4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * lnet/include/lnet/lib-types.h
36 * Types used by the library side routines that do not need to be
37 * exposed to the user application
40 #ifndef __LNET_LIB_TYPES_H__
41 #define __LNET_LIB_TYPES_H__
43 #if defined(__linux__)
44 #include <lnet/linux/lib-types.h>
45 #elif defined(__APPLE__)
46 #include <lnet/darwin/lib-types.h>
47 #elif defined(__WINNT__)
48 #include <lnet/winnt/lib-types.h>
50 #error Unsupported Operating System
53 #include <libcfs/libcfs.h>
54 #include <libcfs/list.h>
55 #include <lnet/types.h>
57 #define WIRE_ATTR __attribute__((packed))
59 /* Packed version of lnet_process_id_t to transfer via network */
60 #include <libcfs/libcfs_pack.h>
63 lnet_pid_t pid; /* node id / process id */
64 } WIRE_ATTR lnet_process_id_packed_t;
66 /* The wire handle's interface cookie only matches one network interface in
67 * one epoch (i.e. new cookie when the interface restarts or the node
68 * reboots). The object cookie only matches one object on that interface
69 * during that object's lifetime (i.e. no cookie re-use). */
71 __u64 wh_interface_cookie;
72 __u64 wh_object_cookie;
73 } WIRE_ATTR lnet_handle_wire_t;
74 #include <libcfs/libcfs_unpack.h>
84 /* The variant fields of the portals message header are aligned on an 8
85 * byte boundary in the message header. Note that all types used in these
86 * wire structs MUST be fixed size and the smaller types are placed at the
88 #include <libcfs/libcfs_pack.h>
89 typedef struct lnet_ack {
90 lnet_handle_wire_t dst_wmd;
93 } WIRE_ATTR lnet_ack_t;
95 typedef struct lnet_put {
96 lnet_handle_wire_t ack_wmd;
101 } WIRE_ATTR lnet_put_t;
103 typedef struct lnet_get {
104 lnet_handle_wire_t return_wmd;
109 } WIRE_ATTR lnet_get_t;
111 typedef struct lnet_reply {
112 lnet_handle_wire_t dst_wmd;
113 } WIRE_ATTR lnet_reply_t;
115 typedef struct lnet_hello {
118 } WIRE_ATTR lnet_hello_t;
125 __u32 type; /* lnet_msg_type_t */
126 __u32 payload_length; /* payload data to follow */
127 /*<------__u64 aligned------->*/
135 } WIRE_ATTR lnet_hdr_t;
137 /* A HELLO message contains a magic number and protocol version
138 * code in the header's dest_nid, the peer's NID in the src_nid, and
139 * LNET_MSG_HELLO in the type field. All other common fields are zero
140 * (including payload_size; i.e. no payload).
141 * This is for use by byte-stream LNDs (e.g. TCP/IP) to check the peer is
142 * running the same protocol and to find out its NID. These LNDs should
143 * exchange HELLO messages when a connection is first established. Individual
144 * LNDs can put whatever else they fancy in lnet_hdr_t::msg.
147 __u32 magic; /* LNET_PROTO_TCP_MAGIC */
148 __u16 version_major; /* increment on incompatible change */
149 __u16 version_minor; /* increment on compatible change */
150 } WIRE_ATTR lnet_magicversion_t;
152 /* PROTO MAGIC for LNDs */
153 #define LNET_PROTO_IB_MAGIC 0x0be91b91
154 #define LNET_PROTO_RA_MAGIC 0x0be91b92
155 #define LNET_PROTO_QSW_MAGIC 0x0be91b93
156 #define LNET_PROTO_GNI_MAGIC 0xb00fbabe /* ask Kim */
157 #define LNET_PROTO_TCP_MAGIC 0xeebc0ded
158 #define LNET_PROTO_PTL_MAGIC 0x50746C4E /* 'PtlN' unique magic */
159 #define LNET_PROTO_MX_MAGIC 0x4d583130 /* 'MX10'! */
160 #define LNET_PROTO_ACCEPTOR_MAGIC 0xacce7100
161 #define LNET_PROTO_PING_MAGIC 0x70696E67 /* 'ping' */
163 /* Placeholder for a future "unified" protocol across all LNDs */
164 /* Current LNDs that receive a request with this magic will respond with a
165 * "stub" reply using their current protocol */
166 #define LNET_PROTO_MAGIC 0x45726963 /* ! */
169 #define LNET_PROTO_TCP_VERSION_MAJOR 1
170 #define LNET_PROTO_TCP_VERSION_MINOR 0
172 /* Acceptor connection request */
174 __u32 acr_magic; /* PTL_ACCEPTOR_PROTO_MAGIC */
