4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lnet/include/lnet/lib-types.h
34 * Types used by the library side routines that do not need to be
35 * exposed to the user application
38 #ifndef __LNET_LIB_TYPES_H__
39 #define __LNET_LIB_TYPES_H__
42 # error This include is only for kernel use.
45 #include <linux/kthread.h>
46 #include <linux/uio.h>
47 #include <linux/semaphore.h>
48 #include <linux/types.h>
50 #include <uapi/linux/lnet/lnet-dlc.h>
51 #include <uapi/linux/lnet/lnetctl.h>
53 /* Max payload size */
54 #define LNET_MAX_PAYLOAD LNET_MTU
56 /** limit on the number of fragments in discontiguous MDs */
57 #define LNET_MAX_IOV 256
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 struct list_head eq_list;
187 struct lnet_libhandle eq_lh;
188 unsigned long eq_enq_seq;
189 unsigned long eq_deq_seq;
190 unsigned int eq_size;
191 lnet_eq_handler_t eq_callback;
192 struct lnet_event *eq_events;
193 int **eq_refs; /* percpt refcount for EQ */
197 struct list_head me_list;
198 struct lnet_libhandle me_lh;
199 struct lnet_process_id me_match_id;
200 unsigned int me_portal;
201 unsigned int me_pos; /* hash offset in mt_hash */
203 __u64 me_ignore_bits;
204 enum lnet_unlink me_unlink;
205 struct lnet_libmd *me_md;
209 struct list_head md_list;
210 struct lnet_libhandle md_lh;
211 struct lnet_me *md_me;
213 unsigned int md_offset;
214 unsigned int md_length;
215 unsigned int md_max_size;
218 unsigned int md_options;
219 unsigned int md_flags;
220 unsigned int md_niov; /* # frags at end of struct */
222 struct lnet_rsp_tracker *md_rspt_ptr;
223 struct lnet_eq *md_eq;
224 struct lnet_handle_md md_bulk_handle;
226 struct kvec iov[LNET_MAX_IOV];
227 lnet_kiov_t kiov[LNET_MAX_IOV];
231 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
232 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
233 #define LNET_MD_FLAG_ABORTED (1 << 2)
235 struct lnet_test_peer {
236 /* info about peers we are trying to fail */
237 struct list_head tp_list; /* ln_test_peers */
238 lnet_nid_t tp_nid; /* matching nid */
239 unsigned int tp_threshold; /* # failures to simulate */
242 #define LNET_COOKIE_TYPE_MD 1
243 #define LNET_COOKIE_TYPE_ME 2
244 #define LNET_COOKIE_TYPE_EQ 3
245 #define LNET_COOKIE_TYPE_BITS 2
246 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
248 struct lnet_ni; /* forward ref */
252 /* fields managed by portals */
253 struct list_head lnd_list; /* stash in the LND table */
254 int lnd_refcount; /* # active instances */
256 /* fields initialized by the LND */
259 int (*lnd_startup)(struct lnet_ni *ni);
260 void (*lnd_shutdown)(struct lnet_ni *ni);
261 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
263 /* In data movement APIs below, payload buffers are described as a set
264 * of 'niov' fragments which are...
266 * in virtual memory (struct kvec *iov != NULL)
268 * in pages (kernel only: plt_kiov_t *kiov != NULL).
269 * The LND may NOT overwrite these fragment descriptors.
