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 #define LNET_MAX_IOV (LNET_MAX_PAYLOAD >> PAGE_SHIFT)
59 * This is the maximum health value.
60 * All local and peer NIs created have their health default to this value.
62 #define LNET_MAX_HEALTH_VALUE 1000
67 enum lnet_msg_hstatus {
68 LNET_MSG_STATUS_OK = 0,
69 LNET_MSG_STATUS_LOCAL_INTERRUPT,
70 LNET_MSG_STATUS_LOCAL_DROPPED,
71 LNET_MSG_STATUS_LOCAL_ABORTED,
72 LNET_MSG_STATUS_LOCAL_NO_ROUTE,
73 LNET_MSG_STATUS_LOCAL_ERROR,
74 LNET_MSG_STATUS_LOCAL_TIMEOUT,
75 LNET_MSG_STATUS_REMOTE_ERROR,
76 LNET_MSG_STATUS_REMOTE_DROPPED,
77 LNET_MSG_STATUS_REMOTE_TIMEOUT,
78 LNET_MSG_STATUS_NETWORK_TIMEOUT,
82 struct lnet_rsp_tracker {
83 /* chain on the waiting list */
84 struct list_head rspt_on_list;
88 lnet_nid_t rspt_next_hop_nid;
89 /* deadline of the REPLY/ACK */
90 ktime_t rspt_deadline;
92 struct lnet_handle_md rspt_mdh;
96 struct list_head msg_activelist;
97 struct list_head msg_list; /* Q for credits/MD */
99 struct lnet_process_id msg_target;
100 /* Primary NID of the source. */
101 lnet_nid_t msg_initiator;
102 /* where is it from, it's only for building event */
107 * hold parameters in case message is with held due
110 lnet_nid_t msg_src_nid_param;
111 lnet_nid_t msg_rtr_nid_param;
114 * Deadline for the message after which it will be finalized if it
117 ktime_t msg_deadline;
119 /* The message health status. */
120 enum lnet_msg_hstatus msg_health_status;
121 /* This is a recovery message */
123 /* the number of times a transmission has been retried */
125 /* flag to indicate that we do not want to resend this message */
128 /* committed for sending */
129 unsigned int msg_tx_committed:1;
130 /* CPT # this message committed for sending */
131 unsigned int msg_tx_cpt:15;
132 /* committed for receiving */
133 unsigned int msg_rx_committed:1;
134 /* CPT # this message committed for receiving */
135 unsigned int msg_rx_cpt:15;
136 /* queued for tx credit */
137 unsigned int msg_tx_delayed:1;
138 /* queued for RX buffer */
139 unsigned int msg_rx_delayed:1;
140 /* ready for pending on RX delay list */
141 unsigned int msg_rx_ready_delay:1;
143 unsigned int msg_vmflush:1; /* VM trying to free memory */
144 unsigned int msg_target_is_router:1; /* sending to a router */
145 unsigned int msg_routing:1; /* being forwarded */
146 unsigned int msg_ack:1; /* ack on finalize (PUT) */
147 unsigned int msg_sending:1; /* outgoing message */
148 unsigned int msg_receiving:1; /* being received */
149 unsigned int msg_txcredit:1; /* taken an NI send credit */
150 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
151 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
152 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
153 unsigned int msg_onactivelist:1; /* on the activelist */
154 unsigned int msg_rdma_get:1;
156 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
157 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
160 struct lnet_libmd *msg_md;
161 /* the NI the message was sent or received over */
162 struct lnet_ni *msg_txni;
163 struct lnet_ni *msg_rxni;
165 unsigned int msg_len;
166 unsigned int msg_wanted;
167 unsigned int msg_offset;
168 unsigned int msg_niov;
169 struct kvec *msg_iov;
170 lnet_kiov_t *msg_kiov;
172 struct lnet_event msg_ev;
173 struct lnet_hdr msg_hdr;
176 struct lnet_libhandle {
177 struct list_head lh_hash_chain;
181 #define lh_entry(ptr, type, member) \
182 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
185 struct list_head eq_list;
186 struct lnet_libhandle eq_lh;
187 unsigned long eq_enq_seq;
188 unsigned long eq_deq_seq;
189 unsigned int eq_size;
190 lnet_eq_handler_t eq_callback;
191 struct lnet_event *eq_events;
192 int **eq_refs; /* percpt refcount for EQ */
196 struct list_head me_list;
