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 bio_vec *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 me_list;
187 struct lnet_process_id me_match_id;
188 unsigned int me_portal;
189 unsigned int me_pos; /* hash offset in mt_hash */
191 __u64 me_ignore_bits;
192 enum lnet_unlink me_unlink;
193 struct lnet_libmd *me_md;
197 struct list_head md_list;
198 struct lnet_libhandle md_lh;
199 struct lnet_me *md_me;
201 unsigned int md_offset;
202 unsigned int md_length;
203 unsigned int md_max_size;
206 unsigned int md_options;
207 unsigned int md_flags;
208 unsigned int md_niov; /* # frags at end of struct */
210 struct lnet_rsp_tracker *md_rspt_ptr;
211 lnet_handler_t md_handler;
212 struct lnet_handle_md md_bulk_handle;
213 struct bio_vec md_kiov[LNET_MAX_IOV];
216 #define LNET_MD_FLAG_ZOMBIE BIT(0)
217 #define LNET_MD_FLAG_AUTO_UNLINK BIT(1)
218 #define LNET_MD_FLAG_ABORTED BIT(2)
220 struct lnet_test_peer {
221 /* info about peers we are trying to fail */
222 struct list_head tp_list; /* ln_test_peers */
223 lnet_nid_t tp_nid; /* matching nid */
224 unsigned int tp_threshold; /* # failures to simulate */
227 #define LNET_COOKIE_TYPE_MD 1
228 #define LNET_COOKIE_TYPE_ME 2
229 #define LNET_COOKIE_TYPE_EQ 3
230 #define LNET_COOKIE_TYPE_BITS 2
231 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
233 struct lnet_ni; /* forward ref */
237 /* fields initialized by the LND */
240 int (*lnd_startup)(struct lnet_ni *ni);
241 void (*lnd_shutdown)(struct lnet_ni *ni);
242 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
244 /* In data movement APIs below, payload buffers are described as a set
245 * of 'niov' fragments which are in pages.
246 * The LND may NOT overwrite these fragment descriptors.
247 * An 'offset' and may specify a byte offset within the set of
248 * fragments to start from
251 /* Start sending a preformatted message. 'private' is NULL for PUT and
252 * GET messages; otherwise this is a response to an incoming message
253 * and 'private' is the 'private' passed to lnet_parse(). Return
254 * non-zero for immediate failure, otherwise complete later with
256 int (*lnd_send)(struct lnet_ni *ni, void *private,
257 struct lnet_msg *msg);
259 /* Start receiving 'mlen' bytes of payload data, skipping the following
260 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
261 * lnet_parse(). Return non-zero for immedaite failure, otherwise
262 * complete later with lnet_finalize(). This also gives back a receive
263 * credit if the LND does flow control. */
264 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
265 int delayed, unsigned int niov,
266 struct bio_vec *kiov,
267 unsigned int offset, unsigned int mlen, unsigned int rlen);
269 /* lnet_parse() has had to delay processing of this message
270 * (e.g. waiting for a forwarding buffer or send credits). Give the
271 * LND a chance to free urgently needed resources. If called, return 0
272 * for success and do NOT give back a receive credit; that has to wait
273 * until lnd_recv() gets called. On failure return < 0 and
274 * release resources; lnd_recv() will not be called. */
275 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
276 struct lnet_msg *msg, void **new_privatep);
278 /* notification of peer down */
279 void (*lnd_notify_peer_down)(lnet_nid_t peer);
281 /* accept a new connection */
282 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
285 struct lnet_tx_queue {
286 int tq_credits; /* # tx credits free */
287 int tq_credits_min; /* lowest it's been */
288 int tq_credits_max; /* total # tx credits */
289 struct list_head tq_delayed; /* delayed TXs */
