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, 2017, 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/semaphore.h>
48 #include <linux/types.h>
50 #include <uapi/linux/lnet/lnet-dlc.h>
51 #include <uapi/linux/lnet/lnetctl.h>
53 /* Max payload size */
54 #define LNET_MAX_PAYLOAD LNET_MTU
56 #define LNET_MAX_IOV (LNET_MAX_PAYLOAD >> PAGE_SHIFT)
59 * This is the maximum health value.
60 * All local and peer NIs created have their health default to this value.
62 #define LNET_MAX_HEALTH_VALUE 1000
67 enum lnet_msg_hstatus {
68 LNET_MSG_STATUS_OK = 0,
69 LNET_MSG_STATUS_LOCAL_INTERRUPT,
70 LNET_MSG_STATUS_LOCAL_DROPPED,
71 LNET_MSG_STATUS_LOCAL_ABORTED,
72 LNET_MSG_STATUS_LOCAL_NO_ROUTE,
73 LNET_MSG_STATUS_LOCAL_ERROR,
74 LNET_MSG_STATUS_LOCAL_TIMEOUT,
75 LNET_MSG_STATUS_REMOTE_ERROR,
76 LNET_MSG_STATUS_REMOTE_DROPPED,
77 LNET_MSG_STATUS_REMOTE_TIMEOUT,
78 LNET_MSG_STATUS_NETWORK_TIMEOUT,
82 struct lnet_rsp_tracker {
83 /* chain on the waiting list */
84 struct list_head rspt_on_list;
87 /* deadline of the REPLY/ACK */
88 ktime_t rspt_deadline;
90 struct lnet_handle_md rspt_mdh;
94 struct list_head msg_activelist;
95 struct list_head msg_list; /* Q for credits/MD */
97 struct lnet_process_id msg_target;
98 /* Primary NID of the source. */
99 lnet_nid_t msg_initiator;
100 /* where is it from, it's only for building event */
105 * hold parameters in case message is with held due
108 lnet_nid_t msg_src_nid_param;
109 lnet_nid_t msg_rtr_nid_param;
112 * Deadline for the message after which it will be finalized if it
115 ktime_t msg_deadline;
117 /* The message health status. */
118 enum lnet_msg_hstatus msg_health_status;
119 /* This is a recovery message */
121 /* the number of times a transmission has been retried */
123 /* flag to indicate that we do not want to resend this message */
126 /* committed for sending */
127 unsigned int msg_tx_committed:1;
128 /* CPT # this message committed for sending */
129 unsigned int msg_tx_cpt:15;
130 /* committed for receiving */
131 unsigned int msg_rx_committed:1;
132 /* CPT # this message committed for receiving */
133 unsigned int msg_rx_cpt:15;
134 /* queued for tx credit */
135 unsigned int msg_tx_delayed:1;
136 /* queued for RX buffer */
137 unsigned int msg_rx_delayed:1;
138 /* ready for pending on RX delay list */
139 unsigned int msg_rx_ready_delay:1;
141 unsigned int msg_vmflush:1; /* VM trying to free memory */
142 unsigned int msg_target_is_router:1; /* sending to a router */
143 unsigned int msg_routing:1; /* being forwarded */
144 unsigned int msg_ack:1; /* ack on finalize (PUT) */
145 unsigned int msg_sending:1; /* outgoing message */
146 unsigned int msg_receiving:1; /* being received */
147 unsigned int msg_txcredit:1; /* taken an NI send credit */
148 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
149 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
150 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
