drivers/staging/lustre/lnet/klnds/socklnd/socklnd.h

   1 /*
   2  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
   3  *
   4  * Copyright (c) 2011, 2012, Intel Corporation.
   5  *
   6  *   Author: Zach Brown <zab@zabbo.net>
   7  *   Author: Peter J. Braam <braam@clusterfs.com>
   8  *   Author: Phil Schwan <phil@clusterfs.com>
   9  *   Author: Eric Barton <eric@bartonsoftware.com>
  10  *
  11  *   This file is part of Lustre, http://www.lustre.org
  12  *
  13  *   Portals is free software; you can redistribute it and/or
  14  *   modify it under the terms of version 2 of the GNU General Public
  15  *   License as published by the Free Software Foundation.
  16  *
  17  *   Portals is distributed in the hope that it will be useful,
  18  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  *   GNU General Public License for more details.
  21  *
  22  *   You should have received a copy of the GNU General Public License
  23  *   along with Portals; if not, write to the Free Software
  24  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  25  *
  26  */
  27
  28 #ifndef _SOCKLND_SOCKLND_H_
  29 #define _SOCKLND_SOCKLND_H_
  30
  31 #define DEBUG_PORTAL_ALLOC
  32 #define DEBUG_SUBSYSTEM S_LND
  33
  34 #include <linux/crc32.h>
  35 #include <linux/errno.h>
  36 #include <linux/if.h>
  37 #include <linux/init.h>
  38 #include <linux/kernel.h>
  39 #include <linux/kmod.h>
  40 #include <linux/list.h>
  41 #include <linux/mm.h>
  42 #include <linux/module.h>
  43 #include <linux/stat.h>
  44 #include <linux/string.h>
  45 #include <linux/syscalls.h>
  46 #include <linux/sysctl.h>
  47 #include <linux/uio.h>
  48 #include <linux/unistd.h>
  49 #include <net/sock.h>
  50 #include <net/tcp.h>
  51
  52 #include "../../../include/linux/libcfs/libcfs.h"
  53 #include "../../../include/linux/lnet/lnet.h"
  54 #include "../../../include/linux/lnet/lib-lnet.h"
  55 #include "../../../include/linux/lnet/socklnd.h"
  56
  57 /* assume one thread for each connection type */
  58 #define SOCKNAL_NSCHEDS         3
  59 #define SOCKNAL_NSCHEDS_HIGH    (SOCKNAL_NSCHEDS << 1)
  60
  61 #define SOCKNAL_PEER_HASH_SIZE  101   /* # peer lists */
  62 #define SOCKNAL_RESCHED         100   /* # scheduler loops before reschedule */
  63 #define SOCKNAL_INSANITY_RECONN 5000  /* connd is trying on reconn infinitely */
  64 #define SOCKNAL_ENOMEM_RETRY    CFS_TICK /* jiffies between retries */
  65
  66 #define SOCKNAL_SINGLE_FRAG_TX  0     /* disable multi-fragment sends */
  67 #define SOCKNAL_SINGLE_FRAG_RX  0     /* disable multi-fragment receives */
  68
  69 #define SOCKNAL_VERSION_DEBUG   0     /* enable protocol version debugging */
  70
  71 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
  72  * no risk if we're not running on a CONFIG_HIGHMEM platform. */
  73 #ifdef CONFIG_HIGHMEM
  74 # define SOCKNAL_RISK_KMAP_DEADLOCK  0
  75 #else
  76 # define SOCKNAL_RISK_KMAP_DEADLOCK  1
  77 #endif
  78
  79 struct ksock_sched_info;
  80
  81 typedef struct                            /* per scheduler state */
  82 {
  83         spinlock_t              kss_lock;       /* serialise */
  84         struct list_head        kss_rx_conns;   /* conn waiting to be read */
  85         struct list_head        kss_tx_conns;   /* conn waiting to be written */
  86         struct list_head        kss_zombie_noop_txs; /* zombie noop tx list */
  87         wait_queue_head_t       kss_waitq;      /* where scheduler sleeps */
  88         int                     kss_nconns;     /* # connections assigned to
  89                                                  * this scheduler */
  90         struct ksock_sched_info *kss_info;      /* owner of it */
  91         struct page             *kss_rx_scratch_pgs[LNET_MAX_IOV];
  92         struct kvec             kss_scratch_iov[LNET_MAX_IOV];
  93 } ksock_sched_t;
  94
  95 struct ksock_sched_info {
  96         int                     ksi_nthreads_max; /* max allowed threads */
  97         int                     ksi_nthreads;     /* number of threads */
  98         int                     ksi_cpt;          /* CPT id */
  99         ksock_sched_t           *ksi_scheds;      /* array of schedulers */
 100 };
 101
 102 #define KSOCK_CPT_SHIFT           16
 103 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
 104 #define KSOCK_THREAD_CPT(id)      ((id) >> KSOCK_CPT_SHIFT)