197 struct lnet_libhandle me_lh;
198 struct lnet_process_id me_match_id;
199 unsigned int me_portal;
200 unsigned int me_pos; /* hash offset in mt_hash */
202 __u64 me_ignore_bits;
203 enum lnet_unlink me_unlink;
204 struct lnet_libmd *me_md;
208 struct list_head md_list;
209 struct lnet_libhandle md_lh;
210 struct lnet_me *md_me;
212 unsigned int md_offset;
213 unsigned int md_length;
214 unsigned int md_max_size;
217 unsigned int md_options;
218 unsigned int md_flags;
219 unsigned int md_niov; /* # frags at end of struct */
221 struct lnet_rsp_tracker *md_rspt_ptr;
222 struct lnet_eq *md_eq;
223 struct lnet_handle_md md_bulk_handle;
225 struct kvec iov[LNET_MAX_IOV];
226 lnet_kiov_t kiov[LNET_MAX_IOV];
230 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
231 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
232 #define LNET_MD_FLAG_ABORTED (1 << 2)
234 struct lnet_test_peer {
235 /* info about peers we are trying to fail */
236 struct list_head tp_list; /* ln_test_peers */
237 lnet_nid_t tp_nid; /* matching nid */
238 unsigned int tp_threshold; /* # failures to simulate */
241 #define LNET_COOKIE_TYPE_MD 1
242 #define LNET_COOKIE_TYPE_ME 2
243 #define LNET_COOKIE_TYPE_EQ 3
244 #define LNET_COOKIE_TYPE_BITS 2
245 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
247 struct lnet_ni; /* forward ref */
251 /* fields managed by portals */
252 struct list_head lnd_list; /* stash in the LND table */
253 int lnd_refcount; /* # active instances */
255 /* fields initialized by the LND */
258 int (*lnd_startup)(struct lnet_ni *ni);
259 void (*lnd_shutdown)(struct lnet_ni *ni);
260 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
262 /* In data movement APIs below, payload buffers are described as a set
263 * of 'niov' fragments which are...
265 * in virtual memory (struct kvec *iov != NULL)
267 * in pages (kernel only: plt_kiov_t *kiov != NULL).
268 * The LND may NOT overwrite these fragment descriptors.
269 * An 'offset' and may specify a byte offset within the set of
270 * fragments to start from
273 /* Start sending a preformatted message. 'private' is NULL for PUT and
274 * GET messages; otherwise this is a response to an incoming message
275 * and 'private' is the 'private' passed to lnet_parse(). Return
276 * non-zero for immediate failure, otherwise complete later with
278 int (*lnd_send)(struct lnet_ni *ni, void *private,
279 struct lnet_msg *msg);
281 /* Start receiving 'mlen' bytes of payload data, skipping the following
282 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
283 * lnet_parse(). Return non-zero for immedaite failure, otherwise
284 * complete later with lnet_finalize(). This also gives back a receive
285 * credit if the LND does flow control. */
286 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
287 int delayed, unsigned int niov,
288 struct kvec *iov, lnet_kiov_t *kiov,
289 unsigned int offset, unsigned int mlen, unsigned int rlen);
291 /* lnet_parse() has had to delay processing of this message
292 * (e.g. waiting for a forwarding buffer or send credits). Give the
293 * LND a chance to free urgently needed resources. If called, return 0
294 * for success and do NOT give back a receive credit; that has to wait
295 * until lnd_recv() gets called. On failure return < 0 and
296 * release resources; lnd_recv() will not be called. */
297 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
298 struct lnet_msg *msg, void **new_privatep);
300 /* notification of peer down */
301 void (*lnd_notify_peer_down)(lnet_nid_t peer);
303 /* query of peer aliveness */
304 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, time64_t *when);
306 /* accept a new connection */
307 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
310 struct lnet_tx_queue {
311 int tq_credits; /* # tx credits free */
