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 struct lnet_nid 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 struct lnet_nid 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 void (*nf_addr2str_size)(const __be32 *addr, size_t asize,
255 char *str, size_t size);
256 int (*nf_str2addr)(const char *str, int nob, u32 *addr);
257 int (*nf_str2addr_size)(const char *str, int nob,
258 __be32 *addr, size_t *asize);
259 int (*nf_parse_addrlist)(char *str, int len,
260 struct list_head *list);
261 int (*nf_print_addrlist)(char *buffer, int count,
262 struct list_head *list);
263 int (*nf_match_addr)(u32 addr, struct list_head *list);
264 int (*nf_min_max)(struct list_head *nidlist, u32 *min_nid,
268 struct lnet_ni; /* forward ref */
272 /* fields initialized by the LND */
275 int (*lnd_startup)(struct lnet_ni *ni);
276 void (*lnd_shutdown)(struct lnet_ni *ni);
277 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
279 /* In data movement APIs below, payload buffers are described as a set
280 * of 'niov' fragments which are in pages.
281 * The LND may NOT overwrite these fragment descriptors.
282 * An 'offset' and may specify a byte offset within the set of
283 * fragments to start from
286 /* Start sending a preformatted message. 'private' is NULL for PUT and
287 * GET messages; otherwise this is a response to an incoming message
288 * and 'private' is the 'private' passed to lnet_parse(). Return
289 * non-zero for immediate failure, otherwise complete later with
291 int (*lnd_send)(struct lnet_ni *ni, void *private,
292 struct lnet_msg *msg);
294 /* Start receiving 'mlen' bytes of payload data, skipping the following
295 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
296 * lnet_parse(). Return non-zero for immedaite failure, otherwise
297 * complete later with lnet_finalize(). This also gives back a receive
298 * credit if the LND does flow control. */
299 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
300 int delayed, unsigned int niov,
301 struct bio_vec *kiov,
302 unsigned int offset, unsigned int mlen, unsigned int rlen);
304 /* lnet_parse() has had to delay processing of this message
305 * (e.g. waiting for a forwarding buffer or send credits). Give the
306 * LND a chance to free urgently needed resources. If called, return 0
307 * for success and do NOT give back a receive credit; that has to wait
308 * until lnd_recv() gets called. On failure return < 0 and
309 * release resources; lnd_recv() will not be called. */
310 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
311 struct lnet_msg *msg, void **new_privatep);
313 /* notification of peer down */
314 void (*lnd_notify_peer_down)(lnet_nid_t peer);
316 /* accept a new connection */
317 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
319 /* get dma_dev priority */
320 unsigned int (*lnd_get_dev_prio)(struct lnet_ni *ni,
321 unsigned int dev_idx);
324 struct lnet_tx_queue {
325 int tq_credits; /* # tx credits free */
326 int tq_credits_min; /* lowest it's been */
327 int tq_credits_max; /* total # tx credits */
328 struct list_head tq_delayed; /* delayed TXs */
331 enum lnet_net_state {
332 /* set when net block is allocated */
333 LNET_NET_STATE_INIT = 0,
334 /* set when NIs in net are started successfully */
335 LNET_NET_STATE_ACTIVE,
336 /* set if all NIs in net are in FAILED state */
337 LNET_NET_STATE_INACTIVE,
338 /* set when shutting down a NET */
339 LNET_NET_STATE_DELETING
343 /* initial state when NI is created */
344 LNET_NI_STATE_INIT = 0,
345 /* set when NI is brought up */
346 LNET_NI_STATE_ACTIVE,
347 /* set when NI is being shutdown */
348 LNET_NI_STATE_DELETING,
351 #define LNET_NI_RECOVERY_PENDING BIT(0)
352 #define LNET_NI_RECOVERY_FAILED BIT(1)
354 enum lnet_stats_type {
355 LNET_STATS_TYPE_SEND = 0,
356 LNET_STATS_TYPE_RECV,
360 struct lnet_comm_count {
361 atomic_t co_get_count;
362 atomic_t co_put_count;
363 atomic_t co_reply_count;
364 atomic_t co_ack_count;
365 atomic_t co_hello_count;
368 struct lnet_element_stats {
369 struct lnet_comm_count el_send_stats;
370 struct lnet_comm_count el_recv_stats;
371 struct lnet_comm_count el_drop_stats;
374 struct lnet_health_local_stats {
375 atomic_t hlt_local_interrupt;
376 atomic_t hlt_local_dropped;
377 atomic_t hlt_local_aborted;
378 atomic_t hlt_local_no_route;
379 atomic_t hlt_local_timeout;
380 atomic_t hlt_local_error;
383 struct lnet_health_remote_stats {
384 atomic_t hlt_remote_dropped;
385 atomic_t hlt_remote_timeout;
386 atomic_t hlt_remote_error;
387 atomic_t hlt_network_timeout;
391 /* chain on the ln_nets */
392 struct list_head net_list;
394 /* net ID, which is composed of
395 * (net_type << 16) | net_num.
