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);
315 /* get dma_dev priority */
316 unsigned int (*lnd_get_dev_prio)(struct lnet_ni *ni,
317 unsigned int dev_idx);
320 struct lnet_tx_queue {
321 int tq_credits; /* # tx credits free */
322 int tq_credits_min; /* lowest it's been */
323 int tq_credits_max; /* total # tx credits */
324 struct list_head tq_delayed; /* delayed TXs */
327 enum lnet_net_state {
328 /* set when net block is allocated */
329 LNET_NET_STATE_INIT = 0,
330 /* set when NIs in net are started successfully */
331 LNET_NET_STATE_ACTIVE,
332 /* set if all NIs in net are in FAILED state */
333 LNET_NET_STATE_INACTIVE,
334 /* set when shutting down a NET */
335 LNET_NET_STATE_DELETING
339 /* initial state when NI is created */
340 LNET_NI_STATE_INIT = 0,
341 /* set when NI is brought up */
342 LNET_NI_STATE_ACTIVE,
343 /* set when NI is being shutdown */
344 LNET_NI_STATE_DELETING,
347 #define LNET_NI_RECOVERY_PENDING BIT(0)
348 #define LNET_NI_RECOVERY_FAILED BIT(1)
350 enum lnet_stats_type {
351 LNET_STATS_TYPE_SEND = 0,
352 LNET_STATS_TYPE_RECV,
356 struct lnet_comm_count {
357 atomic_t co_get_count;
358 atomic_t co_put_count;
359 atomic_t co_reply_count;
360 atomic_t co_ack_count;
361 atomic_t co_hello_count;
364 struct lnet_element_stats {
365 struct lnet_comm_count el_send_stats;
366 struct lnet_comm_count el_recv_stats;
367 struct lnet_comm_count el_drop_stats;
370 struct lnet_health_local_stats {
371 atomic_t hlt_local_interrupt;
372 atomic_t hlt_local_dropped;
373 atomic_t hlt_local_aborted;
374 atomic_t hlt_local_no_route;
375 atomic_t hlt_local_timeout;
376 atomic_t hlt_local_error;
379 struct lnet_health_remote_stats {
380 atomic_t hlt_remote_dropped;
381 atomic_t hlt_remote_timeout;
382 atomic_t hlt_remote_error;
383 atomic_t hlt_network_timeout;
387 /* chain on the ln_nets */
388 struct list_head net_list;
390 /* net ID, which is composed of
391 * (net_type << 16) | net_num.
392 * net_type can be one of the enumerated types defined in
393 * lnet/include/lnet/nidstr.h */
396 /* round robin selection */
399 /* total number of CPTs in the array */
402 /* cumulative CPTs of all NIs in this net */
405 /* relative net selection priority */
406 __u32 net_sel_priority;
408 /* network tunables */
409 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
412 * boolean to indicate that the tunables have been set and
415 bool net_tunables_set;
417 /* procedural interface */
418 const struct lnet_lnd *net_lnd;
420 /* list of NIs on this net */
421 struct list_head net_ni_list;
423 /* list of NIs being added, but not started yet */
424 struct list_head net_ni_added;
426 /* dying LND instances */
427 struct list_head net_ni_zombie;
429 /* when I was last alive */
430 time64_t net_last_alive;
432 /* protects access to net_last_alive */
435 /* list of router nids preferred for this network */
436 struct list_head net_rtr_pref_nids;
440 /* chain on the lnet_net structure */
441 struct list_head ni_netlist;
443 /* chain on the recovery queue */
444 struct list_head ni_recovery;
446 /* MD handle for recovery ping */
447 struct lnet_handle_md ni_ping_mdh;
454 /* bond NI on some CPTs */
457 /* interface's NID */
458 struct lnet_nid ni_nid;
460 /* instance-specific data */
464 atomic_t ni_tx_credits;
466 /* percpt TX queues */
467 struct lnet_tx_queue **ni_tx_queues;
469 /* percpt reference count */
472 /* pointer to parent network */
473 struct lnet_net *ni_net;
475 /* my health status */
476 struct lnet_ni_status *ni_status;
478 /* NI FSM. Protected by lnet_ni_lock() */
479 enum lnet_ni_state ni_state;
481 /* Recovery state. Protected by lnet_ni_lock() */
482 __u32 ni_recovery_state;
484 /* When to send the next recovery ping */
485 time64_t ni_next_ping;
486 /* How many pings sent during current recovery period did not receive
487 * a reply. NB: reset whenever _any_ message arrives on this NI
489 unsigned int ni_ping_count;
491 /* per NI LND tunables */
492 struct lnet_lnd_tunables ni_lnd_tunables;
494 /* lnd tunables set explicitly */
495 bool ni_lnd_tunables_set;
498 struct lnet_element_stats ni_stats;
499 struct lnet_health_local_stats ni_hstats;
501 /* physical device CPT */
504 /* sequence number used to round robin over nis within a net */
509 * initialized to LNET_MAX_HEALTH_VALUE
510 * Value is decremented every time we fail to send a message over
511 * this NI because of a NI specific failure.