312 int tq_credits_min; /* lowest it's been */
313 int tq_credits_max; /* total # tx credits */
314 struct list_head tq_delayed; /* delayed TXs */
317 enum lnet_net_state {
318 /* set when net block is allocated */
319 LNET_NET_STATE_INIT = 0,
320 /* set when NIs in net are started successfully */
321 LNET_NET_STATE_ACTIVE,
322 /* set if all NIs in net are in FAILED state */
323 LNET_NET_STATE_INACTIVE,
324 /* set when shutting down a NET */
325 LNET_NET_STATE_DELETING
329 /* initial state when NI is created */
330 LNET_NI_STATE_INIT = 0,
331 /* set when NI is brought up */
332 LNET_NI_STATE_ACTIVE,
333 /* set when NI is being shutdown */
334 LNET_NI_STATE_DELETING,
337 #define LNET_NI_RECOVERY_PENDING BIT(0)
338 #define LNET_NI_RECOVERY_FAILED BIT(1)
340 enum lnet_stats_type {
341 LNET_STATS_TYPE_SEND = 0,
342 LNET_STATS_TYPE_RECV,
346 struct lnet_comm_count {
347 atomic_t co_get_count;
348 atomic_t co_put_count;
349 atomic_t co_reply_count;
350 atomic_t co_ack_count;
351 atomic_t co_hello_count;
354 struct lnet_element_stats {
355 struct lnet_comm_count el_send_stats;
356 struct lnet_comm_count el_recv_stats;
357 struct lnet_comm_count el_drop_stats;
360 struct lnet_health_local_stats {
361 atomic_t hlt_local_interrupt;
362 atomic_t hlt_local_dropped;
363 atomic_t hlt_local_aborted;
364 atomic_t hlt_local_no_route;
365 atomic_t hlt_local_timeout;
366 atomic_t hlt_local_error;
369 struct lnet_health_remote_stats {
370 atomic_t hlt_remote_dropped;
371 atomic_t hlt_remote_timeout;
372 atomic_t hlt_remote_error;
373 atomic_t hlt_network_timeout;
377 /* chain on the ln_nets */
378 struct list_head net_list;
380 /* net ID, which is composed of
381 * (net_type << 16) | net_num.
382 * net_type can be one of the enumerated types defined in
383 * lnet/include/lnet/nidstr.h */
386 /* priority of the network */
389 /* total number of CPTs in the array */
392 /* cumulative CPTs of all NIs in this net */
395 /* network tunables */
396 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
399 * boolean to indicate that the tunables have been set and
402 bool net_tunables_set;
404 /* procedural interface */
405 struct lnet_lnd *net_lnd;
407 /* list of NIs on this net */
408 struct list_head net_ni_list;
410 /* list of NIs being added, but not started yet */
411 struct list_head net_ni_added;
413 /* dying LND instances */
414 struct list_head net_ni_zombie;
417 enum lnet_net_state net_state;
419 /* when I was last alive */
420 time64_t net_last_alive;
422 /* protects access to net_last_alive */
427 /* chain on the lnet_net structure */
428 struct list_head ni_netlist;
430 /* chain on the recovery queue */
431 struct list_head ni_recovery;
433 /* MD handle for recovery ping */
434 struct lnet_handle_md ni_ping_mdh;
441 /* bond NI on some CPTs */
444 /* interface's NID */
447 /* instance-specific data */
451 atomic_t ni_tx_credits;
453 /* percpt TX queues */
454 struct lnet_tx_queue **ni_tx_queues;
456 /* percpt reference count */
459 /* pointer to parent network */
460 struct lnet_net *ni_net;
462 /* my health status */
463 struct lnet_ni_status *ni_status;
465 /* NI FSM. Protected by lnet_ni_lock() */
466 enum lnet_ni_state ni_state;
468 /* Recovery state. Protected by lnet_ni_lock() */
469 __u32 ni_recovery_state;
471 /* per NI LND tunables */
472 struct lnet_lnd_tunables ni_lnd_tunables;
474 /* lnd tunables set explicitly */
475 bool ni_lnd_tunables_set;
478 struct lnet_element_stats ni_stats;
479 struct lnet_health_local_stats ni_hstats;
481 /* physical device CPT */
484 /* sequence number used to round robin over nis within a net */
489 * initialized to LNET_MAX_HEALTH_VALUE
490 * Value is decremented every time we fail to send a message over
491 * this NI because of a NI specific failure.
