
 387 int mptcp_token_new_accept(u32 token, struct sock *conn);
 388 struct mptcp_sock *mptcp_token_get_sock(u32 token);
 389 void mptcp_token_destroy(u32 token);
 390
 391 void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn);
 392 static inline void mptcp_crypto_key_gen_sha(u64 *key, u32 *token, u64 *idsn)
 393 {
 394         /* we might consider a faster version that computes the key as a
 395          * hash of some information available in the MPTCP socket. Use
 396          * random data at the moment, as it's probably the safest option
 397          * in case multiple sockets are opened in different namespaces at
 398          * the same time.
 399          */
 400         get_random_bytes(key, sizeof(u64));
 401         mptcp_crypto_key_sha(*key, token, idsn);
 402 }
 403
 404 void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac);
 405
 406 void mptcp_pm_init(void);
 407 void mptcp_pm_data_init(struct mptcp_sock *msk);
 408 void mptcp_pm_close(struct mptcp_sock *msk);
 409 void mptcp_pm_new_connection(struct mptcp_sock *msk, int server_side);
 410 void mptcp_pm_fully_established(struct mptcp_sock *msk);
 411 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk);
 412 void mptcp_pm_connection_closed(struct mptcp_sock *msk);
 413 void mptcp_pm_subflow_established(struct mptcp_sock *msk,
 414                                   struct mptcp_subflow_context *subflow);
 415 void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id);
 416 void mptcp_pm_add_addr_received(struct mptcp_sock *msk,
 417                                 const struct mptcp_addr_info *addr);
 418
 419 int mptcp_pm_announce_addr(struct mptcp_sock *msk,
 420                            const struct mptcp_addr_info *addr);
 421 int mptcp_pm_remove_addr(struct mptcp_sock *msk, u8 local_id);
 422 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, u8 remote_id);
 423
 424 static inline bool mptcp_pm_should_signal(struct mptcp_sock *msk)
 425 {
 426         return READ_ONCE(msk->pm.addr_signal);
 427 }
 428
 429 static inline unsigned int mptcp_add_addr_len(int family)
 430 {
 431         if (family == AF_INET)
 432                 return TCPOLEN_MPTCP_ADD_ADDR;
 433         return TCPOLEN_MPTCP_ADD_ADDR6;
 434 }
 435
 436 bool mptcp_pm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
 437                           struct mptcp_addr_info *saddr);
 438 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
 439
 440 void mptcp_pm_nl_init(void);
 441 void mptcp_pm_nl_data_init(struct mptcp_sock *msk);
 442 void mptcp_pm_nl_fully_established(struct mptcp_sock *msk);
 443 void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk);
 444 void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk);
 445 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc);
 446
 447 static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb)
 448 {
 449         return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
 450 }
 451
 452 static inline bool before64(__u64 seq1, __u64 seq2)
 453 {
 454         return (__s64)(seq1 - seq2) < 0;
 455 }
 456
 457 #define after64(seq2, seq1)     before64(seq1, seq2)
 458
 459 void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops);
 460
 461 #endif /* __MPTCP_PROTOCOL_H */