1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lnet/include/lnet/lib-types.h
38 * Types used by the library side routines that do not need to be
39 * exposed to the user application
42 #ifndef __LNET_LIB_TYPES_H__
43 #define __LNET_LIB_TYPES_H__
45 #if defined(__linux__)
46 #include <lnet/linux/lib-types.h>
47 #elif defined(__APPLE__)
48 #include <lnet/darwin/lib-types.h>
49 #elif defined(__WINNT__)
50 #include <lnet/winnt/lib-types.h>
52 #error Unsupported Operating System
55 #include <libcfs/libcfs.h>
56 #include <libcfs/list.h>
57 #include <lnet/types.h>
59 #define WIRE_ATTR __attribute__((packed))
61 /* Packed version of lnet_process_id_t to transfer via network */
62 #include <libcfs/libcfs_pack.h>
65 lnet_pid_t pid; /* node id / process id */
66 } WIRE_ATTR lnet_process_id_packed_t;
68 /* The wire handle's interface cookie only matches one network interface in
69 * one epoch (i.e. new cookie when the interface restarts or the node
70 * reboots). The object cookie only matches one object on that interface
71 * during that object's lifetime (i.e. no cookie re-use). */
73 __u64 wh_interface_cookie;
74 __u64 wh_object_cookie;
75 } WIRE_ATTR lnet_handle_wire_t;
76 #include <libcfs/libcfs_unpack.h>
86 /* The variant fields of the portals message header are aligned on an 8
87 * byte boundary in the message header. Note that all types used in these
88 * wire structs MUST be fixed size and the smaller types are placed at the
90 #include <libcfs/libcfs_pack.h>
91 typedef struct lnet_ack {
92 lnet_handle_wire_t dst_wmd;
95 } WIRE_ATTR lnet_ack_t;
97 typedef struct lnet_put {
98 lnet_handle_wire_t ack_wmd;
103 } WIRE_ATTR lnet_put_t;
105 typedef struct lnet_get {
106 lnet_handle_wire_t return_wmd;
111 } WIRE_ATTR lnet_get_t;
113 typedef struct lnet_reply {
114 lnet_handle_wire_t dst_wmd;
115 } WIRE_ATTR lnet_reply_t;
117 typedef struct lnet_hello {
120 } WIRE_ATTR lnet_hello_t;
127 __u32 type; /* lnet_msg_type_t */
128 __u32 payload_length; /* payload data to follow */
129 /*<------__u64 aligned------->*/
137 } WIRE_ATTR lnet_hdr_t;
139 /* A HELLO message contains a magic number and protocol version
140 * code in the header's dest_nid, the peer's NID in the src_nid, and
141 * LNET_MSG_HELLO in the type field. All other common fields are zero
142 * (including payload_size; i.e. no payload).
143 * This is for use by byte-stream LNDs (e.g. TCP/IP) to check the peer is
144 * running the same protocol and to find out its NID. These LNDs should
145 * exchange HELLO messages when a connection is first established. Individual
146 * LNDs can put whatever else they fancy in lnet_hdr_t::msg.
