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/
31 * lnet/include/lnet/lib-types.h
33 * Types used by the library side routines that do not need to be
34 * exposed to the user application
37 #ifndef __LNET_LIB_TYPES_H__
38 #define __LNET_LIB_TYPES_H__
41 # error This include is only for kernel use.
44 #include <linux/kthread.h>
45 #include <linux/uio.h>
46 #include <linux/semaphore.h>
47 #include <linux/types.h>
48 #include <linux/kref.h>
49 #include <net/genetlink.h>
51 #include <uapi/linux/lnet/lnet-nl.h>
52 #include <uapi/linux/lnet/lnet-dlc.h>
53 #include <uapi/linux/lnet/lnetctl.h>
54 #include <uapi/linux/lnet/nidstr.h>
56 /* Max payload size */
57 #define LNET_MAX_PAYLOAD LNET_MTU
59 /** limit on the number of fragments in discontiguous MDs */
60 #define LNET_MAX_IOV 256
63 * This is the maximum health value.
64 * All local and peer NIs created have their health default to this value.
66 #define LNET_MAX_HEALTH_VALUE 1000
67 #define LNET_MAX_SELECTION_PRIORITY UINT_MAX
72 enum lnet_msg_hstatus {
73 LNET_MSG_STATUS_OK = 0,
74 LNET_MSG_STATUS_LOCAL_INTERRUPT,
75 LNET_MSG_STATUS_LOCAL_DROPPED,
76 LNET_MSG_STATUS_LOCAL_ABORTED,
77 LNET_MSG_STATUS_LOCAL_NO_ROUTE,
78 LNET_MSG_STATUS_LOCAL_ERROR,
79 LNET_MSG_STATUS_LOCAL_TIMEOUT,
80 LNET_MSG_STATUS_REMOTE_ERROR,
81 LNET_MSG_STATUS_REMOTE_DROPPED,
82 LNET_MSG_STATUS_REMOTE_TIMEOUT,
83 LNET_MSG_STATUS_NETWORK_TIMEOUT,
87 struct lnet_rsp_tracker {
88 /* chain on the waiting list */
89 struct list_head rspt_on_list;
93 lnet_nid_t rspt_next_hop_nid;
94 /* deadline of the REPLY/ACK */
95 ktime_t rspt_deadline;
97 struct lnet_handle_md rspt_mdh;
101 struct list_head msg_activelist;
102 struct list_head msg_list; /* Q for credits/MD */
104 struct lnet_process_id msg_target;
105 /* Primary NID of the source. */
106 lnet_nid_t msg_initiator;
107 /* where is it from, it's only for building event */
112 * hold parameters in case message is with held due
115 lnet_nid_t msg_src_nid_param;
116 lnet_nid_t msg_rtr_nid_param;
119 * Deadline for the message after which it will be finalized if it
122 ktime_t msg_deadline;
124 /* The message health status. */
125 enum lnet_msg_hstatus msg_health_status;
126 /* This is a recovery message */
128 /* the number of times a transmission has been retried */
130 /* flag to indicate that we do not want to resend this message */
133 /* committed for sending */
134 unsigned int msg_tx_committed:1;
135 /* CPT # this message committed for sending */
136 unsigned int msg_tx_cpt:15;
137 /* committed for receiving */
138 unsigned int msg_rx_committed:1;
139 /* CPT # this message committed for receiving */
140 unsigned int msg_rx_cpt:15;
141 /* queued for tx credit */
142 unsigned int msg_tx_delayed:1;
143 /* queued for RX buffer */
144 unsigned int msg_rx_delayed:1;
145 /* ready for pending on RX delay list */
146 unsigned int msg_rx_ready_delay:1;
148 unsigned int msg_vmflush:1; /* VM trying to free memory */
149 unsigned int msg_target_is_router:1; /* sending to a router */
150 unsigned int msg_routing:1; /* being forwarded */
151 unsigned int msg_ack:1; /* ack on finalize (PUT) */
152 unsigned int msg_sending:1; /* outgoing message */
153 unsigned int msg_receiving:1; /* being received */
154 unsigned int msg_txcredit:1; /* taken an NI send credit */
155 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
156 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
157 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
158 unsigned int msg_onactivelist:1; /* on the activelist */
159 unsigned int msg_rdma_get:1;
161 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
162 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
165 struct lnet_libmd *msg_md;
166 /* the NI the message was sent or received over */
167 struct lnet_ni *msg_txni;
168 struct lnet_ni *msg_rxni;
170 unsigned int msg_len;
171 unsigned int msg_wanted;
172 unsigned int msg_offset;
173 unsigned int msg_niov;
174 struct bio_vec *msg_kiov;
176 struct lnet_event msg_ev;
177 struct lnet_hdr msg_hdr;
180 struct lnet_libhandle {
181 struct list_head lh_hash_chain;
185 #define lh_entry(ptr, type, member) \
186 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
