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 /* force an RDMA even if the message size is < 4K */
130 /* the number of times a transmission has been retried */
132 /* flag to indicate that we do not want to resend this message */
135 /* committed for sending */
136 unsigned int msg_tx_committed:1;
137 /* CPT # this message committed for sending */
138 unsigned int msg_tx_cpt:15;
139 /* committed for receiving */
140 unsigned int msg_rx_committed:1;
141 /* CPT # this message committed for receiving */
142 unsigned int msg_rx_cpt:15;
143 /* queued for tx credit */
144 unsigned int msg_tx_delayed:1;
145 /* queued for RX buffer */
146 unsigned int msg_rx_delayed:1;
147 /* ready for pending on RX delay list */
148 unsigned int msg_rx_ready_delay:1;
150 unsigned int msg_vmflush:1; /* VM trying to free memory */
151 unsigned int msg_target_is_router:1; /* sending to a router */
152 unsigned int msg_routing:1; /* being forwarded */
153 unsigned int msg_ack:1; /* ack on finalize (PUT) */
154 unsigned int msg_sending:1; /* outgoing message */
155 unsigned int msg_receiving:1; /* being received */
156 unsigned int msg_txcredit:1; /* taken an NI send credit */
157 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
158 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
159 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
160 unsigned int msg_onactivelist:1; /* on the activelist */
161 unsigned int msg_rdma_get:1;
163 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
164 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
167 struct lnet_libmd *msg_md;
168 /* the NI the message was sent or received over */
169 struct lnet_ni *msg_txni;
170 struct lnet_ni *msg_rxni;
172 unsigned int msg_len;
173 unsigned int msg_wanted;
174 unsigned int msg_offset;
175 unsigned int msg_niov;
176 struct bio_vec *msg_kiov;
178 struct lnet_event msg_ev;
179 struct lnet_hdr msg_hdr;
182 struct lnet_libhandle {
183 struct list_head lh_hash_chain;
187 #define lh_entry(ptr, type, member) \
188 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
191 struct list_head me_list;
193 struct lnet_process_id me_match_id;
194 unsigned int me_portal;
195 unsigned int me_pos; /* hash offset in mt_hash */
197 __u64 me_ignore_bits;
198 enum lnet_unlink me_unlink;
199 struct lnet_libmd *me_md;
203 struct list_head md_list;
204 struct lnet_libhandle md_lh;
205 struct lnet_me *md_me;
207 unsigned int md_offset;
208 unsigned int md_length;
209 unsigned int md_max_size;
212 unsigned int md_options;
213 unsigned int md_flags;
214 unsigned int md_niov; /* # frags at end of struct */
216 struct lnet_rsp_tracker *md_rspt_ptr;
217 lnet_handler_t md_handler;
218 struct lnet_handle_md md_bulk_handle;
219 struct bio_vec md_kiov[LNET_MAX_IOV];
222 #define LNET_MD_FLAG_ZOMBIE BIT(0)
223 #define LNET_MD_FLAG_AUTO_UNLINK BIT(1)
224 #define LNET_MD_FLAG_ABORTED BIT(2)
225 /* LNET_MD_FLAG_HANDLING is set when a non-unlink event handler
226 * is being called for an event relating to the md.
227 * It ensures only one such handler runs at a time.
228 * The final "unlink" event is only called once the
229 * md_refcount has reached zero, and this flag has been cleared,
230 * ensuring that it doesn't race with any other event handler
233 #define LNET_MD_FLAG_HANDLING BIT(3)
234 #define LNET_MD_FLAG_DISCARD BIT(4)
236 struct lnet_test_peer {
237 /* info about peers we are trying to fail */
238 struct list_head tp_list; /* ln_test_peers */
239 lnet_nid_t tp_nid; /* matching nid */
240 unsigned int tp_threshold; /* # failures to simulate */
243 #define LNET_COOKIE_TYPE_MD 1
244 #define LNET_COOKIE_TYPE_ME 2
245 #define LNET_COOKIE_TYPE_EQ 3
246 #define LNET_COOKIE_TYPE_BITS 2
247 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
253 void (*nf_addr2str)(u32 addr, char *str, size_t size);
254 int (*nf_str2addr)(const char *str, int nob, u32 *addr);
255 int (*nf_parse_addrlist)(char *str, int len,
256 struct list_head *list);
257 int (*nf_print_addrlist)(char *buffer, int count,
258 struct list_head *list);
259 int (*nf_match_addr)(u32 addr, struct list_head *list);
260 int (*nf_min_max)(struct list_head *nidlist, u32 *min_nid,
264 struct lnet_ni; /* forward ref */
268 /* fields initialized by the LND */
271 int (*lnd_startup)(struct lnet_ni *ni);
272 void (*lnd_shutdown)(struct lnet_ni *ni);
273 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
275 /* In data movement APIs below, payload buffers are described as a set
276 * of 'niov' fragments which are in pages.
