bsd/netinet/mptcp_var.h

   1 /*
   2  * Copyright (c) 2012-2015 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28
  29 #ifndef _NETINET_MPTCP_VAR_H_
  30 #define _NETINET_MPTCP_VAR_H_
  31
  32 #ifdef PRIVATE
  33 #include <netinet/in.h>
  34 #include <netinet/tcp.h>
  35 #endif
  36
  37 #ifdef BSD_KERNEL_PRIVATE
  38 #include <sys/queue.h>
  39 #include <sys/protosw.h>
  40 #include <kern/locks.h>
  41 #include <mach/boolean.h>
  42 #include <netinet/mp_pcb.h>
  43
  44 /*
  45  * MPTCP Session
  46  *
  47  * This is an extension to the multipath PCB specific for MPTCP, protected by
  48  * the per-PCB mpp_lock (also the socket's lock); MPTCP thread signalling uses
  49  * its own mpte_thread_lock due to lock ordering constraints.
  50  */
  51 struct mptses {
  52         struct mppcb    *mpte_mppcb;            /* back ptr to multipath PCB */
  53         struct mptcb    *mpte_mptcb;            /* ptr to MPTCP PCB */
  54         TAILQ_HEAD(, mptopt) mpte_sopts;        /* list of socket options */
  55         TAILQ_HEAD(, mptsub) mpte_subflows;     /* list of subflows */
  56         uint16_t        mpte_numflows;          /* # of subflows in list */
  57         uint16_t        mpte_nummpcapflows;     /* # of MP_CAP subflows */
  58         sae_associd_t   mpte_associd;           /* MPTCP association ID */
  59         sae_connid_t    mpte_connid_last;       /* last used connection ID */
  60         /*
  61          * Threading (protected by mpte_thread_lock)
  62          */
  63         decl_lck_mtx_data(, mpte_thread_lock);  /* thread lock */
  64         struct thread   *mpte_thread;           /* worker thread */
  65         uint32_t        mpte_thread_active;     /* thread is running */
  66         uint32_t        mpte_thread_reqs;       /* # of requests for thread */
  67         struct mptsub   *mpte_active_sub;       /* ptr to last active subf */
  68         uint8_t mpte_flags;                     /* per mptcp session flags */
  69         uint8_t mpte_lost_aid;                  /* storing lost address id */
  70         uint8_t mpte_addrid_last;               /* storing address id parm */
  71 };
  72
  73 /*
  74  * Valid values for mpte_flags.
  75  */
  76 #define MPTE_SND_REM_ADDR       0x01            /* Send Remove_addr option */
  77
  78 #define mptompte(mp)    ((struct mptses *)(mp)->mpp_pcbe)
  79
  80 #define MPTE_LOCK_ASSERT_HELD(_mpte)                                    \
  81         lck_mtx_assert(&(_mpte)->mpte_mppcb->mpp_lock, LCK_MTX_ASSERT_OWNED)
  82
  83 #define MPTE_LOCK_ASSERT_NOTHELD(_mpte)                                 \
  84         lck_mtx_assert(&(_mpte)->mpte_mppcb->mpp_lock, LCK_MTX_ASSERT_NOTOWNED)
  85
  86 #define MPTE_LOCK(_mpte)                                                \
  87         lck_mtx_lock(&(_mpte)->mpte_mppcb->mpp_lock)
  88
  89 #define MPTE_LOCK_SPIN(_mpte)                                           \
  90         lck_mtx_lock_spin(&(_mpte)->mpte_mppcb->mpp_lock)
  91
  92 #define MPTE_CONVERT_LOCK(_mpte) do {                                   \
  93         MPTE_LOCK_ASSERT_HELD(_mpte);                                   \
  94         lck_mtx_convert_spin(&(_mpte)->mpte_mppcb->mpp_lock);           \
  95 } while (0)
  96
  97 #define MPTE_UNLOCK(_mpte)                                              \
  98         lck_mtx_unlock(&(_mpte)->mpte_mppcb->mpp_lock)
  99
 100 /*
 101  * MPTCP socket options
 102  */
 103 struct mptopt {
 104         TAILQ_ENTRY(mptopt)     mpo_entry;      /* glue to other options */
 105         uint32_t                mpo_flags;      /* see flags below */
 106         int                     mpo_level;      /* sopt_level */
 107         int                     mpo_name;       /* sopt_name */
 108         int                     mpo_intval;     /* sopt_val */
 109 };
 110
 111 #define MPOF_ATTACHED           0x1     /* attached to MP socket */
 112 #define MPOF_SUBFLOW_OK         0x2     /* can be issued on subflow socket */
 113 #define MPOF_INTERIM            0x4     /* has not been issued on any subflow */
 114
 115 /*
 116  * Structure passed down to TCP during subflow connection establishment
 117  * containing information pertaining to the MPTCP.
