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
81 struct lnet_rsp_tracker {
82 /* chain on the waiting list */
83 struct list_head rspt_on_list;
86 /* deadline of the REPLY/ACK */
87 ktime_t rspt_deadline;
89 struct lnet_handle_md rspt_mdh;
93 struct list_head msg_activelist;
94 struct list_head msg_list; /* Q for credits/MD */
96 struct lnet_process_id msg_target;
97 /* Primary NID of the source. */
98 lnet_nid_t msg_initiator;
99 /* where is it from, it's only for building event */
104 * hold parameters in case message is with held due
107 lnet_nid_t msg_src_nid_param;
108 lnet_nid_t msg_rtr_nid_param;
111 * Deadline for the message after which it will be finalized if it
114 ktime_t msg_deadline;
116 /* The message health status. */
117 enum lnet_msg_hstatus msg_health_status;
118 /* This is a recovery message */
120 /* the number of times a transmission has been retried */
122 /* flag to indicate that we do not want to resend this message */
125 /* committed for sending */
126 unsigned int msg_tx_committed:1;
127 /* CPT # this message committed for sending */
128 unsigned int msg_tx_cpt:15;
129 /* committed for receiving */
130 unsigned int msg_rx_committed:1;
131 /* CPT # this message committed for receiving */
132 unsigned int msg_rx_cpt:15;
133 /* queued for tx credit */
134 unsigned int msg_tx_delayed:1;
135 /* queued for RX buffer */
136 unsigned int msg_rx_delayed:1;
137 /* ready for pending on RX delay list */
138 unsigned int msg_rx_ready_delay:1;
140 unsigned int msg_vmflush:1; /* VM trying to free memory */
141 unsigned int msg_target_is_router:1; /* sending to a router */
142 unsigned int msg_routing:1; /* being forwarded */
143 unsigned int msg_ack:1; /* ack on finalize (PUT) */
144 unsigned int msg_sending:1; /* outgoing message */
145 unsigned int msg_receiving:1; /* being received */
146 unsigned int msg_txcredit:1; /* taken an NI send credit */
147 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
148 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
149 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
150 unsigned int msg_onactivelist:1; /* on the activelist */
151 unsigned int msg_rdma_get:1;
153 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
154 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
157 struct lnet_libmd *msg_md;
158 /* the NI the message was sent or received over */
159 struct lnet_ni *msg_txni;
160 struct lnet_ni *msg_rxni;
162 unsigned int msg_len;
163 unsigned int msg_wanted;
164 unsigned int msg_offset;
165 unsigned int msg_niov;
166 struct kvec *msg_iov;
167 lnet_kiov_t *msg_kiov;
169 struct lnet_event msg_ev;
170 struct lnet_hdr msg_hdr;
173 struct lnet_libhandle {
174 struct list_head lh_hash_chain;
178 #define lh_entry(ptr, type, member) \
179 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
182 struct list_head eq_list;
183 struct lnet_libhandle eq_lh;
184 unsigned long eq_enq_seq;
185 unsigned long eq_deq_seq;
186 unsigned int eq_size;
187 lnet_eq_handler_t eq_callback;
188 struct lnet_event *eq_events;
189 int **eq_refs; /* percpt refcount for EQ */
193 struct list_head me_list;
194 struct lnet_libhandle me_lh;
195 struct lnet_process_id me_match_id;
196 unsigned int me_portal;
197 unsigned int me_pos; /* hash offset in mt_hash */
199 __u64 me_ignore_bits;
200 enum lnet_unlink me_unlink;
201 struct lnet_libmd *me_md;
205 struct list_head md_list;
206 struct lnet_libhandle md_lh;
207 struct lnet_me *md_me;
209 unsigned int md_offset;
210 unsigned int md_length;
211 unsigned int md_max_size;
214 unsigned int md_options;
215 unsigned int md_flags;
216 unsigned int md_niov; /* # frags at end of struct */
218 struct lnet_rsp_tracker *md_rspt_ptr;
219 struct lnet_eq *md_eq;
220 struct lnet_handle_md md_bulk_handle;
222 struct kvec iov[LNET_MAX_IOV];
223 lnet_kiov_t kiov[LNET_MAX_IOV];
227 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
228 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
229 #define LNET_MD_FLAG_ABORTED (1 << 2)
231 struct lnet_test_peer {
232 /* info about peers we are trying to fail */
233 struct list_head tp_list; /* ln_test_peers */
234 lnet_nid_t tp_nid; /* matching nid */
235 unsigned int tp_threshold; /* # failures to simulate */
238 #define LNET_COOKIE_TYPE_MD 1
239 #define LNET_COOKIE_TYPE_ME 2
240 #define LNET_COOKIE_TYPE_EQ 3
241 #define LNET_COOKIE_TYPE_BITS 2
242 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
244 struct lnet_ni; /* forward ref */
248 /* fields managed by portals */
249 struct list_head lnd_list; /* stash in the LND table */
250 int lnd_refcount; /* # active instances */
252 /* fields initialized by the LND */
255 int (*lnd_startup)(struct lnet_ni *ni);
256 void (*lnd_shutdown)(struct lnet_ni *ni);
257 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
259 /* In data movement APIs below, payload buffers are described as a set
260 * of 'niov' fragments which are...