277 * The LND may NOT overwrite these fragment descriptors.
278 * An 'offset' and may specify a byte offset within the set of
279 * fragments to start from
282 /* Start sending a preformatted message. 'private' is NULL for PUT and
283 * GET messages; otherwise this is a response to an incoming message
284 * and 'private' is the 'private' passed to lnet_parse(). Return
285 * non-zero for immediate failure, otherwise complete later with
287 int (*lnd_send)(struct lnet_ni *ni, void *private,
288 struct lnet_msg *msg);
290 /* Start receiving 'mlen' bytes of payload data, skipping the following
291 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
292 * lnet_parse(). Return non-zero for immedaite failure, otherwise
293 * complete later with lnet_finalize(). This also gives back a receive
294 * credit if the LND does flow control. */
295 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
296 int delayed, unsigned int niov,
297 struct bio_vec *kiov,
298 unsigned int offset, unsigned int mlen, unsigned int rlen);
300 /* lnet_parse() has had to delay processing of this message
301 * (e.g. waiting for a forwarding buffer or send credits). Give the
302 * LND a chance to free urgently needed resources. If called, return 0
303 * for success and do NOT give back a receive credit; that has to wait
304 * until lnd_recv() gets called. On failure return < 0 and
305 * release resources; lnd_recv() will not be called. */
306 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
307 struct lnet_msg *msg, void **new_privatep);
309 /* notification of peer down */
310 void (*lnd_notify_peer_down)(lnet_nid_t peer);
312 /* accept a new connection */
313 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
316 struct lnet_tx_queue {
317 int tq_credits; /* # tx credits free */
318 int tq_credits_min; /* lowest it's been */
319 int tq_credits_max; /* total # tx credits */
320 struct list_head tq_delayed; /* delayed TXs */
323 enum lnet_net_state {
324 /* set when net block is allocated */
325 LNET_NET_STATE_INIT = 0,
326 /* set when NIs in net are started successfully */
327 LNET_NET_STATE_ACTIVE,
328 /* set if all NIs in net are in FAILED state */
329 LNET_NET_STATE_INACTIVE,
330 /* set when shutting down a NET */
331 LNET_NET_STATE_DELETING
335 /* initial state when NI is created */
336 LNET_NI_STATE_INIT = 0,
337 /* set when NI is brought up */
338 LNET_NI_STATE_ACTIVE,
339 /* set when NI is being shutdown */
340 LNET_NI_STATE_DELETING,
343 #define LNET_NI_RECOVERY_PENDING BIT(0)
344 #define LNET_NI_RECOVERY_FAILED BIT(1)
346 enum lnet_stats_type {
347 LNET_STATS_TYPE_SEND = 0,
348 LNET_STATS_TYPE_RECV,
352 struct lnet_comm_count {
353 atomic_t co_get_count;
354 atomic_t co_put_count;
355 atomic_t co_reply_count;
356 atomic_t co_ack_count;
357 atomic_t co_hello_count;
360 struct lnet_element_stats {
361 struct lnet_comm_count el_send_stats;
362 struct lnet_comm_count el_recv_stats;
363 struct lnet_comm_count el_drop_stats;
366 struct lnet_health_local_stats {
367 atomic_t hlt_local_interrupt;
368 atomic_t hlt_local_dropped;
369 atomic_t hlt_local_aborted;
370 atomic_t hlt_local_no_route;
371 atomic_t hlt_local_timeout;
372 atomic_t hlt_local_error;
375 struct lnet_health_remote_stats {
376 atomic_t hlt_remote_dropped;
377 atomic_t hlt_remote_timeout;
378 atomic_t hlt_remote_error;
379 atomic_t hlt_network_timeout;
383 /* chain on the ln_nets */
384 struct list_head net_list;
386 /* net ID, which is composed of
387 * (net_type << 16) | net_num.
