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
68 struct list_head msg_activelist;
69 struct list_head msg_list; /* Q for credits/MD */
71 struct lnet_process_id msg_target;
72 /* Primary NID of the source. */
73 lnet_nid_t msg_initiator;
74 /* where is it from, it's only for building event */
79 * hold parameters in case message is with held due
82 lnet_nid_t msg_src_nid_param;
83 lnet_nid_t msg_rtr_nid_param;
86 * Deadline for the message after which it will be finalized if it
91 /* committed for sending */
92 unsigned int msg_tx_committed:1;
93 /* CPT # this message committed for sending */
94 unsigned int msg_tx_cpt:15;
95 /* committed for receiving */
96 unsigned int msg_rx_committed:1;
97 /* CPT # this message committed for receiving */
98 unsigned int msg_rx_cpt:15;
99 /* queued for tx credit */
100 unsigned int msg_tx_delayed:1;
101 /* queued for RX buffer */
102 unsigned int msg_rx_delayed:1;
103 /* ready for pending on RX delay list */
104 unsigned int msg_rx_ready_delay:1;
106 unsigned int msg_vmflush:1; /* VM trying to free memory */
107 unsigned int msg_target_is_router:1; /* sending to a router */
108 unsigned int msg_routing:1; /* being forwarded */
109 unsigned int msg_ack:1; /* ack on finalize (PUT) */
110 unsigned int msg_sending:1; /* outgoing message */
111 unsigned int msg_receiving:1; /* being received */
112 unsigned int msg_txcredit:1; /* taken an NI send credit */
113 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
114 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
115 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
116 unsigned int msg_onactivelist:1; /* on the activelist */
117 unsigned int msg_rdma_get:1;
119 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
120 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
123 struct lnet_libmd *msg_md;
124 /* the NI the message was sent or received over */
125 struct lnet_ni *msg_txni;
126 struct lnet_ni *msg_rxni;
128 unsigned int msg_len;
129 unsigned int msg_wanted;
130 unsigned int msg_offset;
131 unsigned int msg_niov;
132 struct kvec *msg_iov;
133 lnet_kiov_t *msg_kiov;
135 struct lnet_event msg_ev;
136 struct lnet_hdr msg_hdr;
139 struct lnet_libhandle {
140 struct list_head lh_hash_chain;
144 #define lh_entry(ptr, type, member) \
145 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
148 struct list_head eq_list;
149 struct lnet_libhandle eq_lh;
150 unsigned long eq_enq_seq;
151 unsigned long eq_deq_seq;
152 unsigned int eq_size;
153 lnet_eq_handler_t eq_callback;
154 struct lnet_event *eq_events;
155 int **eq_refs; /* percpt refcount for EQ */
159 struct list_head me_list;
160 struct lnet_libhandle me_lh;
161 struct lnet_process_id me_match_id;
162 unsigned int me_portal;
163 unsigned int me_pos; /* hash offset in mt_hash */
165 __u64 me_ignore_bits;
166 enum lnet_unlink me_unlink;
167 struct lnet_libmd *me_md;
171 struct list_head md_list;
172 struct lnet_libhandle md_lh;
173 struct lnet_me *md_me;
175 unsigned int md_offset;
176 unsigned int md_length;
177 unsigned int md_max_size;
180 unsigned int md_options;
181 unsigned int md_flags;
182 unsigned int md_niov; /* # frags at end of struct */
184 struct lnet_eq *md_eq;
185 struct lnet_handle_md md_bulk_handle;
187 struct kvec iov[LNET_MAX_IOV];
188 lnet_kiov_t kiov[LNET_MAX_IOV];
192 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
193 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
194 #define LNET_MD_FLAG_ABORTED (1 << 2)
196 struct lnet_test_peer {
197 /* info about peers we are trying to fail */
198 struct list_head tp_list; /* ln_test_peers */
199 lnet_nid_t tp_nid; /* matching nid */
200 unsigned int tp_threshold; /* # failures to simulate */
203 #define LNET_COOKIE_TYPE_MD 1
204 #define LNET_COOKIE_TYPE_ME 2
205 #define LNET_COOKIE_TYPE_EQ 3
206 #define LNET_COOKIE_TYPE_BITS 2
207 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
209 struct lnet_ni; /* forward ref */
213 /* fields managed by portals */
214 struct list_head lnd_list; /* stash in the LND table */
215 int lnd_refcount; /* # active instances */
217 /* fields initialized by the LND */
220 int (*lnd_startup)(struct lnet_ni *ni);
221 void (*lnd_shutdown)(struct lnet_ni *ni);
222 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
224 /* In data movement APIs below, payload buffers are described as a set
225 * of 'niov' fragments which are...
