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 2008 Sun Microsystems, Inc. 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 /* The wire handle's interface cookie only matches one network interface in
62 * one epoch (i.e. new cookie when the interface restarts or the node
63 * reboots). The object cookie only matches one object on that interface
64 * during that object's lifetime (i.e. no cookie re-use). */
65 #include <libcfs/libcfs_pack.h>
67 __u64 wh_interface_cookie;
68 __u64 wh_object_cookie;
69 } WIRE_ATTR lnet_handle_wire_t;
70 #include <libcfs/libcfs_unpack.h>
80 /* The variant fields of the portals message header are aligned on an 8
81 * byte boundary in the message header. Note that all types used in these
82 * wire structs MUST be fixed size and the smaller types are placed at the
84 #include <libcfs/libcfs_pack.h>
85 typedef struct lnet_ack {
86 lnet_handle_wire_t dst_wmd;
89 } WIRE_ATTR lnet_ack_t;
91 typedef struct lnet_put {
92 lnet_handle_wire_t ack_wmd;
97 } WIRE_ATTR lnet_put_t;
99 typedef struct lnet_get {
100 lnet_handle_wire_t return_wmd;
105 } WIRE_ATTR lnet_get_t;
107 typedef struct lnet_reply {
108 lnet_handle_wire_t dst_wmd;
109 } WIRE_ATTR lnet_reply_t;
111 typedef struct lnet_hello {
114 } WIRE_ATTR lnet_hello_t;
121 __u32 type; /* lnet_msg_type_t */
122 __u32 payload_length; /* payload data to follow */
123 /*<------__u64 aligned------->*/
131 } WIRE_ATTR lnet_hdr_t;
133 /* A HELLO message contains a magic number and protocol version
134 * code in the header's dest_nid, the peer's NID in the src_nid, and
135 * LNET_MSG_HELLO in the type field. All other common fields are zero
136 * (including payload_size; i.e. no payload).
137 * This is for use by byte-stream LNDs (e.g. TCP/IP) to check the peer is
138 * running the same protocol and to find out its NID. These LNDs should
139 * exchange HELLO messages when a connection is first established. Individual
140 * LNDs can put whatever else they fancy in lnet_hdr_t::msg.
143 __u32 magic; /* LNET_PROTO_TCP_MAGIC */
144 __u16 version_major; /* increment on incompatible change */
145 __u16 version_minor; /* increment on compatible change */
146 } WIRE_ATTR lnet_magicversion_t;
148 /* PROTO MAGIC for LNDs */
149 #define LNET_PROTO_IB_MAGIC 0x0be91b91
150 #define LNET_PROTO_OPENIB_MAGIC LNET_PROTO_IB_MAGIC
151 #define LNET_PROTO_IIB_MAGIC LNET_PROTO_IB_MAGIC
152 #define LNET_PROTO_VIB_MAGIC LNET_PROTO_IB_MAGIC
153 #define LNET_PROTO_RA_MAGIC 0x0be91b92
154 #define LNET_PROTO_QSW_MAGIC 0x0be91b93
155 #define LNET_PROTO_TCP_MAGIC 0xeebc0ded
156 #define LNET_PROTO_PTL_MAGIC 0x50746C4E /* 'PtlN' unique magic */
157 #define LNET_PROTO_GM_MAGIC 0x6d797269 /* 'myri'! */
158 #define LNET_PROTO_MX_MAGIC 0x4d583130 /* 'MX10'! */
159 #define LNET_PROTO_ACCEPTOR_MAGIC 0xacce7100
160 #define LNET_PROTO_PING_MAGIC 0x70696E67 /* 'ping' */
162 /* Placeholder for a future "unified" protocol across all LNDs */
163 /* Current LNDs that receive a request with this magic will respond with a
164 * "stub" reply using their current protocol */
165 #define LNET_PROTO_MAGIC 0x45726963 /* ! */
168 #define LNET_PROTO_TCP_VERSION_MAJOR 1
169 #define LNET_PROTO_TCP_VERSION_MINOR 0
171 /* Acceptor connection request */
173 __u32 acr_magic; /* PTL_ACCEPTOR_PROTO_MAGIC */
174 __u32 acr_version; /* protocol version */
175 __u64 acr_nid; /* target NID */
176 } WIRE_ATTR lnet_acceptor_connreq_t;
177 #include <libcfs/libcfs_unpack.h>
179 #define LNET_PROTO_ACCEPTOR_VERSION 1
184 typedef struct lnet_msg {
185 struct list_head msg_activelist;
186 struct list_head msg_list; /* Q for credits/MD */
188 lnet_process_id_t msg_target;
