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;
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 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;
392 enum lnet_net_state net_state;
396 /* chain on the lnet_net structure */
397 struct list_head ni_netlist;
399 /* chain on net_ni_cpt */
400 struct list_head ni_cptlist;
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 /* when I was last alive */
432 time64_t ni_last_alive;
434 /* pointer to parent network */
435 struct lnet_net *ni_net;
437 /* my health status */
438 struct lnet_ni_status *ni_status;
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;
452 /* physical device CPT */
455 /* sequence number used to round robin over nis within a net */
460 * initialized to LNET_MAX_HEALTH_VALUE
461 * Value is decremented every time we fail to send a message over
462 * this NI because of a NI specific failure.
463 * Value is incremented if we successfully send a message.
468 * equivalent interfaces to use
469 * This is an array because socklnd bonding can still be configured
471 char *ni_interfaces[LNET_INTERFACES_NUM];
472 struct net *ni_net_ns; /* original net namespace */
475 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
478 * Descriptor of a ping info buffer: keep a separate indicator of the
479 * size and a reference count. The type is used both as a source and
480 * sink of data, so we need to keep some information outside of the
481 * area that may be overwritten by network data.
483 struct lnet_ping_buffer {
486 struct lnet_ping_info pb_info;
489 #define LNET_PING_BUFFER_SIZE(NNIDS) \
490 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
491 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
492 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
494 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
495 container_of((PINFO), struct lnet_ping_buffer, pb_info)
497 /* router checker data, per router */
498 struct lnet_rc_data {
499 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
500 struct list_head rcd_list;
501 struct lnet_handle_md rcd_mdh; /* ping buffer MD */
502 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
503 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
504 int rcd_nnis; /* desired size of buffer */
507 struct lnet_peer_ni {
508 /* chain on lpn_peer_nis */
509 struct list_head lpni_peer_nis;
510 /* chain on remote peer list */
511 struct list_head lpni_on_remote_peer_ni_list;
512 /* chain on recovery queue */
513 struct list_head lpni_recovery;
514 /* chain on peer hash */
515 struct list_head lpni_hashlist;
516 /* messages blocking for tx credits */
517 struct list_head lpni_txq;
518 /* messages blocking for router credits */
519 struct list_head lpni_rtrq;
520 /* chain on router list */
521 struct list_head lpni_rtr_list;
522 /* pointer to peer net I'm part of */
523 struct lnet_peer_net *lpni_peer_net;
524 /* statistics kept on each peer NI */
525 struct lnet_element_stats lpni_stats;
526 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
527 spinlock_t lpni_lock;
528 /* # tx credits available */
531 int lpni_mintxcredits;
532 /* # router credits */
535 int lpni_minrtrcredits;
536 /* bytes queued for sending */
540 /* notification outstanding? */
542 /* outstanding notification for LND? */
544 /* some thread is handling notification */
546 /* SEND event outstanding from ping */
547 bool lpni_ping_notsent;
548 /* # times router went dead<->alive. Protected with lpni_lock */
549 int lpni_alive_count;
550 /* time of last aliveness news */
551 time64_t lpni_timestamp;
552 /* time of last ping attempt */
553 time64_t lpni_ping_timestamp;
554 /* != 0 if ping reply expected */
555 time64_t lpni_ping_deadline;
556 /* when I was last alive */
557 time64_t lpni_last_alive;
558 /* when lpni_ni was queried last time */
559 time64_t lpni_last_query;
560 /* network peer is on */
561 struct lnet_net *lpni_net;
565 atomic_t lpni_refcount;
