4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lnet/include/lnet/lib-types.h
34 * Types used by the library side routines that do not need to be
35 * exposed to the user application
38 #ifndef __LNET_LIB_TYPES_H__
39 #define __LNET_LIB_TYPES_H__
42 # error This include is only for kernel use.
45 #include <linux/kthread.h>
46 #include <linux/uio.h>
47 #include <linux/semaphore.h>
48 #include <linux/types.h>
50 #include <uapi/linux/lnet/lnet-dlc.h>
51 #include <uapi/linux/lnet/lnetctl.h>
52 #include <uapi/linux/lnet/nidstr.h>
54 /* Max payload size */
55 #define LNET_MAX_PAYLOAD LNET_MTU
57 /** limit on the number of fragments in discontiguous MDs */
58 #define LNET_MAX_IOV 256
61 * This is the maximum health value.
62 * All local and peer NIs created have their health default to this value.
64 #define LNET_MAX_HEALTH_VALUE 1000
65 #define LNET_MAX_SELECTION_PRIORITY UINT_MAX
70 enum lnet_msg_hstatus {
71 LNET_MSG_STATUS_OK = 0,
72 LNET_MSG_STATUS_LOCAL_INTERRUPT,
73 LNET_MSG_STATUS_LOCAL_DROPPED,
74 LNET_MSG_STATUS_LOCAL_ABORTED,
75 LNET_MSG_STATUS_LOCAL_NO_ROUTE,
76 LNET_MSG_STATUS_LOCAL_ERROR,
77 LNET_MSG_STATUS_LOCAL_TIMEOUT,
78 LNET_MSG_STATUS_REMOTE_ERROR,
79 LNET_MSG_STATUS_REMOTE_DROPPED,
80 LNET_MSG_STATUS_REMOTE_TIMEOUT,
81 LNET_MSG_STATUS_NETWORK_TIMEOUT,
85 struct lnet_rsp_tracker {
86 /* chain on the waiting list */
87 struct list_head rspt_on_list;
91 lnet_nid_t rspt_next_hop_nid;
92 /* deadline of the REPLY/ACK */
93 ktime_t rspt_deadline;
95 struct lnet_handle_md rspt_mdh;
99 struct list_head msg_activelist;
100 struct list_head msg_list; /* Q for credits/MD */
102 struct lnet_process_id msg_target;
103 /* Primary NID of the source. */
104 lnet_nid_t msg_initiator;
105 /* where is it from, it's only for building event */
110 * hold parameters in case message is with held due
113 lnet_nid_t msg_src_nid_param;
114 lnet_nid_t msg_rtr_nid_param;
117 * Deadline for the message after which it will be finalized if it
120 ktime_t msg_deadline;
122 /* The message health status. */
123 enum lnet_msg_hstatus msg_health_status;
124 /* This is a recovery message */
126 /* the number of times a transmission has been retried */
128 /* flag to indicate that we do not want to resend this message */
131 /* committed for sending */
132 unsigned int msg_tx_committed:1;
133 /* CPT # this message committed for sending */
134 unsigned int msg_tx_cpt:15;
135 /* committed for receiving */
136 unsigned int msg_rx_committed:1;
137 /* CPT # this message committed for receiving */
138 unsigned int msg_rx_cpt:15;
139 /* queued for tx credit */
140 unsigned int msg_tx_delayed:1;
141 /* queued for RX buffer */
142 unsigned int msg_rx_delayed:1;
143 /* ready for pending on RX delay list */
144 unsigned int msg_rx_ready_delay:1;
146 unsigned int msg_vmflush:1; /* VM trying to free memory */
147 unsigned int msg_target_is_router:1; /* sending to a router */
148 unsigned int msg_routing:1; /* being forwarded */
149 unsigned int msg_ack:1; /* ack on finalize (PUT) */
150 unsigned int msg_sending:1; /* outgoing message */
151 unsigned int msg_receiving:1; /* being received */
152 unsigned int msg_txcredit:1; /* taken an NI send credit */
153 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
154 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
155 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
156 unsigned int msg_onactivelist:1; /* on the activelist */
157 unsigned int msg_rdma_get:1;
159 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
160 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
163 struct lnet_libmd *msg_md;
164 /* the NI the message was sent or received over */
165 struct lnet_ni *msg_txni;
166 struct lnet_ni *msg_rxni;
168 unsigned int msg_len;
169 unsigned int msg_wanted;
170 unsigned int msg_offset;
171 unsigned int msg_niov;
172 struct bio_vec *msg_kiov;
174 struct lnet_event msg_ev;
175 struct lnet_hdr msg_hdr;