151 unsigned int msg_onactivelist:1; /* on the activelist */
152 unsigned int msg_rdma_get:1;
154 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
155 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
158 struct lnet_libmd *msg_md;
159 /* the NI the message was sent or received over */
160 struct lnet_ni *msg_txni;
161 struct lnet_ni *msg_rxni;
163 unsigned int msg_len;
164 unsigned int msg_wanted;
165 unsigned int msg_offset;
166 unsigned int msg_niov;
167 struct kvec *msg_iov;
168 lnet_kiov_t *msg_kiov;
170 struct lnet_event msg_ev;
171 struct lnet_hdr msg_hdr;
174 struct lnet_libhandle {
175 struct list_head lh_hash_chain;
179 #define lh_entry(ptr, type, member) \
180 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
183 struct list_head eq_list;
184 struct lnet_libhandle eq_lh;
185 unsigned long eq_enq_seq;
186 unsigned long eq_deq_seq;
187 unsigned int eq_size;
188 lnet_eq_handler_t eq_callback;
189 struct lnet_event *eq_events;
190 int **eq_refs; /* percpt refcount for EQ */
194 struct list_head me_list;
195 struct lnet_libhandle me_lh;
196 struct lnet_process_id me_match_id;
197 unsigned int me_portal;
198 unsigned int me_pos; /* hash offset in mt_hash */
200 __u64 me_ignore_bits;
201 enum lnet_unlink me_unlink;
202 struct lnet_libmd *me_md;
206 struct list_head md_list;
207 struct lnet_libhandle md_lh;
208 struct lnet_me *md_me;
210 unsigned int md_offset;
211 unsigned int md_length;
212 unsigned int md_max_size;
215 unsigned int md_options;
216 unsigned int md_flags;
217 unsigned int md_niov; /* # frags at end of struct */
219 struct lnet_rsp_tracker *md_rspt_ptr;
220 struct lnet_eq *md_eq;
221 struct lnet_handle_md md_bulk_handle;
223 struct kvec iov[LNET_MAX_IOV];
224 lnet_kiov_t kiov[LNET_MAX_IOV];
228 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
229 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
230 #define LNET_MD_FLAG_ABORTED (1 << 2)
232 struct lnet_test_peer {
233 /* info about peers we are trying to fail */
234 struct list_head tp_list; /* ln_test_peers */
235 lnet_nid_t tp_nid; /* matching nid */
236 unsigned int tp_threshold; /* # failures to simulate */
239 #define LNET_COOKIE_TYPE_MD 1
240 #define LNET_COOKIE_TYPE_ME 2
241 #define LNET_COOKIE_TYPE_EQ 3
242 #define LNET_COOKIE_TYPE_BITS 2
243 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
245 struct lnet_ni; /* forward ref */
249 /* fields managed by portals */
250 struct list_head lnd_list; /* stash in the LND table */
251 int lnd_refcount; /* # active instances */
253 /* fields initialized by the LND */
256 int (*lnd_startup)(struct lnet_ni *ni);
257 void (*lnd_shutdown)(struct lnet_ni *ni);
258 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
260 /* In data movement APIs below, payload buffers are described as a set
261 * of 'niov' fragments which are...
263 * in virtual memory (struct kvec *iov != NULL)
265 * in pages (kernel only: plt_kiov_t *kiov != NULL).
266 * The LND may NOT overwrite these fragment descriptors.
