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>
52 #include <uapi/linux/lnet/nidstr.h>
54 /* Max payload size */
55 #define LNET_MAX_PAYLOAD LNET_MTU
57 #define LNET_MAX_IOV (LNET_MAX_PAYLOAD >> PAGE_SHIFT)
60 * This is the maximum health value.
61 * All local and peer NIs created have their health default to this value.
63 #define LNET_MAX_HEALTH_VALUE 1000
68 enum lnet_msg_hstatus {
69 LNET_MSG_STATUS_OK = 0,
70 LNET_MSG_STATUS_LOCAL_INTERRUPT,
71 LNET_MSG_STATUS_LOCAL_DROPPED,
72 LNET_MSG_STATUS_LOCAL_ABORTED,
73 LNET_MSG_STATUS_LOCAL_NO_ROUTE,
74 LNET_MSG_STATUS_LOCAL_ERROR,
75 LNET_MSG_STATUS_LOCAL_TIMEOUT,
76 LNET_MSG_STATUS_REMOTE_ERROR,
77 LNET_MSG_STATUS_REMOTE_DROPPED,
78 LNET_MSG_STATUS_REMOTE_TIMEOUT,
79 LNET_MSG_STATUS_NETWORK_TIMEOUT,
83 struct lnet_rsp_tracker {
84 /* chain on the waiting list */
85 struct list_head rspt_on_list;
89 lnet_nid_t rspt_next_hop_nid;
90 /* deadline of the REPLY/ACK */
91 ktime_t rspt_deadline;
93 struct lnet_handle_md rspt_mdh;
97 struct list_head msg_activelist;
98 struct list_head msg_list; /* Q for credits/MD */
100 struct lnet_process_id msg_target;
101 /* Primary NID of the source. */
102 lnet_nid_t msg_initiator;
103 /* where is it from, it's only for building event */
108 * hold parameters in case message is with held due
111 lnet_nid_t msg_src_nid_param;
112 lnet_nid_t msg_rtr_nid_param;
115 * Deadline for the message after which it will be finalized if it
118 ktime_t msg_deadline;
120 /* The message health status. */
121 enum lnet_msg_hstatus msg_health_status;
122 /* This is a recovery message */
124 /* the number of times a transmission has been retried */
126 /* flag to indicate that we do not want to resend this message */
129 /* committed for sending */
130 unsigned int msg_tx_committed:1;
131 /* CPT # this message committed for sending */
132 unsigned int msg_tx_cpt:15;
133 /* committed for receiving */
134 unsigned int msg_rx_committed:1;
135 /* CPT # this message committed for receiving */
136 unsigned int msg_rx_cpt:15;
137 /* queued for tx credit */
138 unsigned int msg_tx_delayed:1;
139 /* queued for RX buffer */
140 unsigned int msg_rx_delayed:1;
141 /* ready for pending on RX delay list */
142 unsigned int msg_rx_ready_delay:1;
144 unsigned int msg_vmflush:1; /* VM trying to free memory */
145 unsigned int msg_target_is_router:1; /* sending to a router */
146 unsigned int msg_routing:1; /* being forwarded */
147 unsigned int msg_ack:1; /* ack on finalize (PUT) */
148 unsigned int msg_sending:1; /* outgoing message */
149 unsigned int msg_receiving:1; /* being received */
150 unsigned int msg_txcredit:1; /* taken an NI send credit */
151 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
152 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
153 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
154 unsigned int msg_onactivelist:1; /* on the activelist */
155 unsigned int msg_rdma_get:1;
157 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
158 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
161 struct lnet_libmd *msg_md;
162 /* the NI the message was sent or received over */
163 struct lnet_ni *msg_txni;
164 struct lnet_ni *msg_rxni;
166 unsigned int msg_len;
167 unsigned int msg_wanted;
168 unsigned int msg_offset;
169 unsigned int msg_niov;
170 struct kvec *msg_iov;
171 lnet_kiov_t *msg_kiov;
173 struct lnet_event msg_ev;
174 struct lnet_hdr msg_hdr;
177 struct lnet_libhandle {