262 * in virtual memory (struct kvec *iov != NULL)
264 * in pages (kernel only: plt_kiov_t *kiov != NULL).
265 * The LND may NOT overwrite these fragment descriptors.
266 * An 'offset' and may specify a byte offset within the set of
267 * fragments to start from
270 /* Start sending a preformatted message. 'private' is NULL for PUT and
271 * GET messages; otherwise this is a response to an incoming message
272 * and 'private' is the 'private' passed to lnet_parse(). Return
273 * non-zero for immediate failure, otherwise complete later with
275 int (*lnd_send)(struct lnet_ni *ni, void *private,
276 struct lnet_msg *msg);
278 /* Start receiving 'mlen' bytes of payload data, skipping the following
279 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
280 * lnet_parse(). Return non-zero for immedaite failure, otherwise
281 * complete later with lnet_finalize(). This also gives back a receive
282 * credit if the LND does flow control. */
283 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
284 int delayed, unsigned int niov,
285 struct kvec *iov, lnet_kiov_t *kiov,
286 unsigned int offset, unsigned int mlen, unsigned int rlen);
288 /* lnet_parse() has had to delay processing of this message
289 * (e.g. waiting for a forwarding buffer or send credits). Give the
290 * LND a chance to free urgently needed resources. If called, return 0
291 * for success and do NOT give back a receive credit; that has to wait
292 * until lnd_recv() gets called. On failure return < 0 and
293 * release resources; lnd_recv() will not be called. */
294 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
295 struct lnet_msg *msg, void **new_privatep);
297 /* notification of peer health */
298 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
300 /* query of peer aliveness */
301 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, time64_t *when);
303 /* accept a new connection */
304 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
307 struct lnet_tx_queue {
308 int tq_credits; /* # tx credits free */
309 int tq_credits_min; /* lowest it's been */
310 int tq_credits_max; /* total # tx credits */
311 struct list_head tq_delayed; /* delayed TXs */
314 enum lnet_net_state {
315 /* set when net block is allocated */
316 LNET_NET_STATE_INIT = 0,
317 /* set when NIs in net are started successfully */
318 LNET_NET_STATE_ACTIVE,
319 /* set if all NIs in net are in FAILED state */
320 LNET_NET_STATE_INACTIVE,
321 /* set when shutting down a NET */
322 LNET_NET_STATE_DELETING
325 #define LNET_NI_STATE_INIT (1 << 0)
326 #define LNET_NI_STATE_ACTIVE (1 << 1)
327 #define LNET_NI_STATE_FAILED (1 << 2)
328 #define LNET_NI_STATE_RECOVERY_PENDING (1 << 3)
329 #define LNET_NI_STATE_DELETING (1 << 4)
331 enum lnet_stats_type {
332 LNET_STATS_TYPE_SEND = 0,
333 LNET_STATS_TYPE_RECV,
337 struct lnet_comm_count {
338 atomic_t co_get_count;
339 atomic_t co_put_count;
340 atomic_t co_reply_count;
341 atomic_t co_ack_count;
342 atomic_t co_hello_count;
345 struct lnet_element_stats {
346 struct lnet_comm_count el_send_stats;
347 struct lnet_comm_count el_recv_stats;
348 struct lnet_comm_count el_drop_stats;
351 struct lnet_health_local_stats {
352 atomic_t hlt_local_interrupt;
353 atomic_t hlt_local_dropped;
354 atomic_t hlt_local_aborted;
355 atomic_t hlt_local_no_route;
356 atomic_t hlt_local_timeout;
357 atomic_t hlt_local_error;
360 struct lnet_health_remote_stats {
361 atomic_t hlt_remote_dropped;
362 atomic_t hlt_remote_timeout;
363 atomic_t hlt_remote_error;
364 atomic_t hlt_network_timeout;
368 /* chain on the ln_nets */
369 struct list_head net_list;
371 /* net ID, which is composed of
372 * (net_type << 16) | net_num.
