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 lnet_eq_handler_t eq_callback;
187 int **eq_refs; /* percpt refcount for EQ */
191 struct list_head me_list;
193 struct lnet_process_id me_match_id;
194 unsigned int me_portal;
195 unsigned int me_pos; /* hash offset in mt_hash */
197 __u64 me_ignore_bits;
198 enum lnet_unlink me_unlink;
199 struct lnet_libmd *me_md;
203 struct list_head md_list;
204 struct lnet_libhandle md_lh;
205 struct lnet_me *md_me;
207 unsigned int md_offset;
208 unsigned int md_length;
209 unsigned int md_max_size;
212 unsigned int md_options;
213 unsigned int md_flags;
214 unsigned int md_niov; /* # frags at end of struct */
216 struct lnet_rsp_tracker *md_rspt_ptr;
217 struct lnet_eq *md_eq;
218 struct lnet_handle_md md_bulk_handle;
220 struct kvec iov[LNET_MAX_IOV];
221 lnet_kiov_t kiov[LNET_MAX_IOV];
225 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
226 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
227 #define LNET_MD_FLAG_ABORTED (1 << 2)
229 struct lnet_test_peer {
230 /* info about peers we are trying to fail */
231 struct list_head tp_list; /* ln_test_peers */
232 lnet_nid_t tp_nid; /* matching nid */
233 unsigned int tp_threshold; /* # failures to simulate */
236 #define LNET_COOKIE_TYPE_MD 1
237 #define LNET_COOKIE_TYPE_ME 2
238 #define LNET_COOKIE_TYPE_EQ 3
239 #define LNET_COOKIE_TYPE_BITS 2
240 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
242 struct lnet_ni; /* forward ref */
246 /* fields initialized by the LND */
249 int (*lnd_startup)(struct lnet_ni *ni);
250 void (*lnd_shutdown)(struct lnet_ni *ni);
251 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
253 /* In data movement APIs below, payload buffers are described as a set
254 * of 'niov' fragments which are...
256 * in virtual memory (struct kvec *iov != NULL)
258 * in pages (kernel only: plt_kiov_t *kiov != NULL).
259 * The LND may NOT overwrite these fragment descriptors.
260 * An 'offset' and may specify a byte offset within the set of
261 * fragments to start from
264 /* Start sending a preformatted message. 'private' is NULL for PUT and
265 * GET messages; otherwise this is a response to an incoming message
266 * and 'private' is the 'private' passed to lnet_parse(). Return
267 * non-zero for immediate failure, otherwise complete later with
269 int (*lnd_send)(struct lnet_ni *ni, void *private,
270 struct lnet_msg *msg);
272 /* Start receiving 'mlen' bytes of payload data, skipping the following
273 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
274 * lnet_parse(). Return non-zero for immedaite failure, otherwise
275 * complete later with lnet_finalize(). This also gives back a receive
276 * credit if the LND does flow control. */
277 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
278 int delayed, unsigned int niov,
279 struct kvec *iov, lnet_kiov_t *kiov,
280 unsigned int offset, unsigned int mlen, unsigned int rlen);
282 /* lnet_parse() has had to delay processing of this message
283 * (e.g. waiting for a forwarding buffer or send credits). Give the
284 * LND a chance to free urgently needed resources. If called, return 0
285 * for success and do NOT give back a receive credit; that has to wait
286 * until lnd_recv() gets called. On failure return < 0 and
287 * release resources; lnd_recv() will not be called. */
288 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
289 struct lnet_msg *msg, void **new_privatep);
291 /* notification of peer down */
292 void (*lnd_notify_peer_down)(lnet_nid_t peer);
294 /* accept a new connection */
295 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
298 struct lnet_tx_queue {
299 int tq_credits; /* # tx credits free */
300 int tq_credits_min; /* lowest it's been */
