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;
190 /* where is it from, it's only for building event */
194 unsigned int msg_vmflush:1; /* VM trying to free memory */
195 unsigned int msg_target_is_router:1; /* sending to a router */
196 unsigned int msg_routing:1; /* being forwarded */
197 unsigned int msg_ack:1; /* ack on finalize (PUT) */
198 unsigned int msg_sending:1; /* outgoing message */
199 unsigned int msg_receiving:1; /* being received */
200 unsigned int msg_delayed:1; /* had to Q for buffer or tx credit */
201 unsigned int msg_txcredit:1; /* taken an NI send credit */
202 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
203 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
204 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
205 unsigned int msg_onactivelist:1; /* on the activelist */
207 struct lnet_peer *msg_txpeer; /* peer I'm sending to */
208 struct lnet_peer *msg_rxpeer; /* peer I received from */
211 struct lnet_libmd *msg_md;
213 unsigned int msg_len;
214 unsigned int msg_wanted;
215 unsigned int msg_offset;
216 unsigned int msg_niov;
217 struct iovec *msg_iov;
218 lnet_kiov_t *msg_kiov;
225 typedef struct lnet_libhandle {
226 cfs_list_t lh_hash_chain;
230 #define lh_entry(ptr, type, member) \
231 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
233 typedef struct lnet_eq {
235 lnet_libhandle_t eq_lh;
236 lnet_seq_t eq_enq_seq;
237 lnet_seq_t eq_deq_seq;
238 unsigned int eq_size;
239 lnet_event_t *eq_events;
241 lnet_eq_handler_t eq_callback;
244 typedef struct lnet_me {
246 lnet_libhandle_t me_lh;
247 lnet_process_id_t me_match_id;
248 unsigned int me_portal;
250 __u64 me_ignore_bits;
251 lnet_unlink_t me_unlink;
252 struct lnet_libmd *me_md;
255 typedef struct lnet_libmd {
257 lnet_libhandle_t md_lh;
260 unsigned int md_offset;
261 unsigned int md_length;
262 unsigned int md_max_size;
265 unsigned int md_options;
266 unsigned int md_flags;
269 unsigned int md_niov; /* # frags */
271 struct iovec iov[LNET_MAX_IOV];
272 lnet_kiov_t kiov[LNET_MAX_IOV];
276 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
277 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
279 #ifdef LNET_USE_LIB_FREELIST
282 void *fl_objs; /* single contiguous array of objects */
283 int fl_nobjs; /* the number of them */
284 int fl_objsize; /* the size (including overhead) of each of them */
285 cfs_list_t fl_list; /* where they are enqueued */
290 cfs_list_t fo_list; /* enqueue on fl_list */
291 void *fo_contents; /* aligned contents */
296 /* info about peers we are trying to fail */
297 cfs_list_t tp_list; /* ln_test_peers */
298 lnet_nid_t tp_nid; /* matching nid */
299 unsigned int tp_threshold; /* # failures to simulate */
302 #define LNET_COOKIE_TYPE_MD 1
303 #define LNET_COOKIE_TYPE_ME 2
304 #define LNET_COOKIE_TYPE_EQ 3
305 #define LNET_COOKIE_TYPE_BITS 2
306 #define LNET_COOKIE_TYPES (1 << LNET_COOKIE_TYPE_BITS)
307 /* LNET_COOKIE_TYPES must be a power of 2, so the cookie type can be
308 * extracted by masking with (LNET_COOKIE_TYPES - 1) */
310 struct lnet_ni; /* forward ref */
312 typedef struct lnet_lnd
314 /* fields managed by portals */
315 cfs_list_t lnd_list; /* stash in the LND table */
316 int lnd_refcount; /* # active instances */
318 /* fields initialised by the LND */
319 unsigned int lnd_type;
321 int (*lnd_startup) (struct lnet_ni *ni);
322 void (*lnd_shutdown) (struct lnet_ni *ni);
323 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
325 /* In data movement APIs below, payload buffers are described as a set
326 * of 'niov' fragments which are...
328 * in virtual memory (struct iovec *iov != NULL)
330 * in pages (kernel only: plt_kiov_t *kiov != NULL).
331 * The LND may NOT overwrite these fragment descriptors.
