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 /** limit on the number of fragments in discontiguous MDs */
58 #define LNET_MAX_IOV 256
61 * This is the maximum health value.
62 * All local and peer NIs created have their health default to this value.
64 #define LNET_MAX_HEALTH_VALUE 1000
69 enum lnet_msg_hstatus {
70 LNET_MSG_STATUS_OK = 0,
71 LNET_MSG_STATUS_LOCAL_INTERRUPT,
72 LNET_MSG_STATUS_LOCAL_DROPPED,
73 LNET_MSG_STATUS_LOCAL_ABORTED,
74 LNET_MSG_STATUS_LOCAL_NO_ROUTE,
75 LNET_MSG_STATUS_LOCAL_ERROR,
76 LNET_MSG_STATUS_LOCAL_TIMEOUT,
77 LNET_MSG_STATUS_REMOTE_ERROR,
78 LNET_MSG_STATUS_REMOTE_DROPPED,
79 LNET_MSG_STATUS_REMOTE_TIMEOUT,
80 LNET_MSG_STATUS_NETWORK_TIMEOUT,
84 struct lnet_rsp_tracker {
85 /* chain on the waiting list */
86 struct list_head rspt_on_list;
90 lnet_nid_t rspt_next_hop_nid;
91 /* deadline of the REPLY/ACK */
92 ktime_t rspt_deadline;
94 struct lnet_handle_md rspt_mdh;
98 struct list_head msg_activelist;
99 struct list_head msg_list; /* Q for credits/MD */
101 struct lnet_process_id msg_target;
102 /* Primary NID of the source. */
103 lnet_nid_t msg_initiator;
104 /* where is it from, it's only for building event */
109 * hold parameters in case message is with held due
112 lnet_nid_t msg_src_nid_param;
113 lnet_nid_t msg_rtr_nid_param;
116 * Deadline for the message after which it will be finalized if it
119 ktime_t msg_deadline;
121 /* The message health status. */
122 enum lnet_msg_hstatus msg_health_status;
123 /* This is a recovery message */
125 /* the number of times a transmission has been retried */
127 /* flag to indicate that we do not want to resend this message */
130 /* committed for sending */
131 unsigned int msg_tx_committed:1;
132 /* CPT # this message committed for sending */
133 unsigned int msg_tx_cpt:15;
134 /* committed for receiving */
135 unsigned int msg_rx_committed:1;
136 /* CPT # this message committed for receiving */
137 unsigned int msg_rx_cpt:15;
138 /* queued for tx credit */
139 unsigned int msg_tx_delayed:1;
140 /* queued for RX buffer */
141 unsigned int msg_rx_delayed:1;
142 /* ready for pending on RX delay list */
143 unsigned int msg_rx_ready_delay:1;
145 unsigned int msg_vmflush:1; /* VM trying to free memory */
146 unsigned int msg_target_is_router:1; /* sending to a router */
147 unsigned int msg_routing:1; /* being forwarded */
148 unsigned int msg_ack:1; /* ack on finalize (PUT) */
149 unsigned int msg_sending:1; /* outgoing message */
150 unsigned int msg_receiving:1; /* being received */
151 unsigned int msg_txcredit:1; /* taken an NI send credit */
152 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
153 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
154 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
155 unsigned int msg_onactivelist:1; /* on the activelist */
156 unsigned int msg_rdma_get:1;
158 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
159 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
162 struct lnet_libmd *msg_md;
163 /* the NI the message was sent or received over */
164 struct lnet_ni *msg_txni;
165 struct lnet_ni *msg_rxni;
167 unsigned int msg_len;
168 unsigned int msg_wanted;
169 unsigned int msg_offset;
170 unsigned int msg_niov;
171 struct bio_vec *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 struct list_head me_list;
188 struct lnet_process_id me_match_id;
189 unsigned int me_portal;
190 unsigned int me_pos; /* hash offset in mt_hash */
192 __u64 me_ignore_bits;
193 enum lnet_unlink me_unlink;
194 struct lnet_libmd *me_md;
198 struct list_head md_list;
199 struct lnet_libhandle md_lh;
200 struct lnet_me *md_me;
202 unsigned int md_offset;
203 unsigned int md_length;
204 unsigned int md_max_size;
207 unsigned int md_options;
208 unsigned int md_flags;
209 unsigned int md_niov; /* # frags at end of struct */
211 struct lnet_rsp_tracker *md_rspt_ptr;
212 lnet_handler_t md_handler;
213 struct lnet_handle_md md_bulk_handle;
214 struct bio_vec md_kiov[LNET_MAX_IOV];
217 #define LNET_MD_FLAG_ZOMBIE BIT(0)
218 #define LNET_MD_FLAG_AUTO_UNLINK BIT(1)
219 #define LNET_MD_FLAG_ABORTED BIT(2)
220 /* LNET_MD_FLAG_HANDLING is set when a non-unlink event handler
221 * is being called for an event relating to the md.
