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;
88 lnet_nid_t rspt_next_hop_nid;
89 /* deadline of the REPLY/ACK */
90 ktime_t rspt_deadline;
92 struct lnet_handle_md rspt_mdh;
96 struct list_head msg_activelist;
97 struct list_head msg_list; /* Q for credits/MD */
99 struct lnet_process_id msg_target;
100 /* Primary NID of the source. */
101 lnet_nid_t msg_initiator;
102 /* where is it from, it's only for building event */
107 * hold parameters in case message is with held due
110 lnet_nid_t msg_src_nid_param;
111 lnet_nid_t msg_rtr_nid_param;
114 * Deadline for the message after which it will be finalized if it
117 ktime_t msg_deadline;
119 /* The message health status. */
120 enum lnet_msg_hstatus msg_health_status;
121 /* This is a recovery message */
123 /* the number of times a transmission has been retried */
125 /* flag to indicate that we do not want to resend this message */
128 /* committed for sending */
129 unsigned int msg_tx_committed:1;
130 /* CPT # this message committed for sending */
131 unsigned int msg_tx_cpt:15;
132 /* committed for receiving */
133 unsigned int msg_rx_committed:1;
134 /* CPT # this message committed for receiving */
135 unsigned int msg_rx_cpt:15;
136 /* queued for tx credit */
137 unsigned int msg_tx_delayed:1;
138 /* queued for RX buffer */
139 unsigned int msg_rx_delayed:1;
140 /* ready for pending on RX delay list */
141 unsigned int msg_rx_ready_delay:1;
143 unsigned int msg_vmflush:1; /* VM trying to free memory */
144 unsigned int msg_target_is_router:1; /* sending to a router */
145 unsigned int msg_routing:1; /* being forwarded */
146 unsigned int msg_ack:1; /* ack on finalize (PUT) */
147 unsigned int msg_sending:1; /* outgoing message */
148 unsigned int msg_receiving:1; /* being received */
149 unsigned int msg_txcredit:1; /* taken an NI send credit */
150 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
151 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
152 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
153 unsigned int msg_onactivelist:1; /* on the activelist */
154 unsigned int msg_rdma_get:1;
156 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
157 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
160 struct lnet_libmd *msg_md;
161 /* the NI the message was sent or received over */
162 struct lnet_ni *msg_txni;
163 struct lnet_ni *msg_rxni;
165 unsigned int msg_len;
166 unsigned int msg_wanted;
167 unsigned int msg_offset;
168 unsigned int msg_niov;
169 struct kvec *msg_iov;
170 lnet_kiov_t *msg_kiov;
172 struct lnet_event msg_ev;
173 struct lnet_hdr msg_hdr;
176 struct lnet_libhandle {
177 struct list_head lh_hash_chain;
181 #define lh_entry(ptr, type, member) \
182 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
185 struct list_head eq_list;
186 struct lnet_libhandle eq_lh;
187 unsigned long eq_enq_seq;
188 unsigned long eq_deq_seq;
189 unsigned int eq_size;
190 lnet_eq_handler_t eq_callback;
191 struct lnet_event *eq_events;
192 int **eq_refs; /* percpt refcount for EQ */
196 struct list_head me_list;
197 struct lnet_libhandle me_lh;
198 struct lnet_process_id me_match_id;
199 unsigned int me_portal;
200 unsigned int me_pos; /* hash offset in mt_hash */
202 __u64 me_ignore_bits;
203 enum lnet_unlink me_unlink;
204 struct lnet_libmd *me_md;
208 struct list_head md_list;
209 struct lnet_libhandle md_lh;
210 struct lnet_me *md_me;
212 unsigned int md_offset;
213 unsigned int md_length;
214 unsigned int md_max_size;
217 unsigned int md_options;
218 unsigned int md_flags;
219 unsigned int md_niov; /* # frags at end of struct */