 118  */
 119 struct mptsub_connreq {
 120         uint32_t        mpcr_type;      /* see MPTSUB_CONNREQ_* below */
 121         uint32_t        mpcr_ifscope;   /* ifscope parameter to connectx(2) */
 122         struct proc     *mpcr_proc;     /* process issuing connectx(2) */
 123 };
 124
 125 /* valid values for mpcr_type */
 126 #define MPTSUB_CONNREQ_MP_ENABLE        1       /* enable MPTCP */
 127 #define MPTSUB_CONNREQ_MP_ADD           2       /* join an existing MPTCP */
 128
 129 /*
 130  * MPTCP subflow
 131  *
 132  * Protected by the the per-subflow mpts_lock.  Note that mpts_flags
 133  * and mpts_evctl are modified via atomic operations.
 134  */
 135 struct mptsub {
 136         decl_lck_mtx_data(, mpts_lock);         /* per-subflow lock */
 137         TAILQ_ENTRY(mptsub)     mpts_entry;     /* glue to peer subflows */
 138         uint32_t                mpts_refcnt;    /* reference count */
 139         uint32_t                mpts_flags;     /* see flags below */
 140         uint32_t                mpts_evctl;     /* subflow control events */
 141         uint32_t                mpts_family;    /* address family */
 142         sae_connid_t            mpts_connid;    /* subflow connection ID */
 143         int                     mpts_oldintval; /* sopt_val before sosetopt  */
 144         uint32_t                mpts_rank;      /* subflow priority/rank */
 145         int32_t                 mpts_soerror;   /* most recent subflow error */
 146         struct mptses           *mpts_mpte;     /* back ptr to MPTCP session */
 147         struct socket           *mpts_socket;   /* subflow socket */
 148         struct sockaddr         *mpts_src;      /* source address */
 149         struct sockaddr         *mpts_dst;      /* destination address */
 150         struct ifnet            *mpts_outif;    /* outbound interface */
 151         u_int64_t               mpts_sndnxt;    /* next byte to send in mp so */
 152         u_int32_t               mpts_rel_seq;   /* running count of subflow # */
 153         struct protosw          *mpts_oprotosw; /* original protosw */
 154         struct mptsub_connreq   mpts_mpcr;      /* connection request */
 155         int32_t                 mpts_srtt;      /* tcp's rtt estimate */
 156         int32_t                 mpts_rxtcur;    /* tcp's rto estimate */
 157         uint32_t                mpts_probesoon; /* send probe after probeto */
 158         uint32_t                mpts_probecnt;  /* number of probes sent */
 159         uint32_t                mpts_maxseg;    /* cached value of t_maxseg */
 160         uint32_t                mpts_peerswitch;/* no of uses of backup so */
 161 #define MPTSL_WIRED             0x01
 162 #define MPTSL_WIFI              0x02
 163 #define MPTSL_CELL              0x04
 164         uint32_t                mpts_linktype;  /* wired, wifi, cell */
 165 };
 166
 167 /*
 168  * Valid values for mpts_flags.  In particular:
 169  *
 170  *    - MP_CAPABLE means that the connection is successfully established as
 171  *      MPTCP and data transfer may occur, but is not yet ready for multipath-
 172  *      related semantics until MP_READY.  I.e. if this is on the first subflow,
 173  *      it causes the MPTCP socket to transition to a connected state, except
 174  *      that additional subflows will not be established; they will be marked
 175  *      with PENDING and will be processed when the first subflow is marked
 176  *      with MP_READY.
 177  *
 178  *    - MP_READY implies that an MP_CAPABLE connection has been confirmed as
 179  *      an MPTCP connection.  See notes above.
 180  *
 181  *    - MP_DEGRADED implies that the connection has lost its MPTCP capabilities
 182  *      but data transfer on the MPTCP socket is unaffected.  Any existing
 183  *      PENDING subflows will be disconnected, and further attempts to connect
 184  *      additional subflows will be rejected.
 185  *
 186  * Note that these are per-subflow flags.  The setting and clearing of MP_READY
 187  * reflects the state of the MPTCP connection with regards to its multipath
 188  * semantics, via the MPTCPF_JOIN_READY flag.  Until that flag is set (meaning
 189  * until at least a subflow is marked with MP_READY), further connectx(2)
 190  * attempts to join will be queued.  When the flag is cleared (after it has
 191  * been set), further connectx(2) will fail (and existing queued ones will be
 192  * aborted) and the MPTCP connection loses all of its multipath semantics.
 193  *
 194  * Keep in sync with bsd/dev/dtrace/scripts/mptcp.d.