492 * Value is incremented if we successfully send a message.
497 * Set to 1 by the LND when it receives an event telling it the device
498 * has gone into a fatal state. Set to 0 when the LND receives an
499 * even telling it the device is back online.
501 atomic_t ni_fatal_error_on;
504 * equivalent interfaces to use
505 * This is an array because socklnd bonding can still be configured
507 char *ni_interfaces[LNET_INTERFACES_NUM];
508 struct net *ni_net_ns; /* original net namespace */
511 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
514 * Descriptor of a ping info buffer: keep a separate indicator of the
515 * size and a reference count. The type is used both as a source and
516 * sink of data, so we need to keep some information outside of the
517 * area that may be overwritten by network data.
519 struct lnet_ping_buffer {
522 struct lnet_ping_info pb_info;
525 #define LNET_PING_BUFFER_SIZE(NNIDS) \
526 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
527 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
528 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
530 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
531 container_of((PINFO), struct lnet_ping_buffer, pb_info)
533 struct lnet_peer_ni {
534 /* chain on lpn_peer_nis */
535 struct list_head lpni_peer_nis;
536 /* chain on remote peer list */
537 struct list_head lpni_on_remote_peer_ni_list;
538 /* chain on recovery queue */
539 struct list_head lpni_recovery;
540 /* chain on peer hash */
541 struct list_head lpni_hashlist;
542 /* messages blocking for tx credits */
543 struct list_head lpni_txq;
544 /* pointer to peer net I'm part of */
545 struct lnet_peer_net *lpni_peer_net;
546 /* statistics kept on each peer NI */
547 struct lnet_element_stats lpni_stats;
548 struct lnet_health_remote_stats lpni_hstats;
549 /* spin lock protecting credits and lpni_txq */
550 spinlock_t lpni_lock;
551 /* # tx credits available */
554 int lpni_mintxcredits;
556 * Each peer_ni in a gateway maintains its own credits. This
557 * allows more traffic to gateways that have multiple interfaces.
559 /* # router credits */
562 int lpni_minrtrcredits;
563 /* bytes queued for sending */
565 /* network peer is on */
566 struct lnet_net *lpni_net;
570 atomic_t lpni_refcount;
571 /* health value for the peer */
572 atomic_t lpni_healthv;
573 /* recovery ping mdh */
574 struct lnet_handle_md lpni_recovery_ping_mdh;
575 /* CPT this peer attached on */
577 /* state flags -- protected by lpni_lock */
579 /* status of the peer NI as reported by the peer */
580 __u32 lpni_ns_status;
581 /* sequence number used to round robin over peer nis within a net */
583 /* sequence number used to round robin over gateways */
585 /* returned RC ping features. Protected with lpni_lock */
586 unsigned int lpni_ping_feats;
587 /* time last message was received from the peer */
588 time64_t lpni_last_alive;
589 /* preferred local nids: if only one, use lpni_pref.nid */
594 /* number of preferred NIDs in lnpi_pref_nids */
595 __u32 lpni_pref_nnids;
598 /* Preferred path added due to traffic on non-MR peer_ni */
599 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
600 /* peer is being recovered. */
601 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
602 /* recovery ping failed */
603 #define LNET_PEER_NI_RECOVERY_FAILED (1 << 2)
604 /* peer is being deleted */
605 #define LNET_PEER_NI_DELETING (1 << 3)
608 /* chain on pt_peer_list */
609 struct list_head lp_peer_list;
611 /* list of peer nets */
612 struct list_head lp_peer_nets;
614 /* list of messages pending discovery*/