175 __u32 acr_version; /* protocol version */
176 __u64 acr_nid; /* target NID */
177 } WIRE_ATTR lnet_acceptor_connreq_t;
178 #include <libcfs/libcfs_unpack.h>
180 #define LNET_PROTO_ACCEPTOR_VERSION 1
185 typedef struct lnet_msg {
186 cfs_list_t msg_activelist;
187 cfs_list_t msg_list; /* Q for credits/MD */
189 lnet_process_id_t msg_target;
192 unsigned int msg_vmflush:1; /* VM trying to free memory */
193 unsigned int msg_target_is_router:1; /* sending to a router */
194 unsigned int msg_routing:1; /* being forwarded */
195 unsigned int msg_ack:1; /* ack on finalize (PUT) */
196 unsigned int msg_sending:1; /* outgoing message */
197 unsigned int msg_receiving:1; /* being received */
198 unsigned int msg_delayed:1; /* had to Q for buffer or tx credit */
199 unsigned int msg_txcredit:1; /* taken an NI send credit */
200 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
201 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
202 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
203 unsigned int msg_onactivelist:1; /* on the activelist */
205 struct lnet_peer *msg_txpeer; /* peer I'm sending to */
206 struct lnet_peer *msg_rxpeer; /* peer I received from */
209 struct lnet_libmd *msg_md;
211 unsigned int msg_len;
212 unsigned int msg_wanted;
213 unsigned int msg_offset;
214 unsigned int msg_niov;
215 struct iovec *msg_iov;
216 lnet_kiov_t *msg_kiov;
223 typedef struct lnet_libhandle {
224 cfs_list_t lh_hash_chain;
228 #define lh_entry(ptr, type, member) \
229 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
231 typedef struct lnet_eq {
233 lnet_libhandle_t eq_lh;
234 lnet_seq_t eq_enq_seq;
235 lnet_seq_t eq_deq_seq;
236 unsigned int eq_size;
237 lnet_event_t *eq_events;
239 lnet_eq_handler_t eq_callback;
242 typedef struct lnet_me {
244 lnet_libhandle_t me_lh;
245 lnet_process_id_t me_match_id;
246 unsigned int me_portal;
248 __u64 me_ignore_bits;
249 lnet_unlink_t me_unlink;
250 struct lnet_libmd *me_md;
253 typedef struct lnet_libmd {
255 lnet_libhandle_t md_lh;
258 unsigned int md_offset;
259 unsigned int md_length;
260 unsigned int md_max_size;
263 unsigned int md_options;
264 unsigned int md_flags;
267 unsigned int md_niov; /* # frags */
269 struct iovec iov[LNET_MAX_IOV];
270 lnet_kiov_t kiov[LNET_MAX_IOV];
274 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
275 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
277 #ifdef LNET_USE_LIB_FREELIST
280 void *fl_objs; /* single contiguous array of objects */
281 int fl_nobjs; /* the number of them */
282 int fl_objsize; /* the size (including overhead) of each of them */
283 cfs_list_t fl_list; /* where they are enqueued */
288 cfs_list_t fo_list; /* enqueue on fl_list */
289 void *fo_contents; /* aligned contents */
294 /* info about peers we are trying to fail */
295 cfs_list_t tp_list; /* ln_test_peers */
296 lnet_nid_t tp_nid; /* matching nid */
297 unsigned int tp_threshold; /* # failures to simulate */
300 #define LNET_COOKIE_TYPE_MD 1
301 #define LNET_COOKIE_TYPE_ME 2
302 #define LNET_COOKIE_TYPE_EQ 3
303 #define LNET_COOKIE_TYPE_BITS 2
304 #define LNET_COOKIE_TYPES (1 << LNET_COOKIE_TYPE_BITS)
305 /* LNET_COOKIE_TYPES must be a power of 2, so the cookie type can be
306 * extracted by masking with (LNET_COOKIE_TYPES - 1) */
308 struct lnet_ni; /* forward ref */
310 typedef struct lnet_lnd
312 /* fields managed by portals */
313 cfs_list_t lnd_list; /* stash in the LND table */
314 int lnd_refcount; /* # active instances */
316 /* fields initialised by the LND */
317 unsigned int lnd_type;
319 int (*lnd_startup) (struct lnet_ni *ni);
320 void (*lnd_shutdown) (struct lnet_ni *ni);
321 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
323 /* In data movement APIs below, payload buffers are described as a set
324 * of 'niov' fragments which are...