 195  */
 196 #define MPTSF_ATTACHED          0x1     /* attached to MPTCP PCB */
 197 #define MPTSF_CONNECTING        0x2     /* connection was attempted */
 198 #define MPTSF_CONNECT_PENDING   0x4     /* will connect when MPTCP is ready */
 199 #define MPTSF_CONNECTED         0x8     /* connection is established */
 200 #define MPTSF_DISCONNECTING     0x10    /* disconnection was attempted */
 201 #define MPTSF_DISCONNECTED      0x20    /* has been disconnected */
 202 #define MPTSF_MP_CAPABLE        0x40    /* connected as a MPTCP subflow */
 203 #define MPTSF_MP_READY          0x80    /* MPTCP has been confirmed */
 204 #define MPTSF_MP_DEGRADED       0x100   /* has lost its MPTCP capabilities */
 205 #define MPTSF_SUSPENDED         0x200   /* write-side is flow controlled */
 206 #define MPTSF_BOUND_IF          0x400   /* subflow bound to an interface */
 207 #define MPTSF_BOUND_IP          0x800   /* subflow bound to a src address */
 208 #define MPTSF_BOUND_PORT        0x1000  /* subflow bound to a src port */
 209 #define MPTSF_PREFERRED         0x2000  /* primary/preferred subflow */
 210 #define MPTSF_SOPT_OLDVAL       0x4000  /* old option value is valid */
 211 #define MPTSF_SOPT_INPROG       0x8000  /* sosetopt in progress */
 212 #define MPTSF_DELETEOK          0x10000 /* subflow can be deleted */
 213 #define MPTSF_FAILINGOVER       0x20000 /* subflow not used for output */
 214 #define MPTSF_ACTIVE            0x40000 /* subflow currently in use */
 215 #define MPTSF_MPCAP_CTRSET      0x80000 /* mpcap counter */
 216 #define MPTSF_FASTJ_SEND        0x100000 /* send data after SYN in MP_JOIN */
 217 #define MPTSF_FASTJ_REQD        0x200000 /* fastjoin required */
 218 #define MPTSF_USER_DISCONNECT   0x400000 /* User triggered disconnect */
 219 #define MPTSF_TFO_REQD          0x800000 /* TFO requested */
 220
 221 #define MPTSF_BITS \
 222         "\020\1ATTACHED\2CONNECTING\3PENDING\4CONNECTED\5DISCONNECTING" \
 223         "\6DISCONNECTED\7MP_CAPABLE\10MP_READY\11MP_DEGRADED\12SUSPENDED" \
 224         "\13BOUND_IF\14BOUND_IP\15BOUND_PORT\16PREFERRED\17SOPT_OLDVAL" \
 225         "\20SOPT_INPROG\21NOLINGER\22FAILINGOVER\23ACTIVE\24MPCAP_CTRSET" \
 226         "\25FASTJ_SEND\26FASTJ_REQD\27USER_DISCONNECT"
 227
 228 #define MPTS_LOCK_ASSERT_HELD(_mpts)                                    \
 229         lck_mtx_assert(&(_mpts)->mpts_lock, LCK_MTX_ASSERT_OWNED)
 230
 231 #define MPTS_LOCK_ASSERT_NOTHELD(_mpts)                                 \
 232         lck_mtx_assert(&(_mpts)->mpts_lock, LCK_MTX_ASSERT_NOTOWNED)
 233
 234 #define MPTS_LOCK(_mpts)                                                \
 235         lck_mtx_lock(&(_mpts)->mpts_lock)
 236
 237 #define MPTS_UNLOCK(_mpts)                                              \
 238         lck_mtx_unlock(&(_mpts)->mpts_lock)
 239
 240 #define MPTS_ADDREF(_mpts)                                              \
 241         mptcp_subflow_addref(_mpts, 0)
 242
 243 #define MPTS_ADDREF_LOCKED(_mpts)                                       \
 244         mptcp_subflow_addref(_mpts, 1)
 245
 246 #define MPTS_REMREF(_mpts)                                              \
 247         mptcp_subflow_remref(_mpts)
 248
 249 /*
 250  * MPTCP states
 251  * Keep in sync with bsd/dev/dtrace/mptcp.d
 252  */
 253 typedef enum mptcp_state {
 254         MPTCPS_CLOSED           = 0,    /* closed */
 255         MPTCPS_LISTEN           = 1,    /* not yet implemented */
 256         MPTCPS_ESTABLISHED      = 2,    /* MPTCP connection established */
 257         MPTCPS_CLOSE_WAIT       = 3,    /* rcvd DFIN, waiting for close */
 258         MPTCPS_FIN_WAIT_1       = 4,    /* have closed, sent DFIN */
 259         MPTCPS_CLOSING          = 5,    /* closed xchd DFIN, waiting DFIN ACK */
 260         MPTCPS_LAST_ACK         = 6,    /* had DFIN and close; await DFIN ACK */
 261         MPTCPS_FIN_WAIT_2       = 7,    /* have closed, DFIN is acked */
 262         MPTCPS_TIME_WAIT        = 8,    /* in 2*MSL quiet wait after close */
 263         MPTCPS_TERMINATE        = 9,    /* terminal state */
 264 } mptcp_state_t;
 265
 266 typedef u_int64_t       mptcp_key_t;
 267 typedef u_int32_t       mptcp_token_t;
 268 typedef u_int8_t        mptcp_addr_id;
 269
 270
 271 /* Address ID list */
 272 struct mptcp_subf_auth_entry {
 273         LIST_ENTRY(mptcp_subf_auth_entry) msae_next;
 274         u_int32_t       msae_laddr_rand;        /* Local nonce */
 275         u_int32_t       msae_raddr_rand;        /* Remote nonce */
 276         mptcp_addr_id   msae_laddr_id;          /* Local addr ID */
 277         mptcp_addr_id   msae_raddr_id;          /* Remote addr ID */
 278 };
 279
 280 /*
 281  * MPTCP Protocol Control Block
 282  *
 283  * Protected by per-MPTCP mpt_lock.
