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 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 (1 << 0)
218 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
219 #define LNET_MD_FLAG_ABORTED (1 << 2)
221 struct lnet_test_peer {
222 /* info about peers we are trying to fail */
223 struct list_head tp_list; /* ln_test_peers */
224 lnet_nid_t tp_nid; /* matching nid */
225 unsigned int tp_threshold; /* # failures to simulate */
228 #define LNET_COOKIE_TYPE_MD 1
229 #define LNET_COOKIE_TYPE_ME 2
230 #define LNET_COOKIE_TYPE_EQ 3
231 #define LNET_COOKIE_TYPE_BITS 2
232 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
234 struct lnet_ni; /* forward ref */
238 /* fields initialized by the LND */
241 int (*lnd_startup)(struct lnet_ni *ni);
242 void (*lnd_shutdown)(struct lnet_ni *ni);
243 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
245 /* In data movement APIs below, payload buffers are described as a set
246 * of 'niov' fragments which are...
248 * in virtual memory (struct kvec *iov != NULL)
250 * in pages (kernel only: plt_kiov_t *kiov != NULL).
251 * The LND may NOT overwrite these fragment descriptors.
252 * An 'offset' and may specify a byte offset within the set of
253 * fragments to start from
256 /* Start sending a preformatted message. 'private' is NULL for PUT and
257 * GET messages; otherwise this is a response to an incoming message
258 * and 'private' is the 'private' passed to lnet_parse(). Return
259 * non-zero for immediate failure, otherwise complete later with
261 int (*lnd_send)(struct lnet_ni *ni, void *private,
262 struct lnet_msg *msg);
264 /* Start receiving 'mlen' bytes of payload data, skipping the following
265 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
266 * lnet_parse(). Return non-zero for immedaite failure, otherwise
267 * complete later with lnet_finalize(). This also gives back a receive
268 * credit if the LND does flow control. */
269 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
270 int delayed, unsigned int niov,
271 struct kvec *iov, struct bio_vec *kiov,
272 unsigned int offset, unsigned int mlen, unsigned int rlen);
274 /* lnet_parse() has had to delay processing of this message
275 * (e.g. waiting for a forwarding buffer or send credits). Give the
276 * LND a chance to free urgently needed resources. If called, return 0
277 * for success and do NOT give back a receive credit; that has to wait
278 * until lnd_recv() gets called. On failure return < 0 and
279 * release resources; lnd_recv() will not be called. */
280 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
281 struct lnet_msg *msg, void **new_privatep);
283 /* notification of peer down */
284 void (*lnd_notify_peer_down)(lnet_nid_t peer);
286 /* accept a new connection */
287 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
290 struct lnet_tx_queue {
291 int tq_credits; /* # tx credits free */
292 int tq_credits_min; /* lowest it's been */
293 int tq_credits_max; /* total # tx credits */
294 struct list_head tq_delayed; /* delayed TXs */
297 enum lnet_net_state {
298 /* set when net block is allocated */
299 LNET_NET_STATE_INIT = 0,
300 /* set when NIs in net are started successfully */
301 LNET_NET_STATE_ACTIVE,
302 /* set if all NIs in net are in FAILED state */
303 LNET_NET_STATE_INACTIVE,
304 /* set when shutting down a NET */
305 LNET_NET_STATE_DELETING
309 /* initial state when NI is created */
310 LNET_NI_STATE_INIT = 0,
311 /* set when NI is brought up */
312 LNET_NI_STATE_ACTIVE,
313 /* set when NI is being shutdown */
314 LNET_NI_STATE_DELETING,
317 #define LNET_NI_RECOVERY_PENDING BIT(0)
318 #define LNET_NI_RECOVERY_FAILED BIT(1)
320 enum lnet_stats_type {
321 LNET_STATS_TYPE_SEND = 0,
322 LNET_STATS_TYPE_RECV,
326 struct lnet_comm_count {
327 atomic_t co_get_count;
328 atomic_t co_put_count;
329 atomic_t co_reply_count;
330 atomic_t co_ack_count;
331 atomic_t co_hello_count;
334 struct lnet_element_stats {
335 struct lnet_comm_count el_send_stats;
336 struct lnet_comm_count el_recv_stats;
337 struct lnet_comm_count el_drop_stats;
340 struct lnet_health_local_stats {
341 atomic_t hlt_local_interrupt;
342 atomic_t hlt_local_dropped;
343 atomic_t hlt_local_aborted;
344 atomic_t hlt_local_no_route;
345 atomic_t hlt_local_timeout;
346 atomic_t hlt_local_error;
349 struct lnet_health_remote_stats {
350 atomic_t hlt_remote_dropped;
351 atomic_t hlt_remote_timeout;
352 atomic_t hlt_remote_error;
353 atomic_t hlt_network_timeout;
357 /* chain on the ln_nets */
358 struct list_head net_list;
360 /* net ID, which is composed of
361 * (net_type << 16) | net_num.