267 * An 'offset' and may specify a byte offset within the set of
268 * fragments to start from
271 /* Start sending a preformatted message. 'private' is NULL for PUT and
272 * GET messages; otherwise this is a response to an incoming message
273 * and 'private' is the 'private' passed to lnet_parse(). Return
274 * non-zero for immediate failure, otherwise complete later with
276 int (*lnd_send)(struct lnet_ni *ni, void *private,
277 struct lnet_msg *msg);
279 /* Start receiving 'mlen' bytes of payload data, skipping the following
280 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
281 * lnet_parse(). Return non-zero for immedaite failure, otherwise
282 * complete later with lnet_finalize(). This also gives back a receive
283 * credit if the LND does flow control. */
284 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
285 int delayed, unsigned int niov,
286 struct kvec *iov, lnet_kiov_t *kiov,
287 unsigned int offset, unsigned int mlen, unsigned int rlen);
289 /* lnet_parse() has had to delay processing of this message
290 * (e.g. waiting for a forwarding buffer or send credits). Give the
291 * LND a chance to free urgently needed resources. If called, return 0
292 * for success and do NOT give back a receive credit; that has to wait
293 * until lnd_recv() gets called. On failure return < 0 and
294 * release resources; lnd_recv() will not be called. */
295 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
296 struct lnet_msg *msg, void **new_privatep);
298 /* notification of peer health */
299 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
301 /* query of peer aliveness */
302 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, time64_t *when);
304 /* accept a new connection */
305 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
308 struct lnet_tx_queue {
309 int tq_credits; /* # tx credits free */
310 int tq_credits_min; /* lowest it's been */
311 int tq_credits_max; /* total # tx credits */
312 struct list_head tq_delayed; /* delayed TXs */
315 enum lnet_net_state {
316 /* set when net block is allocated */
317 LNET_NET_STATE_INIT = 0,
318 /* set when NIs in net are started successfully */
319 LNET_NET_STATE_ACTIVE,
320 /* set if all NIs in net are in FAILED state */
321 LNET_NET_STATE_INACTIVE,
322 /* set when shutting down a NET */
323 LNET_NET_STATE_DELETING
326 #define LNET_NI_STATE_INIT (1 << 0)
327 #define LNET_NI_STATE_ACTIVE (1 << 1)
328 #define LNET_NI_STATE_FAILED (1 << 2)
329 #define LNET_NI_STATE_RECOVERY_PENDING (1 << 3)
330 #define LNET_NI_STATE_DELETING (1 << 4)
332 enum lnet_stats_type {
333 LNET_STATS_TYPE_SEND = 0,
334 LNET_STATS_TYPE_RECV,
338 struct lnet_comm_count {
339 atomic_t co_get_count;
340 atomic_t co_put_count;
341 atomic_t co_reply_count;
342 atomic_t co_ack_count;
343 atomic_t co_hello_count;
346 struct lnet_element_stats {
347 struct lnet_comm_count el_send_stats;
348 struct lnet_comm_count el_recv_stats;
349 struct lnet_comm_count el_drop_stats;
352 struct lnet_health_local_stats {
353 atomic_t hlt_local_interrupt;
354 atomic_t hlt_local_dropped;
355 atomic_t hlt_local_aborted;
356 atomic_t hlt_local_no_route;
357 atomic_t hlt_local_timeout;
358 atomic_t hlt_local_error;
361 struct lnet_health_remote_stats {
362 atomic_t hlt_remote_dropped;
363 atomic_t hlt_remote_timeout;
364 atomic_t hlt_remote_error;
365 atomic_t hlt_network_timeout;
369 /* chain on the ln_nets */
370 struct list_head net_list;
372 /* net ID, which is composed of
373 * (net_type << 16) | net_num.
374 * net_type can be one of the enumerated types defined in
375 * lnet/include/lnet/nidstr.h */
378 /* priority of the network */
381 /* total number of CPTs in the array */
384 /* cumulative CPTs of all NIs in this net */
387 /* network tunables */
388 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
391 * boolean to indicate that the tunables have been set and
394 bool net_tunables_set;
396 /* procedural interface */
397 struct lnet_lnd *net_lnd;
399 /* list of NIs on this net */
400 struct list_head net_ni_list;
402 /* list of NIs being added, but not started yet */
403 struct list_head net_ni_added;
405 /* dying LND instances */
406 struct list_head net_ni_zombie;
409 enum lnet_net_state net_state;
413 /* chain on the lnet_net structure */
414 struct list_head ni_netlist;
416 /* chain on the recovery queue */
417 struct list_head ni_recovery;
419 /* MD handle for recovery ping */
420 struct lnet_handle_md ni_ping_mdh;
427 /* bond NI on some CPTs */
430 /* interface's NID */
433 /* instance-specific data */
437 atomic_t ni_tx_credits;
439 /* percpt TX queues */
440 struct lnet_tx_queue **ni_tx_queues;
442 /* percpt reference count */
445 /* when I was last alive */
446 time64_t ni_last_alive;
448 /* pointer to parent network */
449 struct lnet_net *ni_net;
451 /* my health status */
452 struct lnet_ni_status *ni_status;
457 /* per NI LND tunables */
458 struct lnet_lnd_tunables ni_lnd_tunables;
460 /* lnd tunables set explicitly */
461 bool ni_lnd_tunables_set;
464 struct lnet_element_stats ni_stats;
465 struct lnet_health_local_stats ni_hstats;
467 /* physical device CPT */
470 /* sequence number used to round robin over nis within a net */
475 * initialized to LNET_MAX_HEALTH_VALUE
476 * Value is decremented every time we fail to send a message over
477 * this NI because of a NI specific failure.