 105 #define KSOCK_THREAD_SID(id)      ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
 106
 107 typedef struct                                  /* in-use interface */
 108 {
 109         __u32           ksni_ipaddr;            /* interface's IP address */
 110         __u32           ksni_netmask;           /* interface's network mask */
 111         int             ksni_nroutes;           /* # routes using (active) */
 112         int             ksni_npeers;            /* # peers using (passive) */
 113         char            ksni_name[IFNAMSIZ];    /* interface name */
 114 } ksock_interface_t;
 115
 116 typedef struct {
 117         int          *ksnd_timeout;            /* "stuck" socket timeout
 118                                                 * (seconds) */
 119         int          *ksnd_nscheds;            /* # scheduler threads in each
 120                                                 * pool while starting */
 121         int          *ksnd_nconnds;            /* # connection daemons */
 122         int          *ksnd_nconnds_max;        /* max # connection daemons */
 123         int          *ksnd_min_reconnectms;    /* first connection retry after
 124                                                 * (ms)... */
 125         int          *ksnd_max_reconnectms;    /* ...exponentially increasing to
 126                                                 * this */
 127         int          *ksnd_eager_ack;          /* make TCP ack eagerly? */
 128         int          *ksnd_typed_conns;        /* drive sockets by type? */
 129         int          *ksnd_min_bulk;           /* smallest "large" message */
 130         int          *ksnd_tx_buffer_size;     /* socket tx buffer size */
 131         int          *ksnd_rx_buffer_size;     /* socket rx buffer size */
 132         int          *ksnd_nagle;              /* enable NAGLE? */
 133         int          *ksnd_round_robin;        /* round robin for multiple
 134                                                 * interfaces */
 135         int          *ksnd_keepalive;          /* # secs for sending keepalive
 136                                                 * NOOP */
 137         int          *ksnd_keepalive_idle;     /* # idle secs before 1st probe
 138                                                 */
 139         int          *ksnd_keepalive_count;    /* # probes */
 140         int          *ksnd_keepalive_intvl;    /* time between probes */
 141         int          *ksnd_credits;            /* # concurrent sends */
 142         int          *ksnd_peertxcredits;      /* # concurrent sends to 1 peer
 143                                                 */
 144         int          *ksnd_peerrtrcredits;     /* # per-peer router buffer
 145                                                 * credits */
 146         int          *ksnd_peertimeout;        /* seconds to consider peer dead
 147                                                 */
 148         int          *ksnd_enable_csum;        /* enable check sum */
 149         int          *ksnd_inject_csum_error;  /* set non-zero to inject
 150                                                 * checksum error */
 151         int          *ksnd_nonblk_zcack;       /* always send zc-ack on
 152                                                 * non-blocking connection */
 153         unsigned int *ksnd_zc_min_payload;     /* minimum zero copy payload
 154                                                 * size */
 155         int          *ksnd_zc_recv;            /* enable ZC receive (for
 156                                                 * Chelsio TOE) */
 157         int          *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to
 158                                                 * enable ZC receive */
 159 } ksock_tunables_t;
 160
 161 typedef struct {
 162         __u64             ksnn_incarnation;     /* my epoch */
 163         spinlock_t        ksnn_lock;            /* serialise */
 164         struct list_head          ksnn_list;            /* chain on global list */
 165         int               ksnn_npeers;          /* # peers */
 166         int               ksnn_shutdown;        /* shutting down? */
 167         int               ksnn_ninterfaces;     /* IP interfaces */
 168         ksock_interface_t ksnn_interfaces[LNET_MAX_INTERFACES];
 169 } ksock_net_t;
 170
 171 /** connd timeout */
 172 #define SOCKNAL_CONND_TIMEOUT  120