227 * in virtual memory (struct kvec *iov != NULL)
229 * in pages (kernel only: plt_kiov_t *kiov != NULL).
230 * The LND may NOT overwrite these fragment descriptors.
231 * An 'offset' and may specify a byte offset within the set of
232 * fragments to start from
235 /* Start sending a preformatted message. 'private' is NULL for PUT and
236 * GET messages; otherwise this is a response to an incoming message
237 * and 'private' is the 'private' passed to lnet_parse(). Return
238 * non-zero for immediate failure, otherwise complete later with
240 int (*lnd_send)(struct lnet_ni *ni, void *private,
241 struct lnet_msg *msg);
243 /* Start receiving 'mlen' bytes of payload data, skipping the following
244 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
245 * lnet_parse(). Return non-zero for immedaite failure, otherwise
246 * complete later with lnet_finalize(). This also gives back a receive
247 * credit if the LND does flow control. */
248 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
249 int delayed, unsigned int niov,
250 struct kvec *iov, lnet_kiov_t *kiov,
251 unsigned int offset, unsigned int mlen, unsigned int rlen);
253 /* lnet_parse() has had to delay processing of this message
254 * (e.g. waiting for a forwarding buffer or send credits). Give the
255 * LND a chance to free urgently needed resources. If called, return 0
256 * for success and do NOT give back a receive credit; that has to wait
257 * until lnd_recv() gets called. On failure return < 0 and
258 * release resources; lnd_recv() will not be called. */
259 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
260 struct lnet_msg *msg, void **new_privatep);
262 /* notification of peer health */
263 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
265 /* query of peer aliveness */
266 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, time64_t *when);
268 /* accept a new connection */
269 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
272 struct lnet_tx_queue {
273 int tq_credits; /* # tx credits free */
274 int tq_credits_min; /* lowest it's been */
275 int tq_credits_max; /* total # tx credits */
276 struct list_head tq_delayed; /* delayed TXs */
279 enum lnet_net_state {
280 /* set when net block is allocated */
281 LNET_NET_STATE_INIT = 0,
282 /* set when NIs in net are started successfully */
283 LNET_NET_STATE_ACTIVE,
284 /* set if all NIs in net are in FAILED state */
285 LNET_NET_STATE_INACTIVE,
286 /* set when shutting down a NET */
287 LNET_NET_STATE_DELETING
291 /* set when NI block is allocated */
292 LNET_NI_STATE_INIT = 0,
293 /* set when NI is started successfully */
294 LNET_NI_STATE_ACTIVE,
295 /* set when LND notifies NI failed */
296 LNET_NI_STATE_FAILED,
297 /* set when LND notifies NI degraded */
298 LNET_NI_STATE_DEGRADED,
299 /* set when shuttding down NI */
300 LNET_NI_STATE_DELETING
303 enum lnet_stats_type {
304 LNET_STATS_TYPE_SEND = 0,
305 LNET_STATS_TYPE_RECV,
309 struct lnet_comm_count {
310 atomic_t co_get_count;
311 atomic_t co_put_count;
312 atomic_t co_reply_count;
313 atomic_t co_ack_count;
314 atomic_t co_hello_count;
317 struct lnet_element_stats {
318 struct lnet_comm_count el_send_stats;
319 struct lnet_comm_count el_recv_stats;
320 struct lnet_comm_count el_drop_stats;
324 /* chain on the ln_nets */
325 struct list_head net_list;
327 /* net ID, which is composed of
328 * (net_type << 16) | net_num.