566 /* health value for the peer */
567 atomic_t lpni_healthv;
568 /* recovery ping mdh */
569 struct lnet_handle_md lpni_recovery_ping_mdh;
570 /* CPT this peer attached on */
572 /* state flags -- protected by lpni_lock */
574 /* # refs from lnet_route_t::lr_gateway */
575 int lpni_rtr_refcount;
576 /* sequence number used to round robin over peer nis within a net */
578 /* sequence number used to round robin over gateways */
582 /* returned RC ping features. Protected with lpni_lock */
583 unsigned int lpni_ping_feats;
584 /* routes on this peer */
585 struct list_head lpni_routes;
586 /* preferred local nids: if only one, use lpni_pref.nid */
591 /* number of preferred NIDs in lnpi_pref_nids */
592 __u32 lpni_pref_nnids;
593 /* router checker state */
594 struct lnet_rc_data *lpni_rcd;
597 /* Preferred path added due to traffic on non-MR peer_ni */
598 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
599 /* peer is being recovered. */
600 #define LNET_PEER_NI_RECOVERY_PENDING (1 << 1)
601 /* peer is being deleted */
602 #define LNET_PEER_NI_DELETING (1 << 2)
605 /* chain on pt_peer_list */
606 struct list_head lp_peer_list;
608 /* list of peer nets */
609 struct list_head lp_peer_nets;
611 /* list of messages pending discovery*/
612 struct list_head lp_dc_pendq;
614 /* primary NID of the peer */
615 lnet_nid_t lp_primary_nid;
617 /* CPT of peer_table */
620 /* number of NIDs on this peer */
623 /* reference count */
624 atomic_t lp_refcount;
626 /* lock protecting peer state flags */
629 /* peer state flags */
632 /* buffer for data pushed by peer */
633 struct lnet_ping_buffer *lp_data;
635 /* MD handle for ping in progress */
636 struct lnet_handle_md lp_ping_mdh;
638 /* MD handle for push in progress */
639 struct lnet_handle_md lp_push_mdh;
641 /* number of NIDs for sizing push data */
644 /* NI config sequence number of peer */
647 /* Local NI config sequence number acked by peer */
650 /* Local NI config sequence number sent to peer */
651 __u32 lp_node_seqno_sent;
653 /* Ping error encountered during discovery. */
656 /* Push error encountered during discovery. */
659 /* Error encountered during discovery. */
662 /* time it was put on the ln_dc_working queue */
663 time64_t lp_last_queued;
665 /* link on discovery-related lists */
666 struct list_head lp_dc_list;
668 /* tasks waiting on discovery of this peer */
669 wait_queue_head_t lp_dc_waitq;
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 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
683 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
685 * A peer is marked CONFIGURED if it was configured by DLC.
687 * In addition, a peer is marked DISCOVERED if it has fully passed
688 * through Peer Discovery.
690 * When Peer Discovery is disabled, the discovery thread will mark
691 * peers REDISCOVER to indicate that they should be re-examined if
692 * discovery is (re)enabled on the node.
694 * A peer that was created as the result of inbound traffic will not
697 #define LNET_PEER_CONFIGURED (1 << 2) /* Configured via DLC */
698 #define LNET_PEER_DISCOVERED (1 << 3) /* Peer was discovered */
699 #define LNET_PEER_REDISCOVER (1 << 4) /* Discovery was disabled */
701 * A peer is marked DISCOVERING when discovery is in progress.
702 * The other flags below correspond to stages of discovery.
704 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
705 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
706 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
707 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
708 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
709 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
710 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
712 * A ping can be forced as a way to fix up state, or as a manual
713 * intervention by an admin.
714 * A push can be forced in circumstances that would normally not
715 * allow for one to happen.
