bsd/netinet/mptcp_var.h

   1 /*
   2  * Copyright (c) 2012-2014 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         associd_t       mpte_associd;           /* MPTCP association ID */
  59         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         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_list    *mpts_src_sl;   /* source list */
 149         struct sockaddr_list    *mpts_dst_sl;   /* destination list */
 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 {
 154                 u_int64_t       mptsl_dsn;      /* Data Sequence Number */
 155                 u_int32_t       mptsl_sseq;     /* Corresponding Data Seq */
 156                 u_int32_t       mptsl_len;      /* length of mapping */
 157         } mpts_lastmap;
 158         struct protosw          *mpts_oprotosw; /* original protosw */
 159         struct mptsub_connreq   mpts_mpcr;      /* connection request */
 160 };
 161
 162 /*
 163  * Valid values for mpts_flags.  In particular:
 164  *
 165  *    - MP_CAPABLE means that the connection is successfully established as
 166  *      MPTCP and data transfer may occur, but is not yet ready for multipath-
 167  *      related semantics until MP_READY.  I.e. if this is on the first subflow,
 168  *      it causes the MPTCP socket to transition to a connected state, except
 169  *      that additional subflows will not be established; they will be marked
 170  *      with PENDING and will be processed when the first subflow is marked
 171  *      with MP_READY.
 172  *
 173  *    - MP_READY implies that an MP_CAPABLE connection has been confirmed as
 174  *      an MPTCP connection.  See notes above.
 175  *
 176  *    - MP_DEGRADED implies that the connection has lost its MPTCP capabilities
 177  *      but data transfer on the MPTCP socket is unaffected.  Any existing
 178  *      PENDING subflows will be disconnected, and further attempts to connect
 179  *      additional subflows will be rejected.
 180  *
 181  * Note that these are per-subflow flags.  The setting and clearing of MP_READY
 182  * reflects the state of the MPTCP connection with regards to its multipath
 183  * semantics, via the MPTCPF_JOIN_READY flag.  Until that flag is set (meaning
 184  * until at least a subflow is marked with MP_READY), further connectx(2)
 185  * attempts to join will be queued.  When the flag is cleared (after it has
 186  * been set), further connectx(2) will fail (and existing queued ones will be
 187  * aborted) and the MPTCP connection loses all of its multipath semantics.
 188  *
 189  * Keep in sync with bsd/dev/dtrace/scripts/mptcp.d.
 190  */
 191 #define MPTSF_ATTACHED          0x1     /* attached to MPTCP PCB */
 192 #define MPTSF_CONNECTING        0x2     /* connection was attempted */
 193 #define MPTSF_CONNECT_PENDING   0x4     /* will connect when MPTCP is ready */
 194 #define MPTSF_CONNECTED         0x8     /* connection is established */
 195 #define MPTSF_DISCONNECTING     0x10    /* disconnection was attempted */
 196 #define MPTSF_DISCONNECTED      0x20    /* has been disconnected */
 197 #define MPTSF_MP_CAPABLE        0x40    /* connected as a MPTCP subflow */
 198 #define MPTSF_MP_READY          0x80    /* MPTCP has been confirmed */
 199 #define MPTSF_MP_DEGRADED       0x100   /* has lost its MPTCP capabilities */
 200 #define MPTSF_SUSPENDED         0x200   /* write-side is flow controlled */
 201 #define MPTSF_BOUND_IF          0x400   /* subflow bound to an interface */
 202 #define MPTSF_BOUND_IP          0x800   /* subflow bound to a src address */
 203 #define MPTSF_BOUND_PORT        0x1000  /* subflow bound to a src port */
 204 #define MPTSF_PREFERRED         0x2000  /* primary/preferred subflow */