 284  * Keep in sync with bsd/dev/dtrace/scripts/mptcp.d.
 285  */
 286 struct mptcb {
 287         decl_lck_mtx_data(, mpt_lock);          /* per MPTCP PCB lock */
 288         struct mptses   *mpt_mpte;              /* back ptr to MPTCP session */
 289         mptcp_state_t   mpt_state;              /* MPTCP state */
 290         u_int32_t       mpt_flags;              /* see flags below */
 291         u_int32_t       mpt_refcnt;             /* references held on mptcb */
 292         u_int32_t       mpt_version;            /* MPTCP proto version */
 293         int             mpt_softerror;          /* error not yet reported */
 294         /*
 295          * Authentication and metadata invariants
 296          */
 297         mptcp_key_t     *mpt_localkey;          /* in network byte order */
 298         mptcp_key_t     mpt_remotekey;          /* in network byte order */
 299         mptcp_token_t   mpt_localtoken;         /* HMAC SHA1 of local key */
 300         mptcp_token_t   mpt_remotetoken;        /* HMAC SHA1 of remote key */
 301
 302         /*
 303          * Timer vars for scenarios where subflow level acks arrive, but
 304          * Data ACKs do not.
 305          */
 306         int             mpt_rxtshift;           /* num of consecutive retrans */
 307         u_int32_t       mpt_rxtstart;           /* time at which rxt started */
 308         u_int64_t       mpt_rtseq;              /* seq # being tracked */
 309         u_int32_t       mpt_timer_vals;         /* timer related values */
 310         u_int32_t       mpt_timewait;           /* timewait */
 311         /*
 312          * Sending side
 313          */
 314         u_int64_t       mpt_snduna;             /* DSN of last unacked byte */
 315         u_int64_t       mpt_sndnxt;             /* DSN of next byte to send */
 316         u_int64_t       mpt_sndmax;             /* DSN of max byte sent */
 317         u_int64_t       mpt_local_idsn;         /* First byte's DSN */
 318         u_int32_t       mpt_sndwnd;
 319         /*
 320          * Receiving side
 321          */
 322         u_int64_t       mpt_rcvnxt;             /* Next expected DSN */
 323         u_int64_t       mpt_rcvatmark;          /* mpsocket marker of rcvnxt */
 324         u_int64_t       mpt_remote_idsn;        /* Peer's IDSN */
 325         u_int32_t       mpt_rcvwnd;
 326         LIST_HEAD(, mptcp_subf_auth_entry) mpt_subauth_list; /* address IDs */
 327         /*
 328          * Fastclose
 329          */
 330         u_int64_t       mpt_dsn_at_csum_fail;   /* MPFail Opt DSN */
 331         u_int32_t       mpt_ssn_at_csum_fail;   /* MPFail Subflow Seq */
 332         /*
 333          * Zombie handling
 334          */
 335 #define MPT_GC_TICKS            (30)
 336 #define MPT_GC_TICKS_FAST       (10)
 337         int32_t         mpt_gc_ticks;           /* Used for zombie deletion */
 338
 339         u_int32_t       mpt_notsent_lowat;      /* TCP_NOTSENT_LOWAT support */
 340         u_int32_t       mpt_peer_version;       /* Version from peer */
 341 };
 342
 343 /* valid values for mpt_flags (see also notes on mpts_flags above) */
 344 #define MPTCPF_CHECKSUM         0x1     /* checksum DSS option */
 345 #define MPTCPF_FALLBACK_TO_TCP  0x2     /* Fallback to TCP */
 346 #define MPTCPF_JOIN_READY       0x4     /* Ready to start 2 or more subflows */
 347 #define MPTCPF_RECVD_MPFAIL     0x8     /* Received MP_FAIL option */
 348 #define MPTCPF_PEEL_OFF         0x10    /* Peel off this socket */
 349 #define MPTCPF_SND_64BITDSN     0x20    /* Send full 64-bit DSN */
 350 #define MPTCPF_SND_64BITACK     0x40    /* Send 64-bit ACK response */
 351 #define MPTCPF_RCVD_64BITACK    0x80    /* Received 64-bit Data ACK */
 352 #define MPTCPF_POST_FALLBACK_SYNC       0x100   /* Post fallback resend data */
 353 #define MPTCPF_FALLBACK_HEURISTIC       0x200   /* Send SYN without MP_CAPABLE due to heuristic */
 354 #define MPTCPF_HEURISTIC_TRAC           0x400   /* Tracked this connection in the heuristics as a failure */
 355
 356 #define MPTCPF_BITS \
 357         "\020\1CHECKSUM\2FALLBACK_TO_TCP\3JOIN_READY\4RECVD_MPFAIL\5PEEL_OFF" \
 358         "\6SND_64BITDSN\7SND_64BITACK\10RCVD_64BITACK\11POST_FALLBACK_SYNC" \
 359         "\12FALLBACK_HEURISTIC\13HEURISTIC_TRAC"
 360
 361 /* valid values for mpt_timer_vals */
 362 #define MPTT_REXMT              0x01    /* Starting Retransmit Timer */