388 * net_type can be one of the enumerated types defined in
389 * lnet/include/lnet/nidstr.h */
392 /* round robin selection */
395 /* total number of CPTs in the array */
398 /* cumulative CPTs of all NIs in this net */
401 /* relative net selection priority */
402 __u32 net_sel_priority;
404 /* network tunables */
405 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
408 * boolean to indicate that the tunables have been set and
411 bool net_tunables_set;
413 /* procedural interface */
414 const struct lnet_lnd *net_lnd;
416 /* list of NIs on this net */
417 struct list_head net_ni_list;
419 /* list of NIs being added, but not started yet */
420 struct list_head net_ni_added;
422 /* dying LND instances */
423 struct list_head net_ni_zombie;
425 /* when I was last alive */
426 time64_t net_last_alive;
428 /* protects access to net_last_alive */
431 /* list of router nids preferred for this network */
432 struct list_head net_rtr_pref_nids;
436 /* chain on the lnet_net structure */
437 struct list_head ni_netlist;
439 /* chain on the recovery queue */
440 struct list_head ni_recovery;
442 /* MD handle for recovery ping */
443 struct lnet_handle_md ni_ping_mdh;
450 /* bond NI on some CPTs */
453 /* interface's NID */
456 /* instance-specific data */
460 atomic_t ni_tx_credits;
462 /* percpt TX queues */
463 struct lnet_tx_queue **ni_tx_queues;
465 /* percpt reference count */
468 /* pointer to parent network */
469 struct lnet_net *ni_net;
471 /* my health status */
472 struct lnet_ni_status *ni_status;
474 /* NI FSM. Protected by lnet_ni_lock() */
475 enum lnet_ni_state ni_state;
477 /* Recovery state. Protected by lnet_ni_lock() */
478 __u32 ni_recovery_state;
480 /* When to send the next recovery ping */
481 time64_t ni_next_ping;
482 /* How many pings sent during current recovery period did not receive
483 * a reply. NB: reset whenever _any_ message arrives on this NI
485 unsigned int ni_ping_count;
487 /* per NI LND tunables */
488 struct lnet_lnd_tunables ni_lnd_tunables;
490 /* lnd tunables set explicitly */
491 bool ni_lnd_tunables_set;
494 struct lnet_element_stats ni_stats;
495 struct lnet_health_local_stats ni_hstats;
497 /* physical device CPT */
500 /* sequence number used to round robin over nis within a net */
505 * initialized to LNET_MAX_HEALTH_VALUE
506 * Value is decremented every time we fail to send a message over
507 * this NI because of a NI specific failure.
508 * Value is incremented if we successfully send a message.
513 * Set to 1 by the LND when it receives an event telling it the device
514 * has gone into a fatal state. Set to 0 when the LND receives an
515 * even telling it the device is back online.
517 atomic_t ni_fatal_error_on;
519 /* the relative selection priority of this NI */
520 __u32 ni_sel_priority;
523 * equivalent interface to use
526 struct net *ni_net_ns; /* original net namespace */
529 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
532 * Descriptor of a ping info buffer: keep a separate indicator of the
533 * size and a reference count. The type is used both as a source and
534 * sink of data, so we need to keep some information outside of the
535 * area that may be overwritten by network data.
537 struct lnet_ping_buffer {
541 struct lnet_ping_info pb_info;
544 #define LNET_PING_BUFFER_SIZE(NNIDS) \
545 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
546 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
547 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
549 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
550 container_of((PINFO), struct lnet_ping_buffer, pb_info)
552 struct lnet_nid_list {
553 struct list_head nl_list;
557 struct lnet_peer_ni {
558 /* chain on lpn_peer_nis */
559 struct list_head lpni_peer_nis;
560 /* chain on remote peer list */
561 struct list_head lpni_on_remote_peer_ni_list;
562 /* chain on recovery queue */
563 struct list_head lpni_recovery;
564 /* chain on peer hash */
565 struct list_head lpni_hashlist;
566 /* messages blocking for tx credits */
567 struct list_head lpni_txq;
568 /* pointer to peer net I'm part of */
569 struct lnet_peer_net *lpni_peer_net;
570 /* statistics kept on each peer NI */
571 struct lnet_element_stats lpni_stats;
572 struct lnet_health_remote_stats lpni_hstats;
573 /* spin lock protecting credits and lpni_txq */
574 spinlock_t lpni_lock;
575 /* # tx credits available */
578 int lpni_mintxcredits;
580 * Each peer_ni in a gateway maintains its own credits. This
581 * allows more traffic to gateways that have multiple interfaces.