149 __u32 magic; /* LNET_PROTO_TCP_MAGIC */
150 __u16 version_major; /* increment on incompatible change */
151 __u16 version_minor; /* increment on compatible change */
152 } WIRE_ATTR lnet_magicversion_t;
154 /* PROTO MAGIC for LNDs */
155 #define LNET_PROTO_IB_MAGIC 0x0be91b91
156 #define LNET_PROTO_RA_MAGIC 0x0be91b92
157 #define LNET_PROTO_QSW_MAGIC 0x0be91b93
158 #define LNET_PROTO_TCP_MAGIC 0xeebc0ded
159 #define LNET_PROTO_PTL_MAGIC 0x50746C4E /* 'PtlN' unique magic */
160 #define LNET_PROTO_MX_MAGIC 0x4d583130 /* 'MX10'! */
161 #define LNET_PROTO_ACCEPTOR_MAGIC 0xacce7100
162 #define LNET_PROTO_PING_MAGIC 0x70696E67 /* 'ping' */
164 /* Placeholder for a future "unified" protocol across all LNDs */
165 /* Current LNDs that receive a request with this magic will respond with a
166 * "stub" reply using their current protocol */
167 #define LNET_PROTO_MAGIC 0x45726963 /* ! */
170 #define LNET_PROTO_TCP_VERSION_MAJOR 1
171 #define LNET_PROTO_TCP_VERSION_MINOR 0
173 /* Acceptor connection request */
175 __u32 acr_magic; /* PTL_ACCEPTOR_PROTO_MAGIC */
176 __u32 acr_version; /* protocol version */
177 __u64 acr_nid; /* target NID */
178 } WIRE_ATTR lnet_acceptor_connreq_t;
179 #include <libcfs/libcfs_unpack.h>
181 #define LNET_PROTO_ACCEPTOR_VERSION 1
186 typedef struct lnet_msg {
187 cfs_list_t msg_activelist;
188 cfs_list_t msg_list; /* Q for credits/MD */
190 lnet_process_id_t msg_target;
193 unsigned int msg_vmflush:1; /* VM trying to free memory */
194 unsigned int msg_target_is_router:1; /* sending to a router */
195 unsigned int msg_routing:1; /* being forwarded */
196 unsigned int msg_ack:1; /* ack on finalize (PUT) */
197 unsigned int msg_sending:1; /* outgoing message */
198 unsigned int msg_receiving:1; /* being received */
199 unsigned int msg_delayed:1; /* had to Q for buffer or tx credit */
200 unsigned int msg_txcredit:1; /* taken an NI send credit */
201 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
202 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
203 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
204 unsigned int msg_onactivelist:1; /* on the activelist */
206 struct lnet_peer *msg_txpeer; /* peer I'm sending to */
207 struct lnet_peer *msg_rxpeer; /* peer I received from */
210 struct lnet_libmd *msg_md;
212 unsigned int msg_len;
213 unsigned int msg_wanted;
214 unsigned int msg_offset;
215 unsigned int msg_niov;
216 struct iovec *msg_iov;
217 lnet_kiov_t *msg_kiov;
224 typedef struct lnet_libhandle {
225 cfs_list_t lh_hash_chain;
229 #define lh_entry(ptr, type, member) \
230 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
232 typedef struct lnet_eq {
234 lnet_libhandle_t eq_lh;
235 lnet_seq_t eq_enq_seq;
236 lnet_seq_t eq_deq_seq;
237 unsigned int eq_size;
238 lnet_event_t *eq_events;
240 lnet_eq_handler_t eq_callback;
243 typedef struct lnet_me {
245 lnet_libhandle_t me_lh;
246 lnet_process_id_t me_match_id;
247 unsigned int me_portal;
249 __u64 me_ignore_bits;
250 lnet_unlink_t me_unlink;
251 struct lnet_libmd *me_md;
254 typedef struct lnet_libmd {
256 lnet_libhandle_t md_lh;
259 unsigned int md_offset;
260 unsigned int md_length;
261 unsigned int md_max_size;
264 unsigned int md_options;
265 unsigned int md_flags;
268 unsigned int md_niov; /* # frags */
270 struct iovec iov[LNET_MAX_IOV];
271 lnet_kiov_t kiov[LNET_MAX_IOV];
275 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
276 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
278 #ifdef LNET_USE_LIB_FREELIST
281 void *fl_objs; /* single contiguous array of objects */
282 int fl_nobjs; /* the number of them */
283 int fl_objsize; /* the size (including overhead) of each of them */
284 cfs_list_t fl_list; /* where they are enqueued */
289 cfs_list_t fo_list; /* enqueue on fl_list */
290 void *fo_contents; /* aligned contents */
295 /* info about peers we are trying to fail */
296 cfs_list_t tp_list; /* ln_test_peers */
297 lnet_nid_t tp_nid; /* matching nid */
298 unsigned int tp_threshold; /* # failures to simulate */
301 #define LNET_COOKIE_TYPE_MD 1
302 #define LNET_COOKIE_TYPE_ME 2
303 #define LNET_COOKIE_TYPE_EQ 3
304 #define LNET_COOKIE_TYPE_BITS 2
305 #define LNET_COOKIE_TYPES (1 << LNET_COOKIE_TYPE_BITS)
306 /* LNET_COOKIE_TYPES must be a power of 2, so the cookie type can be
307 * extracted by masking with (LNET_COOKIE_TYPES - 1) */
309 struct lnet_ni; /* forward ref */
311 typedef struct lnet_lnd
313 /* fields managed by portals */
314 cfs_list_t lnd_list; /* stash in the LND table */
315 int lnd_refcount; /* # active instances */
317 /* fields initialised by the LND */
318 unsigned int lnd_type;
320 int (*lnd_startup) (struct lnet_ni *ni);
321 void (*lnd_shutdown) (struct lnet_ni *ni);
322 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
324 /* In data movement APIs below, payload buffers are described as a set
325 * of 'niov' fragments which are...