396 * net_type can be one of the enumerated types defined in
397 * lnet/include/lnet/nidstr.h */
400 /* round robin selection */
403 /* total number of CPTs in the array */
406 /* cumulative CPTs of all NIs in this net */
409 /* relative net selection priority */
410 __u32 net_sel_priority;
412 /* network tunables */
413 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
416 * boolean to indicate that the tunables have been set and
419 bool net_tunables_set;
421 /* procedural interface */
422 const struct lnet_lnd *net_lnd;
424 /* list of NIs on this net */
425 struct list_head net_ni_list;
427 /* list of NIs being added, but not started yet */
428 struct list_head net_ni_added;
430 /* dying LND instances */
431 struct list_head net_ni_zombie;
433 /* when I was last alive */
434 time64_t net_last_alive;
436 /* protects access to net_last_alive */
439 /* list of router nids preferred for this network */
440 struct list_head net_rtr_pref_nids;
444 /* chain on the lnet_net structure */
445 struct list_head ni_netlist;
447 /* chain on the recovery queue */
448 struct list_head ni_recovery;
450 /* MD handle for recovery ping */
451 struct lnet_handle_md ni_ping_mdh;
458 /* bond NI on some CPTs */
461 /* interface's NID */
462 struct lnet_nid ni_nid;
464 /* instance-specific data */
468 atomic_t ni_tx_credits;
470 /* percpt TX queues */
471 struct lnet_tx_queue **ni_tx_queues;
473 /* percpt reference count */
476 /* pointer to parent network */
477 struct lnet_net *ni_net;
479 /* my health status */
480 struct lnet_ni_status *ni_status;
482 /* NI FSM. Protected by lnet_ni_lock() */
483 enum lnet_ni_state ni_state;
485 /* Recovery state. Protected by lnet_ni_lock() */
486 __u32 ni_recovery_state;
488 /* When to send the next recovery ping */
489 time64_t ni_next_ping;
490 /* How many pings sent during current recovery period did not receive
491 * a reply. NB: reset whenever _any_ message arrives on this NI
493 unsigned int ni_ping_count;
495 /* per NI LND tunables */
496 struct lnet_lnd_tunables ni_lnd_tunables;
498 /* lnd tunables set explicitly */
499 bool ni_lnd_tunables_set;
502 struct lnet_element_stats ni_stats;
503 struct lnet_health_local_stats ni_hstats;
505 /* physical device CPT */
508 /* sequence number used to round robin over nis within a net */
513 * initialized to LNET_MAX_HEALTH_VALUE
514 * Value is decremented every time we fail to send a message over
515 * this NI because of a NI specific failure.
516 * Value is incremented if we successfully send a message.
521 * Set to 1 by the LND when it receives an event telling it the device
522 * has gone into a fatal state. Set to 0 when the LND receives an
523 * even telling it the device is back online.