301 int tq_credits_max; /* total # tx credits */
302 struct list_head tq_delayed; /* delayed TXs */
305 enum lnet_net_state {
306 /* set when net block is allocated */
307 LNET_NET_STATE_INIT = 0,
308 /* set when NIs in net are started successfully */
309 LNET_NET_STATE_ACTIVE,
310 /* set if all NIs in net are in FAILED state */
311 LNET_NET_STATE_INACTIVE,
312 /* set when shutting down a NET */
313 LNET_NET_STATE_DELETING
317 /* initial state when NI is created */
318 LNET_NI_STATE_INIT = 0,
319 /* set when NI is brought up */
320 LNET_NI_STATE_ACTIVE,
321 /* set when NI is being shutdown */
322 LNET_NI_STATE_DELETING,
325 #define LNET_NI_RECOVERY_PENDING BIT(0)
326 #define LNET_NI_RECOVERY_FAILED BIT(1)
328 enum lnet_stats_type {
329 LNET_STATS_TYPE_SEND = 0,
330 LNET_STATS_TYPE_RECV,
334 struct lnet_comm_count {
335 atomic_t co_get_count;
336 atomic_t co_put_count;
337 atomic_t co_reply_count;
338 atomic_t co_ack_count;
339 atomic_t co_hello_count;
342 struct lnet_element_stats {
343 struct lnet_comm_count el_send_stats;
344 struct lnet_comm_count el_recv_stats;
345 struct lnet_comm_count el_drop_stats;
348 struct lnet_health_local_stats {
349 atomic_t hlt_local_interrupt;
350 atomic_t hlt_local_dropped;
351 atomic_t hlt_local_aborted;
352 atomic_t hlt_local_no_route;
353 atomic_t hlt_local_timeout;
354 atomic_t hlt_local_error;
357 struct lnet_health_remote_stats {
358 atomic_t hlt_remote_dropped;
359 atomic_t hlt_remote_timeout;
360 atomic_t hlt_remote_error;
361 atomic_t hlt_network_timeout;
365 /* chain on the ln_nets */
366 struct list_head net_list;
368 /* net ID, which is composed of
369 * (net_type << 16) | net_num.
370 * net_type can be one of the enumerated types defined in
371 * lnet/include/lnet/nidstr.h */
374 /* priority of the network */
377 /* total number of CPTs in the array */
380 /* cumulative CPTs of all NIs in this net */
383 /* network tunables */
384 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
387 * boolean to indicate that the tunables have been set and
390 bool net_tunables_set;
392 /* procedural interface */
393 const struct lnet_lnd *net_lnd;
395 /* list of NIs on this net */
396 struct list_head net_ni_list;
398 /* list of NIs being added, but not started yet */
399 struct list_head net_ni_added;
401 /* dying LND instances */
402 struct list_head net_ni_zombie;
404 /* when I was last alive */
405 time64_t net_last_alive;
407 /* protects access to net_last_alive */
412 /* chain on the lnet_net structure */
413 struct list_head ni_netlist;
415 /* chain on the recovery queue */
416 struct list_head ni_recovery;
418 /* MD handle for recovery ping */
419 struct lnet_handle_md ni_ping_mdh;
426 /* bond NI on some CPTs */
429 /* interface's NID */
432 /* instance-specific data */
436 atomic_t ni_tx_credits;
438 /* percpt TX queues */
439 struct lnet_tx_queue **ni_tx_queues;
441 /* percpt reference count */
444 /* pointer to parent network */
445 struct lnet_net *ni_net;
447 /* my health status */
448 struct lnet_ni_status *ni_status;
450 /* NI FSM. Protected by lnet_ni_lock() */
451 enum lnet_ni_state ni_state;
453 /* Recovery state. Protected by lnet_ni_lock() */
454 __u32 ni_recovery_state;
456 /* per NI LND tunables */
457 struct lnet_lnd_tunables ni_lnd_tunables;
459 /* lnd tunables set explicitly */
460 bool ni_lnd_tunables_set;
463 struct lnet_element_stats ni_stats;
464 struct lnet_health_local_stats ni_hstats;
466 /* physical device CPT */
469 /* sequence number used to round robin over nis within a net */
474 * initialized to LNET_MAX_HEALTH_VALUE
475 * Value is decremented every time we fail to send a message over
476 * this NI because of a NI specific failure.
477 * Value is incremented if we successfully send a message.