512 * Value is incremented if we successfully send a message.
517 * Set to 1 by the LND when it receives an event telling it the device
518 * has gone into a fatal state. Set to 0 when the LND receives an
519 * even telling it the device is back online.
521 atomic_t ni_fatal_error_on;
523 /* the relative selection priority of this NI */
524 __u32 ni_sel_priority;
527 * equivalent interface to use
530 struct net *ni_net_ns; /* original net namespace */
533 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
536 * Descriptor of a ping info buffer: keep a separate indicator of the
537 * size and a reference count. The type is used both as a source and
538 * sink of data, so we need to keep some information outside of the
539 * area that may be overwritten by network data.
541 struct lnet_ping_buffer {
545 struct lnet_ping_info pb_info;
548 #define LNET_PING_BUFFER_SIZE(NNIDS) \
549 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
550 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
551 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
553 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
554 container_of((PINFO), struct lnet_ping_buffer, pb_info)
556 struct lnet_nid_list {
557 struct list_head nl_list;
561 struct lnet_peer_ni {
562 /* chain on lpn_peer_nis */
563 struct list_head lpni_peer_nis;
564 /* chain on remote peer list */
565 struct list_head lpni_on_remote_peer_ni_list;
566 /* chain on recovery queue */
567 struct list_head lpni_recovery;
568 /* chain on peer hash */
569 struct list_head lpni_hashlist;
570 /* messages blocking for tx credits */
571 struct list_head lpni_txq;
572 /* pointer to peer net I'm part of */
573 struct lnet_peer_net *lpni_peer_net;
574 /* statistics kept on each peer NI */
575 struct lnet_element_stats lpni_stats;
576 struct lnet_health_remote_stats lpni_hstats;
577 /* spin lock protecting credits and lpni_txq */
578 spinlock_t lpni_lock;
579 /* # tx credits available */
582 int lpni_mintxcredits;
584 * Each peer_ni in a gateway maintains its own credits. This
585 * allows more traffic to gateways that have multiple interfaces.
