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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Seagate, Inc.
32 * lnet/include/lnet/lib-types.h
35 #ifndef __LNET_LIB_TYPES_H__
36 #define __LNET_LIB_TYPES_H__
38 #include <linux/kthread.h>
39 #include <linux/uio.h>
40 #include <linux/types.h>
44 /* Max payload size */
45 #define LNET_MAX_PAYLOAD CONFIG_LNET_MAX_PAYLOAD
46 #if (LNET_MAX_PAYLOAD < LNET_MTU)
47 # error "LNET_MAX_PAYLOAD too small - error in configure --with-max-payload-mb"
48 #elif (LNET_MAX_PAYLOAD > (PAGE_SIZE * LNET_MAX_IOV))
49 # error "LNET_MAX_PAYLOAD too large - error in configure --with-max-payload-mb"
55 typedef struct lnet_msg {
56 struct list_head msg_activelist;
57 struct list_head msg_list; /* Q for credits/MD */
59 lnet_process_id_t msg_target;
60 /* where is it from, it's only for building event */
64 /* committed for sending */
65 unsigned int msg_tx_committed:1;
66 /* CPT # this message committed for sending */
67 unsigned int msg_tx_cpt:15;
68 /* committed for receiving */
69 unsigned int msg_rx_committed:1;
70 /* CPT # this message committed for receiving */
71 unsigned int msg_rx_cpt:15;
72 /* queued for tx credit */
73 unsigned int msg_tx_delayed:1;
74 /* queued for RX buffer */
75 unsigned int msg_rx_delayed:1;
76 /* ready for pending on RX delay list */
77 unsigned int msg_rx_ready_delay:1;
79 unsigned int msg_vmflush:1; /* VM trying to free memory */
80 unsigned int msg_target_is_router:1; /* sending to a router */
81 unsigned int msg_routing:1; /* being forwarded */
82 unsigned int msg_ack:1; /* ack on finalize (PUT) */
83 unsigned int msg_sending:1; /* outgoing message */
84 unsigned int msg_receiving:1; /* being received */
85 unsigned int msg_txcredit:1; /* taken an NI send credit */
86 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
87 unsigned int msg_rtrcredit:1; /* taken a global router credit */
88 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
89 unsigned int msg_onactivelist:1; /* on the activelist */
91 struct lnet_peer *msg_txpeer; /* peer I'm sending to */
92 struct lnet_peer *msg_rxpeer; /* peer I received from */
95 struct lnet_libmd *msg_md;
98 unsigned int msg_wanted;
99 unsigned int msg_offset;
100 unsigned int msg_niov;
101 struct kvec *msg_iov;
102 lnet_kiov_t *msg_kiov;
108 typedef struct lnet_libhandle {
109 struct list_head lh_hash_chain;
113 #define lh_entry(ptr, type, member) \
114 ((type *)((char *)(ptr) - (char *)(&((type *)0)->member)))
116 typedef struct lnet_eq {
117 struct list_head eq_list;
118 lnet_libhandle_t eq_lh;
119 lnet_seq_t eq_enq_seq;
120 lnet_seq_t eq_deq_seq;
121 unsigned int eq_size;
122 lnet_eq_handler_t eq_callback;
123 lnet_event_t *eq_events;
124 int **eq_refs; /* percpt refcount for EQ */
127 typedef struct lnet_me {
128 struct list_head me_list;
129 lnet_libhandle_t me_lh;
130 lnet_process_id_t me_match_id;
131 unsigned int me_portal;
132 unsigned int me_pos; /* hash offset in mt_hash */
134 __u64 me_ignore_bits;
135 lnet_unlink_t me_unlink;
136 struct lnet_libmd *me_md;
139 typedef struct lnet_libmd {
140 struct list_head md_list;
141 lnet_libhandle_t md_lh;
144 unsigned int md_offset;
145 unsigned int md_length;
146 unsigned int md_max_size;
149 unsigned int md_options;
150 unsigned int md_flags;
153 unsigned int md_niov; /* # frags */
155 struct kvec iov[LNET_MAX_IOV];
156 lnet_kiov_t kiov[LNET_MAX_IOV];
160 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
161 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
162 #define LNET_MD_FLAG_ABORTED (1 << 2)
165 /* info about peers we are trying to fail */
166 struct list_head tp_list; /* ln_test_peers */
167 lnet_nid_t tp_nid; /* matching nid */
168 unsigned int tp_threshold; /* # failures to simulate */
171 #define LNET_COOKIE_TYPE_MD 1
172 #define LNET_COOKIE_TYPE_ME 2
173 #define LNET_COOKIE_TYPE_EQ 3
174 #define LNET_COOKIE_TYPE_BITS 2
175 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
177 struct lnet_ni; /* forward ref */
179 typedef struct lnet_lnd {
180 /* fields managed by portals */
181 struct list_head lnd_list; /* stash in the LND table */
182 int lnd_refcount; /* # active instances */
184 /* fields initialised by the LND */
187 int (*lnd_startup)(struct lnet_ni *ni);
188 void (*lnd_shutdown)(struct lnet_ni *ni);
189 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
192 * In data movement APIs below, payload buffers are described as a set
193 * of 'niov' fragments which are...