373 * net_type can be one of the enumerated types defined in
374 * lnet/include/lnet/nidstr.h */
377 /* priority of the network */
380 /* total number of CPTs in the array */
383 /* cumulative CPTs of all NIs in this net */
386 /* network tunables */
387 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
390 * boolean to indicate that the tunables have been set and
393 bool net_tunables_set;
395 /* procedural interface */
396 struct lnet_lnd *net_lnd;
398 /* list of NIs on this net */
399 struct list_head net_ni_list;
401 /* list of NIs being added, but not started yet */
402 struct list_head net_ni_added;
404 /* dying LND instances */
405 struct list_head net_ni_zombie;
408 enum lnet_net_state net_state;
412 /* chain on the lnet_net structure */
413 struct list_head ni_netlist;
415 /* chain on net_ni_cpt */
416 struct list_head ni_cptlist;
418 /* chain on the recovery queue */
419 struct list_head ni_recovery;
421 /* MD handle for recovery ping */
422 struct lnet_handle_md ni_ping_mdh;
429 /* bond NI on some CPTs */
432 /* interface's NID */
435 /* instance-specific data */
439 atomic_t ni_tx_credits;
441 /* percpt TX queues */
442 struct lnet_tx_queue **ni_tx_queues;
444 /* percpt reference count */
447 /* when I was last alive */
448 time64_t ni_last_alive;
450 /* pointer to parent network */
451 struct lnet_net *ni_net;
453 /* my health status */
454 struct lnet_ni_status *ni_status;
459 /* per NI LND tunables */
460 struct lnet_lnd_tunables ni_lnd_tunables;
462 /* lnd tunables set explicitly */
463 bool ni_lnd_tunables_set;
466 struct lnet_element_stats ni_stats;
467 struct lnet_health_local_stats ni_hstats;
469 /* physical device CPT */
472 /* sequence number used to round robin over nis within a net */
477 * initialized to LNET_MAX_HEALTH_VALUE
478 * Value is decremented every time we fail to send a message over
479 * this NI because of a NI specific failure.
480 * Value is incremented if we successfully send a message.
485 * Set to 1 by the LND when it receives an event telling it the device
486 * has gone into a fatal state. Set to 0 when the LND receives an
487 * even telling it the device is back online.
489 atomic_t ni_fatal_error_on;
492 * equivalent interfaces to use
493 * This is an array because socklnd bonding can still be configured
495 char *ni_interfaces[LNET_INTERFACES_NUM];
496 struct net *ni_net_ns; /* original net namespace */
499 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
502 * Descriptor of a ping info buffer: keep a separate indicator of the
503 * size and a reference count. The type is used both as a source and
504 * sink of data, so we need to keep some information outside of the
505 * area that may be overwritten by network data.
507 struct lnet_ping_buffer {
510 struct lnet_ping_info pb_info;
513 #define LNET_PING_BUFFER_SIZE(NNIDS) \
514 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
515 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
516 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
518 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
519 container_of((PINFO), struct lnet_ping_buffer, pb_info)
521 /* router checker data, per router */
522 struct lnet_rc_data {
523 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
524 struct list_head rcd_list;
525 struct lnet_handle_md rcd_mdh; /* ping buffer MD */
526 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
527 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
528 int rcd_nnis; /* desired size of buffer */
531 struct lnet_peer_ni {
532 /* chain on lpn_peer_nis */
533 struct list_head lpni_peer_nis;
534 /* chain on remote peer list */
535 struct list_head lpni_on_remote_peer_ni_list;
536 /* chain on recovery queue */
537 struct list_head lpni_recovery;
538 /* chain on peer hash */
539 struct list_head lpni_hashlist;
540 /* messages blocking for tx credits */
541 struct list_head lpni_txq;
542 /* messages blocking for router credits */
543 struct list_head lpni_rtrq;
544 /* chain on router list */
545 struct list_head lpni_rtr_list;
546 /* pointer to peer net I'm part of */
547 struct lnet_peer_net *lpni_peer_net;
548 /* statistics kept on each peer NI */
549 struct lnet_element_stats lpni_stats;
550 struct lnet_health_remote_stats lpni_hstats;
551 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
552 spinlock_t lpni_lock;
553 /* # tx credits available */
556 int lpni_mintxcredits;
