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 list_head msg_activelist;
83 struct list_head msg_list; /* Q for credits/MD */
85 struct lnet_process_id msg_target;
86 /* Primary NID of the source. */
87 lnet_nid_t msg_initiator;
88 /* where is it from, it's only for building event */
93 * hold parameters in case message is with held due
96 lnet_nid_t msg_src_nid_param;
97 lnet_nid_t msg_rtr_nid_param;
100 * Deadline for the message after which it will be finalized if it
103 ktime_t msg_deadline;
105 /* The message health status. */
106 enum lnet_msg_hstatus msg_health_status;
107 /* This is a recovery message */
109 /* the number of times a transmission has been retried */
111 /* flag to indicate that we do not want to resend this message */
114 /* committed for sending */
115 unsigned int msg_tx_committed:1;
116 /* CPT # this message committed for sending */
117 unsigned int msg_tx_cpt:15;
118 /* committed for receiving */
119 unsigned int msg_rx_committed:1;
120 /* CPT # this message committed for receiving */
121 unsigned int msg_rx_cpt:15;
122 /* queued for tx credit */
123 unsigned int msg_tx_delayed:1;
124 /* queued for RX buffer */
125 unsigned int msg_rx_delayed:1;
126 /* ready for pending on RX delay list */
127 unsigned int msg_rx_ready_delay:1;
129 unsigned int msg_vmflush:1; /* VM trying to free memory */
130 unsigned int msg_target_is_router:1; /* sending to a router */
131 unsigned int msg_routing:1; /* being forwarded */
132 unsigned int msg_ack:1; /* ack on finalize (PUT) */
133 unsigned int msg_sending:1; /* outgoing message */
134 unsigned int msg_receiving:1; /* being received */
135 unsigned int msg_txcredit:1; /* taken an NI send credit */
136 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
137 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
138 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
139 unsigned int msg_onactivelist:1; /* on the activelist */
140 unsigned int msg_rdma_get:1;
142 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
143 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
146 struct lnet_libmd *msg_md;
147 /* the NI the message was sent or received over */
148 struct lnet_ni *msg_txni;
149 struct lnet_ni *msg_rxni;
151 unsigned int msg_len;
152 unsigned int msg_wanted;
153 unsigned int msg_offset;
154 unsigned int msg_niov;
155 struct kvec *msg_iov;
156 lnet_kiov_t *msg_kiov;
158 struct lnet_event msg_ev;
159 struct lnet_hdr msg_hdr;
162 struct lnet_libhandle {
163 struct list_head lh_hash_chain;
167 #define lh_entry(ptr, type, member) \
168 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
171 struct list_head eq_list;
172 struct lnet_libhandle eq_lh;
173 unsigned long eq_enq_seq;
174 unsigned long eq_deq_seq;
175 unsigned int eq_size;
176 lnet_eq_handler_t eq_callback;
177 struct lnet_event *eq_events;
178 int **eq_refs; /* percpt refcount for EQ */
182 struct list_head me_list;
183 struct lnet_libhandle me_lh;
184 struct lnet_process_id me_match_id;
185 unsigned int me_portal;
186 unsigned int me_pos; /* hash offset in mt_hash */
188 __u64 me_ignore_bits;
189 enum lnet_unlink me_unlink;
190 struct lnet_libmd *me_md;
194 struct list_head md_list;
195 struct lnet_libhandle md_lh;
196 struct lnet_me *md_me;
198 unsigned int md_offset;
199 unsigned int md_length;
200 unsigned int md_max_size;
203 unsigned int md_options;
204 unsigned int md_flags;
205 unsigned int md_niov; /* # frags at end of struct */
207 struct lnet_eq *md_eq;
208 struct lnet_handle_md md_bulk_handle;
210 struct kvec iov[LNET_MAX_IOV];
211 lnet_kiov_t kiov[LNET_MAX_IOV];
215 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
216 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
217 #define LNET_MD_FLAG_ABORTED (1 << 2)
219 struct lnet_test_peer {
220 /* info about peers we are trying to fail */
221 struct list_head tp_list; /* ln_test_peers */
222 lnet_nid_t tp_nid; /* matching nid */
223 unsigned int tp_threshold; /* # failures to simulate */
226 #define LNET_COOKIE_TYPE_MD 1
227 #define LNET_COOKIE_TYPE_ME 2
228 #define LNET_COOKIE_TYPE_EQ 3
229 #define LNET_COOKIE_TYPE_BITS 2
230 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
232 struct lnet_ni; /* forward ref */
236 /* fields managed by portals */
237 struct list_head lnd_list; /* stash in the LND table */
238 int lnd_refcount; /* # active instances */
240 /* fields initialized by the LND */
243 int (*lnd_startup)(struct lnet_ni *ni);
244 void (*lnd_shutdown)(struct lnet_ni *ni);
245 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
247 /* In data movement APIs below, payload buffers are described as a set
248 * of 'niov' fragments which are...
