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 #ifndef CONFIG_LNET_MAX_PAYLOAD
55 # error "CONFIG_LNET_MAX_PAYLOAD must be defined in config.h"
58 #define LNET_MAX_PAYLOAD CONFIG_LNET_MAX_PAYLOAD
59 #if (LNET_MAX_PAYLOAD < LNET_MTU)
60 # error "LNET_MAX_PAYLOAD too small - error in configure --with-max-payload-mb"
61 #elif (LNET_MAX_PAYLOAD > (PAGE_SIZE * LNET_MAX_IOV))
62 # error "LNET_MAX_PAYLOAD too large - error in configure --with-max-payload-mb"
69 struct list_head msg_activelist;
70 struct list_head msg_list; /* Q for credits/MD */
72 struct lnet_process_id msg_target;
73 /* Primary NID of the source. */
74 lnet_nid_t msg_initiator;
75 /* where is it from, it's only for building event */
80 * hold parameters in case message is with held due
83 lnet_nid_t msg_src_nid_param;
84 lnet_nid_t msg_rtr_nid_param;
86 /* committed for sending */
87 unsigned int msg_tx_committed:1;
88 /* CPT # this message committed for sending */
89 unsigned int msg_tx_cpt:15;
90 /* committed for receiving */
91 unsigned int msg_rx_committed:1;
92 /* CPT # this message committed for receiving */
93 unsigned int msg_rx_cpt:15;
94 /* queued for tx credit */
95 unsigned int msg_tx_delayed:1;
96 /* queued for RX buffer */
97 unsigned int msg_rx_delayed:1;
98 /* ready for pending on RX delay list */
99 unsigned int msg_rx_ready_delay:1;
101 unsigned int msg_vmflush:1; /* VM trying to free memory */
102 unsigned int msg_target_is_router:1; /* sending to a router */
103 unsigned int msg_routing:1; /* being forwarded */
104 unsigned int msg_ack:1; /* ack on finalize (PUT) */
105 unsigned int msg_sending:1; /* outgoing message */
106 unsigned int msg_receiving:1; /* being received */
107 unsigned int msg_txcredit:1; /* taken an NI send credit */
108 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
109 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
110 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
111 unsigned int msg_onactivelist:1; /* on the activelist */
112 unsigned int msg_rdma_get:1;
114 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
115 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
118 struct lnet_libmd *msg_md;
119 /* the NI the message was sent or received over */
120 struct lnet_ni *msg_txni;
121 struct lnet_ni *msg_rxni;
123 unsigned int msg_len;
124 unsigned int msg_wanted;
125 unsigned int msg_offset;
126 unsigned int msg_niov;
127 struct kvec *msg_iov;
128 lnet_kiov_t *msg_kiov;
130 struct lnet_event msg_ev;
131 struct lnet_hdr msg_hdr;
134 struct lnet_libhandle {
135 struct list_head lh_hash_chain;
139 #define lh_entry(ptr, type, member) \
140 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
143 struct list_head eq_list;
144 struct lnet_libhandle eq_lh;
145 unsigned long eq_enq_seq;
146 unsigned long eq_deq_seq;
147 unsigned int eq_size;
148 lnet_eq_handler_t eq_callback;
149 struct lnet_event *eq_events;
150 int **eq_refs; /* percpt refcount for EQ */
154 struct list_head me_list;
155 struct lnet_libhandle me_lh;
156 struct lnet_process_id me_match_id;
157 unsigned int me_portal;
158 unsigned int me_pos; /* hash offset in mt_hash */
160 __u64 me_ignore_bits;
161 enum lnet_unlink me_unlink;
162 struct lnet_libmd *me_md;
166 struct list_head md_list;
167 struct lnet_libhandle md_lh;
168 struct lnet_me *md_me;
170 unsigned int md_offset;
171 unsigned int md_length;
172 unsigned int md_max_size;
175 unsigned int md_options;
176 unsigned int md_flags;
177 unsigned int md_niov; /* # frags at end of struct */
179 struct lnet_eq *md_eq;
180 struct lnet_handle_md md_bulk_handle;
182 struct kvec iov[LNET_MAX_IOV];
183 lnet_kiov_t kiov[LNET_MAX_IOV];
187 #define LNET_MD_FLAG_ZOMBIE (1 << 0)
188 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1)
189 #define LNET_MD_FLAG_ABORTED (1 << 2)
191 struct lnet_test_peer {
192 /* info about peers we are trying to fail */
193 struct list_head tp_list; /* ln_test_peers */
194 lnet_nid_t tp_nid; /* matching nid */
195 unsigned int tp_threshold; /* # failures to simulate */
198 #define LNET_COOKIE_TYPE_MD 1
199 #define LNET_COOKIE_TYPE_ME 2
200 #define LNET_COOKIE_TYPE_EQ 3
201 #define LNET_COOKIE_TYPE_BITS 2
202 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
204 struct lnet_ni; /* forward ref */
208 /* fields managed by portals */
209 struct list_head lnd_list; /* stash in the LND table */
210 int lnd_refcount; /* # active instances */
212 /* fields initialized by the LND */
215 int (*lnd_startup)(struct lnet_ni *ni);
216 void (*lnd_shutdown)(struct lnet_ni *ni);
217 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
219 /* In data movement APIs below, payload buffers are described as a set
220 * of 'niov' fragments which are...