717 #define LNET_PEER_FORCE_PING (1 << 12) /* Forced Ping */
718 #define LNET_PEER_FORCE_PUSH (1 << 13) /* Forced Push */
720 struct lnet_peer_net {
721 /* chain on lp_peer_nets */
722 struct list_head lpn_peer_nets;
724 /* list of peer_nis on this network */
725 struct list_head lpn_peer_nis;
727 /* pointer to the peer I'm part of */
728 struct lnet_peer *lpn_peer;
733 /* reference count */
734 atomic_t lpn_refcount;
738 #define LNET_PEER_HASH_BITS 9
739 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
742 * peer hash table - one per CPT
744 * protected by lnet_net_lock/EX for update
750 * protected by pt_zombie_lock:
754 * pt_zombie lock nests inside lnet_net_lock
756 struct lnet_peer_table {
757 int pt_version; /* /proc validity stamp */
758 int pt_number; /* # peers_ni extant */
759 struct list_head *pt_hash; /* NID->peer hash */
760 struct list_head pt_peer_list; /* peers */
761 int pt_peers; /* # peers */
762 struct list_head pt_zombie_list; /* zombie peer_ni */
763 int pt_zombies; /* # zombie peers_ni */
764 spinlock_t pt_zombie_lock; /* protect list and count */
767 /* peer aliveness is enabled only on routers for peers in a network where the
768 * struct lnet_ni::ni_peertimeout has been set to a positive value
770 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
771 ((lp)->lpni_net) && \
772 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
775 struct list_head lr_list; /* chain on net */
776 struct list_head lr_gwlist; /* chain on gateway */
777 struct lnet_peer_ni *lr_gateway; /* router node */
778 __u32 lr_net; /* remote network number */
779 int lr_seq; /* sequence for round-robin */
780 unsigned int lr_downis; /* number of down NIs */
781 __u32 lr_hops; /* how far I am */
782 unsigned int lr_priority; /* route priority */
785 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
786 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
787 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
789 struct lnet_remotenet {
790 /* chain on ln_remote_nets_hash */
791 struct list_head lrn_list;
793 struct list_head lrn_routes;
798 /** lnet message has credit and can be submitted to lnd for send/receive */
799 #define LNET_CREDIT_OK 0
800 /** lnet message is waiting for credit */
801 #define LNET_CREDIT_WAIT 1
802 /** lnet message is waiting for discovery */
803 #define LNET_DC_WAIT 2
805 struct lnet_rtrbufpool {
806 /* my free buffer pool */
807 struct list_head rbp_bufs;
808 /* messages blocking for a buffer */
809 struct list_head rbp_msgs;
810 /* # pages in each buffer */
812 /* requested number of buffers */
813 int rbp_req_nbuffers;
814 /* # buffers actually allocated */
816 /* # free buffers / blocked messages */
823 struct list_head rb_list; /* chain on rbp_bufs */
824 struct lnet_rtrbufpool *rb_pool; /* owning pool */
825 lnet_kiov_t rb_kiov[0]; /* the buffer space */
828 #define LNET_PEER_HASHSIZE 503 /* prime! */
830 enum lnet_match_flags {
831 /* Didn't match anything */
832 LNET_MATCHMD_NONE = (1 << 0),
834 LNET_MATCHMD_OK = (1 << 1),
835 /* Must be discarded */
836 LNET_MATCHMD_DROP = (1 << 2),
837 /* match and buffer is exhausted */
838 LNET_MATCHMD_EXHAUSTED = (1 << 3),
840 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
843 /* Options for struct lnet_portal::ptl_options */
844 #define LNET_PTL_LAZY (1 << 0)
845 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
846 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
848 /* parameter for matching operations (GET, PUT) */
849 struct lnet_match_info {
851 struct lnet_process_id mi_id;
854 unsigned int mi_portal;
855 unsigned int mi_rlength;
856 unsigned int mi_roffset;
859 /* ME hash of RDMA portal */
860 #define LNET_MT_HASH_BITS 8
861 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
862 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
863 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
864 * the last entry is reserved for MEs with ignore-bits */
865 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
866 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
867 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
868 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
869 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
870 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
871 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
873 /* portal match table */
874 struct lnet_match_table {
875 /* reserved for upcoming patches, CPU partition ID */
877 unsigned int mt_portal; /* portal index */
878 /* match table is set as "enabled" if there's non-exhausted MD
879 * attached on mt_mhash, it's only valid for wildcard portal */
880 unsigned int mt_enabled;
881 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