178 struct lnet_libhandle {
179 struct list_head lh_hash_chain;
183 #define lh_entry(ptr, type, member) \
184 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
187 struct list_head me_list;
189 struct lnet_process_id me_match_id;
190 unsigned int me_portal;
191 unsigned int me_pos; /* hash offset in mt_hash */
193 __u64 me_ignore_bits;
194 enum lnet_unlink me_unlink;
195 struct lnet_libmd *me_md;
199 struct list_head md_list;
200 struct lnet_libhandle md_lh;
201 struct lnet_me *md_me;
203 unsigned int md_offset;
204 unsigned int md_length;
205 unsigned int md_max_size;
208 unsigned int md_options;
209 unsigned int md_flags;
210 unsigned int md_niov; /* # frags at end of struct */
212 struct lnet_rsp_tracker *md_rspt_ptr;
213 lnet_handler_t md_handler;
214 struct lnet_handle_md md_bulk_handle;
215 struct bio_vec md_kiov[LNET_MAX_IOV];
218 #define LNET_MD_FLAG_ZOMBIE BIT(0)
219 #define LNET_MD_FLAG_AUTO_UNLINK BIT(1)
220 #define LNET_MD_FLAG_ABORTED BIT(2)
221 /* LNET_MD_FLAG_HANDLING is set when a non-unlink event handler
222 * is being called for an event relating to the md.
223 * It ensures only one such handler runs at a time.
224 * The final "unlink" event is only called once the
225 * md_refcount has reached zero, and this flag has been cleared,
226 * ensuring that it doesn't race with any other event handler
229 #define LNET_MD_FLAG_HANDLING BIT(3)
230 #define LNET_MD_FLAG_DISCARD BIT(4)
232 struct lnet_test_peer {
233 /* info about peers we are trying to fail */
234 struct list_head tp_list; /* ln_test_peers */
235 lnet_nid_t tp_nid; /* matching nid */
236 unsigned int tp_threshold; /* # failures to simulate */
239 #define LNET_COOKIE_TYPE_MD 1
240 #define LNET_COOKIE_TYPE_ME 2
241 #define LNET_COOKIE_TYPE_EQ 3
242 #define LNET_COOKIE_TYPE_BITS 2
243 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
245 struct lnet_ni; /* forward ref */
249 /* fields initialized by the LND */
252 int (*lnd_startup)(struct lnet_ni *ni);
253 void (*lnd_shutdown)(struct lnet_ni *ni);
254 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
256 /* In data movement APIs below, payload buffers are described as a set
257 * of 'niov' fragments which are in pages.
258 * The LND may NOT overwrite these fragment descriptors.
259 * An 'offset' and may specify a byte offset within the set of
260 * fragments to start from
263 /* Start sending a preformatted message. 'private' is NULL for PUT and
264 * GET messages; otherwise this is a response to an incoming message
265 * and 'private' is the 'private' passed to lnet_parse(). Return
266 * non-zero for immediate failure, otherwise complete later with
268 int (*lnd_send)(struct lnet_ni *ni, void *private,
269 struct lnet_msg *msg);
271 /* Start receiving 'mlen' bytes of payload data, skipping the following
272 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
273 * lnet_parse(). Return non-zero for immedaite failure, otherwise
274 * complete later with lnet_finalize(). This also gives back a receive
275 * credit if the LND does flow control. */
276 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
277 int delayed, unsigned int niov,
278 struct bio_vec *kiov,
279 unsigned int offset, unsigned int mlen, unsigned int rlen);
281 /* lnet_parse() has had to delay processing of this message
282 * (e.g. waiting for a forwarding buffer or send credits). Give the
283 * LND a chance to free urgently needed resources. If called, return 0
284 * for success and do NOT give back a receive credit; that has to wait
285 * until lnd_recv() gets called. On failure return < 0 and
286 * release resources; lnd_recv() will not be called. */
287 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
288 struct lnet_msg *msg, void **new_privatep);
290 /* notification of peer down */
291 void (*lnd_notify_peer_down)(lnet_nid_t peer);
293 /* accept a new connection */
294 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