250 * in virtual memory (struct kvec *iov != NULL)
252 * in pages (kernel only: plt_kiov_t *kiov != NULL).
253 * The LND may NOT overwrite these fragment descriptors.
254 * An 'offset' and may specify a byte offset within the set of
255 * fragments to start from
258 /* Start sending a preformatted message. 'private' is NULL for PUT and
259 * GET messages; otherwise this is a response to an incoming message
260 * and 'private' is the 'private' passed to lnet_parse(). Return
261 * non-zero for immediate failure, otherwise complete later with
263 int (*lnd_send)(struct lnet_ni *ni, void *private,
264 struct lnet_msg *msg);
266 /* Start receiving 'mlen' bytes of payload data, skipping the following
267 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
268 * lnet_parse(). Return non-zero for immedaite failure, otherwise
269 * complete later with lnet_finalize(). This also gives back a receive
270 * credit if the LND does flow control. */
271 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
272 int delayed, unsigned int niov,
273 struct kvec *iov, lnet_kiov_t *kiov,
274 unsigned int offset, unsigned int mlen, unsigned int rlen);
276 /* lnet_parse() has had to delay processing of this message
277 * (e.g. waiting for a forwarding buffer or send credits). Give the
278 * LND a chance to free urgently needed resources. If called, return 0
279 * for success and do NOT give back a receive credit; that has to wait
280 * until lnd_recv() gets called. On failure return < 0 and
281 * release resources; lnd_recv() will not be called. */
282 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
283 struct lnet_msg *msg, void **new_privatep);
285 /* notification of peer health */
286 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
288 /* query of peer aliveness */
289 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, time64_t *when);
291 /* accept a new connection */
292 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
295 struct lnet_tx_queue {
296 int tq_credits; /* # tx credits free */
297 int tq_credits_min; /* lowest it's been */
298 int tq_credits_max; /* total # tx credits */
299 struct list_head tq_delayed; /* delayed TXs */
302 enum lnet_net_state {
303 /* set when net block is allocated */
304 LNET_NET_STATE_INIT = 0,
305 /* set when NIs in net are started successfully */
306 LNET_NET_STATE_ACTIVE,
307 /* set if all NIs in net are in FAILED state */
308 LNET_NET_STATE_INACTIVE,
309 /* set when shutting down a NET */
310 LNET_NET_STATE_DELETING
313 #define LNET_NI_STATE_INIT (1 << 0)
314 #define LNET_NI_STATE_ACTIVE (1 << 1)
315 #define LNET_NI_STATE_FAILED (1 << 2)
316 #define LNET_NI_STATE_RECOVERY_PENDING (1 << 3)
317 #define LNET_NI_STATE_DELETING (1 << 4)
319 enum lnet_stats_type {
320 LNET_STATS_TYPE_SEND = 0,
321 LNET_STATS_TYPE_RECV,
325 struct lnet_comm_count {
326 atomic_t co_get_count;
327 atomic_t co_put_count;
328 atomic_t co_reply_count;
329 atomic_t co_ack_count;
330 atomic_t co_hello_count;
333 struct lnet_element_stats {
334 struct lnet_comm_count el_send_stats;
335 struct lnet_comm_count el_recv_stats;
336 struct lnet_comm_count el_drop_stats;
340 /* chain on the ln_nets */
341 struct list_head net_list;
343 /* net ID, which is composed of
344 * (net_type << 16) | net_num.