189 struct list_head me_list;
191 struct lnet_process_id me_match_id;
192 unsigned int me_portal;
193 unsigned int me_pos; /* hash offset in mt_hash */
195 __u64 me_ignore_bits;
196 enum lnet_unlink me_unlink;
197 struct lnet_libmd *me_md;
201 struct list_head md_list;
202 struct lnet_libhandle md_lh;
203 struct lnet_me *md_me;
205 unsigned int md_offset;
206 unsigned int md_length;
207 unsigned int md_max_size;
210 unsigned int md_options;
211 unsigned int md_flags;
212 unsigned int md_niov; /* # frags at end of struct */
214 struct lnet_rsp_tracker *md_rspt_ptr;
215 lnet_handler_t md_handler;
216 struct lnet_handle_md md_bulk_handle;
217 struct bio_vec md_kiov[LNET_MAX_IOV];
220 #define LNET_MD_FLAG_ZOMBIE BIT(0)
221 #define LNET_MD_FLAG_AUTO_UNLINK BIT(1)
222 #define LNET_MD_FLAG_ABORTED BIT(2)
223 /* LNET_MD_FLAG_HANDLING is set when a non-unlink event handler
224 * is being called for an event relating to the md.
225 * It ensures only one such handler runs at a time.
226 * The final "unlink" event is only called once the
227 * md_refcount has reached zero, and this flag has been cleared,
228 * ensuring that it doesn't race with any other event handler
231 #define LNET_MD_FLAG_HANDLING BIT(3)
232 #define LNET_MD_FLAG_DISCARD BIT(4)
234 struct lnet_test_peer {
235 /* info about peers we are trying to fail */
236 struct list_head tp_list; /* ln_test_peers */
237 lnet_nid_t tp_nid; /* matching nid */
238 unsigned int tp_threshold; /* # failures to simulate */
241 #define LNET_COOKIE_TYPE_MD 1
242 #define LNET_COOKIE_TYPE_ME 2
243 #define LNET_COOKIE_TYPE_EQ 3
244 #define LNET_COOKIE_TYPE_BITS 2
245 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
251 void (*nf_addr2str)(u32 addr, char *str, size_t size);
252 int (*nf_str2addr)(const char *str, int nob, u32 *addr);
253 int (*nf_parse_addrlist)(char *str, int len,
254 struct list_head *list);
255 int (*nf_print_addrlist)(char *buffer, int count,
256 struct list_head *list);
257 int (*nf_match_addr)(u32 addr, struct list_head *list);
258 int (*nf_min_max)(struct list_head *nidlist, u32 *min_nid,
262 struct lnet_ni; /* forward ref */
266 /* fields initialized by the LND */
269 int (*lnd_startup)(struct lnet_ni *ni);
270 void (*lnd_shutdown)(struct lnet_ni *ni);
271 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
273 /* In data movement APIs below, payload buffers are described as a set
274 * of 'niov' fragments which are in pages.
275 * The LND may NOT overwrite these fragment descriptors.
276 * An 'offset' and may specify a byte offset within the set of
277 * fragments to start from
280 /* Start sending a preformatted message. 'private' is NULL for PUT and
281 * GET messages; otherwise this is a response to an incoming message
282 * and 'private' is the 'private' passed to lnet_parse(). Return
283 * non-zero for immediate failure, otherwise complete later with
285 int (*lnd_send)(struct lnet_ni *ni, void *private,
286 struct lnet_msg *msg);
288 /* Start receiving 'mlen' bytes of payload data, skipping the following
289 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
290 * lnet_parse(). Return non-zero for immedaite failure, otherwise
291 * complete later with lnet_finalize(). This also gives back a receive
292 * credit if the LND does flow control. */
293 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
294 int delayed, unsigned int niov,
295 struct bio_vec *kiov,
296 unsigned int offset, unsigned int mlen, unsigned int rlen);
298 /* lnet_parse() has had to delay processing of this message
299 * (e.g. waiting for a forwarding buffer or send credits). Give the
300 * LND a chance to free urgently needed resources. If called, return 0
301 * for success and do NOT give back a receive credit; that has to wait
302 * until lnd_recv() gets called. On failure return < 0 and
303 * release resources; lnd_recv() will not be called. */
304 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
305 struct lnet_msg *msg, void **new_privatep);
307 /* notification of peer down */
308 void (*lnd_notify_peer_down)(lnet_nid_t peer);
310 /* accept a new connection */
311 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
314 struct lnet_tx_queue {