525 atomic_t ni_fatal_error_on;
527 /* the relative selection priority of this NI */
528 __u32 ni_sel_priority;
531 * equivalent interface to use
534 struct net *ni_net_ns; /* original net namespace */
537 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
540 * Descriptor of a ping info buffer: keep a separate indicator of the
541 * size and a reference count. The type is used both as a source and
542 * sink of data, so we need to keep some information outside of the
543 * area that may be overwritten by network data.
545 struct lnet_ping_buffer {
549 struct lnet_ping_info pb_info;
552 #define LNET_PING_BUFFER_SIZE(NNIDS) \
553 offsetof(struct lnet_ping_buffer, pb_info.pi_ni[NNIDS])
554 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
555 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
557 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
558 container_of((PINFO), struct lnet_ping_buffer, pb_info)
560 struct lnet_nid_list {
561 struct list_head nl_list;
562 struct lnet_nid nl_nid;
565 struct lnet_peer_ni {
566 /* chain on lpn_peer_nis */
567 struct list_head lpni_peer_nis;
568 /* chain on remote peer list */
569 struct list_head lpni_on_remote_peer_ni_list;
570 /* chain on recovery queue */
571 struct list_head lpni_recovery;
572 /* chain on peer hash */
573 struct list_head lpni_hashlist;
574 /* messages blocking for tx credits */
575 struct list_head lpni_txq;
576 /* pointer to peer net I'm part of */
577 struct lnet_peer_net *lpni_peer_net;
578 /* statistics kept on each peer NI */
579 struct lnet_element_stats lpni_stats;
580 struct lnet_health_remote_stats lpni_hstats;
581 /* spin lock protecting credits and lpni_txq */
582 spinlock_t lpni_lock;
583 /* # tx credits available */
586 int lpni_mintxcredits;
588 * Each peer_ni in a gateway maintains its own credits. This
589 * allows more traffic to gateways that have multiple interfaces.
591 /* # router credits */
594 int lpni_minrtrcredits;
595 /* bytes queued for sending */
597 /* network peer is on */
598 struct lnet_net *lpni_net;
600 struct lnet_nid lpni_nid;
602 struct kref lpni_kref;
603 /* health value for the peer */
604 atomic_t lpni_healthv;
605 /* recovery ping mdh */
606 struct lnet_handle_md lpni_recovery_ping_mdh;
607 /* When to send the next recovery ping */
608 time64_t lpni_next_ping;
609 /* How many pings sent during current recovery period did not receive
610 * a reply. NB: reset whenever _any_ message arrives from this peer NI
612 unsigned int lpni_ping_count;
613 /* CPT this peer attached on */
615 /* state flags -- protected by lpni_lock */
617 /* status of the peer NI as reported by the peer */
618 __u32 lpni_ns_status;
619 /* sequence number used to round robin over peer nis within a net */
621 /* sequence number used to round robin over gateways */
623 /* returned RC ping features. Protected with lpni_lock */
624 unsigned int lpni_ping_feats;
625 /* time last message was received from the peer */
626 time64_t lpni_last_alive;
627 /* preferred local nids: if only one, use lpni_pref.nid */
630 struct list_head nids;
632 /* list of router nids preferred for this peer NI */
633 struct list_head lpni_rtr_pref_nids;
634 /* The relative selection priority of this peer NI */
635 __u32 lpni_sel_priority;
636 /* number of preferred NIDs in lnpi_pref_nids */
637 __u32 lpni_pref_nnids;
640 /* Preferred path added due to traffic on non-MR peer_ni */
641 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
642 /* peer is being recovered. */
643 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
644 /* recovery ping failed */
645 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
646 /* peer is being deleted */
647 #define LNET_PEER_NI_DELETING BIT(3)
650 /* chain on pt_peer_list */
651 struct list_head lp_peer_list;
653 /* list of peer nets */
654 struct list_head lp_peer_nets;
656 /* list of messages pending discovery*/
657 struct list_head lp_dc_pendq;
659 /* chain on router list */
660 struct list_head lp_rtr_list;
662 /* primary NID of the peer */
663 struct lnet_nid lp_primary_nid;
665 /* source NID to use during discovery */
666 struct lnet_nid lp_disc_src_nid;
667 /* destination NID to use during discovery */
668 struct lnet_nid lp_disc_dst_nid;
670 /* net to perform discovery on */
671 __u32 lp_disc_net_id;