 363 #define MPTT_TW                 0x02    /* Starting Timewait Timer */
 364 #define MPTT_FASTCLOSE          0x04    /* Starting Fastclose wait timer */
 365 //#define MPTT_PROBE_TIMER      0x08    /* Timer for probing preferred path */
 366
 367 #define MPT_LOCK_ASSERT_HELD(_mpt)                                      \
 368         lck_mtx_assert(&(_mpt)->mpt_lock, LCK_MTX_ASSERT_OWNED)
 369
 370 #define MPT_LOCK_ASSERT_NOTHELD(_mpt)                                   \
 371         lck_mtx_assert(&(_mpt)->mpt_lock, LCK_MTX_ASSERT_NOTOWNED)
 372
 373 #define MPT_LOCK(_mpt)                                                  \
 374         lck_mtx_lock(&(_mpt)->mpt_lock)
 375
 376 #define MPT_LOCK_SPIN(_mpt)                                             \
 377         lck_mtx_lock_spin(&(_mpt)->mpt_lock)
 378
 379 #define MPT_CONVERT_LOCK(_mpt) do {                                     \
 380         MPT_LOCK_ASSERT_HELD(_mpt);                                     \
 381         lck_mtx_convert_spin(&(_mpt)->mpt_lock);                        \
 382 } while (0)
 383
 384 #define MPT_UNLOCK(_mpt)                                                \
 385         lck_mtx_unlock(&(_mpt)->mpt_lock)
 386
 387 /* events for close FSM */
 388 #define MPCE_CLOSE              0x1
 389 #define MPCE_RECV_DATA_ACK      0x2
 390 #define MPCE_RECV_DATA_FIN      0x4
 391
 392 /* mptcb manipulation */
 393 #define tptomptp(tp)    ((struct mptcb *)((tp)->t_mptcb))
 394
 395 /*
 396  * MPTCP control block and state structures are allocated along with
 397  * the MP protocol control block; the folllowing represents the layout.
 398  */
 399 struct mpp_mtp {
 400         struct mppcb            mpp;            /* Multipath PCB */
 401         struct mptses           mpp_ses;        /* MPTCP session */
 402         struct mptcb            mtcb;           /* MPTCP PCB */
 403 };
 404
 405 #ifdef SYSCTL_DECL
 406 SYSCTL_DECL(_net_inet_mptcp);
 407 #endif /* SYSCTL_DECL */
 408
 409 extern struct mppcbinfo mtcbinfo;
 410 extern struct pr_usrreqs mptcp_usrreqs;
 411
 412 /* Encryption algorithm related definitions */
 413 #define MPTCP_SHA1_RESULTLEN    20
 414 #define SHA1_TRUNCATED          8
 415
 416 /* List of valid keys to use for MPTCP connections */
 417 #define MPTCP_KEY_DIGEST_LEN            (MPTCP_SHA1_RESULTLEN)
 418 #define MPTCP_MX_KEY_ALLOCS             (256)
 419 #define MPTCP_KEY_PREALLOCS_MX          (16)
 420 #define MPTCP_MX_PREALLOC_ZONE_SZ       (8192)
 421
 422 struct mptcp_key_entry {
 423         LIST_ENTRY(mptcp_key_entry)     mkey_next;
 424         mptcp_key_t                     mkey_value;
 425 #define MKEYF_FREE      0x0
 426 #define MKEYF_INUSE     0x1
 427         u_int32_t                       mkey_flags;
 428         char                            mkey_digest[MPTCP_KEY_DIGEST_LEN];
 429 };
 430
 431 /* structure for managing unique key list */
 432 struct mptcp_keys_pool_head {
 433         struct mptcp_key_entry *lh_first;       /* list of keys */
 434         u_int32_t       mkph_count;             /* total keys in pool */
 435         vm_size_t       mkph_key_elm_sz;        /* size of key entry */
 436         struct zone     *mkph_key_entry_zone;   /* zone for key entry */
 437         decl_lck_mtx_data(, mkph_lock);         /* lock for key list */
 438 };
 439
 440 /* MPTCP Receive Window */
 441 #define MPTCP_RWIN_MAX  (1<<16)
 442
 443 /* MPTCP Debugging Levels */
 444 #define MPTCP_LOGLVL_NONE       0x0     /* No debug logging */
 445 #define MPTCP_LOGLVL_ERR        0x1     /* Errors in execution are logged */
 446 #define MPTCP_LOGLVL_LOG        0x2     /* Important logs */
 447 #define MPTCP_LOGLVL_VERBOSE    0x3     /* Verbose logs */
 448
 449 /* MPTCP sub-components for debug logging */
 450 #define MPTCP_NO_DBG            0x00    /* No areas are logged */
 451 #define MPTCP_STATE_DBG         0x01    /* State machine logging */
 452 #define MPTCP_SOCKET_DBG        0x02    /* Socket call logging */
 453 #define MPTCP_SENDER_DBG        0x04    /* Sender side logging */
 454 #define MPTCP_RECEIVER_DBG      0x08    /* Receiver logging */
 455 #define MPTCP_EVENTS_DBG        0x10    /* Subflow events logging */
 456 #define MPTCP_ALL_DBG           (MPTCP_STATE_DBG | MPTCP_SOCKET_DBG | \
 457     MPTCP_SENDER_DBG | MPTCP_RECEIVER_DBG | MPTCP_EVENTS_DBG)
 458
 459 /* Mask to obtain 32-bit portion of data sequence number */
 460 #define MPTCP_DATASEQ_LOW32_MASK        (0xffffffff)