329 * net_type can be one of the enumerated types defined in
330 * lnet/include/lnet/nidstr.h */
333 /* priority of the network */
336 /* total number of CPTs in the array */
339 /* cumulative CPTs of all NIs in this net */
342 /* network tunables */
343 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
346 * boolean to indicate that the tunables have been set and
349 bool net_tunables_set;
351 /* procedural interface */
352 struct lnet_lnd *net_lnd;
354 /* list of NIs on this net */
355 struct list_head net_ni_list;
357 /* list of NIs being added, but not started yet */
358 struct list_head net_ni_added;
360 /* dying LND instances */
361 struct list_head net_ni_zombie;
364 enum lnet_net_state net_state;
368 /* chain on the lnet_net structure */
369 struct list_head ni_netlist;
371 /* chain on net_ni_cpt */
372 struct list_head ni_cptlist;
379 /* bond NI on some CPTs */
382 /* interface's NID */
385 /* instance-specific data */
389 atomic_t ni_tx_credits;
391 /* percpt TX queues */
392 struct lnet_tx_queue **ni_tx_queues;
394 /* percpt reference count */
397 /* when I was last alive */
398 time64_t ni_last_alive;
400 /* pointer to parent network */
401 struct lnet_net *ni_net;
403 /* my health status */
404 struct lnet_ni_status *ni_status;
407 enum lnet_ni_state ni_state;
409 /* per NI LND tunables */
410 struct lnet_lnd_tunables ni_lnd_tunables;
412 /* lnd tunables set explicitly */
413 bool ni_lnd_tunables_set;
416 struct lnet_element_stats ni_stats;
418 /* physical device CPT */
421 /* sequence number used to round robin over nis within a net */
426 * initialized to LNET_MAX_HEALTH_VALUE
427 * Value is decremented every time we fail to send a message over
428 * this NI because of a NI specific failure.
429 * Value is incremented if we successfully send a message.
434 * equivalent interfaces to use
435 * This is an array because socklnd bonding can still be configured
437 char *ni_interfaces[LNET_INTERFACES_NUM];
438 struct net *ni_net_ns; /* original net namespace */
441 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
444 * Descriptor of a ping info buffer: keep a separate indicator of the
445 * size and a reference count. The type is used both as a source and
446 * sink of data, so we need to keep some information outside of the
447 * area that may be overwritten by network data.
449 struct lnet_ping_buffer {
452 struct lnet_ping_info pb_info;
455 #define LNET_PING_BUFFER_SIZE(NNIDS) \
456 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
457 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
458 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
460 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
461 container_of((PINFO), struct lnet_ping_buffer, pb_info)
463 /* router checker data, per router */
464 struct lnet_rc_data {
465 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
466 struct list_head rcd_list;
467 struct lnet_handle_md rcd_mdh; /* ping buffer MD */
468 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
469 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
470 int rcd_nnis; /* desired size of buffer */
473 struct lnet_peer_ni {
474 /* chain on lpn_peer_nis */
475 struct list_head lpni_peer_nis;
476 /* chain on remote peer list */
477 struct list_head lpni_on_remote_peer_ni_list;
478 /* chain on peer hash */
479 struct list_head lpni_hashlist;
480 /* messages blocking for tx credits */
481 struct list_head lpni_txq;
482 /* messages blocking for router credits */
483 struct list_head lpni_rtrq;
484 /* chain on router list */
485 struct list_head lpni_rtr_list;
486 /* pointer to peer net I'm part of */
487 struct lnet_peer_net *lpni_peer_net;
488 /* statistics kept on each peer NI */
489 struct lnet_element_stats lpni_stats;
490 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
491 spinlock_t lpni_lock;
492 /* # tx credits available */
495 int lpni_mintxcredits;
496 /* # router credits */
499 int lpni_minrtrcredits;
500 /* bytes queued for sending */