 205 #define MPTSF_SOPT_OLDVAL       0x4000  /* old option value is valid */
 206 #define MPTSF_SOPT_INPROG       0x8000  /* sosetopt in progress */
 207 #define MPTSF_DELETEOK          0x10000 /* subflow can be deleted */
 208 #define MPTSF_FAILINGOVER       0x20000 /* subflow not used for output */
 209 #define MPTSF_ACTIVE            0x40000 /* subflow currently in use */
 210 #define MPTSF_MPCAP_CTRSET      0x80000 /* mpcap counter */
 211 #define MPTSF_FASTJ_SEND        0x100000 /* send data after SYN in MP_JOIN */
 212 #define MPTSF_FASTJ_REQD        0x200000 /* fastjoin required */
 213 #define MPTSF_USER_DISCONNECT   0x400000 /* User triggered disconnect */
 214
 215 #define MPTSF_BITS \
 216         "\020\1ATTACHED\2CONNECTING\3PENDING\4CONNECTED\5DISCONNECTING" \
 217         "\6DISCONNECTED\7MP_CAPABLE\10MP_READY\11MP_DEGRADED\12SUSPENDED" \
 218         "\13BOUND_IF\14BOUND_IP\15BOUND_PORT\16PREFERRED\17SOPT_OLDVAL" \
 219         "\20SOPT_INPROG\21NOLINGER\22FAILINGOVER\23ACTIVE\24MPCAP_CTRSET" \
 220         "\25FASTJ_SEND\26FASTJ_REQD\27USER_DISCONNECT"
 221
 222 #define MPTS_LOCK_ASSERT_HELD(_mpts)                                    \
 223         lck_mtx_assert(&(_mpts)->mpts_lock, LCK_MTX_ASSERT_OWNED)
 224
 225 #define MPTS_LOCK_ASSERT_NOTHELD(_mpts)                                 \
 226         lck_mtx_assert(&(_mpts)->mpts_lock, LCK_MTX_ASSERT_NOTOWNED)
 227
 228 #define MPTS_LOCK(_mpts)                                                \
 229         lck_mtx_lock(&(_mpts)->mpts_lock)
 230
 231 #define MPTS_UNLOCK(_mpts)                                              \
 232         lck_mtx_unlock(&(_mpts)->mpts_lock)
 233
 234 #define MPTS_ADDREF(_mpts)                                              \
 235         mptcp_subflow_addref(_mpts, 0)
 236
 237 #define MPTS_ADDREF_LOCKED(_mpts)                                       \
 238         mptcp_subflow_addref(_mpts, 1)
 239
 240 #define MPTS_REMREF(_mpts)                                              \
 241         mptcp_subflow_remref(_mpts)
 242
 243 /*
 244  * MPTCP states
 245  * Keep in sync with bsd/dev/dtrace/mptcp.d
 246  */
 247 typedef enum mptcp_state {
 248         MPTCPS_CLOSED           = 0,    /* closed */
 249         MPTCPS_LISTEN           = 1,    /* not yet implemented */
 250         MPTCPS_ESTABLISHED      = 2,    /* MPTCP connection established */
 251         MPTCPS_CLOSE_WAIT       = 3,    /* rcvd DFIN, waiting for close */
 252         MPTCPS_FIN_WAIT_1       = 4,    /* have closed, sent DFIN */
 253         MPTCPS_CLOSING          = 5,    /* closed xchd DFIN, waiting DFIN ACK */
 254         MPTCPS_LAST_ACK         = 6,    /* had DFIN and close; await DFIN ACK */
 255         MPTCPS_FIN_WAIT_2       = 7,    /* have closed, DFIN is acked */
 256         MPTCPS_TIME_WAIT        = 8,    /* in 2*MSL quiet wait after close */
 257         MPTCPS_FASTCLOSE_WAIT   = 9,    /* sent MP_FASTCLOSE */
 258         MPTCPS_TERMINATE        = 10,   /* terminal state */
 259 } mptcp_state_t;
 260
 261 typedef u_int64_t       mptcp_key_t;
 262 typedef u_int32_t       mptcp_token_t;
 263 typedef u_int8_t        mptcp_addr_id;
 264
 265
 266 /* Address ID list */
 267 struct mptcp_subf_auth_entry {
 268         LIST_ENTRY(mptcp_subf_auth_entry) msae_next;
 269         u_int32_t       msae_laddr_rand;        /* Local nonce */
 270         u_int32_t       msae_raddr_rand;        /* Remote nonce */
 271         mptcp_addr_id   msae_laddr_id;          /* Local addr ID */
 272         mptcp_addr_id   msae_raddr_id;          /* Remote addr ID */
 273 };
 274
 275 /*
 276  * MPTCP Protocol Control Block
 277  *
 278  * Protected by per-MPTCP mpt_lock.
