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