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, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lnet/include/lnet/lib-types.h
34 * Types used by the library side routines that do not need to be
35 * exposed to the user application
38 #ifndef __LNET_LIB_TYPES_H__
39 #define __LNET_LIB_TYPES_H__
42 # error This include is only for kernel use.
45 #include <linux/kthread.h>
46 #include <linux/uio.h>
47 #include <linux/types.h>
49 #include <lnet/lnetctl.h>
51 /* Max payload size */
52 #ifndef CONFIG_LNET_MAX_PAYLOAD
53 # error "CONFIG_LNET_MAX_PAYLOAD must be defined in config.h"
56 #define LNET_MAX_PAYLOAD CONFIG_LNET_MAX_PAYLOAD
57 #if (LNET_MAX_PAYLOAD < LNET_MTU)
58 # error "LNET_MAX_PAYLOAD too small - error in configure --with-max-payload-mb"
59 #elif (LNET_MAX_PAYLOAD > (PAGE_SIZE * LNET_MAX_IOV))
60 # error "LNET_MAX_PAYLOAD too large - error in configure --with-max-payload-mb"
66 typedef struct lnet_msg {
67 struct list_head msg_activelist;
68 struct list_head msg_list; /* Q for credits/MD */
70 lnet_process_id_t msg_target;
71 /* Primary NID of the source. */
72 lnet_nid_t msg_initiator;
73 /* where is it from, it's only for building event */
77 /* committed for sending */
78 unsigned int msg_tx_committed:1;
79 /* CPT # this message committed for sending */
80 unsigned int msg_tx_cpt:15;
81 /* committed for receiving */
82 unsigned int msg_rx_committed:1;
83 /* CPT # this message committed for receiving */
84 unsigned int msg_rx_cpt:15;
85 /* queued for tx credit */
86 unsigned int msg_tx_delayed:1;
87 /* queued for RX buffer */
88 unsigned int msg_rx_delayed:1;
89 /* ready for pending on RX delay list */
90 unsigned int msg_rx_ready_delay:1;
92 unsigned int msg_vmflush:1; /* VM trying to free memory */
93 unsigned int msg_target_is_router:1; /* sending to a router */
94 unsigned int msg_routing:1; /* being forwarded */
95 unsigned int msg_ack:1; /* ack on finalize (PUT) */
96 unsigned int msg_sending:1; /* outgoing message */
97 unsigned int msg_receiving:1; /* being received */
98 unsigned int msg_txcredit:1; /* taken an NI send credit */
99 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
100 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
101 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
102 unsigned int msg_onactivelist:1; /* on the activelist */
103 unsigned int msg_rdma_get:1;
105 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
106 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
109 struct lnet_libmd *msg_md;
110 /* the NI the message was sent or received over */
111 struct lnet_ni *msg_txni;
112 struct lnet_ni *msg_rxni;
114 unsigned int msg_len;
115 unsigned int msg_wanted;
116 unsigned int msg_offset;
117 unsigned int msg_niov;
118 struct kvec *msg_iov;
119 lnet_kiov_t *msg_kiov;
126 typedef struct lnet_libhandle {
127 struct list_head lh_hash_chain;
131 #define lh_entry(ptr, type, member) \
132 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
134 typedef struct lnet_eq {
135 struct list_head eq_list;
136 lnet_libhandle_t eq_lh;
137 lnet_seq_t eq_enq_seq;
138 lnet_seq_t eq_deq_seq;
139 unsigned int eq_size;
140 lnet_eq_handler_t eq_callback;
141 lnet_event_t *eq_events;
142 int **eq_refs; /* percpt refcount for EQ */
145 typedef struct lnet_me {
146 struct list_head me_list;
147 lnet_libhandle_t me_lh;
148 lnet_process_id_t me_match_id;
149 unsigned int me_portal;
150 unsigned int me_pos; /* hash offset in mt_hash */
152 __u64 me_ignore_bits;
153 lnet_unlink_t me_unlink;
154 struct lnet_libmd *me_md;
157 typedef struct lnet_libmd {
158 struct list_head md_list;
159 lnet_libhandle_t md_lh;
162 unsigned int md_offset;
163 unsigned int md_length;
164 unsigned int md_max_size;
167 unsigned int md_options;
168 unsigned int md_flags;
169 unsigned int md_niov; /* # frags at end of struct */
172 lnet_handle_md_t md_bulk_handle;
174 struct kvec iov[LNET_MAX_IOV];
175 lnet_kiov_t kiov[LNET_MAX_IOV];
179 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
180 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
181 #define LNET_MD_FLAG_ABORTED (1 << 2)
184 /* info about peers we are trying to fail */
185 struct list_head tp_list; /* ln_test_peers */
186 lnet_nid_t tp_nid; /* matching nid */
187 unsigned int tp_threshold; /* # failures to simulate */
190 #define LNET_COOKIE_TYPE_MD 1
191 #define LNET_COOKIE_TYPE_ME 2
192 #define LNET_COOKIE_TYPE_EQ 3
193 #define LNET_COOKIE_TYPE_BITS 2
194 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
196 struct lnet_ni; /* forward ref */
199 typedef struct lnet_lnd
201 /* fields managed by portals */
202 struct list_head lnd_list; /* stash in the LND table */
203 int lnd_refcount; /* # active instances */
205 /* fields initialized by the LND */
208 int (*lnd_startup)(struct lnet_ni *ni);
209 void (*lnd_shutdown)(struct lnet_ni *ni);
210 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
212 /* In data movement APIs below, payload buffers are described as a set
213 * of 'niov' fragments which are...