504 /* notification outstanding? */
506 /* outstanding notification for LND? */
508 /* some thread is handling notification */
510 /* SEND event outstanding from ping */
511 bool lpni_ping_notsent;
512 /* # times router went dead<->alive. Protected with lpni_lock */
513 int lpni_alive_count;
514 /* time of last aliveness news */
515 time64_t lpni_timestamp;
516 /* time of last ping attempt */
517 time64_t lpni_ping_timestamp;
518 /* != 0 if ping reply expected */
519 time64_t lpni_ping_deadline;
520 /* when I was last alive */
521 time64_t lpni_last_alive;
522 /* when lpni_ni was queried last time */
523 time64_t lpni_last_query;
524 /* network peer is on */
525 struct lnet_net *lpni_net;
529 atomic_t lpni_refcount;
530 /* CPT this peer attached on */
532 /* state flags -- protected by lpni_lock */
534 /* # refs from lnet_route_t::lr_gateway */
535 int lpni_rtr_refcount;
536 /* sequence number used to round robin over peer nis within a net */
538 /* sequence number used to round robin over gateways */
542 /* returned RC ping features. Protected with lpni_lock */
543 unsigned int lpni_ping_feats;
544 /* routes on this peer */
545 struct list_head lpni_routes;
546 /* preferred local nids: if only one, use lpni_pref.nid */
551 /* number of preferred NIDs in lnpi_pref_nids */
552 __u32 lpni_pref_nnids;
553 /* router checker state */
554 struct lnet_rc_data *lpni_rcd;
557 /* Preferred path added due to traffic on non-MR peer_ni */
558 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
561 /* chain on pt_peer_list */
562 struct list_head lp_peer_list;
564 /* list of peer nets */
565 struct list_head lp_peer_nets;
567 /* list of messages pending discovery*/
568 struct list_head lp_dc_pendq;
570 /* primary NID of the peer */
571 lnet_nid_t lp_primary_nid;
573 /* CPT of peer_table */
576 /* number of NIDs on this peer */
579 /* reference count */
580 atomic_t lp_refcount;
582 /* lock protecting peer state flags */
585 /* peer state flags */
588 /* buffer for data pushed by peer */
589 struct lnet_ping_buffer *lp_data;
591 /* MD handle for ping in progress */
592 struct lnet_handle_md lp_ping_mdh;
594 /* MD handle for push in progress */
595 struct lnet_handle_md lp_push_mdh;
597 /* number of NIDs for sizing push data */
600 /* NI config sequence number of peer */
603 /* Local NI config sequence number acked by peer */
606 /* Local NI config sequence number sent to peer */
607 __u32 lp_node_seqno_sent;
609 /* Ping error encountered during discovery. */
612 /* Push error encountered during discovery. */
615 /* Error encountered during discovery. */
618 /* time it was put on the ln_dc_working queue */
619 time64_t lp_last_queued;
621 /* link on discovery-related lists */
622 struct list_head lp_dc_list;
624 /* tasks waiting on discovery of this peer */
625 wait_queue_head_t lp_dc_waitq;
629 * The status flags in lp_state. Their semantics have chosen so that
630 * lp_state can be zero-initialized.
632 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
633 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
635 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
636 * NOT set when the peer was pinged by discovery.
638 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
639 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
641 * A peer is marked CONFIGURED if it was configured by DLC.
643 * In addition, a peer is marked DISCOVERED if it has fully passed
644 * through Peer Discovery.
646 * When Peer Discovery is disabled, the discovery thread will mark
647 * peers REDISCOVER to indicate that they should be re-examined if
648 * discovery is (re)enabled on the node.
650 * A peer that was created as the result of inbound traffic will not
653 #define LNET_PEER_CONFIGURED (1 << 2) /* Configured via DLC */
654 #define LNET_PEER_DISCOVERED (1 << 3) /* Peer was discovered */
655 #define LNET_PEER_REDISCOVER (1 << 4) /* Discovery was disabled */
657 * A peer is marked DISCOVERING when discovery is in progress.