478 * Value is incremented if we successfully send a message.
483 * Set to 1 by the LND when it receives an event telling it the device
484 * has gone into a fatal state. Set to 0 when the LND receives an
485 * even telling it the device is back online.
487 atomic_t ni_fatal_error_on;
490 * equivalent interfaces to use
491 * This is an array because socklnd bonding can still be configured
493 char *ni_interfaces[LNET_INTERFACES_NUM];
494 struct net *ni_net_ns; /* original net namespace */
497 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
500 * Descriptor of a ping info buffer: keep a separate indicator of the
501 * size and a reference count. The type is used both as a source and
502 * sink of data, so we need to keep some information outside of the
503 * area that may be overwritten by network data.
505 struct lnet_ping_buffer {
508 struct lnet_ping_info pb_info;
511 #define LNET_PING_BUFFER_SIZE(NNIDS) \
512 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
513 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
514 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
516 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
517 container_of((PINFO), struct lnet_ping_buffer, pb_info)
519 /* router checker data, per router */
520 struct lnet_rc_data {
521 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
522 struct list_head rcd_list;
523 struct lnet_handle_md rcd_mdh; /* ping buffer MD */
524 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
525 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
526 int rcd_nnis; /* desired size of buffer */
529 struct lnet_peer_ni {
530 /* chain on lpn_peer_nis */
531 struct list_head lpni_peer_nis;
532 /* chain on remote peer list */
533 struct list_head lpni_on_remote_peer_ni_list;
534 /* chain on recovery queue */
535 struct list_head lpni_recovery;
536 /* chain on peer hash */
537 struct list_head lpni_hashlist;
538 /* messages blocking for tx credits */
539 struct list_head lpni_txq;
540 /* messages blocking for router credits */
541 struct list_head lpni_rtrq;
542 /* chain on router list */
543 struct list_head lpni_rtr_list;
544 /* pointer to peer net I'm part of */
545 struct lnet_peer_net *lpni_peer_net;
546 /* statistics kept on each peer NI */
547 struct lnet_element_stats lpni_stats;
548 struct lnet_health_remote_stats lpni_hstats;
549 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
550 spinlock_t lpni_lock;
551 /* # tx credits available */
554 int lpni_mintxcredits;
555 /* # router credits */
558 int lpni_minrtrcredits;
559 /* bytes queued for sending */
563 /* notification outstanding? */
565 /* outstanding notification for LND? */
567 /* some thread is handling notification */
569 /* SEND event outstanding from ping */
570 bool lpni_ping_notsent;
571 /* # times router went dead<->alive. Protected with lpni_lock */
572 int lpni_alive_count;
573 /* time of last aliveness news */
574 time64_t lpni_timestamp;
575 /* time of last ping attempt */
576 time64_t lpni_ping_timestamp;
577 /* != 0 if ping reply expected */
578 time64_t lpni_ping_deadline;
579 /* when I was last alive */
580 time64_t lpni_last_alive;
581 /* when lpni_ni was queried last time */
582 time64_t lpni_last_query;
583 /* network peer is on */
584 struct lnet_net *lpni_net;
588 atomic_t lpni_refcount;
589 /* health value for the peer */
590 atomic_t lpni_healthv;
591 /* recovery ping mdh */
592 struct lnet_handle_md lpni_recovery_ping_mdh;
593 /* CPT this peer attached on */
595 /* state flags -- protected by lpni_lock */
597 /* # refs from lnet_route_t::lr_gateway */
598 int lpni_rtr_refcount;
599 /* sequence number used to round robin over peer nis within a net */
601 /* sequence number used to round robin over gateways */