345 * net_type can be one of the enumerated types defined in
346 * lnet/include/lnet/nidstr.h */
349 /* priority of the network */
352 /* total number of CPTs in the array */
355 /* cumulative CPTs of all NIs in this net */
358 /* network tunables */
359 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
362 * boolean to indicate that the tunables have been set and
365 bool net_tunables_set;
367 /* procedural interface */
368 struct lnet_lnd *net_lnd;
370 /* list of NIs on this net */
371 struct list_head net_ni_list;
373 /* list of NIs being added, but not started yet */
374 struct list_head net_ni_added;
376 /* dying LND instances */
377 struct list_head net_ni_zombie;
380 enum lnet_net_state net_state;
384 /* chain on the lnet_net structure */
385 struct list_head ni_netlist;
387 /* chain on net_ni_cpt */
388 struct list_head ni_cptlist;
390 /* chain on the recovery queue */
391 struct list_head ni_recovery;
393 /* MD handle for recovery ping */
394 struct lnet_handle_md ni_ping_mdh;
401 /* bond NI on some CPTs */
404 /* interface's NID */
407 /* instance-specific data */
411 atomic_t ni_tx_credits;
413 /* percpt TX queues */
414 struct lnet_tx_queue **ni_tx_queues;
416 /* percpt reference count */
419 /* when I was last alive */
420 time64_t ni_last_alive;
422 /* pointer to parent network */
423 struct lnet_net *ni_net;
425 /* my health status */
426 struct lnet_ni_status *ni_status;
431 /* per NI LND tunables */
432 struct lnet_lnd_tunables ni_lnd_tunables;
434 /* lnd tunables set explicitly */
435 bool ni_lnd_tunables_set;
438 struct lnet_element_stats ni_stats;
440 /* physical device CPT */
443 /* sequence number used to round robin over nis within a net */
448 * initialized to LNET_MAX_HEALTH_VALUE
449 * Value is decremented every time we fail to send a message over
450 * this NI because of a NI specific failure.
451 * Value is incremented if we successfully send a message.
456 * equivalent interfaces to use
457 * This is an array because socklnd bonding can still be configured
459 char *ni_interfaces[LNET_INTERFACES_NUM];
460 struct net *ni_net_ns; /* original net namespace */
463 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
466 * Descriptor of a ping info buffer: keep a separate indicator of the
467 * size and a reference count. The type is used both as a source and
468 * sink of data, so we need to keep some information outside of the
469 * area that may be overwritten by network data.
471 struct lnet_ping_buffer {
474 struct lnet_ping_info pb_info;
477 #define LNET_PING_BUFFER_SIZE(NNIDS) \
478 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
479 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
480 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
482 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
483 container_of((PINFO), struct lnet_ping_buffer, pb_info)
485 /* router checker data, per router */
486 struct lnet_rc_data {
487 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
488 struct list_head rcd_list;
489 struct lnet_handle_md rcd_mdh; /* ping buffer MD */
490 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
491 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
492 int rcd_nnis; /* desired size of buffer */
495 struct lnet_peer_ni {
496 /* chain on lpn_peer_nis */
497 struct list_head lpni_peer_nis;
498 /* chain on remote peer list */
499 struct list_head lpni_on_remote_peer_ni_list;
500 /* chain on recovery queue */
501 struct list_head lpni_recovery;
502 /* chain on peer hash */
503 struct list_head lpni_hashlist;
504 /* messages blocking for tx credits */
505 struct list_head lpni_txq;
506 /* messages blocking for router credits */
507 struct list_head lpni_rtrq;
508 /* chain on router list */
509 struct list_head lpni_rtr_list;
510 /* pointer to peer net I'm part of */
511 struct lnet_peer_net *lpni_peer_net;
512 /* statistics kept on each peer NI */
513 struct lnet_element_stats lpni_stats;
514 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
515 spinlock_t lpni_lock;
516 /* # tx credits available */
519 int lpni_mintxcredits;
520 /* # router credits */
523 int lpni_minrtrcredits;
524 /* bytes queued for sending */
528 /* notification outstanding? */
530 /* outstanding notification for LND? */