 461 #define MPTCP_DATASEQ_LOW32(seq)        (seq & MPTCP_DATASEQ_LOW32_MASK)
 462
 463 /* Mask to obtain upper 32-bit portion of data sequence number */
 464 #define MPTCP_DATASEQ_HIGH32_MASK       (0xffffffff00000000)
 465 #define MPTCP_DATASEQ_HIGH32(seq)       (seq & MPTCP_DATASEQ_HIGH32_MASK)
 466
 467 /* Mask to obtain 32-bit portion of data ack */
 468 #define MPTCP_DATAACK_LOW32_MASK        (0xffffffff)
 469 #define MPTCP_DATAACK_LOW32(ack)        (ack & MPTCP_DATAACK_LOW32_MASK)
 470
 471 /* Mask to obtain upper 32-bit portion of data ack */
 472 #define MPTCP_DATAACK_HIGH32_MASK       (0xffffffff00000000)
 473 #define MPTCP_DATAACK_HIGH32(ack)       (ack & MPTCP_DATAACK_HIGH32_MASK)
 474
 475 /*
 476  * x is the 64-bit data sequence number, y the 32-bit data seq number to be
 477  * extended. z is y extended to the appropriate 64-bit value.
 478  * This algorithm is based on the fact that subflow level window sizes are
 479  * at the maximum 2**30 (in reality, they are a lot lesser). A high throughput
 480  * application sending on a large number of subflows can in theory have very
 481  * large MPTCP level send and receive windows. In which case, 64 bit DSNs
 482  * must be sent in place of 32 bit DSNs on wire. For us, with 2 subflows at
 483  * 512K each, sequence wraparound detection can be done by checking whether
 484  * the 32-bit value obtained on wire is 2**31 bytes apart from the stored
 485  * lower 32-bits of the Data Sequence Number. Bogus DSNs are dropped by
 486  * comparing against rwnd. Bogus DSNs within rwnd cannot be protected against
 487  * and are as weak as bogus TCP sequence numbers.
 488  */
 489 #define MPTCP_EXTEND_DSN(x, y, z) {                                     \
 490         if ((MPTCP_DATASEQ_LOW32(x) > y) &&                             \
 491             ((((u_int32_t)MPTCP_DATASEQ_LOW32(x)) - (u_int32_t)y) >=    \
 492             (u_int32_t)(1 << 31))) {                                    \
 493                 /*                                                      \
 494                  * y wrapped around and x and y are 2**31 bytes  apart  \
 495                  */                                                     \
 496                 z = MPTCP_DATASEQ_HIGH32(x) + 0x100000000;              \
 497                 z |= y;                                                 \
 498         } else if ((MPTCP_DATASEQ_LOW32(x) < y) &&                      \
 499             (((u_int32_t)y -                                            \
 500             ((u_int32_t)MPTCP_DATASEQ_LOW32(x))) >=                     \
 501             (u_int32_t)(1 << 31))) {                                    \
 502                 /*                                                      \
 503                  * x wrapped around and x and y are 2**31 apart         \
 504                  */                                                     \
 505                 z = MPTCP_DATASEQ_HIGH32(x) - 0x100000000;              \
 506                 z |= y;                                                 \
 507         } else {                                                        \
 508                 z = MPTCP_DATASEQ_HIGH32(x) | y;                        \
 509         }                                                               \
 510 }
 511
 512 #define mptcplog(x, y, z)       do {                                    \
 513         if ((mptcp_dbg_area & y) &&                                     \
 514             (mptcp_dbg_level >= z))                                     \
 515                 log x;                                                  \
 516 } while (0)
 517
 518 extern int mptcp_enable;        /* Multipath TCP */
 519 extern int mptcp_mpcap_retries; /* Multipath TCP retries */
 520 extern int mptcp_join_retries;  /* Multipath TCP Join retries */
 521 extern int mptcp_dss_csum;      /* Multipath DSS Option checksum */
 522 extern int mptcp_fail_thresh;   /* Multipath failover thresh of retransmits */
 523 extern int mptcp_subflow_keeptime; /* Multipath subflow TCP_KEEPALIVE opt */
 524 extern int mptcp_mpprio_enable; /* MP_PRIO option enable/disable */
 525 extern int mptcp_remaddr_enable;/* REMOVE_ADDR option enable/disable */
 526 extern int mptcp_fastjoin;      /* Enable FastJoin */
 527 extern int mptcp_zerortt_fastjoin; /* Enable Data after SYN Fast Join */
 528 extern int mptcp_rwnotify;      /* Enable RW notification on resume */