315 int tq_credits; /* # tx credits free */
316 int tq_credits_min; /* lowest it's been */
317 int tq_credits_max; /* total # tx credits */
318 struct list_head tq_delayed; /* delayed TXs */
321 enum lnet_net_state {
322 /* set when net block is allocated */
323 LNET_NET_STATE_INIT = 0,
324 /* set when NIs in net are started successfully */
325 LNET_NET_STATE_ACTIVE,
326 /* set if all NIs in net are in FAILED state */
327 LNET_NET_STATE_INACTIVE,
328 /* set when shutting down a NET */
329 LNET_NET_STATE_DELETING
333 /* initial state when NI is created */
334 LNET_NI_STATE_INIT = 0,
335 /* set when NI is brought up */
336 LNET_NI_STATE_ACTIVE,
337 /* set when NI is being shutdown */
338 LNET_NI_STATE_DELETING,
341 #define LNET_NI_RECOVERY_PENDING BIT(0)
342 #define LNET_NI_RECOVERY_FAILED BIT(1)
344 enum lnet_stats_type {
345 LNET_STATS_TYPE_SEND = 0,
346 LNET_STATS_TYPE_RECV,
350 struct lnet_comm_count {
351 atomic_t co_get_count;
352 atomic_t co_put_count;
353 atomic_t co_reply_count;
354 atomic_t co_ack_count;
355 atomic_t co_hello_count;
358 struct lnet_element_stats {
359 struct lnet_comm_count el_send_stats;
360 struct lnet_comm_count el_recv_stats;
361 struct lnet_comm_count el_drop_stats;
364 struct lnet_health_local_stats {
365 atomic_t hlt_local_interrupt;
366 atomic_t hlt_local_dropped;
367 atomic_t hlt_local_aborted;
368 atomic_t hlt_local_no_route;
369 atomic_t hlt_local_timeout;
370 atomic_t hlt_local_error;
373 struct lnet_health_remote_stats {
374 atomic_t hlt_remote_dropped;
375 atomic_t hlt_remote_timeout;
376 atomic_t hlt_remote_error;
377 atomic_t hlt_network_timeout;
381 /* chain on the ln_nets */
382 struct list_head net_list;
384 /* net ID, which is composed of
385 * (net_type << 16) | net_num.
386 * net_type can be one of the enumerated types defined in
387 * lnet/include/lnet/nidstr.h */
390 /* round robin selection */
393 /* total number of CPTs in the array */
396 /* cumulative CPTs of all NIs in this net */
399 /* relative net selection priority */
400 __u32 net_sel_priority;
402 /* network tunables */
403 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
406 * boolean to indicate that the tunables have been set and
409 bool net_tunables_set;
411 /* procedural interface */
412 const struct lnet_lnd *net_lnd;
414 /* list of NIs on this net */
415 struct list_head net_ni_list;
417 /* list of NIs being added, but not started yet */
418 struct list_head net_ni_added;
420 /* dying LND instances */
421 struct list_head net_ni_zombie;
423 /* when I was last alive */
424 time64_t net_last_alive;
426 /* protects access to net_last_alive */
429 /* list of router nids preferred for this network */
430 struct list_head net_rtr_pref_nids;
434 /* chain on the lnet_net structure */
435 struct list_head ni_netlist;
437 /* chain on the recovery queue */
438 struct list_head ni_recovery;
440 /* MD handle for recovery ping */
441 struct lnet_handle_md ni_ping_mdh;
448 /* bond NI on some CPTs */
451 /* interface's NID */
454 /* instance-specific data */
458 atomic_t ni_tx_credits;
460 /* percpt TX queues */
461 struct lnet_tx_queue **ni_tx_queues;
463 /* percpt reference count */
466 /* pointer to parent network */
467 struct lnet_net *ni_net;
469 /* my health status */
470 struct lnet_ni_status *ni_status;
472 /* NI FSM. Protected by lnet_ni_lock() */
473 enum lnet_ni_state ni_state;
475 /* Recovery state. Protected by lnet_ni_lock() */
476 __u32 ni_recovery_state;
478 /* When to send the next recovery ping */
479 time64_t ni_next_ping;
480 /* How many pings sent during current recovery period did not receive
481 * a reply. NB: reset whenever _any_ message arrives on this NI
483 unsigned int ni_ping_count;
485 /* per NI LND tunables */
486 struct lnet_lnd_tunables ni_lnd_tunables;
488 /* lnd tunables set explicitly */
489 bool ni_lnd_tunables_set;
492 struct lnet_element_stats ni_stats;
493 struct lnet_health_local_stats ni_hstats;
495 /* physical device CPT */
498 /* sequence number used to round robin over nis within a net */
503 * initialized to LNET_MAX_HEALTH_VALUE
504 * Value is decremented every time we fail to send a message over
505 * this NI because of a NI specific failure.