332 * An 'offset' and may specify a byte offset within the set of
333 * fragments to start from
336 /* Start sending a preformatted message. 'private' is NULL for PUT and
337 * GET messages; otherwise this is a response to an incoming message
338 * and 'private' is the 'private' passed to lnet_parse(). Return
339 * non-zero for immediate failure, otherwise complete later with
341 int (*lnd_send)(struct lnet_ni *ni, void *private, lnet_msg_t *msg);
343 /* Start receiving 'mlen' bytes of payload data, skipping the following
344 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
345 * lnet_parse(). Return non-zero for immedaite failure, otherwise
346 * complete later with lnet_finalize(). This also gives back a receive
347 * credit if the LND does flow control. */
348 int (*lnd_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
349 int delayed, unsigned int niov,
350 struct iovec *iov, lnet_kiov_t *kiov,
351 unsigned int offset, unsigned int mlen, unsigned int rlen);
353 /* lnet_parse() has had to delay processing of this message
354 * (e.g. waiting for a forwarding buffer or send credits). Give the
355 * LND a chance to free urgently needed resources. If called, return 0
356 * for success and do NOT give back a receive credit; that has to wait
357 * until lnd_recv() gets called. On failure return < 0 and
358 * release resources; lnd_recv() will not be called. */
359 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
360 void **new_privatep);
362 /* notification of peer health */
363 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
365 /* query of peer aliveness */
366 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, cfs_time_t *when);
368 #if defined(__KERNEL__) || defined(HAVE_LIBPTHREAD)
369 /* accept a new connection */
370 int (*lnd_accept)(struct lnet_ni *ni, cfs_socket_t *sock);
374 /* wait for something to happen */
375 void (*lnd_wait)(struct lnet_ni *ni, int milliseconds);
377 /* ensure non-RDMA messages can be received outside liblustre */
378 int (*lnd_setasync)(struct lnet_ni *ni, lnet_process_id_t id, int nasync);
382 #define LNET_NI_STATUS_UP 0x15aac0de
383 #define LNET_NI_STATUS_DOWN 0xdeadface
384 #define LNET_NI_STATUS_INVALID 0x00000000
389 } WIRE_ATTR lnet_ni_status_t;
391 #define LNET_MAX_INTERFACES 16
393 typedef struct lnet_ni {
394 cfs_list_t ni_list; /* chain on ln_nis */
395 cfs_list_t ni_txq; /* messages waiting for tx credits */
396 int ni_maxtxcredits; /* # tx credits */
397 int ni_txcredits; /* # tx credits free */
398 int ni_mintxcredits; /* lowest it's been */
399 int ni_peertxcredits; /* # per-peer send credits */
400 int ni_peerrtrcredits; /* # per-peer router buffer credits */
401 int ni_peertimeout; /* seconds to consider peer dead */
402 lnet_nid_t ni_nid; /* interface's NID */
403 void *ni_data; /* instance-specific data */
404 lnd_t *ni_lnd; /* procedural interface */
405 int ni_refcount; /* reference count */
406 /* when I was last alive */
408 lnet_ni_status_t *ni_status; /* my health status */
409 /* equivalent interfaces to use */
410 char *ni_interfaces[LNET_MAX_INTERFACES];
413 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
415 LNET_PROTO_PING_UNKNOWN = 0, /* unknown */
416 LNET_PROTO_PING_VERSION_1 = 1, /* old version */
417 LNET_PROTO_PING_VERSION = 2, /* current version */
425 lnet_ni_status_t pi_ni[0];
426 } WIRE_ATTR lnet_ping_info_t;
428 /* router checker data, per router */
429 #define LNET_MAX_RTR_NIS 16
430 #define LNET_PINGINFO_SIZE offsetof(lnet_ping_info_t, pi_ni[LNET_MAX_RTR_NIS])
432 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
434 lnet_handle_md_t rcd_mdh; /* ping buffer MD */
435 struct lnet_peer *rcd_gateway; /* reference to gateway */
436 lnet_ping_info_t *rcd_pinginfo; /* ping buffer */
439 typedef struct lnet_peer {
440 cfs_list_t lp_hashlist; /* chain on peer hash */