270 * An 'offset' and may specify a byte offset within the set of
271 * fragments to start from
274 /* Start sending a preformatted message. 'private' is NULL for PUT and
275 * GET messages; otherwise this is a response to an incoming message
276 * and 'private' is the 'private' passed to lnet_parse(). Return
277 * non-zero for immediate failure, otherwise complete later with
279 int (*lnd_send)(struct lnet_ni *ni, void *private,
280 struct lnet_msg *msg);
282 /* Start receiving 'mlen' bytes of payload data, skipping the following
283 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
284 * lnet_parse(). Return non-zero for immedaite failure, otherwise
285 * complete later with lnet_finalize(). This also gives back a receive
286 * credit if the LND does flow control. */
287 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
288 int delayed, unsigned int niov,
289 struct kvec *iov, lnet_kiov_t *kiov,
290 unsigned int offset, unsigned int mlen, unsigned int rlen);
292 /* lnet_parse() has had to delay processing of this message
293 * (e.g. waiting for a forwarding buffer or send credits). Give the
294 * LND a chance to free urgently needed resources. If called, return 0
295 * for success and do NOT give back a receive credit; that has to wait
296 * until lnd_recv() gets called. On failure return < 0 and
297 * release resources; lnd_recv() will not be called. */
298 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
299 struct lnet_msg *msg, void **new_privatep);
301 /* notification of peer health */
302 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
304 /* query of peer aliveness */
305 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, time64_t *when);
307 /* accept a new connection */
308 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
311 struct lnet_tx_queue {
312 int tq_credits; /* # tx credits free */
313 int tq_credits_min; /* lowest it's been */
314 int tq_credits_max; /* total # tx credits */
315 struct list_head tq_delayed; /* delayed TXs */
318 enum lnet_net_state {
319 /* set when net block is allocated */
320 LNET_NET_STATE_INIT = 0,
321 /* set when NIs in net are started successfully */
322 LNET_NET_STATE_ACTIVE,
323 /* set if all NIs in net are in FAILED state */
324 LNET_NET_STATE_INACTIVE,
325 /* set when shutting down a NET */
326 LNET_NET_STATE_DELETING
330 /* initial state when NI is created */
331 LNET_NI_STATE_INIT = 0,
332 /* set when NI is brought up */
333 LNET_NI_STATE_ACTIVE,
334 /* set when NI is being shutdown */
335 LNET_NI_STATE_DELETING,
338 #define LNET_NI_RECOVERY_PENDING BIT(0)
339 #define LNET_NI_RECOVERY_FAILED BIT(1)
341 enum lnet_stats_type {
342 LNET_STATS_TYPE_SEND = 0,
343 LNET_STATS_TYPE_RECV,
347 struct lnet_comm_count {
348 atomic_t co_get_count;
349 atomic_t co_put_count;
350 atomic_t co_reply_count;
351 atomic_t co_ack_count;
352 atomic_t co_hello_count;
355 struct lnet_element_stats {
356 struct lnet_comm_count el_send_stats;
357 struct lnet_comm_count el_recv_stats;
358 struct lnet_comm_count el_drop_stats;
361 struct lnet_health_local_stats {
362 atomic_t hlt_local_interrupt;
363 atomic_t hlt_local_dropped;
364 atomic_t hlt_local_aborted;
365 atomic_t hlt_local_no_route;
366 atomic_t hlt_local_timeout;
367 atomic_t hlt_local_error;
370 struct lnet_health_remote_stats {
371 atomic_t hlt_remote_dropped;
372 atomic_t hlt_remote_timeout;
373 atomic_t hlt_remote_error;
374 atomic_t hlt_network_timeout;
378 /* chain on the ln_nets */
379 struct list_head net_list;
381 /* net ID, which is composed of
382 * (net_type << 16) | net_num.
383 * net_type can be one of the enumerated types defined in
384 * lnet/include/lnet/nidstr.h */
387 /* priority of the network */
390 /* total number of CPTs in the array */
393 /* cumulative CPTs of all NIs in this net */
396 /* network tunables */
397 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
400 * boolean to indicate that the tunables have been set and
403 bool net_tunables_set;
405 /* procedural interface */
406 struct lnet_lnd *net_lnd;
408 /* list of NIs on this net */
409 struct list_head net_ni_list;
411 /* list of NIs being added, but not started yet */
412 struct list_head net_ni_added;
414 /* dying LND instances */
415 struct list_head net_ni_zombie;
418 enum lnet_net_state net_state;
422 /* chain on the lnet_net structure */
423 struct list_head ni_netlist;
425 /* chain on the recovery queue */
426 struct list_head ni_recovery;
428 /* MD handle for recovery ping */
429 struct lnet_handle_md ni_ping_mdh;
436 /* bond NI on some CPTs */
439 /* interface's NID */
442 /* instance-specific data */
446 atomic_t ni_tx_credits;
448 /* percpt TX queues */
449 struct lnet_tx_queue **ni_tx_queues;
451 /* percpt reference count */
454 /* when I was last alive */
455 time64_t ni_last_alive;
457 /* pointer to parent network */
458 struct lnet_net *ni_net;
460 /* my health status */
461 struct lnet_ni_status *ni_status;
463 /* NI FSM. Protected by lnet_ni_lock() */
464 enum lnet_ni_state ni_state;
466 /* Recovery state. Protected by lnet_ni_lock() */
467 __u32 ni_recovery_state;
469 /* per NI LND tunables */
470 struct lnet_lnd_tunables ni_lnd_tunables;
472 /* lnd tunables set explicitly */
473 bool ni_lnd_tunables_set;
476 struct lnet_element_stats ni_stats;
477 struct lnet_health_local_stats ni_hstats;
479 /* physical device CPT */
482 /* sequence number used to round robin over nis within a net */
487 * initialized to LNET_MAX_HEALTH_VALUE
488 * Value is decremented every time we fail to send a message over
489 * this NI because of a NI specific failure.