292 enum lnet_net_state {
293 /* set when net block is allocated */
294 LNET_NET_STATE_INIT = 0,
295 /* set when NIs in net are started successfully */
296 LNET_NET_STATE_ACTIVE,
297 /* set if all NIs in net are in FAILED state */
298 LNET_NET_STATE_INACTIVE,
299 /* set when shutting down a NET */
300 LNET_NET_STATE_DELETING
304 /* initial state when NI is created */
305 LNET_NI_STATE_INIT = 0,
306 /* set when NI is brought up */
307 LNET_NI_STATE_ACTIVE,
308 /* set when NI is being shutdown */
309 LNET_NI_STATE_DELETING,
312 #define LNET_NI_RECOVERY_PENDING BIT(0)
313 #define LNET_NI_RECOVERY_FAILED BIT(1)
315 enum lnet_stats_type {
316 LNET_STATS_TYPE_SEND = 0,
317 LNET_STATS_TYPE_RECV,
321 struct lnet_comm_count {
322 atomic_t co_get_count;
323 atomic_t co_put_count;
324 atomic_t co_reply_count;
325 atomic_t co_ack_count;
326 atomic_t co_hello_count;
329 struct lnet_element_stats {
330 struct lnet_comm_count el_send_stats;
331 struct lnet_comm_count el_recv_stats;
332 struct lnet_comm_count el_drop_stats;
335 struct lnet_health_local_stats {
336 atomic_t hlt_local_interrupt;
337 atomic_t hlt_local_dropped;
338 atomic_t hlt_local_aborted;
339 atomic_t hlt_local_no_route;
340 atomic_t hlt_local_timeout;
341 atomic_t hlt_local_error;
344 struct lnet_health_remote_stats {
345 atomic_t hlt_remote_dropped;
346 atomic_t hlt_remote_timeout;
347 atomic_t hlt_remote_error;
348 atomic_t hlt_network_timeout;
352 /* chain on the ln_nets */
353 struct list_head net_list;
355 /* net ID, which is composed of
356 * (net_type << 16) | net_num.
357 * net_type can be one of the enumerated types defined in
358 * lnet/include/lnet/nidstr.h */
361 /* priority of the network */
364 /* total number of CPTs in the array */
367 /* cumulative CPTs of all NIs in this net */
370 /* network tunables */
371 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
374 * boolean to indicate that the tunables have been set and
377 bool net_tunables_set;
379 /* procedural interface */
380 const struct lnet_lnd *net_lnd;
382 /* list of NIs on this net */
383 struct list_head net_ni_list;
385 /* list of NIs being added, but not started yet */
386 struct list_head net_ni_added;
388 /* dying LND instances */
389 struct list_head net_ni_zombie;
391 /* when I was last alive */
392 time64_t net_last_alive;
394 /* protects access to net_last_alive */
399 /* chain on the lnet_net structure */
400 struct list_head ni_netlist;
402 /* chain on the recovery queue */
403 struct list_head ni_recovery;
405 /* MD handle for recovery ping */
406 struct lnet_handle_md ni_ping_mdh;
413 /* bond NI on some CPTs */
416 /* interface's NID */
419 /* instance-specific data */
423 atomic_t ni_tx_credits;
425 /* percpt TX queues */
426 struct lnet_tx_queue **ni_tx_queues;
428 /* percpt reference count */
431 /* pointer to parent network */
432 struct lnet_net *ni_net;
434 /* my health status */
435 struct lnet_ni_status *ni_status;
437 /* NI FSM. Protected by lnet_ni_lock() */
438 enum lnet_ni_state ni_state;
440 /* Recovery state. Protected by lnet_ni_lock() */
441 __u32 ni_recovery_state;
443 /* per NI LND tunables */
444 struct lnet_lnd_tunables ni_lnd_tunables;
446 /* lnd tunables set explicitly */
447 bool ni_lnd_tunables_set;
450 struct lnet_element_stats ni_stats;
451 struct lnet_health_local_stats ni_hstats;
453 /* physical device CPT */
456 /* sequence number used to round robin over nis within a net */
461 * initialized to LNET_MAX_HEALTH_VALUE
462 * Value is decremented every time we fail to send a message over
463 * this NI because of a NI specific failure.