441 cfs_list_t lp_txq; /* messages blocking for tx credits */
442 cfs_list_t lp_rtrq; /* messages blocking for router credits */
443 cfs_list_t lp_rtr_list; /* chain on router list */
444 int lp_txcredits; /* # tx credits available */
445 int lp_mintxcredits; /* low water mark */
446 int lp_rtrcredits; /* # router credits */
447 int lp_minrtrcredits; /* low water mark */
448 unsigned int lp_alive:1; /* alive/dead? */
449 unsigned int lp_notify:1; /* notification outstanding? */
450 unsigned int lp_notifylnd:1; /* outstanding notification for LND? */
451 unsigned int lp_notifying:1; /* some thread is handling notification */
452 unsigned int lp_ping_notsent; /* SEND event outstanding from ping */
453 int lp_alive_count; /* # times router went dead<->alive */
454 long lp_txqnob; /* bytes queued for sending */
455 cfs_time_t lp_timestamp; /* time of last aliveness news */
456 cfs_time_t lp_ping_timestamp; /* time of last ping attempt */
457 cfs_time_t lp_ping_deadline; /* != 0 if ping reply expected */
458 cfs_time_t lp_last_alive; /* when I was last alive */
459 cfs_time_t lp_last_query; /* when lp_ni was queried last time */
460 lnet_ni_t *lp_ni; /* interface peer is on */
461 lnet_nid_t lp_nid; /* peer's NID */
462 int lp_refcount; /* # refs */
463 int lp_rtr_refcount; /* # refs from lnet_route_t::lr_gateway */
464 /* returned RC ping version */
465 unsigned int lp_ping_version;
466 cfs_list_t lp_routes; /* routers on this peer */
467 lnet_rc_data_t *lp_rcd; /* router checker state */
472 #define LNET_PEER_HASH_BITS 9
473 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
475 /* peer hash table */
476 struct lnet_peer_table {
477 int pt_version; /* /proc validity stamp */
478 int pt_number; /* # peers extant */
479 cfs_list_t pt_deathrow; /* zombie peers */
480 cfs_list_t *pt_hash; /* NID->peer hash */
483 #define lnet_peer_aliveness_enabled(lp) ((lp)->lp_ni->ni_peertimeout > 0)
486 cfs_list_t lr_list; /* chain on net */
487 cfs_list_t lr_gwlist; /* chain on gateway */
488 lnet_peer_t *lr_gateway; /* router node */
489 __u32 lr_net; /* remote network number */
490 unsigned int lr_downis; /* number of down NIs */
491 unsigned int lr_hops; /* how far I am */
495 cfs_list_t lrn_list; /* chain on ln_remote_nets */
496 cfs_list_t lrn_routes; /* routes to me */
497 __u32 lrn_net; /* my net number */
501 cfs_list_t rbp_bufs; /* my free buffer pool */
502 cfs_list_t rbp_msgs; /* messages blocking for a buffer */
503 int rbp_npages; /* # pages in each buffer */
504 int rbp_nbuffers; /* # buffers */
505 int rbp_credits; /* # free buffers / blocked messages */
506 int rbp_mincredits; /* low water mark */
510 cfs_list_t rb_list; /* chain on rbp_bufs */
511 lnet_rtrbufpool_t *rb_pool; /* owning pool */
512 lnet_kiov_t rb_kiov[0]; /* the buffer space */
515 #include <libcfs/libcfs_pack.h>
528 } WIRE_ATTR lnet_counters_t;
529 #include <libcfs/libcfs_unpack.h>
531 #define LNET_PEER_HASHSIZE 503 /* prime! */
533 #define LNET_NRBPOOLS 3 /* # different router buffer pools */
536 /* Didn't match anything */
537 LNET_MATCHMD_NONE = (1 << 0),
539 LNET_MATCHMD_OK = (1 << 1),
540 /* Must be discarded */
541 LNET_MATCHMD_DROP = (1 << 2),
544 /* Options for lnet_portal_t::ptl_options */
545 #define LNET_PTL_LAZY (1 << 0)
546 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
547 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
549 /* ME hash of RDMA portal */
550 #define LNET_PORTAL_HASH_BITS 8
551 #define LNET_PORTAL_HASH_SIZE (1 << LNET_PORTAL_HASH_BITS)
553 typedef struct lnet_portal {
554 unsigned int ptl_index; /* portal ID, reserved */
555 cfs_list_t *ptl_mhash; /* match hash */
556 cfs_list_t ptl_mlist; /* match list */
557 cfs_list_t ptl_msgq; /* messages blocking for MD */
558 __u64 ptl_ml_version; /* validity stamp, only changed for new attached MD */