178 struct list_head lh_hash_chain;
182 #define lh_entry(ptr, type, member) \
183 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
186 unsigned long eq_enq_seq;
187 unsigned long eq_deq_seq;
188 unsigned int eq_size;
189 lnet_eq_handler_t eq_callback;
190 struct lnet_event *eq_events;
191 int **eq_refs; /* percpt refcount for EQ */
195 struct list_head me_list;
197 struct lnet_process_id me_match_id;
198 unsigned int me_portal;
199 unsigned int me_pos; /* hash offset in mt_hash */
201 __u64 me_ignore_bits;
202 enum lnet_unlink me_unlink;
203 struct lnet_libmd *me_md;
207 struct list_head md_list;
208 struct lnet_libhandle md_lh;
209 struct lnet_me *md_me;
211 unsigned int md_offset;
212 unsigned int md_length;
213 unsigned int md_max_size;
216 unsigned int md_options;
217 unsigned int md_flags;
218 unsigned int md_niov; /* # frags at end of struct */
220 struct lnet_rsp_tracker *md_rspt_ptr;
221 struct lnet_eq *md_eq;
222 struct lnet_handle_md md_bulk_handle;
224 struct kvec iov[LNET_MAX_IOV];
225 lnet_kiov_t kiov[LNET_MAX_IOV];
229 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
230 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
231 #define LNET_MD_FLAG_ABORTED (1 << 2)
233 struct lnet_test_peer {
234 /* info about peers we are trying to fail */
235 struct list_head tp_list; /* ln_test_peers */
236 lnet_nid_t tp_nid; /* matching nid */
237 unsigned int tp_threshold; /* # failures to simulate */
240 #define LNET_COOKIE_TYPE_MD 1
241 #define LNET_COOKIE_TYPE_ME 2
242 #define LNET_COOKIE_TYPE_EQ 3
243 #define LNET_COOKIE_TYPE_BITS 2
244 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
246 struct lnet_ni; /* forward ref */
250 /* fields initialized by the LND */
253 int (*lnd_startup)(struct lnet_ni *ni);
254 void (*lnd_shutdown)(struct lnet_ni *ni);
255 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
257 /* In data movement APIs below, payload buffers are described as a set
258 * of 'niov' fragments which are...
260 * in virtual memory (struct kvec *iov != NULL)
262 * in pages (kernel only: plt_kiov_t *kiov != NULL).
263 * The LND may NOT overwrite these fragment descriptors.
264 * An 'offset' and may specify a byte offset within the set of
265 * fragments to start from
268 /* Start sending a preformatted message. 'private' is NULL for PUT and
269 * GET messages; otherwise this is a response to an incoming message
270 * and 'private' is the 'private' passed to lnet_parse(). Return
271 * non-zero for immediate failure, otherwise complete later with
273 int (*lnd_send)(struct lnet_ni *ni, void *private,
274 struct lnet_msg *msg);
276 /* Start receiving 'mlen' bytes of payload data, skipping the following
277 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
278 * lnet_parse(). Return non-zero for immedaite failure, otherwise
279 * complete later with lnet_finalize(). This also gives back a receive
280 * credit if the LND does flow control. */
281 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
282 int delayed, unsigned int niov,
283 struct kvec *iov, lnet_kiov_t *kiov,
284 unsigned int offset, unsigned int mlen, unsigned int rlen);
286 /* lnet_parse() has had to delay processing of this message
287 * (e.g. waiting for a forwarding buffer or send credits). Give the
288 * LND a chance to free urgently needed resources. If called, return 0
289 * for success and do NOT give back a receive credit; that has to wait
290 * until lnd_recv() gets called. On failure return < 0 and