587 /* # router credits */
590 int lpni_minrtrcredits;
591 /* bytes queued for sending */
593 /* network peer is on */
594 struct lnet_net *lpni_net;
598 struct kref lpni_kref;
599 /* health value for the peer */
600 atomic_t lpni_healthv;
601 /* recovery ping mdh */
602 struct lnet_handle_md lpni_recovery_ping_mdh;
603 /* When to send the next recovery ping */
604 time64_t lpni_next_ping;
605 /* How many pings sent during current recovery period did not receive
606 * a reply. NB: reset whenever _any_ message arrives from this peer NI
608 unsigned int lpni_ping_count;
609 /* CPT this peer attached on */
611 /* state flags -- protected by lpni_lock */
613 /* status of the peer NI as reported by the peer */
614 __u32 lpni_ns_status;
615 /* sequence number used to round robin over peer nis within a net */
617 /* sequence number used to round robin over gateways */
619 /* returned RC ping features. Protected with lpni_lock */
620 unsigned int lpni_ping_feats;
621 /* time last message was received from the peer */
622 time64_t lpni_last_alive;
623 /* preferred local nids: if only one, use lpni_pref.nid */
626 struct list_head nids;
628 /* list of router nids preferred for this peer NI */
629 struct list_head lpni_rtr_pref_nids;
630 /* The relative selection priority of this peer NI */
631 __u32 lpni_sel_priority;
632 /* number of preferred NIDs in lnpi_pref_nids */
633 __u32 lpni_pref_nnids;
636 /* Preferred path added due to traffic on non-MR peer_ni */
637 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
638 /* peer is being recovered. */
639 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
640 /* recovery ping failed */
641 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
642 /* peer is being deleted */
643 #define LNET_PEER_NI_DELETING BIT(3)
646 /* chain on pt_peer_list */
647 struct list_head lp_peer_list;
649 /* list of peer nets */
650 struct list_head lp_peer_nets;
652 /* list of messages pending discovery*/
653 struct list_head lp_dc_pendq;
655 /* chain on router list */
656 struct list_head lp_rtr_list;
658 /* primary NID of the peer */
659 lnet_nid_t lp_primary_nid;
661 /* source NID to use during discovery */
662 lnet_nid_t lp_disc_src_nid;
663 /* destination NID to use during discovery */
664 lnet_nid_t lp_disc_dst_nid;
666 /* net to perform discovery on */
667 __u32 lp_disc_net_id;
669 /* CPT of peer_table */
672 /* number of NIDs on this peer */
675 /* # refs from lnet_route::lr_gateway */
679 * peer specific health sensitivity value to decrement peer nis in
680 * this peer with if set to something other than 0
682 __u32 lp_health_sensitivity;
684 /* messages blocking for router credits */
685 struct list_head lp_rtrq;
687 /* routes on this peer */
688 struct list_head lp_routes;
690 /* reference count */
691 atomic_t lp_refcount;
693 /* lock protecting peer state flags and lpni_rtrq */
696 /* peer state flags */
699 /* buffer for data pushed by peer */
700 struct lnet_ping_buffer *lp_data;
702 /* MD handle for ping in progress */
703 struct lnet_handle_md lp_ping_mdh;
705 /* MD handle for push in progress */
706 struct lnet_handle_md lp_push_mdh;
708 /* number of NIDs for sizing push data */
711 /* NI config sequence number of peer */
714 /* Local NI config sequence number acked by peer */
717 /* Local NI config sequence number sent to peer */
718 __u32 lp_node_seqno_sent;
720 /* Ping error encountered during discovery. */
723 /* Push error encountered during discovery. */
726 /* Error encountered during discovery. */
729 /* time it was put on the ln_dc_working queue */
730 time64_t lp_last_queued;
732 /* link on discovery-related lists */
733 struct list_head lp_dc_list;
735 /* tasks waiting on discovery of this peer */
736 wait_queue_head_t lp_dc_waitq;
738 /* cached peer aliveness */
743 * The status flags in lp_state. Their semantics have chosen so that
744 * lp_state can be zero-initialized.
746 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
747 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
749 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
750 * NOT set when the peer was pinged by discovery.
752 * A peer is marked ROUTER if it indicates so in the feature bit.
754 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
755 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
756 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
759 * A peer is marked CONFIGURED if it was configured by DLC.
761 * In addition, a peer is marked DISCOVERED if it has fully passed
762 * through Peer Discovery.
764 * When Peer Discovery is disabled, the discovery thread will mark
765 * peers REDISCOVER to indicate that they should be re-examined if
766 * discovery is (re)enabled on the node.
768 * A peer that was created as the result of inbound traffic will not
771 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
772 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
773 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
775 * A peer is marked DISCOVERING when discovery is in progress.
776 * The other flags below correspond to stages of discovery.
778 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
779 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
780 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
781 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
782 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
783 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
784 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
786 * A ping can be forced as a way to fix up state, or as a manual
787 * intervention by an admin.
788 * A push can be forced in circumstances that would normally not
789 * allow for one to happen.