327 * in virtual memory (struct iovec *iov != NULL)
329 * in pages (kernel only: plt_kiov_t *kiov != NULL).
330 * The LND may NOT overwrite these fragment descriptors.
331 * An 'offset' and may specify a byte offset within the set of
332 * fragments to start from
335 /* Start sending a preformatted message. 'private' is NULL for PUT and
336 * GET messages; otherwise this is a response to an incoming message
337 * and 'private' is the 'private' passed to lnet_parse(). Return
338 * non-zero for immediate failure, otherwise complete later with
340 int (*lnd_send)(struct lnet_ni *ni, void *private, lnet_msg_t *msg);
342 /* Start receiving 'mlen' bytes of payload data, skipping the following
343 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
344 * lnet_parse(). Return non-zero for immedaite failure, otherwise
345 * complete later with lnet_finalize(). This also gives back a receive
346 * credit if the LND does flow control. */
347 int (*lnd_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
348 int delayed, unsigned int niov,
349 struct iovec *iov, lnet_kiov_t *kiov,
350 unsigned int offset, unsigned int mlen, unsigned int rlen);
352 /* lnet_parse() has had to delay processing of this message
353 * (e.g. waiting for a forwarding buffer or send credits). Give the
354 * LND a chance to free urgently needed resources. If called, return 0
355 * for success and do NOT give back a receive credit; that has to wait
356 * until lnd_recv() gets called. On failure return < 0 and
357 * release resources; lnd_recv() will not be called. */
358 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
359 void **new_privatep);
361 /* notification of peer health */
362 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
364 /* query of peer aliveness */
365 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, cfs_time_t *when);
367 #if defined(__KERNEL__) || defined(HAVE_LIBPTHREAD)
368 /* accept a new connection */
369 int (*lnd_accept)(struct lnet_ni *ni, cfs_socket_t *sock);
373 /* wait for something to happen */
374 void (*lnd_wait)(struct lnet_ni *ni, int milliseconds);
376 /* ensure non-RDMA messages can be received outside liblustre */
377 int (*lnd_setasync)(struct lnet_ni *ni, lnet_process_id_t id, int nasync);
381 #define LNET_NI_STATUS_UP 0x15aac0de
382 #define LNET_NI_STATUS_DOWN 0xdeadface
383 #define LNET_NI_STATUS_INVALID 0x00000000
388 } WIRE_ATTR lnet_ni_status_t;
390 #define LNET_MAX_INTERFACES 16
392 typedef struct lnet_ni {
393 cfs_list_t ni_list; /* chain on ln_nis */
394 cfs_list_t ni_txq; /* messages waiting for tx credits */
395 int ni_maxtxcredits; /* # tx credits */
396 int ni_txcredits; /* # tx credits free */
397 int ni_mintxcredits; /* lowest it's been */
398 int ni_peertxcredits; /* # per-peer send credits */
399 int ni_peerrtrcredits; /* # per-peer router buffer credits */
400 int ni_peertimeout; /* seconds to consider peer dead */
401 lnet_nid_t ni_nid; /* interface's NID */
402 void *ni_data; /* instance-specific data */
403 lnd_t *ni_lnd; /* procedural interface */
404 int ni_refcount; /* reference count */
405 cfs_time_t ni_last_alive; /* when I was last alive */
406 lnet_ni_status_t *ni_status; /* my health status */
407 char *ni_interfaces[LNET_MAX_INTERFACES]; /* equivalent interfaces to use */
410 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
411 #define LNET_PROTO_PING_VERSION 2
412 #define LNET_PROTO_PING_VERSION1 1
418 lnet_ni_status_t pi_ni[0];