291 * release resources; lnd_recv() will not be called. */
292 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
293 struct lnet_msg *msg, void **new_privatep);
295 /* notification of peer down */
296 void (*lnd_notify_peer_down)(lnet_nid_t peer);
298 /* accept a new connection */
299 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
302 struct lnet_tx_queue {
303 int tq_credits; /* # tx credits free */
304 int tq_credits_min; /* lowest it's been */
305 int tq_credits_max; /* total # tx credits */
306 struct list_head tq_delayed; /* delayed TXs */
309 enum lnet_net_state {
310 /* set when net block is allocated */
311 LNET_NET_STATE_INIT = 0,
312 /* set when NIs in net are started successfully */
313 LNET_NET_STATE_ACTIVE,
314 /* set if all NIs in net are in FAILED state */
315 LNET_NET_STATE_INACTIVE,
316 /* set when shutting down a NET */
317 LNET_NET_STATE_DELETING
321 /* initial state when NI is created */
322 LNET_NI_STATE_INIT = 0,
323 /* set when NI is brought up */
324 LNET_NI_STATE_ACTIVE,
325 /* set when NI is being shutdown */
326 LNET_NI_STATE_DELETING,
329 #define LNET_NI_RECOVERY_PENDING BIT(0)
330 #define LNET_NI_RECOVERY_FAILED BIT(1)
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 const 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;
408 /* when I was last alive */
409 time64_t net_last_alive;
411 /* protects access to net_last_alive */
416 /* chain on the lnet_net structure */
417 struct list_head ni_netlist;
419 /* chain on the recovery queue */
420 struct list_head ni_recovery;
422 /* MD handle for recovery ping */
423 struct lnet_handle_md ni_ping_mdh;
430 /* bond NI on some CPTs */
433 /* interface's NID */
436 /* instance-specific data */
440 atomic_t ni_tx_credits;
442 /* percpt TX queues */
443 struct lnet_tx_queue **ni_tx_queues;
445 /* percpt reference count */
448 /* pointer to parent network */
449 struct lnet_net *ni_net;
451 /* my health status */
452 struct lnet_ni_status *ni_status;
454 /* NI FSM. Protected by lnet_ni_lock() */
455 enum lnet_ni_state ni_state;
457 /* Recovery state. Protected by lnet_ni_lock() */
458 __u32 ni_recovery_state;
460 /* per NI LND tunables */
461 struct lnet_lnd_tunables ni_lnd_tunables;
463 /* lnd tunables set explicitly */
464 bool ni_lnd_tunables_set;
467 struct lnet_element_stats ni_stats;
468 struct lnet_health_local_stats ni_hstats;
470 /* physical device CPT */
473 /* sequence number used to round robin over nis within a net */
478 * initialized to LNET_MAX_HEALTH_VALUE
479 * Value is decremented every time we fail to send a message over
480 * this NI because of a NI specific failure.
481 * Value is incremented if we successfully send a message.
486 * Set to 1 by the LND when it receives an event telling it the device
487 * has gone into a fatal state. Set to 0 when the LND receives an
488 * even telling it the device is back online.
490 atomic_t ni_fatal_error_on;
493 * equivalent interfaces to use
494 * This is an array because socklnd bonding can still be configured
496 char *ni_interfaces[LNET_INTERFACES_NUM];
497 struct net *ni_net_ns; /* original net namespace */
500 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
503 * Descriptor of a ping info buffer: keep a separate indicator of the
504 * size and a reference count. The type is used both as a source and
505 * sink of data, so we need to keep some information outside of the
506 * area that may be overwritten by network data.