326 * in virtual memory (struct iovec *iov != NULL)
328 * in pages (kernel only: plt_kiov_t *kiov != NULL).
329 * The LND may NOT overwrite these fragment descriptors.
330 * An 'offset' and may specify a byte offset within the set of
331 * fragments to start from
334 /* Start sending a preformatted message. 'private' is NULL for PUT and
335 * GET messages; otherwise this is a response to an incoming message
336 * and 'private' is the 'private' passed to lnet_parse(). Return
337 * non-zero for immediate failure, otherwise complete later with
339 int (*lnd_send)(struct lnet_ni *ni, void *private, lnet_msg_t *msg);
341 /* Start receiving 'mlen' bytes of payload data, skipping the following
342 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
343 * lnet_parse(). Return non-zero for immedaite failure, otherwise
344 * complete later with lnet_finalize(). This also gives back a receive
345 * credit if the LND does flow control. */
346 int (*lnd_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
347 int delayed, unsigned int niov,
348 struct iovec *iov, lnet_kiov_t *kiov,
349 unsigned int offset, unsigned int mlen, unsigned int rlen);
351 /* lnet_parse() has had to delay processing of this message
352 * (e.g. waiting for a forwarding buffer or send credits). Give the
353 * LND a chance to free urgently needed resources. If called, return 0
354 * for success and do NOT give back a receive credit; that has to wait
355 * until lnd_recv() gets called. On failure return < 0 and
356 * release resources; lnd_recv() will not be called. */
357 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
358 void **new_privatep);
360 /* notification of peer health */
361 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
363 /* query of peer aliveness */
364 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, cfs_time_t *when);
366 #if defined(__KERNEL__) || defined(HAVE_LIBPTHREAD)
367 /* accept a new connection */
368 int (*lnd_accept)(struct lnet_ni *ni, cfs_socket_t *sock);
372 /* wait for something to happen */
373 void (*lnd_wait)(struct lnet_ni *ni, int milliseconds);
375 /* ensure non-RDMA messages can be received outside liblustre */
376 int (*lnd_setasync)(struct lnet_ni *ni, lnet_process_id_t id, int nasync);
380 #define LNET_NI_STATUS_UP 0x15aac0de
381 #define LNET_NI_STATUS_DOWN 0xdeadface
382 #define LNET_NI_STATUS_INVALID 0x00000000
387 } WIRE_ATTR lnet_ni_status_t;
389 #define LNET_MAX_INTERFACES 16
391 typedef struct lnet_ni {
392 cfs_list_t ni_list; /* chain on ln_nis */
393 cfs_list_t ni_txq; /* messages waiting for tx credits */
394 int ni_maxtxcredits; /* # tx credits */
395 int ni_txcredits; /* # tx credits free */
396 int ni_mintxcredits; /* lowest it's been */
397 int ni_peertxcredits; /* # per-peer send credits */
398 int ni_peerrtrcredits; /* # per-peer router buffer credits */
399 int ni_peertimeout; /* seconds to consider peer dead */
400 lnet_nid_t ni_nid; /* interface's NID */
401 void *ni_data; /* instance-specific data */
402 lnd_t *ni_lnd; /* procedural interface */
403 int ni_refcount; /* reference count */
404 cfs_time_t ni_last_alive; /* when I was last alive */
405 lnet_ni_status_t *ni_status; /* my health status */
406 char *ni_interfaces[LNET_MAX_INTERFACES]; /* equivalent interfaces to use */
409 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
410 #define LNET_PROTO_PING_VERSION 2
411 #define LNET_PROTO_PING_VERSION1 1
417 lnet_ni_status_t pi_ni[0];
418 } WIRE_ATTR lnet_ping_info_t;
420 /* router checker data, per router */
421 #define LNET_MAX_RTR_NIS 16
422 #define LNET_PINGINFO_SIZE offsetof(lnet_ping_info_t, pi_ni[LNET_MAX_RTR_NIS])
424 cfs_list_t rcd_list; /* chain on the_lnet.ln_zombie_rcd */