482 * Set to 1 by the LND when it receives an event telling it the device
483 * has gone into a fatal state. Set to 0 when the LND receives an
484 * even telling it the device is back online.
486 atomic_t ni_fatal_error_on;
489 * equivalent interfaces to use
490 * This is an array because socklnd bonding can still be configured
492 char *ni_interfaces[LNET_INTERFACES_NUM];
493 struct net *ni_net_ns; /* original net namespace */
496 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
499 * Descriptor of a ping info buffer: keep a separate indicator of the
500 * size and a reference count. The type is used both as a source and
501 * sink of data, so we need to keep some information outside of the
502 * area that may be overwritten by network data.
504 struct lnet_ping_buffer {
507 struct lnet_ping_info pb_info;
510 #define LNET_PING_BUFFER_SIZE(NNIDS) \
511 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
512 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
513 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
515 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
516 container_of((PINFO), struct lnet_ping_buffer, pb_info)
518 struct lnet_peer_ni {
519 /* chain on lpn_peer_nis */
520 struct list_head lpni_peer_nis;
521 /* chain on remote peer list */
522 struct list_head lpni_on_remote_peer_ni_list;
523 /* chain on recovery queue */
524 struct list_head lpni_recovery;
525 /* chain on peer hash */
526 struct list_head lpni_hashlist;
527 /* messages blocking for tx credits */
528 struct list_head lpni_txq;
529 /* pointer to peer net I'm part of */
530 struct lnet_peer_net *lpni_peer_net;
531 /* statistics kept on each peer NI */
532 struct lnet_element_stats lpni_stats;
533 struct lnet_health_remote_stats lpni_hstats;
534 /* spin lock protecting credits and lpni_txq */
535 spinlock_t lpni_lock;
536 /* # tx credits available */
539 int lpni_mintxcredits;
541 * Each peer_ni in a gateway maintains its own credits. This
542 * allows more traffic to gateways that have multiple interfaces.
544 /* # router credits */
547 int lpni_minrtrcredits;
548 /* bytes queued for sending */
550 /* network peer is on */
551 struct lnet_net *lpni_net;
555 atomic_t lpni_refcount;
556 /* health value for the peer */
557 atomic_t lpni_healthv;
558 /* recovery ping mdh */
559 struct lnet_handle_md lpni_recovery_ping_mdh;
560 /* CPT this peer attached on */
562 /* state flags -- protected by lpni_lock */
564 /* status of the peer NI as reported by the peer */
565 __u32 lpni_ns_status;
566 /* sequence number used to round robin over peer nis within a net */
568 /* sequence number used to round robin over gateways */
570 /* returned RC ping features. Protected with lpni_lock */
571 unsigned int lpni_ping_feats;
572 /* time last message was received from the peer */
573 time64_t lpni_last_alive;
574 /* preferred local nids: if only one, use lpni_pref.nid */
579 /* number of preferred NIDs in lnpi_pref_nids */
580 __u32 lpni_pref_nnids;
583 /* Preferred path added due to traffic on non-MR peer_ni */
584 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
585 /* peer is being recovered. */
586 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
587 /* recovery ping failed */
588 #define LNET_PEER_NI_RECOVERY_FAILED (1 << 2)
589 /* peer is being deleted */
590 #define LNET_PEER_NI_DELETING (1 << 3)
593 /* chain on pt_peer_list */
594 struct list_head lp_peer_list;
596 /* list of peer nets */
597 struct list_head lp_peer_nets;
599 /* list of messages pending discovery*/
600 struct list_head lp_dc_pendq;
602 /* chain on router list */
603 struct list_head lp_rtr_list;