532 /* some thread is handling notification */
534 /* SEND event outstanding from ping */
535 bool lpni_ping_notsent;
536 /* # times router went dead<->alive. Protected with lpni_lock */
537 int lpni_alive_count;
538 /* time of last aliveness news */
539 time64_t lpni_timestamp;
540 /* time of last ping attempt */
541 time64_t lpni_ping_timestamp;
542 /* != 0 if ping reply expected */
543 time64_t lpni_ping_deadline;
544 /* when I was last alive */
545 time64_t lpni_last_alive;
546 /* when lpni_ni was queried last time */
547 time64_t lpni_last_query;
548 /* network peer is on */
549 struct lnet_net *lpni_net;
553 atomic_t lpni_refcount;
554 /* health value for the peer */
555 atomic_t lpni_healthv;
556 /* recovery ping mdh */
557 struct lnet_handle_md lpni_recovery_ping_mdh;
558 /* CPT this peer attached on */
560 /* state flags -- protected by lpni_lock */
562 /* # refs from lnet_route_t::lr_gateway */
563 int lpni_rtr_refcount;
564 /* sequence number used to round robin over peer nis within a net */
566 /* sequence number used to round robin over gateways */
570 /* returned RC ping features. Protected with lpni_lock */
571 unsigned int lpni_ping_feats;
572 /* routes on this peer */
573 struct list_head lpni_routes;
574 /* preferred local nids: if only one, use lpni_pref.nid */
579 /* number of preferred NIDs in lnpi_pref_nids */
580 __u32 lpni_pref_nnids;
581 /* router checker state */
582 struct lnet_rc_data *lpni_rcd;
585 /* Preferred path added due to traffic on non-MR peer_ni */
586 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
587 /* peer is being recovered. */
588 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
589 /* peer is being deleted */
590 #define LNET_PEER_NI_DELETING (1 << 2)
593 /* chain on pt_peer_list */
594 struct list_head lp_peer_list;
596 /* list of peer nets */
597 struct list_head lp_peer_nets;
599 /* list of messages pending discovery*/
600 struct list_head lp_dc_pendq;
602 /* primary NID of the peer */
603 lnet_nid_t lp_primary_nid;
605 /* CPT of peer_table */
608 /* number of NIDs on this peer */
611 /* reference count */
612 atomic_t lp_refcount;
614 /* lock protecting peer state flags */
617 /* peer state flags */
620 /* buffer for data pushed by peer */
621 struct lnet_ping_buffer *lp_data;
623 /* MD handle for ping in progress */
624 struct lnet_handle_md lp_ping_mdh;
626 /* MD handle for push in progress */
627 struct lnet_handle_md lp_push_mdh;
629 /* number of NIDs for sizing push data */
632 /* NI config sequence number of peer */
635 /* Local NI config sequence number acked by peer */
638 /* Local NI config sequence number sent to peer */
639 __u32 lp_node_seqno_sent;
641 /* Ping error encountered during discovery. */
644 /* Push error encountered during discovery. */
647 /* Error encountered during discovery. */
650 /* time it was put on the ln_dc_working queue */
651 time64_t lp_last_queued;
653 /* link on discovery-related lists */
654 struct list_head lp_dc_list;
656 /* tasks waiting on discovery of this peer */
657 wait_queue_head_t lp_dc_waitq;
661 * The status flags in lp_state. Their semantics have chosen so that
662 * lp_state can be zero-initialized.
664 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
665 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
667 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
668 * NOT set when the peer was pinged by discovery.
670 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
671 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
673 * A peer is marked CONFIGURED if it was configured by DLC.
675 * In addition, a peer is marked DISCOVERED if it has fully passed
676 * through Peer Discovery.
678 * When Peer Discovery is disabled, the discovery thread will mark
679 * peers REDISCOVER to indicate that they should be re-examined if
680 * discovery is (re)enabled on the node.
682 * A peer that was created as the result of inbound traffic will not
685 #define LNET_PEER_CONFIGURED (1 << 2) /* Configured via DLC */
686 #define LNET_PEER_DISCOVERED (1 << 3) /* Peer was discovered */
687 #define LNET_PEER_REDISCOVER (1 << 4) /* Discovery was disabled */
689 * A peer is marked DISCOVERING when discovery is in progress.