 173 /** reserved thread for accepting & creating new connd */
 174 #define SOCKNAL_CONND_RESV     1
 175
 176 typedef struct {
 177         int                     ksnd_init;              /* initialisation state
 178                                                          */
 179         int                     ksnd_nnets;             /* # networks set up */
 180         struct list_head        ksnd_nets;              /* list of nets */
 181         rwlock_t                ksnd_global_lock;       /* stabilize peer/conn
 182                                                          * ops */
 183         struct list_head        *ksnd_peers;            /* hash table of all my
 184                                                          * known peers */
 185         int                     ksnd_peer_hash_size;    /* size of ksnd_peers */
 186
 187         int                     ksnd_nthreads;          /* # live threads */
 188         int                     ksnd_shuttingdown;      /* tell threads to exit
 189                                                          */
 190         struct ksock_sched_info **ksnd_sched_info;      /* schedulers info */
 191
 192         atomic_t                ksnd_nactive_txs;       /* #active txs */
 193
 194         struct list_head        ksnd_deathrow_conns;    /* conns to close:
 195                                                          * reaper_lock*/
 196         struct list_head        ksnd_zombie_conns;      /* conns to free:
 197                                                          * reaper_lock */
 198         struct list_head        ksnd_enomem_conns;      /* conns to retry:
 199                                                          * reaper_lock*/
 200         wait_queue_head_t       ksnd_reaper_waitq;      /* reaper sleeps here */
 201         unsigned long           ksnd_reaper_waketime;   /* when reaper will wake
 202                                                          */
 203         spinlock_t              ksnd_reaper_lock;       /* serialise */
 204
 205         int                     ksnd_enomem_tx;         /* test ENOMEM sender */
 206         int                     ksnd_stall_tx;          /* test sluggish sender
 207                                                          */
 208         int                     ksnd_stall_rx;          /* test sluggish
 209                                                          * receiver */
 210
 211         struct list_head        ksnd_connd_connreqs;    /* incoming connection
 212                                                          * requests */
 213         struct list_head        ksnd_connd_routes;      /* routes waiting to be
 214                                                          * connected */
 215         wait_queue_head_t       ksnd_connd_waitq;       /* connds sleep here */
 216         int                     ksnd_connd_connecting;  /* # connds connecting
 217                                                          */
 218         long                    ksnd_connd_failed_stamp;/* time stamp of the
 219                                                          * last failed
 220                                                          * connecting attempt */
 221         unsigned                ksnd_connd_starting;    /* # starting connd */
 222         long                    ksnd_connd_starting_stamp;/* time stamp of the
 223                                                            * last starting connd
 224                                                            */
 225         unsigned                ksnd_connd_running;     /* # running connd */
 226         spinlock_t              ksnd_connd_lock;        /* serialise */
 227
 228         struct list_head        ksnd_idle_noop_txs;     /* list head for freed
 229                                                          * noop tx */
 230         spinlock_t              ksnd_tx_lock;           /* serialise, g_lock
 231                                                          * unsafe */
 232
 233 } ksock_nal_data_t;
 234
 235 #define SOCKNAL_INIT_NOTHING 0
 236 #define SOCKNAL_INIT_DATA    1
 237 #define SOCKNAL_INIT_ALL     2
 238
 239 /* A packet just assembled for transmission is represented by 1 or more
 240  * struct iovec fragments (the first frag contains the portals header),
 241  * followed by 0 or more lnet_kiov_t fragments.