508 struct lnet_ping_buffer {
511 struct lnet_ping_info pb_info;
514 #define LNET_PING_BUFFER_SIZE(NNIDS) \
515 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
516 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
517 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
519 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
520 container_of((PINFO), struct lnet_ping_buffer, pb_info)
522 struct lnet_peer_ni {
523 /* chain on lpn_peer_nis */
524 struct list_head lpni_peer_nis;
525 /* chain on remote peer list */
526 struct list_head lpni_on_remote_peer_ni_list;
527 /* chain on recovery queue */
528 struct list_head lpni_recovery;
529 /* chain on peer hash */
530 struct list_head lpni_hashlist;
531 /* messages blocking for tx credits */
532 struct list_head lpni_txq;
533 /* pointer to peer net I'm part of */
534 struct lnet_peer_net *lpni_peer_net;
535 /* statistics kept on each peer NI */
536 struct lnet_element_stats lpni_stats;
537 struct lnet_health_remote_stats lpni_hstats;
538 /* spin lock protecting credits and lpni_txq */
539 spinlock_t lpni_lock;
540 /* # tx credits available */
543 int lpni_mintxcredits;
545 * Each peer_ni in a gateway maintains its own credits. This
546 * allows more traffic to gateways that have multiple interfaces.
548 /* # router credits */
551 int lpni_minrtrcredits;
552 /* bytes queued for sending */
554 /* network peer is on */
555 struct lnet_net *lpni_net;
559 atomic_t lpni_refcount;
560 /* health value for the peer */
561 atomic_t lpni_healthv;
562 /* recovery ping mdh */
563 struct lnet_handle_md lpni_recovery_ping_mdh;
564 /* CPT this peer attached on */
566 /* state flags -- protected by lpni_lock */
568 /* status of the peer NI as reported by the peer */
569 __u32 lpni_ns_status;
570 /* sequence number used to round robin over peer nis within a net */
572 /* sequence number used to round robin over gateways */
574 /* returned RC ping features. Protected with lpni_lock */
575 unsigned int lpni_ping_feats;
576 /* time last message was received from the peer */
577 time64_t lpni_last_alive;
578 /* preferred local nids: if only one, use lpni_pref.nid */
583 /* number of preferred NIDs in lnpi_pref_nids */
584 __u32 lpni_pref_nnids;
587 /* Preferred path added due to traffic on non-MR peer_ni */
588 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
589 /* peer is being recovered. */
590 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
591 /* recovery ping failed */
592 #define LNET_PEER_NI_RECOVERY_FAILED (1 << 2)
593 /* peer is being deleted */
594 #define LNET_PEER_NI_DELETING (1 << 3)
597 /* chain on pt_peer_list */
598 struct list_head lp_peer_list;
600 /* list of peer nets */
601 struct list_head lp_peer_nets;
603 /* list of messages pending discovery*/
604 struct list_head lp_dc_pendq;
606 /* chain on router list */
607 struct list_head lp_rtr_list;
609 /* primary NID of the peer */
610 lnet_nid_t lp_primary_nid;
612 /* net to perform discovery on */
613 __u32 lp_disc_net_id;
615 /* CPT of peer_table */
618 /* number of NIDs on this peer */
621 /* # refs from lnet_route::lr_gateway */
625 * peer specific health sensitivity value to decrement peer nis in
626 * this peer with if set to something other than 0
628 __u32 lp_health_sensitivity;
630 /* messages blocking for router credits */
631 struct list_head lp_rtrq;
633 /* routes on this peer */
634 struct list_head lp_routes;
636 /* reference count */
637 atomic_t lp_refcount;
639 /* lock protecting peer state flags and lpni_rtrq */
642 /* peer state flags */
645 /* buffer for data pushed by peer */
646 struct lnet_ping_buffer *lp_data;
648 /* MD handle for ping in progress */
649 struct lnet_handle_md lp_ping_mdh;
651 /* MD handle for push in progress */
652 struct lnet_handle_md lp_push_mdh;
654 /* number of NIDs for sizing push data */
657 /* NI config sequence number of peer */
660 /* Local NI config sequence number acked by peer */
663 /* Local NI config sequence number sent to peer */
664 __u32 lp_node_seqno_sent;
666 /* Ping error encountered during discovery. */
669 /* Push error encountered during discovery. */
672 /* Error encountered during discovery. */
675 /* time it was put on the ln_dc_working queue */
676 time64_t lp_last_queued;
678 /* link on discovery-related lists */
679 struct list_head lp_dc_list;
681 /* tasks waiting on discovery of this peer */
682 wait_queue_head_t lp_dc_waitq;
684 /* cached peer aliveness */
689 * The status flags in lp_state. Their semantics have chosen so that
690 * lp_state can be zero-initialized.