297 struct lnet_tx_queue {
298 int tq_credits; /* # tx credits free */
299 int tq_credits_min; /* lowest it's been */
300 int tq_credits_max; /* total # tx credits */
301 struct list_head tq_delayed; /* delayed TXs */
304 enum lnet_net_state {
305 /* set when net block is allocated */
306 LNET_NET_STATE_INIT = 0,
307 /* set when NIs in net are started successfully */
308 LNET_NET_STATE_ACTIVE,
309 /* set if all NIs in net are in FAILED state */
310 LNET_NET_STATE_INACTIVE,
311 /* set when shutting down a NET */
312 LNET_NET_STATE_DELETING
316 /* initial state when NI is created */
317 LNET_NI_STATE_INIT = 0,
318 /* set when NI is brought up */
319 LNET_NI_STATE_ACTIVE,
320 /* set when NI is being shutdown */
321 LNET_NI_STATE_DELETING,
324 #define LNET_NI_RECOVERY_PENDING BIT(0)
325 #define LNET_NI_RECOVERY_FAILED BIT(1)
327 enum lnet_stats_type {
328 LNET_STATS_TYPE_SEND = 0,
329 LNET_STATS_TYPE_RECV,
333 struct lnet_comm_count {
334 atomic_t co_get_count;
335 atomic_t co_put_count;
336 atomic_t co_reply_count;
337 atomic_t co_ack_count;
338 atomic_t co_hello_count;
341 struct lnet_element_stats {
342 struct lnet_comm_count el_send_stats;
343 struct lnet_comm_count el_recv_stats;
344 struct lnet_comm_count el_drop_stats;
347 struct lnet_health_local_stats {
348 atomic_t hlt_local_interrupt;
349 atomic_t hlt_local_dropped;
350 atomic_t hlt_local_aborted;
351 atomic_t hlt_local_no_route;
352 atomic_t hlt_local_timeout;
353 atomic_t hlt_local_error;
356 struct lnet_health_remote_stats {
357 atomic_t hlt_remote_dropped;
358 atomic_t hlt_remote_timeout;
359 atomic_t hlt_remote_error;
360 atomic_t hlt_network_timeout;
364 /* chain on the ln_nets */
365 struct list_head net_list;
367 /* net ID, which is composed of
368 * (net_type << 16) | net_num.
369 * net_type can be one of the enumerated types defined in
370 * lnet/include/lnet/nidstr.h */
373 /* total number of CPTs in the array */
376 /* cumulative CPTs of all NIs in this net */
379 /* relative net selection priority */
380 __u32 net_sel_priority;
382 /* network tunables */
383 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
386 * boolean to indicate that the tunables have been set and
389 bool net_tunables_set;
391 /* procedural interface */
392 const struct lnet_lnd *net_lnd;
394 /* list of NIs on this net */
395 struct list_head net_ni_list;
397 /* list of NIs being added, but not started yet */
398 struct list_head net_ni_added;
400 /* dying LND instances */
401 struct list_head net_ni_zombie;
403 /* when I was last alive */
404 time64_t net_last_alive;
406 /* protects access to net_last_alive */
409 /* list of router nids preferred for this network */
410 struct list_head net_rtr_pref_nids;
414 /* chain on the lnet_net structure */
415 struct list_head ni_netlist;
417 /* chain on the recovery queue */
418 struct list_head ni_recovery;
420 /* MD handle for recovery ping */
421 struct lnet_handle_md ni_ping_mdh;
428 /* bond NI on some CPTs */
431 /* interface's NID */
434 /* instance-specific data */
438 atomic_t ni_tx_credits;
440 /* percpt TX queues */
441 struct lnet_tx_queue **ni_tx_queues;
443 /* percpt reference count */
446 /* pointer to parent network */
447 struct lnet_net *ni_net;
449 /* my health status */
450 struct lnet_ni_status *ni_status;
452 /* NI FSM. Protected by lnet_ni_lock() */
453 enum lnet_ni_state ni_state;
455 /* Recovery state. Protected by lnet_ni_lock() */
456 __u32 ni_recovery_state;
458 /* per NI LND tunables */
459 struct lnet_lnd_tunables ni_lnd_tunables;
461 /* lnd tunables set explicitly */
462 bool ni_lnd_tunables_set;
465 struct lnet_element_stats ni_stats;
466 struct lnet_health_local_stats ni_hstats;
468 /* physical device CPT */
471 /* sequence number used to round robin over nis within a net */
476 * initialized to LNET_MAX_HEALTH_VALUE
477 * Value is decremented every time we fail to send a message over
478 * this NI because of a NI specific failure.