605 /* primary NID of the peer */
606 lnet_nid_t lp_primary_nid;
608 /* net to perform discovery on */
609 __u32 lp_disc_net_id;
611 /* CPT of peer_table */
614 /* number of NIDs on this peer */
617 /* # refs from lnet_route::lr_gateway */
621 * peer specific health sensitivity value to decrement peer nis in
622 * this peer with if set to something other than 0
624 __u32 lp_health_sensitivity;
626 /* messages blocking for router credits */
627 struct list_head lp_rtrq;
629 /* routes on this peer */
630 struct list_head lp_routes;
632 /* reference count */
633 atomic_t lp_refcount;
635 /* lock protecting peer state flags and lpni_rtrq */
638 /* peer state flags */
641 /* buffer for data pushed by peer */
642 struct lnet_ping_buffer *lp_data;
644 /* MD handle for ping in progress */
645 struct lnet_handle_md lp_ping_mdh;
647 /* MD handle for push in progress */
648 struct lnet_handle_md lp_push_mdh;
650 /* number of NIDs for sizing push data */
653 /* NI config sequence number of peer */
656 /* Local NI config sequence number acked by peer */
659 /* Local NI config sequence number sent to peer */
660 __u32 lp_node_seqno_sent;
662 /* Ping error encountered during discovery. */
665 /* Push error encountered during discovery. */
668 /* Error encountered during discovery. */
671 /* time it was put on the ln_dc_working queue */
672 time64_t lp_last_queued;
674 /* link on discovery-related lists */
675 struct list_head lp_dc_list;
677 /* tasks waiting on discovery of this peer */
678 wait_queue_head_t lp_dc_waitq;
680 /* cached peer aliveness */
685 * The status flags in lp_state. Their semantics have chosen so that
686 * lp_state can be zero-initialized.
688 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
689 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
691 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
692 * NOT set when the peer was pinged by discovery.
694 * A peer is marked ROUTER if it indicates so in the feature bit.
696 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
697 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
698 #define LNET_PEER_ROUTER_ENABLED (1 << 2) /* router feature enabled */
701 * A peer is marked CONFIGURED if it was configured by DLC.
703 * In addition, a peer is marked DISCOVERED if it has fully passed
704 * through Peer Discovery.
706 * When Peer Discovery is disabled, the discovery thread will mark
707 * peers REDISCOVER to indicate that they should be re-examined if
708 * discovery is (re)enabled on the node.
710 * A peer that was created as the result of inbound traffic will not
713 #define LNET_PEER_CONFIGURED (1 << 3) /* Configured via DLC */
714 #define LNET_PEER_DISCOVERED (1 << 4) /* Peer was discovered */
715 #define LNET_PEER_REDISCOVER (1 << 5) /* Discovery was disabled */
717 * A peer is marked DISCOVERING when discovery is in progress.
718 * The other flags below correspond to stages of discovery.
720 #define LNET_PEER_DISCOVERING (1 << 6) /* Discovering */
721 #define LNET_PEER_DATA_PRESENT (1 << 7) /* Remote peer data present */
722 #define LNET_PEER_NIDS_UPTODATE (1 << 8) /* Remote peer info uptodate */
723 #define LNET_PEER_PING_SENT (1 << 9) /* Waiting for REPLY to Ping */
724 #define LNET_PEER_PUSH_SENT (1 << 10) /* Waiting for ACK of Push */
725 #define LNET_PEER_PING_FAILED (1 << 11) /* Ping send failure */
726 #define LNET_PEER_PUSH_FAILED (1 << 12) /* Push send failure */
728 * A ping can be forced as a way to fix up state, or as a manual
729 * intervention by an admin.
730 * A push can be forced in circumstances that would normally not
731 * allow for one to happen.