191 unsigned int msg_target_is_router:1; /* sending to a router */
192 unsigned int msg_routing:1; /* being forwarded */
193 unsigned int msg_ack:1; /* ack on finalize (PUT) */
194 unsigned int msg_sending:1; /* outgoing message */
195 unsigned int msg_receiving:1; /* being received */
196 unsigned int msg_delayed:1; /* had to Q for buffer or tx credit */
197 unsigned int msg_txcredit:1; /* taken an NI send credit */
198 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
199 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
200 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
201 unsigned int msg_onactivelist:1; /* on the activelist */
203 struct lnet_peer *msg_txpeer; /* peer I'm sending to */
204 struct lnet_peer *msg_rxpeer; /* peer I received from */
207 struct lnet_libmd *msg_md;
209 unsigned int msg_len;
210 unsigned int msg_wanted;
211 unsigned int msg_offset;
212 unsigned int msg_niov;
213 struct iovec *msg_iov;
214 lnet_kiov_t *msg_kiov;
221 typedef struct lnet_libhandle {
222 struct list_head lh_hash_chain;
226 #define lh_entry(ptr, type, member) \
227 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
229 typedef struct lnet_eq {
230 struct list_head eq_list;
231 lnet_libhandle_t eq_lh;
232 lnet_seq_t eq_enq_seq;
233 lnet_seq_t eq_deq_seq;
234 unsigned int eq_size;
235 lnet_event_t *eq_events;
237 lnet_eq_handler_t eq_callback;
240 typedef struct lnet_me {
241 struct list_head me_list;
242 lnet_libhandle_t me_lh;
243 lnet_process_id_t me_match_id;
244 unsigned int me_portal;
246 __u64 me_ignore_bits;
247 lnet_unlink_t me_unlink;
248 struct lnet_libmd *me_md;
251 typedef struct lnet_libmd {
252 struct list_head md_list;
253 lnet_libhandle_t md_lh;
256 unsigned int md_offset;
257 unsigned int md_length;
258 unsigned int md_max_size;
261 unsigned int md_options;
262 unsigned int md_flags;
265 unsigned int md_niov; /* # frags */
267 struct iovec iov[LNET_MAX_IOV];
268 lnet_kiov_t kiov[LNET_MAX_IOV];
272 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
273 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
275 #ifdef LNET_USE_LIB_FREELIST
278 void *fl_objs; /* single contiguous array of objects */
279 int fl_nobjs; /* the number of them */
280 int fl_objsize; /* the size (including overhead) of each of them */
281 struct list_head fl_list; /* where they are enqueued */
286 struct list_head fo_list; /* enqueue on fl_list */
287 void *fo_contents; /* aligned contents */
292 /* info about peers we are trying to fail */
293 struct list_head tp_list; /* ln_test_peers */
294 lnet_nid_t tp_nid; /* matching nid */
295 unsigned int tp_threshold; /* # failures to simulate */
298 #define LNET_COOKIE_TYPE_MD 1
299 #define LNET_COOKIE_TYPE_ME 2
300 #define LNET_COOKIE_TYPE_EQ 3
301 #define LNET_COOKIE_TYPES 4
302 /* LNET_COOKIE_TYPES must be a power of 2, so the cookie type can be
303 * extracted by masking with (LNET_COOKIE_TYPES - 1) */
305 struct lnet_ni; /* forward ref */
307 typedef struct lnet_lnd
309 /* fields managed by portals */
310 struct list_head lnd_list; /* stash in the LND table */
311 int lnd_refcount; /* # active instances */
313 /* fields initialised by the LND */
314 unsigned int lnd_type;
316 int (*lnd_startup) (struct lnet_ni *ni);
317 void (*lnd_shutdown) (struct lnet_ni *ni);
318 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
320 /* In data movement APIs below, payload buffers are described as a set
321 * of 'niov' fragments which are...
323 * in virtual memory (struct iovec *iov != NULL)
325 * in pages (kernel only: plt_kiov_t *kiov != NULL).
326 * The LND may NOT overwrite these fragment descriptors.