222 * in virtual memory (struct kvec *iov != NULL)
224 * in pages (kernel only: plt_kiov_t *kiov != NULL).
225 * The LND may NOT overwrite these fragment descriptors.
226 * An 'offset' and may specify a byte offset within the set of
227 * fragments to start from
230 /* Start sending a preformatted message. 'private' is NULL for PUT and
231 * GET messages; otherwise this is a response to an incoming message
232 * and 'private' is the 'private' passed to lnet_parse(). Return
233 * non-zero for immediate failure, otherwise complete later with
235 int (*lnd_send)(struct lnet_ni *ni, void *private,
236 struct lnet_msg *msg);
238 /* Start receiving 'mlen' bytes of payload data, skipping the following
239 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
240 * lnet_parse(). Return non-zero for immedaite failure, otherwise
241 * complete later with lnet_finalize(). This also gives back a receive
242 * credit if the LND does flow control. */
243 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
244 int delayed, unsigned int niov,
245 struct kvec *iov, lnet_kiov_t *kiov,
246 unsigned int offset, unsigned int mlen, unsigned int rlen);
248 /* lnet_parse() has had to delay processing of this message
249 * (e.g. waiting for a forwarding buffer or send credits). Give the
250 * LND a chance to free urgently needed resources. If called, return 0
251 * for success and do NOT give back a receive credit; that has to wait
252 * until lnd_recv() gets called. On failure return < 0 and
253 * release resources; lnd_recv() will not be called. */
254 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
255 struct lnet_msg *msg, void **new_privatep);
257 /* notification of peer health */
258 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
260 /* query of peer aliveness */
261 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, cfs_time_t *when);
263 /* accept a new connection */
264 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
267 struct lnet_tx_queue {
268 int tq_credits; /* # tx credits free */
269 int tq_credits_min; /* lowest it's been */
270 int tq_credits_max; /* total # tx credits */
271 struct list_head tq_delayed; /* delayed TXs */
274 enum lnet_net_state {
275 /* set when net block is allocated */
276 LNET_NET_STATE_INIT = 0,
277 /* set when NIs in net are started successfully */
278 LNET_NET_STATE_ACTIVE,
279 /* set if all NIs in net are in FAILED state */
280 LNET_NET_STATE_INACTIVE,
281 /* set when shutting down a NET */
282 LNET_NET_STATE_DELETING
286 /* set when NI block is allocated */
287 LNET_NI_STATE_INIT = 0,
288 /* set when NI is started successfully */
289 LNET_NI_STATE_ACTIVE,
290 /* set when LND notifies NI failed */
291 LNET_NI_STATE_FAILED,
292 /* set when LND notifies NI degraded */
293 LNET_NI_STATE_DEGRADED,
294 /* set when shuttding down NI */
295 LNET_NI_STATE_DELETING
298 enum lnet_stats_type {
299 LNET_STATS_TYPE_SEND = 0,
300 LNET_STATS_TYPE_RECV,
304 struct lnet_comm_count {
305 atomic_t co_get_count;
306 atomic_t co_put_count;
307 atomic_t co_reply_count;
308 atomic_t co_ack_count;
309 atomic_t co_hello_count;
312 struct lnet_element_stats {
313 struct lnet_comm_count el_send_stats;
314 struct lnet_comm_count el_recv_stats;
315 struct lnet_comm_count el_drop_stats;
319 /* chain on the ln_nets */
320 struct list_head net_list;
322 /* net ID, which is composed of
323 * (net_type << 16) | net_num.