690 * The other flags below correspond to stages of discovery.
692 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
693 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
694 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
695 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
696 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
697 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
698 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
700 * A ping can be forced as a way to fix up state, or as a manual
701 * intervention by an admin.
702 * A push can be forced in circumstances that would normally not
703 * allow for one to happen.
705 #define LNET_PEER_FORCE_PING (1 << 12) /* Forced Ping */
706 #define LNET_PEER_FORCE_PUSH (1 << 13) /* Forced Push */
708 struct lnet_peer_net {
709 /* chain on lp_peer_nets */
710 struct list_head lpn_peer_nets;
712 /* list of peer_nis on this network */
713 struct list_head lpn_peer_nis;
715 /* pointer to the peer I'm part of */
716 struct lnet_peer *lpn_peer;
721 /* reference count */
722 atomic_t lpn_refcount;
726 #define LNET_PEER_HASH_BITS 9
727 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
730 * peer hash table - one per CPT
732 * protected by lnet_net_lock/EX for update
738 * protected by pt_zombie_lock:
742 * pt_zombie lock nests inside lnet_net_lock
744 struct lnet_peer_table {
745 int pt_version; /* /proc validity stamp */
746 int pt_number; /* # peers_ni extant */
747 struct list_head *pt_hash; /* NID->peer hash */
748 struct list_head pt_peer_list; /* peers */
749 int pt_peers; /* # peers */
750 struct list_head pt_zombie_list; /* zombie peer_ni */
751 int pt_zombies; /* # zombie peers_ni */
752 spinlock_t pt_zombie_lock; /* protect list and count */
755 /* peer aliveness is enabled only on routers for peers in a network where the
756 * struct lnet_ni::ni_peertimeout has been set to a positive value
758 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
759 ((lp)->lpni_net) && \
760 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
763 struct list_head lr_list; /* chain on net */
764 struct list_head lr_gwlist; /* chain on gateway */
765 struct lnet_peer_ni *lr_gateway; /* router node */
766 __u32 lr_net; /* remote network number */
767 int lr_seq; /* sequence for round-robin */
768 unsigned int lr_downis; /* number of down NIs */
769 __u32 lr_hops; /* how far I am */
770 unsigned int lr_priority; /* route priority */
773 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
774 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
775 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
777 struct lnet_remotenet {
778 /* chain on ln_remote_nets_hash */
779 struct list_head lrn_list;
781 struct list_head lrn_routes;
786 /** lnet message has credit and can be submitted to lnd for send/receive */
787 #define LNET_CREDIT_OK 0
788 /** lnet message is waiting for credit */
789 #define LNET_CREDIT_WAIT 1
790 /** lnet message is waiting for discovery */
791 #define LNET_DC_WAIT 2
793 struct lnet_rtrbufpool {
794 /* my free buffer pool */
795 struct list_head rbp_bufs;
796 /* messages blocking for a buffer */
797 struct list_head rbp_msgs;
798 /* # pages in each buffer */
800 /* requested number of buffers */
801 int rbp_req_nbuffers;
802 /* # buffers actually allocated */
804 /* # free buffers / blocked messages */
811 struct list_head rb_list; /* chain on rbp_bufs */
812 struct lnet_rtrbufpool *rb_pool; /* owning pool */
813 lnet_kiov_t rb_kiov[0]; /* the buffer space */
816 #define LNET_PEER_HASHSIZE 503 /* prime! */
818 enum lnet_match_flags {
819 /* Didn't match anything */
820 LNET_MATCHMD_NONE = (1 << 0),
822 LNET_MATCHMD_OK = (1 << 1),
823 /* Must be discarded */
824 LNET_MATCHMD_DROP = (1 << 2),
825 /* match and buffer is exhausted */
826 LNET_MATCHMD_EXHAUSTED = (1 << 3),
828 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
831 /* Options for struct lnet_portal::ptl_options */
832 #define LNET_PTL_LAZY (1 << 0)
833 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
834 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