583 /* # router credits */
586 int lpni_minrtrcredits;
587 /* bytes queued for sending */
589 /* network peer is on */
590 struct lnet_net *lpni_net;
594 struct kref lpni_kref;
595 /* health value for the peer */
596 atomic_t lpni_healthv;
597 /* recovery ping mdh */
598 struct lnet_handle_md lpni_recovery_ping_mdh;
599 /* When to send the next recovery ping */
600 time64_t lpni_next_ping;
601 /* How many pings sent during current recovery period did not receive
602 * a reply. NB: reset whenever _any_ message arrives from this peer NI
604 unsigned int lpni_ping_count;
605 /* CPT this peer attached on */
607 /* state flags -- protected by lpni_lock */
609 /* status of the peer NI as reported by the peer */
610 __u32 lpni_ns_status;
611 /* sequence number used to round robin over peer nis within a net */
613 /* sequence number used to round robin over gateways */
615 /* returned RC ping features. Protected with lpni_lock */
616 unsigned int lpni_ping_feats;
617 /* time last message was received from the peer */
618 time64_t lpni_last_alive;
619 /* preferred local nids: if only one, use lpni_pref.nid */
622 struct list_head nids;
624 /* list of router nids preferred for this peer NI */
625 struct list_head lpni_rtr_pref_nids;
626 /* The relative selection priority of this peer NI */
627 __u32 lpni_sel_priority;
628 /* number of preferred NIDs in lnpi_pref_nids */
629 __u32 lpni_pref_nnids;
632 /* Preferred path added due to traffic on non-MR peer_ni */
633 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
634 /* peer is being recovered. */
635 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
636 /* recovery ping failed */
637 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
638 /* peer is being deleted */
639 #define LNET_PEER_NI_DELETING BIT(3)
642 /* chain on pt_peer_list */
643 struct list_head lp_peer_list;
645 /* list of peer nets */
646 struct list_head lp_peer_nets;
648 /* list of messages pending discovery*/
649 struct list_head lp_dc_pendq;
651 /* chain on router list */
652 struct list_head lp_rtr_list;
654 /* primary NID of the peer */
655 lnet_nid_t lp_primary_nid;
657 /* source NID to use during discovery */
658 lnet_nid_t lp_disc_src_nid;
659 /* destination NID to use during discovery */
660 lnet_nid_t lp_disc_dst_nid;
662 /* net to perform discovery on */
663 __u32 lp_disc_net_id;
665 /* CPT of peer_table */
668 /* number of NIDs on this peer */
671 /* # refs from lnet_route::lr_gateway */
675 * peer specific health sensitivity value to decrement peer nis in
676 * this peer with if set to something other than 0
678 __u32 lp_health_sensitivity;
680 /* messages blocking for router credits */
681 struct list_head lp_rtrq;
683 /* routes on this peer */
684 struct list_head lp_routes;
686 /* reference count */
687 atomic_t lp_refcount;
689 /* lock protecting peer state flags and lpni_rtrq */
692 /* peer state flags */
695 /* buffer for data pushed by peer */
696 struct lnet_ping_buffer *lp_data;
698 /* MD handle for ping in progress */
699 struct lnet_handle_md lp_ping_mdh;
701 /* MD handle for push in progress */
702 struct lnet_handle_md lp_push_mdh;
704 /* number of NIDs for sizing push data */
707 /* NI config sequence number of peer */
710 /* Local NI config sequence number acked by peer */
713 /* Local NI config sequence number sent to peer */
714 __u32 lp_node_seqno_sent;
716 /* Ping error encountered during discovery. */
719 /* Push error encountered during discovery. */
722 /* Error encountered during discovery. */
725 /* time it was put on the ln_dc_working queue */
726 time64_t lp_last_queued;
728 /* link on discovery-related lists */
729 struct list_head lp_dc_list;
731 /* tasks waiting on discovery of this peer */
732 wait_queue_head_t lp_dc_waitq;
734 /* cached peer aliveness */
739 * The status flags in lp_state. Their semantics have chosen so that
740 * lp_state can be zero-initialized.