 242  *
 243  * On the receive side, initially 1 struct iovec fragment is posted for
 244  * receive (the header).  Once the header has been received, the payload is
 245  * received into either struct iovec or lnet_kiov_t fragments, depending on
 246  * what the header matched or whether the message needs forwarding. */
 247
 248 struct ksock_conn;  /* forward ref */
 249 struct ksock_peer;  /* forward ref */
 250 struct ksock_route; /* forward ref */
 251 struct ksock_proto; /* forward ref */
 252
 253 typedef struct                             /* transmit packet */
 254 {
 255         struct list_head  tx_list;         /* queue on conn for transmission etc
 256                                             */
 257         struct list_head  tx_zc_list;      /* queue on peer for ZC request */
 258         atomic_t          tx_refcount;     /* tx reference count */
 259         int               tx_nob;          /* # packet bytes */
 260         int               tx_resid;        /* residual bytes */
 261         int               tx_niov;         /* # packet iovec frags */
 262         struct kvec       *tx_iov;         /* packet iovec frags */
 263         int               tx_nkiov;        /* # packet page frags */
 264         unsigned short    tx_zc_aborted;   /* aborted ZC request */
 265         unsigned short    tx_zc_capable:1; /* payload is large enough for ZC */
 266         unsigned short    tx_zc_checked:1; /* Have I checked if I should ZC? */
 267         unsigned short    tx_nonblk:1;     /* it's a non-blocking ACK */
 268         lnet_kiov_t       *tx_kiov;        /* packet page frags */
 269         struct ksock_conn *tx_conn;        /* owning conn */
 270         lnet_msg_t        *tx_lnetmsg;     /* lnet message for lnet_finalize()
 271                                             */
 272         unsigned long     tx_deadline;     /* when (in jiffies) tx times out */
 273         ksock_msg_t       tx_msg;          /* socklnd message buffer */
 274         int               tx_desc_size;    /* size of this descriptor */
 275         union {
 276                 struct {
 277                         struct kvec iov;     /* virt hdr */
 278                         lnet_kiov_t kiov[0]; /* paged payload */
 279                 } paged;
 280                 struct {
 281                         struct kvec iov[1];  /* virt hdr + payload */
 282                 } virt;
 283         } tx_frags;
 284 } ksock_tx_t;
 285
 286 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
 287
 288 /* network zero copy callback descriptor embedded in ksock_tx_t */
 289
 290 /* space for the rx frag descriptors; we either read a single contiguous
 291  * header, or up to LNET_MAX_IOV frags of payload of either type. */
 292 typedef union {
 293         struct kvec      iov[LNET_MAX_IOV];
 294         lnet_kiov_t      kiov[LNET_MAX_IOV];
 295 } ksock_rxiovspace_t;
 296
 297 #define SOCKNAL_RX_KSM_HEADER   1 /* reading ksock message header */
 298 #define SOCKNAL_RX_LNET_HEADER  2 /* reading lnet message header */
 299 #define SOCKNAL_RX_PARSE        3 /* Calling lnet_parse() */
 300 #define SOCKNAL_RX_PARSE_WAIT   4 /* waiting to be told to read the body */
 301 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
 302 #define SOCKNAL_RX_SLOP         6 /* skipping body */
 303
 304 typedef struct ksock_conn {
 305         struct ksock_peer  *ksnc_peer;        /* owning peer */
 306         struct ksock_route *ksnc_route;       /* owning route */
 307         struct list_head   ksnc_list;         /* stash on peer's conn list */
 308         struct socket      *ksnc_sock;        /* actual socket */
 309         void               *ksnc_saved_data_ready;  /* socket's original
 310                                                      * data_ready() callback */
 311         void               *ksnc_saved_write_space; /* socket's original
 312                                                      * write_space() callback */
 313         atomic_t           ksnc_conn_refcount;/* conn refcount */
 314         atomic_t           ksnc_sock_refcount;/* sock refcount */
 315         ksock_sched_t      *ksnc_scheduler;   /* who schedules this connection
 316                                                */
 317         __u32              ksnc_myipaddr;     /* my IP */
 318         __u32              ksnc_ipaddr;       /* peer's IP */
 319         int                ksnc_port;         /* peer's port */
 320         signed int         ksnc_type:3;       /* type of connection, should be