506 * Value is incremented if we successfully send a message.
511 * Set to 1 by the LND when it receives an event telling it the device
512 * has gone into a fatal state. Set to 0 when the LND receives an
513 * even telling it the device is back online.
515 atomic_t ni_fatal_error_on;
517 /* the relative selection priority of this NI */
518 __u32 ni_sel_priority;
521 * equivalent interface to use
524 struct net *ni_net_ns; /* original net namespace */
527 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
530 * Descriptor of a ping info buffer: keep a separate indicator of the
531 * size and a reference count. The type is used both as a source and
532 * sink of data, so we need to keep some information outside of the
533 * area that may be overwritten by network data.
535 struct lnet_ping_buffer {
539 struct lnet_ping_info pb_info;
542 #define LNET_PING_BUFFER_SIZE(NNIDS) \
543 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
544 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
545 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
547 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
548 container_of((PINFO), struct lnet_ping_buffer, pb_info)
550 struct lnet_nid_list {
551 struct list_head nl_list;
555 struct lnet_peer_ni {
556 /* chain on lpn_peer_nis */
557 struct list_head lpni_peer_nis;
558 /* chain on remote peer list */
559 struct list_head lpni_on_remote_peer_ni_list;
560 /* chain on recovery queue */
561 struct list_head lpni_recovery;
562 /* chain on peer hash */
563 struct list_head lpni_hashlist;
564 /* messages blocking for tx credits */
565 struct list_head lpni_txq;
566 /* pointer to peer net I'm part of */
567 struct lnet_peer_net *lpni_peer_net;
568 /* statistics kept on each peer NI */
569 struct lnet_element_stats lpni_stats;
570 struct lnet_health_remote_stats lpni_hstats;
571 /* spin lock protecting credits and lpni_txq */
572 spinlock_t lpni_lock;
573 /* # tx credits available */
576 int lpni_mintxcredits;
578 * Each peer_ni in a gateway maintains its own credits. This
579 * allows more traffic to gateways that have multiple interfaces.
581 /* # router credits */
584 int lpni_minrtrcredits;
585 /* bytes queued for sending */
587 /* network peer is on */
588 struct lnet_net *lpni_net;
592 struct kref lpni_kref;
593 /* health value for the peer */
594 atomic_t lpni_healthv;
595 /* recovery ping mdh */
596 struct lnet_handle_md lpni_recovery_ping_mdh;
597 /* When to send the next recovery ping */
598 time64_t lpni_next_ping;
599 /* How many pings sent during current recovery period did not receive
600 * a reply. NB: reset whenever _any_ message arrives from this peer NI
602 unsigned int lpni_ping_count;
603 /* CPT this peer attached on */
605 /* state flags -- protected by lpni_lock */
607 /* status of the peer NI as reported by the peer */
608 __u32 lpni_ns_status;
609 /* sequence number used to round robin over peer nis within a net */
611 /* sequence number used to round robin over gateways */
613 /* returned RC ping features. Protected with lpni_lock */
614 unsigned int lpni_ping_feats;
615 /* time last message was received from the peer */
616 time64_t lpni_last_alive;
617 /* preferred local nids: if only one, use lpni_pref.nid */
620 struct list_head nids;
622 /* list of router nids preferred for this peer NI */
623 struct list_head lpni_rtr_pref_nids;
624 /* The relative selection priority of this peer NI */
625 __u32 lpni_sel_priority;
626 /* number of preferred NIDs in lnpi_pref_nids */
627 __u32 lpni_pref_nnids;
630 /* Preferred path added due to traffic on non-MR peer_ni */
631 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
632 /* peer is being recovered. */
633 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
634 /* recovery ping failed */
635 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
636 /* peer is being deleted */
637 #define LNET_PEER_NI_DELETING BIT(3)
640 /* chain on pt_peer_list */
641 struct list_head lp_peer_list;
643 /* list of peer nets */
644 struct list_head lp_peer_nets;
646 /* list of messages pending discovery*/
647 struct list_head lp_dc_pendq;
649 /* chain on router list */
650 struct list_head lp_rtr_list;
652 /* primary NID of the peer */
653 lnet_nid_t lp_primary_nid;
655 /* source NID to use during discovery */
656 lnet_nid_t lp_disc_src_nid;
657 /* destination NID to use during discovery */
658 lnet_nid_t lp_disc_dst_nid;
660 /* net to perform discovery on */
661 __u32 lp_disc_net_id;
663 /* CPT of peer_table */
666 /* number of NIDs on this peer */
669 /* # refs from lnet_route::lr_gateway */
673 * peer specific health sensitivity value to decrement peer nis in
674 * this peer with if set to something other than 0
676 __u32 lp_health_sensitivity;
678 /* messages blocking for router credits */
679 struct list_head lp_rtrq;
681 /* routes on this peer */
682 struct list_head lp_routes;
684 /* reference count */
685 atomic_t lp_refcount;
687 /* lock protecting peer state flags and lpni_rtrq */
690 /* peer state flags */
693 /* buffer for data pushed by peer */
694 struct lnet_ping_buffer *lp_data;
696 /* MD handle for ping in progress */
697 struct lnet_handle_md lp_ping_mdh;
699 /* MD handle for push in progress */
