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