215 * in virtual memory (struct kvec *iov != NULL)
217 * in pages (kernel only: plt_kiov_t *kiov != NULL).
218 * The LND may NOT overwrite these fragment descriptors.
219 * An 'offset' and may specify a byte offset within the set of
220 * fragments to start from
223 /* Start sending a preformatted message. 'private' is NULL for PUT and
224 * GET messages; otherwise this is a response to an incoming message
225 * and 'private' is the 'private' passed to lnet_parse(). Return
226 * non-zero for immediate failure, otherwise complete later with
228 int (*lnd_send)(struct lnet_ni *ni, void *private, lnet_msg_t *msg);
230 /* Start receiving 'mlen' bytes of payload data, skipping the following
231 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
232 * lnet_parse(). Return non-zero for immedaite failure, otherwise
233 * complete later with lnet_finalize(). This also gives back a receive
234 * credit if the LND does flow control. */
235 int (*lnd_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
236 int delayed, unsigned int niov,
237 struct kvec *iov, lnet_kiov_t *kiov,
238 unsigned int offset, unsigned int mlen, unsigned int rlen);
240 /* lnet_parse() has had to delay processing of this message
241 * (e.g. waiting for a forwarding buffer or send credits). Give the
242 * LND a chance to free urgently needed resources. If called, return 0
243 * for success and do NOT give back a receive credit; that has to wait
244 * until lnd_recv() gets called. On failure return < 0 and
245 * release resources; lnd_recv() will not be called. */
246 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
247 void **new_privatep);
249 /* notification of peer health */
250 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
252 /* query of peer aliveness */
253 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, cfs_time_t *when);
255 /* accept a new connection */
256 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
259 struct lnet_tx_queue {
260 int tq_credits; /* # tx credits free */
261 int tq_credits_min; /* lowest it's been */
262 int tq_credits_max; /* total # tx credits */
263 struct list_head tq_delayed; /* delayed TXs */
266 enum lnet_net_state {
267 /* set when net block is allocated */
268 LNET_NET_STATE_INIT = 0,
269 /* set when NIs in net are started successfully */
270 LNET_NET_STATE_ACTIVE,
271 /* set if all NIs in net are in FAILED state */
272 LNET_NET_STATE_INACTIVE,
273 /* set when shutting down a NET */
274 LNET_NET_STATE_DELETING
278 /* set when NI block is allocated */
279 LNET_NI_STATE_INIT = 0,
280 /* set when NI is started successfully */
281 LNET_NI_STATE_ACTIVE,
282 /* set when LND notifies NI failed */
283 LNET_NI_STATE_FAILED,
284 /* set when LND notifies NI degraded */
285 LNET_NI_STATE_DEGRADED,
286 /* set when shuttding down NI */
287 LNET_NI_STATE_DELETING
290 struct lnet_element_stats {
297 /* chain on the ln_nets */
298 struct list_head net_list;
300 /* net ID, which is composed of
301 * (net_type << 16) | net_num.