605 /* returned RC ping features. Protected with lpni_lock */
606 unsigned int lpni_ping_feats;
607 /* routes on this peer */
608 struct list_head lpni_routes;
609 /* preferred local nids: if only one, use lpni_pref.nid */
614 /* number of preferred NIDs in lnpi_pref_nids */
615 __u32 lpni_pref_nnids;
616 /* router checker state */
617 struct lnet_rc_data *lpni_rcd;
620 /* Preferred path added due to traffic on non-MR peer_ni */
621 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
622 /* peer is being recovered. */
623 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
624 /* peer is being deleted */
625 #define LNET_PEER_NI_DELETING (1 << 2)
628 /* chain on pt_peer_list */
629 struct list_head lp_peer_list;
631 /* list of peer nets */
632 struct list_head lp_peer_nets;
634 /* list of messages pending discovery*/
635 struct list_head lp_dc_pendq;
637 /* primary NID of the peer */
638 lnet_nid_t lp_primary_nid;
640 /* CPT of peer_table */
643 /* number of NIDs on this peer */
646 /* reference count */
647 atomic_t lp_refcount;
649 /* lock protecting peer state flags */
652 /* peer state flags */
655 /* buffer for data pushed by peer */
656 struct lnet_ping_buffer *lp_data;
658 /* MD handle for ping in progress */
659 struct lnet_handle_md lp_ping_mdh;
661 /* MD handle for push in progress */
662 struct lnet_handle_md lp_push_mdh;
664 /* number of NIDs for sizing push data */
667 /* NI config sequence number of peer */
670 /* Local NI config sequence number acked by peer */
673 /* Local NI config sequence number sent to peer */
674 __u32 lp_node_seqno_sent;
676 /* Ping error encountered during discovery. */
679 /* Push error encountered during discovery. */
682 /* Error encountered during discovery. */
685 /* time it was put on the ln_dc_working queue */
686 time64_t lp_last_queued;
688 /* link on discovery-related lists */
689 struct list_head lp_dc_list;
691 /* tasks waiting on discovery of this peer */
692 wait_queue_head_t lp_dc_waitq;
696 * The status flags in lp_state. Their semantics have chosen so that
697 * lp_state can be zero-initialized.
699 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
700 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
702 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
703 * NOT set when the peer was pinged by discovery.
705 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
706 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
708 * A peer is marked CONFIGURED if it was configured by DLC.
710 * In addition, a peer is marked DISCOVERED if it has fully passed
711 * through Peer Discovery.
713 * When Peer Discovery is disabled, the discovery thread will mark
714 * peers REDISCOVER to indicate that they should be re-examined if
715 * discovery is (re)enabled on the node.
717 * A peer that was created as the result of inbound traffic will not
720 #define LNET_PEER_CONFIGURED (1 << 2) /* Configured via DLC */
721 #define LNET_PEER_DISCOVERED (1 << 3) /* Peer was discovered */
722 #define LNET_PEER_REDISCOVER (1 << 4) /* Discovery was disabled */
724 * A peer is marked DISCOVERING when discovery is in progress.
725 * The other flags below correspond to stages of discovery.
727 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
728 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
729 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
730 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
731 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
732 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
733 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
735 * A ping can be forced as a way to fix up state, or as a manual
736 * intervention by an admin.
737 * A push can be forced in circumstances that would normally not
738 * allow for one to happen.