195 * in virtual memory (struct iovec *iov != NULL)
197 * in pages (kernel only: plt_kiov_t *kiov != NULL).
198 * The LND may NOT overwrite these fragment descriptors.
199 * An 'offset' and may specify a byte offset within the set of
200 * fragments to start from
204 * Start sending a preformatted message. 'private' is NULL for PUT and
205 * GET messages; otherwise this is a response to an incoming message
206 * and 'private' is the 'private' passed to lnet_parse(). Return
207 * non-zero for immediate failure, otherwise complete later with
210 int (*lnd_send)(struct lnet_ni *ni, void *private, lnet_msg_t *msg);
213 * Start receiving 'mlen' bytes of payload data, skipping the following
214 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
215 * lnet_parse(). Return non-zero for immediate failure, otherwise
216 * complete later with lnet_finalize(). This also gives back a receive
217 * credit if the LND does flow control.
219 int (*lnd_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
220 int delayed, unsigned int niov,
221 struct kvec *iov, lnet_kiov_t *kiov,
222 unsigned int offset, unsigned int mlen,
226 * lnet_parse() has had to delay processing of this message
227 * (e.g. waiting for a forwarding buffer or send credits). Give the
228 * LND a chance to free urgently needed resources. If called, return 0
229 * for success and do NOT give back a receive credit; that has to wait
230 * until lnd_recv() gets called. On failure return < 0 and
231 * release resources; lnd_recv() will not be called.
233 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
234 lnet_msg_t *msg, void **new_privatep);
236 /* notification of peer health */
237 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
239 /* query of peer aliveness */
240 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer,
241 unsigned long *when);
243 /* accept a new connection */
244 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
247 struct lnet_tx_queue {
248 int tq_credits; /* # tx credits free */
249 int tq_credits_min; /* lowest it's been */
250 int tq_credits_max; /* total # tx credits */
251 struct list_head tq_delayed; /* delayed TXs */
254 typedef struct lnet_ni {
256 struct list_head ni_list; /* chain on ln_nis */
257 struct list_head ni_cptlist; /* chain on ln_nis_cpt */
258 int ni_maxtxcredits; /* # tx credits */
259 /* # per-peer send credits */
260 int ni_peertxcredits;
261 /* # per-peer router buffer credits */
262 int ni_peerrtrcredits;
263 /* seconds to consider peer dead */
265 int ni_ncpts; /* number of CPTs */
266 __u32 *ni_cpts; /* bond NI on some CPTs */
267 lnet_nid_t ni_nid; /* interface's NID */
268 void *ni_data; /* instance-specific data */
269 lnd_t *ni_lnd; /* procedural interface */
270 struct lnet_tx_queue **ni_tx_queues; /* percpt TX queues */
271 int **ni_refs; /* percpt reference count */
272 time64_t ni_last_alive;/* when I was last alive */
273 lnet_ni_status_t *ni_status; /* my health status */
274 /* equivalent interfaces to use */
275 char *ni_interfaces[LNET_MAX_INTERFACES];
278 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
281 * NB: value of these features equal to LNET_PROTO_PING_VERSION_x
282 * of old LNet, so there shouldn't be any compatibility issue
284 #define LNET_PING_FEAT_INVAL (0) /* no feature */
285 #define LNET_PING_FEAT_BASE (1 << 0) /* just a ping */
286 #define LNET_PING_FEAT_NI_STATUS (1 << 1) /* return NI status */
287 #define LNET_PING_FEAT_RTE_DISABLED (1 << 2) /* Routing enabled */
289 #define LNET_PING_FEAT_MASK (LNET_PING_FEAT_BASE | \
290 LNET_PING_FEAT_NI_STATUS)
292 /* router checker data, per router */
293 #define LNET_MAX_RTR_NIS 16