362 * net_type can be one of the enumerated types defined in
363 * lnet/include/lnet/nidstr.h */
366 /* priority of the network */
369 /* total number of CPTs in the array */
372 /* cumulative CPTs of all NIs in this net */
375 /* network tunables */
376 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
379 * boolean to indicate that the tunables have been set and
382 bool net_tunables_set;
384 /* procedural interface */
385 const struct lnet_lnd *net_lnd;
387 /* list of NIs on this net */
388 struct list_head net_ni_list;
390 /* list of NIs being added, but not started yet */
391 struct list_head net_ni_added;
393 /* dying LND instances */
394 struct list_head net_ni_zombie;
396 /* when I was last alive */
397 time64_t net_last_alive;
399 /* protects access to net_last_alive */
404 /* chain on the lnet_net structure */
405 struct list_head ni_netlist;
407 /* chain on the recovery queue */
408 struct list_head ni_recovery;
410 /* MD handle for recovery ping */
411 struct lnet_handle_md ni_ping_mdh;
418 /* bond NI on some CPTs */
421 /* interface's NID */
424 /* instance-specific data */
428 atomic_t ni_tx_credits;
430 /* percpt TX queues */
431 struct lnet_tx_queue **ni_tx_queues;
433 /* percpt reference count */
436 /* pointer to parent network */
437 struct lnet_net *ni_net;
439 /* my health status */
440 struct lnet_ni_status *ni_status;
442 /* NI FSM. Protected by lnet_ni_lock() */
443 enum lnet_ni_state ni_state;
445 /* Recovery state. Protected by lnet_ni_lock() */
446 __u32 ni_recovery_state;
448 /* per NI LND tunables */
449 struct lnet_lnd_tunables ni_lnd_tunables;
451 /* lnd tunables set explicitly */
452 bool ni_lnd_tunables_set;
455 struct lnet_element_stats ni_stats;
456 struct lnet_health_local_stats ni_hstats;
458 /* physical device CPT */
461 /* sequence number used to round robin over nis within a net */
466 * initialized to LNET_MAX_HEALTH_VALUE
467 * Value is decremented every time we fail to send a message over
468 * this NI because of a NI specific failure.
469 * Value is incremented if we successfully send a message.
474 * Set to 1 by the LND when it receives an event telling it the device
475 * has gone into a fatal state. Set to 0 when the LND receives an
476 * even telling it the device is back online.
478 atomic_t ni_fatal_error_on;
481 * equivalent interfaces to use
482 * This is an array because socklnd bonding can still be configured
484 char *ni_interfaces[LNET_INTERFACES_NUM];
485 struct net *ni_net_ns; /* original net namespace */
488 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
491 * Descriptor of a ping info buffer: keep a separate indicator of the
492 * size and a reference count. The type is used both as a source and
493 * sink of data, so we need to keep some information outside of the
494 * area that may be overwritten by network data.