221 struct lnet_rsp_tracker *md_rspt_ptr;
222 struct lnet_eq *md_eq;
223 struct lnet_handle_md md_bulk_handle;
225 struct kvec iov[LNET_MAX_IOV];
226 lnet_kiov_t kiov[LNET_MAX_IOV];
230 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
231 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
232 #define LNET_MD_FLAG_ABORTED (1 << 2)
234 struct lnet_test_peer {
235 /* info about peers we are trying to fail */
236 struct list_head tp_list; /* ln_test_peers */
237 lnet_nid_t tp_nid; /* matching nid */
238 unsigned int tp_threshold; /* # failures to simulate */
241 #define LNET_COOKIE_TYPE_MD 1
242 #define LNET_COOKIE_TYPE_ME 2
243 #define LNET_COOKIE_TYPE_EQ 3
244 #define LNET_COOKIE_TYPE_BITS 2
245 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
247 struct lnet_ni; /* forward ref */
251 /* fields managed by portals */
252 struct list_head lnd_list; /* stash in the LND table */
253 int lnd_refcount; /* # active instances */
255 /* fields initialized by the LND */
258 int (*lnd_startup)(struct lnet_ni *ni);
259 void (*lnd_shutdown)(struct lnet_ni *ni);
260 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
262 /* In data movement APIs below, payload buffers are described as a set
263 * of 'niov' fragments which are...
265 * in virtual memory (struct kvec *iov != NULL)
267 * in pages (kernel only: plt_kiov_t *kiov != NULL).
268 * The LND may NOT overwrite these fragment descriptors.
269 * An 'offset' and may specify a byte offset within the set of
270 * fragments to start from
273 /* Start sending a preformatted message. 'private' is NULL for PUT and
274 * GET messages; otherwise this is a response to an incoming message
275 * and 'private' is the 'private' passed to lnet_parse(). Return
276 * non-zero for immediate failure, otherwise complete later with
278 int (*lnd_send)(struct lnet_ni *ni, void *private,
279 struct lnet_msg *msg);
281 /* Start receiving 'mlen' bytes of payload data, skipping the following
282 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
283 * lnet_parse(). Return non-zero for immedaite failure, otherwise
284 * complete later with lnet_finalize(). This also gives back a receive
285 * credit if the LND does flow control. */
286 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
287 int delayed, unsigned int niov,
288 struct kvec *iov, lnet_kiov_t *kiov,
289 unsigned int offset, unsigned int mlen, unsigned int rlen);
291 /* lnet_parse() has had to delay processing of this message
292 * (e.g. waiting for a forwarding buffer or send credits). Give the
293 * LND a chance to free urgently needed resources. If called, return 0
294 * for success and do NOT give back a receive credit; that has to wait
295 * until lnd_recv() gets called. On failure return < 0 and
296 * release resources; lnd_recv() will not be called. */
297 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
298 struct lnet_msg *msg, void **new_privatep);
300 /* notification of peer health */
301 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
303 /* query of peer aliveness */
304 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, time64_t *when);
306 /* accept a new connection */
307 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
310 struct lnet_tx_queue {
311 int tq_credits; /* # tx credits free */
312 int tq_credits_min; /* lowest it's been */
313 int tq_credits_max; /* total # tx credits */
314 struct list_head tq_delayed; /* delayed TXs */
317 enum lnet_net_state {
318 /* set when net block is allocated */
319 LNET_NET_STATE_INIT = 0,
320 /* set when NIs in net are started successfully */
321 LNET_NET_STATE_ACTIVE,