658 * The other flags below correspond to stages of discovery.
660 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
661 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
662 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
663 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
664 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
665 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
666 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
668 * A ping can be forced as a way to fix up state, or as a manual
669 * intervention by an admin.
670 * A push can be forced in circumstances that would normally not
671 * allow for one to happen.
673 #define LNET_PEER_FORCE_PING (1 << 12) /* Forced Ping */
674 #define LNET_PEER_FORCE_PUSH (1 << 13) /* Forced Push */
676 struct lnet_peer_net {
677 /* chain on lp_peer_nets */
678 struct list_head lpn_peer_nets;
680 /* list of peer_nis on this network */
681 struct list_head lpn_peer_nis;
683 /* pointer to the peer I'm part of */
684 struct lnet_peer *lpn_peer;
689 /* reference count */
690 atomic_t lpn_refcount;
694 #define LNET_PEER_HASH_BITS 9
695 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
698 * peer hash table - one per CPT
700 * protected by lnet_net_lock/EX for update
706 * protected by pt_zombie_lock:
710 * pt_zombie lock nests inside lnet_net_lock
712 struct lnet_peer_table {
713 int pt_version; /* /proc validity stamp */
714 int pt_number; /* # peers_ni extant */
715 struct list_head *pt_hash; /* NID->peer hash */
716 struct list_head pt_peer_list; /* peers */
717 int pt_peers; /* # peers */
718 struct list_head pt_zombie_list; /* zombie peer_ni */
719 int pt_zombies; /* # zombie peers_ni */
720 spinlock_t pt_zombie_lock; /* protect list and count */
723 /* peer aliveness is enabled only on routers for peers in a network where the
724 * struct lnet_ni::ni_peertimeout has been set to a positive value
726 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
727 ((lp)->lpni_net) && \
728 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
731 struct list_head lr_list; /* chain on net */
732 struct list_head lr_gwlist; /* chain on gateway */
733 struct lnet_peer_ni *lr_gateway; /* router node */
734 __u32 lr_net; /* remote network number */
735 int lr_seq; /* sequence for round-robin */
736 unsigned int lr_downis; /* number of down NIs */
737 __u32 lr_hops; /* how far I am */
738 unsigned int lr_priority; /* route priority */
741 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
742 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
743 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
745 struct lnet_remotenet {
746 /* chain on ln_remote_nets_hash */
747 struct list_head lrn_list;
749 struct list_head lrn_routes;
754 /** lnet message has credit and can be submitted to lnd for send/receive */
755 #define LNET_CREDIT_OK 0
756 /** lnet message is waiting for credit */
757 #define LNET_CREDIT_WAIT 1
758 /** lnet message is waiting for discovery */
759 #define LNET_DC_WAIT 2
761 struct lnet_rtrbufpool {
762 /* my free buffer pool */
763 struct list_head rbp_bufs;
764 /* messages blocking for a buffer */
765 struct list_head rbp_msgs;
766 /* # pages in each buffer */
768 /* requested number of buffers */
769 int rbp_req_nbuffers;
770 /* # buffers actually allocated */
772 /* # free buffers / blocked messages */
779 struct list_head rb_list; /* chain on rbp_bufs */
780 struct lnet_rtrbufpool *rb_pool; /* owning pool */
781 lnet_kiov_t rb_kiov[0]; /* the buffer space */
784 #define LNET_PEER_HASHSIZE 503 /* prime! */
786 enum lnet_match_flags {
787 /* Didn't match anything */
788 LNET_MATCHMD_NONE = (1 << 0),
790 LNET_MATCHMD_OK = (1 << 1),
791 /* Must be discarded */
792 LNET_MATCHMD_DROP = (1 << 2),
793 /* match and buffer is exhausted */
794 LNET_MATCHMD_EXHAUSTED = (1 << 3),
796 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
799 /* Options for struct lnet_portal::ptl_options */
800 #define LNET_PTL_LAZY (1 << 0)