733 #define LNET_PEER_FORCE_PING (1 << 13) /* Forced Ping */
734 #define LNET_PEER_FORCE_PUSH (1 << 14) /* Forced Push */
736 /* force delete even if router */
737 #define LNET_PEER_RTR_NI_FORCE_DEL (1 << 15)
739 /* gw undergoing alive discovery */
740 #define LNET_PEER_RTR_DISCOVERY (1 << 16)
741 /* gw has undergone discovery (does not indicate success or failure) */
742 #define LNET_PEER_RTR_DISCOVERED (1 << 17)
744 struct lnet_peer_net {
745 /* chain on lp_peer_nets */
746 struct list_head lpn_peer_nets;
748 /* list of peer_nis on this network */
749 struct list_head lpn_peer_nis;
751 /* pointer to the peer I'm part of */
752 struct lnet_peer *lpn_peer;
757 /* peer net health */
760 /* time of last router net check attempt */
761 time64_t lpn_rtrcheck_timestamp;
763 /* selection sequence number */
766 /* reference count */
767 atomic_t lpn_refcount;
771 #define LNET_PEER_HASH_BITS 9
772 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
775 * peer hash table - one per CPT
777 * protected by lnet_net_lock/EX for update
782 * protected by pt_zombie_lock:
786 * pt_zombie lock nests inside lnet_net_lock
788 struct lnet_peer_table {
789 int pt_version; /* /proc validity stamp */
790 struct list_head *pt_hash; /* NID->peer hash */
791 struct list_head pt_peer_list; /* peers */
792 int pt_peers; /* # peers */
793 struct list_head pt_zombie_list; /* zombie peer_ni */
794 int pt_zombies; /* # zombie peers_ni */
795 spinlock_t pt_zombie_lock; /* protect list and count */
798 /* peer aliveness is enabled only on routers for peers in a network where the
799 * struct lnet_ni::ni_peertimeout has been set to a positive value
801 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
802 ((lp)->lpni_net) && \
803 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
806 struct list_head lr_list; /* chain on net */
807 struct list_head lr_gwlist; /* chain on gateway */
808 struct lnet_peer *lr_gateway; /* router node */
809 lnet_nid_t lr_nid; /* NID used to add route */
810 __u32 lr_net; /* remote network number */
811 __u32 lr_lnet; /* local network number */
812 int lr_seq; /* sequence for round-robin */
813 __u32 lr_hops; /* how far I am */
814 unsigned int lr_priority; /* route priority */
815 bool lr_alive; /* cached route aliveness */
818 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
819 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
820 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
822 struct lnet_remotenet {
823 /* chain on ln_remote_nets_hash */
824 struct list_head lrn_list;
826 struct list_head lrn_routes;
831 /** lnet message has credit and can be submitted to lnd for send/receive */
832 #define LNET_CREDIT_OK 0
833 /** lnet message is waiting for credit */
834 #define LNET_CREDIT_WAIT 1
835 /** lnet message is waiting for discovery */
836 #define LNET_DC_WAIT 2
838 struct lnet_rtrbufpool {
839 /* my free buffer pool */
840 struct list_head rbp_bufs;
841 /* messages blocking for a buffer */
842 struct list_head rbp_msgs;
843 /* # pages in each buffer */
845 /* requested number of buffers */
846 int rbp_req_nbuffers;
847 /* # buffers actually allocated */
849 /* # free buffers / blocked messages */
856 struct list_head rb_list; /* chain on rbp_bufs */
857 struct lnet_rtrbufpool *rb_pool; /* owning pool */
858 lnet_kiov_t rb_kiov[0]; /* the buffer space */
861 #define LNET_PEER_HASHSIZE 503 /* prime! */
863 enum lnet_match_flags {
864 /* Didn't match anything */
865 LNET_MATCHMD_NONE = (1 << 0),
867 LNET_MATCHMD_OK = (1 << 1),
868 /* Must be discarded */
869 LNET_MATCHMD_DROP = (1 << 2),
870 /* match and buffer is exhausted */
871 LNET_MATCHMD_EXHAUSTED = (1 << 3),
873 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
876 /* Options for struct lnet_portal::ptl_options */
877 #define LNET_PTL_LAZY (1 << 0)
878 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
879 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
881 /* parameter for matching operations (GET, PUT) */
882 struct lnet_match_info {
884 struct lnet_process_id mi_id;
887 unsigned int mi_portal;
888 unsigned int mi_rlength;
889 unsigned int mi_roffset;
892 /* ME hash of RDMA portal */
893 #define LNET_MT_HASH_BITS 8
894 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
895 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
896 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
897 * the last entry is reserved for MEs with ignore-bits */
898 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
899 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
900 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
901 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
902 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
903 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
904 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
906 /* portal match table */
907 struct lnet_match_table {
908 /* reserved for upcoming patches, CPU partition ID */
910 unsigned int mt_portal; /* portal index */