322 /* set if all NIs in net are in FAILED state */
323 LNET_NET_STATE_INACTIVE,
324 /* set when shutting down a NET */
325 LNET_NET_STATE_DELETING
329 /* initial state when NI is created */
330 LNET_NI_STATE_INIT = 0,
331 /* set when NI is brought up */
332 LNET_NI_STATE_ACTIVE,
333 /* set when NI is being shutdown */
334 LNET_NI_STATE_DELETING,
337 #define LNET_NI_RECOVERY_PENDING BIT(0)
338 #define LNET_NI_RECOVERY_FAILED BIT(1)
340 enum lnet_stats_type {
341 LNET_STATS_TYPE_SEND = 0,
342 LNET_STATS_TYPE_RECV,
346 struct lnet_comm_count {
347 atomic_t co_get_count;
348 atomic_t co_put_count;
349 atomic_t co_reply_count;
350 atomic_t co_ack_count;
351 atomic_t co_hello_count;
354 struct lnet_element_stats {
355 struct lnet_comm_count el_send_stats;
356 struct lnet_comm_count el_recv_stats;
357 struct lnet_comm_count el_drop_stats;
360 struct lnet_health_local_stats {
361 atomic_t hlt_local_interrupt;
362 atomic_t hlt_local_dropped;
363 atomic_t hlt_local_aborted;
364 atomic_t hlt_local_no_route;
365 atomic_t hlt_local_timeout;
366 atomic_t hlt_local_error;
369 struct lnet_health_remote_stats {
370 atomic_t hlt_remote_dropped;
371 atomic_t hlt_remote_timeout;
372 atomic_t hlt_remote_error;
373 atomic_t hlt_network_timeout;
377 /* chain on the ln_nets */
378 struct list_head net_list;
380 /* net ID, which is composed of
381 * (net_type << 16) | net_num.
382 * net_type can be one of the enumerated types defined in
383 * lnet/include/lnet/nidstr.h */
386 /* priority of the network */
389 /* total number of CPTs in the array */
392 /* cumulative CPTs of all NIs in this net */
395 /* network tunables */
396 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
399 * boolean to indicate that the tunables have been set and
402 bool net_tunables_set;
404 /* procedural interface */
405 struct lnet_lnd *net_lnd;
407 /* list of NIs on this net */
408 struct list_head net_ni_list;
410 /* list of NIs being added, but not started yet */
411 struct list_head net_ni_added;
413 /* dying LND instances */
414 struct list_head net_ni_zombie;
417 enum lnet_net_state net_state;
421 /* chain on the lnet_net structure */
422 struct list_head ni_netlist;
424 /* chain on the recovery queue */
425 struct list_head ni_recovery;
427 /* MD handle for recovery ping */
428 struct lnet_handle_md ni_ping_mdh;
435 /* bond NI on some CPTs */
438 /* interface's NID */
441 /* instance-specific data */
445 atomic_t ni_tx_credits;
447 /* percpt TX queues */
448 struct lnet_tx_queue **ni_tx_queues;
450 /* percpt reference count */
453 /* when I was last alive */
454 time64_t ni_last_alive;
456 /* pointer to parent network */
457 struct lnet_net *ni_net;
459 /* my health status */
460 struct lnet_ni_status *ni_status;
462 /* NI FSM. Protected by lnet_ni_lock() */
463 enum lnet_ni_state ni_state;
465 /* Recovery state. Protected by lnet_ni_lock() */
466 __u32 ni_recovery_state;
468 /* per NI LND tunables */
469 struct lnet_lnd_tunables ni_lnd_tunables;
471 /* lnd tunables set explicitly */
472 bool ni_lnd_tunables_set;
475 struct lnet_element_stats ni_stats;
476 struct lnet_health_local_stats ni_hstats;
478 /* physical device CPT */
481 /* sequence number used to round robin over nis within a net */
486 * initialized to LNET_MAX_HEALTH_VALUE
487 * Value is decremented every time we fail to send a message over
488 * this NI because of a NI specific failure.
489 * Value is incremented if we successfully send a message.
494 * Set to 1 by the LND when it receives an event telling it the device
495 * has gone into a fatal state. Set to 0 when the LND receives an
496 * even telling it the device is back online.