324 * net_type can be one of the enumerated types defined in
325 * lnet/include/lnet/nidstr.h */
328 /* priority of the network */
331 /* total number of CPTs in the array */
334 /* cumulative CPTs of all NIs in this net */
337 /* network tunables */
338 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
341 * boolean to indicate that the tunables have been set and
344 bool net_tunables_set;
346 /* procedural interface */
347 struct lnet_lnd *net_lnd;
349 /* list of NIs on this net */
350 struct list_head net_ni_list;
352 /* list of NIs being added, but not started yet */
353 struct list_head net_ni_added;
355 /* dying LND instances */
356 struct list_head net_ni_zombie;
359 enum lnet_net_state net_state;
363 /* chain on the lnet_net structure */
364 struct list_head ni_netlist;
366 /* chain on net_ni_cpt */
367 struct list_head ni_cptlist;
374 /* bond NI on some CPTs */
377 /* interface's NID */
380 /* instance-specific data */
384 atomic_t ni_tx_credits;
386 /* percpt TX queues */
387 struct lnet_tx_queue **ni_tx_queues;
389 /* percpt reference count */
392 /* when I was last alive */
395 /* pointer to parent network */
396 struct lnet_net *ni_net;
398 /* my health status */
399 struct lnet_ni_status *ni_status;
402 enum lnet_ni_state ni_state;
404 /* per NI LND tunables */
405 struct lnet_lnd_tunables ni_lnd_tunables;
407 /* lnd tunables set explicitly */
408 bool ni_lnd_tunables_set;
411 struct lnet_element_stats ni_stats;
413 /* physical device CPT */
416 /* sequence number used to round robin over nis within a net */
420 * equivalent interfaces to use
421 * This is an array because socklnd bonding can still be configured
423 char *ni_interfaces[LNET_INTERFACES_NUM];
424 struct net *ni_net_ns; /* original net namespace */
427 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
430 * Descriptor of a ping info buffer: keep a separate indicator of the
431 * size and a reference count. The type is used both as a source and
432 * sink of data, so we need to keep some information outside of the
433 * area that may be overwritten by network data.
435 struct lnet_ping_buffer {
438 struct lnet_ping_info pb_info;
441 #define LNET_PING_BUFFER_SIZE(NNIDS) \
442 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
443 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
444 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
446 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
447 container_of((PINFO), struct lnet_ping_buffer, pb_info)
449 /* router checker data, per router */
450 struct lnet_rc_data {
451 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
452 struct list_head rcd_list;
453 struct lnet_handle_md rcd_mdh; /* ping buffer MD */
454 struct lnet_peer_ni *rcd_gateway; /* reference to gateway */
455 struct lnet_ping_buffer *rcd_pingbuffer;/* ping buffer */
456 int rcd_nnis; /* desired size of buffer */
459 struct lnet_peer_ni {
460 /* chain on lpn_peer_nis */
461 struct list_head lpni_peer_nis;
462 /* chain on remote peer list */
463 struct list_head lpni_on_remote_peer_ni_list;
464 /* chain on peer hash */
465 struct list_head lpni_hashlist;
466 /* messages blocking for tx credits */
467 struct list_head lpni_txq;
468 /* messages blocking for router credits */
469 struct list_head lpni_rtrq;
470 /* chain on router list */
471 struct list_head lpni_rtr_list;
472 /* pointer to peer net I'm part of */
473 struct lnet_peer_net *lpni_peer_net;
474 /* statistics kept on each peer NI */
475 struct lnet_element_stats lpni_stats;
476 /* spin lock protecting credits and lpni_txq / lpni_rtrq */
477 spinlock_t lpni_lock;
478 /* # tx credits available */
481 int lpni_mintxcredits;
482 /* # router credits */
485 int lpni_minrtrcredits;
486 /* bytes queued for sending */
490 /* notification outstanding? */
492 /* outstanding notification for LND? */
494 /* some thread is handling notification */