615 struct list_head lp_dc_pendq;
617 /* chain on router list */
618 struct list_head lp_rtr_list;
620 /* primary NID of the peer */
621 lnet_nid_t lp_primary_nid;
623 /* net to perform discovery on */
624 __u32 lp_disc_net_id;
626 /* CPT of peer_table */
629 /* number of NIDs on this peer */
632 /* # refs from lnet_route_t::lr_gateway */
636 * peer specific health sensitivity value to decrement peer nis in
637 * this peer with if set to something other than 0
639 __u32 lp_health_sensitivity;
641 /* messages blocking for router credits */
642 struct list_head lp_rtrq;
644 /* routes on this peer */
645 struct list_head lp_routes;
647 /* reference count */
648 atomic_t lp_refcount;
650 /* lock protecting peer state flags and lpni_rtrq */
653 /* peer state flags */
656 /* buffer for data pushed by peer */
657 struct lnet_ping_buffer *lp_data;
659 /* MD handle for ping in progress */
660 struct lnet_handle_md lp_ping_mdh;
662 /* MD handle for push in progress */
663 struct lnet_handle_md lp_push_mdh;
665 /* number of NIDs for sizing push data */
668 /* NI config sequence number of peer */
671 /* Local NI config sequence number acked by peer */
674 /* Local NI config sequence number sent to peer */
675 __u32 lp_node_seqno_sent;
677 /* Ping error encountered during discovery. */
680 /* Push error encountered during discovery. */
683 /* Error encountered during discovery. */
686 /* time it was put on the ln_dc_working queue */
687 time64_t lp_last_queued;
689 /* link on discovery-related lists */
690 struct list_head lp_dc_list;
692 /* tasks waiting on discovery of this peer */
693 wait_queue_head_t lp_dc_waitq;
697 * The status flags in lp_state. Their semantics have chosen so that
698 * lp_state can be zero-initialized.
700 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
701 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
703 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
704 * NOT set when the peer was pinged by discovery.
706 * A peer is marked ROUTER if it indicates so in the feature bit.
708 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
709 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
710 #define LNET_PEER_ROUTER_ENABLED (1 << 2) /* router feature enabled */
713 * A peer is marked CONFIGURED if it was configured by DLC.
715 * In addition, a peer is marked DISCOVERED if it has fully passed
716 * through Peer Discovery.
718 * When Peer Discovery is disabled, the discovery thread will mark
719 * peers REDISCOVER to indicate that they should be re-examined if
720 * discovery is (re)enabled on the node.
722 * A peer that was created as the result of inbound traffic will not
725 #define LNET_PEER_CONFIGURED (1 << 3) /* Configured via DLC */
726 #define LNET_PEER_DISCOVERED (1 << 4) /* Peer was discovered */
727 #define LNET_PEER_REDISCOVER (1 << 5) /* Discovery was disabled */
729 * A peer is marked DISCOVERING when discovery is in progress.
730 * The other flags below correspond to stages of discovery.
732 #define LNET_PEER_DISCOVERING (1 << 6) /* Discovering */
733 #define LNET_PEER_DATA_PRESENT (1 << 7) /* Remote peer data present */
734 #define LNET_PEER_NIDS_UPTODATE (1 << 8) /* Remote peer info uptodate */
735 #define LNET_PEER_PING_SENT (1 << 9) /* Waiting for REPLY to Ping */
736 #define LNET_PEER_PUSH_SENT (1 << 10) /* Waiting for ACK of Push */
737 #define LNET_PEER_PING_FAILED (1 << 11) /* Ping send failure */
738 #define LNET_PEER_PUSH_FAILED (1 << 12) /* Push send failure */
740 * A ping can be forced as a way to fix up state, or as a manual
741 * intervention by an admin.