479 * Value is incremented if we successfully send a message.
484 * Set to 1 by the LND when it receives an event telling it the device
485 * has gone into a fatal state. Set to 0 when the LND receives an
486 * even telling it the device is back online.
488 atomic_t ni_fatal_error_on;
490 /* the relative selection priority of this NI */
491 __u32 ni_sel_priority;
494 * equivalent interfaces to use
495 * This is an array because socklnd bonding can still be configured
497 char *ni_interfaces[LNET_INTERFACES_NUM];
498 struct net *ni_net_ns; /* original net namespace */
501 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
504 * Descriptor of a ping info buffer: keep a separate indicator of the
505 * size and a reference count. The type is used both as a source and
506 * sink of data, so we need to keep some information outside of the
507 * area that may be overwritten by network data.
509 struct lnet_ping_buffer {
513 struct lnet_ping_info pb_info;
516 #define LNET_PING_BUFFER_SIZE(NNIDS) \
517 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
518 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
519 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
521 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
522 container_of((PINFO), struct lnet_ping_buffer, pb_info)
524 struct lnet_nid_list {
525 struct list_head nl_list;
529 struct lnet_peer_ni {
530 /* chain on lpn_peer_nis */
531 struct list_head lpni_peer_nis;
532 /* chain on remote peer list */
533 struct list_head lpni_on_remote_peer_ni_list;
534 /* chain on recovery queue */
535 struct list_head lpni_recovery;
536 /* chain on peer hash */
537 struct list_head lpni_hashlist;
538 /* messages blocking for tx credits */
539 struct list_head lpni_txq;
540 /* pointer to peer net I'm part of */
541 struct lnet_peer_net *lpni_peer_net;
542 /* statistics kept on each peer NI */
543 struct lnet_element_stats lpni_stats;
544 struct lnet_health_remote_stats lpni_hstats;
545 /* spin lock protecting credits and lpni_txq */
546 spinlock_t lpni_lock;
547 /* # tx credits available */
550 int lpni_mintxcredits;
552 * Each peer_ni in a gateway maintains its own credits. This
553 * allows more traffic to gateways that have multiple interfaces.
555 /* # router credits */
558 int lpni_minrtrcredits;
559 /* bytes queued for sending */
561 /* network peer is on */
562 struct lnet_net *lpni_net;
566 atomic_t lpni_refcount;
567 /* health value for the peer */
568 atomic_t lpni_healthv;
569 /* recovery ping mdh */
570 struct lnet_handle_md lpni_recovery_ping_mdh;
571 /* CPT this peer attached on */
573 /* state flags -- protected by lpni_lock */
575 /* status of the peer NI as reported by the peer */
576 __u32 lpni_ns_status;
577 /* sequence number used to round robin over peer nis within a net */
579 /* sequence number used to round robin over gateways */
581 /* returned RC ping features. Protected with lpni_lock */
582 unsigned int lpni_ping_feats;
583 /* time last message was received from the peer */
584 time64_t lpni_last_alive;
585 /* preferred local nids: if only one, use lpni_pref.nid */
588 struct list_head nids;
590 /* list of router nids preferred for this peer NI */
591 struct list_head lpni_rtr_pref_nids;
592 /* The relative selection priority of this peer NI */
593 __u32 lpni_sel_priority;
594 /* number of preferred NIDs in lnpi_pref_nids */
595 __u32 lpni_pref_nnids;
598 /* Preferred path added due to traffic on non-MR peer_ni */
599 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
600 /* peer is being recovered. */
601 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
602 /* recovery ping failed */
603 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
604 /* peer is being deleted */
605 #define LNET_PEER_NI_DELETING BIT(3)
608 /* chain on pt_peer_list */
609 struct list_head lp_peer_list;
611 /* list of peer nets */
612 struct list_head lp_peer_nets;
614 /* list of messages pending discovery*/
615 struct list_head lp_dc_pendq;
617 /* chain on router list */
618 struct list_head lp_rtr_list;
620 /* primary NID of the peer */
621 lnet_nid_t lp_primary_nid;
623 /* source NID to use during discovery */
624 lnet_nid_t lp_disc_src_nid;
626 /* net to perform discovery on */
627 __u32 lp_disc_net_id;
629 /* CPT of peer_table */
632 /* number of NIDs on this peer */
635 /* # refs from lnet_route::lr_gateway */
639 * peer specific health sensitivity value to decrement peer nis in
640 * this peer with if set to something other than 0
642 __u32 lp_health_sensitivity;
644 /* messages blocking for router credits */
645 struct list_head lp_rtrq;
647 /* routes on this peer */
648 struct list_head lp_routes;
650 /* reference count */
651 atomic_t lp_refcount;