692 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
693 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
695 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
696 * NOT set when the peer was pinged by discovery.
698 * A peer is marked ROUTER if it indicates so in the feature bit.
700 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
701 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
702 #define LNET_PEER_ROUTER_ENABLED (1 << 2) /* router feature enabled */
705 * A peer is marked CONFIGURED if it was configured by DLC.
707 * In addition, a peer is marked DISCOVERED if it has fully passed
708 * through Peer Discovery.
710 * When Peer Discovery is disabled, the discovery thread will mark
711 * peers REDISCOVER to indicate that they should be re-examined if
712 * discovery is (re)enabled on the node.
714 * A peer that was created as the result of inbound traffic will not
717 #define LNET_PEER_CONFIGURED (1 << 3) /* Configured via DLC */
718 #define LNET_PEER_DISCOVERED (1 << 4) /* Peer was discovered */
719 #define LNET_PEER_REDISCOVER (1 << 5) /* Discovery was disabled */
721 * A peer is marked DISCOVERING when discovery is in progress.
722 * The other flags below correspond to stages of discovery.
724 #define LNET_PEER_DISCOVERING (1 << 6) /* Discovering */
725 #define LNET_PEER_DATA_PRESENT (1 << 7) /* Remote peer data present */
726 #define LNET_PEER_NIDS_UPTODATE (1 << 8) /* Remote peer info uptodate */
727 #define LNET_PEER_PING_SENT (1 << 9) /* Waiting for REPLY to Ping */
728 #define LNET_PEER_PUSH_SENT (1 << 10) /* Waiting for ACK of Push */
729 #define LNET_PEER_PING_FAILED (1 << 11) /* Ping send failure */
730 #define LNET_PEER_PUSH_FAILED (1 << 12) /* Push send failure */
732 * A ping can be forced as a way to fix up state, or as a manual
733 * intervention by an admin.
734 * A push can be forced in circumstances that would normally not
735 * allow for one to happen.
737 #define LNET_PEER_FORCE_PING (1 << 13) /* Forced Ping */
738 #define LNET_PEER_FORCE_PUSH (1 << 14) /* Forced Push */
740 /* force delete even if router */
741 #define LNET_PEER_RTR_NI_FORCE_DEL (1 << 15)
743 /* gw undergoing alive discovery */
744 #define LNET_PEER_RTR_DISCOVERY (1 << 16)
745 /* gw has undergone discovery (does not indicate success or failure) */
746 #define LNET_PEER_RTR_DISCOVERED (1 << 17)
748 struct lnet_peer_net {
749 /* chain on lp_peer_nets */
750 struct list_head lpn_peer_nets;
752 /* list of peer_nis on this network */
753 struct list_head lpn_peer_nis;
755 /* pointer to the peer I'm part of */
756 struct lnet_peer *lpn_peer;
761 /* peer net health */
764 /* time of last router net check attempt */
765 time64_t lpn_rtrcheck_timestamp;
767 /* selection sequence number */
770 /* reference count */
771 atomic_t lpn_refcount;
775 #define LNET_PEER_HASH_BITS 9
776 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
779 * peer hash table - one per CPT
781 * protected by lnet_net_lock/EX for update
786 * protected by pt_zombie_lock:
790 * pt_zombie lock nests inside lnet_net_lock
792 struct lnet_peer_table {
793 int pt_version; /* /proc validity stamp */
794 struct list_head *pt_hash; /* NID->peer hash */
795 struct list_head pt_peer_list; /* peers */
796 int pt_peers; /* # peers */
797 struct list_head pt_zombie_list; /* zombie peer_ni */
798 int pt_zombies; /* # zombie peers_ni */
799 spinlock_t pt_zombie_lock; /* protect list and count */
802 /* peer aliveness is enabled only on routers for peers in a network where the
803 * struct lnet_ni::ni_peertimeout has been set to a positive value
805 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
806 ((lp)->lpni_net) && \
807 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
810 struct list_head lr_list; /* chain on net */
811 struct list_head lr_gwlist; /* chain on gateway */
812 struct lnet_peer *lr_gateway; /* router node */
813 lnet_nid_t lr_nid; /* NID used to add route */
814 __u32 lr_net; /* remote network number */