496 struct lnet_ping_buffer {
500 struct lnet_ping_info pb_info;
503 #define LNET_PING_BUFFER_SIZE(NNIDS) \
504 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
505 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
506 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
508 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
509 container_of((PINFO), struct lnet_ping_buffer, pb_info)
511 struct lnet_peer_ni {
512 /* chain on lpn_peer_nis */
513 struct list_head lpni_peer_nis;
514 /* chain on remote peer list */
515 struct list_head lpni_on_remote_peer_ni_list;
516 /* chain on recovery queue */
517 struct list_head lpni_recovery;
518 /* chain on peer hash */
519 struct list_head lpni_hashlist;
520 /* messages blocking for tx credits */
521 struct list_head lpni_txq;
522 /* pointer to peer net I'm part of */
523 struct lnet_peer_net *lpni_peer_net;
524 /* statistics kept on each peer NI */
525 struct lnet_element_stats lpni_stats;
526 struct lnet_health_remote_stats lpni_hstats;
527 /* spin lock protecting credits and lpni_txq */
528 spinlock_t lpni_lock;
529 /* # tx credits available */
532 int lpni_mintxcredits;
534 * Each peer_ni in a gateway maintains its own credits. This
535 * allows more traffic to gateways that have multiple interfaces.
537 /* # router credits */
540 int lpni_minrtrcredits;
541 /* bytes queued for sending */
543 /* network peer is on */
544 struct lnet_net *lpni_net;
548 atomic_t lpni_refcount;
549 /* health value for the peer */
550 atomic_t lpni_healthv;
551 /* recovery ping mdh */
552 struct lnet_handle_md lpni_recovery_ping_mdh;
553 /* CPT this peer attached on */
555 /* state flags -- protected by lpni_lock */
557 /* status of the peer NI as reported by the peer */
558 __u32 lpni_ns_status;
559 /* sequence number used to round robin over peer nis within a net */
561 /* sequence number used to round robin over gateways */
563 /* returned RC ping features. Protected with lpni_lock */
564 unsigned int lpni_ping_feats;
565 /* time last message was received from the peer */
566 time64_t lpni_last_alive;
567 /* preferred local nids: if only one, use lpni_pref.nid */
572 /* number of preferred NIDs in lnpi_pref_nids */
573 __u32 lpni_pref_nnids;
576 /* Preferred path added due to traffic on non-MR peer_ni */
577 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
578 /* peer is being recovered. */
579 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
580 /* recovery ping failed */
581 #define LNET_PEER_NI_RECOVERY_FAILED (1 << 2)
582 /* peer is being deleted */
583 #define LNET_PEER_NI_DELETING (1 << 3)
586 /* chain on pt_peer_list */
587 struct list_head lp_peer_list;
589 /* list of peer nets */
590 struct list_head lp_peer_nets;
592 /* list of messages pending discovery*/
593 struct list_head lp_dc_pendq;
595 /* chain on router list */
596 struct list_head lp_rtr_list;
598 /* primary NID of the peer */
599 lnet_nid_t lp_primary_nid;
601 /* net to perform discovery on */
602 __u32 lp_disc_net_id;
604 /* CPT of peer_table */
607 /* number of NIDs on this peer */
610 /* # refs from lnet_route::lr_gateway */
614 * peer specific health sensitivity value to decrement peer nis in
615 * this peer with if set to something other than 0
617 __u32 lp_health_sensitivity;
619 /* messages blocking for router credits */
620 struct list_head lp_rtrq;
622 /* routes on this peer */
623 struct list_head lp_routes;
625 /* reference count */
626 atomic_t lp_refcount;
628 /* lock protecting peer state flags and lpni_rtrq */
631 /* peer state flags */
634 /* buffer for data pushed by peer */
635 struct lnet_ping_buffer *lp_data;
637 /* MD handle for ping in progress */
638 struct lnet_handle_md lp_ping_mdh;
640 /* MD handle for push in progress */
641 struct lnet_handle_md lp_push_mdh;
643 /* number of NIDs for sizing push data */
646 /* NI config sequence number of peer */
649 /* Local NI config sequence number acked by peer */
652 /* Local NI config sequence number sent to peer */
653 __u32 lp_node_seqno_sent;
655 /* Ping error encountered during discovery. */
658 /* Push error encountered during discovery. */
661 /* Error encountered during discovery. */
664 /* time it was put on the ln_dc_working queue */
665 time64_t lp_last_queued;
667 /* link on discovery-related lists */
668 struct list_head lp_dc_list;
670 /* tasks waiting on discovery of this peer */
671 wait_queue_head_t lp_dc_waitq;
673 /* cached peer aliveness */
678 * The status flags in lp_state. Their semantics have chosen so that
679 * lp_state can be zero-initialized.