222 * It ensures only one such handler runs at a time.
223 * The final "unlink" event is only called once the
224 * md_refcount has reached zero, and this flag has been cleared,
225 * ensuring that it doesn't race with any other event handler
228 #define LNET_MD_FLAG_HANDLING BIT(3)
229 #define LNET_MD_FLAG_DISCARD BIT(4)
231 struct lnet_test_peer {
232 /* info about peers we are trying to fail */
233 struct list_head tp_list; /* ln_test_peers */
234 lnet_nid_t tp_nid; /* matching nid */
235 unsigned int tp_threshold; /* # failures to simulate */
238 #define LNET_COOKIE_TYPE_MD 1
239 #define LNET_COOKIE_TYPE_ME 2
240 #define LNET_COOKIE_TYPE_EQ 3
241 #define LNET_COOKIE_TYPE_BITS 2
242 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
244 struct lnet_ni; /* forward ref */
248 /* fields initialized by the LND */
251 int (*lnd_startup)(struct lnet_ni *ni);
252 void (*lnd_shutdown)(struct lnet_ni *ni);
253 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
255 /* In data movement APIs below, payload buffers are described as a set
256 * of 'niov' fragments which are in pages.
257 * The LND may NOT overwrite these fragment descriptors.
258 * An 'offset' and may specify a byte offset within the set of
259 * fragments to start from
262 /* Start sending a preformatted message. 'private' is NULL for PUT and
263 * GET messages; otherwise this is a response to an incoming message
264 * and 'private' is the 'private' passed to lnet_parse(). Return
265 * non-zero for immediate failure, otherwise complete later with
267 int (*lnd_send)(struct lnet_ni *ni, void *private,
268 struct lnet_msg *msg);
270 /* Start receiving 'mlen' bytes of payload data, skipping the following
271 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
272 * lnet_parse(). Return non-zero for immedaite failure, otherwise
273 * complete later with lnet_finalize(). This also gives back a receive
274 * credit if the LND does flow control. */
275 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
276 int delayed, unsigned int niov,
277 struct bio_vec *kiov,
278 unsigned int offset, unsigned int mlen, unsigned int rlen);
280 /* lnet_parse() has had to delay processing of this message
281 * (e.g. waiting for a forwarding buffer or send credits). Give the
282 * LND a chance to free urgently needed resources. If called, return 0
283 * for success and do NOT give back a receive credit; that has to wait
284 * until lnd_recv() gets called. On failure return < 0 and
285 * release resources; lnd_recv() will not be called. */
286 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
287 struct lnet_msg *msg, void **new_privatep);
289 /* notification of peer down */
290 void (*lnd_notify_peer_down)(lnet_nid_t peer);
292 /* accept a new connection */
293 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
296 struct lnet_tx_queue {
297 int tq_credits; /* # tx credits free */
298 int tq_credits_min; /* lowest it's been */
299 int tq_credits_max; /* total # tx credits */
300 struct list_head tq_delayed; /* delayed TXs */
303 enum lnet_net_state {
304 /* set when net block is allocated */
305 LNET_NET_STATE_INIT = 0,
306 /* set when NIs in net are started successfully */
307 LNET_NET_STATE_ACTIVE,
308 /* set if all NIs in net are in FAILED state */
309 LNET_NET_STATE_INACTIVE,
310 /* set when shutting down a NET */
311 LNET_NET_STATE_DELETING
315 /* initial state when NI is created */
316 LNET_NI_STATE_INIT = 0,
317 /* set when NI is brought up */
318 LNET_NI_STATE_ACTIVE,
319 /* set when NI is being shutdown */
320 LNET_NI_STATE_DELETING,
323 #define LNET_NI_RECOVERY_PENDING BIT(0)
324 #define LNET_NI_RECOVERY_FAILED BIT(1)
326 enum lnet_stats_type {
327 LNET_STATS_TYPE_SEND = 0,
328 LNET_STATS_TYPE_RECV,
332 struct lnet_comm_count {
333 atomic_t co_get_count;
334 atomic_t co_put_count;
335 atomic_t co_reply_count;
336 atomic_t co_ack_count;
337 atomic_t co_hello_count;
340 struct lnet_element_stats {
341 struct lnet_comm_count el_send_stats;
342 struct lnet_comm_count el_recv_stats;
343 struct lnet_comm_count el_drop_stats;
346 struct lnet_health_local_stats {
347 atomic_t hlt_local_interrupt;
348 atomic_t hlt_local_dropped;
349 atomic_t hlt_local_aborted;
350 atomic_t hlt_local_no_route;
351 atomic_t hlt_local_timeout;
352 atomic_t hlt_local_error;
355 struct lnet_health_remote_stats {
356 atomic_t hlt_remote_dropped;
357 atomic_t hlt_remote_timeout;
358 atomic_t hlt_remote_error;
359 atomic_t hlt_network_timeout;
363 /* chain on the ln_nets */
364 struct list_head net_list;
366 /* net ID, which is composed of
367 * (net_type << 16) | net_num.