 321                                                * signed value */
 322         unsigned int       ksnc_closing:1;    /* being shut down */
 323         unsigned int       ksnc_flip:1;       /* flip or not, only for V2.x */
 324         unsigned int       ksnc_zc_capable:1; /* enable to ZC */
 325         struct ksock_proto *ksnc_proto;       /* protocol for the connection */
 326
 327         /* reader */
 328         struct list_head   ksnc_rx_list;      /* where I enq waiting input or a
 329                                                * forwarding descriptor */
 330         unsigned long      ksnc_rx_deadline;  /* when (in jiffies) receive times
 331                                                * out */
 332         __u8               ksnc_rx_started;   /* started receiving a message */
 333         __u8               ksnc_rx_ready;     /* data ready to read */
 334         __u8               ksnc_rx_scheduled; /* being progressed */
 335         __u8               ksnc_rx_state;     /* what is being read */
 336         int                ksnc_rx_nob_left;  /* # bytes to next hdr/body */
 337         int                ksnc_rx_nob_wanted;/* bytes actually wanted */
 338         int                ksnc_rx_niov;      /* # iovec frags */
 339         struct kvec        *ksnc_rx_iov;      /* the iovec frags */
 340         int                ksnc_rx_nkiov;     /* # page frags */
 341         lnet_kiov_t        *ksnc_rx_kiov;     /* the page frags */
 342         ksock_rxiovspace_t ksnc_rx_iov_space; /* space for frag descriptors */
 343         __u32              ksnc_rx_csum;      /* partial checksum for incoming
 344                                                * data */
 345         void               *ksnc_cookie;      /* rx lnet_finalize passthru arg
 346                                                */
 347         ksock_msg_t        ksnc_msg;          /* incoming message buffer:
 348                                                * V2.x message takes the
 349                                                * whole struct
 350                                                * V1.x message is a bare
 351                                                * lnet_hdr_t, it's stored in
 352                                                * ksnc_msg.ksm_u.lnetmsg */
 353
 354         /* WRITER */
 355         struct list_head   ksnc_tx_list;      /* where I enq waiting for output
 356                                                * space */
 357         struct list_head   ksnc_tx_queue;     /* packets waiting to be sent */
 358         ksock_tx_t         *ksnc_tx_carrier;  /* next TX that can carry a LNet
 359                                                * message or ZC-ACK */
 360         unsigned long      ksnc_tx_deadline;  /* when (in jiffies) tx times out
 361                                                */
 362         int                ksnc_tx_bufnob;    /* send buffer marker */
 363         atomic_t           ksnc_tx_nob;       /* # bytes queued */
 364         int                ksnc_tx_ready;     /* write space */
 365         int                ksnc_tx_scheduled; /* being progressed */
 366         unsigned long      ksnc_tx_last_post; /* time stamp of the last posted
 367                                                * TX */
 368 } ksock_conn_t;
 369
 370 typedef struct ksock_route {
 371         struct list_head  ksnr_list;           /* chain on peer route list */
 372         struct list_head  ksnr_connd_list;     /* chain on ksnr_connd_routes */
 373         struct ksock_peer *ksnr_peer;          /* owning peer */
 374         atomic_t          ksnr_refcount;       /* # users */
 375         unsigned long     ksnr_timeout;        /* when (in jiffies) reconnection
 376                                                 * can happen next */
 377         long              ksnr_retry_interval; /* how long between retries */
 378         __u32             ksnr_myipaddr;       /* my IP */
 379         __u32             ksnr_ipaddr;         /* IP address to connect to */
 380         int               ksnr_port;           /* port to connect to */
 381         unsigned int      ksnr_scheduled:1;    /* scheduled for attention */
 382         unsigned int      ksnr_connecting:1;   /* connection establishment in
 383                                                 * progress */
 384         unsigned int      ksnr_connected:4;    /* connections established by