836 /* parameter for matching operations (GET, PUT) */
837 struct lnet_match_info {
839 struct lnet_process_id mi_id;
842 unsigned int mi_portal;
843 unsigned int mi_rlength;
844 unsigned int mi_roffset;
847 /* ME hash of RDMA portal */
848 #define LNET_MT_HASH_BITS 8
849 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
850 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
851 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
852 * the last entry is reserved for MEs with ignore-bits */
853 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
854 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
855 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
856 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
857 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
858 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
859 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
861 /* portal match table */
862 struct lnet_match_table {
863 /* reserved for upcoming patches, CPU partition ID */
865 unsigned int mt_portal; /* portal index */
866 /* match table is set as "enabled" if there's non-exhausted MD
867 * attached on mt_mhash, it's only valid for wildcard portal */
868 unsigned int mt_enabled;
869 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
870 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
871 struct list_head *mt_mhash; /* matching hash */
874 /* these are only useful for wildcard portal */
875 /* Turn off message rotor for wildcard portals */
876 #define LNET_PTL_ROTOR_OFF 0
877 /* round-robin dispatch all PUT messages for wildcard portals */
878 #define LNET_PTL_ROTOR_ON 1
879 /* round-robin dispatch routed PUT message for wildcard portals */
880 #define LNET_PTL_ROTOR_RR_RT 2
881 /* dispatch routed PUT message by hashing source NID for wildcard portals */
882 #define LNET_PTL_ROTOR_HASH_RT 3
886 unsigned int ptl_index; /* portal ID, reserved */
887 /* flags on this portal: lazy, unique... */
888 unsigned int ptl_options;
889 /* list of messages which are stealing buffer */
890 struct list_head ptl_msg_stealing;
891 /* messages blocking for MD */
892 struct list_head ptl_msg_delayed;
893 /* Match table for each CPT */
894 struct lnet_match_table **ptl_mtables;
895 /* spread rotor of incoming "PUT" */
896 unsigned int ptl_rotor;
897 /* # active entries for this portal */
899 /* array of active entries' cpu-partition-id */
903 #define LNET_LH_HASH_BITS 12
904 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
905 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
907 /* resource container (ME, MD, EQ) */
908 struct lnet_res_container {
909 unsigned int rec_type; /* container type */
910 __u64 rec_lh_cookie; /* cookie generator */
911 struct list_head rec_active; /* active resource list */
912 struct list_head *rec_lh_hash; /* handle hash */
915 /* message container */
916 struct lnet_msg_container {
917 int msc_init; /* initialized or not */
918 /* max # threads finalizing */
920 /* msgs waiting to complete finalizing */
921 struct list_head msc_finalizing;
922 struct list_head msc_active; /* active message list */
923 /* threads doing finalization */
924 void **msc_finalizers;
927 /* Peer Discovery states */
928 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
929 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
930 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
932 /* Router Checker states */
933 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
934 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
935 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
938 #define LNET_STATE_SHUTDOWN 0 /* not started */
939 #define LNET_STATE_RUNNING 1 /* started up OK */
940 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
943 /* CPU partition table of LNet */
944 struct cfs_cpt_table *ln_cpt_table;
945 /* number of CPTs in ln_cpt_table */
946 unsigned int ln_cpt_number;
947 unsigned int ln_cpt_bits;
949 /* protect LNet resources (ME/MD/EQ) */
950 struct cfs_percpt_lock *ln_res_lock;
953 /* the vector of portals */
954 struct lnet_portal **ln_portals;
955 /* percpt ME containers */