653 /* lock protecting peer state flags and lpni_rtrq */
656 /* peer state flags */
659 /* buffer for data pushed by peer */
660 struct lnet_ping_buffer *lp_data;
662 /* MD handle for ping in progress */
663 struct lnet_handle_md lp_ping_mdh;
665 /* MD handle for push in progress */
666 struct lnet_handle_md lp_push_mdh;
668 /* number of NIDs for sizing push data */
671 /* NI config sequence number of peer */
674 /* Local NI config sequence number acked by peer */
677 /* Local NI config sequence number sent to peer */
678 __u32 lp_node_seqno_sent;
680 /* Ping error encountered during discovery. */
683 /* Push error encountered during discovery. */
686 /* Error encountered during discovery. */
689 /* time it was put on the ln_dc_working queue */
690 time64_t lp_last_queued;
692 /* link on discovery-related lists */
693 struct list_head lp_dc_list;
695 /* tasks waiting on discovery of this peer */
696 wait_queue_head_t lp_dc_waitq;
698 /* cached peer aliveness */
703 * The status flags in lp_state. Their semantics have chosen so that
704 * lp_state can be zero-initialized.
706 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
707 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
709 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
710 * NOT set when the peer was pinged by discovery.
712 * A peer is marked ROUTER if it indicates so in the feature bit.
714 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
715 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
716 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
719 * A peer is marked CONFIGURED if it was configured by DLC.
721 * In addition, a peer is marked DISCOVERED if it has fully passed
722 * through Peer Discovery.
724 * When Peer Discovery is disabled, the discovery thread will mark
725 * peers REDISCOVER to indicate that they should be re-examined if
726 * discovery is (re)enabled on the node.
728 * A peer that was created as the result of inbound traffic will not
731 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
732 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
733 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
735 * A peer is marked DISCOVERING when discovery is in progress.
736 * The other flags below correspond to stages of discovery.
738 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
739 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
740 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
741 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
742 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
743 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
744 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
746 * A ping can be forced as a way to fix up state, or as a manual
747 * intervention by an admin.
748 * A push can be forced in circumstances that would normally not
749 * allow for one to happen.
751 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
752 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
754 /* force delete even if router */
755 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
757 /* gw undergoing alive discovery */
758 #define LNET_PEER_RTR_DISCOVERY BIT(16)
759 /* gw has undergone discovery (does not indicate success or failure) */
760 #define LNET_PEER_RTR_DISCOVERED BIT(17)
762 /* peer is marked for deletion */
763 #define LNET_PEER_MARK_DELETION BIT(18)
765 struct lnet_peer_net {
766 /* chain on lp_peer_nets */
767 struct list_head lpn_peer_nets;
769 /* list of peer_nis on this network */
770 struct list_head lpn_peer_nis;
772 /* pointer to the peer I'm part of */
773 struct lnet_peer *lpn_peer;
778 /* peer net health */
781 /* time of last router net check attempt */
782 time64_t lpn_rtrcheck_timestamp;
784 /* selection sequence number */
787 /* relative peer net selection priority */
788 __u32 lpn_sel_priority;
790 /* reference count */
791 atomic_t lpn_refcount;
795 #define LNET_PEER_HASH_BITS 9
796 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
799 * peer hash table - one per CPT
801 * protected by lnet_net_lock/EX for update
806 * protected by pt_zombie_lock:
810 * pt_zombie lock nests inside lnet_net_lock
812 struct lnet_peer_table {
813 int pt_version; /* /proc validity stamp */
814 struct list_head *pt_hash; /* NID->peer hash */
815 struct list_head pt_peer_list; /* peers */
816 int pt_peers; /* # peers */
817 struct list_head pt_zombie_list; /* zombie peer_ni */
818 int pt_zombies; /* # zombie peers_ni */
819 spinlock_t pt_zombie_lock; /* protect list and count */
822 /* peer aliveness is enabled only on routers for peers in a network where the
823 * struct lnet_ni::ni_peertimeout has been set to a positive value