700 struct lnet_handle_md lp_push_mdh;
702 /* number of NIDs for sizing push data */
705 /* NI config sequence number of peer */
708 /* Local NI config sequence number acked by peer */
711 /* Local NI config sequence number sent to peer */
712 __u32 lp_node_seqno_sent;
714 /* Ping error encountered during discovery. */
717 /* Push error encountered during discovery. */
720 /* Error encountered during discovery. */
723 /* time it was put on the ln_dc_working queue */
724 time64_t lp_last_queued;
726 /* link on discovery-related lists */
727 struct list_head lp_dc_list;
729 /* tasks waiting on discovery of this peer */
730 wait_queue_head_t lp_dc_waitq;
732 /* cached peer aliveness */
737 * The status flags in lp_state. Their semantics have chosen so that
738 * lp_state can be zero-initialized.
740 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
741 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
743 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
744 * NOT set when the peer was pinged by discovery.
746 * A peer is marked ROUTER if it indicates so in the feature bit.
748 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
749 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
750 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
753 * A peer is marked CONFIGURED if it was configured by DLC.
755 * In addition, a peer is marked DISCOVERED if it has fully passed
756 * through Peer Discovery.
758 * When Peer Discovery is disabled, the discovery thread will mark
759 * peers REDISCOVER to indicate that they should be re-examined if
760 * discovery is (re)enabled on the node.
762 * A peer that was created as the result of inbound traffic will not
765 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
766 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
767 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
769 * A peer is marked DISCOVERING when discovery is in progress.
770 * The other flags below correspond to stages of discovery.
772 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
773 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
774 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
775 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
776 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
777 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
778 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
780 * A ping can be forced as a way to fix up state, or as a manual
781 * intervention by an admin.
782 * A push can be forced in circumstances that would normally not
783 * allow for one to happen.
785 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
786 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
788 /* force delete even if router */
789 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
791 /* gw undergoing alive discovery */
792 #define LNET_PEER_RTR_DISCOVERY BIT(16)
793 /* gw has undergone discovery (does not indicate success or failure) */
794 #define LNET_PEER_RTR_DISCOVERED BIT(17)
796 /* peer is marked for deletion */
797 #define LNET_PEER_MARK_DELETION BIT(18)
798 /* lnet_peer_del()/lnet_peer_del_locked() has been called on the peer */
799 #define LNET_PEER_MARK_DELETED BIT(19)
801 struct lnet_peer_net {
802 /* chain on lp_peer_nets */
803 struct list_head lpn_peer_nets;
805 /* list of peer_nis on this network */
806 struct list_head lpn_peer_nis;
808 /* pointer to the peer I'm part of */
809 struct lnet_peer *lpn_peer;
814 /* peer net health */
817 /* time of next router ping on this net */
818 time64_t lpn_next_ping;
820 /* selection sequence number */
823 /* relative peer net selection priority */
824 __u32 lpn_sel_priority;
826 /* reference count */
827 atomic_t lpn_refcount;
831 #define LNET_PEER_HASH_BITS 9
832 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
835 * peer hash table - one per CPT
837 * protected by lnet_net_lock/EX for update
842 * protected by pt_zombie_lock:
846 * pt_zombie lock nests inside lnet_net_lock
848 struct lnet_peer_table {
849 int pt_version; /* /proc validity stamp */
850 struct list_head *pt_hash; /* NID->peer hash */
851 struct list_head pt_peer_list; /* peers */
852 int pt_peers; /* # peers */
853 struct list_head pt_zombie_list; /* zombie peer_ni */
854 int pt_zombies; /* # zombie peers_ni */
855 spinlock_t pt_zombie_lock; /* protect list and count */
858 /* peer aliveness is enabled only on routers for peers in a network where the
859 * struct lnet_ni::ni_peertimeout has been set to a positive value
861 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
862 ((lp)->lpni_net) && \
863 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
866 struct list_head lr_list; /* chain on net */
867 struct list_head lr_gwlist; /* chain on gateway */
868 struct lnet_peer *lr_gateway; /* router node */
869 lnet_nid_t lr_nid; /* NID used to add route */
870 __u32 lr_net; /* remote network number */
871 __u32 lr_lnet; /* local network number */
872 int lr_seq; /* sequence for round-robin */
873 __u32 lr_hops; /* how far I am */
874 unsigned int lr_priority; /* route priority */
875 atomic_t lr_alive; /* cached route aliveness */
876 bool lr_single_hop; /* this route is single-hop */