815 __u32 lr_lnet; /* local network number */
816 int lr_seq; /* sequence for round-robin */
817 __u32 lr_hops; /* how far I am */
818 unsigned int lr_priority; /* route priority */
819 bool lr_alive; /* cached route aliveness */
822 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
823 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
824 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
826 struct lnet_remotenet {
827 /* chain on ln_remote_nets_hash */
828 struct list_head lrn_list;
830 struct list_head lrn_routes;
835 /** lnet message has credit and can be submitted to lnd for send/receive */
836 #define LNET_CREDIT_OK 0
837 /** lnet message is waiting for credit */
838 #define LNET_CREDIT_WAIT 1
839 /** lnet message is waiting for discovery */
840 #define LNET_DC_WAIT 2
842 struct lnet_rtrbufpool {
843 /* my free buffer pool */
844 struct list_head rbp_bufs;
845 /* messages blocking for a buffer */
846 struct list_head rbp_msgs;
847 /* # pages in each buffer */
849 /* requested number of buffers */
850 int rbp_req_nbuffers;
851 /* # buffers actually allocated */
853 /* # free buffers / blocked messages */
860 struct list_head rb_list; /* chain on rbp_bufs */
861 struct lnet_rtrbufpool *rb_pool; /* owning pool */
862 lnet_kiov_t rb_kiov[0]; /* the buffer space */
865 #define LNET_PEER_HASHSIZE 503 /* prime! */
867 enum lnet_match_flags {
868 /* Didn't match anything */
869 LNET_MATCHMD_NONE = (1 << 0),
871 LNET_MATCHMD_OK = (1 << 1),
872 /* Must be discarded */
873 LNET_MATCHMD_DROP = (1 << 2),
874 /* match and buffer is exhausted */
875 LNET_MATCHMD_EXHAUSTED = (1 << 3),
877 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
880 /* Options for struct lnet_portal::ptl_options */
881 #define LNET_PTL_LAZY (1 << 0)
882 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
883 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
885 /* parameter for matching operations (GET, PUT) */
886 struct lnet_match_info {
888 struct lnet_process_id mi_id;
891 unsigned int mi_portal;
892 unsigned int mi_rlength;
893 unsigned int mi_roffset;
896 /* ME hash of RDMA portal */
897 #define LNET_MT_HASH_BITS 8
898 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
899 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
900 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
901 * the last entry is reserved for MEs with ignore-bits */
902 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
903 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
904 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
905 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
906 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
907 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
908 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
910 /* portal match table */
911 struct lnet_match_table {
912 /* reserved for upcoming patches, CPU partition ID */
914 unsigned int mt_portal; /* portal index */
915 /* match table is set as "enabled" if there's non-exhausted MD
916 * attached on mt_mhash, it's only valid for wildcard portal */
917 unsigned int mt_enabled;
918 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
919 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
920 struct list_head *mt_mhash; /* matching hash */
923 /* these are only useful for wildcard portal */
924 /* Turn off message rotor for wildcard portals */
925 #define LNET_PTL_ROTOR_OFF 0
926 /* round-robin dispatch all PUT messages for wildcard portals */
927 #define LNET_PTL_ROTOR_ON 1
928 /* round-robin dispatch routed PUT message for wildcard portals */
929 #define LNET_PTL_ROTOR_RR_RT 2
930 /* dispatch routed PUT message by hashing source NID for wildcard portals */
931 #define LNET_PTL_ROTOR_HASH_RT 3
935 unsigned int ptl_index; /* portal ID, reserved */
936 /* flags on this portal: lazy, unique... */
937 unsigned int ptl_options;
938 /* list of messages which are stealing buffer */
939 struct list_head ptl_msg_stealing;
940 /* messages blocking for MD */