825 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
826 ((lp)->lpni_net) && \
827 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
830 struct list_head lr_list; /* chain on net */
831 struct list_head lr_gwlist; /* chain on gateway */
832 struct lnet_peer *lr_gateway; /* router node */
833 lnet_nid_t lr_nid; /* NID used to add route */
834 __u32 lr_net; /* remote network number */
835 __u32 lr_lnet; /* local network number */
836 int lr_seq; /* sequence for round-robin */
837 __u32 lr_hops; /* how far I am */
838 unsigned int lr_priority; /* route priority */
839 bool lr_alive; /* cached route aliveness */
840 bool lr_single_hop; /* this route is single-hop */
843 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
844 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
845 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
847 struct lnet_remotenet {
848 /* chain on ln_remote_nets_hash */
849 struct list_head lrn_list;
851 struct list_head lrn_routes;
856 /** lnet message has credit and can be submitted to lnd for send/receive */
857 #define LNET_CREDIT_OK 0
858 /** lnet message is waiting for credit */
859 #define LNET_CREDIT_WAIT 1
860 /** lnet message is waiting for discovery */
861 #define LNET_DC_WAIT 2
863 struct lnet_rtrbufpool {
864 /* my free buffer pool */
865 struct list_head rbp_bufs;
866 /* messages blocking for a buffer */
867 struct list_head rbp_msgs;
868 /* # pages in each buffer */
870 /* requested number of buffers */
871 int rbp_req_nbuffers;
872 /* # buffers actually allocated */
874 /* # free buffers / blocked messages */
881 struct list_head rb_list; /* chain on rbp_bufs */
882 struct lnet_rtrbufpool *rb_pool; /* owning pool */
883 struct bio_vec rb_kiov[0]; /* the buffer space */
886 #define LNET_PEER_HASHSIZE 503 /* prime! */
888 enum lnet_match_flags {
889 /* Didn't match anything */
890 LNET_MATCHMD_NONE = BIT(0),
892 LNET_MATCHMD_OK = BIT(1),
893 /* Must be discarded */
894 LNET_MATCHMD_DROP = BIT(2),
895 /* match and buffer is exhausted */
896 LNET_MATCHMD_EXHAUSTED = BIT(3),
898 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
901 /* Options for struct lnet_portal::ptl_options */
902 #define LNET_PTL_LAZY BIT(0)
903 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
904 #define LNET_PTL_MATCH_WILDCARD BIT(2) /* wildcard match, request portal */
906 /* parameter for matching operations (GET, PUT) */
907 struct lnet_match_info {
909 struct lnet_process_id mi_id;
912 unsigned int mi_portal;
913 unsigned int mi_rlength;
914 unsigned int mi_roffset;
917 /* ME hash of RDMA portal */
918 #define LNET_MT_HASH_BITS 8
919 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
920 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
921 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
922 * the last entry is reserved for MEs with ignore-bits */
923 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
924 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
925 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
926 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
927 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
928 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
929 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
931 /* portal match table */
932 struct lnet_match_table {
933 /* reserved for upcoming patches, CPU partition ID */
935 unsigned int mt_portal; /* portal index */
936 /* match table is set as "enabled" if there's non-exhausted MD
937 * attached on mt_mhash, it's only valid for wildcard portal */
938 unsigned int mt_enabled;
939 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
940 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
941 struct list_head *mt_mhash; /* matching hash */
944 /* these are only useful for wildcard portal */
945 /* Turn off message rotor for wildcard portals */
946 #define LNET_PTL_ROTOR_OFF 0
947 /* round-robin dispatch all PUT messages for wildcard portals */
948 #define LNET_PTL_ROTOR_ON 1
949 /* round-robin dispatch routed PUT message for wildcard portals */
950 #define LNET_PTL_ROTOR_RR_RT 2
951 /* dispatch routed PUT message by hashing source NID for wildcard portals */
952 #define LNET_PTL_ROTOR_HASH_RT 3
956 unsigned int ptl_index; /* portal ID, reserved */
957 /* flags on this portal: lazy, unique... */
958 unsigned int ptl_options;
959 /* list of messages which are stealing buffer */
960 struct list_head ptl_msg_stealing;
961 /* messages blocking for MD */
962 struct list_head ptl_msg_delayed;
963 /* Match table for each CPT */
964 struct lnet_match_table **ptl_mtables;
965 /* spread rotor of incoming "PUT" */
966 unsigned int ptl_rotor;
967 /* # active entries for this portal */