294 #define LNET_PINGINFO_SIZE offsetof(lnet_ping_info_t, pi_ni[LNET_MAX_RTR_NIS])
296 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
297 struct list_head rcd_list;
298 lnet_handle_md_t rcd_mdh; /* ping buffer MD */
299 struct lnet_peer *rcd_gateway; /* reference to gateway */
300 lnet_ping_info_t *rcd_pinginfo; /* ping buffer */
303 typedef struct lnet_peer {
304 struct list_head lp_hashlist; /* chain on peer hash */
305 struct list_head lp_txq; /* messages blocking for
307 struct list_head lp_rtrq; /* messages blocking for
309 struct list_head lp_rtr_list; /* chain on router list */
310 int lp_txcredits; /* # tx credits available */
311 int lp_mintxcredits; /* low water mark */
312 int lp_rtrcredits; /* # router credits */
313 int lp_minrtrcredits; /* low water mark */
314 unsigned int lp_alive:1; /* alive/dead? */
315 unsigned int lp_notify:1; /* notification outstanding? */
316 unsigned int lp_notifylnd:1;/* outstanding notification
318 unsigned int lp_notifying:1; /* some thread is handling
320 unsigned int lp_ping_notsent;/* SEND event outstanding
322 int lp_alive_count; /* # times router went
324 long lp_txqnob; /* bytes queued for sending */
325 unsigned long lp_timestamp; /* time of last aliveness
327 unsigned long lp_ping_timestamp;/* time of last ping
329 unsigned long lp_ping_deadline; /* != 0 if ping reply
331 unsigned long lp_last_alive; /* when I was last alive */
332 unsigned long lp_last_query; /* when lp_ni was queried
334 lnet_ni_t *lp_ni; /* interface peer is on */
335 lnet_nid_t lp_nid; /* peer's NID */
336 int lp_refcount; /* # refs */
337 int lp_cpt; /* CPT this peer attached on */
338 /* # refs from lnet_route_t::lr_gateway */
340 /* returned RC ping features */
341 unsigned int lp_ping_feats;
342 struct list_head lp_routes; /* routers on this peer */
343 lnet_rc_data_t *lp_rcd; /* router checker state */
347 #define LNET_PEER_HASH_BITS 9
348 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
350 /* peer hash table */
351 struct lnet_peer_table {
352 int pt_version; /* /proc validity stamp */
353 int pt_number; /* # peers extant */
354 /* # zombies to go to deathrow (and not there yet) */
356 struct list_head pt_deathrow; /* zombie peers */
357 struct list_head *pt_hash; /* NID->peer hash */
361 * peer aliveness is enabled only on routers for peers in a network where the
362 * lnet_ni_t::ni_peertimeout has been set to a positive value
364 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing && \
365 (lp)->lp_ni->ni_peertimeout > 0)
368 struct list_head lr_list; /* chain on net */
369 struct list_head lr_gwlist; /* chain on gateway */
370 lnet_peer_t *lr_gateway; /* router node */
371 __u32 lr_net; /* remote network number */
372 int lr_seq; /* sequence for round-robin */
373 unsigned int lr_downis; /* number of down NIs */
374 unsigned int lr_hops; /* how far I am */
375 unsigned int lr_priority; /* route priority */
378 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
379 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
380 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
383 struct list_head lrn_list; /* chain on
384 ln_remote_nets_hash */
385 struct list_head lrn_routes; /* routes to me */
386 __u32 lrn_net; /* my net number */
389 /** lnet message has credit and can be submitted to lnd for send/receive */
390 #define LNET_CREDIT_OK 0
391 /** lnet message is waiting for credit */
392 #define LNET_CREDIT_WAIT 1
395 struct list_head rbp_bufs; /* my free buffer pool */
396 struct list_head rbp_msgs; /* messages blocking
398 int rbp_npages; /* # pages in each buffer */
399 int rbp_nbuffers; /* # buffers */