681 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
682 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
684 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
685 * NOT set when the peer was pinged by discovery.
687 * A peer is marked ROUTER if it indicates so in the feature bit.
689 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
690 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
691 #define LNET_PEER_ROUTER_ENABLED (1 << 2) /* router feature enabled */
694 * A peer is marked CONFIGURED if it was configured by DLC.
696 * In addition, a peer is marked DISCOVERED if it has fully passed
697 * through Peer Discovery.
699 * When Peer Discovery is disabled, the discovery thread will mark
700 * peers REDISCOVER to indicate that they should be re-examined if
701 * discovery is (re)enabled on the node.
703 * A peer that was created as the result of inbound traffic will not
706 #define LNET_PEER_CONFIGURED (1 << 3) /* Configured via DLC */
707 #define LNET_PEER_DISCOVERED (1 << 4) /* Peer was discovered */
708 #define LNET_PEER_REDISCOVER (1 << 5) /* Discovery was disabled */
710 * A peer is marked DISCOVERING when discovery is in progress.
711 * The other flags below correspond to stages of discovery.
713 #define LNET_PEER_DISCOVERING (1 << 6) /* Discovering */
714 #define LNET_PEER_DATA_PRESENT (1 << 7) /* Remote peer data present */
715 #define LNET_PEER_NIDS_UPTODATE (1 << 8) /* Remote peer info uptodate */
716 #define LNET_PEER_PING_SENT (1 << 9) /* Waiting for REPLY to Ping */
717 #define LNET_PEER_PUSH_SENT (1 << 10) /* Waiting for ACK of Push */
718 #define LNET_PEER_PING_FAILED (1 << 11) /* Ping send failure */
719 #define LNET_PEER_PUSH_FAILED (1 << 12) /* Push send failure */
721 * A ping can be forced as a way to fix up state, or as a manual
722 * intervention by an admin.
723 * A push can be forced in circumstances that would normally not
724 * allow for one to happen.
726 #define LNET_PEER_FORCE_PING (1 << 13) /* Forced Ping */
727 #define LNET_PEER_FORCE_PUSH (1 << 14) /* Forced Push */
729 /* force delete even if router */
730 #define LNET_PEER_RTR_NI_FORCE_DEL (1 << 15)
732 /* gw undergoing alive discovery */
733 #define LNET_PEER_RTR_DISCOVERY (1 << 16)
734 /* gw has undergone discovery (does not indicate success or failure) */
735 #define LNET_PEER_RTR_DISCOVERED (1 << 17)
737 /* peer is marked for deletion */
738 #define LNET_PEER_MARK_DELETION (1 << 18)
740 struct lnet_peer_net {
741 /* chain on lp_peer_nets */
742 struct list_head lpn_peer_nets;
744 /* list of peer_nis on this network */
745 struct list_head lpn_peer_nis;
747 /* pointer to the peer I'm part of */
748 struct lnet_peer *lpn_peer;
753 /* peer net health */
756 /* time of last router net check attempt */
757 time64_t lpn_rtrcheck_timestamp;
759 /* selection sequence number */
762 /* reference count */
763 atomic_t lpn_refcount;
767 #define LNET_PEER_HASH_BITS 9
768 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
771 * peer hash table - one per CPT
773 * protected by lnet_net_lock/EX for update
778 * protected by pt_zombie_lock:
782 * pt_zombie lock nests inside lnet_net_lock
784 struct lnet_peer_table {
785 int pt_version; /* /proc validity stamp */
786 struct list_head *pt_hash; /* NID->peer hash */
787 struct list_head pt_peer_list; /* peers */
788 int pt_peers; /* # peers */
789 struct list_head pt_zombie_list; /* zombie peer_ni */
790 int pt_zombies; /* # zombie peers_ni */
791 spinlock_t pt_zombie_lock; /* protect list and count */
794 /* peer aliveness is enabled only on routers for peers in a network where the
795 * struct lnet_ni::ni_peertimeout has been set to a positive value
797 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
798 ((lp)->lpni_net) && \
799 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
802 struct list_head lr_list; /* chain on net */
803 struct list_head lr_gwlist; /* chain on gateway */
804 struct lnet_peer *lr_gateway; /* router node */
805 lnet_nid_t lr_nid; /* NID used to add route */