673 /* CPT of peer_table */
676 /* number of NIDs on this peer */
679 /* # refs from lnet_route::lr_gateway */
683 * peer specific health sensitivity value to decrement peer nis in
684 * this peer with if set to something other than 0
686 __u32 lp_health_sensitivity;
688 /* messages blocking for router credits */
689 struct list_head lp_rtrq;
691 /* routes on this peer */
692 struct list_head lp_routes;
694 /* reference count */
695 atomic_t lp_refcount;
697 /* lock protecting peer state flags and lpni_rtrq */
700 /* peer state flags */
703 /* buffer for data pushed by peer */
704 struct lnet_ping_buffer *lp_data;
706 /* MD handle for ping in progress */
707 struct lnet_handle_md lp_ping_mdh;
709 /* MD handle for push in progress */
710 struct lnet_handle_md lp_push_mdh;
712 /* number of NIDs for sizing push data */
715 /* NI config sequence number of peer */
718 /* Local NI config sequence number acked by peer */
721 /* Local NI config sequence number sent to peer */
722 __u32 lp_node_seqno_sent;
724 /* Ping error encountered during discovery. */
727 /* Push error encountered during discovery. */
730 /* Error encountered during discovery. */
733 /* time it was put on the ln_dc_working queue */
734 time64_t lp_last_queued;
736 /* link on discovery-related lists */
737 struct list_head lp_dc_list;
739 /* tasks waiting on discovery of this peer */
740 wait_queue_head_t lp_dc_waitq;
742 /* cached peer aliveness */
747 * The status flags in lp_state. Their semantics have chosen so that
748 * lp_state can be zero-initialized.
750 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
751 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
753 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
754 * NOT set when the peer was pinged by discovery.
756 * A peer is marked ROUTER if it indicates so in the feature bit.
758 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
759 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
760 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
763 * A peer is marked CONFIGURED if it was configured by DLC.
765 * In addition, a peer is marked DISCOVERED if it has fully passed
766 * through Peer Discovery.
768 * When Peer Discovery is disabled, the discovery thread will mark
769 * peers REDISCOVER to indicate that they should be re-examined if
770 * discovery is (re)enabled on the node.
772 * A peer that was created as the result of inbound traffic will not
775 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
776 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
777 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
779 * A peer is marked DISCOVERING when discovery is in progress.
780 * The other flags below correspond to stages of discovery.
782 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
783 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
784 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
785 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
786 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
787 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
788 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
790 * A ping can be forced as a way to fix up state, or as a manual
791 * intervention by an admin.
792 * A push can be forced in circumstances that would normally not
793 * allow for one to happen.
795 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
796 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
798 /* force delete even if router */
799 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
801 /* gw undergoing alive discovery */
802 #define LNET_PEER_RTR_DISCOVERY BIT(16)
803 /* gw has undergone discovery (does not indicate success or failure) */
804 #define LNET_PEER_RTR_DISCOVERED BIT(17)
806 /* peer is marked for deletion */
807 #define LNET_PEER_MARK_DELETION BIT(18)
808 /* lnet_peer_del()/lnet_peer_del_locked() has been called on the peer */
809 #define LNET_PEER_MARK_DELETED BIT(19)
810 /* lock primary NID to what's requested by ULP */
811 #define LNET_PEER_LOCK_PRIMARY BIT(20)
812 /* this is for informational purposes only. It is set if a peer gets
813 * configured from Lustre with a primary NID which belongs to another peer
814 * which is also configured by Lustre as the primary NID.