791 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
792 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
794 /* force delete even if router */
795 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
797 /* gw undergoing alive discovery */
798 #define LNET_PEER_RTR_DISCOVERY BIT(16)
799 /* gw has undergone discovery (does not indicate success or failure) */
800 #define LNET_PEER_RTR_DISCOVERED BIT(17)
802 /* peer is marked for deletion */
803 #define LNET_PEER_MARK_DELETION BIT(18)
804 /* lnet_peer_del()/lnet_peer_del_locked() has been called on the peer */
805 #define LNET_PEER_MARK_DELETED BIT(19)
806 /* lock primary NID to what's requested by ULP */
807 #define LNET_PEER_LOCK_PRIMARY BIT(20)
808 /* this is for informational purposes only. It is set if a peer gets
809 * configured from Lustre with a primary NID which belongs to another peer
810 * which is also configured by Lustre as the primary NID.
812 #define LNET_PEER_BAD_CONFIG BIT(21)
814 struct lnet_peer_net {
815 /* chain on lp_peer_nets */
816 struct list_head lpn_peer_nets;
818 /* list of peer_nis on this network */
819 struct list_head lpn_peer_nis;
821 /* pointer to the peer I'm part of */
822 struct lnet_peer *lpn_peer;
827 /* peer net health */
830 /* time of next router ping on this net */
831 time64_t lpn_next_ping;
833 /* selection sequence number */
836 /* relative peer net selection priority */
837 __u32 lpn_sel_priority;
839 /* reference count */
840 atomic_t lpn_refcount;
844 #define LNET_PEER_HASH_BITS 9
845 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
848 * peer hash table - one per CPT
850 * protected by lnet_net_lock/EX for update
855 * protected by pt_zombie_lock:
859 * pt_zombie lock nests inside lnet_net_lock
861 struct lnet_peer_table {
862 int pt_version; /* /proc validity stamp */
863 struct list_head *pt_hash; /* NID->peer hash */
864 struct list_head pt_peer_list; /* peers */
865 int pt_peers; /* # peers */
866 struct list_head pt_zombie_list; /* zombie peer_ni */
867 int pt_zombies; /* # zombie peers_ni */
868 spinlock_t pt_zombie_lock; /* protect list and count */
871 /* peer aliveness is enabled only on routers for peers in a network where the
872 * struct lnet_ni::ni_peertimeout has been set to a positive value
874 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
875 ((lp)->lpni_net) && \
876 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
879 struct list_head lr_list; /* chain on net */
880 struct list_head lr_gwlist; /* chain on gateway */
881 struct lnet_peer *lr_gateway; /* router node */
882 lnet_nid_t lr_nid; /* NID used to add route */
883 __u32 lr_net; /* remote network number */
884 __u32 lr_lnet; /* local network number */
885 int lr_seq; /* sequence for round-robin */
886 __u32 lr_hops; /* how far I am */
887 unsigned int lr_priority; /* route priority */
888 atomic_t lr_alive; /* cached route aliveness */
889 bool lr_single_hop; /* this route is single-hop */
892 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
893 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
894 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
896 struct lnet_remotenet {
897 /* chain on ln_remote_nets_hash */
898 struct list_head lrn_list;
900 struct list_head lrn_routes;
905 /** lnet message has credit and can be submitted to lnd for send/receive */
906 #define LNET_CREDIT_OK 0
907 /** lnet message is waiting for credit */
908 #define LNET_CREDIT_WAIT 1
909 /** lnet message is waiting for discovery */
910 #define LNET_DC_WAIT 2
912 struct lnet_rtrbufpool {
913 /* my free buffer pool */
914 struct list_head rbp_bufs;
915 /* messages blocking for a buffer */
916 struct list_head rbp_msgs;
917 /* # pages in each buffer */
919 /* requested number of buffers */
920 int rbp_req_nbuffers;
921 /* # buffers actually allocated */
923 /* # free buffers / blocked messages */
930 struct list_head rb_list; /* chain on rbp_bufs */
931 struct lnet_rtrbufpool *rb_pool; /* owning pool */
932 struct bio_vec rb_kiov[0]; /* the buffer space */
935 #define LNET_PEER_HASHSIZE 503 /* prime! */
937 enum lnet_match_flags {