879 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
880 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
881 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
883 struct lnet_remotenet {
884 /* chain on ln_remote_nets_hash */
885 struct list_head lrn_list;
887 struct list_head lrn_routes;
892 /** lnet message has credit and can be submitted to lnd for send/receive */
893 #define LNET_CREDIT_OK 0
894 /** lnet message is waiting for credit */
895 #define LNET_CREDIT_WAIT 1
896 /** lnet message is waiting for discovery */
897 #define LNET_DC_WAIT 2
899 struct lnet_rtrbufpool {
900 /* my free buffer pool */
901 struct list_head rbp_bufs;
902 /* messages blocking for a buffer */
903 struct list_head rbp_msgs;
904 /* # pages in each buffer */
906 /* requested number of buffers */
907 int rbp_req_nbuffers;
908 /* # buffers actually allocated */
910 /* # free buffers / blocked messages */
917 struct list_head rb_list; /* chain on rbp_bufs */
918 struct lnet_rtrbufpool *rb_pool; /* owning pool */
919 struct bio_vec rb_kiov[0]; /* the buffer space */
922 #define LNET_PEER_HASHSIZE 503 /* prime! */
924 enum lnet_match_flags {
925 /* Didn't match anything */
926 LNET_MATCHMD_NONE = BIT(0),
928 LNET_MATCHMD_OK = BIT(1),
929 /* Must be discarded */
930 LNET_MATCHMD_DROP = BIT(2),
931 /* match and buffer is exhausted */
932 LNET_MATCHMD_EXHAUSTED = BIT(3),
934 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
937 /* Options for struct lnet_portal::ptl_options */
938 #define LNET_PTL_LAZY BIT(0)
939 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
940 #define LNET_PTL_MATCH_WILDCARD BIT(2) /* wildcard match, request portal */
942 /* parameter for matching operations (GET, PUT) */
943 struct lnet_match_info {
945 struct lnet_process_id mi_id;
948 unsigned int mi_portal;
949 unsigned int mi_rlength;
950 unsigned int mi_roffset;
953 /* ME hash of RDMA portal */
954 #define LNET_MT_HASH_BITS 8
955 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
956 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
957 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
958 * the last entry is reserved for MEs with ignore-bits */
959 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
960 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
961 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
962 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
963 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
964 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
965 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
967 /* portal match table */
968 struct lnet_match_table {
969 /* reserved for upcoming patches, CPU partition ID */
971 unsigned int mt_portal; /* portal index */
972 /* match table is set as "enabled" if there's non-exhausted MD
973 * attached on mt_mhash, it's only valid for wildcard portal */
974 unsigned int mt_enabled;
975 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
976 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
977 struct list_head *mt_mhash; /* matching hash */
980 /* these are only useful for wildcard portal */
981 /* Turn off message rotor for wildcard portals */
982 #define LNET_PTL_ROTOR_OFF 0
983 /* round-robin dispatch all PUT messages for wildcard portals */
984 #define LNET_PTL_ROTOR_ON 1
985 /* round-robin dispatch routed PUT message for wildcard portals */
986 #define LNET_PTL_ROTOR_RR_RT 2
987 /* dispatch routed PUT message by hashing source NID for wildcard portals */
988 #define LNET_PTL_ROTOR_HASH_RT 3
992 unsigned int ptl_index; /* portal ID, reserved */
993 /* flags on this portal: lazy, unique... */
994 unsigned int ptl_options;
995 /* list of messages which are stealing buffer */
996 struct list_head ptl_msg_stealing;
997 /* messages blocking for MD */
998 struct list_head ptl_msg_delayed;
999 /* Match table for each CPT */
1000 struct lnet_match_table **ptl_mtables;
1001 /* spread rotor of incoming "PUT" */
1002 unsigned int ptl_rotor;
1003 /* # active entries for this portal */
1005 /* array of active entries' cpu-partition-id */
1009 #define LNET_LH_HASH_BITS 12
1010 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
1011 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
1013 /* resource container (ME, MD, EQ) */
1014 struct lnet_res_container {
1015 unsigned int rec_type; /* container type */
1016 __u64 rec_lh_cookie; /* cookie generator */
1017 struct list_head rec_active; /* active resource list */
1018 struct list_head *rec_lh_hash; /* handle hash */
1021 /* message container */
1022 struct lnet_msg_container {
1023 int msc_init; /* initialized or not */
1024 /* max # threads finalizing */
1025 int msc_nfinalizers;
1026 /* msgs waiting to complete finalizing */
1027 struct list_head msc_finalizing;
1028 /* msgs waiting to be resent */
1029 struct list_head msc_resending;
1030 struct list_head msc_active; /* active message list */
1031 /* threads doing finalization */
1032 void **msc_finalizers;
1033 /* threads doing resends */
1034 void **msc_resenders;
1037 /* This UDSP structures need to match the user space liblnetconfig structures
1038 * in order for the marshall and unmarshall functions to be common.