368 * net_type can be one of the enumerated types defined in
369 * lnet/include/lnet/nidstr.h */
372 /* priority of the network */
375 /* total number of CPTs in the array */
378 /* cumulative CPTs of all NIs in this net */
381 /* network tunables */
382 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
385 * boolean to indicate that the tunables have been set and
388 bool net_tunables_set;
390 /* procedural interface */
391 const struct lnet_lnd *net_lnd;
393 /* list of NIs on this net */
394 struct list_head net_ni_list;
396 /* list of NIs being added, but not started yet */
397 struct list_head net_ni_added;
399 /* dying LND instances */
400 struct list_head net_ni_zombie;
402 /* when I was last alive */
403 time64_t net_last_alive;
405 /* protects access to net_last_alive */
410 /* chain on the lnet_net structure */
411 struct list_head ni_netlist;
413 /* chain on the recovery queue */
414 struct list_head ni_recovery;
416 /* MD handle for recovery ping */
417 struct lnet_handle_md ni_ping_mdh;
424 /* bond NI on some CPTs */
427 /* interface's NID */
430 /* instance-specific data */
434 atomic_t ni_tx_credits;
436 /* percpt TX queues */
437 struct lnet_tx_queue **ni_tx_queues;
439 /* percpt reference count */
442 /* pointer to parent network */
443 struct lnet_net *ni_net;
445 /* my health status */
446 struct lnet_ni_status *ni_status;
448 /* NI FSM. Protected by lnet_ni_lock() */
449 enum lnet_ni_state ni_state;
451 /* Recovery state. Protected by lnet_ni_lock() */
452 __u32 ni_recovery_state;
454 /* per NI LND tunables */
455 struct lnet_lnd_tunables ni_lnd_tunables;
457 /* lnd tunables set explicitly */
458 bool ni_lnd_tunables_set;
461 struct lnet_element_stats ni_stats;
462 struct lnet_health_local_stats ni_hstats;
464 /* physical device CPT */
467 /* sequence number used to round robin over nis within a net */
472 * initialized to LNET_MAX_HEALTH_VALUE
473 * Value is decremented every time we fail to send a message over
474 * this NI because of a NI specific failure.
475 * Value is incremented if we successfully send a message.
480 * Set to 1 by the LND when it receives an event telling it the device
481 * has gone into a fatal state. Set to 0 when the LND receives an
482 * even telling it the device is back online.
484 atomic_t ni_fatal_error_on;
487 * equivalent interfaces to use
488 * This is an array because socklnd bonding can still be configured
490 char *ni_interfaces[LNET_INTERFACES_NUM];
491 struct net *ni_net_ns; /* original net namespace */
494 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
497 * Descriptor of a ping info buffer: keep a separate indicator of the
498 * size and a reference count. The type is used both as a source and
499 * sink of data, so we need to keep some information outside of the
500 * area that may be overwritten by network data.