801 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
802 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
804 /* parameter for matching operations (GET, PUT) */
805 struct lnet_match_info {
807 struct lnet_process_id mi_id;
810 unsigned int mi_portal;
811 unsigned int mi_rlength;
812 unsigned int mi_roffset;
815 /* ME hash of RDMA portal */
816 #define LNET_MT_HASH_BITS 8
817 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
818 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
819 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
820 * the last entry is reserved for MEs with ignore-bits */
821 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
822 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
823 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
824 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
825 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
826 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
827 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
829 /* portal match table */
830 struct lnet_match_table {
831 /* reserved for upcoming patches, CPU partition ID */
833 unsigned int mt_portal; /* portal index */
834 /* match table is set as "enabled" if there's non-exhausted MD
835 * attached on mt_mhash, it's only valid for wildcard portal */
836 unsigned int mt_enabled;
837 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
838 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
839 struct list_head *mt_mhash; /* matching hash */
842 /* these are only useful for wildcard portal */
843 /* Turn off message rotor for wildcard portals */
844 #define LNET_PTL_ROTOR_OFF 0
845 /* round-robin dispatch all PUT messages for wildcard portals */
846 #define LNET_PTL_ROTOR_ON 1
847 /* round-robin dispatch routed PUT message for wildcard portals */
848 #define LNET_PTL_ROTOR_RR_RT 2
849 /* dispatch routed PUT message by hashing source NID for wildcard portals */
850 #define LNET_PTL_ROTOR_HASH_RT 3
854 unsigned int ptl_index; /* portal ID, reserved */
855 /* flags on this portal: lazy, unique... */
856 unsigned int ptl_options;
857 /* list of messages which are stealing buffer */
858 struct list_head ptl_msg_stealing;
859 /* messages blocking for MD */
860 struct list_head ptl_msg_delayed;
861 /* Match table for each CPT */
862 struct lnet_match_table **ptl_mtables;
863 /* spread rotor of incoming "PUT" */
864 unsigned int ptl_rotor;
865 /* # active entries for this portal */
867 /* array of active entries' cpu-partition-id */
871 #define LNET_LH_HASH_BITS 12
872 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
873 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
875 /* resource container (ME, MD, EQ) */
876 struct lnet_res_container {
877 unsigned int rec_type; /* container type */
878 __u64 rec_lh_cookie; /* cookie generator */
879 struct list_head rec_active; /* active resource list */
880 struct list_head *rec_lh_hash; /* handle hash */
883 /* message container */
884 struct lnet_msg_container {
885 int msc_init; /* initialized or not */
886 /* max # threads finalizing */
888 /* msgs waiting to complete finalizing */
889 struct list_head msc_finalizing;
890 struct list_head msc_active; /* active message list */
891 /* threads doing finalization */
892 void **msc_finalizers;
895 /* Peer Discovery states */
896 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
897 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
898 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
900 /* Router Checker states */
901 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
902 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
903 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
906 #define LNET_STATE_SHUTDOWN 0 /* not started */
907 #define LNET_STATE_RUNNING 1 /* started up OK */
908 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
911 /* CPU partition table of LNet */
912 struct cfs_cpt_table *ln_cpt_table;
913 /* number of CPTs in ln_cpt_table */
914 unsigned int ln_cpt_number;