464 * Value is incremented if we successfully send a message.
469 * Set to 1 by the LND when it receives an event telling it the device
470 * has gone into a fatal state. Set to 0 when the LND receives an
471 * even telling it the device is back online.
473 atomic_t ni_fatal_error_on;
476 * equivalent interfaces to use
477 * This is an array because socklnd bonding can still be configured
479 char *ni_interfaces[LNET_INTERFACES_NUM];
480 struct net *ni_net_ns; /* original net namespace */
483 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
486 * Descriptor of a ping info buffer: keep a separate indicator of the
487 * size and a reference count. The type is used both as a source and
488 * sink of data, so we need to keep some information outside of the
489 * area that may be overwritten by network data.
491 struct lnet_ping_buffer {
495 struct lnet_ping_info pb_info;
498 #define LNET_PING_BUFFER_SIZE(NNIDS) \
499 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
500 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
501 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
503 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
504 container_of((PINFO), struct lnet_ping_buffer, pb_info)
506 struct lnet_peer_ni {
507 /* chain on lpn_peer_nis */
508 struct list_head lpni_peer_nis;
509 /* chain on remote peer list */
510 struct list_head lpni_on_remote_peer_ni_list;
511 /* chain on recovery queue */
512 struct list_head lpni_recovery;
513 /* chain on peer hash */
514 struct list_head lpni_hashlist;
515 /* messages blocking for tx credits */
516 struct list_head lpni_txq;
517 /* pointer to peer net I'm part of */
518 struct lnet_peer_net *lpni_peer_net;
519 /* statistics kept on each peer NI */
520 struct lnet_element_stats lpni_stats;
521 struct lnet_health_remote_stats lpni_hstats;
522 /* spin lock protecting credits and lpni_txq */
523 spinlock_t lpni_lock;
524 /* # tx credits available */
527 int lpni_mintxcredits;
529 * Each peer_ni in a gateway maintains its own credits. This
530 * allows more traffic to gateways that have multiple interfaces.
532 /* # router credits */
535 int lpni_minrtrcredits;
536 /* bytes queued for sending */
538 /* network peer is on */
539 struct lnet_net *lpni_net;
543 atomic_t lpni_refcount;
544 /* health value for the peer */
545 atomic_t lpni_healthv;
546 /* recovery ping mdh */
547 struct lnet_handle_md lpni_recovery_ping_mdh;
548 /* CPT this peer attached on */
550 /* state flags -- protected by lpni_lock */
552 /* status of the peer NI as reported by the peer */
553 __u32 lpni_ns_status;
554 /* sequence number used to round robin over peer nis within a net */
556 /* sequence number used to round robin over gateways */
558 /* returned RC ping features. Protected with lpni_lock */
559 unsigned int lpni_ping_feats;
560 /* time last message was received from the peer */
561 time64_t lpni_last_alive;
562 /* preferred local nids: if only one, use lpni_pref.nid */
567 /* number of preferred NIDs in lnpi_pref_nids */
568 __u32 lpni_pref_nnids;
571 /* Preferred path added due to traffic on non-MR peer_ni */
572 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
573 /* peer is being recovered. */
574 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
575 /* recovery ping failed */
576 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
577 /* peer is being deleted */
578 #define LNET_PEER_NI_DELETING BIT(3)
581 /* chain on pt_peer_list */
582 struct list_head lp_peer_list;
584 /* list of peer nets */
585 struct list_head lp_peer_nets;
587 /* list of messages pending discovery*/
588 struct list_head lp_dc_pendq;
590 /* chain on router list */
591 struct list_head lp_rtr_list;
593 /* primary NID of the peer */
594 lnet_nid_t lp_primary_nid;
596 /* net to perform discovery on */
597 __u32 lp_disc_net_id;
599 /* CPT of peer_table */
602 /* number of NIDs on this peer */
605 /* # refs from lnet_route::lr_gateway */
609 * peer specific health sensitivity value to decrement peer nis in
610 * this peer with if set to something other than 0
612 __u32 lp_health_sensitivity;
614 /* messages blocking for router credits */
615 struct list_head lp_rtrq;
617 /* routes on this peer */
618 struct list_head lp_routes;
620 /* reference count */
621 atomic_t lp_refcount;
623 /* lock protecting peer state flags and lpni_rtrq */