327 * An 'offset' and may specify a byte offset within the set of
328 * fragments to start from
331 /* Start sending a preformatted message. 'private' is NULL for PUT and
332 * GET messages; otherwise this is a response to an incoming message
333 * and 'private' is the 'private' passed to lnet_parse(). Return
334 * non-zero for immediate failure, otherwise complete later with
336 int (*lnd_send)(struct lnet_ni *ni, void *private, lnet_msg_t *msg);
338 /* Start receiving 'mlen' bytes of payload data, skipping the following
339 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
340 * lnet_parse(). Return non-zero for immedaite failure, otherwise
341 * complete later with lnet_finalize(). This also gives back a receive
342 * credit if the LND does flow control. */
343 int (*lnd_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
344 int delayed, unsigned int niov,
345 struct iovec *iov, lnet_kiov_t *kiov,
346 unsigned int offset, unsigned int mlen, unsigned int rlen);
348 /* lnet_parse() has had to delay processing of this message
349 * (e.g. waiting for a forwarding buffer or send credits). Give the
350 * LND a chance to free urgently needed resources. If called, return 0
351 * for success and do NOT give back a receive credit; that has to wait
352 * until lnd_recv() gets called. On failure return < 0 and
353 * release resources; lnd_recv() will not be called. */
354 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
355 void **new_privatep);
357 /* notification of peer health */
358 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
360 /* query of peer aliveness */
361 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, time_t *when);
363 #if defined(__KERNEL__) || defined(HAVE_LIBPTHREAD)
364 /* accept a new connection */
365 int (*lnd_accept)(struct lnet_ni *ni, cfs_socket_t *sock);
369 /* wait for something to happen */
370 void (*lnd_wait)(struct lnet_ni *ni, int milliseconds);
372 /* ensure non-RDMA messages can be received outside liblustre */
373 int (*lnd_setasync)(struct lnet_ni *ni, lnet_process_id_t id, int nasync);
377 #define LNET_MAX_INTERFACES 16
379 typedef struct lnet_ni {
380 struct list_head ni_list; /* chain on ln_nis */
381 struct list_head ni_txq; /* messages waiting for tx credits */
382 int ni_maxtxcredits; /* # tx credits */
383 int ni_txcredits; /* # tx credits free */
384 int ni_mintxcredits; /* lowest it's been */
385 int ni_peertxcredits; /* # per-peer send credits */
386 int ni_peerrtrcredits; /* # per-peer router buffer credits */
387 int ni_peertimeout; /* seconds to consider peer dead */
388 lnet_nid_t ni_nid; /* interface's NID */
389 void *ni_data; /* instance-specific data */
390 lnd_t *ni_lnd; /* procedural interface */
391 int ni_refcount; /* reference count */
392 char *ni_interfaces[LNET_MAX_INTERFACES]; /* equivalent interfaces to use */
395 typedef struct lnet_peer {
396 struct list_head lp_hashlist; /* chain on peer hash */
397 struct list_head lp_txq; /* messages blocking for tx credits */
398 struct list_head lp_rtrq; /* messages blocking for router credits */
399 struct list_head lp_rtr_list; /* chain on router list */
400 int lp_txcredits; /* # tx credits available */
401 int lp_mintxcredits; /* low water mark */
402 int lp_rtrcredits; /* # router credits */
403 int lp_minrtrcredits; /* low water mark */
404 unsigned int lp_alive:1; /* alive/dead? */
405 unsigned int lp_notify:1; /* notification outstanding? */
406 unsigned int lp_notifylnd:1; /* outstanding notification for LND? */
407 unsigned int lp_notifying:1; /* some thread is handling notification */
408 unsigned int lp_ping_notsent; /* SEND event outstanding from ping */
409 int lp_alive_count; /* # times router went dead<->alive */
410 long lp_txqnob; /* bytes queued for sending */
411 time_t lp_timestamp; /* time of last aliveness news */
412 time_t lp_last_alive; /* when I was last alive */
413 time_t lp_last_query; /* when LND was queried last time */
414 time_t lp_ping_timestamp; /* time of last ping attempt */
415 time_t lp_ping_deadline; /* != 0 if ping reply expected */
416 lnet_ni_t *lp_ni; /* interface peer is on */
417 lnet_nid_t lp_nid; /* peer's NID */
418 int lp_refcount; /* # refs */
419 int lp_rtr_refcount; /* # refs from lnet_route_t::lr_gateway */
423 struct list_head lr_list; /* chain on net */
424 lnet_peer_t *lr_gateway; /* router node */
428 struct list_head lrn_list; /* chain on ln_remote_nets */
429 struct list_head lrn_routes; /* routes to me */
430 __u32 lrn_net; /* my net number */
431 unsigned int lrn_hops; /* how far I am */
435 struct list_head rbp_bufs; /* my free buffer pool */