 385                                                 * type */
 386         unsigned int      ksnr_deleted:1;      /* been removed from peer? */
 387         unsigned int      ksnr_share_count;    /* created explicitly? */
 388         int               ksnr_conn_count;     /* # conns established by this
 389                                                 * route */
 390 } ksock_route_t;
 391
 392 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
 393
 394 typedef struct ksock_peer {
 395         struct list_head   ksnp_list;           /* stash on global peer list */
 396         unsigned long      ksnp_last_alive;     /* when (in jiffies) I was last
 397                                                  * alive */
 398         lnet_process_id_t  ksnp_id;             /* who's on the other end(s) */
 399         atomic_t           ksnp_refcount;       /* # users */
 400         int                ksnp_sharecount;     /* lconf usage counter */
 401         int                ksnp_closing;        /* being closed */
 402         int                ksnp_accepting;      /* # passive connections pending
 403                                                  */
 404         int                ksnp_error;          /* errno on closing last conn */
 405         __u64              ksnp_zc_next_cookie; /* ZC completion cookie */
 406         __u64              ksnp_incarnation;    /* latest known peer incarnation
 407                                                  */
 408         struct ksock_proto *ksnp_proto;         /* latest known peer protocol */
 409         struct list_head   ksnp_conns;          /* all active connections */
 410         struct list_head   ksnp_routes;         /* routes */
 411         struct list_head   ksnp_tx_queue;       /* waiting packets */
 412         spinlock_t         ksnp_lock;           /* serialize, g_lock unsafe */
 413         struct list_head   ksnp_zc_req_list;    /* zero copy requests wait for
 414                                                  * ACK  */
 415         unsigned long      ksnp_send_keepalive; /* time to send keepalive */
 416         lnet_ni_t          *ksnp_ni;            /* which network */
 417         int                ksnp_n_passive_ips;  /* # of... */
 418
 419         /* preferred local interfaces */
 420         __u32              ksnp_passive_ips[LNET_MAX_INTERFACES];
 421 } ksock_peer_t;
 422
 423 typedef struct ksock_connreq {
 424         struct list_head ksncr_list;  /* stash on ksnd_connd_connreqs */
 425         lnet_ni_t        *ksncr_ni;   /* chosen NI */
 426         struct socket    *ksncr_sock; /* accepted socket */
 427 } ksock_connreq_t;
 428
 429 extern ksock_nal_data_t ksocknal_data;
 430 extern ksock_tunables_t ksocknal_tunables;
 431
 432 #define SOCKNAL_MATCH_NO  0 /* TX can't match type of connection */
 433 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
 434 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not
 435                              * preferred */
 436
 437 typedef struct ksock_proto {
 438         /* version number of protocol */
 439         int        pro_version;
 440
 441         /* handshake function */
 442         int        (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *);
 443
 444         /* handshake function */
 445         int        (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);
 446
 447         /* message pack */
 448         void       (*pro_pack)(ksock_tx_t *);
 449
 450         /* message unpack */
 451         void       (*pro_unpack)(ksock_msg_t *);
 452
 453         /* queue tx on the connection */
 454         ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *);
 455
 456         /* queue ZC ack on the connection */
 457         int        (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64);
 458
 459         /* handle ZC request */
 460         int        (*pro_handle_zcreq)(ksock_conn_t *, __u64, int);
 461
 462         /* handle ZC ACK */
 463         int        (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64);
 464
 465         /* msg type matches the connection type:
 466          * return value:
 467          *   return MATCH_NO  : no
 468          *   return MATCH_YES : matching type
 469          *   return MATCH_MAY : can be backup */
 470         int        (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int);
 471 } ksock_proto_t;
 472
 473 extern ksock_proto_t ksocknal_protocol_v1x;
 474 extern ksock_proto_t ksocknal_protocol_v2x;
 475 extern ksock_proto_t ksocknal_protocol_v3x;