626 /* peer state flags */
629 /* buffer for data pushed by peer */
630 struct lnet_ping_buffer *lp_data;
632 /* MD handle for ping in progress */
633 struct lnet_handle_md lp_ping_mdh;
635 /* MD handle for push in progress */
636 struct lnet_handle_md lp_push_mdh;
638 /* number of NIDs for sizing push data */
641 /* NI config sequence number of peer */
644 /* Local NI config sequence number acked by peer */
647 /* Local NI config sequence number sent to peer */
648 __u32 lp_node_seqno_sent;
650 /* Ping error encountered during discovery. */
653 /* Push error encountered during discovery. */
656 /* Error encountered during discovery. */
659 /* time it was put on the ln_dc_working queue */
660 time64_t lp_last_queued;
662 /* link on discovery-related lists */
663 struct list_head lp_dc_list;
665 /* tasks waiting on discovery of this peer */
666 wait_queue_head_t lp_dc_waitq;
668 /* cached peer aliveness */
673 * The status flags in lp_state. Their semantics have chosen so that
674 * lp_state can be zero-initialized.
676 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
677 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
679 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
680 * NOT set when the peer was pinged by discovery.
682 * A peer is marked ROUTER if it indicates so in the feature bit.
684 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
685 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
686 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
689 * A peer is marked CONFIGURED if it was configured by DLC.
691 * In addition, a peer is marked DISCOVERED if it has fully passed
692 * through Peer Discovery.
694 * When Peer Discovery is disabled, the discovery thread will mark
695 * peers REDISCOVER to indicate that they should be re-examined if
696 * discovery is (re)enabled on the node.
698 * A peer that was created as the result of inbound traffic will not
701 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
702 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
703 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
705 * A peer is marked DISCOVERING when discovery is in progress.
706 * The other flags below correspond to stages of discovery.
708 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
709 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
710 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
711 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
712 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
713 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
714 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
716 * A ping can be forced as a way to fix up state, or as a manual
717 * intervention by an admin.
718 * A push can be forced in circumstances that would normally not
719 * allow for one to happen.
721 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
722 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
724 /* force delete even if router */
725 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
727 /* gw undergoing alive discovery */
728 #define LNET_PEER_RTR_DISCOVERY BIT(16)
729 /* gw has undergone discovery (does not indicate success or failure) */
730 #define LNET_PEER_RTR_DISCOVERED BIT(17)
732 /* peer is marked for deletion */
733 #define LNET_PEER_MARK_DELETION BIT(18)
735 struct lnet_peer_net {
736 /* chain on lp_peer_nets */
737 struct list_head lpn_peer_nets;
739 /* list of peer_nis on this network */
740 struct list_head lpn_peer_nis;
742 /* pointer to the peer I'm part of */
743 struct lnet_peer *lpn_peer;
748 /* peer net health */
751 /* time of last router net check attempt */
752 time64_t lpn_rtrcheck_timestamp;
754 /* selection sequence number */
757 /* reference count */
758 atomic_t lpn_refcount;
762 #define LNET_PEER_HASH_BITS 9
763 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
766 * peer hash table - one per CPT
768 * protected by lnet_net_lock/EX for update
773 * protected by pt_zombie_lock:
777 * pt_zombie lock nests inside lnet_net_lock
779 struct lnet_peer_table {
780 int pt_version; /* /proc validity stamp */
781 struct list_head *pt_hash; /* NID->peer hash */
782 struct list_head pt_peer_list; /* peers */
783 int pt_peers; /* # peers */
784 struct list_head pt_zombie_list; /* zombie peer_ni */
785 int pt_zombies; /* # zombie peers_ni */