496 /* SEND event outstanding from ping */
497 bool lpni_ping_notsent;
498 /* # times router went dead<->alive. Protected with lpni_lock */
499 int lpni_alive_count;
500 /* time of last aliveness news */
501 cfs_time_t lpni_timestamp;
502 /* time of last ping attempt */
503 cfs_time_t lpni_ping_timestamp;
504 /* != 0 if ping reply expected */
505 cfs_time_t lpni_ping_deadline;
506 /* when I was last alive */
507 cfs_time_t lpni_last_alive;
508 /* when lpni_ni was queried last time */
509 cfs_time_t lpni_last_query;
510 /* network peer is on */
511 struct lnet_net *lpni_net;
515 atomic_t lpni_refcount;
516 /* CPT this peer attached on */
518 /* state flags -- protected by lpni_lock */
520 /* # refs from lnet_route_t::lr_gateway */
521 int lpni_rtr_refcount;
522 /* sequence number used to round robin over peer nis within a net */
524 /* sequence number used to round robin over gateways */
528 /* returned RC ping features. Protected with lpni_lock */
529 unsigned int lpni_ping_feats;
530 /* routes on this peer */
531 struct list_head lpni_routes;
532 /* preferred local nids: if only one, use lpni_pref.nid */
537 /* number of preferred NIDs in lnpi_pref_nids */
538 __u32 lpni_pref_nnids;
539 /* router checker state */
540 struct lnet_rc_data *lpni_rcd;
543 /* Preferred path added due to traffic on non-MR peer_ni */
544 #define LNET_PEER_NI_NON_MR_PREF (1 << 0)
547 /* chain on pt_peer_list */
548 struct list_head lp_peer_list;
550 /* list of peer nets */
551 struct list_head lp_peer_nets;
553 /* list of messages pending discovery*/
554 struct list_head lp_dc_pendq;
556 /* primary NID of the peer */
557 lnet_nid_t lp_primary_nid;
559 /* CPT of peer_table */
562 /* number of NIDs on this peer */
565 /* reference count */
566 atomic_t lp_refcount;
568 /* lock protecting peer state flags */
571 /* peer state flags */
574 /* buffer for data pushed by peer */
575 struct lnet_ping_buffer *lp_data;
577 /* MD handle for ping in progress */
578 struct lnet_handle_md lp_ping_mdh;
580 /* MD handle for push in progress */
581 struct lnet_handle_md lp_push_mdh;
583 /* number of NIDs for sizing push data */
586 /* NI config sequence number of peer */
589 /* Local NI config sequence number acked by peer */
592 /* Local NI config sequence number sent to peer */
593 __u32 lp_node_seqno_sent;
595 /* Ping error encountered during discovery. */
598 /* Push error encountered during discovery. */
601 /* Error encountered during discovery. */
604 /* time it was put on the ln_dc_working queue */
605 time64_t lp_last_queued;
607 /* link on discovery-related lists */
608 struct list_head lp_dc_list;
610 /* tasks waiting on discovery of this peer */
611 wait_queue_head_t lp_dc_waitq;
615 * The status flags in lp_state. Their semantics have chosen so that
616 * lp_state can be zero-initialized.
618 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
619 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
621 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
622 * NOT set when the peer was pinged by discovery.
624 #define LNET_PEER_MULTI_RAIL (1 << 0) /* Multi-rail aware */
625 #define LNET_PEER_NO_DISCOVERY (1 << 1) /* Peer disabled discovery */
627 * A peer is marked CONFIGURED if it was configured by DLC.
629 * In addition, a peer is marked DISCOVERED if it has fully passed
630 * through Peer Discovery.
632 * When Peer Discovery is disabled, the discovery thread will mark
633 * peers REDISCOVER to indicate that they should be re-examined if
634 * discovery is (re)enabled on the node.
636 * A peer that was created as the result of inbound traffic will not
639 #define LNET_PEER_CONFIGURED (1 << 2) /* Configured via DLC */
640 #define LNET_PEER_DISCOVERED (1 << 3) /* Peer was discovered */
641 #define LNET_PEER_REDISCOVER (1 << 4) /* Discovery was disabled */
643 * A peer is marked DISCOVERING when discovery is in progress.