498 atomic_t ni_fatal_error_on;
501 * equivalent interfaces to use
502 * This is an array because socklnd bonding can still be configured
504 char *ni_interfaces[LNET_INTERFACES_NUM];
505 struct net *ni_net_ns; /* original net namespace */
508 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
511 * Descriptor of a ping info buffer: keep a separate indicator of the
512 * size and a reference count. The type is used both as a source and
513 * sink of data, so we need to keep some information outside of the
514 * area that may be overwritten by network data.
516 struct lnet_ping_buffer {
519 struct lnet_ping_info pb_info;
522 #define LNET_PING_BUFFER_SIZE(NNIDS) \
523 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
524 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
525 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
527 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
528 container_of((PINFO), struct lnet_ping_buffer, pb_info)
530 /* router checker data, per router */
531 struct lnet_rc_data {
532 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
533 struct list_head rcd_list;
534 struct lnet_handle_md rcd_mdh; /* ping buffer MD */
535 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
536 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
537 int rcd_nnis; /* desired size of buffer */
540 struct lnet_peer_ni {
541 /* chain on lpn_peer_nis */
542 struct list_head lpni_peer_nis;
543 /* chain on remote peer list */
544 struct list_head lpni_on_remote_peer_ni_list;
545 /* chain on recovery queue */
546 struct list_head lpni_recovery;
547 /* chain on peer hash */
548 struct list_head lpni_hashlist;
549 /* messages blocking for tx credits */
550 struct list_head lpni_txq;
551 /* messages blocking for router credits */
552 struct list_head lpni_rtrq;
553 /* chain on router list */
554 struct list_head lpni_rtr_list;
555 /* pointer to peer net I'm part of */
556 struct lnet_peer_net *lpni_peer_net;
557 /* statistics kept on each peer NI */
558 struct lnet_element_stats lpni_stats;
559 struct lnet_health_remote_stats lpni_hstats;
560 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
561 spinlock_t lpni_lock;
562 /* # tx credits available */
565 int lpni_mintxcredits;
566 /* # router credits */
569 int lpni_minrtrcredits;
570 /* bytes queued for sending */
574 /* notification outstanding? */
576 /* outstanding notification for LND? */
578 /* some thread is handling notification */
580 /* SEND event outstanding from ping */
581 bool lpni_ping_notsent;
582 /* # times router went dead<->alive. Protected with lpni_lock */
583 int lpni_alive_count;
584 /* time of last aliveness news */
585 time64_t lpni_timestamp;
586 /* time of last ping attempt */
587 time64_t lpni_ping_timestamp;
588 /* != 0 if ping reply expected */
589 time64_t lpni_ping_deadline;
590 /* when I was last alive */
591 time64_t lpni_last_alive;
592 /* when lpni_ni was queried last time */
593 time64_t lpni_last_query;
594 /* network peer is on */
595 struct lnet_net *lpni_net;
599 atomic_t lpni_refcount;
600 /* health value for the peer */
601 atomic_t lpni_healthv;
602 /* recovery ping mdh */
603 struct lnet_handle_md lpni_recovery_ping_mdh;
604 /* CPT this peer attached on */
606 /* state flags -- protected by lpni_lock */
608 /* # refs from lnet_route_t::lr_gateway */
609 int lpni_rtr_refcount;
610 /* sequence number used to round robin over peer nis within a net */
612 /* sequence number used to round robin over gateways */
616 /* returned RC ping features. Protected with lpni_lock */
617 unsigned int lpni_ping_feats;
618 /* routes on this peer */
619 struct list_head lpni_routes;
620 /* preferred local nids: if only one, use lpni_pref.nid */