786 spinlock_t pt_zombie_lock; /* protect list and count */
789 /* peer aliveness is enabled only on routers for peers in a network where the
790 * struct lnet_ni::ni_peertimeout has been set to a positive value
792 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
793 ((lp)->lpni_net) && \
794 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
797 struct list_head lr_list; /* chain on net */
798 struct list_head lr_gwlist; /* chain on gateway */
799 struct lnet_peer *lr_gateway; /* router node */
800 lnet_nid_t lr_nid; /* NID used to add route */
801 __u32 lr_net; /* remote network number */
802 __u32 lr_lnet; /* local network number */
803 int lr_seq; /* sequence for round-robin */
804 __u32 lr_hops; /* how far I am */
805 unsigned int lr_priority; /* route priority */
806 bool lr_alive; /* cached route aliveness */
807 bool lr_single_hop; /* this route is single-hop */
810 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
811 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
812 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
814 struct lnet_remotenet {
815 /* chain on ln_remote_nets_hash */
816 struct list_head lrn_list;
818 struct list_head lrn_routes;
823 /** lnet message has credit and can be submitted to lnd for send/receive */
824 #define LNET_CREDIT_OK 0
825 /** lnet message is waiting for credit */
826 #define LNET_CREDIT_WAIT 1
827 /** lnet message is waiting for discovery */
828 #define LNET_DC_WAIT 2
830 struct lnet_rtrbufpool {
831 /* my free buffer pool */
832 struct list_head rbp_bufs;
833 /* messages blocking for a buffer */
834 struct list_head rbp_msgs;
835 /* # pages in each buffer */
837 /* requested number of buffers */
838 int rbp_req_nbuffers;
839 /* # buffers actually allocated */
841 /* # free buffers / blocked messages */
848 struct list_head rb_list; /* chain on rbp_bufs */
849 struct lnet_rtrbufpool *rb_pool; /* owning pool */
850 struct bio_vec rb_kiov[0]; /* the buffer space */
853 #define LNET_PEER_HASHSIZE 503 /* prime! */
855 enum lnet_match_flags {
856 /* Didn't match anything */
857 LNET_MATCHMD_NONE = BIT(0),
859 LNET_MATCHMD_OK = BIT(1),
860 /* Must be discarded */
861 LNET_MATCHMD_DROP = BIT(2),
862 /* match and buffer is exhausted */
863 LNET_MATCHMD_EXHAUSTED = BIT(3),
865 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
868 /* Options for struct lnet_portal::ptl_options */
869 #define LNET_PTL_LAZY BIT(0)
870 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
871 #define LNET_PTL_MATCH_WILDCARD BIT(2) /* wildcard match, request portal */
873 /* parameter for matching operations (GET, PUT) */
874 struct lnet_match_info {
876 struct lnet_process_id mi_id;
879 unsigned int mi_portal;
880 unsigned int mi_rlength;
881 unsigned int mi_roffset;
884 /* ME hash of RDMA portal */
885 #define LNET_MT_HASH_BITS 8
886 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
887 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
888 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
889 * the last entry is reserved for MEs with ignore-bits */
890 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
891 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
892 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
893 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
894 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
895 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
896 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
898 /* portal match table */
899 struct lnet_match_table {
900 /* reserved for upcoming patches, CPU partition ID */
902 unsigned int mt_portal; /* portal index */
903 /* match table is set as "enabled" if there's non-exhausted MD
904 * attached on mt_mhash, it's only valid for wildcard portal */
905 unsigned int mt_enabled;
906 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
907 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
908 struct list_head *mt_mhash; /* matching hash */
911 /* these are only useful for wildcard portal */
912 /* Turn off message rotor for wildcard portals */
913 #define LNET_PTL_ROTOR_OFF 0
914 /* round-robin dispatch all PUT messages for wildcard portals */
915 #define LNET_PTL_ROTOR_ON 1
916 /* round-robin dispatch routed PUT message for wildcard portals */
917 #define LNET_PTL_ROTOR_RR_RT 2