742 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
743 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
745 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
746 * NOT set when the peer was pinged by discovery.
748 * A peer is marked ROUTER if it indicates so in the feature bit.
750 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
751 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
752 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
755 * A peer is marked CONFIGURED if it was configured by DLC.
757 * In addition, a peer is marked DISCOVERED if it has fully passed
758 * through Peer Discovery.
760 * When Peer Discovery is disabled, the discovery thread will mark
761 * peers REDISCOVER to indicate that they should be re-examined if
762 * discovery is (re)enabled on the node.
764 * A peer that was created as the result of inbound traffic will not
767 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
768 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
769 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
771 * A peer is marked DISCOVERING when discovery is in progress.
772 * The other flags below correspond to stages of discovery.
774 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
775 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
776 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
777 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
778 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
779 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
780 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
782 * A ping can be forced as a way to fix up state, or as a manual
783 * intervention by an admin.
784 * A push can be forced in circumstances that would normally not
785 * allow for one to happen.
787 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
788 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
790 /* force delete even if router */
791 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
793 /* gw undergoing alive discovery */
794 #define LNET_PEER_RTR_DISCOVERY BIT(16)
795 /* gw has undergone discovery (does not indicate success or failure) */
796 #define LNET_PEER_RTR_DISCOVERED BIT(17)
798 /* peer is marked for deletion */
799 #define LNET_PEER_MARK_DELETION BIT(18)
800 /* lnet_peer_del()/lnet_peer_del_locked() has been called on the peer */
801 #define LNET_PEER_MARK_DELETED BIT(19)
803 struct lnet_peer_net {
804 /* chain on lp_peer_nets */
805 struct list_head lpn_peer_nets;
807 /* list of peer_nis on this network */
808 struct list_head lpn_peer_nis;
810 /* pointer to the peer I'm part of */
811 struct lnet_peer *lpn_peer;
816 /* peer net health */
819 /* time of next router ping on this net */
820 time64_t lpn_next_ping;
822 /* selection sequence number */
825 /* relative peer net selection priority */
826 __u32 lpn_sel_priority;
828 /* reference count */
829 atomic_t lpn_refcount;
833 #define LNET_PEER_HASH_BITS 9
834 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
837 * peer hash table - one per CPT
839 * protected by lnet_net_lock/EX for update
844 * protected by pt_zombie_lock:
848 * pt_zombie lock nests inside lnet_net_lock
850 struct lnet_peer_table {
851 int pt_version; /* /proc validity stamp */
852 struct list_head *pt_hash; /* NID->peer hash */
853 struct list_head pt_peer_list; /* peers */
854 int pt_peers; /* # peers */
855 struct list_head pt_zombie_list; /* zombie peer_ni */
856 int pt_zombies; /* # zombie peers_ni */
857 spinlock_t pt_zombie_lock; /* protect list and count */
860 /* peer aliveness is enabled only on routers for peers in a network where the
861 * struct lnet_ni::ni_peertimeout has been set to a positive value
863 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
864 ((lp)->lpni_net) && \
865 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
868 struct list_head lr_list; /* chain on net */
869 struct list_head lr_gwlist; /* chain on gateway */
870 struct lnet_peer *lr_gateway; /* router node */
871 lnet_nid_t lr_nid; /* NID used to add route */
872 __u32 lr_net; /* remote network number */
873 __u32 lr_lnet; /* local network number */
874 int lr_seq; /* sequence for round-robin */
875 __u32 lr_hops; /* how far I am */
876 unsigned int lr_priority; /* route priority */
877 atomic_t lr_alive; /* cached route aliveness */
878 bool lr_single_hop; /* this route is single-hop */
881 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
882 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
883 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
885 struct lnet_remotenet {
886 /* chain on ln_remote_nets_hash */
887 struct list_head lrn_list;
889 struct list_head lrn_routes;
894 /** lnet message has credit and can be submitted to lnd for send/receive */