938 /* Didn't match anything */
939 LNET_MATCHMD_NONE = BIT(0),
941 LNET_MATCHMD_OK = BIT(1),
942 /* Must be discarded */
943 LNET_MATCHMD_DROP = BIT(2),
944 /* match and buffer is exhausted */
945 LNET_MATCHMD_EXHAUSTED = BIT(3),
947 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
950 /* Options for struct lnet_portal::ptl_options */
951 #define LNET_PTL_LAZY BIT(0)
952 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
953 #define LNET_PTL_MATCH_WILDCARD BIT(2) /* wildcard match, request portal */
955 /* parameter for matching operations (GET, PUT) */
956 struct lnet_match_info {
958 struct lnet_process_id mi_id;
961 unsigned int mi_portal;
962 unsigned int mi_rlength;
963 unsigned int mi_roffset;
966 /* ME hash of RDMA portal */
967 #define LNET_MT_HASH_BITS 8
968 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
969 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
970 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
971 * the last entry is reserved for MEs with ignore-bits */
972 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
973 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
974 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
975 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
976 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
977 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
978 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
980 /* portal match table */
981 struct lnet_match_table {
982 /* reserved for upcoming patches, CPU partition ID */
984 unsigned int mt_portal; /* portal index */
985 /* match table is set as "enabled" if there's non-exhausted MD
986 * attached on mt_mhash, it's only valid for wildcard portal */
987 unsigned int mt_enabled;
988 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
989 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
990 struct list_head *mt_mhash; /* matching hash */
993 /* these are only useful for wildcard portal */
994 /* Turn off message rotor for wildcard portals */
995 #define LNET_PTL_ROTOR_OFF 0
996 /* round-robin dispatch all PUT messages for wildcard portals */
997 #define LNET_PTL_ROTOR_ON 1
998 /* round-robin dispatch routed PUT message for wildcard portals */
999 #define LNET_PTL_ROTOR_RR_RT 2
1000 /* dispatch routed PUT message by hashing source NID for wildcard portals */
1001 #define LNET_PTL_ROTOR_HASH_RT 3
1003 struct lnet_portal {
1004 spinlock_t ptl_lock;
1005 unsigned int ptl_index; /* portal ID, reserved */
1006 /* flags on this portal: lazy, unique... */
1007 unsigned int ptl_options;
1008 /* list of messages which are stealing buffer */
1009 struct list_head ptl_msg_stealing;
1010 /* messages blocking for MD */
1011 struct list_head ptl_msg_delayed;
1012 /* Match table for each CPT */
1013 struct lnet_match_table **ptl_mtables;
1014 /* spread rotor of incoming "PUT" */
1015 unsigned int ptl_rotor;
1016 /* # active entries for this portal */
1018 /* array of active entries' cpu-partition-id */
1022 #define LNET_LH_HASH_BITS 12
1023 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
1024 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
1026 /* resource container (ME, MD, EQ) */
1027 struct lnet_res_container {
1028 unsigned int rec_type; /* container type */
1029 __u64 rec_lh_cookie; /* cookie generator */
1030 struct list_head rec_active; /* active resource list */
1031 struct list_head *rec_lh_hash; /* handle hash */
1034 /* message container */
1035 struct lnet_msg_container {
1036 int msc_init; /* initialized or not */
1037 /* max # threads finalizing */
1038 int msc_nfinalizers;
1039 /* msgs waiting to complete finalizing */
1040 struct list_head msc_finalizing;
1041 /* msgs waiting to be resent */
1042 struct list_head msc_resending;
1043 struct list_head msc_active; /* active message list */
1044 /* threads doing finalization */
1045 void **msc_finalizers;
1046 /* threads doing resends */
1047 void **msc_resenders;
1050 /* This UDSP structures need to match the user space liblnetconfig structures
1051 * in order for the marshall and unmarshall functions to be common.