625 /* number of preferred NIDs in lnpi_pref_nids */
626 __u32 lpni_pref_nnids;
627 /* router checker state */
628 struct lnet_rc_data *lpni_rcd;
631 /* Preferred path added due to traffic on non-MR peer_ni */
632 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
633 /* peer is being recovered. */
634 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
635 /* recovery ping failed */
636 #define LNET_PEER_NI_RECOVERY_FAILED (1 << 2)
637 /* peer is being deleted */
638 #define LNET_PEER_NI_DELETING (1 << 3)
641 /* chain on pt_peer_list */
642 struct list_head lp_peer_list;
644 /* list of peer nets */
645 struct list_head lp_peer_nets;
647 /* list of messages pending discovery*/
648 struct list_head lp_dc_pendq;
650 /* primary NID of the peer */
651 lnet_nid_t lp_primary_nid;
653 /* source NID to use during discovery */
654 lnet_nid_t lp_disc_src_nid;
656 /* CPT of peer_table */
659 /* number of NIDs on this peer */
662 /* reference count */
663 atomic_t lp_refcount;
665 /* lock protecting peer state flags */
668 /* peer state flags */
671 /* buffer for data pushed by peer */
672 struct lnet_ping_buffer *lp_data;
674 /* MD handle for ping in progress */
675 struct lnet_handle_md lp_ping_mdh;
677 /* MD handle for push in progress */
678 struct lnet_handle_md lp_push_mdh;
680 /* number of NIDs for sizing push data */
683 /* NI config sequence number of peer */
686 /* Local NI config sequence number acked by peer */
689 /* Local NI config sequence number sent to peer */
690 __u32 lp_node_seqno_sent;
692 /* Ping error encountered during discovery. */
695 /* Push error encountered during discovery. */
698 /* Error encountered during discovery. */
701 /* time it was put on the ln_dc_working queue */
702 time64_t lp_last_queued;
704 /* link on discovery-related lists */
705 struct list_head lp_dc_list;
707 /* tasks waiting on discovery of this peer */
708 wait_queue_head_t lp_dc_waitq;
712 * The status flags in lp_state. Their semantics have chosen so that
713 * lp_state can be zero-initialized.
715 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
716 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
718 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
719 * NOT set when the peer was pinged by discovery.
721 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
722 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
724 * A peer is marked CONFIGURED if it was configured by DLC.
726 * In addition, a peer is marked DISCOVERED if it has fully passed
727 * through Peer Discovery.
729 * When Peer Discovery is disabled, the discovery thread will mark
730 * peers REDISCOVER to indicate that they should be re-examined if
731 * discovery is (re)enabled on the node.
733 * A peer that was created as the result of inbound traffic will not
736 #define LNET_PEER_CONFIGURED (1 << 2) /* Configured via DLC */
737 #define LNET_PEER_DISCOVERED (1 << 3) /* Peer was discovered */
738 #define LNET_PEER_REDISCOVER (1 << 4) /* Discovery was disabled */
740 * A peer is marked DISCOVERING when discovery is in progress.
741 * The other flags below correspond to stages of discovery.
743 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
744 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
745 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
746 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
747 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
748 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
749 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
751 * A ping can be forced as a way to fix up state, or as a manual
752 * intervention by an admin.
753 * A push can be forced in circumstances that would normally not
754 * allow for one to happen.