742 * A push can be forced in circumstances that would normally not
743 * allow for one to happen.
745 #define LNET_PEER_FORCE_PING (1 << 13) /* Forced Ping */
746 #define LNET_PEER_FORCE_PUSH (1 << 14) /* Forced Push */
748 /* force delete even if router */
749 #define LNET_PEER_RTR_NI_FORCE_DEL (1 << 15)
751 /* gw undergoing alive discovery */
752 #define LNET_PEER_RTR_DISCOVERY (1 << 16)
754 struct lnet_peer_net {
755 /* chain on lp_peer_nets */
756 struct list_head lpn_peer_nets;
758 /* list of peer_nis on this network */
759 struct list_head lpn_peer_nis;
761 /* pointer to the peer I'm part of */
762 struct lnet_peer *lpn_peer;
767 /* time of last router net check attempt */
768 time64_t lpn_rtrcheck_timestamp;
770 /* selection sequence number */
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 *lr_gateway; /* router node */
818 lnet_nid_t lr_nid; /* NID used to add route */
819 __u32 lr_net; /* remote network number */
820 __u32 lr_lnet; /* local network number */
821 int lr_seq; /* sequence for round-robin */
822 __u32 lr_hops; /* how far I am */
823 unsigned int lr_priority; /* route priority */
826 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
827 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
828 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
830 struct lnet_remotenet {
831 /* chain on ln_remote_nets_hash */
832 struct list_head lrn_list;
834 struct list_head lrn_routes;
839 /** lnet message has credit and can be submitted to lnd for send/receive */
840 #define LNET_CREDIT_OK 0
841 /** lnet message is waiting for credit */
842 #define LNET_CREDIT_WAIT 1
843 /** lnet message is waiting for discovery */
844 #define LNET_DC_WAIT 2
846 struct lnet_rtrbufpool {
847 /* my free buffer pool */
848 struct list_head rbp_bufs;
849 /* messages blocking for a buffer */
850 struct list_head rbp_msgs;
851 /* # pages in each buffer */
853 /* requested number of buffers */
854 int rbp_req_nbuffers;
855 /* # buffers actually allocated */
857 /* # free buffers / blocked messages */
864 struct list_head rb_list; /* chain on rbp_bufs */
865 struct lnet_rtrbufpool *rb_pool; /* owning pool */
866 lnet_kiov_t rb_kiov[0]; /* the buffer space */
869 #define LNET_PEER_HASHSIZE 503 /* prime! */
871 enum lnet_match_flags {
872 /* Didn't match anything */
873 LNET_MATCHMD_NONE = (1 << 0),
875 LNET_MATCHMD_OK = (1 << 1),
876 /* Must be discarded */
877 LNET_MATCHMD_DROP = (1 << 2),
878 /* match and buffer is exhausted */
879 LNET_MATCHMD_EXHAUSTED = (1 << 3),
881 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
884 /* Options for struct lnet_portal::ptl_options */
885 #define LNET_PTL_LAZY (1 << 0)
886 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
887 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
889 /* parameter for matching operations (GET, PUT) */
890 struct lnet_match_info {
892 struct lnet_process_id mi_id;
895 unsigned int mi_portal;
896 unsigned int mi_rlength;
897 unsigned int mi_roffset;
900 /* ME hash of RDMA portal */
901 #define LNET_MT_HASH_BITS 8
902 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
903 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
904 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
905 * the last entry is reserved for MEs with ignore-bits */
906 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
907 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
908 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
909 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
910 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
911 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
912 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
914 /* portal match table */
915 struct lnet_match_table {
916 /* reserved for upcoming patches, CPU partition ID */
918 unsigned int mt_portal; /* portal index */
919 /* match table is set as "enabled" if there's non-exhausted MD
920 * attached on mt_mhash, it's only valid for wildcard portal */
921 unsigned int mt_enabled;
922 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
923 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
924 struct list_head *mt_mhash; /* matching hash */
927 /* these are only useful for wildcard portal */
928 /* Turn off message rotor for wildcard portals */
929 #define LNET_PTL_ROTOR_OFF 0
930 /* round-robin dispatch all PUT messages for wildcard portals */
931 #define LNET_PTL_ROTOR_ON 1