302 * net_type can be one of the enumerated types defined in
303 * lnet/include/lnet/nidstr.h */
306 /* priority of the network */
309 /* total number of CPTs in the array */
312 /* cumulative CPTs of all NIs in this net */
315 /* network tunables */
316 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
319 * boolean to indicate that the tunables have been set and
322 bool net_tunables_set;
324 /* procedural interface */
327 /* list of NIs on this net */
328 struct list_head net_ni_list;
330 /* list of NIs being added, but not started yet */
331 struct list_head net_ni_added;
333 /* dying LND instances */
334 struct list_head net_ni_zombie;
337 enum lnet_net_state net_state;
340 typedef struct lnet_ni {
341 /* chain on the lnet_net structure */
342 struct list_head ni_netlist;
344 /* chain on net_ni_cpt */
345 struct list_head ni_cptlist;
352 /* bond NI on some CPTs */
355 /* interface's NID */
358 /* instance-specific data */
362 atomic_t ni_tx_credits;
364 /* percpt TX queues */
365 struct lnet_tx_queue **ni_tx_queues;
367 /* percpt reference count */
370 /* when I was last alive */
373 /* pointer to parent network */
374 struct lnet_net *ni_net;
376 /* my health status */
377 lnet_ni_status_t *ni_status;
380 enum lnet_ni_state ni_state;
382 /* per NI LND tunables */
383 struct lnet_lnd_tunables ni_lnd_tunables;
385 /* lnd tunables set explicitly */
386 bool ni_lnd_tunables_set;
389 struct lnet_element_stats ni_stats;
391 /* physical device CPT */
394 /* sequence number used to round robin over nis within a net */
398 * equivalent interfaces to use
399 * This is an array because socklnd bonding can still be configured
401 char *ni_interfaces[LNET_NUM_INTERFACES];
402 struct net *ni_net_ns; /* original net namespace */
405 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
408 * Descriptor of a ping info buffer: keep a separate indicator of the
409 * size and a reference count. The type is used both as a source and
410 * sink of data, so we need to keep some information outside of the
411 * area that may be overwritten by network data.
413 struct lnet_ping_buffer {
416 struct lnet_ping_info pb_info;
419 #define LNET_PING_BUFFER_SIZE(NNIDS) \
420 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
421 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
422 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
424 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
425 container_of((PINFO), struct lnet_ping_buffer, pb_info)
427 /* router checker data, per router */
429 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
430 struct list_head rcd_list;
431 lnet_handle_md_t rcd_mdh; /* ping buffer MD */
432 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
433 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
434 int rcd_nnis; /* desired size of buffer */
437 struct lnet_peer_ni {
438 /* chain on lpn_peer_nis */
439 struct list_head lpni_peer_nis;
440 /* chain on remote peer list */
441 struct list_head lpni_on_remote_peer_ni_list;
442 /* chain on peer hash */
443 struct list_head lpni_hashlist;
444 /* messages blocking for tx credits */
445 struct list_head lpni_txq;
446 /* messages blocking for router credits */
447 struct list_head lpni_rtrq;
448 /* chain on router list */
449 struct list_head lpni_rtr_list;
450 /* pointer to peer net I'm part of */
451 struct lnet_peer_net *lpni_peer_net;
452 /* statistics kept on each peer NI */
453 struct lnet_element_stats lpni_stats;
454 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
455 spinlock_t lpni_lock;
456 /* # tx credits available */
459 int lpni_mintxcredits;
460 /* # router credits */
463 int lpni_minrtrcredits;
464 /* bytes queued for sending */
468 /* notification outstanding? */
470 /* outstanding notification for LND? */
472 /* some thread is handling notification */
474 /* SEND event outstanding from ping */
475 bool lpni_ping_notsent;
476 /* # times router went dead<->alive. Protected with lpni_lock */
477 int lpni_alive_count;
478 /* time of last aliveness news */
479 cfs_time_t lpni_timestamp;
480 /* time of last ping attempt */
481 cfs_time_t lpni_ping_timestamp;
482 /* != 0 if ping reply expected */
483 cfs_time_t lpni_ping_deadline;
484 /* when I was last alive */
485 cfs_time_t lpni_last_alive;
486 /* when lpni_ni was queried last time */
487 cfs_time_t lpni_last_query;
488 /* network peer is on */
489 struct lnet_net *lpni_net;
493 atomic_t lpni_refcount;
494 /* CPT this peer attached on */
496 /* state flags -- protected by lpni_lock */
498 /* # refs from lnet_route_t::lr_gateway */
499 int lpni_rtr_refcount;