956 struct lnet_res_container **ln_me_containers;
957 /* percpt MD container */
958 struct lnet_res_container **ln_md_containers;
960 /* Event Queue container */
961 struct lnet_res_container ln_eq_container;
962 wait_queue_head_t ln_eq_waitq;
963 spinlock_t ln_eq_wait_lock;
965 unsigned int ln_remote_nets_hbits;
967 /* protect NI, peer table, credits, routers, rtrbuf... */
968 struct cfs_percpt_lock *ln_net_lock;
969 /* percpt message containers for active/finalizing/freed message */
970 struct lnet_msg_container **ln_msg_containers;
971 struct lnet_counters **ln_counters;
972 struct lnet_peer_table **ln_peer_tables;
973 /* list of peer nis not on a local network */
974 struct list_head ln_remote_peer_ni_list;
975 /* failure simulation */
976 struct list_head ln_test_peers;
977 struct list_head ln_drop_rules;
978 struct list_head ln_delay_rules;
980 struct list_head ln_nets;
981 /* the loopback NI */
982 struct lnet_ni *ln_loni;
983 /* network zombie list */
984 struct list_head ln_net_zombie;
985 /* resend messages list */
986 struct list_head ln_msg_resend;
987 /* spin lock to protect the msg resend list */
988 spinlock_t ln_msg_resend_lock;
990 /* remote networks with routes to them */
991 struct list_head *ln_remote_nets_hash;
993 __u64 ln_remote_nets_version;
994 /* list of all known routers */
995 struct list_head ln_routers;
997 __u64 ln_routers_version;
998 /* percpt router buffer pools */
999 struct lnet_rtrbufpool **ln_rtrpools;
1002 * Ping target / Push source
1004 * The ping target and push source share a single buffer. The
1005 * ln_ping_target is protected against concurrent updates by
1008 struct lnet_handle_md ln_ping_target_md;
1009 struct lnet_handle_eq ln_ping_target_eq;
1010 struct lnet_ping_buffer *ln_ping_target;
1011 atomic_t ln_ping_target_seqno;
1016 * ln_push_nnis contains the desired size of the push target.
1017 * The lnet_net_lock is used to handle update races. The old
1018 * buffer may linger a while after it has been unlinked, in
1019 * which case the event handler cleans up.
1021 struct lnet_handle_eq ln_push_target_eq;
1022 struct lnet_handle_md ln_push_target_md;
1023 struct lnet_ping_buffer *ln_push_target;
1024 int ln_push_target_nnis;
1026 /* discovery event queue handle */
1027 struct lnet_handle_eq ln_dc_eqh;
1028 /* discovery requests */
1029 struct list_head ln_dc_request;
1030 /* discovery working list */
1031 struct list_head ln_dc_working;
1032 /* discovery expired list */
1033 struct list_head ln_dc_expired;
1034 /* discovery thread wait queue */
1035 wait_queue_head_t ln_dc_waitq;
1036 /* discovery startup/shutdown state */
1039 /* monitor thread startup/shutdown state */
1041 /* router checker's event queue */
1042 struct lnet_handle_eq ln_rc_eqh;
1043 /* rcd still pending on net */
1044 struct list_head ln_rcd_deathrow;
1045 /* rcd ready for free */
1046 struct list_head ln_rcd_zombie;
1047 /* serialise startup/shutdown */
1048 struct semaphore ln_mt_signal;
1050 struct mutex ln_api_mutex;
1051 struct mutex ln_lnd_mutex;
1052 /* Have I called LNetNIInit myself? */
1054 /* LNetNIInit/LNetNIFini counter */
1056 /* SHUTDOWN/RUNNING/STOPPING */
1059 int ln_routing; /* am I a router? */
1060 lnet_pid_t ln_pid; /* requested pid */
1061 /* uniquely identifies this ni in this epoch */
1062 __u64 ln_interface_cookie;
1063 /* registered LNDs */
1064 struct list_head ln_lnds;
1066 /* test protocol compatibility flags */
1067 int ln_testprotocompat;
1069 /* 0 - load the NIs from the mod params
1070 * 1 - do not load the NIs from the mod params
1071 * Reverse logic to ensure that other calls to LNetNIInit
1074 bool ln_nis_from_mod_params;
1077 * waitq for the monitor thread. The monitor thread takes care of
1078 * checking routes, timedout messages and resending messages.
1080 wait_queue_head_t ln_mt_waitq;
1082 /* per-cpt resend queues */
1083 struct list_head **ln_mt_resendqs;
1084 /* local NIs to recover */
1085 struct list_head ln_mt_localNIRecovq;
1086 /* local NIs to recover */
1087 struct list_head ln_mt_peerNIRecovq;
1088 /* recovery eq handler */
1089 struct lnet_handle_eq ln_mt_eqh;