941 struct list_head ptl_msg_delayed;
942 /* Match table for each CPT */
943 struct lnet_match_table **ptl_mtables;
944 /* spread rotor of incoming "PUT" */
945 unsigned int ptl_rotor;
946 /* # active entries for this portal */
948 /* array of active entries' cpu-partition-id */
952 #define LNET_LH_HASH_BITS 12
953 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
954 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
956 /* resource container (ME, MD, EQ) */
957 struct lnet_res_container {
958 unsigned int rec_type; /* container type */
959 __u64 rec_lh_cookie; /* cookie generator */
960 struct list_head rec_active; /* active resource list */
961 struct list_head *rec_lh_hash; /* handle hash */
964 /* message container */
965 struct lnet_msg_container {
966 int msc_init; /* initialized or not */
967 /* max # threads finalizing */
969 /* msgs waiting to complete finalizing */
970 struct list_head msc_finalizing;
971 /* msgs waiting to be resent */
972 struct list_head msc_resending;
973 struct list_head msc_active; /* active message list */
974 /* threads doing finalization */
975 void **msc_finalizers;
976 /* threads doing resends */
977 void **msc_resenders;
980 /* Peer Discovery states */
981 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
982 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
983 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
985 /* Router Checker states */
986 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
987 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
988 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
991 #define LNET_STATE_SHUTDOWN 0 /* not started */
992 #define LNET_STATE_RUNNING 1 /* started up OK */
993 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
996 /* CPU partition table of LNet */
997 struct cfs_cpt_table *ln_cpt_table;
998 /* number of CPTs in ln_cpt_table */
999 unsigned int ln_cpt_number;
1000 unsigned int ln_cpt_bits;
1002 /* protect LNet resources (ME/MD/EQ) */
1003 struct cfs_percpt_lock *ln_res_lock;
1006 /* the vector of portals */
1007 struct lnet_portal **ln_portals;
1008 /* percpt MD container */
1009 struct lnet_res_container **ln_md_containers;
1011 /* Event Queue container */
1012 struct lnet_res_container ln_eq_container;
1013 spinlock_t ln_eq_wait_lock;
1015 unsigned int ln_remote_nets_hbits;
1017 /* protect NI, peer table, credits, routers, rtrbuf... */
1018 struct cfs_percpt_lock *ln_net_lock;
1019 /* percpt message containers for active/finalizing/freed message */
1020 struct lnet_msg_container **ln_msg_containers;
1021 struct lnet_counters **ln_counters;
1022 struct lnet_peer_table **ln_peer_tables;
1023 /* list of peer nis not on a local network */
1024 struct list_head ln_remote_peer_ni_list;
1025 /* failure simulation */
1026 struct list_head ln_test_peers;
1027 struct list_head ln_drop_rules;
1028 struct list_head ln_delay_rules;
1030 struct list_head ln_nets;
1031 /* the loopback NI */
1032 struct lnet_ni *ln_loni;
1033 /* network zombie list */
1034 struct list_head ln_net_zombie;
1035 /* resend messages list */
1036 struct list_head ln_msg_resend;
1037 /* spin lock to protect the msg resend list */
1038 spinlock_t ln_msg_resend_lock;
1040 /* remote networks with routes to them */
1041 struct list_head *ln_remote_nets_hash;
1042 /* validity stamp */
1043 __u64 ln_remote_nets_version;
1044 /* list of all known routers */
1045 struct list_head ln_routers;
1046 /* validity stamp */
1047 __u64 ln_routers_version;
1048 /* percpt router buffer pools */
1049 struct lnet_rtrbufpool **ln_rtrpools;
1052 * Ping target / Push source
1054 * The ping target and push source share a single buffer. The
1055 * ln_ping_target is protected against concurrent updates by
1058 struct lnet_handle_md ln_ping_target_md;
1059 struct lnet_eq *ln_ping_target_eq;
1060 struct lnet_ping_buffer *ln_ping_target;
1061 atomic_t ln_ping_target_seqno;
1066 * ln_push_nnis contains the desired size of the push target.