740 #define LNET_PEER_FORCE_PING (1 << 12) /* Forced Ping */
741 #define LNET_PEER_FORCE_PUSH (1 << 13) /* Forced Push */
743 struct lnet_peer_net {
744 /* chain on lp_peer_nets */
745 struct list_head lpn_peer_nets;
747 /* list of peer_nis on this network */
748 struct list_head lpn_peer_nis;
750 /* pointer to the peer I'm part of */
751 struct lnet_peer *lpn_peer;
756 /* reference count */
757 atomic_t lpn_refcount;
761 #define LNET_PEER_HASH_BITS 9
762 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
765 * peer hash table - one per CPT
767 * protected by lnet_net_lock/EX for update
773 * protected by pt_zombie_lock:
777 * pt_zombie lock nests inside lnet_net_lock
779 struct lnet_peer_table {
780 int pt_version; /* /proc validity stamp */
781 int pt_number; /* # peers_ni extant */
782 struct list_head *pt_hash; /* NID->peer hash */
783 struct list_head pt_peer_list; /* peers */
784 int pt_peers; /* # peers */
785 struct list_head pt_zombie_list; /* zombie peer_ni */
786 int pt_zombies; /* # zombie peers_ni */
787 spinlock_t pt_zombie_lock; /* protect list and count */
790 /* peer aliveness is enabled only on routers for peers in a network where the
791 * struct lnet_ni::ni_peertimeout has been set to a positive value
793 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
794 ((lp)->lpni_net) && \
795 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
798 struct list_head lr_list; /* chain on net */
799 struct list_head lr_gwlist; /* chain on gateway */
800 struct lnet_peer_ni *lr_gateway; /* router node */
801 __u32 lr_net; /* remote network number */
802 int lr_seq; /* sequence for round-robin */
803 unsigned int lr_downis; /* number of down NIs */
804 __u32 lr_hops; /* how far I am */
805 unsigned int lr_priority; /* route priority */
808 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
809 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
810 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
812 struct lnet_remotenet {
813 /* chain on ln_remote_nets_hash */
814 struct list_head lrn_list;
816 struct list_head lrn_routes;
821 /** lnet message has credit and can be submitted to lnd for send/receive */
822 #define LNET_CREDIT_OK 0
823 /** lnet message is waiting for credit */
824 #define LNET_CREDIT_WAIT 1
825 /** lnet message is waiting for discovery */
826 #define LNET_DC_WAIT 2
828 struct lnet_rtrbufpool {
829 /* my free buffer pool */
830 struct list_head rbp_bufs;
831 /* messages blocking for a buffer */
832 struct list_head rbp_msgs;
833 /* # pages in each buffer */
835 /* requested number of buffers */
836 int rbp_req_nbuffers;
837 /* # buffers actually allocated */
839 /* # free buffers / blocked messages */
846 struct list_head rb_list; /* chain on rbp_bufs */
847 struct lnet_rtrbufpool *rb_pool; /* owning pool */
848 lnet_kiov_t rb_kiov[0]; /* the buffer space */
851 #define LNET_PEER_HASHSIZE 503 /* prime! */
853 enum lnet_match_flags {
854 /* Didn't match anything */
855 LNET_MATCHMD_NONE = (1 << 0),
857 LNET_MATCHMD_OK = (1 << 1),
858 /* Must be discarded */
859 LNET_MATCHMD_DROP = (1 << 2),
860 /* match and buffer is exhausted */
861 LNET_MATCHMD_EXHAUSTED = (1 << 3),
863 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
866 /* Options for struct lnet_portal::ptl_options */
867 #define LNET_PTL_LAZY (1 << 0)
868 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
869 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
871 /* parameter for matching operations (GET, PUT) */
872 struct lnet_match_info {
874 struct lnet_process_id mi_id;
877 unsigned int mi_portal;
878 unsigned int mi_rlength;
879 unsigned int mi_roffset;
882 /* ME hash of RDMA portal */
883 #define LNET_MT_HASH_BITS 8
884 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
885 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
886 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
887 * the last entry is reserved for MEs with ignore-bits */
888 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
889 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
890 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
891 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
892 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
893 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
894 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
896 /* portal match table */
897 struct lnet_match_table {
898 /* reserved for upcoming patches, CPU partition ID */
900 unsigned int mt_portal; /* portal index */
901 /* match table is set as "enabled" if there's non-exhausted MD
902 * attached on mt_mhash, it's only valid for wildcard portal */
903 unsigned int mt_enabled;
904 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