 279  * Keep in sync with bsd/dev/dtrace/scripts/mptcp.d.
 280  */
 281 struct mptcb {
 282         decl_lck_mtx_data(, mpt_lock);          /* per MPTCP PCB lock */
 283         struct mptses   *mpt_mpte;              /* back ptr to MPTCP session */
 284         mptcp_state_t   mpt_state;              /* MPTCP state */
 285         u_int32_t       mpt_flags;              /* see flags below */
 286         u_int32_t       mpt_refcnt;             /* references held on mptcb */
 287         u_int32_t       mpt_version;            /* MPTCP proto version */
 288         int             mpt_softerror;          /* error not yet reported */
 289         /*
 290          * Authentication and metadata invariants
 291          */
 292         mptcp_key_t     *mpt_localkey;          /* in network byte order */
 293         mptcp_key_t     mpt_remotekey;          /* in network byte order */
 294         mptcp_token_t   mpt_localtoken;         /* HMAC SHA1 of local key */
 295         mptcp_token_t   mpt_remotetoken;        /* HMAC SHA1 of remote key */
 296
 297         /*
 298          * Timer vars for scenarios where subflow level acks arrive, but
 299          * Data ACKs do not.
 300          */
 301         int             mpt_rxtshift;           /* num of consecutive retrans */
 302         u_int32_t       mpt_rxtstart;           /* time at which rxt started */
 303         u_int64_t       mpt_rtseq;              /* seq # being tracked */
 304         u_int32_t       mpt_timer_vals;         /* timer related values */
 305         u_int32_t       mpt_timewait;           /* timewait */
 306         /*
 307          * Sending side
 308          */
 309         u_int64_t       mpt_snduna;             /* DSN of last unacked byte */
 310         u_int64_t       mpt_sndnxt;             /* DSN of next byte to send */
 311         u_int64_t       mpt_sndmax;             /* DSN of max byte sent */
 312         u_int64_t       mpt_local_idsn;         /* First byte's DSN */
 313         u_int32_t       mpt_sndwnd;
 314         /*
 315          * Receiving side
 316          */
 317         u_int64_t       mpt_rcvnxt;             /* Next expected DSN */
 318         u_int64_t       mpt_rcvatmark;          /* mpsocket marker of rcvnxt */
 319         u_int64_t       mpt_remote_idsn;        /* Peer's IDSN */
 320         u_int32_t       mpt_rcvwnd;
 321         LIST_HEAD(, mptcp_subf_auth_entry) mpt_subauth_list; /* address IDs */
 322         /*
 323          * Fastclose
 324          */
 325         u_int64_t       mpt_dsn_at_csum_fail;   /* MPFail Opt DSN */
 326         u_int32_t       mpt_ssn_at_csum_fail;   /* MPFail Subflow Seq */
 327         /*
 328          * Zombie handling
 329          */
 330 #define MPT_GC_TICKS    (60)
 331         int32_t         mpt_gc_ticks;           /* Used for zombie deletion */
 332
 333         u_int32_t       mpt_notsent_lowat;      /* TCP_NOTSENT_LOWAT support */
 334 };
 335
 336 /* valid values for mpt_flags (see also notes on mpts_flags above) */
 337 #define MPTCPF_CHECKSUM         0x1     /* checksum DSS option */
 338 #define MPTCPF_FALLBACK_TO_TCP  0x2     /* Fallback to TCP */
 339 #define MPTCPF_JOIN_READY       0x4     /* Ready to start 2 or more subflows */
 340 #define MPTCPF_RECVD_MPFAIL     0x8     /* Received MP_FAIL option */
 341 #define MPTCPF_PEEL_OFF         0x10    /* Peel off this socket */
 342 #define MPTCPF_SND_64BITDSN     0x20    /* Send full 64-bit DSN */
 343 #define MPTCPF_SND_64BITACK     0x40    /* Send 64-bit ACK response */