1067 * The lnet_net_lock is used to handle update races. The old
1068 * buffer may linger a while after it has been unlinked, in
1069 * which case the event handler cleans up.
1071 struct lnet_eq *ln_push_target_eq;
1072 struct lnet_handle_md ln_push_target_md;
1073 struct lnet_ping_buffer *ln_push_target;
1074 int ln_push_target_nnis;
1076 /* discovery event queue handle */
1077 struct lnet_eq *ln_dc_eq;
1078 /* discovery requests */
1079 struct list_head ln_dc_request;
1080 /* discovery working list */
1081 struct list_head ln_dc_working;
1082 /* discovery expired list */
1083 struct list_head ln_dc_expired;
1084 /* discovery thread wait queue */
1085 wait_queue_head_t ln_dc_waitq;
1086 /* discovery startup/shutdown state */
1089 /* monitor thread startup/shutdown state */
1091 /* serialise startup/shutdown */
1092 struct semaphore ln_mt_signal;
1094 struct mutex ln_api_mutex;
1095 struct mutex ln_lnd_mutex;
1096 /* Have I called LNetNIInit myself? */
1098 /* LNetNIInit/LNetNIFini counter */
1100 /* SHUTDOWN/RUNNING/STOPPING */
1103 int ln_routing; /* am I a router? */
1104 lnet_pid_t ln_pid; /* requested pid */
1105 /* uniquely identifies this ni in this epoch */
1106 __u64 ln_interface_cookie;
1107 /* registered LNDs */
1108 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1110 /* test protocol compatibility flags */
1111 unsigned long ln_testprotocompat;
1113 /* 0 - load the NIs from the mod params
1114 * 1 - do not load the NIs from the mod params
1115 * Reverse logic to ensure that other calls to LNetNIInit
1118 bool ln_nis_from_mod_params;
1121 * completion for the monitor thread. The monitor thread takes care of
1122 * checking routes, timedout messages and resending messages.
1124 struct completion ln_mt_wait_complete;
1126 /* per-cpt resend queues */
1127 struct list_head **ln_mt_resendqs;
1128 /* local NIs to recover */
1129 struct list_head ln_mt_localNIRecovq;
1130 /* local NIs to recover */
1131 struct list_head ln_mt_peerNIRecovq;
1133 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1134 * There are CPT number of queues. Since response trackers will be
1135 * added on the fast path we can't afford to grab the exclusive
1136 * net lock to protect these queues. The CPT will be calculated
1137 * based on the mdh cookie.
1139 struct list_head **ln_mt_rstq;
1141 * A response tracker becomes a zombie when the associated MD is queued
1142 * for unlink before the response tracker is detached from the MD. An
1143 * entry on a zombie list can be freed when either the remaining
1144 * operations on the MD complete or when LNet has shut down.
1146 struct list_head **ln_mt_zombie_rstqs;
1147 /* recovery eq handler */
1148 struct lnet_eq *ln_mt_eq;
1151 * Completed when the discovery and monitor threads can enter their
1154 struct completion ln_started;