905 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
906 struct list_head *mt_mhash; /* matching hash */
909 /* these are only useful for wildcard portal */
910 /* Turn off message rotor for wildcard portals */
911 #define LNET_PTL_ROTOR_OFF 0
912 /* round-robin dispatch all PUT messages for wildcard portals */
913 #define LNET_PTL_ROTOR_ON 1
914 /* round-robin dispatch routed PUT message for wildcard portals */
915 #define LNET_PTL_ROTOR_RR_RT 2
916 /* dispatch routed PUT message by hashing source NID for wildcard portals */
917 #define LNET_PTL_ROTOR_HASH_RT 3
921 unsigned int ptl_index; /* portal ID, reserved */
922 /* flags on this portal: lazy, unique... */
923 unsigned int ptl_options;
924 /* list of messages which are stealing buffer */
925 struct list_head ptl_msg_stealing;
926 /* messages blocking for MD */
927 struct list_head ptl_msg_delayed;
928 /* Match table for each CPT */
929 struct lnet_match_table **ptl_mtables;
930 /* spread rotor of incoming "PUT" */
931 unsigned int ptl_rotor;
932 /* # active entries for this portal */
934 /* array of active entries' cpu-partition-id */
938 #define LNET_LH_HASH_BITS 12
939 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
940 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
942 /* resource container (ME, MD, EQ) */
943 struct lnet_res_container {
944 unsigned int rec_type; /* container type */
945 __u64 rec_lh_cookie; /* cookie generator */
946 struct list_head rec_active; /* active resource list */
947 struct list_head *rec_lh_hash; /* handle hash */
950 /* message container */
951 struct lnet_msg_container {
952 int msc_init; /* initialized or not */
953 /* max # threads finalizing */
955 /* msgs waiting to complete finalizing */
956 struct list_head msc_finalizing;
957 struct list_head msc_active; /* active message list */
958 /* threads doing finalization */
959 void **msc_finalizers;
962 /* Peer Discovery states */
963 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
964 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
965 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
967 /* Router Checker states */
968 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
969 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
970 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
973 #define LNET_STATE_SHUTDOWN 0 /* not started */
974 #define LNET_STATE_RUNNING 1 /* started up OK */
975 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
978 /* CPU partition table of LNet */
979 struct cfs_cpt_table *ln_cpt_table;
980 /* number of CPTs in ln_cpt_table */
981 unsigned int ln_cpt_number;
982 unsigned int ln_cpt_bits;
984 /* protect LNet resources (ME/MD/EQ) */
985 struct cfs_percpt_lock *ln_res_lock;
988 /* the vector of portals */
989 struct lnet_portal **ln_portals;
990 /* percpt ME containers */
991 struct lnet_res_container **ln_me_containers;
992 /* percpt MD container */
993 struct lnet_res_container **ln_md_containers;
995 /* Event Queue container */
996 struct lnet_res_container ln_eq_container;
997 wait_queue_head_t ln_eq_waitq;
998 spinlock_t ln_eq_wait_lock;
1000 unsigned int ln_remote_nets_hbits;
1002 /* protect NI, peer table, credits, routers, rtrbuf... */
1003 struct cfs_percpt_lock *ln_net_lock;
1004 /* percpt message containers for active/finalizing/freed message */
1005 struct lnet_msg_container **ln_msg_containers;
1006 struct lnet_counters **ln_counters;
1007 struct lnet_peer_table **ln_peer_tables;
1008 /* list of peer nis not on a local network */
1009 struct list_head ln_remote_peer_ni_list;
1010 /* failure simulation */
1011 struct list_head ln_test_peers;
1012 struct list_head ln_drop_rules;
1013 struct list_head ln_delay_rules;
1015 struct list_head ln_nets;
1016 /* the loopback NI */
1017 struct lnet_ni *ln_loni;
1018 /* network zombie list */
1019 struct list_head ln_net_zombie;
1020 /* resend messages list */
1021 struct list_head ln_msg_resend;
1022 /* spin lock to protect the msg resend list */
1023 spinlock_t ln_msg_resend_lock;
1025 /* remote networks with routes to them */
1026 struct list_head *ln_remote_nets_hash;
1027 /* validity stamp */
1028 __u64 ln_remote_nets_version;
1029 /* list of all known routers */
1030 struct list_head ln_routers;
1031 /* validity stamp */
1032 __u64 ln_routers_version;
1033 /* percpt router buffer pools */
1034 struct lnet_rtrbufpool **ln_rtrpools;
1037 * Ping target / Push source
1039 * The ping target and push source share a single buffer. The
1040 * ln_ping_target is protected against concurrent updates by
1043 struct lnet_handle_md ln_ping_target_md;
1044 struct lnet_handle_eq ln_ping_target_eq;
1045 struct lnet_ping_buffer *ln_ping_target;
1046 atomic_t ln_ping_target_seqno;
1051 * ln_push_nnis contains the desired size of the push target.