500 /* sequence number used to round robin over peer nis within a net */
502 /* sequence number used to round robin over gateways */
506 /* returned RC ping features. Protected with lpni_lock */
507 unsigned int lpni_ping_feats;
508 /* routes on this peer */
509 struct list_head lpni_routes;
510 /* preferred local nids: if only one, use lpni_pref.nid */
515 /* number of preferred NIDs in lnpi_pref_nids */
516 __u32 lpni_pref_nnids;
517 /* router checker state */
518 lnet_rc_data_t *lpni_rcd;
521 /* Preferred path added due to traffic on non-MR peer_ni */
522 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
525 /* chain on pt_peer_list */
526 struct list_head lp_peer_list;
528 /* list of peer nets */
529 struct list_head lp_peer_nets;
531 /* primary NID of the peer */
532 lnet_nid_t lp_primary_nid;
534 /* CPT of peer_table */
537 /* number of NIDs on this peer */
540 /* reference count */
541 atomic_t lp_refcount;
543 /* lock protecting peer state flags */
546 /* peer state flags */
549 /* buffer for data pushed by peer */
550 struct lnet_ping_buffer *lp_data;
552 /* MD handle for ping in progress */
553 lnet_handle_md_t lp_ping_mdh;
555 /* MD handle for push in progress */
556 lnet_handle_md_t lp_push_mdh;
558 /* number of NIDs for sizing push data */
561 /* NI config sequence number of peer */
564 /* Local NI config sequence number acked by peer */
567 /* Local NI config sequence number sent to peer */
568 __u32 lp_node_seqno_sent;
570 /* Ping error encountered during discovery. */
573 /* Push error encountered during discovery. */
576 /* Error encountered during discovery. */
579 /* link on discovery-related lists */
580 struct list_head lp_dc_list;
582 /* tasks waiting on discovery of this peer */
583 wait_queue_head_t lp_dc_waitq;
586 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
589 * A peer can be CONFIGURED by DLC, DISCOVERED by peer discovery, or
590 * UNDISCOVERED if discovery was disabled. The UNDISCOVERED state
591 * exists to keep a peer from being requeued to discovery.
593 * A peer that was created as the result of inbound traffic will not
596 #define LNET_PEER_CONFIGURED (1 << 1) /* Configured via DLC */
597 #define LNET_PEER_DISCOVERED (1 << 2) /* Peer was discovered */
598 #define LNET_PEER_UNDISCOVERED (1 << 3) /* Discovery was disabled */
600 #define LNET_PEER_QUEUED (1 << 4) /* Queued for discovery */
601 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
602 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
603 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
604 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
605 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
606 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
607 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
608 #define LNET_PEER_PING_REQUIRED (1 << 12) /* Must ping to fix up state */
610 struct lnet_peer_net {
611 /* chain on lp_peer_nets */
612 struct list_head lpn_peer_nets;
614 /* list of peer_nis on this network */
615 struct list_head lpn_peer_nis;
617 /* pointer to the peer I'm part of */
618 struct lnet_peer *lpn_peer;
623 /* reference count */
624 atomic_t lpn_refcount;
628 #define LNET_PEER_HASH_BITS 9
629 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
632 * peer hash table - one per CPT
634 * protected by lnet_net_lock/EX for update
641 * protected by pt_zombie_lock:
645 * pt_zombie lock nests inside lnet_net_lock
647 struct lnet_peer_table {
648 int pt_version; /* /proc validity stamp */
649 int pt_number; /* # peers_ni extant */
650 struct list_head *pt_hash; /* NID->peer hash */
651 struct list_head pt_peer_list; /* peers */
652 int pt_peers; /* # peers */
653 int pt_peer_nnids; /* # NIDS on listed peers */
654 struct list_head pt_zombie_list; /* zombie peer_ni */
655 int pt_zombies; /* # zombie peers_ni */
656 spinlock_t pt_zombie_lock; /* protect list and count */
659 /* peer aliveness is enabled only on routers for peers in a network where the
660 * lnet_ni_t::ni_peertimeout has been set to a positive value */
661 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
662 ((lp)->lpni_net) && \
663 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
666 struct list_head lr_list; /* chain on net */
667 struct list_head lr_gwlist; /* chain on gateway */
668 struct lnet_peer_ni *lr_gateway; /* router node */
669 __u32 lr_net; /* remote network number */
670 int lr_seq; /* sequence for round-robin */
671 unsigned int lr_downis; /* number of down NIs */
672 __u32 lr_hops; /* how far I am */
673 unsigned int lr_priority; /* route priority */