502 struct lnet_ping_buffer {
506 struct lnet_ping_info pb_info;
509 #define LNET_PING_BUFFER_SIZE(NNIDS) \
510 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
511 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
512 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
514 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
515 container_of((PINFO), struct lnet_ping_buffer, pb_info)
517 struct lnet_peer_ni {
518 /* chain on lpn_peer_nis */
519 struct list_head lpni_peer_nis;
520 /* chain on remote peer list */
521 struct list_head lpni_on_remote_peer_ni_list;
522 /* chain on recovery queue */
523 struct list_head lpni_recovery;
524 /* chain on peer hash */
525 struct list_head lpni_hashlist;
526 /* messages blocking for tx credits */
527 struct list_head lpni_txq;
528 /* pointer to peer net I'm part of */
529 struct lnet_peer_net *lpni_peer_net;
530 /* statistics kept on each peer NI */
531 struct lnet_element_stats lpni_stats;
532 struct lnet_health_remote_stats lpni_hstats;
533 /* spin lock protecting credits and lpni_txq */
534 spinlock_t lpni_lock;
535 /* # tx credits available */
538 int lpni_mintxcredits;
540 * Each peer_ni in a gateway maintains its own credits. This
541 * allows more traffic to gateways that have multiple interfaces.
543 /* # router credits */
546 int lpni_minrtrcredits;
547 /* bytes queued for sending */
549 /* network peer is on */
550 struct lnet_net *lpni_net;
554 atomic_t lpni_refcount;
555 /* health value for the peer */
556 atomic_t lpni_healthv;
557 /* recovery ping mdh */
558 struct lnet_handle_md lpni_recovery_ping_mdh;
559 /* CPT this peer attached on */
561 /* state flags -- protected by lpni_lock */
563 /* status of the peer NI as reported by the peer */
564 __u32 lpni_ns_status;
565 /* sequence number used to round robin over peer nis within a net */
567 /* sequence number used to round robin over gateways */
569 /* returned RC ping features. Protected with lpni_lock */
570 unsigned int lpni_ping_feats;
571 /* time last message was received from the peer */
572 time64_t lpni_last_alive;
573 /* preferred local nids: if only one, use lpni_pref.nid */
578 /* number of preferred NIDs in lnpi_pref_nids */
579 __u32 lpni_pref_nnids;
582 /* Preferred path added due to traffic on non-MR peer_ni */
583 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
584 /* peer is being recovered. */
585 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
586 /* recovery ping failed */
587 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
588 /* peer is being deleted */
589 #define LNET_PEER_NI_DELETING BIT(3)
592 /* chain on pt_peer_list */
593 struct list_head lp_peer_list;
595 /* list of peer nets */
596 struct list_head lp_peer_nets;
598 /* list of messages pending discovery*/
599 struct list_head lp_dc_pendq;
601 /* chain on router list */
602 struct list_head lp_rtr_list;
604 /* primary NID of the peer */
605 lnet_nid_t lp_primary_nid;
607 /* source NID to use during discovery */
608 lnet_nid_t lp_disc_src_nid;
610 /* net to perform discovery on */
611 __u32 lp_disc_net_id;
613 /* CPT of peer_table */
616 /* number of NIDs on this peer */
619 /* # refs from lnet_route::lr_gateway */
623 * peer specific health sensitivity value to decrement peer nis in
624 * this peer with if set to something other than 0
626 __u32 lp_health_sensitivity;
628 /* messages blocking for router credits */
629 struct list_head lp_rtrq;
631 /* routes on this peer */
632 struct list_head lp_routes;
634 /* reference count */
635 atomic_t lp_refcount;
637 /* lock protecting peer state flags and lpni_rtrq */
640 /* peer state flags */
643 /* buffer for data pushed by peer */
644 struct lnet_ping_buffer *lp_data;
646 /* MD handle for ping in progress */
647 struct lnet_handle_md lp_ping_mdh;
649 /* MD handle for push in progress */
650 struct lnet_handle_md lp_push_mdh;
652 /* number of NIDs for sizing push data */
655 /* NI config sequence number of peer */
658 /* Local NI config sequence number acked by peer */
661 /* Local NI config sequence number sent to peer */
662 __u32 lp_node_seqno_sent;
664 /* Ping error encountered during discovery. */
667 /* Push error encountered during discovery. */
670 /* Error encountered during discovery. */
673 /* time it was put on the ln_dc_working queue */
674 time64_t lp_last_queued;
676 /* link on discovery-related lists */
677 struct list_head lp_dc_list;
679 /* tasks waiting on discovery of this peer */
680 wait_queue_head_t lp_dc_waitq;
682 /* cached peer aliveness */
687 * The status flags in lp_state. Their semantics have chosen so that
688 * lp_state can be zero-initialized.
690 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
691 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
693 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
694 * NOT set when the peer was pinged by discovery.
696 * A peer is marked ROUTER if it indicates so in the feature bit.