816 #define LNET_PEER_BAD_CONFIG BIT(21)
818 struct lnet_peer_net {
819 /* chain on lp_peer_nets */
820 struct list_head lpn_peer_nets;
822 /* list of peer_nis on this network */
823 struct list_head lpn_peer_nis;
825 /* pointer to the peer I'm part of */
826 struct lnet_peer *lpn_peer;
831 /* peer net health */
834 /* time of next router ping on this net */
835 time64_t lpn_next_ping;
837 /* selection sequence number */
840 /* relative peer net selection priority */
841 __u32 lpn_sel_priority;
843 /* reference count */
844 atomic_t lpn_refcount;
848 #define LNET_PEER_HASH_BITS 9
849 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
852 * peer hash table - one per CPT
854 * protected by lnet_net_lock/EX for update
859 * protected by pt_zombie_lock:
863 * pt_zombie lock nests inside lnet_net_lock
865 struct lnet_peer_table {
866 int pt_version; /* /proc validity stamp */
867 struct list_head *pt_hash; /* NID->peer hash */
868 struct list_head pt_peer_list; /* peers */
869 int pt_peers; /* # peers */
870 struct list_head pt_zombie_list; /* zombie peer_ni */
871 int pt_zombies; /* # zombie peers_ni */
872 spinlock_t pt_zombie_lock; /* protect list and count */
875 /* peer aliveness is enabled only on routers for peers in a network where the
876 * struct lnet_ni::ni_peertimeout has been set to a positive value
878 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
879 ((lp)->lpni_net) && \
880 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
883 struct list_head lr_list; /* chain on net */
884 struct list_head lr_gwlist; /* chain on gateway */
885 struct lnet_peer *lr_gateway; /* router node */
886 struct lnet_nid lr_nid; /* NID used to add route */
887 __u32 lr_net; /* remote network number */
888 __u32 lr_lnet; /* local network number */
889 int lr_seq; /* sequence for round-robin */
890 __u32 lr_hops; /* how far I am */
891 unsigned int lr_priority; /* route priority */
892 atomic_t lr_alive; /* cached route aliveness */
893 bool lr_single_hop; /* this route is single-hop */
896 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
897 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
898 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
900 struct lnet_remotenet {
901 /* chain on ln_remote_nets_hash */
902 struct list_head lrn_list;
904 struct list_head lrn_routes;
909 /** lnet message has credit and can be submitted to lnd for send/receive */
910 #define LNET_CREDIT_OK 0
911 /** lnet message is waiting for credit */
912 #define LNET_CREDIT_WAIT 1
913 /** lnet message is waiting for discovery */
914 #define LNET_DC_WAIT 2
916 struct lnet_rtrbufpool {
917 /* my free buffer pool */
918 struct list_head rbp_bufs;
919 /* messages blocking for a buffer */
920 struct list_head rbp_msgs;
921 /* # pages in each buffer */
923 /* requested number of buffers */
924 int rbp_req_nbuffers;
925 /* # buffers actually allocated */
927 /* # free buffers / blocked messages */
934 struct list_head rb_list; /* chain on rbp_bufs */
935 struct lnet_rtrbufpool *rb_pool; /* owning pool */
936 struct bio_vec rb_kiov[0]; /* the buffer space */
939 #define LNET_PEER_HASHSIZE 503 /* prime! */
941 enum lnet_match_flags {
942 /* Didn't match anything */
943 LNET_MATCHMD_NONE = BIT(0),
945 LNET_MATCHMD_OK = BIT(1),
946 /* Must be discarded */
947 LNET_MATCHMD_DROP = BIT(2),
948 /* match and buffer is exhausted */
949 LNET_MATCHMD_EXHAUSTED = BIT(3),
951 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
954 /* Options for struct lnet_portal::ptl_options */
955 #define LNET_PTL_LAZY BIT(0)
956 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