557 /* # router credits */
560 int lpni_minrtrcredits;
561 /* bytes queued for sending */
565 /* notification outstanding? */
567 /* outstanding notification for LND? */
569 /* some thread is handling notification */
571 /* SEND event outstanding from ping */
572 bool lpni_ping_notsent;
573 /* # times router went dead<->alive. Protected with lpni_lock */
574 int lpni_alive_count;
575 /* time of last aliveness news */
576 time64_t lpni_timestamp;
577 /* time of last ping attempt */
578 time64_t lpni_ping_timestamp;
579 /* != 0 if ping reply expected */
580 time64_t lpni_ping_deadline;
581 /* when I was last alive */
582 time64_t lpni_last_alive;
583 /* when lpni_ni was queried last time */
584 time64_t lpni_last_query;
585 /* network peer is on */
586 struct lnet_net *lpni_net;
590 atomic_t lpni_refcount;
591 /* health value for the peer */
592 atomic_t lpni_healthv;
593 /* recovery ping mdh */
594 struct lnet_handle_md lpni_recovery_ping_mdh;
595 /* CPT this peer attached on */
597 /* state flags -- protected by lpni_lock */
599 /* # refs from lnet_route_t::lr_gateway */
600 int lpni_rtr_refcount;
601 /* sequence number used to round robin over peer nis within a net */
603 /* sequence number used to round robin over gateways */
607 /* returned RC ping features. Protected with lpni_lock */
608 unsigned int lpni_ping_feats;
609 /* routes on this peer */
610 struct list_head lpni_routes;
611 /* preferred local nids: if only one, use lpni_pref.nid */
616 /* number of preferred NIDs in lnpi_pref_nids */
617 __u32 lpni_pref_nnids;
618 /* router checker state */
619 struct lnet_rc_data *lpni_rcd;
622 /* Preferred path added due to traffic on non-MR peer_ni */
623 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
624 /* peer is being recovered. */
625 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
626 /* peer is being deleted */
627 #define LNET_PEER_NI_DELETING (1 << 2)
630 /* chain on pt_peer_list */
631 struct list_head lp_peer_list;
633 /* list of peer nets */
634 struct list_head lp_peer_nets;
636 /* list of messages pending discovery*/
637 struct list_head lp_dc_pendq;
639 /* primary NID of the peer */
640 lnet_nid_t lp_primary_nid;
642 /* CPT of peer_table */
645 /* number of NIDs on this peer */
648 /* reference count */
649 atomic_t lp_refcount;
651 /* lock protecting peer state flags */
654 /* peer state flags */
657 /* buffer for data pushed by peer */
658 struct lnet_ping_buffer *lp_data;
660 /* MD handle for ping in progress */
661 struct lnet_handle_md lp_ping_mdh;
663 /* MD handle for push in progress */
664 struct lnet_handle_md lp_push_mdh;
666 /* number of NIDs for sizing push data */
669 /* NI config sequence number of peer */
672 /* Local NI config sequence number acked by peer */
675 /* Local NI config sequence number sent to peer */
676 __u32 lp_node_seqno_sent;
678 /* Ping error encountered during discovery. */
681 /* Push error encountered during discovery. */
684 /* Error encountered during discovery. */
687 /* time it was put on the ln_dc_working queue */
688 time64_t lp_last_queued;
690 /* link on discovery-related lists */
691 struct list_head lp_dc_list;
693 /* tasks waiting on discovery of this peer */
694 wait_queue_head_t lp_dc_waitq;
698 * The status flags in lp_state. Their semantics have chosen so that
699 * lp_state can be zero-initialized.
701 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
702 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
704 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
705 * NOT set when the peer was pinged by discovery.
707 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
708 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
710 * A peer is marked CONFIGURED if it was configured by DLC.
712 * In addition, a peer is marked DISCOVERED if it has fully passed
713 * through Peer Discovery.
715 * When Peer Discovery is disabled, the discovery thread will mark
716 * peers REDISCOVER to indicate that they should be re-examined if
717 * discovery is (re)enabled on the node.
719 * A peer that was created as the result of inbound traffic will not
722 #define LNET_PEER_CONFIGURED (1 << 2) /* Configured via DLC */
723 #define LNET_PEER_DISCOVERED (1 << 3) /* Peer was discovered */
724 #define LNET_PEER_REDISCOVER (1 << 4) /* Discovery was disabled */
726 * A peer is marked DISCOVERING when discovery is in progress.