644 * The other flags below correspond to stages of discovery.
646 #define LNET_PEER_DISCOVERING (1 << 5) /* Discovering */
647 #define LNET_PEER_DATA_PRESENT (1 << 6) /* Remote peer data present */
648 #define LNET_PEER_NIDS_UPTODATE (1 << 7) /* Remote peer info uptodate */
649 #define LNET_PEER_PING_SENT (1 << 8) /* Waiting for REPLY to Ping */
650 #define LNET_PEER_PUSH_SENT (1 << 9) /* Waiting for ACK of Push */
651 #define LNET_PEER_PING_FAILED (1 << 10) /* Ping send failure */
652 #define LNET_PEER_PUSH_FAILED (1 << 11) /* Push send failure */
654 * A ping can be forced as a way to fix up state, or as a manual
655 * intervention by an admin.
656 * A push can be forced in circumstances that would normally not
657 * allow for one to happen.
659 #define LNET_PEER_FORCE_PING (1 << 12) /* Forced Ping */
660 #define LNET_PEER_FORCE_PUSH (1 << 13) /* Forced Push */
662 struct lnet_peer_net {
663 /* chain on lp_peer_nets */
664 struct list_head lpn_peer_nets;
666 /* list of peer_nis on this network */
667 struct list_head lpn_peer_nis;
669 /* pointer to the peer I'm part of */
670 struct lnet_peer *lpn_peer;
675 /* reference count */
676 atomic_t lpn_refcount;
680 #define LNET_PEER_HASH_BITS 9
681 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
684 * peer hash table - one per CPT
686 * protected by lnet_net_lock/EX for update
692 * protected by pt_zombie_lock:
696 * pt_zombie lock nests inside lnet_net_lock
698 struct lnet_peer_table {
699 int pt_version; /* /proc validity stamp */
700 int pt_number; /* # peers_ni extant */
701 struct list_head *pt_hash; /* NID->peer hash */
702 struct list_head pt_peer_list; /* peers */
703 int pt_peers; /* # peers */
704 struct list_head pt_zombie_list; /* zombie peer_ni */
705 int pt_zombies; /* # zombie peers_ni */
706 spinlock_t pt_zombie_lock; /* protect list and count */
709 /* peer aliveness is enabled only on routers for peers in a network where the
710 * struct lnet_ni::ni_peertimeout has been set to a positive value
712 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
713 ((lp)->lpni_net) && \
714 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
717 struct list_head lr_list; /* chain on net */
718 struct list_head lr_gwlist; /* chain on gateway */
719 struct lnet_peer_ni *lr_gateway; /* router node */
720 __u32 lr_net; /* remote network number */
721 int lr_seq; /* sequence for round-robin */
722 unsigned int lr_downis; /* number of down NIs */
723 __u32 lr_hops; /* how far I am */
724 unsigned int lr_priority; /* route priority */
727 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
728 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
729 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
731 struct lnet_remotenet {
732 /* chain on ln_remote_nets_hash */
733 struct list_head lrn_list;
735 struct list_head lrn_routes;
740 /** lnet message has credit and can be submitted to lnd for send/receive */
741 #define LNET_CREDIT_OK 0
742 /** lnet message is waiting for credit */
743 #define LNET_CREDIT_WAIT 1
744 /** lnet message is waiting for discovery */
745 #define LNET_DC_WAIT 2
747 struct lnet_rtrbufpool {
748 /* my free buffer pool */
749 struct list_head rbp_bufs;
750 /* messages blocking for a buffer */
751 struct list_head rbp_msgs;
752 /* # pages in each buffer */
754 /* requested number of buffers */
755 int rbp_req_nbuffers;
756 /* # buffers actually allocated */
758 /* # free buffers / blocked messages */
765 struct list_head rb_list; /* chain on rbp_bufs */
766 struct lnet_rtrbufpool *rb_pool; /* owning pool */
767 lnet_kiov_t rb_kiov[0]; /* the buffer space */
770 #define LNET_PEER_HASHSIZE 503 /* prime! */
772 enum lnet_match_flags {
773 /* Didn't match anything */
774 LNET_MATCHMD_NONE = (1 << 0),
776 LNET_MATCHMD_OK = (1 << 1),
777 /* Must be discarded */
778 LNET_MATCHMD_DROP = (1 << 2),
779 /* match and buffer is exhausted */
780 LNET_MATCHMD_EXHAUSTED = (1 << 3),
782 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
785 /* Options for struct lnet_portal::ptl_options */