490 * Value is incremented if we successfully send a message.
495 * Set to 1 by the LND when it receives an event telling it the device
496 * has gone into a fatal state. Set to 0 when the LND receives an
497 * even telling it the device is back online.
499 atomic_t ni_fatal_error_on;
502 * equivalent interfaces to use
503 * This is an array because socklnd bonding can still be configured
505 char *ni_interfaces[LNET_INTERFACES_NUM];
506 struct net *ni_net_ns; /* original net namespace */
509 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
512 * Descriptor of a ping info buffer: keep a separate indicator of the
513 * size and a reference count. The type is used both as a source and
514 * sink of data, so we need to keep some information outside of the
515 * area that may be overwritten by network data.
517 struct lnet_ping_buffer {
520 struct lnet_ping_info pb_info;
523 #define LNET_PING_BUFFER_SIZE(NNIDS) \
524 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
525 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
526 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
528 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
529 container_of((PINFO), struct lnet_ping_buffer, pb_info)
531 /* router checker data, per router */
532 struct lnet_rc_data {
533 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
534 struct list_head rcd_list;
535 struct lnet_handle_md rcd_mdh; /* ping buffer MD */
536 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
537 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
538 int rcd_nnis; /* desired size of buffer */
541 struct lnet_peer_ni {
542 /* chain on lpn_peer_nis */
543 struct list_head lpni_peer_nis;
544 /* chain on remote peer list */
545 struct list_head lpni_on_remote_peer_ni_list;
546 /* chain on recovery queue */
547 struct list_head lpni_recovery;
548 /* chain on peer hash */
549 struct list_head lpni_hashlist;
550 /* messages blocking for tx credits */
551 struct list_head lpni_txq;
552 /* messages blocking for router credits */
553 struct list_head lpni_rtrq;
554 /* chain on router list */
555 struct list_head lpni_rtr_list;
556 /* pointer to peer net I'm part of */
557 struct lnet_peer_net *lpni_peer_net;
558 /* statistics kept on each peer NI */
559 struct lnet_element_stats lpni_stats;
560 struct lnet_health_remote_stats lpni_hstats;
561 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
562 spinlock_t lpni_lock;
563 /* # tx credits available */
566 int lpni_mintxcredits;
567 /* # router credits */
570 int lpni_minrtrcredits;
571 /* bytes queued for sending */
575 /* notification outstanding? */
577 /* outstanding notification for LND? */
579 /* some thread is handling notification */
581 /* SEND event outstanding from ping */
582 bool lpni_ping_notsent;
583 /* # times router went dead<->alive. Protected with lpni_lock */
584 int lpni_alive_count;
585 /* time of last aliveness news */
586 time64_t lpni_timestamp;
587 /* time of last ping attempt */
588 time64_t lpni_ping_timestamp;
589 /* != 0 if ping reply expected */
590 time64_t lpni_ping_deadline;
591 /* when I was last alive */
592 time64_t lpni_last_alive;
593 /* when lpni_ni was queried last time */
594 time64_t lpni_last_query;
595 /* network peer is on */
596 struct lnet_net *lpni_net;
600 atomic_t lpni_refcount;
601 /* health value for the peer */
602 atomic_t lpni_healthv;
603 /* recovery ping mdh */
604 struct lnet_handle_md lpni_recovery_ping_mdh;
605 /* CPT this peer attached on */
607 /* state flags -- protected by lpni_lock */
609 /* # refs from lnet_route_t::lr_gateway */
610 int lpni_rtr_refcount;
611 /* sequence number used to round robin over peer nis within a net */
613 /* sequence number used to round robin over gateways */
617 /* returned RC ping features. Protected with lpni_lock */
618 unsigned int lpni_ping_feats;