419 } WIRE_ATTR lnet_ping_info_t;
421 /* router checker data, per router */
422 #define LNET_MAX_RTR_NIS 16
423 #define LNET_PINGINFO_SIZE offsetof(lnet_ping_info_t, pi_ni[LNET_MAX_RTR_NIS])
425 cfs_list_t rcd_list; /* chain on the_lnet.ln_zombie_rcd */
426 lnet_handle_md_t rcd_mdh; /* ping buffer MD */
427 lnet_ping_info_t *rcd_pinginfo; /* ping buffer */
430 typedef struct lnet_peer {
431 cfs_list_t lp_hashlist; /* chain on peer hash */
432 cfs_list_t lp_txq; /* messages blocking for tx credits */
433 cfs_list_t lp_rtrq; /* messages blocking for router credits */
434 cfs_list_t lp_rtr_list; /* chain on router list */
435 int lp_txcredits; /* # tx credits available */
436 int lp_mintxcredits; /* low water mark */
437 int lp_rtrcredits; /* # router credits */
438 int lp_minrtrcredits; /* low water mark */
439 unsigned int lp_alive:1; /* alive/dead? */
440 unsigned int lp_notify:1; /* notification outstanding? */
441 unsigned int lp_notifylnd:1; /* outstanding notification for LND? */
442 unsigned int lp_notifying:1; /* some thread is handling notification */
443 unsigned int lp_ping_notsent; /* SEND event outstanding from ping */
444 int lp_alive_count; /* # times router went dead<->alive */
445 long lp_txqnob; /* bytes queued for sending */
446 cfs_time_t lp_timestamp; /* time of last aliveness news */
447 cfs_time_t lp_ping_timestamp; /* time of last ping attempt */
448 cfs_time_t lp_ping_deadline; /* != 0 if ping reply expected */
449 cfs_time_t lp_last_alive; /* when I was last alive */
450 cfs_time_t lp_last_query; /* when lp_ni was queried last time */
451 lnet_ni_t *lp_ni; /* interface peer is on */
452 lnet_nid_t lp_nid; /* peer's NID */
453 int lp_refcount; /* # refs */
454 int lp_rtr_refcount; /* # refs from lnet_route_t::lr_gateway */
455 lnet_rc_data_t *lp_rcd; /* router checker state */
458 #define lnet_peer_aliveness_enabled(lp) ((lp)->lp_ni->ni_peertimeout > 0)
461 cfs_list_t lr_list; /* chain on net */
462 lnet_peer_t *lr_gateway; /* router node */
463 unsigned int lr_hops; /* how far I am */
467 cfs_list_t lrn_list; /* chain on ln_remote_nets */
468 cfs_list_t lrn_routes; /* routes to me */
469 __u32 lrn_net; /* my net number */
473 cfs_list_t rbp_bufs; /* my free buffer pool */
474 cfs_list_t rbp_msgs; /* messages blocking for a buffer */
475 int rbp_npages; /* # pages in each buffer */
476 int rbp_nbuffers; /* # buffers */
477 int rbp_credits; /* # free buffers / blocked messages */
478 int rbp_mincredits; /* low water mark */
482 cfs_list_t rb_list; /* chain on rbp_bufs */
483 lnet_rtrbufpool_t *rb_pool; /* owning pool */
484 lnet_kiov_t rb_kiov[0]; /* the buffer space */
487 #include <libcfs/libcfs_pack.h>
500 } WIRE_ATTR lnet_counters_t;
501 #include <libcfs/libcfs_unpack.h>
503 #define LNET_PEER_HASHSIZE 503 /* prime! */
505 #define LNET_NRBPOOLS 3 /* # different router buffer pools */
507 /* Options for lnet_portal_t::ptl_options */
508 #define LNET_PTL_LAZY (1 << 0)
509 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
510 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
512 /* ME hash of RDMA portal */
513 #define LNET_PORTAL_HASH_BITS 8
514 #define LNET_PORTAL_HASH_SIZE (1 << LNET_PORTAL_HASH_BITS)
517 cfs_list_t *ptl_mhash; /* match hash */
518 cfs_list_t ptl_mlist; /* match list */
519 cfs_list_t ptl_msgq; /* messages blocking for MD */
520 __u64 ptl_ml_version; /* validity stamp, only changed for new attached MD */
521 __u64 ptl_msgq_version; /* validity stamp */
522 unsigned int ptl_options;