 344 #define MPTCPF_RCVD_64BITACK    0x80    /* Received 64-bit Data ACK */
 345 #define MPTCPF_POST_FALLBACK_SYNC       0x100   /* Post fallback resend data */
 346
 347 #define MPTCPF_BITS \
 348         "\020\1CHECKSUM\2FALLBACK_TO_TCP\3JOIN_READY\4RECVD_MPFAIL\5PEEL_OFF" \
 349         "\6SND_64BITDSN\7SND_64BITACK\10RCVD_64BITACK\11POST_FALLBACK_SYNC"
 350
 351 /* valid values for mpt_timer_vals */
 352 #define MPTT_REXMT      0x01    /* Starting Retransmit Timer */
 353 #define MPTT_TW         0x02    /* Starting Timewait Timer */
 354 #define MPTT_FASTCLOSE  0x04    /* Starting Fastclose wait timer */
 355
 356 #define MPT_LOCK_ASSERT_HELD(_mpt)                                      \
 357         lck_mtx_assert(&(_mpt)->mpt_lock, LCK_MTX_ASSERT_OWNED)
 358
 359 #define MPT_LOCK_ASSERT_NOTHELD(_mpt)                                   \
 360         lck_mtx_assert(&(_mpt)->mpt_lock, LCK_MTX_ASSERT_NOTOWNED)
 361
 362 #define MPT_LOCK(_mpt)                                                  \
 363         lck_mtx_lock(&(_mpt)->mpt_lock)
 364
 365 #define MPT_LOCK_SPIN(_mpt)                                             \
 366         lck_mtx_lock_spin(&(_mpt)->mpt_lock)
 367
 368 #define MPT_CONVERT_LOCK(_mpt) do {                                     \
 369         MPT_LOCK_ASSERT_HELD(_mpt);                                     \
 370         lck_mtx_convert_spin(&(_mpt)->mpt_lock);                        \
 371 } while (0)
 372
 373 #define MPT_UNLOCK(_mpt)                                                \
 374         lck_mtx_unlock(&(_mpt)->mpt_lock)
 375
 376 /* events for close FSM */
 377 #define MPCE_CLOSE              0x1
 378 #define MPCE_RECV_DATA_ACK      0x2
 379 #define MPCE_RECV_DATA_FIN      0x4
 380
 381 /* mptcb manipulation */
 382 #define tptomptp(tp)    ((struct mptcb *)((tp)->t_mptcb))
 383
 384 /*
 385  * MPTCP control block and state structures are allocated along with
 386  * the MP protocol control block; the folllowing represents the layout.
 387  */
 388 struct mpp_mtp {
 389         struct mppcb            mpp;            /* Multipath PCB */
 390         struct mptses           mpp_ses;        /* MPTCP session */
 391         struct mptcb            mtcb;           /* MPTCP PCB */
 392 };
 393
 394 #ifdef SYSCTL_DECL
 395 SYSCTL_DECL(_net_inet_mptcp);
 396 #endif /* SYSCTL_DECL */
 397
 398 extern struct mppcbinfo mtcbinfo;
 399 extern struct pr_usrreqs mptcp_usrreqs;
 400
 401 /* Encryption algorithm related definitions */
 402 #define MPTCP_SHA1_RESULTLEN    20
 403 #define SHA1_TRUNCATED          8
 404
 405 /* List of valid keys to use for MPTCP connections */
 406 #define MPTCP_KEY_DIGEST_LEN            (MPTCP_SHA1_RESULTLEN)
 407 #define MPTCP_MX_KEY_ALLOCS             (256)
 408 #define MPTCP_KEY_PREALLOCS_MX          (16)
 409 #define MPTCP_MX_PREALLOC_ZONE_SZ       (8192)
 410
 411 struct mptcp_key_entry {
 412         LIST_ENTRY(mptcp_key_entry)     mkey_next;
 413         mptcp_key_t                     mkey_value;
 414 #define MKEYF_FREE      0x0
 415 #define MKEYF_INUSE     0x1
 416         u_int32_t                       mkey_flags;
 417         char                            mkey_digest[MPTCP_KEY_DIGEST_LEN];
 418 };
 419
 420 /* structure for managing unique key list */
 421 struct mptcp_keys_pool_head {
 422         struct mptcp_key_entry *lh_first;       /* list of keys */
 423         u_int32_t       mkph_count;             /* total keys in pool */
 424         vm_size_t       mkph_key_elm_sz;        /* size of key entry */