915 unsigned int ln_cpt_bits;
917 /* protect LNet resources (ME/MD/EQ) */
918 struct cfs_percpt_lock *ln_res_lock;
921 /* the vector of portals */
922 struct lnet_portal **ln_portals;
923 /* percpt ME containers */
924 struct lnet_res_container **ln_me_containers;
925 /* percpt MD container */
926 struct lnet_res_container **ln_md_containers;
928 /* Event Queue container */
929 struct lnet_res_container ln_eq_container;
930 wait_queue_head_t ln_eq_waitq;
931 spinlock_t ln_eq_wait_lock;
933 unsigned int ln_remote_nets_hbits;
935 /* protect NI, peer table, credits, routers, rtrbuf... */
936 struct cfs_percpt_lock *ln_net_lock;
937 /* percpt message containers for active/finalizing/freed message */
938 struct lnet_msg_container **ln_msg_containers;
939 struct lnet_counters **ln_counters;
940 struct lnet_peer_table **ln_peer_tables;
941 /* list of peer nis not on a local network */
942 struct list_head ln_remote_peer_ni_list;
943 /* failure simulation */
944 struct list_head ln_test_peers;
945 struct list_head ln_drop_rules;
946 struct list_head ln_delay_rules;
948 struct list_head ln_nets;
949 /* the loopback NI */
950 struct lnet_ni *ln_loni;
951 /* network zombie list */
952 struct list_head ln_net_zombie;
953 /* resend messages list */
954 struct list_head ln_msg_resend;
955 /* spin lock to protect the msg resend list */
956 spinlock_t ln_msg_resend_lock;
958 /* remote networks with routes to them */
959 struct list_head *ln_remote_nets_hash;
961 __u64 ln_remote_nets_version;
962 /* list of all known routers */
963 struct list_head ln_routers;
965 __u64 ln_routers_version;
966 /* percpt router buffer pools */
967 struct lnet_rtrbufpool **ln_rtrpools;
970 * Ping target / Push source
972 * The ping target and push source share a single buffer. The
973 * ln_ping_target is protected against concurrent updates by
976 struct lnet_handle_md ln_ping_target_md;
977 struct lnet_handle_eq ln_ping_target_eq;
978 struct lnet_ping_buffer *ln_ping_target;
979 atomic_t ln_ping_target_seqno;
984 * ln_push_nnis contains the desired size of the push target.
985 * The lnet_net_lock is used to handle update races. The old
986 * buffer may linger a while after it has been unlinked, in
987 * which case the event handler cleans up.
989 struct lnet_handle_eq ln_push_target_eq;
990 struct lnet_handle_md ln_push_target_md;
991 struct lnet_ping_buffer *ln_push_target;
992 int ln_push_target_nnis;
994 /* discovery event queue handle */
995 struct lnet_handle_eq ln_dc_eqh;
996 /* discovery requests */
997 struct list_head ln_dc_request;
998 /* discovery working list */
999 struct list_head ln_dc_working;
1000 /* discovery expired list */
1001 struct list_head ln_dc_expired;
1002 /* discovery thread wait queue */
1003 wait_queue_head_t ln_dc_waitq;
1004 /* discovery startup/shutdown state */
1007 /* monitor thread startup/shutdown state */
1009 /* router checker's event queue */
1010 struct lnet_handle_eq ln_rc_eqh;
1011 /* rcd still pending on net */
1012 struct list_head ln_rcd_deathrow;
1013 /* rcd ready for free */
1014 struct list_head ln_rcd_zombie;
1015 /* serialise startup/shutdown */
1016 struct semaphore ln_mt_signal;
1018 struct mutex ln_api_mutex;
1019 struct mutex ln_lnd_mutex;
1020 /* Have I called LNetNIInit myself? */
1022 /* LNetNIInit/LNetNIFini counter */
1024 /* SHUTDOWN/RUNNING/STOPPING */
1027 int ln_routing; /* am I a router? */
1028 lnet_pid_t ln_pid; /* requested pid */
1029 /* uniquely identifies this ni in this epoch */
1030 __u64 ln_interface_cookie;
1031 /* registered LNDs */
1032 struct list_head ln_lnds;
1034 /* test protocol compatibility flags */
1035 int ln_testprotocompat;
1037 /* 0 - load the NIs from the mod params
1038 * 1 - do not load the NIs from the mod params
1039 * Reverse logic to ensure that other calls to LNetNIInit
1042 bool ln_nis_from_mod_params;
1045 * waitq for the monitor thread. The monitor thread takes care of
1046 * checking routes, timedout messages and resending messages.
1048 wait_queue_head_t ln_mt_waitq;