698 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
699 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
700 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
703 * A peer is marked CONFIGURED if it was configured by DLC.
705 * In addition, a peer is marked DISCOVERED if it has fully passed
706 * through Peer Discovery.
708 * When Peer Discovery is disabled, the discovery thread will mark
709 * peers REDISCOVER to indicate that they should be re-examined if
710 * discovery is (re)enabled on the node.
712 * A peer that was created as the result of inbound traffic will not
715 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
716 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
717 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
719 * A peer is marked DISCOVERING when discovery is in progress.
720 * The other flags below correspond to stages of discovery.
722 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
723 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
724 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
725 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
726 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
727 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
728 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
730 * A ping can be forced as a way to fix up state, or as a manual
731 * intervention by an admin.
732 * A push can be forced in circumstances that would normally not
733 * allow for one to happen.
735 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
736 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
738 /* force delete even if router */
739 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
741 /* gw undergoing alive discovery */
742 #define LNET_PEER_RTR_DISCOVERY BIT(16)
743 /* gw has undergone discovery (does not indicate success or failure) */
744 #define LNET_PEER_RTR_DISCOVERED BIT(17)
746 /* peer is marked for deletion */
747 #define LNET_PEER_MARK_DELETION BIT(18)
749 struct lnet_peer_net {
750 /* chain on lp_peer_nets */
751 struct list_head lpn_peer_nets;
753 /* list of peer_nis on this network */
754 struct list_head lpn_peer_nis;
756 /* pointer to the peer I'm part of */
757 struct lnet_peer *lpn_peer;
762 /* peer net health */
765 /* time of last router net check attempt */
766 time64_t lpn_rtrcheck_timestamp;
768 /* selection sequence number */
771 /* reference count */
772 atomic_t lpn_refcount;
776 #define LNET_PEER_HASH_BITS 9
777 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
780 * peer hash table - one per CPT
782 * protected by lnet_net_lock/EX for update
787 * protected by pt_zombie_lock:
791 * pt_zombie lock nests inside lnet_net_lock
793 struct lnet_peer_table {
794 int pt_version; /* /proc validity stamp */
795 struct list_head *pt_hash; /* NID->peer hash */
796 struct list_head pt_peer_list; /* peers */
797 int pt_peers; /* # peers */
798 struct list_head pt_zombie_list; /* zombie peer_ni */
799 int pt_zombies; /* # zombie peers_ni */
800 spinlock_t pt_zombie_lock; /* protect list and count */
803 /* peer aliveness is enabled only on routers for peers in a network where the
804 * struct lnet_ni::ni_peertimeout has been set to a positive value
806 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
807 ((lp)->lpni_net) && \
808 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
811 struct list_head lr_list; /* chain on net */
812 struct list_head lr_gwlist; /* chain on gateway */
813 struct lnet_peer *lr_gateway; /* router node */
814 lnet_nid_t lr_nid; /* NID used to add route */
815 __u32 lr_net; /* remote network number */
816 __u32 lr_lnet; /* local network number */
817 int lr_seq; /* sequence for round-robin */
818 __u32 lr_hops; /* how far I am */
819 unsigned int lr_priority; /* route priority */
820 bool lr_alive; /* cached route aliveness */
821 bool lr_single_hop; /* this route is single-hop */
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 struct bio_vec 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 = BIT(0),
873 LNET_MATCHMD_OK = BIT(1),
874 /* Must be discarded */
875 LNET_MATCHMD_DROP = BIT(2),
876 /* match and buffer is exhausted */
877 LNET_MATCHMD_EXHAUSTED = BIT(3),
879 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
882 /* Options for struct lnet_portal::ptl_options */
883 #define LNET_PTL_LAZY BIT(0)
884 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
885 #define LNET_PTL_MATCH_WILDCARD BIT(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 MD container */
1011 struct lnet_res_container **ln_md_containers;
1013 /* Event Queue container */