786 #define LNET_PTL_LAZY (1 << 0)
787 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */
788 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */
790 /* parameter for matching operations (GET, PUT) */
791 struct lnet_match_info {
793 struct lnet_process_id mi_id;
796 unsigned int mi_portal;
797 unsigned int mi_rlength;
798 unsigned int mi_roffset;
801 /* ME hash of RDMA portal */
802 #define LNET_MT_HASH_BITS 8
803 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
804 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
805 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
806 * the last entry is reserved for MEs with ignore-bits */
807 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
808 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
809 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
810 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
811 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
812 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
813 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
815 /* portal match table */
816 struct lnet_match_table {
817 /* reserved for upcoming patches, CPU partition ID */
819 unsigned int mt_portal; /* portal index */
820 /* match table is set as "enabled" if there's non-exhausted MD
821 * attached on mt_mhash, it's only valid for wildcard portal */
822 unsigned int mt_enabled;
823 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
824 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
825 struct list_head *mt_mhash; /* matching hash */
828 /* these are only useful for wildcard portal */
829 /* Turn off message rotor for wildcard portals */
830 #define LNET_PTL_ROTOR_OFF 0
831 /* round-robin dispatch all PUT messages for wildcard portals */
832 #define LNET_PTL_ROTOR_ON 1
833 /* round-robin dispatch routed PUT message for wildcard portals */
834 #define LNET_PTL_ROTOR_RR_RT 2
835 /* dispatch routed PUT message by hashing source NID for wildcard portals */
836 #define LNET_PTL_ROTOR_HASH_RT 3
840 unsigned int ptl_index; /* portal ID, reserved */
841 /* flags on this portal: lazy, unique... */
842 unsigned int ptl_options;
843 /* list of messages which are stealing buffer */
844 struct list_head ptl_msg_stealing;
845 /* messages blocking for MD */
846 struct list_head ptl_msg_delayed;
847 /* Match table for each CPT */
848 struct lnet_match_table **ptl_mtables;
849 /* spread rotor of incoming "PUT" */
850 unsigned int ptl_rotor;
851 /* # active entries for this portal */
853 /* array of active entries' cpu-partition-id */
857 #define LNET_LH_HASH_BITS 12
858 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
859 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
861 /* resource container (ME, MD, EQ) */
862 struct lnet_res_container {
863 unsigned int rec_type; /* container type */
864 __u64 rec_lh_cookie; /* cookie generator */
865 struct list_head rec_active; /* active resource list */
866 struct list_head *rec_lh_hash; /* handle hash */
869 /* message container */
870 struct lnet_msg_container {
871 int msc_init; /* initialized or not */
872 /* max # threads finalizing */
874 /* msgs waiting to complete finalizing */
875 struct list_head msc_finalizing;
876 struct list_head msc_active; /* active message list */
877 /* threads doing finalization */
878 void **msc_finalizers;
881 /* Peer Discovery states */
882 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
883 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
884 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
886 /* Router Checker states */
887 #define LNET_RC_STATE_SHUTDOWN 0 /* not started */
888 #define LNET_RC_STATE_RUNNING 1 /* started up OK */
889 #define LNET_RC_STATE_STOPPING 2 /* telling thread to stop */
892 #define LNET_STATE_SHUTDOWN 0 /* not started */
893 #define LNET_STATE_RUNNING 1 /* started up OK */
894 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
897 /* CPU partition table of LNet */
898 struct cfs_cpt_table *ln_cpt_table;
899 /* number of CPTs in ln_cpt_table */
900 unsigned int ln_cpt_number;