969 /* array of active entries' cpu-partition-id */
973 #define LNET_LH_HASH_BITS 12
974 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
975 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
977 /* resource container (ME, MD, EQ) */
978 struct lnet_res_container {
979 unsigned int rec_type; /* container type */
980 __u64 rec_lh_cookie; /* cookie generator */
981 struct list_head rec_active; /* active resource list */
982 struct list_head *rec_lh_hash; /* handle hash */
985 /* message container */
986 struct lnet_msg_container {
987 int msc_init; /* initialized or not */
988 /* max # threads finalizing */
990 /* msgs waiting to complete finalizing */
991 struct list_head msc_finalizing;
992 /* msgs waiting to be resent */
993 struct list_head msc_resending;
994 struct list_head msc_active; /* active message list */
995 /* threads doing finalization */
996 void **msc_finalizers;
997 /* threads doing resends */
998 void **msc_resenders;
1001 /* This UDSP structures need to match the user space liblnetconfig structures
1002 * in order for the marshall and unmarshall functions to be common.
1005 /* Net is described as a
1009 struct lnet_ud_net_descr {
1011 struct list_head udn_net_num_range;
1014 /* each NID range is defined as
1016 * 2. address range descriptor
1018 struct lnet_ud_nid_descr {
1019 struct lnet_ud_net_descr ud_net_id;
1020 struct list_head ud_addr_range;
1024 /* a UDSP rule can have up to three user defined NID descriptors
1025 * - src: defines the local NID range for the rule
1026 * - dst: defines the peer NID range for the rule
1027 * - rte: defines the router NID range for the rule
1029 * An action union defines the action to take when the rule
1033 struct list_head udsp_on_list;
1035 struct lnet_ud_nid_descr udsp_src;
1036 struct lnet_ud_nid_descr udsp_dst;
1037 struct lnet_ud_nid_descr udsp_rte;
1038 enum lnet_udsp_action_type udsp_action_type;
1040 __u32 udsp_priority;
1044 /* Peer Discovery states */
1045 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
1046 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
1047 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
1049 /* Router Checker states */
1050 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
1051 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
1052 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
1055 #define LNET_STATE_SHUTDOWN 0 /* not started */
1056 #define LNET_STATE_RUNNING 1 /* started up OK */
1057 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
1060 /* CPU partition table of LNet */
1061 struct cfs_cpt_table *ln_cpt_table;
1062 /* number of CPTs in ln_cpt_table */
1063 unsigned int ln_cpt_number;
1064 unsigned int ln_cpt_bits;
1066 /* protect LNet resources (ME/MD/EQ) */
1067 struct cfs_percpt_lock *ln_res_lock;
1070 /* the vector of portals */
1071 struct lnet_portal **ln_portals;
1072 /* percpt MD container */
1073 struct lnet_res_container **ln_md_containers;
1075 /* Event Queue container */
1076 struct lnet_res_container ln_eq_container;
1077 spinlock_t ln_eq_wait_lock;
1079 unsigned int ln_remote_nets_hbits;
1081 /* protect NI, peer table, credits, routers, rtrbuf... */
1082 struct cfs_percpt_lock *ln_net_lock;
1083 /* percpt message containers for active/finalizing/freed message */
1084 struct lnet_msg_container **ln_msg_containers;
1085 struct lnet_counters **ln_counters;
1086 struct lnet_peer_table **ln_peer_tables;
1087 /* list of peer nis not on a local network */
1088 struct list_head ln_remote_peer_ni_list;
1089 /* failure simulation */
1090 struct list_head ln_test_peers;
1091 struct list_head ln_drop_rules;
1092 struct list_head ln_delay_rules;
1094 struct list_head ln_nets;
1095 /* the loopback NI */
1096 struct lnet_ni *ln_loni;
1097 /* network zombie list */
1098 struct list_head ln_net_zombie;
1099 /* resend messages list */
1100 struct list_head ln_msg_resend;
1101 /* spin lock to protect the msg resend list */
1102 spinlock_t ln_msg_resend_lock;
1104 /* remote networks with routes to them */
1105 struct list_head *ln_remote_nets_hash;
1106 /* validity stamp */
1107 __u64 ln_remote_nets_version;
1108 /* list of all known routers */
1109 struct list_head ln_routers;
1110 /* validity stamp */
1111 __u64 ln_routers_version;
1112 /* percpt router buffer pools */
1113 struct lnet_rtrbufpool **ln_rtrpools;
1116 * Ping target / Push source
1118 * The ping target and push source share a single buffer. The
1119 * ln_ping_target is protected against concurrent updates by
1122 struct lnet_handle_md ln_ping_target_md;
1123 lnet_handler_t ln_ping_target_handler;
1124 struct lnet_ping_buffer *ln_ping_target;
1125 atomic_t ln_ping_target_seqno;
1130 * ln_push_nnis contains the desired size of the push target.