1041 /* Net is described as a
1045 struct lnet_ud_net_descr {
1047 struct list_head udn_net_num_range;
1050 /* each NID range is defined as
1052 * 2. address range descriptor
1054 struct lnet_ud_nid_descr {
1055 struct lnet_ud_net_descr ud_net_id;
1056 struct list_head ud_addr_range;
1060 /* a UDSP rule can have up to three user defined NID descriptors
1061 * - src: defines the local NID range for the rule
1062 * - dst: defines the peer NID range for the rule
1063 * - rte: defines the router NID range for the rule
1065 * An action union defines the action to take when the rule
1069 struct list_head udsp_on_list;
1071 struct lnet_ud_nid_descr udsp_src;
1072 struct lnet_ud_nid_descr udsp_dst;
1073 struct lnet_ud_nid_descr udsp_rte;
1074 enum lnet_udsp_action_type udsp_action_type;
1076 __u32 udsp_priority;
1080 /* Peer Discovery states */
1081 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
1082 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
1083 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
1085 /* Router Checker states */
1086 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
1087 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
1088 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
1091 #define LNET_STATE_SHUTDOWN 0 /* not started */
1092 #define LNET_STATE_RUNNING 1 /* started up OK */
1093 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
1096 /* CPU partition table of LNet */
1097 struct cfs_cpt_table *ln_cpt_table;
1098 /* number of CPTs in ln_cpt_table */
1099 unsigned int ln_cpt_number;
1100 unsigned int ln_cpt_bits;
1102 /* protect LNet resources (ME/MD/EQ) */
1103 struct cfs_percpt_lock *ln_res_lock;
1106 /* the vector of portals */
1107 struct lnet_portal **ln_portals;
1108 /* percpt MD container */
1109 struct lnet_res_container **ln_md_containers;
1111 /* Event Queue container */
1112 struct lnet_res_container ln_eq_container;
1113 spinlock_t ln_eq_wait_lock;
1115 unsigned int ln_remote_nets_hbits;
1117 /* protect NI, peer table, credits, routers, rtrbuf... */
1118 struct cfs_percpt_lock *ln_net_lock;
1119 /* percpt message containers for active/finalizing/freed message */
1120 struct lnet_msg_container **ln_msg_containers;
1121 struct lnet_counters **ln_counters;
1122 struct lnet_peer_table **ln_peer_tables;
1123 /* list of peer nis not on a local network */
1124 struct list_head ln_remote_peer_ni_list;
1125 /* failure simulation */
1126 struct list_head ln_test_peers;
1127 struct list_head ln_drop_rules;
1128 struct list_head ln_delay_rules;
1130 struct list_head ln_nets;
1131 /* the loopback NI */
1132 struct lnet_ni *ln_loni;
1133 /* network zombie list */
1134 struct list_head ln_net_zombie;
1135 /* resend messages list */
1136 struct list_head ln_msg_resend;
1137 /* spin lock to protect the msg resend list */
1138 spinlock_t ln_msg_resend_lock;
1140 /* remote networks with routes to them */
1141 struct list_head *ln_remote_nets_hash;
1142 /* validity stamp */
1143 __u64 ln_remote_nets_version;
1144 /* list of all known routers */
1145 struct list_head ln_routers;
1146 /* validity stamp */
1147 __u64 ln_routers_version;
1148 /* percpt router buffer pools */
1149 struct lnet_rtrbufpool **ln_rtrpools;
1152 * Ping target / Push source
1154 * The ping target and push source share a single buffer. The
1155 * ln_ping_target is protected against concurrent updates by
1158 struct lnet_handle_md ln_ping_target_md;
1159 lnet_handler_t ln_ping_target_handler;
1160 struct lnet_ping_buffer *ln_ping_target;
1161 atomic_t ln_ping_target_seqno;
1166 * ln_push_nnis contains the desired size of the push target.