901 unsigned int ln_cpt_bits;
903 /* protect LNet resources (ME/MD/EQ) */
904 struct cfs_percpt_lock *ln_res_lock;
907 /* the vector of portals */
908 struct lnet_portal **ln_portals;
909 /* percpt ME containers */
910 struct lnet_res_container **ln_me_containers;
911 /* percpt MD container */
912 struct lnet_res_container **ln_md_containers;
914 /* Event Queue container */
915 struct lnet_res_container ln_eq_container;
916 wait_queue_head_t ln_eq_waitq;
917 spinlock_t ln_eq_wait_lock;
919 unsigned int ln_remote_nets_hbits;
921 /* protect NI, peer table, credits, routers, rtrbuf... */
922 struct cfs_percpt_lock *ln_net_lock;
923 /* percpt message containers for active/finalizing/freed message */
924 struct lnet_msg_container **ln_msg_containers;
925 struct lnet_counters **ln_counters;
926 struct lnet_peer_table **ln_peer_tables;
927 /* list of peer nis not on a local network */
928 struct list_head ln_remote_peer_ni_list;
929 /* failure simulation */
930 struct list_head ln_test_peers;
931 struct list_head ln_drop_rules;
932 struct list_head ln_delay_rules;
934 struct list_head ln_nets;
935 /* the loopback NI */
936 struct lnet_ni *ln_loni;
937 /* network zombie list */
938 struct list_head ln_net_zombie;
939 /* resend messages list */
940 struct list_head ln_msg_resend;
941 /* spin lock to protect the msg resend list */
942 spinlock_t ln_msg_resend_lock;
944 /* remote networks with routes to them */
945 struct list_head *ln_remote_nets_hash;
947 __u64 ln_remote_nets_version;
948 /* list of all known routers */
949 struct list_head ln_routers;
951 __u64 ln_routers_version;
952 /* percpt router buffer pools */
953 struct lnet_rtrbufpool **ln_rtrpools;
956 * Ping target / Push source
958 * The ping target and push source share a single buffer. The
959 * ln_ping_target is protected against concurrent updates by
962 struct lnet_handle_md ln_ping_target_md;
963 struct lnet_handle_eq ln_ping_target_eq;
964 struct lnet_ping_buffer *ln_ping_target;
965 atomic_t ln_ping_target_seqno;
970 * ln_push_nnis contains the desired size of the push target.
971 * The lnet_net_lock is used to handle update races. The old
972 * buffer may linger a while after it has been unlinked, in
973 * which case the event handler cleans up.
975 struct lnet_handle_eq ln_push_target_eq;
976 struct lnet_handle_md ln_push_target_md;
977 struct lnet_ping_buffer *ln_push_target;
978 int ln_push_target_nnis;
980 /* discovery event queue handle */
981 struct lnet_handle_eq ln_dc_eqh;
982 /* discovery requests */
983 struct list_head ln_dc_request;
984 /* discovery working list */
985 struct list_head ln_dc_working;
986 /* discovery expired list */
987 struct list_head ln_dc_expired;
988 /* discovery thread wait queue */
989 wait_queue_head_t ln_dc_waitq;
990 /* discovery startup/shutdown state */
993 /* router checker startup/shutdown state */
995 /* router checker's event queue */
996 struct lnet_handle_eq ln_rc_eqh;
997 /* rcd still pending on net */
998 struct list_head ln_rcd_deathrow;
999 /* rcd ready for free */
1000 struct list_head ln_rcd_zombie;
1001 /* serialise startup/shutdown */
1002 struct semaphore ln_rc_signal;
1004 struct mutex ln_api_mutex;
1005 struct mutex ln_lnd_mutex;
1006 /* Have I called LNetNIInit myself? */
1008 /* LNetNIInit/LNetNIFini counter */
1010 /* SHUTDOWN/RUNNING/STOPPING */
1013 int ln_routing; /* am I a router? */
1014 lnet_pid_t ln_pid; /* requested pid */
1015 /* uniquely identifies this ni in this epoch */
1016 __u64 ln_interface_cookie;
1017 /* registered LNDs */
1018 struct list_head ln_lnds;
1020 /* test protocol compatibility flags */
1021 int ln_testprotocompat;
1023 /* 0 - load the NIs from the mod params
1024 * 1 - do not load the NIs from the mod params
1025 * Reverse logic to ensure that other calls to LNetNIInit
1028 bool ln_nis_from_mod_params;
1030 /* waitq for router checker. As long as there are no routes in
1031 * the list, the router checker will sleep on this queue. when
1032 * routes are added the thread will wake up */
1033 wait_queue_head_t ln_rc_waitq;