895 #define LNET_CREDIT_OK 0
896 /** lnet message is waiting for credit */
897 #define LNET_CREDIT_WAIT 1
898 /** lnet message is waiting for discovery */
899 #define LNET_DC_WAIT 2
901 struct lnet_rtrbufpool {
902 /* my free buffer pool */
903 struct list_head rbp_bufs;
904 /* messages blocking for a buffer */
905 struct list_head rbp_msgs;
906 /* # pages in each buffer */
908 /* requested number of buffers */
909 int rbp_req_nbuffers;
910 /* # buffers actually allocated */
912 /* # free buffers / blocked messages */
919 struct list_head rb_list; /* chain on rbp_bufs */
920 struct lnet_rtrbufpool *rb_pool; /* owning pool */
921 struct bio_vec rb_kiov[0]; /* the buffer space */
924 #define LNET_PEER_HASHSIZE 503 /* prime! */
926 enum lnet_match_flags {
927 /* Didn't match anything */
928 LNET_MATCHMD_NONE = BIT(0),
930 LNET_MATCHMD_OK = BIT(1),
931 /* Must be discarded */
932 LNET_MATCHMD_DROP = BIT(2),
933 /* match and buffer is exhausted */
934 LNET_MATCHMD_EXHAUSTED = BIT(3),
936 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
939 /* Options for struct lnet_portal::ptl_options */
940 #define LNET_PTL_LAZY BIT(0)
941 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
942 #define LNET_PTL_MATCH_WILDCARD BIT(2) /* wildcard match, request portal */
944 /* parameter for matching operations (GET, PUT) */
945 struct lnet_match_info {
947 struct lnet_process_id mi_id;
950 unsigned int mi_portal;
951 unsigned int mi_rlength;
952 unsigned int mi_roffset;
955 /* ME hash of RDMA portal */
956 #define LNET_MT_HASH_BITS 8
957 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
958 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
959 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
960 * the last entry is reserved for MEs with ignore-bits */
961 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
962 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
963 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
964 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
965 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
966 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
967 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
969 /* portal match table */
970 struct lnet_match_table {
971 /* reserved for upcoming patches, CPU partition ID */
973 unsigned int mt_portal; /* portal index */
974 /* match table is set as "enabled" if there's non-exhausted MD
975 * attached on mt_mhash, it's only valid for wildcard portal */
976 unsigned int mt_enabled;
977 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
978 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
979 struct list_head *mt_mhash; /* matching hash */
982 /* these are only useful for wildcard portal */
983 /* Turn off message rotor for wildcard portals */
984 #define LNET_PTL_ROTOR_OFF 0
985 /* round-robin dispatch all PUT messages for wildcard portals */
986 #define LNET_PTL_ROTOR_ON 1
987 /* round-robin dispatch routed PUT message for wildcard portals */
988 #define LNET_PTL_ROTOR_RR_RT 2
989 /* dispatch routed PUT message by hashing source NID for wildcard portals */
990 #define LNET_PTL_ROTOR_HASH_RT 3
994 unsigned int ptl_index; /* portal ID, reserved */
995 /* flags on this portal: lazy, unique... */
996 unsigned int ptl_options;
997 /* list of messages which are stealing buffer */
998 struct list_head ptl_msg_stealing;
999 /* messages blocking for MD */
1000 struct list_head ptl_msg_delayed;
1001 /* Match table for each CPT */
1002 struct lnet_match_table **ptl_mtables;
1003 /* spread rotor of incoming "PUT" */
1004 unsigned int ptl_rotor;
1005 /* # active entries for this portal */
1007 /* array of active entries' cpu-partition-id */
1011 #define LNET_LH_HASH_BITS 12
1012 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
1013 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
1015 /* resource container (ME, MD, EQ) */
1016 struct lnet_res_container {
1017 unsigned int rec_type; /* container type */
1018 __u64 rec_lh_cookie; /* cookie generator */
1019 struct list_head rec_active; /* active resource list */
1020 struct list_head *rec_lh_hash; /* handle hash */
1023 /* message container */
1024 struct lnet_msg_container {
1025 int msc_init; /* initialized or not */
1026 /* max # threads finalizing */
1027 int msc_nfinalizers;
1028 /* msgs waiting to complete finalizing */
1029 struct list_head msc_finalizing;
1030 /* msgs waiting to be resent */
1031 struct list_head msc_resending;
1032 struct list_head msc_active; /* active message list */
1033 /* threads doing finalization */
1034 void **msc_finalizers;
1035 /* threads doing resends */
1036 void **msc_resenders;
1039 /* This UDSP structures need to match the user space liblnetconfig structures
1040 * in order for the marshall and unmarshall functions to be common.