882 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
883 struct list_head *mt_mhash; /* matching hash */
886 /* these are only useful for wildcard portal */
887 /* Turn off message rotor for wildcard portals */
888 #define LNET_PTL_ROTOR_OFF 0
889 /* round-robin dispatch all PUT messages for wildcard portals */
890 #define LNET_PTL_ROTOR_ON 1
891 /* round-robin dispatch routed PUT message for wildcard portals */
892 #define LNET_PTL_ROTOR_RR_RT 2
893 /* dispatch routed PUT message by hashing source NID for wildcard portals */
894 #define LNET_PTL_ROTOR_HASH_RT 3
898 unsigned int ptl_index; /* portal ID, reserved */
899 /* flags on this portal: lazy, unique... */
900 unsigned int ptl_options;
901 /* list of messages which are stealing buffer */
902 struct list_head ptl_msg_stealing;
903 /* messages blocking for MD */
904 struct list_head ptl_msg_delayed;
905 /* Match table for each CPT */
906 struct lnet_match_table **ptl_mtables;
907 /* spread rotor of incoming "PUT" */
908 unsigned int ptl_rotor;
909 /* # active entries for this portal */
911 /* array of active entries' cpu-partition-id */
915 #define LNET_LH_HASH_BITS 12
916 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
917 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
919 /* resource container (ME, MD, EQ) */
920 struct lnet_res_container {
921 unsigned int rec_type; /* container type */
922 __u64 rec_lh_cookie; /* cookie generator */
923 struct list_head rec_active; /* active resource list */
924 struct list_head *rec_lh_hash; /* handle hash */
927 /* message container */
928 struct lnet_msg_container {
929 int msc_init; /* initialized or not */
930 /* max # threads finalizing */
932 /* msgs waiting to complete finalizing */
933 struct list_head msc_finalizing;
934 struct list_head msc_active; /* active message list */
935 /* threads doing finalization */
936 void **msc_finalizers;
939 /* Peer Discovery states */
940 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
941 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
942 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
944 /* Router Checker states */
945 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
946 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
947 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
950 #define LNET_STATE_SHUTDOWN 0 /* not started */
951 #define LNET_STATE_RUNNING 1 /* started up OK */
952 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
955 /* CPU partition table of LNet */
956 struct cfs_cpt_table *ln_cpt_table;
957 /* number of CPTs in ln_cpt_table */
958 unsigned int ln_cpt_number;
959 unsigned int ln_cpt_bits;
961 /* protect LNet resources (ME/MD/EQ) */
962 struct cfs_percpt_lock *ln_res_lock;
965 /* the vector of portals */
966 struct lnet_portal **ln_portals;
967 /* percpt ME containers */
968 struct lnet_res_container **ln_me_containers;
969 /* percpt MD container */
970 struct lnet_res_container **ln_md_containers;
972 /* Event Queue container */
973 struct lnet_res_container ln_eq_container;
974 wait_queue_head_t ln_eq_waitq;
975 spinlock_t ln_eq_wait_lock;
977 unsigned int ln_remote_nets_hbits;
979 /* protect NI, peer table, credits, routers, rtrbuf... */
980 struct cfs_percpt_lock *ln_net_lock;
981 /* percpt message containers for active/finalizing/freed message */
982 struct lnet_msg_container **ln_msg_containers;
983 struct lnet_counters **ln_counters;
984 struct lnet_peer_table **ln_peer_tables;
985 /* list of peer nis not on a local network */
986 struct list_head ln_remote_peer_ni_list;
987 /* failure simulation */
988 struct list_head ln_test_peers;
989 struct list_head ln_drop_rules;
990 struct list_head ln_delay_rules;
992 struct list_head ln_nets;
993 /* the loopback NI */
994 struct lnet_ni *ln_loni;
995 /* network zombie list */
996 struct list_head ln_net_zombie;
997 /* resend messages list */
998 struct list_head ln_msg_resend;
999 /* spin lock to protect the msg resend list */
1000 spinlock_t ln_msg_resend_lock;
1002 /* remote networks with routes to them */
1003 struct list_head *ln_remote_nets_hash;
1004 /* validity stamp */
1005 __u64 ln_remote_nets_version;
1006 /* list of all known routers */
1007 struct list_head ln_routers;
1008 /* validity stamp */
1009 __u64 ln_routers_version;
1010 /* percpt router buffer pools */
1011 struct lnet_rtrbufpool **ln_rtrpools;
1014 * Ping target / Push source
1016 * The ping target and push source share a single buffer. The
1017 * ln_ping_target is protected against concurrent updates by
1020 struct lnet_handle_md ln_ping_target_md;
1021 struct lnet_handle_eq ln_ping_target_eq;
1022 struct lnet_ping_buffer *ln_ping_target;
1023 atomic_t ln_ping_target_seqno;
1028 * ln_push_nnis contains the desired size of the push target.