676 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
677 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
678 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
681 /* chain on ln_remote_nets_hash */
682 struct list_head lrn_list;
684 struct list_head lrn_routes;
689 /** lnet message has credit and can be submitted to lnd for send/receive */
690 #define LNET_CREDIT_OK 0
691 /** lnet message is waiting for credit */
692 #define LNET_CREDIT_WAIT 1
695 /* my free buffer pool */
696 struct list_head rbp_bufs;
697 /* messages blocking for a buffer */
698 struct list_head rbp_msgs;
699 /* # pages in each buffer */
701 /* requested number of buffers */
702 int rbp_req_nbuffers;
703 /* # buffers actually allocated */
705 /* # free buffers / blocked messages */
712 struct list_head rb_list; /* chain on rbp_bufs */
713 lnet_rtrbufpool_t *rb_pool; /* owning pool */
714 lnet_kiov_t rb_kiov[0]; /* the buffer space */
717 #define LNET_PEER_HASHSIZE 503 /* prime! */
720 /* Didn't match anything */
721 LNET_MATCHMD_NONE = (1 << 0),
723 LNET_MATCHMD_OK = (1 << 1),
724 /* Must be discarded */
725 LNET_MATCHMD_DROP = (1 << 2),
726 /* match and buffer is exhausted */
727 LNET_MATCHMD_EXHAUSTED = (1 << 3),
729 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
732 /* Options for lnet_portal_t::ptl_options */
733 #define LNET_PTL_LAZY (1 << 0)
734 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
735 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
737 /* parameter for matching operations (GET, PUT) */
738 struct lnet_match_info {
740 lnet_process_id_t mi_id;
743 unsigned int mi_portal;
744 unsigned int mi_rlength;
745 unsigned int mi_roffset;
748 /* ME hash of RDMA portal */
749 #define LNET_MT_HASH_BITS 8
750 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
751 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
752 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
753 * the last entry is reserved for MEs with ignore-bits */
754 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
755 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
756 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
757 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
758 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
759 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
760 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
762 /* portal match table */
763 struct lnet_match_table {
764 /* reserved for upcoming patches, CPU partition ID */
766 unsigned int mt_portal; /* portal index */
767 /* match table is set as "enabled" if there's non-exhausted MD
768 * attached on mt_mhash, it's only valid for wildcard portal */
769 unsigned int mt_enabled;
770 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
771 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
772 struct list_head *mt_mhash; /* matching hash */
775 /* these are only useful for wildcard portal */
776 /* Turn off message rotor for wildcard portals */
777 #define LNET_PTL_ROTOR_OFF 0
778 /* round-robin dispatch all PUT messages for wildcard portals */
779 #define LNET_PTL_ROTOR_ON 1
780 /* round-robin dispatch routed PUT message for wildcard portals */
781 #define LNET_PTL_ROTOR_RR_RT 2
782 /* dispatch routed PUT message by hashing source NID for wildcard portals */
783 #define LNET_PTL_ROTOR_HASH_RT 3
785 typedef struct lnet_portal {
787 unsigned int ptl_index; /* portal ID, reserved */
788 /* flags on this portal: lazy, unique... */
789 unsigned int ptl_options;
790 /* list of messages which are stealing buffer */
791 struct list_head ptl_msg_stealing;
792 /* messages blocking for MD */
793 struct list_head ptl_msg_delayed;
794 /* Match table for each CPT */
795 struct lnet_match_table **ptl_mtables;
796 /* spread rotor of incoming "PUT" */
797 unsigned int ptl_rotor;
798 /* # active entries for this portal */
800 /* array of active entries' cpu-partition-id */
804 #define LNET_LH_HASH_BITS 12
805 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
806 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
808 /* resource container (ME, MD, EQ) */
809 struct lnet_res_container {
810 unsigned int rec_type; /* container type */
811 __u64 rec_lh_cookie; /* cookie generator */
812 struct list_head rec_active; /* active resource list */
813 struct list_head *rec_lh_hash; /* handle hash */
816 /* message container */
817 struct lnet_msg_container {
818 int msc_init; /* initialized or not */
819 /* max # threads finalizing */
821 /* msgs waiting to complete finalizing */
822 struct list_head msc_finalizing;