756 #define LNET_PEER_FORCE_PING (1 << 12) /* Forced Ping */
757 #define LNET_PEER_FORCE_PUSH (1 << 13) /* Forced Push */
759 struct lnet_peer_net {
760 /* chain on lp_peer_nets */
761 struct list_head lpn_peer_nets;
763 /* list of peer_nis on this network */
764 struct list_head lpn_peer_nis;
766 /* pointer to the peer I'm part of */
767 struct lnet_peer *lpn_peer;
772 /* reference count */
773 atomic_t lpn_refcount;
777 #define LNET_PEER_HASH_BITS 9
778 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
781 * peer hash table - one per CPT
783 * protected by lnet_net_lock/EX for update
789 * protected by pt_zombie_lock:
793 * pt_zombie lock nests inside lnet_net_lock
795 struct lnet_peer_table {
796 int pt_version; /* /proc validity stamp */
797 int pt_number; /* # peers_ni extant */
798 struct list_head *pt_hash; /* NID->peer hash */
799 struct list_head pt_peer_list; /* peers */
800 int pt_peers; /* # peers */
801 struct list_head pt_zombie_list; /* zombie peer_ni */
802 int pt_zombies; /* # zombie peers_ni */
803 spinlock_t pt_zombie_lock; /* protect list and count */
806 /* peer aliveness is enabled only on routers for peers in a network where the
807 * struct lnet_ni::ni_peertimeout has been set to a positive value
809 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
810 ((lp)->lpni_net) && \
811 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
814 struct list_head lr_list; /* chain on net */
815 struct list_head lr_gwlist; /* chain on gateway */
816 struct lnet_peer_ni *lr_gateway; /* router node */
817 __u32 lr_net; /* remote network number */
818 int lr_seq; /* sequence for round-robin */
819 unsigned int lr_downis; /* number of down NIs */
820 __u32 lr_hops; /* how far I am */
821 unsigned int lr_priority; /* route priority */
824 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
825 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
826 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
828 struct lnet_remotenet {
829 /* chain on ln_remote_nets_hash */
830 struct list_head lrn_list;
832 struct list_head lrn_routes;
837 /** lnet message has credit and can be submitted to lnd for send/receive */
838 #define LNET_CREDIT_OK 0
839 /** lnet message is waiting for credit */
840 #define LNET_CREDIT_WAIT 1
841 /** lnet message is waiting for discovery */
842 #define LNET_DC_WAIT 2
844 struct lnet_rtrbufpool {
845 /* my free buffer pool */
846 struct list_head rbp_bufs;
847 /* messages blocking for a buffer */
848 struct list_head rbp_msgs;
849 /* # pages in each buffer */
851 /* requested number of buffers */
852 int rbp_req_nbuffers;
853 /* # buffers actually allocated */
855 /* # free buffers / blocked messages */
862 struct list_head rb_list; /* chain on rbp_bufs */
863 struct lnet_rtrbufpool *rb_pool; /* owning pool */
864 lnet_kiov_t rb_kiov[0]; /* the buffer space */
867 #define LNET_PEER_HASHSIZE 503 /* prime! */
869 enum lnet_match_flags {
870 /* Didn't match anything */
871 LNET_MATCHMD_NONE = (1 << 0),
873 LNET_MATCHMD_OK = (1 << 1),
874 /* Must be discarded */
875 LNET_MATCHMD_DROP = (1 << 2),
876 /* match and buffer is exhausted */
877 LNET_MATCHMD_EXHAUSTED = (1 << 3),
879 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
882 /* Options for struct lnet_portal::ptl_options */
883 #define LNET_PTL_LAZY (1 << 0)
884 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
885 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
887 /* parameter for matching operations (GET, PUT) */
888 struct lnet_match_info {
890 struct lnet_process_id mi_id;
893 unsigned int mi_portal;
894 unsigned int mi_rlength;
895 unsigned int mi_roffset;
898 /* ME hash of RDMA portal */
899 #define LNET_MT_HASH_BITS 8
900 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
901 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
902 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
903 * the last entry is reserved for MEs with ignore-bits */
904 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
905 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
906 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
907 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
908 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
909 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
910 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
912 /* portal match table */
913 struct lnet_match_table {
914 /* reserved for upcoming patches, CPU partition ID */
916 unsigned int mt_portal; /* portal index */
917 /* match table is set as "enabled" if there's non-exhausted MD
918 * attached on mt_mhash, it's only valid for wildcard portal */
919 unsigned int mt_enabled;
920 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
921 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
922 struct list_head *mt_mhash; /* matching hash */