911 /* match table is set as "enabled" if there's non-exhausted MD
912 * attached on mt_mhash, it's only valid for wildcard portal */
913 unsigned int mt_enabled;
914 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
915 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
916 struct list_head *mt_mhash; /* matching hash */
919 /* these are only useful for wildcard portal */
920 /* Turn off message rotor for wildcard portals */
921 #define LNET_PTL_ROTOR_OFF 0
922 /* round-robin dispatch all PUT messages for wildcard portals */
923 #define LNET_PTL_ROTOR_ON 1
924 /* round-robin dispatch routed PUT message for wildcard portals */
925 #define LNET_PTL_ROTOR_RR_RT 2
926 /* dispatch routed PUT message by hashing source NID for wildcard portals */
927 #define LNET_PTL_ROTOR_HASH_RT 3
931 unsigned int ptl_index; /* portal ID, reserved */
932 /* flags on this portal: lazy, unique... */
933 unsigned int ptl_options;
934 /* list of messages which are stealing buffer */
935 struct list_head ptl_msg_stealing;
936 /* messages blocking for MD */
937 struct list_head ptl_msg_delayed;
938 /* Match table for each CPT */
939 struct lnet_match_table **ptl_mtables;
940 /* spread rotor of incoming "PUT" */
941 unsigned int ptl_rotor;
942 /* # active entries for this portal */
944 /* array of active entries' cpu-partition-id */
948 #define LNET_LH_HASH_BITS 12
949 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
950 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
952 /* resource container (ME, MD, EQ) */
953 struct lnet_res_container {
954 unsigned int rec_type; /* container type */
955 __u64 rec_lh_cookie; /* cookie generator */
956 struct list_head rec_active; /* active resource list */
957 struct list_head *rec_lh_hash; /* handle hash */
960 /* message container */
961 struct lnet_msg_container {
962 int msc_init; /* initialized or not */
963 /* max # threads finalizing */
965 /* msgs waiting to complete finalizing */
966 struct list_head msc_finalizing;
967 /* msgs waiting to be resent */
968 struct list_head msc_resending;
969 struct list_head msc_active; /* active message list */
970 /* threads doing finalization */
971 void **msc_finalizers;
972 /* threads doing resends */
973 void **msc_resenders;
976 /* Peer Discovery states */
977 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
978 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
979 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
981 /* Router Checker states */
982 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
983 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
984 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
987 #define LNET_STATE_SHUTDOWN 0 /* not started */
988 #define LNET_STATE_RUNNING 1 /* started up OK */
989 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
992 /* CPU partition table of LNet */
993 struct cfs_cpt_table *ln_cpt_table;
994 /* number of CPTs in ln_cpt_table */
995 unsigned int ln_cpt_number;
996 unsigned int ln_cpt_bits;
998 /* protect LNet resources (ME/MD/EQ) */
999 struct cfs_percpt_lock *ln_res_lock;
1002 /* the vector of portals */
1003 struct lnet_portal **ln_portals;
1004 /* percpt MD container */
1005 struct lnet_res_container **ln_md_containers;
1007 /* Event Queue container */
1008 struct lnet_res_container ln_eq_container;
1009 spinlock_t ln_eq_wait_lock;
1011 unsigned int ln_remote_nets_hbits;
1013 /* protect NI, peer table, credits, routers, rtrbuf... */
1014 struct cfs_percpt_lock *ln_net_lock;
1015 /* percpt message containers for active/finalizing/freed message */
1016 struct lnet_msg_container **ln_msg_containers;
1017 struct lnet_counters **ln_counters;
1018 struct lnet_peer_table **ln_peer_tables;
1019 /* list of peer nis not on a local network */
1020 struct list_head ln_remote_peer_ni_list;
1021 /* failure simulation */
1022 struct list_head ln_test_peers;
1023 struct list_head ln_drop_rules;
1024 struct list_head ln_delay_rules;
1026 struct list_head ln_nets;
1027 /* the loopback NI */
1028 struct lnet_ni *ln_loni;
1029 /* network zombie list */
1030 struct list_head ln_net_zombie;
1031 /* resend messages list */
1032 struct list_head ln_msg_resend;
1033 /* spin lock to protect the msg resend list */
1034 spinlock_t ln_msg_resend_lock;
1036 /* remote networks with routes to them */
1037 struct list_head *ln_remote_nets_hash;
1038 /* validity stamp */
1039 __u64 ln_remote_nets_version;
1040 /* list of all known routers */
1041 struct list_head ln_routers;
1042 /* validity stamp */
1043 __u64 ln_routers_version;
1044 /* percpt router buffer pools */
1045 struct lnet_rtrbufpool **ln_rtrpools;
1048 * Ping target / Push source
1050 * The ping target and push source share a single buffer. The
1051 * ln_ping_target is protected against concurrent updates by
1054 struct lnet_handle_md ln_ping_target_md;
1055 struct lnet_eq *ln_ping_target_eq;
1056 struct lnet_ping_buffer *ln_ping_target;
1057 atomic_t ln_ping_target_seqno;
1062 * ln_push_nnis contains the desired size of the push target.