1054 /* Net is described as a
1058 struct lnet_ud_net_descr {
1060 struct list_head udn_net_num_range;
1063 /* each NID range is defined as
1065 * 2. address range descriptor
1067 struct lnet_ud_nid_descr {
1068 struct lnet_ud_net_descr ud_net_id;
1069 struct list_head ud_addr_range;
1073 /* a UDSP rule can have up to three user defined NID descriptors
1074 * - src: defines the local NID range for the rule
1075 * - dst: defines the peer NID range for the rule
1076 * - rte: defines the router NID range for the rule
1078 * An action union defines the action to take when the rule
1082 struct list_head udsp_on_list;
1084 struct lnet_ud_nid_descr udsp_src;
1085 struct lnet_ud_nid_descr udsp_dst;
1086 struct lnet_ud_nid_descr udsp_rte;
1087 enum lnet_udsp_action_type udsp_action_type;
1089 __u32 udsp_priority;
1093 /* Peer Discovery states */
1094 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
1095 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
1096 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
1098 /* Router Checker states */
1099 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
1100 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
1101 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
1104 #define LNET_STATE_SHUTDOWN 0 /* not started */
1105 #define LNET_STATE_RUNNING 1 /* started up OK */
1106 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
1109 /* CPU partition table of LNet */
1110 struct cfs_cpt_table *ln_cpt_table;
1111 /* number of CPTs in ln_cpt_table */
1112 unsigned int ln_cpt_number;
1113 unsigned int ln_cpt_bits;
1115 /* protect LNet resources (ME/MD/EQ) */
1116 struct cfs_percpt_lock *ln_res_lock;
1119 /* the vector of portals */
1120 struct lnet_portal **ln_portals;
1121 /* percpt MD container */
1122 struct lnet_res_container **ln_md_containers;
1124 /* Event Queue container */
1125 struct lnet_res_container ln_eq_container;
1126 spinlock_t ln_eq_wait_lock;
1128 unsigned int ln_remote_nets_hbits;
1130 /* protect NI, peer table, credits, routers, rtrbuf... */
1131 struct cfs_percpt_lock *ln_net_lock;
1132 /* percpt message containers for active/finalizing/freed message */
1133 struct lnet_msg_container **ln_msg_containers;
1134 struct lnet_counters **ln_counters;
1135 struct lnet_peer_table **ln_peer_tables;
1136 /* list of peer nis not on a local network */
1137 struct list_head ln_remote_peer_ni_list;
1138 /* failure simulation */
1139 struct list_head ln_test_peers;
1140 struct list_head ln_drop_rules;
1141 struct list_head ln_delay_rules;
1143 struct list_head ln_nets;
1144 /* the loopback NI */
1145 struct lnet_ni *ln_loni;
1146 /* network zombie list */
1147 struct list_head ln_net_zombie;
1148 /* resend messages list */
1149 struct list_head ln_msg_resend;
1150 /* spin lock to protect the msg resend list */
1151 spinlock_t ln_msg_resend_lock;
1153 /* remote networks with routes to them */
1154 struct list_head *ln_remote_nets_hash;
1155 /* validity stamp */
1156 __u64 ln_remote_nets_version;
1157 /* list of all known routers */
1158 struct list_head ln_routers;
1159 /* validity stamp */
1160 __u64 ln_routers_version;
1161 /* percpt router buffer pools */
1162 struct lnet_rtrbufpool **ln_rtrpools;
1165 * Ping target / Push source
1167 * The ping target and push source share a single buffer. The
1168 * ln_ping_target is protected against concurrent updates by
1171 struct lnet_handle_md ln_ping_target_md;
1172 lnet_handler_t ln_ping_target_handler;
1173 struct lnet_ping_buffer *ln_ping_target;
1174 atomic_t ln_ping_target_seqno;
1179 * ln_push_nnis contains the desired size of the push target.
1180 * The lnet_net_lock is used to handle update races. The old
1181 * buffer may linger a while after it has been unlinked, in
1182 * which case the event handler cleans up.