1131 * The lnet_net_lock is used to handle update races. The old
1132 * buffer may linger a while after it has been unlinked, in
1133 * which case the event handler cleans up.
1135 lnet_handler_t ln_push_target_handler;
1136 struct lnet_handle_md ln_push_target_md;
1137 struct lnet_ping_buffer *ln_push_target;
1138 int ln_push_target_nnis;
1140 /* discovery event queue handle */
1141 lnet_handler_t ln_dc_handler;
1142 /* discovery requests */
1143 struct list_head ln_dc_request;
1144 /* discovery working list */
1145 struct list_head ln_dc_working;
1146 /* discovery expired list */
1147 struct list_head ln_dc_expired;
1148 /* discovery thread wait queue */
1149 wait_queue_head_t ln_dc_waitq;
1150 /* discovery startup/shutdown state */
1153 /* monitor thread startup/shutdown state */
1155 /* serialise startup/shutdown */
1156 struct semaphore ln_mt_signal;
1158 struct mutex ln_api_mutex;
1159 struct mutex ln_lnd_mutex;
1160 /* Have I called LNetNIInit myself? */
1162 /* LNetNIInit/LNetNIFini counter */
1164 /* SHUTDOWN/RUNNING/STOPPING */
1167 int ln_routing; /* am I a router? */
1168 lnet_pid_t ln_pid; /* requested pid */
1169 /* uniquely identifies this ni in this epoch */
1170 __u64 ln_interface_cookie;
1171 /* registered LNDs */
1172 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1174 /* test protocol compatibility flags */
1175 unsigned long ln_testprotocompat;
1177 /* 0 - load the NIs from the mod params
1178 * 1 - do not load the NIs from the mod params
1179 * Reverse logic to ensure that other calls to LNetNIInit
1182 bool ln_nis_from_mod_params;
1185 * completion for the monitor thread. The monitor thread takes care of
1186 * checking routes, timedout messages and resending messages.
1188 struct completion ln_mt_wait_complete;
1190 /* per-cpt resend queues */
1191 struct list_head **ln_mt_resendqs;
1192 /* local NIs to recover */
1193 struct list_head ln_mt_localNIRecovq;
1194 /* local NIs to recover */
1195 struct list_head ln_mt_peerNIRecovq;
1197 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1198 * There are CPT number of queues. Since response trackers will be
1199 * added on the fast path we can't afford to grab the exclusive
1200 * net lock to protect these queues. The CPT will be calculated
1201 * based on the mdh cookie.
1203 struct list_head **ln_mt_rstq;
1205 * A response tracker becomes a zombie when the associated MD is queued
1206 * for unlink before the response tracker is detached from the MD. An
1207 * entry on a zombie list can be freed when either the remaining
1208 * operations on the MD complete or when LNet has shut down.
1210 struct list_head **ln_mt_zombie_rstqs;
1211 /* recovery handler */
1212 lnet_handler_t ln_mt_handler;
1215 * Completed when the discovery and monitor threads can enter their
1218 struct completion ln_started;
1220 struct list_head ln_udsp_list;