436 struct list_head rbp_msgs; /* messages blocking for a buffer */
437 int rbp_npages; /* # pages in each buffer */
438 int rbp_nbuffers; /* # buffers */
439 int rbp_credits; /* # free buffers / blocked messages */
440 int rbp_mincredits; /* low water mark */
444 struct list_head rb_list; /* chain on rbp_bufs */
445 lnet_rtrbufpool_t *rb_pool; /* owning pool */
446 lnet_kiov_t rb_kiov[0]; /* the buffer space */
449 #include <libcfs/libcfs_pack.h>
462 } WIRE_ATTR lnet_counters_t;
463 #include <libcfs/libcfs_unpack.h>
465 #define LNET_PEER_HASHSIZE 503 /* prime! */
467 #define LNET_NRBPOOLS 3 /* # different router buffer pools */
469 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
470 #define LNET_PROTO_PING_VERSION 1
476 lnet_nid_t pi_nid[0];
477 } WIRE_ATTR lnet_ping_info_t;
479 /* Options for lnet_portal_t::ptl_options */
480 #define LNET_PTL_LAZY (1 << 0)
482 struct list_head ptl_ml; /* match list */
483 struct list_head ptl_msgq; /* messages blocking for MD */
484 __u64 ptl_ml_version; /* validity stamp, only changed for new attached MD */
485 __u64 ptl_msgq_version; /* validity stamp */
486 unsigned int ptl_options;
490 /* < 0 == startup error */
491 #define LNET_RC_STATE_SHUTDOWN 0 /* not started */
492 #define LNET_RC_STATE_RUNNING 1 /* started up OK */
493 #define LNET_RC_STATE_STOPTHREAD 2 /* telling thread to stop */
494 #define LNET_RC_STATE_UNLINKING 3 /* unlinking RC MD */
495 #define LNET_RC_STATE_UNLINKED 4 /* RC's MD has been unlinked */
499 /* Stuff initialised at LNetInit() */
500 int ln_init; /* LNetInit() called? */
501 int ln_refcount; /* LNetNIInit/LNetNIFini counter */
502 int ln_niinit_self; /* Have I called LNetNIInit myself? */
504 struct list_head ln_lnds; /* registered LNDs */
508 cfs_waitq_t ln_waitq;
509 struct semaphore ln_api_mutex;
510 struct semaphore ln_lnd_mutex;
512 # ifndef HAVE_LIBPTHREAD
517 pthread_cond_t ln_cond;
518 pthread_mutex_t ln_lock;
519 pthread_mutex_t ln_api_mutex;
520 pthread_mutex_t ln_lnd_mutex;
524 /* Stuff initialised at LNetNIInit() */
526 int ln_shutdown; /* shutdown in progress */
527 int ln_nportals; /* # portals */
528 lnet_portal_t *ln_portals; /* the vector of portals */
530 lnet_pid_t ln_pid; /* requested pid */
532 struct list_head ln_nis; /* LND instances */
533 lnet_ni_t *ln_loni; /* the loopback NI */
534 lnet_ni_t *ln_eqwaitni; /* NI to wait for events in */
535 struct list_head ln_zombie_nis; /* dying LND instances */
536 int ln_nzombie_nis; /* # of NIs to wait for */
538 struct list_head ln_remote_nets; /* remote networks with routes to them */
539 __u64 ln_remote_nets_version; /* validity stamp */
541 struct list_head ln_routers; /* list of all known routers */
542 __u64 ln_routers_version; /* validity stamp */
544 struct list_head *ln_peer_hash; /* NID->peer hash */
545 int ln_npeers; /* # peers extant */
546 int ln_peertable_version; /* /proc validity stamp */
548 int ln_routing; /* am I a router? */
549 lnet_rtrbufpool_t ln_rtrpools[LNET_NRBPOOLS]; /* router buffer pools */
551 int ln_lh_hash_size; /* size of lib handle hash table */
552 struct list_head *ln_lh_hash_table; /* all extant lib handles, this interface */
553 __u64 ln_next_object_cookie; /* cookie generator */
554 __u64 ln_interface_cookie; /* uniquely identifies this ni in this epoch */
556 char *ln_network_tokens; /* space for network names */
557 int ln_network_tokens_nob;
559 int ln_testprotocompat; /* test protocol compatibility flags */
561 struct list_head ln_finalizeq; /* msgs waiting to complete finalizing */
563 void **ln_finalizers; /* threads doing finalization */
564 int ln_nfinalizers; /* max # threads finalizing */
568 struct list_head ln_test_peers; /* failure simulation */
570 lnet_handle_md_t ln_ping_target_md;
571 lnet_handle_eq_t ln_ping_target_eq;
572 lnet_ping_info_t *ln_ping_info;
575 int ln_rc_state; /* router checker startup/shutdown state */
576 struct semaphore ln_rc_signal; /* serialise startup/shutdown */
577 lnet_handle_eq_t ln_rc_eqh; /* router checker's event queue */
580 #ifdef LNET_USE_LIB_FREELIST
581 lnet_freelist_t ln_free_mes;
582 lnet_freelist_t ln_free_msgs;
583 lnet_freelist_t ln_free_mds;
584 lnet_freelist_t ln_free_eqs;
586 struct list_head ln_active_msgs;
587 struct list_head ln_active_mds;
588 struct list_head ln_active_eqs;
590 lnet_counters_t ln_counters;
593 /* Temporary workaround to allow uOSS and test programs force
594 * server mode in userspace. The only place where we use it is
595 * lnet_prepare(). The only way to turn this flag on is to
596 * call lnet_server_mode() */
598 int ln_server_mode_flag;