619 /* routes on this peer */
620 struct list_head lpni_routes;
621 /* preferred local nids: if only one, use lpni_pref.nid */
626 /* number of preferred NIDs in lnpi_pref_nids */
627 __u32 lpni_pref_nnids;
628 /* router checker state */
629 struct lnet_rc_data *lpni_rcd;
632 /* Preferred path added due to traffic on non-MR peer_ni */
633 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
634 /* peer is being recovered. */
635 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
636 /* recovery ping failed */
637 #define LNET_PEER_NI_RECOVERY_FAILED (1 << 2)
638 /* peer is being deleted */
639 #define LNET_PEER_NI_DELETING (1 << 3)
642 /* chain on pt_peer_list */
643 struct list_head lp_peer_list;
645 /* list of peer nets */
646 struct list_head lp_peer_nets;
648 /* list of messages pending discovery*/
649 struct list_head lp_dc_pendq;
651 /* primary NID of the peer */
652 lnet_nid_t lp_primary_nid;
654 /* source NID to use during discovery */
655 lnet_nid_t lp_disc_src_nid;
657 /* CPT of peer_table */
660 /* number of NIDs on this peer */
663 /* reference count */
664 atomic_t lp_refcount;
666 /* lock protecting peer state flags */
669 /* peer state flags */
672 /* buffer for data pushed by peer */
673 struct lnet_ping_buffer *lp_data;
675 /* MD handle for ping in progress */
676 struct lnet_handle_md lp_ping_mdh;
678 /* MD handle for push in progress */
679 struct lnet_handle_md lp_push_mdh;
681 /* number of NIDs for sizing push data */
684 /* NI config sequence number of peer */
687 /* Local NI config sequence number acked by peer */
690 /* Local NI config sequence number sent to peer */
691 __u32 lp_node_seqno_sent;
693 /* Ping error encountered during discovery. */
696 /* Push error encountered during discovery. */
699 /* Error encountered during discovery. */
702 /* time it was put on the ln_dc_working queue */
703 time64_t lp_last_queued;
705 /* link on discovery-related lists */
706 struct list_head lp_dc_list;
708 /* tasks waiting on discovery of this peer */
709 wait_queue_head_t lp_dc_waitq;
713 * The status flags in lp_state. Their semantics have chosen so that
714 * lp_state can be zero-initialized.
716 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
717 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
719 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
720 * NOT set when the peer was pinged by discovery.
722 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
723 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
725 * A peer is marked CONFIGURED if it was configured by DLC.
727 * In addition, a peer is marked DISCOVERED if it has fully passed
728 * through Peer Discovery.
730 * When Peer Discovery is disabled, the discovery thread will mark
731 * peers REDISCOVER to indicate that they should be re-examined if
732 * discovery is (re)enabled on the node.
734 * A peer that was created as the result of inbound traffic will not
737 #define LNET_PEER_CONFIGURED (1 << 2) /* Configured via DLC */
738 #define LNET_PEER_DISCOVERED (1 << 3) /* Peer was discovered */
739 #define LNET_PEER_REDISCOVER (1 << 4) /* Discovery was disabled */
741 * A peer is marked DISCOVERING when discovery is in progress.
742 * The other flags below correspond to stages of discovery.
744 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
745 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
746 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
747 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
748 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
749 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
750 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
752 * A ping can be forced as a way to fix up state, or as a manual
753 * intervention by an admin.
754 * A push can be forced in circumstances that would normally not
755 * allow for one to happen.