 476
 477 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
 478 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
 479 #define KSOCK_PROTO_V1       KSOCK_PROTO_V1_MAJOR
 480
 481 #ifndef CPU_MASK_NONE
 482 #define CPU_MASK_NONE   0UL
 483 #endif
 484
 485 static inline __u32 ksocknal_csum(__u32 crc, unsigned char const *p, size_t len)
 486 {
 487 #if 1
 488         return crc32_le(crc, p, len);
 489 #else
 490         while (len-- > 0)
 491                 crc = ((crc + 0x100) & ~0xff) | ((crc + *p++) & 0xff) ;
 492         return crc;
 493 #endif
 494 }
 495
 496 static inline int
 497 ksocknal_route_mask(void)
 498 {
 499         if (!*ksocknal_tunables.ksnd_typed_conns)
 500                 return (1 << SOCKLND_CONN_ANY);
 501
 502         return ((1 << SOCKLND_CONN_CONTROL) |
 503                 (1 << SOCKLND_CONN_BULK_IN) |
 504                 (1 << SOCKLND_CONN_BULK_OUT));
 505 }
 506
 507 static inline struct list_head *
 508 ksocknal_nid2peerlist(lnet_nid_t nid)
 509 {
 510         unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
 511
 512         return &ksocknal_data.ksnd_peers[hash];
 513 }
 514
 515 static inline void
 516 ksocknal_conn_addref(ksock_conn_t *conn)
 517 {
 518         LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
 519         atomic_inc(&conn->ksnc_conn_refcount);
 520 }
 521
 522 extern void ksocknal_queue_zombie_conn(ksock_conn_t *conn);
 523 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
 524
 525 static inline void
 526 ksocknal_conn_decref(ksock_conn_t *conn)
 527 {
 528         LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
 529         if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
 530                 ksocknal_queue_zombie_conn(conn);
 531 }
 532
 533 static inline int
 534 ksocknal_connsock_addref(ksock_conn_t *conn)
 535 {
 536         int rc = -ESHUTDOWN;
 537
 538         read_lock(&ksocknal_data.ksnd_global_lock);
 539         if (!conn->ksnc_closing) {
 540                 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
 541                 atomic_inc(&conn->ksnc_sock_refcount);
 542                 rc = 0;
 543         }
 544         read_unlock(&ksocknal_data.ksnd_global_lock);
 545
 546         return rc;
 547 }
 548
 549 static inline void
 550 ksocknal_connsock_decref(ksock_conn_t *conn)
 551 {
 552         LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
 553         if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
 554                 LASSERT(conn->ksnc_closing);
 555                 sock_release(conn->ksnc_sock);
 556                 conn->ksnc_sock = NULL;
 557                 ksocknal_finalize_zcreq(conn);
 558         }
 559 }
 560
 561 static inline void
 562 ksocknal_tx_addref(ksock_tx_t *tx)
 563 {
 564         LASSERT(atomic_read(&tx->tx_refcount) > 0);
 565         atomic_inc(&tx->tx_refcount);
 566 }
 567
 568 extern void ksocknal_tx_prep(ksock_conn_t *, ksock_tx_t *tx);
 569 extern void ksocknal_tx_done(lnet_ni_t *ni, ksock_tx_t *tx);
 570
 571 static inline void
 572 ksocknal_tx_decref(ksock_tx_t *tx)
 573 {
 574         LASSERT(atomic_read(&tx->tx_refcount) > 0);
 575         if (atomic_dec_and_test(&tx->tx_refcount))
 576                 ksocknal_tx_done(NULL, tx);
 577 }
 578
 579 static inline void
 580 ksocknal_route_addref(ksock_route_t *route)
 581 {
 582         LASSERT(atomic_read(&route->ksnr_refcount) > 0);
 583         atomic_inc(&route->ksnr_refcount);
 584 }
 585
 586 extern void ksocknal_destroy_route(ksock_route_t *route);
 587
 588 static inline void
 589 ksocknal_route_decref(ksock_route_t *route)
 590 {
 591         LASSERT(atomic_read(&route->ksnr_refcount) > 0);
 592         if (atomic_dec_and_test(&route->ksnr_refcount))
 593                 ksocknal_destroy_route(route);
 594 }
 595
 596 static inline void
 597 ksocknal_peer_addref(ksock_peer_t *peer)
 598 {
 599         LASSERT(atomic_read(&peer->ksnp_refcount) > 0);
 600         atomic_inc(&peer->ksnp_refcount);
 601 }
 602
 603 extern void ksocknal_destroy_peer(ksock_peer_t *peer);
 604
 605 static inline void
 606 ksocknal_peer_decref(ksock_peer_t *peer)
 607 {
 608         LASSERT(atomic_read(&peer->ksnp_refcount) > 0);
 609         if (atomic_dec_and_test(&peer->ksnp_refcount))
 610                 ksocknal_destroy_peer(peer);
 611 }
 612
 613 int ksocknal_startup(lnet_ni_t *ni);
 614 void ksocknal_shutdown(lnet_ni_t *ni);
 615 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