925 /* these are only useful for wildcard portal */
926 /* Turn off message rotor for wildcard portals */
927 #define LNET_PTL_ROTOR_OFF 0
928 /* round-robin dispatch all PUT messages for wildcard portals */
929 #define LNET_PTL_ROTOR_ON 1
930 /* round-robin dispatch routed PUT message for wildcard portals */
931 #define LNET_PTL_ROTOR_RR_RT 2
932 /* dispatch routed PUT message by hashing source NID for wildcard portals */
933 #define LNET_PTL_ROTOR_HASH_RT 3
937 unsigned int ptl_index; /* portal ID, reserved */
938 /* flags on this portal: lazy, unique... */
939 unsigned int ptl_options;
940 /* list of messages which are stealing buffer */
941 struct list_head ptl_msg_stealing;
942 /* messages blocking for MD */
943 struct list_head ptl_msg_delayed;
944 /* Match table for each CPT */
945 struct lnet_match_table **ptl_mtables;
946 /* spread rotor of incoming "PUT" */
947 unsigned int ptl_rotor;
948 /* # active entries for this portal */
950 /* array of active entries' cpu-partition-id */
954 #define LNET_LH_HASH_BITS 12
955 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
956 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
958 /* resource container (ME, MD, EQ) */
959 struct lnet_res_container {
960 unsigned int rec_type; /* container type */
961 __u64 rec_lh_cookie; /* cookie generator */
962 struct list_head rec_active; /* active resource list */
963 struct list_head *rec_lh_hash; /* handle hash */
966 /* message container */
967 struct lnet_msg_container {
968 int msc_init; /* initialized or not */
969 /* max # threads finalizing */
971 /* msgs waiting to complete finalizing */
972 struct list_head msc_finalizing;
973 /* msgs waiting to be resent */
974 struct list_head msc_resending;
975 struct list_head msc_active; /* active message list */
976 /* threads doing finalization */
977 void **msc_finalizers;
978 /* threads doing resends */
979 void **msc_resenders;
982 /* Peer Discovery states */
983 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
984 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
985 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
987 /* Router Checker states */
988 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
989 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
990 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
993 #define LNET_STATE_SHUTDOWN 0 /* not started */
994 #define LNET_STATE_RUNNING 1 /* started up OK */
995 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
998 /* CPU partition table of LNet */
999 struct cfs_cpt_table *ln_cpt_table;
1000 /* number of CPTs in ln_cpt_table */
1001 unsigned int ln_cpt_number;
1002 unsigned int ln_cpt_bits;
1004 /* protect LNet resources (ME/MD/EQ) */
1005 struct cfs_percpt_lock *ln_res_lock;
1008 /* the vector of portals */
1009 struct lnet_portal **ln_portals;
1010 /* percpt ME containers */
1011 struct lnet_res_container **ln_me_containers;
1012 /* percpt MD container */
1013 struct lnet_res_container **ln_md_containers;
1015 /* Event Queue container */
1016 struct lnet_res_container ln_eq_container;
1017 wait_queue_head_t ln_eq_waitq;
1018 spinlock_t ln_eq_wait_lock;
1020 unsigned int ln_remote_nets_hbits;
1022 /* protect NI, peer table, credits, routers, rtrbuf... */
1023 struct cfs_percpt_lock *ln_net_lock;
1024 /* percpt message containers for active/finalizing/freed message */
1025 struct lnet_msg_container **ln_msg_containers;
1026 struct lnet_counters **ln_counters;
1027 struct lnet_peer_table **ln_peer_tables;
1028 /* list of peer nis not on a local network */
1029 struct list_head ln_remote_peer_ni_list;
1030 /* failure simulation */
1031 struct list_head ln_test_peers;
1032 struct list_head ln_drop_rules;
1033 struct list_head ln_delay_rules;
1035 struct list_head ln_nets;
1036 /* the loopback NI */
1037 struct lnet_ni *ln_loni;
1038 /* network zombie list */
1039 struct list_head ln_net_zombie;
1040 /* resend messages list */
1041 struct list_head ln_msg_resend;
1042 /* spin lock to protect the msg resend list */
1043 spinlock_t ln_msg_resend_lock;
1045 /* remote networks with routes to them */
1046 struct list_head *ln_remote_nets_hash;
1047 /* validity stamp */
1048 __u64 ln_remote_nets_version;
1049 /* list of all known routers */
1050 struct list_head ln_routers;
1051 /* validity stamp */
1052 __u64 ln_routers_version;
1053 /* percpt router buffer pools */
1054 struct lnet_rtrbufpool **ln_rtrpools;
1057 * Ping target / Push source
1059 * The ping target and push source share a single buffer. The
1060 * ln_ping_target is protected against concurrent updates by
1063 struct lnet_handle_md ln_ping_target_md;
1064 struct lnet_handle_eq ln_ping_target_eq;
1065 struct lnet_ping_buffer *ln_ping_target;
1066 atomic_t ln_ping_target_seqno;
1071 * ln_push_nnis contains the desired size of the push target.