400 int rbp_credits; /* # free buffers /
402 int rbp_mincredits; /* low water mark */
406 struct list_head rb_list; /* chain on rbp_bufs */
407 lnet_rtrbufpool_t *rb_pool; /* owning pool */
408 lnet_kiov_t rb_kiov[0]; /* the buffer space */
411 #define LNET_PEER_HASHSIZE 503 /* prime! */
413 #define LNET_TINY_BUF_IDX 0
414 #define LNET_SMALL_BUF_IDX 1
415 #define LNET_LARGE_BUF_IDX 2
417 /* # different router buffer pools */
418 #define LNET_NRBPOOLS (LNET_LARGE_BUF_IDX + 1)
421 /* Didn't match anything */
422 LNET_MATCHMD_NONE = (1 << 0),
424 LNET_MATCHMD_OK = (1 << 1),
425 /* Must be discarded */
426 LNET_MATCHMD_DROP = (1 << 2),
427 /* match and buffer is exhausted */
428 LNET_MATCHMD_EXHAUSTED = (1 << 3),
430 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
433 /* Options for lnet_portal_t::ptl_options */
434 #define LNET_PTL_LAZY (1 << 0)
435 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
436 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match,
439 /* parameter for matching operations (GET, PUT) */
440 struct lnet_match_info {
442 lnet_process_id_t mi_id;
444 unsigned int mi_portal;
445 unsigned int mi_rlength;
446 unsigned int mi_roffset;
449 /* ME hash of RDMA portal */
450 #define LNET_MT_HASH_BITS 8
451 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
452 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
454 * we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
455 * the last entry is reserved for MEs with ignore-bits
457 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
459 * __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
460 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
461 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE]
463 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
464 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
465 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
467 /* portal match table */
468 struct lnet_match_table {
469 /* reserved for upcoming patches, CPU partition ID */
471 unsigned int mt_portal; /* portal index */
473 * match table is set as "enabled" if there's non-exhausted MD
474 * attached on mt_mhash, it's only valid for wildcard portal
476 unsigned int mt_enabled;
477 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
478 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
479 struct list_head *mt_mhash; /* matching hash */
482 /* these are only useful for wildcard portal */
483 /* Turn off message rotor for wildcard portals */
484 #define LNET_PTL_ROTOR_OFF 0
485 /* round-robin dispatch all PUT messages for wildcard portals */
486 #define LNET_PTL_ROTOR_ON 1
487 /* round-robin dispatch routed PUT message for wildcard portals */
488 #define LNET_PTL_ROTOR_RR_RT 2
489 /* dispatch routed PUT message by hashing source NID for wildcard portals */
490 #define LNET_PTL_ROTOR_HASH_RT 3
492 typedef struct lnet_portal {
494 unsigned int ptl_index; /* portal ID, reserved */
495 /* flags on this portal: lazy, unique... */
496 unsigned int ptl_options;
497 /* list of messages which are stealing buffer */
498 struct list_head ptl_msg_stealing;
499 /* messages blocking for MD */
500 struct list_head ptl_msg_delayed;
501 /* Match table for each CPT */
502 struct lnet_match_table **ptl_mtables;
503 /* spread rotor of incoming "PUT" */
504 unsigned int ptl_rotor;
505 /* # active entries for this portal */
507 /* array of active entries' cpu-partition-id */
511 #define LNET_LH_HASH_BITS 12
512 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
513 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
515 /* resource container (ME, MD, EQ) */
516 struct lnet_res_container {