757 #define LNET_PEER_FORCE_PING (1 << 12) /* Forced Ping */
758 #define LNET_PEER_FORCE_PUSH (1 << 13) /* Forced Push */
760 struct lnet_peer_net {
761 /* chain on lp_peer_nets */
762 struct list_head lpn_peer_nets;
764 /* list of peer_nis on this network */
765 struct list_head lpn_peer_nis;
767 /* pointer to the peer I'm part of */
768 struct lnet_peer *lpn_peer;
773 /* reference count */
774 atomic_t lpn_refcount;
778 #define LNET_PEER_HASH_BITS 9
779 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
782 * peer hash table - one per CPT
784 * protected by lnet_net_lock/EX for update
790 * protected by pt_zombie_lock:
794 * pt_zombie lock nests inside lnet_net_lock
796 struct lnet_peer_table {
797 int pt_version; /* /proc validity stamp */
798 int pt_number; /* # peers_ni extant */
799 struct list_head *pt_hash; /* NID->peer hash */
800 struct list_head pt_peer_list; /* peers */
801 int pt_peers; /* # peers */
802 struct list_head pt_zombie_list; /* zombie peer_ni */
803 int pt_zombies; /* # zombie peers_ni */
804 spinlock_t pt_zombie_lock; /* protect list and count */
807 /* peer aliveness is enabled only on routers for peers in a network where the
808 * struct lnet_ni::ni_peertimeout has been set to a positive value
810 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
811 ((lp)->lpni_net) && \
812 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
815 struct list_head lr_list; /* chain on net */
816 struct list_head lr_gwlist; /* chain on gateway */
817 struct lnet_peer_ni *lr_gateway; /* router node */
818 __u32 lr_net; /* remote network number */
819 int lr_seq; /* sequence for round-robin */
820 unsigned int lr_downis; /* number of down NIs */
821 __u32 lr_hops; /* how far I am */
822 unsigned int lr_priority; /* route priority */
825 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
826 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
827 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
829 struct lnet_remotenet {
830 /* chain on ln_remote_nets_hash */
831 struct list_head lrn_list;
833 struct list_head lrn_routes;
838 /** lnet message has credit and can be submitted to lnd for send/receive */
839 #define LNET_CREDIT_OK 0
840 /** lnet message is waiting for credit */
841 #define LNET_CREDIT_WAIT 1
842 /** lnet message is waiting for discovery */
843 #define LNET_DC_WAIT 2
845 struct lnet_rtrbufpool {
846 /* my free buffer pool */
847 struct list_head rbp_bufs;
848 /* messages blocking for a buffer */
849 struct list_head rbp_msgs;
850 /* # pages in each buffer */
852 /* requested number of buffers */
853 int rbp_req_nbuffers;
854 /* # buffers actually allocated */
856 /* # free buffers / blocked messages */
863 struct list_head rb_list; /* chain on rbp_bufs */
864 struct lnet_rtrbufpool *rb_pool; /* owning pool */
865 lnet_kiov_t rb_kiov[0]; /* the buffer space */
868 #define LNET_PEER_HASHSIZE 503 /* prime! */
870 enum lnet_match_flags {
871 /* Didn't match anything */
872 LNET_MATCHMD_NONE = (1 << 0),
874 LNET_MATCHMD_OK = (1 << 1),
875 /* Must be discarded */
876 LNET_MATCHMD_DROP = (1 << 2),
877 /* match and buffer is exhausted */
878 LNET_MATCHMD_EXHAUSTED = (1 << 3),
880 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
883 /* Options for struct lnet_portal::ptl_options */
884 #define LNET_PTL_LAZY (1 << 0)
885 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
886 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
888 /* parameter for matching operations (GET, PUT) */
889 struct lnet_match_info {
891 struct lnet_process_id mi_id;
894 unsigned int mi_portal;
895 unsigned int mi_rlength;
896 unsigned int mi_roffset;
899 /* ME hash of RDMA portal */
900 #define LNET_MT_HASH_BITS 8
901 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
902 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
903 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
904 * the last entry is reserved for MEs with ignore-bits */
905 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
906 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
907 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
908 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
909 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
910 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
911 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
913 /* portal match table */
914 struct lnet_match_table {
915 /* reserved for upcoming patches, CPU partition ID */
917 unsigned int mt_portal; /* portal index */
918 /* match table is set as "enabled" if there's non-exhausted MD
919 * attached on mt_mhash, it's only valid for wildcard portal */
920 unsigned int mt_enabled;
921 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
922 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
923 struct list_head *mt_mhash; /* matching hash */