727 * The other flags below correspond to stages of discovery.
729 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
730 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
731 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
732 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
733 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
734 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
735 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
737 * A ping can be forced as a way to fix up state, or as a manual
738 * intervention by an admin.
739 * A push can be forced in circumstances that would normally not
740 * allow for one to happen.
742 #define LNET_PEER_FORCE_PING (1 << 12) /* Forced Ping */
743 #define LNET_PEER_FORCE_PUSH (1 << 13) /* Forced Push */
745 struct lnet_peer_net {
746 /* chain on lp_peer_nets */
747 struct list_head lpn_peer_nets;
749 /* list of peer_nis on this network */
750 struct list_head lpn_peer_nis;
752 /* pointer to the peer I'm part of */
753 struct lnet_peer *lpn_peer;
758 /* reference count */
759 atomic_t lpn_refcount;
763 #define LNET_PEER_HASH_BITS 9
764 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
767 * peer hash table - one per CPT
769 * protected by lnet_net_lock/EX for update
775 * protected by pt_zombie_lock:
779 * pt_zombie lock nests inside lnet_net_lock
781 struct lnet_peer_table {
782 int pt_version; /* /proc validity stamp */
783 int pt_number; /* # peers_ni extant */
784 struct list_head *pt_hash; /* NID->peer hash */
785 struct list_head pt_peer_list; /* peers */
786 int pt_peers; /* # peers */
787 struct list_head pt_zombie_list; /* zombie peer_ni */
788 int pt_zombies; /* # zombie peers_ni */
789 spinlock_t pt_zombie_lock; /* protect list and count */
792 /* peer aliveness is enabled only on routers for peers in a network where the
793 * struct lnet_ni::ni_peertimeout has been set to a positive value
795 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
796 ((lp)->lpni_net) && \
797 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
800 struct list_head lr_list; /* chain on net */
801 struct list_head lr_gwlist; /* chain on gateway */
802 struct lnet_peer_ni *lr_gateway; /* router node */
803 __u32 lr_net; /* remote network number */
804 int lr_seq; /* sequence for round-robin */
805 unsigned int lr_downis; /* number of down NIs */
806 __u32 lr_hops; /* how far I am */
807 unsigned int lr_priority; /* route priority */
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 lnet_kiov_t 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 = (1 << 0),
859 LNET_MATCHMD_OK = (1 << 1),
860 /* Must be discarded */
861 LNET_MATCHMD_DROP = (1 << 2),
862 /* match and buffer is exhausted */
863 LNET_MATCHMD_EXHAUSTED = (1 << 3),
865 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
868 /* Options for struct lnet_portal::ptl_options */
869 #define LNET_PTL_LAZY (1 << 0)
870 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
871 #define LNET_PTL_MATCH_WILDCARD (1 << 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 struct list_head msc_active; /* active message list */
960 /* threads doing finalization */
961 void **msc_finalizers;
964 /* Peer Discovery states */
965 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
966 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
967 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
969 /* Router Checker states */
970 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
971 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
972 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
975 #define LNET_STATE_SHUTDOWN 0 /* not started */
976 #define LNET_STATE_RUNNING 1 /* started up OK */
977 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
980 /* CPU partition table of LNet */
981 struct cfs_cpt_table *ln_cpt_table;
982 /* number of CPTs in ln_cpt_table */
983 unsigned int ln_cpt_number;
984 unsigned int ln_cpt_bits;
986 /* protect LNet resources (ME/MD/EQ) */
987 struct cfs_percpt_lock *ln_res_lock;
990 /* the vector of portals */
991 struct lnet_portal **ln_portals;
992 /* percpt ME containers */
993 struct lnet_res_container **ln_me_containers;
994 /* percpt MD container */
995 struct lnet_res_container **ln_md_containers;
997 /* Event Queue container */
998 struct lnet_res_container ln_eq_container;