932 /* round-robin dispatch routed PUT message for wildcard portals */
933 #define LNET_PTL_ROTOR_RR_RT 2
934 /* dispatch routed PUT message by hashing source NID for wildcard portals */
935 #define LNET_PTL_ROTOR_HASH_RT 3
939 unsigned int ptl_index; /* portal ID, reserved */
940 /* flags on this portal: lazy, unique... */
941 unsigned int ptl_options;
942 /* list of messages which are stealing buffer */
943 struct list_head ptl_msg_stealing;
944 /* messages blocking for MD */
945 struct list_head ptl_msg_delayed;
946 /* Match table for each CPT */
947 struct lnet_match_table **ptl_mtables;
948 /* spread rotor of incoming "PUT" */
949 unsigned int ptl_rotor;
950 /* # active entries for this portal */
952 /* array of active entries' cpu-partition-id */
956 #define LNET_LH_HASH_BITS 12
957 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
958 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
960 /* resource container (ME, MD, EQ) */
961 struct lnet_res_container {
962 unsigned int rec_type; /* container type */
963 __u64 rec_lh_cookie; /* cookie generator */
964 struct list_head rec_active; /* active resource list */
965 struct list_head *rec_lh_hash; /* handle hash */
968 /* message container */
969 struct lnet_msg_container {
970 int msc_init; /* initialized or not */
971 /* max # threads finalizing */
973 /* msgs waiting to complete finalizing */
974 struct list_head msc_finalizing;
975 struct list_head msc_active; /* active message list */
976 /* threads doing finalization */
977 void **msc_finalizers;
980 /* Peer Discovery states */
981 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
982 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
983 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
985 /* Router Checker states */
986 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
987 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
988 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
991 #define LNET_STATE_SHUTDOWN 0 /* not started */
992 #define LNET_STATE_RUNNING 1 /* started up OK */
993 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
996 /* CPU partition table of LNet */
997 struct cfs_cpt_table *ln_cpt_table;
998 /* number of CPTs in ln_cpt_table */
999 unsigned int ln_cpt_number;
1000 unsigned int ln_cpt_bits;
1002 /* protect LNet resources (ME/MD/EQ) */
1003 struct cfs_percpt_lock *ln_res_lock;
1006 /* the vector of portals */
1007 struct lnet_portal **ln_portals;
1008 /* percpt ME containers */
1009 struct lnet_res_container **ln_me_containers;
1010 /* percpt MD container */
1011 struct lnet_res_container **ln_md_containers;
1013 /* Event Queue container */
1014 struct lnet_res_container ln_eq_container;
1015 wait_queue_head_t ln_eq_waitq;
1016 spinlock_t ln_eq_wait_lock;
1018 unsigned int ln_remote_nets_hbits;
1020 /* protect NI, peer table, credits, routers, rtrbuf... */
1021 struct cfs_percpt_lock *ln_net_lock;
1022 /* percpt message containers for active/finalizing/freed message */
1023 struct lnet_msg_container **ln_msg_containers;
1024 struct lnet_counters **ln_counters;
1025 struct lnet_peer_table **ln_peer_tables;
1026 /* list of peer nis not on a local network */
1027 struct list_head ln_remote_peer_ni_list;
1028 /* failure simulation */
1029 struct list_head ln_test_peers;
1030 struct list_head ln_drop_rules;
1031 struct list_head ln_delay_rules;
1033 struct list_head ln_nets;
1034 /* the loopback NI */
1035 struct lnet_ni *ln_loni;
1036 /* network zombie list */
1037 struct list_head ln_net_zombie;
1038 /* resend messages list */
1039 struct list_head ln_msg_resend;
1040 /* spin lock to protect the msg resend list */
1041 spinlock_t ln_msg_resend_lock;
1043 /* remote networks with routes to them */
1044 struct list_head *ln_remote_nets_hash;
1045 /* validity stamp */
1046 __u64 ln_remote_nets_version;
1047 /* list of all known routers */
1048 struct list_head ln_routers;
1049 /* validity stamp */
1050 __u64 ln_routers_version;
1051 /* percpt router buffer pools */
1052 struct lnet_rtrbufpool **ln_rtrpools;
1055 * Ping target / Push source
1057 * The ping target and push source share a single buffer. The
1058 * ln_ping_target is protected against concurrent updates by
1061 struct lnet_handle_md ln_ping_target_md;
1062 struct lnet_handle_eq ln_ping_target_eq;
1063 struct lnet_ping_buffer *ln_ping_target;
1064 atomic_t ln_ping_target_seqno;
1069 * ln_push_nnis contains the desired size of the push target.