926 /* these are only useful for wildcard portal */
927 /* Turn off message rotor for wildcard portals */
928 #define LNET_PTL_ROTOR_OFF 0
929 /* round-robin dispatch all PUT messages for wildcard portals */
930 #define LNET_PTL_ROTOR_ON 1
931 /* round-robin dispatch routed PUT message for wildcard portals */
932 #define LNET_PTL_ROTOR_RR_RT 2
933 /* dispatch routed PUT message by hashing source NID for wildcard portals */
934 #define LNET_PTL_ROTOR_HASH_RT 3
938 unsigned int ptl_index; /* portal ID, reserved */
939 /* flags on this portal: lazy, unique... */
940 unsigned int ptl_options;
941 /* list of messages which are stealing buffer */
942 struct list_head ptl_msg_stealing;
943 /* messages blocking for MD */
944 struct list_head ptl_msg_delayed;
945 /* Match table for each CPT */
946 struct lnet_match_table **ptl_mtables;
947 /* spread rotor of incoming "PUT" */
948 unsigned int ptl_rotor;
949 /* # active entries for this portal */
951 /* array of active entries' cpu-partition-id */
955 #define LNET_LH_HASH_BITS 12
956 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
957 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
959 /* resource container (ME, MD, EQ) */
960 struct lnet_res_container {
961 unsigned int rec_type; /* container type */
962 __u64 rec_lh_cookie; /* cookie generator */
963 struct list_head rec_active; /* active resource list */
964 struct list_head *rec_lh_hash; /* handle hash */
967 /* message container */
968 struct lnet_msg_container {
969 int msc_init; /* initialized or not */
970 /* max # threads finalizing */
972 /* msgs waiting to complete finalizing */
973 struct list_head msc_finalizing;
974 /* msgs waiting to be resent */
975 struct list_head msc_resending;
976 struct list_head msc_active; /* active message list */
977 /* threads doing finalization */
978 void **msc_finalizers;
979 /* threads doing resends */
980 void **msc_resenders;
983 /* Peer Discovery states */
984 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
985 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
986 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
988 /* Router Checker states */
989 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
990 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
991 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
994 #define LNET_STATE_SHUTDOWN 0 /* not started */
995 #define LNET_STATE_RUNNING 1 /* started up OK */
996 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
999 /* CPU partition table of LNet */
1000 struct cfs_cpt_table *ln_cpt_table;
1001 /* number of CPTs in ln_cpt_table */
1002 unsigned int ln_cpt_number;
1003 unsigned int ln_cpt_bits;
1005 /* protect LNet resources (ME/MD/EQ) */
1006 struct cfs_percpt_lock *ln_res_lock;
1009 /* the vector of portals */
1010 struct lnet_portal **ln_portals;
1011 /* percpt ME containers */
1012 struct lnet_res_container **ln_me_containers;
1013 /* percpt MD container */
1014 struct lnet_res_container **ln_md_containers;
1016 /* Event Queue container */
1017 struct lnet_res_container ln_eq_container;
1018 wait_queue_head_t ln_eq_waitq;
1019 spinlock_t ln_eq_wait_lock;
1021 unsigned int ln_remote_nets_hbits;
1023 /* protect NI, peer table, credits, routers, rtrbuf... */
1024 struct cfs_percpt_lock *ln_net_lock;
1025 /* percpt message containers for active/finalizing/freed message */
1026 struct lnet_msg_container **ln_msg_containers;
1027 struct lnet_counters **ln_counters;
1028 struct lnet_peer_table **ln_peer_tables;
1029 /* list of peer nis not on a local network */
1030 struct list_head ln_remote_peer_ni_list;
1031 /* failure simulation */
1032 struct list_head ln_test_peers;
1033 struct list_head ln_drop_rules;
1034 struct list_head ln_delay_rules;
1036 struct list_head ln_nets;
1037 /* the loopback NI */
1038 struct lnet_ni *ln_loni;
1039 /* network zombie list */
1040 struct list_head ln_net_zombie;
1041 /* resend messages list */
1042 struct list_head ln_msg_resend;
1043 /* spin lock to protect the msg resend list */
1044 spinlock_t ln_msg_resend_lock;
1046 /* remote networks with routes to them */
1047 struct list_head *ln_remote_nets_hash;
1048 /* validity stamp */
1049 __u64 ln_remote_nets_version;
1050 /* list of all known routers */
1051 struct list_head ln_routers;
1052 /* validity stamp */
1053 __u64 ln_routers_version;
1054 /* percpt router buffer pools */
1055 struct lnet_rtrbufpool **ln_rtrpools;
1058 * Ping target / Push source
1060 * The ping target and push source share a single buffer. The
1061 * ln_ping_target is protected against concurrent updates by
1064 struct lnet_handle_md ln_ping_target_md;
1065 struct lnet_handle_eq ln_ping_target_eq;
1066 struct lnet_ping_buffer *ln_ping_target;
1067 atomic_t ln_ping_target_seqno;
1072 * ln_push_nnis contains the desired size of the push target.