918 /* dispatch routed PUT message by hashing source NID for wildcard portals */
919 #define LNET_PTL_ROTOR_HASH_RT 3
923 unsigned int ptl_index; /* portal ID, reserved */
924 /* flags on this portal: lazy, unique... */
925 unsigned int ptl_options;
926 /* list of messages which are stealing buffer */
927 struct list_head ptl_msg_stealing;
928 /* messages blocking for MD */
929 struct list_head ptl_msg_delayed;
930 /* Match table for each CPT */
931 struct lnet_match_table **ptl_mtables;
932 /* spread rotor of incoming "PUT" */
933 unsigned int ptl_rotor;
934 /* # active entries for this portal */
936 /* array of active entries' cpu-partition-id */
940 #define LNET_LH_HASH_BITS 12
941 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
942 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
944 /* resource container (ME, MD, EQ) */
945 struct lnet_res_container {
946 unsigned int rec_type; /* container type */
947 __u64 rec_lh_cookie; /* cookie generator */
948 struct list_head rec_active; /* active resource list */
949 struct list_head *rec_lh_hash; /* handle hash */
952 /* message container */
953 struct lnet_msg_container {
954 int msc_init; /* initialized or not */
955 /* max # threads finalizing */
957 /* msgs waiting to complete finalizing */
958 struct list_head msc_finalizing;
959 /* msgs waiting to be resent */
960 struct list_head msc_resending;
961 struct list_head msc_active; /* active message list */
962 /* threads doing finalization */
963 void **msc_finalizers;
964 /* threads doing resends */
965 void **msc_resenders;
968 /* Peer Discovery states */
969 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
970 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
971 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
973 /* Router Checker states */
974 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
975 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
976 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
979 #define LNET_STATE_SHUTDOWN 0 /* not started */
980 #define LNET_STATE_RUNNING 1 /* started up OK */
981 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
984 /* CPU partition table of LNet */
985 struct cfs_cpt_table *ln_cpt_table;
986 /* number of CPTs in ln_cpt_table */
987 unsigned int ln_cpt_number;
988 unsigned int ln_cpt_bits;
990 /* protect LNet resources (ME/MD/EQ) */
991 struct cfs_percpt_lock *ln_res_lock;
994 /* the vector of portals */
995 struct lnet_portal **ln_portals;
996 /* percpt MD container */
997 struct lnet_res_container **ln_md_containers;
999 /* Event Queue container */
1000 struct lnet_res_container ln_eq_container;
1001 spinlock_t ln_eq_wait_lock;
1003 unsigned int ln_remote_nets_hbits;
1005 /* protect NI, peer table, credits, routers, rtrbuf... */
1006 struct cfs_percpt_lock *ln_net_lock;
1007 /* percpt message containers for active/finalizing/freed message */
1008 struct lnet_msg_container **ln_msg_containers;
1009 struct lnet_counters **ln_counters;
1010 struct lnet_peer_table **ln_peer_tables;
1011 /* list of peer nis not on a local network */
1012 struct list_head ln_remote_peer_ni_list;
1013 /* failure simulation */
1014 struct list_head ln_test_peers;
1015 struct list_head ln_drop_rules;
1016 struct list_head ln_delay_rules;
1018 struct list_head ln_nets;
1019 /* the loopback NI */
1020 struct lnet_ni *ln_loni;
1021 /* network zombie list */
1022 struct list_head ln_net_zombie;
1023 /* resend messages list */
1024 struct list_head ln_msg_resend;
1025 /* spin lock to protect the msg resend list */
1026 spinlock_t ln_msg_resend_lock;
1028 /* remote networks with routes to them */
1029 struct list_head *ln_remote_nets_hash;
1030 /* validity stamp */
1031 __u64 ln_remote_nets_version;
1032 /* list of all known routers */
1033 struct list_head ln_routers;
1034 /* validity stamp */
1035 __u64 ln_routers_version;
1036 /* percpt router buffer pools */
1037 struct lnet_rtrbufpool **ln_rtrpools;
1040 * Ping target / Push source
1042 * The ping target and push source share a single buffer. The
1043 * ln_ping_target is protected against concurrent updates by
1046 struct lnet_handle_md ln_ping_target_md;
1047 lnet_handler_t ln_ping_target_handler;
1048 struct lnet_ping_buffer *ln_ping_target;
1049 atomic_t ln_ping_target_seqno;
1054 * ln_push_nnis contains the desired size of the push target.