1184 lnet_handler_t ln_push_target_handler;
1185 struct lnet_handle_md ln_push_target_md;
1186 struct lnet_ping_buffer *ln_push_target;
1187 int ln_push_target_nnis;
1189 /* discovery event queue handle */
1190 lnet_handler_t ln_dc_handler;
1191 /* discovery requests */
1192 struct list_head ln_dc_request;
1193 /* discovery working list */
1194 struct list_head ln_dc_working;
1195 /* discovery expired list */
1196 struct list_head ln_dc_expired;
1197 /* discovery thread wait queue */
1198 wait_queue_head_t ln_dc_waitq;
1199 /* discovery startup/shutdown state */
1202 /* monitor thread startup/shutdown state */
1204 /* serialise startup/shutdown */
1205 struct semaphore ln_mt_signal;
1207 struct mutex ln_api_mutex;
1208 struct mutex ln_lnd_mutex;
1209 /* Have I called LNetNIInit myself? */
1211 /* LNetNIInit/LNetNIFini counter */
1213 /* SHUTDOWN/RUNNING/STOPPING */
1216 int ln_routing; /* am I a router? */
1217 lnet_pid_t ln_pid; /* requested pid */
1218 /* uniquely identifies this ni in this epoch */
1219 __u64 ln_interface_cookie;
1220 /* registered LNDs */
1221 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1223 /* test protocol compatibility flags */
1224 unsigned long ln_testprotocompat;
1226 /* 0 - load the NIs from the mod params
1227 * 1 - do not load the NIs from the mod params
1228 * Reverse logic to ensure that other calls to LNetNIInit
1231 bool ln_nis_from_mod_params;
1234 * completion for the monitor thread. The monitor thread takes care of
1235 * checking routes, timedout messages and resending messages.
1237 struct completion ln_mt_wait_complete;
1239 /* per-cpt resend queues */
1240 struct list_head **ln_mt_resendqs;
1241 /* local NIs to recover */
1242 struct list_head ln_mt_localNIRecovq;
1243 /* local NIs to recover */
1244 struct list_head ln_mt_peerNIRecovq;
1246 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1247 * There are CPT number of queues. Since response trackers will be
1248 * added on the fast path we can't afford to grab the exclusive
1249 * net lock to protect these queues. The CPT will be calculated
1250 * based on the mdh cookie.
1252 struct list_head **ln_mt_rstq;
1254 * A response tracker becomes a zombie when the associated MD is queued
1255 * for unlink before the response tracker is detached from the MD. An
1256 * entry on a zombie list can be freed when either the remaining
1257 * operations on the MD complete or when LNet has shut down.
1259 struct list_head **ln_mt_zombie_rstqs;
1260 /* recovery handler */
1261 lnet_handler_t ln_mt_handler;
1264 * Completed when the discovery and monitor threads can enter their
1267 struct completion ln_started;
1269 struct list_head ln_udsp_list;
1272 static const struct nla_policy scalar_attr_policy[LN_SCALAR_CNT + 1] = {
1273 [LN_SCALAR_ATTR_LIST] = { .type = NLA_NESTED },
1274 [LN_SCALAR_ATTR_LIST_SIZE] = { .type = NLA_U16 },
1275 [LN_SCALAR_ATTR_INDEX] = { .type = NLA_U16 },
1276 [LN_SCALAR_ATTR_NLA_TYPE] = { .type = NLA_U16 },
1277 [LN_SCALAR_ATTR_VALUE] = { .type = NLA_STRING },
1278 [LN_SCALAR_ATTR_KEY_FORMAT] = { .type = NLA_U16 },
1281 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
1282 const struct genl_family *family, int flags,
1283 u8 cmd, const struct ln_key_list *data[]);
1285 /* Special workaround for pre-4.19 kernels to send error messages
1286 * from dumpit routines. Newer kernels will send message with
1287 * NL_SET_ERR_MSG information by default if NETLINK_EXT_ACK is set.
1289 static inline int lnet_nl_send_error(struct sk_buff *msg, int portid, int seq,
1292 #ifndef HAVE_NL_DUMP_WITH_EXT_ACK
1293 struct nlmsghdr *nlh;
1298 nlh = nlmsg_put(msg, portid, seq, NLMSG_ERROR, sizeof(error), 0);
1301 #ifdef HAVE_NL_PARSE_WITH_EXT_ACK
1302 netlink_ack(msg, nlh, error, NULL);
1304 netlink_ack(msg, nlh, error);
1306 return nlmsg_len(nlh);