1043 /* Net is described as a
1047 struct lnet_ud_net_descr {
1049 struct list_head udn_net_num_range;
1052 /* each NID range is defined as
1054 * 2. address range descriptor
1056 struct lnet_ud_nid_descr {
1057 struct lnet_ud_net_descr ud_net_id;
1058 struct list_head ud_addr_range;
1062 /* a UDSP rule can have up to three user defined NID descriptors
1063 * - src: defines the local NID range for the rule
1064 * - dst: defines the peer NID range for the rule
1065 * - rte: defines the router NID range for the rule
1067 * An action union defines the action to take when the rule
1071 struct list_head udsp_on_list;
1073 struct lnet_ud_nid_descr udsp_src;
1074 struct lnet_ud_nid_descr udsp_dst;
1075 struct lnet_ud_nid_descr udsp_rte;
1076 enum lnet_udsp_action_type udsp_action_type;
1078 __u32 udsp_priority;
1082 /* Peer Discovery states */
1083 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
1084 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
1085 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
1087 /* Router Checker states */
1088 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
1089 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
1090 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
1093 #define LNET_STATE_SHUTDOWN 0 /* not started */
1094 #define LNET_STATE_RUNNING 1 /* started up OK */
1095 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
1098 /* CPU partition table of LNet */
1099 struct cfs_cpt_table *ln_cpt_table;
1100 /* number of CPTs in ln_cpt_table */
1101 unsigned int ln_cpt_number;
1102 unsigned int ln_cpt_bits;
1104 /* protect LNet resources (ME/MD/EQ) */
1105 struct cfs_percpt_lock *ln_res_lock;
1108 /* the vector of portals */
1109 struct lnet_portal **ln_portals;
1110 /* percpt MD container */
1111 struct lnet_res_container **ln_md_containers;
1113 /* Event Queue container */
1114 struct lnet_res_container ln_eq_container;
1115 spinlock_t ln_eq_wait_lock;
1117 unsigned int ln_remote_nets_hbits;
1119 /* protect NI, peer table, credits, routers, rtrbuf... */
1120 struct cfs_percpt_lock *ln_net_lock;
1121 /* percpt message containers for active/finalizing/freed message */
1122 struct lnet_msg_container **ln_msg_containers;
1123 struct lnet_counters **ln_counters;
1124 struct lnet_peer_table **ln_peer_tables;
1125 /* list of peer nis not on a local network */
1126 struct list_head ln_remote_peer_ni_list;
1127 /* failure simulation */
1128 struct list_head ln_test_peers;
1129 struct list_head ln_drop_rules;
1130 struct list_head ln_delay_rules;
1132 struct list_head ln_nets;
1133 /* the loopback NI */
1134 struct lnet_ni *ln_loni;
1135 /* network zombie list */
1136 struct list_head ln_net_zombie;
1137 /* resend messages list */
1138 struct list_head ln_msg_resend;
1139 /* spin lock to protect the msg resend list */
1140 spinlock_t ln_msg_resend_lock;
1142 /* remote networks with routes to them */
1143 struct list_head *ln_remote_nets_hash;
1144 /* validity stamp */
1145 __u64 ln_remote_nets_version;
1146 /* list of all known routers */
1147 struct list_head ln_routers;
1148 /* validity stamp */
1149 __u64 ln_routers_version;
1150 /* percpt router buffer pools */
1151 struct lnet_rtrbufpool **ln_rtrpools;
1154 * Ping target / Push source
1156 * The ping target and push source share a single buffer. The
1157 * ln_ping_target is protected against concurrent updates by
1160 struct lnet_handle_md ln_ping_target_md;
1161 lnet_handler_t ln_ping_target_handler;
1162 struct lnet_ping_buffer *ln_ping_target;
1163 atomic_t ln_ping_target_seqno;
1168 * ln_push_nnis contains the desired size of the push target.
1169 * The lnet_net_lock is used to handle update races. The old
1170 * buffer may linger a while after it has been unlinked, in
1171 * which case the event handler cleans up.