1167 * The lnet_net_lock is used to handle update races. The old
1168 * buffer may linger a while after it has been unlinked, in
1169 * which case the event handler cleans up.
1171 lnet_handler_t ln_push_target_handler;
1172 struct lnet_handle_md ln_push_target_md;
1173 struct lnet_ping_buffer *ln_push_target;
1174 int ln_push_target_nnis;
1176 /* discovery event queue handle */
1177 lnet_handler_t ln_dc_handler;
1178 /* discovery requests */
1179 struct list_head ln_dc_request;
1180 /* discovery working list */
1181 struct list_head ln_dc_working;
1182 /* discovery expired list */
1183 struct list_head ln_dc_expired;
1184 /* discovery thread wait queue */
1185 wait_queue_head_t ln_dc_waitq;
1186 /* discovery startup/shutdown state */
1189 /* monitor thread startup/shutdown state */
1191 /* serialise startup/shutdown */
1192 struct semaphore ln_mt_signal;
1194 struct mutex ln_api_mutex;
1195 struct mutex ln_lnd_mutex;
1196 /* Have I called LNetNIInit myself? */
1198 /* LNetNIInit/LNetNIFini counter */
1200 /* SHUTDOWN/RUNNING/STOPPING */
1203 int ln_routing; /* am I a router? */
1204 lnet_pid_t ln_pid; /* requested pid */
1205 /* uniquely identifies this ni in this epoch */
1206 __u64 ln_interface_cookie;
1207 /* registered LNDs */
1208 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1210 /* test protocol compatibility flags */
1211 unsigned long ln_testprotocompat;
1213 /* 0 - load the NIs from the mod params
1214 * 1 - do not load the NIs from the mod params
1215 * Reverse logic to ensure that other calls to LNetNIInit
1218 bool ln_nis_from_mod_params;
1221 * completion for the monitor thread. The monitor thread takes care of
1222 * checking routes, timedout messages and resending messages.
1224 struct completion ln_mt_wait_complete;
1226 /* per-cpt resend queues */
1227 struct list_head **ln_mt_resendqs;
1228 /* local NIs to recover */
1229 struct list_head ln_mt_localNIRecovq;
1230 /* local NIs to recover */
1231 struct list_head ln_mt_peerNIRecovq;
1233 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1234 * There are CPT number of queues. Since response trackers will be
1235 * added on the fast path we can't afford to grab the exclusive
1236 * net lock to protect these queues. The CPT will be calculated
1237 * based on the mdh cookie.
1239 struct list_head **ln_mt_rstq;
1241 * A response tracker becomes a zombie when the associated MD is queued
1242 * for unlink before the response tracker is detached from the MD. An
1243 * entry on a zombie list can be freed when either the remaining
1244 * operations on the MD complete or when LNet has shut down.
1246 struct list_head **ln_mt_zombie_rstqs;
1247 /* recovery handler */
1248 lnet_handler_t ln_mt_handler;
1251 * Completed when the discovery and monitor threads can enter their
1254 struct completion ln_started;
1256 struct list_head ln_udsp_list;
1259 static const struct nla_policy scalar_attr_policy[LN_SCALAR_CNT + 1] = {
1260 [LN_SCALAR_ATTR_LIST] = { .type = NLA_NESTED },
1261 [LN_SCALAR_ATTR_LIST_SIZE] = { .type = NLA_U16 },
1262 [LN_SCALAR_ATTR_INDEX] = { .type = NLA_U16 },
1263 [LN_SCALAR_ATTR_NLA_TYPE] = { .type = NLA_U16 },
1264 [LN_SCALAR_ATTR_VALUE] = { .type = NLA_STRING },
1265 [LN_SCALAR_ATTR_KEY_FORMAT] = { .type = NLA_U16 },
1268 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
1269 const struct genl_family *family, int flags,
1270 u8 cmd, const struct ln_key_list *data[]);
1272 /* Special workaround for pre-4.19 kernels to send error messages
1273 * from dumpit routines. Newer kernels will send message with
1274 * NL_SET_ERR_MSG information by default if NETLINK_EXT_ACK is set.
1276 static inline int lnet_nl_send_error(struct sk_buff *msg, int portid, int seq,
1279 #ifndef HAVE_NL_DUMP_WITH_EXT_ACK
1280 struct nlmsghdr *nlh;
1285 nlh = nlmsg_put(msg, portid, seq, NLMSG_ERROR, sizeof(error), 0);
1288 #ifdef HAVE_NL_PARSE_WITH_EXT_ACK
1289 netlink_ack(msg, nlh, error, NULL);
1291 netlink_ack(msg, nlh, error);
1293 return nlmsg_len(nlh);