 529 extern uint32_t mptcp_dbg_level;        /* Multipath TCP debugging level */
 530 extern uint32_t mptcp_dbg_area; /* Multipath TCP debugging area */
 531
 532 #define MPPCB_LIMIT     32
 533 extern uint32_t mptcp_socket_limit; /* max number of mptcp sockets allowed */
 534 extern uint32_t mptcp_delayed_subf_start; /* delayed cellular subflow start */
 535 extern int tcp_jack_rxmt;       /* Join ACK retransmission value in msecs */
 536
 537 __BEGIN_DECLS
 538 extern void mptcp_init(struct protosw *, struct domain *);
 539 extern int mptcp_ctloutput(struct socket *, struct sockopt *);
 540 extern void *mptcp_sescreate(struct socket *, struct mppcb *);
 541 extern void mptcp_drain(void);
 542 extern struct mptses *mptcp_drop(struct mptses *, struct mptcb *, int);
 543 extern struct mptses *mptcp_close(struct mptses *, struct mptcb *);
 544 extern int mptcp_lock(struct socket *, int, void *);
 545 extern int mptcp_unlock(struct socket *, int, void *);
 546 extern lck_mtx_t *mptcp_getlock(struct socket *, int);
 547 extern void mptcp_thread_signal(struct mptses *);
 548 extern void mptcp_flush_sopts(struct mptses *);
 549 extern int mptcp_setconnorder(struct mptses *, sae_connid_t, uint32_t);
 550 extern int mptcp_getconnorder(struct mptses *, sae_connid_t, uint32_t *);
 551
 552 extern struct mptopt *mptcp_sopt_alloc(int);
 553 extern const char *mptcp_sopt2str(int, int, char *, int);
 554 extern void mptcp_sopt_free(struct mptopt *);
 555 extern void mptcp_sopt_insert(struct mptses *, struct mptopt *);
 556 extern void mptcp_sopt_remove(struct mptses *, struct mptopt *);
 557 extern struct mptopt *mptcp_sopt_find(struct mptses *, struct sockopt *);
 558
 559 extern struct mptsub *mptcp_subflow_alloc(int);
 560 extern void mptcp_subflow_free(struct mptsub *);
 561 extern void mptcp_subflow_addref(struct mptsub *, int);
 562 extern int mptcp_subflow_add(struct mptses *, struct mptsub *,
 563     struct proc *, uint32_t);
 564 extern void mptcp_subflow_del(struct mptses *, struct mptsub *, boolean_t);
 565 extern void mptcp_subflow_remref(struct mptsub *);
 566 extern int mptcp_subflow_output(struct mptses *, struct mptsub *);
 567 extern void mptcp_subflow_disconnect(struct mptses *, struct mptsub *,
 568     boolean_t);
 569 extern void mptcp_subflow_sopeeloff(struct mptses *, struct mptsub *,
 570     struct socket *);
 571 extern int mptcp_subflow_sosetopt(struct mptses *, struct socket *,
 572     struct mptopt *);
 573 extern int mptcp_subflow_sogetopt(struct mptses *, struct socket *,
 574     struct mptopt *);
 575
 576 extern void mptcp_input(struct mptses *, struct mbuf *);
 577 extern int mptcp_output(struct mptses *);
 578 extern void mptcp_close_fsm(struct mptcb *, uint32_t);
 579
 580 extern mptcp_token_t mptcp_get_localtoken(void *);
 581 extern mptcp_token_t mptcp_get_remotetoken(void *);
 582
 583 extern u_int64_t mptcp_get_localkey(void *);
 584 extern u_int64_t mptcp_get_remotekey(void *);
 585
 586 extern void mptcp_free_key(mptcp_key_t *key);
 587 extern void mptcp_hmac_sha1(mptcp_key_t, mptcp_key_t, u_int32_t, u_int32_t,
 588     u_char*, int);
 589 extern void mptcp_get_hmac(mptcp_addr_id, struct mptcb *, u_char *, int);
 590 extern void mptcp_get_rands(mptcp_addr_id, struct mptcb *, u_int32_t *,
 591     u_int32_t *);
 592 extern void mptcp_set_raddr_rand(mptcp_addr_id, struct mptcb *, mptcp_addr_id,
 593     u_int32_t);
 594 extern u_int64_t mptcp_get_trunced_hmac(mptcp_addr_id, struct mptcb *mp_tp);
 595 extern void mptcp_generate_token(char *, int, caddr_t, int);
 596 extern void mptcp_generate_idsn(char *, int, caddr_t, int);
 597 extern int mptcp_init_remote_parms(struct mptcb *);
 598 extern boolean_t mptcp_ok_to_keepalive(struct mptcb *);
 599 extern void mptcp_insert_dsn(struct mppcb *, struct mbuf *);
 600 extern void  mptcp_output_getm_dsnmap32(struct socket *, int, uint32_t,
 601     u_int32_t *, u_int32_t *, u_int16_t *, u_int64_t *);
 602 extern void  mptcp_output_getm_dsnmap64(struct socket *, int, uint32_t,
 603     u_int64_t *, u_int32_t *, u_int16_t *);
 604 extern void mptcp_send_dfin(struct socket *);
 605 extern void mptcp_act_on_txfail(struct socket *);
 606 extern struct mptsub *mptcp_get_subflow(struct mptses *, struct mptsub *,
 607     struct mptsub **);
 608 extern struct mptsub *mptcp_get_pending_subflow(struct mptses *,
 609     struct mptsub *);
 610 extern struct mptsub* mptcp_use_symptoms_hints(struct mptsub*,