517 unsigned int rec_type; /* container type */
518 __u64 rec_lh_cookie; /* cookie generator */
519 struct list_head rec_active; /* active resource list */
520 struct list_head *rec_lh_hash; /* handle hash */
523 /* message container */
524 struct lnet_msg_container {
525 int msc_init; /* initialized or not */
526 /* max # threads finalizing */
528 /* msgs waiting to complete finalizing */
529 struct list_head msc_finalizing;
530 struct list_head msc_active; /* active message list */
531 /* threads doing finalization */
532 void **msc_finalizers;
535 /* Router Checker states */
536 #define LNET_RC_STATE_SHUTDOWN 0 /* not started */
537 #define LNET_RC_STATE_RUNNING 1 /* started up OK */
538 #define LNET_RC_STATE_STOPPING 2 /* telling thread to stop */
541 /* CPU partition table of LNet */
542 struct cfs_cpt_table *ln_cpt_table;
543 /* number of CPTs in ln_cpt_table */
544 unsigned int ln_cpt_number;
545 unsigned int ln_cpt_bits;
547 /* protect LNet resources (ME/MD/EQ) */
548 struct cfs_percpt_lock *ln_res_lock;
551 /* the vector of portals */
552 lnet_portal_t **ln_portals;
553 /* percpt ME containers */
554 struct lnet_res_container **ln_me_containers;
555 /* percpt MD container */
556 struct lnet_res_container **ln_md_containers;
558 /* Event Queue container */
559 struct lnet_res_container ln_eq_container;
560 wait_queue_head_t ln_eq_waitq;
561 spinlock_t ln_eq_wait_lock;
562 unsigned int ln_remote_nets_hbits;
564 /* protect NI, peer table, credits, routers, rtrbuf... */
565 struct cfs_percpt_lock *ln_net_lock;
566 /* percpt message containers for active/finalizing/freed message */
567 struct lnet_msg_container **ln_msg_containers;
568 lnet_counters_t **ln_counters;
569 struct lnet_peer_table **ln_peer_tables;
570 /* failure simulation */
571 struct list_head ln_test_peers;
573 struct list_head ln_nis; /* LND instances */
574 /* NIs bond on specific CPT(s) */
575 struct list_head ln_nis_cpt;
576 /* dying LND instances */
577 struct list_head ln_nis_zombie;
578 lnet_ni_t *ln_loni; /* the loopback NI */
580 /* remote networks with routes to them */
581 struct list_head *ln_remote_nets_hash;
583 __u64 ln_remote_nets_version;
584 /* list of all known routers */
585 struct list_head ln_routers;
587 __u64 ln_routers_version;
588 /* percpt router buffer pools */
589 lnet_rtrbufpool_t **ln_rtrpools;
591 lnet_handle_md_t ln_ping_target_md;
592 lnet_handle_eq_t ln_ping_target_eq;
593 lnet_ping_info_t *ln_ping_info;
595 /* router checker startup/shutdown state */
597 /* router checker's event queue */
598 lnet_handle_eq_t ln_rc_eqh;
599 /* rcd still pending on net */
600 struct list_head ln_rcd_deathrow;
601 /* rcd ready for free */
602 struct list_head ln_rcd_zombie;
603 /* serialise startup/shutdown */
604 struct semaphore ln_rc_signal;
606 struct mutex ln_api_mutex;
607 struct mutex ln_lnd_mutex;
608 /* Have I called LNetNIInit myself? */
610 /* LNetNIInit/LNetNIFini counter */
612 /* shutdown in progress */
615 int ln_routing; /* am I a router? */
616 lnet_pid_t ln_pid; /* requested pid */
617 /* uniquely identifies this ni in this epoch */
618 __u64 ln_interface_cookie;
619 /* registered LNDs */
620 struct list_head ln_lnds;
622 /* test protocol compatibility flags */
623 int ln_testprotocompat;
626 * 0 - load the NIs from the mod params
627 * 1 - do not load the NIs from the mod params
628 * Reverse logic to ensure that other calls to LNetNIInit
631 bool ln_nis_from_mod_params;
634 * waitq for router checker. As long as there are no routes in
635 * the list, the router checker will sleep on this queue. when
636 * routes are added the thread will wake up
638 wait_queue_head_t ln_rc_waitq;