806 __u32 lr_net; /* remote network number */
807 __u32 lr_lnet; /* local network number */
808 int lr_seq; /* sequence for round-robin */
809 __u32 lr_hops; /* how far I am */
810 unsigned int lr_priority; /* route priority */
811 bool lr_alive; /* cached route aliveness */
812 bool lr_single_hop; /* this route is single-hop */
815 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
816 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
817 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
819 struct lnet_remotenet {
820 /* chain on ln_remote_nets_hash */
821 struct list_head lrn_list;
823 struct list_head lrn_routes;
828 /** lnet message has credit and can be submitted to lnd for send/receive */
829 #define LNET_CREDIT_OK 0
830 /** lnet message is waiting for credit */
831 #define LNET_CREDIT_WAIT 1
832 /** lnet message is waiting for discovery */
833 #define LNET_DC_WAIT 2
835 struct lnet_rtrbufpool {
836 /* my free buffer pool */
837 struct list_head rbp_bufs;
838 /* messages blocking for a buffer */
839 struct list_head rbp_msgs;
840 /* # pages in each buffer */
842 /* requested number of buffers */
843 int rbp_req_nbuffers;
844 /* # buffers actually allocated */
846 /* # free buffers / blocked messages */
853 struct list_head rb_list; /* chain on rbp_bufs */
854 struct lnet_rtrbufpool *rb_pool; /* owning pool */
855 struct bio_vec rb_kiov[0]; /* the buffer space */
858 #define LNET_PEER_HASHSIZE 503 /* prime! */
860 enum lnet_match_flags {
861 /* Didn't match anything */
862 LNET_MATCHMD_NONE = (1 << 0),
864 LNET_MATCHMD_OK = (1 << 1),
865 /* Must be discarded */
866 LNET_MATCHMD_DROP = (1 << 2),
867 /* match and buffer is exhausted */
868 LNET_MATCHMD_EXHAUSTED = (1 << 3),
870 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
873 /* Options for struct lnet_portal::ptl_options */
874 #define LNET_PTL_LAZY (1 << 0)
875 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
876 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
878 /* parameter for matching operations (GET, PUT) */
879 struct lnet_match_info {
881 struct lnet_process_id mi_id;
884 unsigned int mi_portal;
885 unsigned int mi_rlength;
886 unsigned int mi_roffset;
889 /* ME hash of RDMA portal */
890 #define LNET_MT_HASH_BITS 8
891 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
892 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
893 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
894 * the last entry is reserved for MEs with ignore-bits */
895 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
896 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
897 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
898 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
899 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
900 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
901 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
903 /* portal match table */
904 struct lnet_match_table {
905 /* reserved for upcoming patches, CPU partition ID */
907 unsigned int mt_portal; /* portal index */
908 /* match table is set as "enabled" if there's non-exhausted MD
909 * attached on mt_mhash, it's only valid for wildcard portal */
910 unsigned int mt_enabled;
911 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
912 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
913 struct list_head *mt_mhash; /* matching hash */
916 /* these are only useful for wildcard portal */
917 /* Turn off message rotor for wildcard portals */
918 #define LNET_PTL_ROTOR_OFF 0
919 /* round-robin dispatch all PUT messages for wildcard portals */
920 #define LNET_PTL_ROTOR_ON 1
921 /* round-robin dispatch routed PUT message for wildcard portals */
922 #define LNET_PTL_ROTOR_RR_RT 2
923 /* dispatch routed PUT message by hashing source NID for wildcard portals */
924 #define LNET_PTL_ROTOR_HASH_RT 3
928 unsigned int ptl_index; /* portal ID, reserved */
929 /* flags on this portal: lazy, unique... */
930 unsigned int ptl_options;
931 /* list of messages which are stealing buffer */