1052 * The lnet_net_lock is used to handle update races. The old
1053 * buffer may linger a while after it has been unlinked, in
1054 * which case the event handler cleans up.
1056 struct lnet_handle_eq ln_push_target_eq;
1057 struct lnet_handle_md ln_push_target_md;
1058 struct lnet_ping_buffer *ln_push_target;
1059 int ln_push_target_nnis;
1061 /* discovery event queue handle */
1062 struct lnet_handle_eq ln_dc_eqh;
1063 /* discovery requests */
1064 struct list_head ln_dc_request;
1065 /* discovery working list */
1066 struct list_head ln_dc_working;
1067 /* discovery expired list */
1068 struct list_head ln_dc_expired;
1069 /* discovery thread wait queue */
1070 wait_queue_head_t ln_dc_waitq;
1071 /* discovery startup/shutdown state */
1074 /* monitor thread startup/shutdown state */
1076 /* router checker's event queue */
1077 struct lnet_handle_eq ln_rc_eqh;
1078 /* rcd still pending on net */
1079 struct list_head ln_rcd_deathrow;
1080 /* rcd ready for free */
1081 struct list_head ln_rcd_zombie;
1082 /* serialise startup/shutdown */
1083 struct semaphore ln_mt_signal;
1085 struct mutex ln_api_mutex;
1086 struct mutex ln_lnd_mutex;
1087 /* Have I called LNetNIInit myself? */
1089 /* LNetNIInit/LNetNIFini counter */
1091 /* SHUTDOWN/RUNNING/STOPPING */
1094 int ln_routing; /* am I a router? */
1095 lnet_pid_t ln_pid; /* requested pid */
1096 /* uniquely identifies this ni in this epoch */
1097 __u64 ln_interface_cookie;
1098 /* registered LNDs */
1099 struct list_head ln_lnds;
1101 /* test protocol compatibility flags */
1102 int ln_testprotocompat;
1104 /* 0 - load the NIs from the mod params
1105 * 1 - do not load the NIs from the mod params
1106 * Reverse logic to ensure that other calls to LNetNIInit
1109 bool ln_nis_from_mod_params;
1112 * waitq for the monitor thread. The monitor thread takes care of
1113 * checking routes, timedout messages and resending messages.
1115 wait_queue_head_t ln_mt_waitq;
1117 /* per-cpt resend queues */
1118 struct list_head **ln_mt_resendqs;
1119 /* local NIs to recover */
1120 struct list_head ln_mt_localNIRecovq;
1121 /* local NIs to recover */
1122 struct list_head ln_mt_peerNIRecovq;
1124 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1125 * There are CPT number of queues. Since response trackers will be
1126 * added on the fast path we can't afford to grab the exclusive
1127 * net lock to protect these queues. The CPT will be calculated
1128 * based on the mdh cookie.
1130 struct list_head **ln_mt_rstq;
1131 /* recovery eq handler */
1132 struct lnet_handle_eq ln_mt_eqh;