957 #define LNET_PTL_MATCH_WILDCARD BIT(2) /* wildcard match, request portal */
959 /* parameter for matching operations (GET, PUT) */
960 struct lnet_match_info {
962 struct lnet_process_id mi_id;
965 unsigned int mi_portal;
966 unsigned int mi_rlength;
967 unsigned int mi_roffset;
970 /* ME hash of RDMA portal */
971 #define LNET_MT_HASH_BITS 8
972 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
973 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
974 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
975 * the last entry is reserved for MEs with ignore-bits */
976 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
977 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
978 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
979 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
980 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
981 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
982 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
984 /* portal match table */
985 struct lnet_match_table {
986 /* reserved for upcoming patches, CPU partition ID */
988 unsigned int mt_portal; /* portal index */
989 /* match table is set as "enabled" if there's non-exhausted MD
990 * attached on mt_mhash, it's only valid for wildcard portal */
991 unsigned int mt_enabled;
992 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
993 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
994 struct list_head *mt_mhash; /* matching hash */
997 /* these are only useful for wildcard portal */
998 /* Turn off message rotor for wildcard portals */
999 #define LNET_PTL_ROTOR_OFF 0
1000 /* round-robin dispatch all PUT messages for wildcard portals */
1001 #define LNET_PTL_ROTOR_ON 1
1002 /* round-robin dispatch routed PUT message for wildcard portals */
1003 #define LNET_PTL_ROTOR_RR_RT 2
1004 /* dispatch routed PUT message by hashing source NID for wildcard portals */
1005 #define LNET_PTL_ROTOR_HASH_RT 3
1007 struct lnet_portal {
1008 spinlock_t ptl_lock;
1009 unsigned int ptl_index; /* portal ID, reserved */
1010 /* flags on this portal: lazy, unique... */
1011 unsigned int ptl_options;
1012 /* list of messages which are stealing buffer */
1013 struct list_head ptl_msg_stealing;
1014 /* messages blocking for MD */
1015 struct list_head ptl_msg_delayed;
1016 /* Match table for each CPT */
1017 struct lnet_match_table **ptl_mtables;
1018 /* spread rotor of incoming "PUT" */
1019 unsigned int ptl_rotor;
1020 /* # active entries for this portal */
1022 /* array of active entries' cpu-partition-id */
1026 #define LNET_LH_HASH_BITS 12
1027 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
1028 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
1030 /* resource container (ME, MD, EQ) */
1031 struct lnet_res_container {
1032 unsigned int rec_type; /* container type */
1033 __u64 rec_lh_cookie; /* cookie generator */
1034 struct list_head rec_active; /* active resource list */
1035 struct list_head *rec_lh_hash; /* handle hash */
1038 /* message container */
1039 struct lnet_msg_container {
1040 int msc_init; /* initialized or not */
1041 /* max # threads finalizing */
1042 int msc_nfinalizers;
1043 /* msgs waiting to complete finalizing */
1044 struct list_head msc_finalizing;
1045 /* msgs waiting to be resent */
1046 struct list_head msc_resending;
1047 struct list_head msc_active; /* active message list */
1048 /* threads doing finalization */
1049 void **msc_finalizers;
1050 /* threads doing resends */
1051 void **msc_resenders;
1054 /* This UDSP structures need to match the user space liblnetconfig structures
1055 * in order for the marshall and unmarshall functions to be common.