932 struct list_head ptl_msg_stealing;
933 /* messages blocking for MD */
934 struct list_head ptl_msg_delayed;
935 /* Match table for each CPT */
936 struct lnet_match_table **ptl_mtables;
937 /* spread rotor of incoming "PUT" */
938 unsigned int ptl_rotor;
939 /* # active entries for this portal */
941 /* array of active entries' cpu-partition-id */
945 #define LNET_LH_HASH_BITS 12
946 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
947 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
949 /* resource container (ME, MD, EQ) */
950 struct lnet_res_container {
951 unsigned int rec_type; /* container type */
952 __u64 rec_lh_cookie; /* cookie generator */
953 struct list_head rec_active; /* active resource list */
954 struct list_head *rec_lh_hash; /* handle hash */
957 /* message container */
958 struct lnet_msg_container {
959 int msc_init; /* initialized or not */
960 /* max # threads finalizing */
962 /* msgs waiting to complete finalizing */
963 struct list_head msc_finalizing;
964 /* msgs waiting to be resent */
965 struct list_head msc_resending;
966 struct list_head msc_active; /* active message list */
967 /* threads doing finalization */
968 void **msc_finalizers;
969 /* threads doing resends */
970 void **msc_resenders;
973 /* Peer Discovery states */
974 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
975 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
976 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
978 /* Router Checker states */
979 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
980 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
981 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
984 #define LNET_STATE_SHUTDOWN 0 /* not started */
985 #define LNET_STATE_RUNNING 1 /* started up OK */
986 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
989 /* CPU partition table of LNet */
990 struct cfs_cpt_table *ln_cpt_table;
991 /* number of CPTs in ln_cpt_table */
992 unsigned int ln_cpt_number;
993 unsigned int ln_cpt_bits;
995 /* protect LNet resources (ME/MD/EQ) */
996 struct cfs_percpt_lock *ln_res_lock;
999 /* the vector of portals */
1000 struct lnet_portal **ln_portals;
1001 /* percpt MD container */
1002 struct lnet_res_container **ln_md_containers;
1004 /* Event Queue container */
1005 struct lnet_res_container ln_eq_container;
1006 spinlock_t ln_eq_wait_lock;
1008 unsigned int ln_remote_nets_hbits;
1010 /* protect NI, peer table, credits, routers, rtrbuf... */
1011 struct cfs_percpt_lock *ln_net_lock;
1012 /* percpt message containers for active/finalizing/freed message */
1013 struct lnet_msg_container **ln_msg_containers;
1014 struct lnet_counters **ln_counters;
1015 struct lnet_peer_table **ln_peer_tables;
1016 /* list of peer nis not on a local network */
1017 struct list_head ln_remote_peer_ni_list;
1018 /* failure simulation */
1019 struct list_head ln_test_peers;
1020 struct list_head ln_drop_rules;
1021 struct list_head ln_delay_rules;
1023 struct list_head ln_nets;
1024 /* the loopback NI */
1025 struct lnet_ni *ln_loni;
1026 /* network zombie list */
1027 struct list_head ln_net_zombie;
1028 /* resend messages list */
1029 struct list_head ln_msg_resend;
1030 /* spin lock to protect the msg resend list */
1031 spinlock_t ln_msg_resend_lock;
1033 /* remote networks with routes to them */
1034 struct list_head *ln_remote_nets_hash;
1035 /* validity stamp */
1036 __u64 ln_remote_nets_version;
1037 /* list of all known routers */
1038 struct list_head ln_routers;
1039 /* validity stamp */
1040 __u64 ln_routers_version;
1041 /* percpt router buffer pools */
1042 struct lnet_rtrbufpool **ln_rtrpools;
1045 * Ping target / Push source
1047 * The ping target and push source share a single buffer. The
1048 * ln_ping_target is protected against concurrent updates by
1051 struct lnet_handle_md ln_ping_target_md;
1052 lnet_handler_t ln_ping_target_handler;
1053 struct lnet_ping_buffer *ln_ping_target;
1054 atomic_t ln_ping_target_seqno;
1059 * ln_push_nnis contains the desired size of the push target.