 425         struct zone     *mkph_key_entry_zone;   /* zone for key entry */
 426         decl_lck_mtx_data(, mkph_lock);         /* lock for key list */
 427 };
 428
 429 /* MPTCP Receive Window */
 430 #define MPTCP_RWIN_MAX  (1<<16)
 431
 432 /* MPTCP Debugging Levels */
 433 #define MP_NODEBUG              0x0
 434 #define MP_ERR_DEBUG            0x1
 435 #define MP_VERBOSE_DEBUG_1      0x2
 436 #define MP_VERBOSE_DEBUG_2      0x3
 437 #define MP_VERBOSE_DEBUG_3      0x4
 438 #define MP_VERBOSE_DEBUG_4      0x5     /* output path debugging */
 439
 440 /* Mask to obtain 32-bit portion of data sequence number */
 441 #define MPTCP_DATASEQ_LOW32_MASK        (0xffffffff)
 442 #define MPTCP_DATASEQ_LOW32(seq)        (seq & MPTCP_DATASEQ_LOW32_MASK)
 443
 444 /* Mask to obtain upper 32-bit portion of data sequence number */
 445 #define MPTCP_DATASEQ_HIGH32_MASK       (0xffffffff00000000)
 446 #define MPTCP_DATASEQ_HIGH32(seq)       (seq & MPTCP_DATASEQ_HIGH32_MASK)
 447
 448 /* Mask to obtain 32-bit portion of data ack */
 449 #define MPTCP_DATAACK_LOW32_MASK        (0xffffffff)
 450 #define MPTCP_DATAACK_LOW32(ack)        (ack & MPTCP_DATAACK_LOW32_MASK)
 451
 452 /* Mask to obtain upper 32-bit portion of data ack */
 453 #define MPTCP_DATAACK_HIGH32_MASK       (0xffffffff00000000)
 454 #define MPTCP_DATAACK_HIGH32(ack)       (ack & MPTCP_DATAACK_HIGH32_MASK)
 455
 456 /*
 457  * x is the 64-bit data sequence number, y the 32-bit data seq number to be
 458  * extended. z is y extended to the appropriate 64-bit value.
 459  * This algorithm is based on the fact that subflow level window sizes are
 460  * at the maximum 2**30 (in reality, they are a lot lesser). A high throughput
 461  * application sending on a large number of subflows can in theory have very
 462  * large MPTCP level send and receive windows. In which case, 64 bit DSNs
 463  * must be sent in place of 32 bit DSNs on wire. For us, with 2 subflows at
 464  * 512K each, sequence wraparound detection can be done by checking whether
 465  * the 32-bit value obtained on wire is 2**31 bytes apart from the stored
 466  * lower 32-bits of the Data Sequence Number. Bogus DSNs are dropped by
 467  * comparing against rwnd. Bogus DSNs within rwnd cannot be protected against
 468  * and are as weak as bogus TCP sequence numbers.
 469  */
 470 #define MPTCP_EXTEND_DSN(x, y, z) {                                     \
 471         if ((MPTCP_DATASEQ_LOW32(x) > y) &&                             \
 472             ((((u_int32_t)MPTCP_DATASEQ_LOW32(x)) - (u_int32_t)y) >=    \
 473             (u_int32_t)(1 << 31))) {                                    \
 474                 /*                                                      \
 475                  * y wrapped around and x and y are 2**31 bytes  apart  \
 476                  */                                                     \
 477                 z = MPTCP_DATASEQ_HIGH32(x) + 0x100000000;              \
 478                 z |= y;                                                 \
 479         } else if ((MPTCP_DATASEQ_LOW32(x) < y) &&                      \
 480             (((u_int32_t)y -                                            \
 481             ((u_int32_t)MPTCP_DATASEQ_LOW32(x))) >=                     \
 482             (u_int32_t)(1 << 31))) {                                    \
 483                 /*                                                      \
 484                  * x wrapped around and x and y are 2**31 apart         \