425 lnet_handle_md_t rcd_mdh; /* ping buffer MD */
426 lnet_ping_info_t *rcd_pinginfo; /* ping buffer */
429 typedef struct lnet_peer {
430 cfs_list_t lp_hashlist; /* chain on peer hash */
431 cfs_list_t lp_txq; /* messages blocking for tx credits */
432 cfs_list_t lp_rtrq; /* messages blocking for router credits */
433 cfs_list_t lp_rtr_list; /* chain on router list */
434 int lp_txcredits; /* # tx credits available */
435 int lp_mintxcredits; /* low water mark */
436 int lp_rtrcredits; /* # router credits */
437 int lp_minrtrcredits; /* low water mark */
438 unsigned int lp_alive:1; /* alive/dead? */
439 unsigned int lp_notify:1; /* notification outstanding? */
440 unsigned int lp_notifylnd:1; /* outstanding notification for LND? */
441 unsigned int lp_notifying:1; /* some thread is handling notification */
442 unsigned int lp_ping_notsent; /* SEND event outstanding from ping */
443 int lp_alive_count; /* # times router went dead<->alive */
444 long lp_txqnob; /* bytes queued for sending */
445 cfs_time_t lp_timestamp; /* time of last aliveness news */
446 cfs_time_t lp_ping_timestamp; /* time of last ping attempt */
447 cfs_time_t lp_ping_deadline; /* != 0 if ping reply expected */
448 cfs_time_t lp_last_alive; /* when I was last alive */
449 cfs_time_t lp_last_query; /* when lp_ni was queried last time */
450 lnet_ni_t *lp_ni; /* interface peer is on */
451 lnet_nid_t lp_nid; /* peer's NID */
452 int lp_refcount; /* # refs */
453 int lp_rtr_refcount; /* # refs from lnet_route_t::lr_gateway */
454 lnet_rc_data_t *lp_rcd; /* router checker state */
459 #define LNET_PEER_HASH_BITS 9
460 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
462 /* peer hash table */
463 struct lnet_peer_table {
464 int pt_version; /* /proc validity stamp */
465 int pt_number; /* # peers extant */
466 cfs_list_t pt_deathrow; /* zombie peers */
467 cfs_list_t *pt_hash; /* NID->peer hash */
470 #define lnet_peer_aliveness_enabled(lp) ((lp)->lp_ni->ni_peertimeout > 0)
473 cfs_list_t lr_list; /* chain on net */
474 lnet_peer_t *lr_gateway; /* router node */
475 unsigned int lr_hops; /* how far I am */
479 cfs_list_t lrn_list; /* chain on ln_remote_nets */
480 cfs_list_t lrn_routes; /* routes to me */
481 __u32 lrn_net; /* my net number */
485 cfs_list_t rbp_bufs; /* my free buffer pool */
486 cfs_list_t rbp_msgs; /* messages blocking for a buffer */
487 int rbp_npages; /* # pages in each buffer */
488 int rbp_nbuffers; /* # buffers */
489 int rbp_credits; /* # free buffers / blocked messages */
490 int rbp_mincredits; /* low water mark */
494 cfs_list_t rb_list; /* chain on rbp_bufs */
495 lnet_rtrbufpool_t *rb_pool; /* owning pool */
496 lnet_kiov_t rb_kiov[0]; /* the buffer space */
499 #include <libcfs/libcfs_pack.h>
512 } WIRE_ATTR lnet_counters_t;
513 #include <libcfs/libcfs_unpack.h>
515 #define LNET_PEER_HASHSIZE 503 /* prime! */
517 #define LNET_NRBPOOLS 3 /* # different router buffer pools */
519 /* Options for lnet_portal_t::ptl_options */
520 #define LNET_PTL_LAZY (1 << 0)
521 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
522 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
524 /* ME hash of RDMA portal */
525 #define LNET_PORTAL_HASH_BITS 8
526 #define LNET_PORTAL_HASH_SIZE (1 << LNET_PORTAL_HASH_BITS)
529 cfs_list_t *ptl_mhash; /* match hash */
530 cfs_list_t ptl_mlist; /* match list */
531 cfs_list_t ptl_msgq; /* messages blocking for MD */
532 __u64 ptl_ml_version; /* validity stamp, only changed for new attached MD */
533 __u64 ptl_msgq_version; /* validity stamp */
534 unsigned int ptl_options;
537 #define LNET_LH_HASH_BITS 12
538 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
539 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
541 /* resource container (ME, MD, EQ) */