525 /* Router Checker states */
526 #define LNET_RC_STATE_SHUTDOWN 0 /* not started */
527 #define LNET_RC_STATE_RUNNING 1 /* started up OK */
528 #define LNET_RC_STATE_STOPTHREAD 2 /* telling thread to stop */
529 #define LNET_RC_STATE_UNLINKING 3 /* unlinking RC MD */
530 #define LNET_RC_STATE_UNLINKED 4 /* RC's MD has been unlinked */
534 /* Stuff initialised at LNetInit() */
535 int ln_init; /* LNetInit() called? */
536 int ln_refcount; /* LNetNIInit/LNetNIFini counter */
537 int ln_niinit_self; /* Have I called LNetNIInit myself? */
539 cfs_list_t ln_lnds; /* registered LNDs */
542 cfs_spinlock_t ln_lock;
543 cfs_waitq_t ln_waitq;
544 cfs_semaphore_t ln_api_mutex;
545 cfs_semaphore_t ln_lnd_mutex;
547 # ifndef HAVE_LIBPTHREAD
552 pthread_cond_t ln_cond;
553 pthread_mutex_t ln_lock;
554 pthread_mutex_t ln_api_mutex;
555 pthread_mutex_t ln_lnd_mutex;
559 /* Stuff initialised at LNetNIInit() */
561 int ln_shutdown; /* shutdown in progress */
562 int ln_nportals; /* # portals */
563 lnet_portal_t *ln_portals; /* the vector of portals */
565 lnet_pid_t ln_pid; /* requested pid */
567 cfs_list_t ln_nis; /* LND instances */
568 lnet_ni_t *ln_loni; /* the loopback NI */
569 lnet_ni_t *ln_eqwaitni; /* NI to wait for events in */
570 cfs_list_t ln_zombie_nis; /* dying LND instances */
571 int ln_nzombie_nis; /* # of NIs to wait for */
573 cfs_list_t ln_remote_nets; /* remote networks with routes to them */
574 __u64 ln_remote_nets_version; /* validity stamp */
576 cfs_list_t ln_routers; /* list of all known routers */
577 __u64 ln_routers_version; /* validity stamp */
579 cfs_list_t *ln_peer_hash; /* NID->peer hash */
580 int ln_npeers; /* # peers extant */
581 int ln_peertable_version; /* /proc validity stamp */
583 int ln_routing; /* am I a router? */
584 lnet_rtrbufpool_t ln_rtrpools[LNET_NRBPOOLS]; /* router buffer pools */
586 int ln_lh_hash_size; /* size of lib handle hash table */
587 cfs_list_t *ln_lh_hash_table; /* all extant lib handles, this interface */
588 __u64 ln_next_object_cookie; /* cookie generator */
589 __u64 ln_interface_cookie; /* uniquely identifies this ni in this epoch */
591 char *ln_network_tokens; /* space for network names */
592 int ln_network_tokens_nob;
594 int ln_testprotocompat; /* test protocol compatibility flags */
596 cfs_list_t ln_finalizeq; /* msgs waiting to complete finalizing */
598 void **ln_finalizers; /* threads doing finalization */
599 int ln_nfinalizers; /* max # threads finalizing */
603 cfs_list_t ln_test_peers; /* failure simulation */
605 lnet_handle_md_t ln_ping_target_md;
606 lnet_handle_eq_t ln_ping_target_eq;
607 lnet_ping_info_t *ln_ping_info;
610 cfs_semaphore_t ln_rc_signal; /* serialise startup/shutdown */
612 int ln_rc_state; /* router checker startup/shutdown state */
613 lnet_handle_eq_t ln_rc_eqh; /* router checker's event queue */
614 lnet_handle_md_t ln_rc_mdh;
615 cfs_list_t ln_zombie_rcd;
617 #ifdef LNET_USE_LIB_FREELIST
618 lnet_freelist_t ln_free_mes;
619 lnet_freelist_t ln_free_msgs;
620 lnet_freelist_t ln_free_mds;
621 lnet_freelist_t ln_free_eqs;
623 cfs_list_t ln_active_msgs;
624 cfs_list_t ln_active_mds;
625 cfs_list_t ln_active_eqs;
627 lnet_counters_t ln_counters;
630 /* Temporary workaround to allow uOSS and test programs force
631 * server mode in userspace. The only place where we use it is
632 * lnet_prepare(). The only way to turn this flag on is to
633 * call lnet_server_mode() */
635 int ln_server_mode_flag;