 485                  */                                                     \
 486                 z = MPTCP_DATASEQ_HIGH32(x) - 0x100000000;              \
 487                 z |= y;                                                 \
 488         } else {                                                        \
 489                 z = MPTCP_DATASEQ_HIGH32(x) | y;                        \
 490         }                                                               \
 491 }
 492
 493 #define mptcplog(x)     do { if (mptcp_verbose >= 1) log x; } while (0)
 494 #define mptcplog2(x)    do { if (mptcp_verbose >= 2) log x; } while (0)
 495 #define mptcplog3(x)    do { if (mptcp_verbose >= 3) log x; } while (0)
 496
 497 extern int mptcp_enable;        /* Multipath TCP */
 498 extern int mptcp_dbg;           /* Multipath TCP DBG */
 499 extern int mptcp_mpcap_retries; /* Multipath TCP retries */
 500 extern int mptcp_join_retries;  /* Multipath TCP Join retries */
 501 extern int mptcp_dss_csum;      /* Multipath DSS Option checksum */
 502 extern int mptcp_fail_thresh;   /* Multipath failover thresh of retransmits */
 503 extern int mptcp_subflow_keeptime; /* Multipath subflow TCP_KEEPALIVE opt */
 504 extern int mptcp_mpprio_enable; /* MP_PRIO option enable/disable */
 505 extern int mptcp_remaddr_enable;/* REMOVE_ADDR option enable/disable */
 506 extern int mptcp_fastjoin;      /* Enable FastJoin */
 507 extern int mptcp_zerortt_fastjoin; /* Enable Data after SYN Fast Join */
 508 extern int mptcp_rwnotify;      /* Enable RW notification on resume */
 509 extern uint32_t mptcp_verbose;  /* verbose and mptcp_dbg must be unified */
 510 #define MPPCB_LIMIT     16
 511 extern uint32_t mptcp_socket_limit; /* max number of mptcp sockets allowed */
 512 extern uint32_t mptcp_delayed_subf_start; /* delayed cellular subflow start */
 513 extern int tcp_jack_rxmt;       /* Join ACK retransmission value in msecs */
 514
 515 __BEGIN_DECLS
 516 extern void mptcp_init(struct protosw *, struct domain *);
 517 extern int mptcp_ctloutput(struct socket *, struct sockopt *);
 518 extern struct mptses *mptcp_sescreate(struct socket *, struct mppcb *);
 519 extern void mptcp_drain(void);
 520 extern struct mptses *mptcp_drop(struct mptses *, struct mptcb *, int);
 521 extern struct mptses *mptcp_close(struct mptses *, struct mptcb *);
 522 extern int mptcp_lock(struct socket *, int, void *);
 523 extern int mptcp_unlock(struct socket *, int, void *);
 524 extern lck_mtx_t *mptcp_getlock(struct socket *, int);
 525 extern void mptcp_thread_signal(struct mptses *);
 526 extern void mptcp_flush_sopts(struct mptses *);
 527 extern int mptcp_setconnorder(struct mptses *, connid_t, uint32_t);
 528 extern int mptcp_getconnorder(struct mptses *, connid_t, uint32_t *);
 529
 530 extern struct mptopt *mptcp_sopt_alloc(int);
 531 extern const char *mptcp_sopt2str(int, int, char *, int);
 532 extern void mptcp_sopt_free(struct mptopt *);
 533 extern void mptcp_sopt_insert(struct mptses *, struct mptopt *);
 534 extern void mptcp_sopt_remove(struct mptses *, struct mptopt *);
 535 extern struct mptopt *mptcp_sopt_find(struct mptses *, struct sockopt *);
 536
 537 extern struct mptsub *mptcp_subflow_alloc(int);
 538 extern void mptcp_subflow_free(struct mptsub *);
 539 extern void mptcp_subflow_addref(struct mptsub *, int);
 540 extern int mptcp_subflow_add(struct mptses *, struct mptsub *,
 541     struct proc *, uint32_t);
 542 extern void mptcp_subflow_del(struct mptses *, struct mptsub *, boolean_t);
 543 extern void mptcp_subflow_remref(struct mptsub *);