1060 * The lnet_net_lock is used to handle update races. The old
1061 * buffer may linger a while after it has been unlinked, in
1062 * which case the event handler cleans up.
1064 lnet_handler_t ln_push_target_handler;
1065 struct lnet_handle_md ln_push_target_md;
1066 struct lnet_ping_buffer *ln_push_target;
1067 int ln_push_target_nnis;
1069 /* discovery event queue handle */
1070 lnet_handler_t ln_dc_handler;
1071 /* discovery requests */
1072 struct list_head ln_dc_request;
1073 /* discovery working list */
1074 struct list_head ln_dc_working;
1075 /* discovery expired list */
1076 struct list_head ln_dc_expired;
1077 /* discovery thread wait queue */
1078 wait_queue_head_t ln_dc_waitq;
1079 /* discovery startup/shutdown state */
1082 /* monitor thread startup/shutdown state */
1084 /* serialise startup/shutdown */
1085 struct semaphore ln_mt_signal;
1087 struct mutex ln_api_mutex;
1088 struct mutex ln_lnd_mutex;
1089 /* Have I called LNetNIInit myself? */
1091 /* LNetNIInit/LNetNIFini counter */
1093 /* SHUTDOWN/RUNNING/STOPPING */
1096 int ln_routing; /* am I a router? */
1097 lnet_pid_t ln_pid; /* requested pid */
1098 /* uniquely identifies this ni in this epoch */
1099 __u64 ln_interface_cookie;
1100 /* registered LNDs */
1101 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1103 /* test protocol compatibility flags */
1104 unsigned long ln_testprotocompat;
1106 /* 0 - load the NIs from the mod params
1107 * 1 - do not load the NIs from the mod params
1108 * Reverse logic to ensure that other calls to LNetNIInit
1111 bool ln_nis_from_mod_params;
1114 * completion for the monitor thread. The monitor thread takes care of
1115 * checking routes, timedout messages and resending messages.
1117 struct completion ln_mt_wait_complete;
1119 /* per-cpt resend queues */
1120 struct list_head **ln_mt_resendqs;
1121 /* local NIs to recover */
1122 struct list_head ln_mt_localNIRecovq;
1123 /* local NIs to recover */
1124 struct list_head ln_mt_peerNIRecovq;
1126 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1127 * There are CPT number of queues. Since response trackers will be
1128 * added on the fast path we can't afford to grab the exclusive
1129 * net lock to protect these queues. The CPT will be calculated
1130 * based on the mdh cookie.
1132 struct list_head **ln_mt_rstq;
1134 * A response tracker becomes a zombie when the associated MD is queued
1135 * for unlink before the response tracker is detached from the MD. An
1136 * entry on a zombie list can be freed when either the remaining
1137 * operations on the MD complete or when LNet has shut down.
1139 struct list_head **ln_mt_zombie_rstqs;
1140 /* recovery handler */
1141 lnet_handler_t ln_mt_handler;
1144 * Completed when the discovery and monitor threads can enter their
1147 struct completion ln_started;