1173 lnet_handler_t ln_push_target_handler;
1174 struct lnet_handle_md ln_push_target_md;
1175 struct lnet_ping_buffer *ln_push_target;
1176 int ln_push_target_nnis;
1178 /* discovery event queue handle */
1179 lnet_handler_t ln_dc_handler;
1180 /* discovery requests */
1181 struct list_head ln_dc_request;
1182 /* discovery working list */
1183 struct list_head ln_dc_working;
1184 /* discovery expired list */
1185 struct list_head ln_dc_expired;
1186 /* discovery thread wait queue */
1187 wait_queue_head_t ln_dc_waitq;
1188 /* discovery startup/shutdown state */
1191 /* monitor thread startup/shutdown state */
1193 /* serialise startup/shutdown */
1194 struct semaphore ln_mt_signal;
1196 struct mutex ln_api_mutex;
1197 struct mutex ln_lnd_mutex;
1198 /* Have I called LNetNIInit myself? */
1200 /* LNetNIInit/LNetNIFini counter */
1202 /* SHUTDOWN/RUNNING/STOPPING */
1205 int ln_routing; /* am I a router? */
1206 lnet_pid_t ln_pid; /* requested pid */
1207 /* uniquely identifies this ni in this epoch */
1208 __u64 ln_interface_cookie;
1209 /* registered LNDs */
1210 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1212 /* test protocol compatibility flags */
1213 unsigned long ln_testprotocompat;
1215 /* 0 - load the NIs from the mod params
1216 * 1 - do not load the NIs from the mod params
1217 * Reverse logic to ensure that other calls to LNetNIInit
1220 bool ln_nis_from_mod_params;
1223 * completion for the monitor thread. The monitor thread takes care of
1224 * checking routes, timedout messages and resending messages.
1226 struct completion ln_mt_wait_complete;
1228 /* per-cpt resend queues */
1229 struct list_head **ln_mt_resendqs;
1230 /* local NIs to recover */
1231 struct list_head ln_mt_localNIRecovq;
1232 /* local NIs to recover */
1233 struct list_head ln_mt_peerNIRecovq;
1235 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1236 * There are CPT number of queues. Since response trackers will be
1237 * added on the fast path we can't afford to grab the exclusive
1238 * net lock to protect these queues. The CPT will be calculated
1239 * based on the mdh cookie.
1241 struct list_head **ln_mt_rstq;
1243 * A response tracker becomes a zombie when the associated MD is queued
1244 * for unlink before the response tracker is detached from the MD. An
1245 * entry on a zombie list can be freed when either the remaining
1246 * operations on the MD complete or when LNet has shut down.
1248 struct list_head **ln_mt_zombie_rstqs;
1249 /* recovery handler */
1250 lnet_handler_t ln_mt_handler;
1253 * Completed when the discovery and monitor threads can enter their
1256 struct completion ln_started;
1258 struct list_head ln_udsp_list;
1261 static const struct nla_policy scalar_attr_policy[LN_SCALAR_CNT + 1] = {
1262 [LN_SCALAR_ATTR_LIST] = { .type = NLA_NESTED },
1263 [LN_SCALAR_ATTR_LIST_SIZE] = { .type = NLA_U16 },
1264 [LN_SCALAR_ATTR_INDEX] = { .type = NLA_U16 },
1265 [LN_SCALAR_ATTR_NLA_TYPE] = { .type = NLA_U16 },
1266 [LN_SCALAR_ATTR_VALUE] = { .type = NLA_STRING },
1267 [LN_SCALAR_ATTR_KEY_FORMAT] = { .type = NLA_U16 },
1270 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
1271 const struct genl_family *family, int flags,
1272 u8 cmd, const struct ln_key_list *data[]);
1274 /* Special workaround for pre-4.19 kernels to send error messages
1275 * from dumpit routines. Newer kernels will send message with
1276 * NL_SET_ERR_MSG information by default if NETLINK_EXT_ACK is set.
1278 static inline int lnet_nl_send_error(struct sk_buff *msg, int portid, int seq,
1281 #ifndef HAVE_NL_DUMP_WITH_EXT_ACK
1282 struct nlmsghdr *nlh;
1287 nlh = nlmsg_put(msg, portid, seq, NLMSG_ERROR, sizeof(error), 0);
1290 #ifdef HAVE_NL_PARSE_WITH_EXT_ACK
1291 netlink_ack(msg, nlh, error, NULL);
1293 netlink_ack(msg, nlh, error);
1295 return nlmsg_len(nlh);