1055 * The lnet_net_lock is used to handle update races. The old
1056 * buffer may linger a while after it has been unlinked, in
1057 * which case the event handler cleans up.
1059 lnet_handler_t ln_push_target_handler;
1060 struct lnet_handle_md ln_push_target_md;
1061 struct lnet_ping_buffer *ln_push_target;
1062 int ln_push_target_nnis;
1064 /* discovery event queue handle */
1065 lnet_handler_t ln_dc_handler;
1066 /* discovery requests */
1067 struct list_head ln_dc_request;
1068 /* discovery working list */
1069 struct list_head ln_dc_working;
1070 /* discovery expired list */
1071 struct list_head ln_dc_expired;
1072 /* discovery thread wait queue */
1073 wait_queue_head_t ln_dc_waitq;
1074 /* discovery startup/shutdown state */
1077 /* monitor thread startup/shutdown state */
1079 /* serialise startup/shutdown */
1080 struct semaphore ln_mt_signal;
1082 struct mutex ln_api_mutex;
1083 struct mutex ln_lnd_mutex;
1084 /* Have I called LNetNIInit myself? */
1086 /* LNetNIInit/LNetNIFini counter */
1088 /* SHUTDOWN/RUNNING/STOPPING */
1091 int ln_routing; /* am I a router? */
1092 lnet_pid_t ln_pid; /* requested pid */
1093 /* uniquely identifies this ni in this epoch */
1094 __u64 ln_interface_cookie;
1095 /* registered LNDs */
1096 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1098 /* test protocol compatibility flags */
1099 unsigned long ln_testprotocompat;
1101 /* 0 - load the NIs from the mod params
1102 * 1 - do not load the NIs from the mod params
1103 * Reverse logic to ensure that other calls to LNetNIInit
1106 bool ln_nis_from_mod_params;
1109 * completion for the monitor thread. The monitor thread takes care of
1110 * checking routes, timedout messages and resending messages.
1112 struct completion ln_mt_wait_complete;
1114 /* per-cpt resend queues */
1115 struct list_head **ln_mt_resendqs;
1116 /* local NIs to recover */
1117 struct list_head ln_mt_localNIRecovq;
1118 /* local NIs to recover */
1119 struct list_head ln_mt_peerNIRecovq;
1121 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1122 * There are CPT number of queues. Since response trackers will be
1123 * added on the fast path we can't afford to grab the exclusive
1124 * net lock to protect these queues. The CPT will be calculated
1125 * based on the mdh cookie.
1127 struct list_head **ln_mt_rstq;
1129 * A response tracker becomes a zombie when the associated MD is queued
1130 * for unlink before the response tracker is detached from the MD. An
1131 * entry on a zombie list can be freed when either the remaining
1132 * operations on the MD complete or when LNet has shut down.
1134 struct list_head **ln_mt_zombie_rstqs;
1135 /* recovery handler */
1136 lnet_handler_t ln_mt_handler;
1139 * Completed when the discovery and monitor threads can enter their
1142 struct completion ln_started;