1063 * The lnet_net_lock is used to handle update races. The old
1064 * buffer may linger a while after it has been unlinked, in
1065 * which case the event handler cleans up.
1067 struct lnet_eq *ln_push_target_eq;
1068 struct lnet_handle_md ln_push_target_md;
1069 struct lnet_ping_buffer *ln_push_target;
1070 int ln_push_target_nnis;
1072 /* discovery event queue handle */
1073 struct lnet_eq *ln_dc_eq;
1074 /* discovery requests */
1075 struct list_head ln_dc_request;
1076 /* discovery working list */
1077 struct list_head ln_dc_working;
1078 /* discovery expired list */
1079 struct list_head ln_dc_expired;
1080 /* discovery thread wait queue */
1081 wait_queue_head_t ln_dc_waitq;
1082 /* discovery startup/shutdown state */
1085 /* monitor thread startup/shutdown state */
1087 /* serialise startup/shutdown */
1088 struct semaphore ln_mt_signal;
1090 struct mutex ln_api_mutex;
1091 struct mutex ln_lnd_mutex;
1092 /* Have I called LNetNIInit myself? */
1094 /* LNetNIInit/LNetNIFini counter */
1096 /* SHUTDOWN/RUNNING/STOPPING */
1099 int ln_routing; /* am I a router? */
1100 lnet_pid_t ln_pid; /* requested pid */
1101 /* uniquely identifies this ni in this epoch */
1102 __u64 ln_interface_cookie;
1103 /* registered LNDs */
1104 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1106 /* test protocol compatibility flags */
1107 unsigned long ln_testprotocompat;
1109 /* 0 - load the NIs from the mod params
1110 * 1 - do not load the NIs from the mod params
1111 * Reverse logic to ensure that other calls to LNetNIInit
1114 bool ln_nis_from_mod_params;
1117 * completion for the monitor thread. The monitor thread takes care of
1118 * checking routes, timedout messages and resending messages.
1120 struct completion ln_mt_wait_complete;
1122 /* per-cpt resend queues */
1123 struct list_head **ln_mt_resendqs;
1124 /* local NIs to recover */
1125 struct list_head ln_mt_localNIRecovq;
1126 /* local NIs to recover */
1127 struct list_head ln_mt_peerNIRecovq;
1129 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1130 * There are CPT number of queues. Since response trackers will be
1131 * added on the fast path we can't afford to grab the exclusive
1132 * net lock to protect these queues. The CPT will be calculated
1133 * based on the mdh cookie.
1135 struct list_head **ln_mt_rstq;
1137 * A response tracker becomes a zombie when the associated MD is queued
1138 * for unlink before the response tracker is detached from the MD. An
1139 * entry on a zombie list can be freed when either the remaining
1140 * operations on the MD complete or when LNet has shut down.
1142 struct list_head **ln_mt_zombie_rstqs;
1143 /* recovery eq handler */
1144 struct lnet_eq *ln_mt_eq;
1147 * Completed when the discovery and monitor threads can enter their
1150 struct completion ln_started;