999 wait_queue_head_t ln_eq_waitq;
1000 spinlock_t ln_eq_wait_lock;
1002 unsigned int ln_remote_nets_hbits;
1004 /* protect NI, peer table, credits, routers, rtrbuf... */
1005 struct cfs_percpt_lock *ln_net_lock;
1006 /* percpt message containers for active/finalizing/freed message */
1007 struct lnet_msg_container **ln_msg_containers;
1008 struct lnet_counters **ln_counters;
1009 struct lnet_peer_table **ln_peer_tables;
1010 /* list of peer nis not on a local network */
1011 struct list_head ln_remote_peer_ni_list;
1012 /* failure simulation */
1013 struct list_head ln_test_peers;
1014 struct list_head ln_drop_rules;
1015 struct list_head ln_delay_rules;
1017 struct list_head ln_nets;
1018 /* the loopback NI */
1019 struct lnet_ni *ln_loni;
1020 /* network zombie list */
1021 struct list_head ln_net_zombie;
1022 /* resend messages list */
1023 struct list_head ln_msg_resend;
1024 /* spin lock to protect the msg resend list */
1025 spinlock_t ln_msg_resend_lock;
1027 /* remote networks with routes to them */
1028 struct list_head *ln_remote_nets_hash;
1029 /* validity stamp */
1030 __u64 ln_remote_nets_version;
1031 /* list of all known routers */
1032 struct list_head ln_routers;
1033 /* validity stamp */
1034 __u64 ln_routers_version;
1035 /* percpt router buffer pools */
1036 struct lnet_rtrbufpool **ln_rtrpools;
1039 * Ping target / Push source
1041 * The ping target and push source share a single buffer. The
1042 * ln_ping_target is protected against concurrent updates by
1045 struct lnet_handle_md ln_ping_target_md;
1046 struct lnet_handle_eq ln_ping_target_eq;
1047 struct lnet_ping_buffer *ln_ping_target;
1048 atomic_t ln_ping_target_seqno;
1053 * ln_push_nnis contains the desired size of the push target.
1054 * The lnet_net_lock is used to handle update races. The old
1055 * buffer may linger a while after it has been unlinked, in
1056 * which case the event handler cleans up.
1058 struct lnet_handle_eq ln_push_target_eq;
1059 struct lnet_handle_md ln_push_target_md;
1060 struct lnet_ping_buffer *ln_push_target;
1061 int ln_push_target_nnis;
1063 /* discovery event queue handle */
1064 struct lnet_handle_eq ln_dc_eqh;
1065 /* discovery requests */
1066 struct list_head ln_dc_request;
1067 /* discovery working list */
1068 struct list_head ln_dc_working;
1069 /* discovery expired list */
1070 struct list_head ln_dc_expired;
1071 /* discovery thread wait queue */
1072 wait_queue_head_t ln_dc_waitq;
1073 /* discovery startup/shutdown state */
1076 /* monitor thread startup/shutdown state */
1078 /* router checker's event queue */
1079 struct lnet_handle_eq ln_rc_eqh;
1080 /* rcd still pending on net */
1081 struct list_head ln_rcd_deathrow;
1082 /* rcd ready for free */
1083 struct list_head ln_rcd_zombie;
1084 /* serialise startup/shutdown */
1085 struct semaphore ln_mt_signal;
1087 struct mutex ln_api_mutex;
1088 struct mutex ln_lnd_mutex;
1089 /* Have I called LNetNIInit myself? */
1091 /* LNetNIInit/LNetNIFini counter */
1093 /* SHUTDOWN/RUNNING/STOPPING */
1096 int ln_routing; /* am I a router? */
1097 lnet_pid_t ln_pid; /* requested pid */
1098 /* uniquely identifies this ni in this epoch */
1099 __u64 ln_interface_cookie;
1100 /* registered LNDs */
1101 struct list_head ln_lnds;
1103 /* test protocol compatibility flags */
1104 int ln_testprotocompat;
1106 /* 0 - load the NIs from the mod params
1107 * 1 - do not load the NIs from the mod params
1108 * Reverse logic to ensure that other calls to LNetNIInit
1111 bool ln_nis_from_mod_params;
1114 * waitq for the monitor thread. The monitor thread takes care of
1115 * checking routes, timedout messages and resending messages.
1117 wait_queue_head_t ln_mt_waitq;
1119 /* per-cpt resend queues */
1120 struct list_head **ln_mt_resendqs;
1121 /* local NIs to recover */
1122 struct list_head ln_mt_localNIRecovq;
1123 /* local NIs to recover */
1124 struct list_head ln_mt_peerNIRecovq;
1126 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1127 * There are CPT number of queues. Since response trackers will be
1128 * added on the fast path we can't afford to grab the exclusive
1129 * net lock to protect these queues. The CPT will be calculated
1130 * based on the mdh cookie.
1132 struct list_head **ln_mt_rstq;
1133 /* recovery eq handler */
1134 struct lnet_handle_eq ln_mt_eqh;