1058 /* Net is described as a
1062 struct lnet_ud_net_descr {
1064 struct list_head udn_net_num_range;
1067 /* each NID range is defined as
1069 * 2. address range descriptor
1071 struct lnet_ud_nid_descr {
1072 struct lnet_ud_net_descr ud_net_id;
1073 struct list_head ud_addr_range;
1077 /* a UDSP rule can have up to three user defined NID descriptors
1078 * - src: defines the local NID range for the rule
1079 * - dst: defines the peer NID range for the rule
1080 * - rte: defines the router NID range for the rule
1082 * An action union defines the action to take when the rule
1086 struct list_head udsp_on_list;
1088 struct lnet_ud_nid_descr udsp_src;
1089 struct lnet_ud_nid_descr udsp_dst;
1090 struct lnet_ud_nid_descr udsp_rte;
1091 enum lnet_udsp_action_type udsp_action_type;
1093 __u32 udsp_priority;
1097 /* Peer Discovery states */
1098 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
1099 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
1100 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
1102 /* Router Checker states */
1103 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
1104 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
1105 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
1108 #define LNET_STATE_SHUTDOWN 0 /* not started */
1109 #define LNET_STATE_RUNNING 1 /* started up OK */
1110 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
1113 /* CPU partition table of LNet */
1114 struct cfs_cpt_table *ln_cpt_table;
1115 /* number of CPTs in ln_cpt_table */
1116 unsigned int ln_cpt_number;
1117 unsigned int ln_cpt_bits;
1119 /* protect LNet resources (ME/MD/EQ) */
1120 struct cfs_percpt_lock *ln_res_lock;
1123 /* the vector of portals */
1124 struct lnet_portal **ln_portals;
1125 /* percpt MD container */
1126 struct lnet_res_container **ln_md_containers;
1128 /* Event Queue container */
1129 struct lnet_res_container ln_eq_container;
1130 spinlock_t ln_eq_wait_lock;
1132 unsigned int ln_remote_nets_hbits;
1134 /* protect NI, peer table, credits, routers, rtrbuf... */
1135 struct cfs_percpt_lock *ln_net_lock;
1136 /* percpt message containers for active/finalizing/freed message */
1137 struct lnet_msg_container **ln_msg_containers;
1138 struct lnet_counters **ln_counters;
1139 struct lnet_peer_table **ln_peer_tables;
1140 /* list of peer nis not on a local network */
1141 struct list_head ln_remote_peer_ni_list;
1142 /* failure simulation */
1143 struct list_head ln_test_peers;
1144 struct list_head ln_drop_rules;
1145 struct list_head ln_delay_rules;
1147 struct list_head ln_nets;
1148 /* the loopback NI */
1149 struct lnet_ni *ln_loni;
1150 /* network zombie list */
1151 struct list_head ln_net_zombie;
1152 /* resend messages list */
1153 struct list_head ln_msg_resend;
1154 /* spin lock to protect the msg resend list */
1155 spinlock_t ln_msg_resend_lock;
1157 /* remote networks with routes to them */
1158 struct list_head *ln_remote_nets_hash;
1159 /* validity stamp */
1160 __u64 ln_remote_nets_version;
1161 /* list of all known routers */
1162 struct list_head ln_routers;
1163 /* validity stamp */
1164 __u64 ln_routers_version;
1165 /* percpt router buffer pools */
1166 struct lnet_rtrbufpool **ln_rtrpools;
1169 * Ping target / Push source
1171 * The ping target and push source share a single buffer. The
1172 * ln_ping_target is protected against concurrent updates by
1175 struct lnet_handle_md ln_ping_target_md;
1176 lnet_handler_t ln_ping_target_handler;
1177 struct lnet_ping_buffer *ln_ping_target;
1178 atomic_t ln_ping_target_seqno;
1183 * ln_push_nnis contains the desired size of the push target.
1184 * The lnet_net_lock is used to handle update races. The old
1185 * buffer may linger a while after it has been unlinked, in
1186 * which case the event handler cleans up.