1029 * The lnet_net_lock is used to handle update races. The old
1030 * buffer may linger a while after it has been unlinked, in
1031 * which case the event handler cleans up.
1033 struct lnet_handle_eq ln_push_target_eq;
1034 struct lnet_handle_md ln_push_target_md;
1035 struct lnet_ping_buffer *ln_push_target;
1036 int ln_push_target_nnis;
1038 /* discovery event queue handle */
1039 struct lnet_handle_eq ln_dc_eqh;
1040 /* discovery requests */
1041 struct list_head ln_dc_request;
1042 /* discovery working list */
1043 struct list_head ln_dc_working;
1044 /* discovery expired list */
1045 struct list_head ln_dc_expired;
1046 /* discovery thread wait queue */
1047 wait_queue_head_t ln_dc_waitq;
1048 /* discovery startup/shutdown state */
1051 /* monitor thread startup/shutdown state */
1053 /* router checker's event queue */
1054 struct lnet_handle_eq ln_rc_eqh;
1055 /* rcd still pending on net */
1056 struct list_head ln_rcd_deathrow;
1057 /* rcd ready for free */
1058 struct list_head ln_rcd_zombie;
1059 /* serialise startup/shutdown */
1060 struct semaphore ln_mt_signal;
1062 struct mutex ln_api_mutex;
1063 struct mutex ln_lnd_mutex;
1064 /* Have I called LNetNIInit myself? */
1066 /* LNetNIInit/LNetNIFini counter */
1068 /* SHUTDOWN/RUNNING/STOPPING */
1071 int ln_routing; /* am I a router? */
1072 lnet_pid_t ln_pid; /* requested pid */
1073 /* uniquely identifies this ni in this epoch */
1074 __u64 ln_interface_cookie;
1075 /* registered LNDs */
1076 struct list_head ln_lnds;
1078 /* test protocol compatibility flags */
1079 int ln_testprotocompat;
1081 /* 0 - load the NIs from the mod params
1082 * 1 - do not load the NIs from the mod params
1083 * Reverse logic to ensure that other calls to LNetNIInit
1086 bool ln_nis_from_mod_params;
1089 * waitq for the monitor thread. The monitor thread takes care of
1090 * checking routes, timedout messages and resending messages.
1092 wait_queue_head_t ln_mt_waitq;
1094 /* per-cpt resend queues */
1095 struct list_head **ln_mt_resendqs;
1096 /* local NIs to recover */
1097 struct list_head ln_mt_localNIRecovq;
1098 /* local NIs to recover */
1099 struct list_head ln_mt_peerNIRecovq;
1101 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1102 * There are CPT number of queues. Since response trackers will be
1103 * added on the fast path we can't afford to grab the exclusive
1104 * net lock to protect these queues. The CPT will be calculated
1105 * based on the mdh cookie.
1107 struct list_head **ln_mt_rstq;
1108 /* recovery eq handler */
1109 struct lnet_handle_eq ln_mt_eqh;