1072 * The lnet_net_lock is used to handle update races. The old
1073 * buffer may linger a while after it has been unlinked, in
1074 * which case the event handler cleans up.
1076 struct lnet_handle_eq ln_push_target_eq;
1077 struct lnet_handle_md ln_push_target_md;
1078 struct lnet_ping_buffer *ln_push_target;
1079 int ln_push_target_nnis;
1081 /* discovery event queue handle */
1082 struct lnet_handle_eq ln_dc_eqh;
1083 /* discovery requests */
1084 struct list_head ln_dc_request;
1085 /* discovery working list */
1086 struct list_head ln_dc_working;
1087 /* discovery expired list */
1088 struct list_head ln_dc_expired;
1089 /* discovery thread wait queue */
1090 wait_queue_head_t ln_dc_waitq;
1091 /* discovery startup/shutdown state */
1094 /* monitor thread startup/shutdown state */
1096 /* router checker's event queue */
1097 struct lnet_handle_eq ln_rc_eqh;
1098 /* rcd still pending on net */
1099 struct list_head ln_rcd_deathrow;
1100 /* rcd ready for free */
1101 struct list_head ln_rcd_zombie;
1102 /* serialise startup/shutdown */
1103 struct semaphore ln_mt_signal;
1105 struct mutex ln_api_mutex;
1106 struct mutex ln_lnd_mutex;
1107 /* Have I called LNetNIInit myself? */
1109 /* LNetNIInit/LNetNIFini counter */
1111 /* SHUTDOWN/RUNNING/STOPPING */
1114 int ln_routing; /* am I a router? */
1115 lnet_pid_t ln_pid; /* requested pid */
1116 /* uniquely identifies this ni in this epoch */
1117 __u64 ln_interface_cookie;
1118 /* registered LNDs */
1119 struct list_head ln_lnds;
1121 /* test protocol compatibility flags */
1122 int ln_testprotocompat;
1124 /* 0 - load the NIs from the mod params
1125 * 1 - do not load the NIs from the mod params
1126 * Reverse logic to ensure that other calls to LNetNIInit
1129 bool ln_nis_from_mod_params;
1132 * waitq for the monitor thread. The monitor thread takes care of
1133 * checking routes, timedout messages and resending messages.
1135 wait_queue_head_t ln_mt_waitq;
1137 /* per-cpt resend queues */
1138 struct list_head **ln_mt_resendqs;
1139 /* local NIs to recover */
1140 struct list_head ln_mt_localNIRecovq;
1141 /* local NIs to recover */
1142 struct list_head ln_mt_peerNIRecovq;
1144 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1145 * There are CPT number of queues. Since response trackers will be
1146 * added on the fast path we can't afford to grab the exclusive
1147 * net lock to protect these queues. The CPT will be calculated
1148 * based on the mdh cookie.
1150 struct list_head **ln_mt_rstq;
1152 * A response tracker becomes a zombie when the associated MD is queued
1153 * for unlink before the response tracker is detached from the MD. An
1154 * entry on a zombie list can be freed when either the remaining
1155 * operations on the MD complete or when LNet has shut down.
1157 struct list_head **ln_mt_zombie_rstqs;
1158 /* recovery eq handler */
1159 struct lnet_handle_eq ln_mt_eqh;