1188 lnet_handler_t ln_push_target_handler;
1189 struct lnet_handle_md ln_push_target_md;
1190 struct lnet_ping_buffer *ln_push_target;
1191 int ln_push_target_nnis;
1193 /* discovery event queue handle */
1194 lnet_handler_t ln_dc_handler;
1195 /* discovery requests */
1196 struct list_head ln_dc_request;
1197 /* discovery working list */
1198 struct list_head ln_dc_working;
1199 /* discovery expired list */
1200 struct list_head ln_dc_expired;
1201 /* discovery thread wait queue */
1202 wait_queue_head_t ln_dc_waitq;
1203 /* discovery startup/shutdown state */
1206 /* monitor thread startup/shutdown state */
1208 /* serialise startup/shutdown */
1209 struct semaphore ln_mt_signal;
1211 struct mutex ln_api_mutex;
1212 struct mutex ln_lnd_mutex;
1213 /* Have I called LNetNIInit myself? */
1215 /* LNetNIInit/LNetNIFini counter */
1217 /* SHUTDOWN/RUNNING/STOPPING */
1220 int ln_routing; /* am I a router? */
1221 lnet_pid_t ln_pid; /* requested pid */
1222 /* uniquely identifies this ni in this epoch */
1223 __u64 ln_interface_cookie;
1224 /* registered LNDs */
1225 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1227 /* test protocol compatibility flags */
1228 unsigned long ln_testprotocompat;
1230 /* 0 - load the NIs from the mod params
1231 * 1 - do not load the NIs from the mod params
1232 * Reverse logic to ensure that other calls to LNetNIInit
1235 bool ln_nis_from_mod_params;
1238 * completion for the monitor thread. The monitor thread takes care of
1239 * checking routes, timedout messages and resending messages.
1241 struct completion ln_mt_wait_complete;
1243 /* per-cpt resend queues */
1244 struct list_head **ln_mt_resendqs;
1245 /* local NIs to recover */
1246 struct list_head ln_mt_localNIRecovq;
1247 /* local NIs to recover */
1248 struct list_head ln_mt_peerNIRecovq;
1250 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1251 * There are CPT number of queues. Since response trackers will be
1252 * added on the fast path we can't afford to grab the exclusive
1253 * net lock to protect these queues. The CPT will be calculated
1254 * based on the mdh cookie.
1256 struct list_head **ln_mt_rstq;
1258 * A response tracker becomes a zombie when the associated MD is queued
1259 * for unlink before the response tracker is detached from the MD. An
1260 * entry on a zombie list can be freed when either the remaining
1261 * operations on the MD complete or when LNet has shut down.
1263 struct list_head **ln_mt_zombie_rstqs;
1264 /* recovery handler */
1265 lnet_handler_t ln_mt_handler;
1268 * Completed when the discovery and monitor threads can enter their
1271 struct completion ln_started;
1273 struct list_head ln_udsp_list;
1276 static const struct nla_policy scalar_attr_policy[LN_SCALAR_CNT + 1] = {
1277 [LN_SCALAR_ATTR_LIST] = { .type = NLA_NESTED },
1278 [LN_SCALAR_ATTR_LIST_SIZE] = { .type = NLA_U16 },
1279 [LN_SCALAR_ATTR_INDEX] = { .type = NLA_U16 },
1280 [LN_SCALAR_ATTR_NLA_TYPE] = { .type = NLA_U16 },
1281 [LN_SCALAR_ATTR_VALUE] = { .type = NLA_STRING },
1282 [LN_SCALAR_ATTR_KEY_FORMAT] = { .type = NLA_U16 },
1285 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
1286 const struct genl_family *family, int flags,
1287 u8 cmd, const struct ln_key_list *data[]);
1289 /* Special workaround for pre-4.19 kernels to send error messages
1290 * from dumpit routines. Newer kernels will send message with
1291 * NL_SET_ERR_MSG information by default if NETLINK_EXT_ACK is set.
1293 static inline int lnet_nl_send_error(struct sk_buff *msg, int portid, int seq,
1296 #ifndef HAVE_NL_DUMP_WITH_EXT_ACK
1297 struct nlmsghdr *nlh;
1302 nlh = nlmsg_put(msg, portid, seq, NLMSG_ERROR, sizeof(error), 0);
1305 #ifdef HAVE_NL_PARSE_WITH_EXT_ACK
1306 netlink_ack(msg, nlh, error, NULL);
1308 netlink_ack(msg, nlh, error);
1310 return nlmsg_len(nlh);