1073 * The lnet_net_lock is used to handle update races. The old
1074 * buffer may linger a while after it has been unlinked, in
1075 * which case the event handler cleans up.
1077 struct lnet_handle_eq ln_push_target_eq;
1078 struct lnet_handle_md ln_push_target_md;
1079 struct lnet_ping_buffer *ln_push_target;
1080 int ln_push_target_nnis;
1082 /* discovery event queue handle */
1083 struct lnet_handle_eq ln_dc_eqh;
1084 /* discovery requests */
1085 struct list_head ln_dc_request;
1086 /* discovery working list */
1087 struct list_head ln_dc_working;
1088 /* discovery expired list */
1089 struct list_head ln_dc_expired;
1090 /* discovery thread wait queue */
1091 wait_queue_head_t ln_dc_waitq;
1092 /* discovery startup/shutdown state */
1095 /* monitor thread startup/shutdown state */
1097 /* router checker's event queue */
1098 struct lnet_handle_eq ln_rc_eqh;
1099 /* rcd still pending on net */
1100 struct list_head ln_rcd_deathrow;
1101 /* rcd ready for free */
1102 struct list_head ln_rcd_zombie;
1103 /* serialise startup/shutdown */
1104 struct semaphore ln_mt_signal;
1106 struct mutex ln_api_mutex;
1107 struct mutex ln_lnd_mutex;
1108 /* Have I called LNetNIInit myself? */
1110 /* LNetNIInit/LNetNIFini counter */
1112 /* SHUTDOWN/RUNNING/STOPPING */
1115 int ln_routing; /* am I a router? */
1116 lnet_pid_t ln_pid; /* requested pid */
1117 /* uniquely identifies this ni in this epoch */
1118 __u64 ln_interface_cookie;
1119 /* registered LNDs */
1120 struct list_head ln_lnds;
1122 /* test protocol compatibility flags */
1123 int ln_testprotocompat;
1125 /* 0 - load the NIs from the mod params
1126 * 1 - do not load the NIs from the mod params
1127 * Reverse logic to ensure that other calls to LNetNIInit
1130 bool ln_nis_from_mod_params;
1133 * waitq for the monitor thread. The monitor thread takes care of
1134 * checking routes, timedout messages and resending messages.
1136 wait_queue_head_t ln_mt_waitq;
1138 /* per-cpt resend queues */
1139 struct list_head **ln_mt_resendqs;
1140 /* local NIs to recover */
1141 struct list_head ln_mt_localNIRecovq;
1142 /* local NIs to recover */
1143 struct list_head ln_mt_peerNIRecovq;
1145 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1146 * There are CPT number of queues. Since response trackers will be
1147 * added on the fast path we can't afford to grab the exclusive
1148 * net lock to protect these queues. The CPT will be calculated
1149 * based on the mdh cookie.
1151 struct list_head **ln_mt_rstq;
1153 * A response tracker becomes a zombie when the associated MD is queued
1154 * for unlink before the response tracker is detached from the MD. An
1155 * entry on a zombie list can be freed when either the remaining
1156 * operations on the MD complete or when LNet has shut down.
1158 struct list_head **ln_mt_zombie_rstqs;
1159 /* recovery eq handler */
1160 struct lnet_handle_eq ln_mt_eqh;