 611     struct mptsub *);
 612 extern int mptcp_get_map_for_dsn(struct socket *, u_int64_t, u_int32_t *);
 613 extern int32_t mptcp_adj_sendlen(struct socket *so, int32_t off, int32_t len);
 614 extern int32_t mptcp_sbspace(struct mptcb *);
 615 extern void mptcp_notify_mpready(struct socket *);
 616 extern void mptcp_notify_mpfail(struct socket *);
 617 extern void mptcp_notify_close(struct socket *);
 618 extern boolean_t mptcp_no_rto_spike(struct socket*);
 619 extern int mptcp_set_notsent_lowat(struct mptses *mpte, int optval);
 620 extern u_int32_t mptcp_get_notsent_lowat(struct mptses *mpte);
 621 extern int mptcp_notsent_lowat_check(struct socket *so);
 622 extern void mptcp_control_register(void);
 623 extern int mptcp_is_wifi_unusable(void);
 624 extern int mptcp_is_cell_unusable(void);
 625 __END_DECLS
 626
 627 #endif /* BSD_KERNEL_PRIVATE */
 628 #ifdef PRIVATE
 629
 630 typedef struct mptcp_flow {
 631         size_t                  flow_len;
 632         size_t                  flow_tcpci_offset;
 633         uint32_t                flow_flags;
 634         sae_connid_t            flow_cid;
 635         struct sockaddr_storage flow_src;
 636         struct sockaddr_storage flow_dst;
 637         uint64_t                flow_sndnxt;    /* subflow's sndnxt snapshot */
 638         uint32_t                flow_relseq;    /* last subflow rel seq# */
 639         int32_t                 flow_soerror;   /* subflow level error */
 640         uint32_t                flow_probecnt;  /* number of probes sent */
 641         uint32_t                flow_peerswitch;/* did peer switch */
 642         conninfo_tcp_t          flow_ci;        /* must be the last field */
 643 } mptcp_flow_t;
 644
 645 typedef struct conninfo_mptcp {
 646         size_t          mptcpci_len;
 647         size_t          mptcpci_flow_offset;    /* offsetof first flow */
 648         size_t          mptcpci_nflows;         /* number of subflows */
 649         uint32_t        mptcpci_state;          /* MPTCP level state */
 650         uint32_t        mptcpci_mpte_flags;     /* Session flags */
 651         uint32_t        mptcpci_flags;          /* MPTCB flags */
 652         uint32_t        mptcpci_ltoken;         /* local token */
 653         uint32_t        mptcpci_rtoken;         /* remote token */
 654         uint32_t        mptcpci_notsent_lowat;  /* NOTSENT_LOWAT */
 655
 656         /* Send side */
 657         uint64_t        mptcpci_snduna;         /* DSN of last unacked byte */
 658         uint64_t        mptcpci_sndnxt;         /* DSN of next byte to send */
 659         uint64_t        mptcpci_sndmax;         /* DSN of max byte sent */
 660         uint64_t        mptcpci_lidsn;          /* Local IDSN */
 661         uint32_t        mptcpci_sndwnd;         /* Send window snapshot */
 662
 663         /* Receive side */
 664         uint64_t        mptcpci_rcvnxt;         /* Next expected DSN */
 665         uint64_t        mptcpci_rcvatmark;      /* Session level rcvnxt */
 666         uint64_t        mptcpci_ridsn;          /* Peer's IDSN */
 667         uint32_t        mptcpci_rcvwnd;         /* Receive window */
 668
 669         uint8_t         mptcpci_mpte_addrid;    /* last addr id */
 670
 671         mptcp_flow_t    mptcpci_flows[1];
 672 } conninfo_mptcp_t;
 673
 674 /* Use SymptomsD notifications of wifi and cell status in subflow selection */
 675 #define MPTCP_KERN_CTL_NAME    "com.apple.network.advisory"
 676 typedef struct symptoms_advisory {
 677         union {
 678                 uint32_t        sa_nwk_status_int;
 679                 struct {
 680                         union {
 681 #define SYMPTOMS_ADVISORY_NOCOMMENT    0x00
 682                                 uint16_t        sa_nwk_status;
 683                                 struct {
 684 #define SYMPTOMS_ADVISORY_WIFI_BAD     0x01
 685 #define SYMPTOMS_ADVISORY_WIFI_OK      0x02
 686                                         uint8_t sa_wifi_status;
 687 #define SYMPTOMS_ADVISORY_CELL_BAD     0x01
 688 #define SYMPTOMS_ADVISORY_CELL_OK      0x02
 689                                         uint8_t sa_cell_status;
 690                                 };
 691                         };
 692                         uint16_t        sa_unused;
 693                 };
 694         };
 695 } symptoms_advisory_t;
 696
 697
 698 #endif /* PRIVATE */
 699 #endif /* _NETINET_MPTCP_VAR_H_ */