542 struct lnet_res_container {
543 unsigned int rec_type; /* container type */
544 __u64 rec_lh_cookie; /* cookie generator */
545 cfs_list_t rec_active; /* active resource list */
546 cfs_list_t *rec_lh_hash; /* handle hash */
547 #ifdef LNET_USE_LIB_FREELIST
548 lnet_freelist_t rec_freelist; /* freelist for resources */
552 /* message container */
553 struct lnet_msg_container {
554 int msc_init; /* initialized or not */
555 /* max # threads finalizing */
557 /* msgs waiting to complete finalizing */
558 cfs_list_t msc_finalizing;
559 cfs_list_t msc_active; /* active message list */
560 /* threads doing finalization */
561 void **msc_finalizers;
562 #ifdef LNET_USE_LIB_FREELIST
563 lnet_freelist_t msc_freelist; /* freelist for messages */
567 /* Router Checker states */
568 #define LNET_RC_STATE_SHUTDOWN 0 /* not started */
569 #define LNET_RC_STATE_RUNNING 1 /* started up OK */
570 #define LNET_RC_STATE_STOPTHREAD 2 /* telling thread to stop */
571 #define LNET_RC_STATE_UNLINKING 3 /* unlinking RC MD */
572 #define LNET_RC_STATE_UNLINKED 4 /* RC's MD has been unlinked */
576 /* Stuff initialised at LNetInit() */
577 int ln_init; /* LNetInit() called? */
578 int ln_refcount; /* LNetNIInit/LNetNIFini counter */
579 int ln_niinit_self; /* Have I called LNetNIInit myself? */
581 cfs_list_t ln_lnds; /* registered LNDs */
584 cfs_spinlock_t ln_lock;
585 cfs_waitq_t ln_waitq;
586 cfs_mutex_t ln_api_mutex;
587 cfs_mutex_t ln_lnd_mutex;
589 # ifndef HAVE_LIBPTHREAD
594 pthread_cond_t ln_cond;
595 pthread_mutex_t ln_lock;
596 pthread_mutex_t ln_api_mutex;
597 pthread_mutex_t ln_lnd_mutex;
601 struct lnet_res_container ln_me_container;
603 struct lnet_res_container ln_md_container;
604 /* Event Queue container */
605 struct lnet_res_container ln_eq_container;
607 /* Stuff initialised at LNetNIInit() */
609 int ln_shutdown; /* shutdown in progress */
610 int ln_nportals; /* # portals */
611 lnet_portal_t *ln_portals; /* the vector of portals */
613 lnet_pid_t ln_pid; /* requested pid */
615 cfs_list_t ln_nis; /* LND instances */
616 lnet_ni_t *ln_loni; /* the loopback NI */
617 lnet_ni_t *ln_eqwaitni; /* NI to wait for events in */
618 cfs_list_t ln_zombie_nis; /* dying LND instances */
619 int ln_nzombie_nis; /* # of NIs to wait for */
621 cfs_list_t ln_remote_nets; /* remote networks with routes to them */
622 __u64 ln_remote_nets_version; /* validity stamp */
624 cfs_list_t ln_routers; /* list of all known routers */
625 __u64 ln_routers_version; /* validity stamp */
627 int ln_routing; /* am I a router? */
628 lnet_rtrbufpool_t ln_rtrpools[LNET_NRBPOOLS]; /* router buffer pools */
630 __u64 ln_interface_cookie; /* uniquely identifies this ni in this epoch */
632 char *ln_network_tokens; /* space for network names */
633 int ln_network_tokens_nob;
635 int ln_testprotocompat; /* test protocol compatibility flags */
637 cfs_list_t ln_test_peers; /* failure simulation */
639 /* message container */
640 struct lnet_peer_table *ln_peer_table;
641 struct lnet_msg_container ln_msg_container;
643 lnet_handle_md_t ln_ping_target_md;
644 lnet_handle_eq_t ln_ping_target_eq;
645 lnet_ping_info_t *ln_ping_info;
648 cfs_semaphore_t ln_rc_signal; /* serialise startup/shutdown */
650 int ln_rc_state; /* router checker startup/shutdown state */
651 lnet_handle_eq_t ln_rc_eqh; /* router checker's event queue */
652 lnet_handle_md_t ln_rc_mdh;
653 cfs_list_t ln_zombie_rcd;
655 lnet_counters_t ln_counters;
658 /* Temporary workaround to allow uOSS and test programs force
659 * server mode in userspace. The only place where we use it is
660 * lnet_prepare(). The only way to turn this flag on is to
661 * call lnet_server_mode() */
663 int ln_server_mode_flag;