823 struct list_head msc_active; /* active message list */
824 /* threads doing finalization */
825 void **msc_finalizers;
828 /* Peer Discovery states */
829 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
830 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
831 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
833 /* Router Checker states */
834 #define LNET_RC_STATE_SHUTDOWN 0 /* not started */
835 #define LNET_RC_STATE_RUNNING 1 /* started up OK */
836 #define LNET_RC_STATE_STOPPING 2 /* telling thread to stop */
839 #define LNET_STATE_SHUTDOWN 0 /* not started */
840 #define LNET_STATE_RUNNING 1 /* started up OK */
841 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
845 /* CPU partition table of LNet */
846 struct cfs_cpt_table *ln_cpt_table;
847 /* number of CPTs in ln_cpt_table */
848 unsigned int ln_cpt_number;
849 unsigned int ln_cpt_bits;
851 /* protect LNet resources (ME/MD/EQ) */
852 struct cfs_percpt_lock *ln_res_lock;
855 /* the vector of portals */
856 lnet_portal_t **ln_portals;
857 /* percpt ME containers */
858 struct lnet_res_container **ln_me_containers;
859 /* percpt MD container */
860 struct lnet_res_container **ln_md_containers;
862 /* Event Queue container */
863 struct lnet_res_container ln_eq_container;
864 wait_queue_head_t ln_eq_waitq;
865 spinlock_t ln_eq_wait_lock;
867 unsigned int ln_remote_nets_hbits;
869 /* protect NI, peer table, credits, routers, rtrbuf... */
870 struct cfs_percpt_lock *ln_net_lock;
871 /* percpt message containers for active/finalizing/freed message */
872 struct lnet_msg_container **ln_msg_containers;
873 lnet_counters_t **ln_counters;
874 struct lnet_peer_table **ln_peer_tables;
875 /* list of peer nis not on a local network */
876 struct list_head ln_remote_peer_ni_list;
877 /* failure simulation */
878 struct list_head ln_test_peers;
879 struct list_head ln_drop_rules;
880 struct list_head ln_delay_rules;
882 struct list_head ln_nets;
883 /* the loopback NI */
884 struct lnet_ni *ln_loni;
885 /* network zombie list */
886 struct list_head ln_net_zombie;
888 /* remote networks with routes to them */
889 struct list_head *ln_remote_nets_hash;
891 __u64 ln_remote_nets_version;
892 /* list of all known routers */
893 struct list_head ln_routers;
895 __u64 ln_routers_version;
896 /* percpt router buffer pools */
897 lnet_rtrbufpool_t **ln_rtrpools;
900 * Ping target / Push source
902 * The ping target and push source share a single buffer. The
903 * ln_ping_target is protected against concurrent updates by
906 lnet_handle_eq_t ln_ping_target_eq;
907 lnet_handle_md_t ln_ping_target_md;
908 struct lnet_ping_buffer *ln_ping_target;
909 atomic_t ln_ping_target_seqno;
914 * ln_push_nnis contains the desired size of the push target.
915 * The lnet_net_lock is used to handle update races. The old
916 * buffer may linger a while after it has been unlinked, in
917 * which case the event handler cleans up.
919 lnet_handle_eq_t ln_push_target_eq;
920 lnet_handle_md_t ln_push_target_md;
921 struct lnet_ping_buffer *ln_push_target;
922 int ln_push_target_nnis;
924 /* discovery event queue handle */
925 lnet_handle_eq_t ln_dc_eqh;
926 /* discovery requests */
927 struct list_head ln_dc_request;
928 /* discovery working list */
929 struct list_head ln_dc_working;
930 /* discovery thread wait queue */
931 wait_queue_head_t ln_dc_waitq;
932 /* discovery startup/shutdown state */
935 /* router checker startup/shutdown state */
937 /* router checker's event queue */
938 lnet_handle_eq_t ln_rc_eqh;
939 /* rcd still pending on net */
940 struct list_head ln_rcd_deathrow;
941 /* rcd ready for free */
942 struct list_head ln_rcd_zombie;
943 /* serialise startup/shutdown */
944 struct semaphore ln_rc_signal;
946 struct mutex ln_api_mutex;
947 struct mutex ln_lnd_mutex;
948 /* Have I called LNetNIInit myself? */
950 /* LNetNIInit/LNetNIFini counter */
952 /* SHUTDOWN/RUNNING/STOPPING */
955 int ln_routing; /* am I a router? */
956 lnet_pid_t ln_pid; /* requested pid */
957 /* uniquely identifies this ni in this epoch */
958 __u64 ln_interface_cookie;
959 /* registered LNDs */
960 struct list_head ln_lnds;
962 /* test protocol compatibility flags */
963 int ln_testprotocompat;
965 /* 0 - load the NIs from the mod params
966 * 1 - do not load the NIs from the mod params
967 * Reverse logic to ensure that other calls to LNetNIInit
970 bool ln_nis_from_mod_params;
972 /* waitq for router checker. As long as there are no routes in
973 * the list, the router checker will sleep on this queue. when
974 * routes are added the thread will wake up */
975 wait_queue_head_t ln_rc_waitq;