 616 int ksocknal_send(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
 617 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
 618                   int delayed, unsigned int niov,
 619                   struct kvec *iov, lnet_kiov_t *kiov,
 620                   unsigned int offset, unsigned int mlen, unsigned int rlen);
 621 int ksocknal_accept(lnet_ni_t *ni, struct socket *sock);
 622
 623 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
 624 extern ksock_peer_t *ksocknal_find_peer_locked(lnet_ni_t *ni, lnet_process_id_t id);
 625 extern ksock_peer_t *ksocknal_find_peer(lnet_ni_t *ni, lnet_process_id_t id);
 626 extern void ksocknal_peer_failed(ksock_peer_t *peer);
 627 extern int ksocknal_create_conn(lnet_ni_t *ni, ksock_route_t *route,
 628                                  struct socket *sock, int type);
 629 extern void ksocknal_close_conn_locked(ksock_conn_t *conn, int why);
 630 extern void ksocknal_terminate_conn(ksock_conn_t *conn);
 631 extern void ksocknal_destroy_conn(ksock_conn_t *conn);
 632 extern int  ksocknal_close_peer_conns_locked(ksock_peer_t *peer,
 633                                               __u32 ipaddr, int why);
 634 extern int ksocknal_close_conn_and_siblings(ksock_conn_t *conn, int why);
 635 extern int ksocknal_close_matching_conns(lnet_process_id_t id, __u32 ipaddr);
 636 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
 637                                                ksock_tx_t *tx, int nonblk);
 638
 639 extern int  ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
 640                                    lnet_process_id_t id);
 641 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
 642 extern void ksocknal_free_tx(ksock_tx_t *tx);
 643 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
 644 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
 645 extern void ksocknal_queue_tx_locked(ksock_tx_t *tx, ksock_conn_t *conn);
 646 extern void ksocknal_txlist_done(lnet_ni_t *ni, struct list_head *txlist,
 647                                   int error);
 648 extern void ksocknal_notify(lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
 649 extern void ksocknal_query(struct lnet_ni *ni, lnet_nid_t nid, unsigned long *when);
 650 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
 651 extern void ksocknal_thread_fini(void);
 652 extern void ksocknal_launch_all_connections_locked(ksock_peer_t *peer);
 653 extern ksock_route_t *ksocknal_find_connectable_route_locked(ksock_peer_t *peer);
 654 extern ksock_route_t *ksocknal_find_connecting_route_locked(ksock_peer_t *peer);
 655 extern int ksocknal_new_packet(ksock_conn_t *conn, int skip);
 656 extern int ksocknal_scheduler(void *arg);
 657 extern int ksocknal_connd(void *arg);
 658 extern int ksocknal_reaper(void *arg);
 659 extern int ksocknal_send_hello(lnet_ni_t *ni, ksock_conn_t *conn,
 660                                 lnet_nid_t peer_nid, ksock_hello_msg_t *hello);
 661 extern int ksocknal_recv_hello(lnet_ni_t *ni, ksock_conn_t *conn,
 662                                 ksock_hello_msg_t *hello, lnet_process_id_t *id,
 663                                 __u64 *incarnation);
 664 extern void ksocknal_read_callback(ksock_conn_t *conn);
 665 extern void ksocknal_write_callback(ksock_conn_t *conn);
 666
 667 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
 668 extern void ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn);
 669 extern void ksocknal_lib_set_callback(struct socket *sock,  ksock_conn_t *conn);
 670 extern void ksocknal_lib_reset_callback(struct socket *sock, ksock_conn_t *conn);
 671 extern void ksocknal_lib_push_conn(ksock_conn_t *conn);
 672 extern int ksocknal_lib_get_conn_addrs(ksock_conn_t *conn);
 673 extern int ksocknal_lib_setup_sock(struct socket *so);
 674 extern int ksocknal_lib_send_iov(ksock_conn_t *conn, ksock_tx_t *tx);
 675 extern int ksocknal_lib_send_kiov(ksock_conn_t *conn, ksock_tx_t *tx);
 676 extern void ksocknal_lib_eager_ack(ksock_conn_t *conn);
 677 extern int ksocknal_lib_recv_iov(ksock_conn_t *conn);
 678 extern int ksocknal_lib_recv_kiov(ksock_conn_t *conn);
 679 extern int ksocknal_lib_get_conn_tunables(ksock_conn_t *conn, int *txmem,
 680                                            int *rxmem, int *nagle);
 681
 682 extern int ksocknal_tunables_init(void);
 683
 684 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
 685
 686 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
 687 extern int ksocknal_lib_bind_thread_to_cpu(int id);
 688
 689 #endif /* _SOCKLND_SOCKLND_H_ */