1014 struct lnet_res_container ln_eq_container;
1015 spinlock_t ln_eq_wait_lock;
1017 unsigned int ln_remote_nets_hbits;
1019 /* protect NI, peer table, credits, routers, rtrbuf... */
1020 struct cfs_percpt_lock *ln_net_lock;
1021 /* percpt message containers for active/finalizing/freed message */
1022 struct lnet_msg_container **ln_msg_containers;
1023 struct lnet_counters **ln_counters;
1024 struct lnet_peer_table **ln_peer_tables;
1025 /* list of peer nis not on a local network */
1026 struct list_head ln_remote_peer_ni_list;
1027 /* failure simulation */
1028 struct list_head ln_test_peers;
1029 struct list_head ln_drop_rules;
1030 struct list_head ln_delay_rules;
1032 struct list_head ln_nets;
1033 /* the loopback NI */
1034 struct lnet_ni *ln_loni;
1035 /* network zombie list */
1036 struct list_head ln_net_zombie;
1037 /* resend messages list */
1038 struct list_head ln_msg_resend;
1039 /* spin lock to protect the msg resend list */
1040 spinlock_t ln_msg_resend_lock;
1042 /* remote networks with routes to them */
1043 struct list_head *ln_remote_nets_hash;
1044 /* validity stamp */
1045 __u64 ln_remote_nets_version;
1046 /* list of all known routers */
1047 struct list_head ln_routers;
1048 /* validity stamp */
1049 __u64 ln_routers_version;
1050 /* percpt router buffer pools */
1051 struct lnet_rtrbufpool **ln_rtrpools;
1054 * Ping target / Push source
1056 * The ping target and push source share a single buffer. The
1057 * ln_ping_target is protected against concurrent updates by
1060 struct lnet_handle_md ln_ping_target_md;
1061 lnet_handler_t ln_ping_target_handler;
1062 struct lnet_ping_buffer *ln_ping_target;
1063 atomic_t ln_ping_target_seqno;
1068 * ln_push_nnis contains the desired size of the push target.
1069 * The lnet_net_lock is used to handle update races. The old
1070 * buffer may linger a while after it has been unlinked, in
1071 * which case the event handler cleans up.
1073 lnet_handler_t ln_push_target_handler;
1074 struct lnet_handle_md ln_push_target_md;
1075 struct lnet_ping_buffer *ln_push_target;
1076 int ln_push_target_nnis;
1078 /* discovery event queue handle */
1079 lnet_handler_t ln_dc_handler;
1080 /* discovery requests */
1081 struct list_head ln_dc_request;
1082 /* discovery working list */
1083 struct list_head ln_dc_working;
1084 /* discovery expired list */
1085 struct list_head ln_dc_expired;
1086 /* discovery thread wait queue */
1087 wait_queue_head_t ln_dc_waitq;
1088 /* discovery startup/shutdown state */
1091 /* monitor thread startup/shutdown state */
1093 /* serialise startup/shutdown */
1094 struct semaphore ln_mt_signal;
1096 struct mutex ln_api_mutex;
1097 struct mutex ln_lnd_mutex;
1098 /* Have I called LNetNIInit myself? */
1100 /* LNetNIInit/LNetNIFini counter */
1102 /* SHUTDOWN/RUNNING/STOPPING */
1105 int ln_routing; /* am I a router? */
1106 lnet_pid_t ln_pid; /* requested pid */
1107 /* uniquely identifies this ni in this epoch */
1108 __u64 ln_interface_cookie;
1109 /* registered LNDs */
1110 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1112 /* test protocol compatibility flags */
1113 unsigned long ln_testprotocompat;
1115 /* 0 - load the NIs from the mod params
1116 * 1 - do not load the NIs from the mod params
1117 * Reverse logic to ensure that other calls to LNetNIInit
1120 bool ln_nis_from_mod_params;
1123 * completion for the monitor thread. The monitor thread takes care of
1124 * checking routes, timedout messages and resending messages.
1126 struct completion ln_mt_wait_complete;
1128 /* per-cpt resend queues */
1129 struct list_head **ln_mt_resendqs;
1130 /* local NIs to recover */
1131 struct list_head ln_mt_localNIRecovq;
1132 /* local NIs to recover */
1133 struct list_head ln_mt_peerNIRecovq;
1135 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1136 * There are CPT number of queues. Since response trackers will be
1137 * added on the fast path we can't afford to grab the exclusive
1138 * net lock to protect these queues. The CPT will be calculated
1139 * based on the mdh cookie.
1141 struct list_head **ln_mt_rstq;
1143 * A response tracker becomes a zombie when the associated MD is queued
1144 * for unlink before the response tracker is detached from the MD. An
1145 * entry on a zombie list can be freed when either the remaining
1146 * operations on the MD complete or when LNet has shut down.
1148 struct list_head **ln_mt_zombie_rstqs;
1149 /* recovery handler */
1150 lnet_handler_t ln_mt_handler;
1153 * Completed when the discovery and monitor threads can enter their
1156 struct completion ln_started;