559 __u64 ptl_msgq_version; /* validity stamp */
560 unsigned int ptl_options;
563 #define LNET_LH_HASH_BITS 12
564 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
565 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
567 /* resource container (ME, MD, EQ) */
568 struct lnet_res_container {
569 unsigned int rec_type; /* container type */
570 __u64 rec_lh_cookie; /* cookie generator */
571 cfs_list_t rec_active; /* active resource list */
572 cfs_list_t *rec_lh_hash; /* handle hash */
573 #ifdef LNET_USE_LIB_FREELIST
574 lnet_freelist_t rec_freelist; /* freelist for resources */
578 /* message container */
579 struct lnet_msg_container {
580 int msc_init; /* initialized or not */
581 /* max # threads finalizing */
583 /* msgs waiting to complete finalizing */
584 cfs_list_t msc_finalizing;
585 cfs_list_t msc_active; /* active message list */
586 /* threads doing finalization */
587 void **msc_finalizers;
588 #ifdef LNET_USE_LIB_FREELIST
589 lnet_freelist_t msc_freelist; /* freelist for messages */
593 /* Router Checker states */
594 #define LNET_RC_STATE_SHUTDOWN 0 /* not started */
595 #define LNET_RC_STATE_RUNNING 1 /* started up OK */
596 #define LNET_RC_STATE_STOPPING 2 /* telling thread to stop */
600 /* Stuff initialised at LNetInit() */
601 int ln_init; /* LNetInit() called? */
602 int ln_refcount; /* LNetNIInit/LNetNIFini counter */
603 int ln_niinit_self; /* Have I called LNetNIInit myself? */
604 /* shutdown in progress */
607 cfs_list_t ln_lnds; /* registered LNDs */
610 cfs_spinlock_t ln_lock;
611 cfs_mutex_t ln_api_mutex;
612 cfs_mutex_t ln_lnd_mutex;
613 cfs_waitq_t ln_eq_waitq;
615 # ifndef HAVE_LIBPTHREAD
620 pthread_mutex_t ln_lock;
621 pthread_mutex_t ln_api_mutex;
622 pthread_mutex_t ln_lnd_mutex;
623 pthread_cond_t ln_eq_cond;
627 struct lnet_res_container ln_me_container;
629 struct lnet_res_container ln_md_container;
630 /* Event Queue container */
631 struct lnet_res_container ln_eq_container;
635 /* the vector of portals */
636 lnet_portal_t **ln_portals;
638 lnet_pid_t ln_pid; /* requested pid */
640 cfs_list_t ln_nis; /* LND instances */
641 lnet_ni_t *ln_loni; /* the loopback NI */
642 /* NI to wait for events in */
643 lnet_ni_t *ln_eq_waitni;
645 cfs_list_t ln_zombie_nis; /* dying LND instances */
646 int ln_nzombie_nis; /* # of NIs to wait for */
648 cfs_list_t ln_remote_nets; /* remote networks with routes to them */
649 __u64 ln_remote_nets_version; /* validity stamp */
651 cfs_list_t ln_routers; /* list of all known routers */
652 __u64 ln_routers_version; /* validity stamp */
654 int ln_routing; /* am I a router? */
655 lnet_rtrbufpool_t ln_rtrpools[LNET_NRBPOOLS]; /* router buffer pools */
657 __u64 ln_interface_cookie; /* uniquely identifies this ni in this epoch */
659 char *ln_network_tokens; /* space for network names */
660 int ln_network_tokens_nob;
662 int ln_testprotocompat; /* test protocol compatibility flags */
664 cfs_list_t ln_test_peers; /* failure simulation */
666 /* message container */
667 struct lnet_peer_table *ln_peer_table;
668 struct lnet_msg_container ln_msg_container;
670 lnet_handle_md_t ln_ping_target_md;
671 lnet_handle_eq_t ln_ping_target_eq;
672 lnet_ping_info_t *ln_ping_info;
675 cfs_semaphore_t ln_rc_signal; /* serialise startup/shutdown */
677 /* router checker startup/shutdown state */
679 /* router checker's event queue */
680 lnet_handle_eq_t ln_rc_eqh;
681 /* rcd still pending on net */
682 cfs_list_t ln_rcd_deathrow;
683 /* rcd ready for free */
684 cfs_list_t ln_rcd_zombie;
686 lnet_counters_t ln_counters;
689 /* Temporary workaround to allow uOSS and test programs force
690 * server mode in userspace. The only place where we use it is
691 * lnet_prepare(). The only way to turn this flag on is to
692 * call lnet_server_mode() */
694 int ln_server_mode_flag;