1070 * The lnet_net_lock is used to handle update races. The old
1071 * buffer may linger a while after it has been unlinked, in
1072 * which case the event handler cleans up.
1074 struct lnet_handle_eq ln_push_target_eq;
1075 struct lnet_handle_md ln_push_target_md;
1076 struct lnet_ping_buffer *ln_push_target;
1077 int ln_push_target_nnis;
1079 /* discovery event queue handle */
1080 struct lnet_handle_eq ln_dc_eqh;
1081 /* discovery requests */
1082 struct list_head ln_dc_request;
1083 /* discovery working list */
1084 struct list_head ln_dc_working;
1085 /* discovery expired list */
1086 struct list_head ln_dc_expired;
1087 /* discovery thread wait queue */
1088 wait_queue_head_t ln_dc_waitq;
1089 /* discovery startup/shutdown state */
1092 /* monitor thread startup/shutdown state */
1094 /* serialise startup/shutdown */
1095 struct semaphore ln_mt_signal;
1097 struct mutex ln_api_mutex;
1098 struct mutex ln_lnd_mutex;
1099 /* Have I called LNetNIInit myself? */
1101 /* LNetNIInit/LNetNIFini counter */
1103 /* SHUTDOWN/RUNNING/STOPPING */
1106 int ln_routing; /* am I a router? */
1107 lnet_pid_t ln_pid; /* requested pid */
1108 /* uniquely identifies this ni in this epoch */
1109 __u64 ln_interface_cookie;
1110 /* registered LNDs */
1111 struct list_head ln_lnds;
1113 /* test protocol compatibility flags */
1114 int ln_testprotocompat;
1116 /* 0 - load the NIs from the mod params
1117 * 1 - do not load the NIs from the mod params
1118 * Reverse logic to ensure that other calls to LNetNIInit
1121 bool ln_nis_from_mod_params;
1124 * waitq for the monitor thread. The monitor thread takes care of
1125 * checking routes, timedout messages and resending messages.
1127 wait_queue_head_t ln_mt_waitq;
1129 /* per-cpt resend queues */
1130 struct list_head **ln_mt_resendqs;
1131 /* local NIs to recover */
1132 struct list_head ln_mt_localNIRecovq;
1133 /* local NIs to recover */
1134 struct list_head ln_mt_peerNIRecovq;
1136 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1137 * There are CPT number of queues. Since response trackers will be
1138 * added on the fast path we can't afford to grab the exclusive
1139 * net lock to protect these queues. The CPT will be calculated
1140 * based on the mdh cookie.
1142 struct list_head **ln_mt_rstq;
1144 * A response tracker becomes a zombie when the associated MD is queued
1145 * for unlink before the response tracker is detached from the MD. An
1146 * entry on a zombie list can be freed when either the remaining
1147 * operations on the MD complete or when LNet has shut down.
1149 struct list_head **ln_mt_zombie_rstqs;
1150 /* recovery eq handler */
1151 struct lnet_handle_eq ln_mt_eqh;
1154 * Completed when the discovery and monitor threads can enter their
1157 struct completion ln_started;