 544 extern int mptcp_subflow_output(struct mptses *, struct mptsub *);
 545 extern void mptcp_subflow_disconnect(struct mptses *, struct mptsub *,
 546     boolean_t);
 547 extern void mptcp_subflow_sopeeloff(struct mptses *, struct mptsub *,
 548     struct socket *);
 549 extern int mptcp_subflow_sosetopt(struct mptses *, struct socket *,
 550     struct mptopt *);
 551 extern int mptcp_subflow_sogetopt(struct mptses *, struct socket *,
 552     struct mptopt *);
 553
 554 extern void mptcp_input(struct mptses *, struct mbuf *);
 555 extern int mptcp_output(struct mptses *);
 556 extern void mptcp_close_fsm(struct mptcb *, uint32_t);
 557
 558 extern mptcp_token_t mptcp_get_localtoken(void *);
 559 extern mptcp_token_t mptcp_get_remotetoken(void *);
 560
 561 extern u_int64_t mptcp_get_localkey(void *);
 562 extern u_int64_t mptcp_get_remotekey(void *);
 563
 564 extern void mptcp_free_key(mptcp_key_t *key);
 565 extern void mptcp_hmac_sha1(mptcp_key_t, mptcp_key_t, u_int32_t, u_int32_t,
 566     u_char*, int);
 567 extern void mptcp_get_hmac(mptcp_addr_id, struct mptcb *, u_char *, int);
 568 extern void mptcp_get_rands(mptcp_addr_id, struct mptcb *, u_int32_t *,
 569     u_int32_t *);
 570 extern void mptcp_set_raddr_rand(mptcp_addr_id, struct mptcb *, mptcp_addr_id,
 571     u_int32_t);
 572 extern u_int64_t mptcp_get_trunced_hmac(mptcp_addr_id, struct mptcb *mp_tp);
 573 extern int mptcp_generate_token(char *, int, caddr_t, int);
 574 extern int mptcp_generate_idsn(char *, int, caddr_t, int);
 575 extern boolean_t mptcp_ok_to_keepalive(struct mptcb *);
 576 extern void mptcp_insert_dsn(struct mppcb *, struct mbuf *);
 577 extern void  mptcp_output_getm_dsnmap32(struct socket *, int, uint32_t,
 578     u_int32_t *, u_int32_t *, u_int16_t *, u_int64_t *);
 579 extern void  mptcp_output_getm_dsnmap64(struct socket *, int, uint32_t,
 580     u_int64_t *, u_int32_t *, u_int16_t *);
 581 extern void mptcp_send_dfin(struct socket *);
 582 extern void mptcp_act_on_txfail(struct socket *);
 583 extern struct mptsub *mptcp_get_subflow(struct mptses *, struct mptsub *);
 584 extern struct mptsub *mptcp_get_pending_subflow(struct mptses *,
 585     struct mptsub *);
 586 extern int mptcp_get_map_for_dsn(struct socket *, u_int64_t, u_int32_t *);
 587 extern int32_t mptcp_adj_sendlen(struct socket *so, int32_t off, int32_t len);
 588 extern int32_t mptcp_sbspace(struct mptcb *);
 589 extern void mptcp_notify_mpready(struct socket *);
 590 extern void mptcp_notify_mpfail(struct socket *);
 591 extern void mptcp_notify_close(struct socket *);
 592 extern boolean_t mptcp_no_rto_spike(struct socket*);
 593 extern int mptcp_set_notsent_lowat(struct mptses *mpte, int optval);
 594 extern u_int32_t mptcp_get_notsent_lowat(struct mptses *mpte);
 595 extern int mptcp_notsent_lowat_check(struct socket *so);
 596
 597 __END_DECLS
 598
 599 #endif /* BSD_KERNEL_PRIVATE */
 600 #ifdef PRIVATE
 601 typedef struct mptcp_flow {
 602         uint32_t                flow_flags;
 603         connid_t                flow_cid;
 604         struct sockaddr_storage flow_src;
 605         struct sockaddr_storage flow_dst;
 606         conninfo_tcp_t          flow_ci;
 607 } mptcp_flow_t;
 608
 609 typedef struct conninfo_mptcp {
 610         size_t          mptcpci_len;
 611         size_t          mptcpci_nflows;
 612         uint32_t        mptcpci_state;
 613         mptcp_flow_t    mptcpci_flows[1];
 614 } conninfo_mptcp_t;
 615
 616 #endif /* PRIVATE */
 617 #endif /* _NETINET_MPTCP_VAR_H_ */