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 int libcfs_strid(struct lnet_processid *id, const char *str);
58 int cfs_match_nid_net(struct lnet_nid *nid, u32 net,
59 struct list_head *net_num_list,
60 struct list_head *addr);
62 /* Max payload size */
63 #define LNET_MAX_PAYLOAD LNET_MTU
65 /** limit on the number of fragments in discontiguous MDs */
66 #define LNET_MAX_IOV 256
69 * This is the maximum health value.
70 * All local and peer NIs created have their health default to this value.
72 #define LNET_MAX_HEALTH_VALUE 1000
73 #define LNET_MAX_SELECTION_PRIORITY UINT_MAX
78 enum lnet_msg_hstatus {
79 LNET_MSG_STATUS_OK = 0,
80 LNET_MSG_STATUS_LOCAL_INTERRUPT,
81 LNET_MSG_STATUS_LOCAL_DROPPED,
82 LNET_MSG_STATUS_LOCAL_ABORTED,
83 LNET_MSG_STATUS_LOCAL_NO_ROUTE,
84 LNET_MSG_STATUS_LOCAL_ERROR,
85 LNET_MSG_STATUS_LOCAL_TIMEOUT,
86 LNET_MSG_STATUS_REMOTE_ERROR,
87 LNET_MSG_STATUS_REMOTE_DROPPED,
88 LNET_MSG_STATUS_REMOTE_TIMEOUT,
89 LNET_MSG_STATUS_NETWORK_TIMEOUT,
93 struct lnet_rsp_tracker {
94 /* chain on the waiting list */
95 struct list_head rspt_on_list;
99 struct lnet_nid rspt_next_hop_nid;
100 /* deadline of the REPLY/ACK */
101 ktime_t rspt_deadline;
103 struct lnet_handle_md rspt_mdh;
107 struct list_head msg_activelist;
108 struct list_head msg_list; /* Q for credits/MD */
110 struct lnet_processid msg_target;
111 /* Primary NID of the source. */
112 struct lnet_nid msg_initiator;
113 /* where is it from, it's only for building event */
114 struct lnet_nid msg_from;
118 * hold parameters in case message is with held due
121 struct lnet_nid msg_src_nid_param;
122 struct lnet_nid msg_rtr_nid_param;
125 * Deadline for the message after which it will be finalized if it
128 ktime_t msg_deadline;
130 /* The message health status. */
131 enum lnet_msg_hstatus msg_health_status;
132 /* This is a recovery message */
134 /* the number of times a transmission has been retried */
136 /* flag to indicate that we do not want to resend this message */
139 /* committed for sending */
140 unsigned int msg_tx_committed:1;
141 /* CPT # this message committed for sending */
142 unsigned int msg_tx_cpt:15;
143 /* committed for receiving */
144 unsigned int msg_rx_committed:1;
145 /* CPT # this message committed for receiving */
146 unsigned int msg_rx_cpt:15;
147 /* queued for tx credit */
148 unsigned int msg_tx_delayed:1;
149 /* queued for RX buffer */
150 unsigned int msg_rx_delayed:1;
151 /* ready for pending on RX delay list */
152 unsigned int msg_rx_ready_delay:1;
154 unsigned int msg_vmflush:1; /* VM trying to free memory */
155 unsigned int msg_target_is_router:1; /* sending to a router */
156 unsigned int msg_routing:1; /* being forwarded */
157 unsigned int msg_ack:1; /* ack on finalize (PUT) */
158 unsigned int msg_sending:1; /* outgoing message */
159 unsigned int msg_receiving:1; /* being received */
160 unsigned int msg_txcredit:1; /* taken an NI send credit */
161 unsigned int msg_peertxcredit:1; /* taken a peer send credit */
162 unsigned int msg_rtrcredit:1; /* taken a globel router credit */
163 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */
164 unsigned int msg_onactivelist:1; /* on the activelist */
165 unsigned int msg_rdma_get:1;
167 struct lnet_peer_ni *msg_txpeer; /* peer I'm sending to */
168 struct lnet_peer_ni *msg_rxpeer; /* peer I received from */
171 struct lnet_libmd *msg_md;
172 /* the NI the message was sent or received over */
173 struct lnet_ni *msg_txni;
174 struct lnet_ni *msg_rxni;
176 unsigned int msg_len;
177 unsigned int msg_wanted;
178 unsigned int msg_offset;
179 unsigned int msg_niov;
180 struct bio_vec *msg_kiov;
182 struct lnet_event msg_ev;
183 struct lnet_hdr msg_hdr;
186 struct lnet_libhandle {
187 struct list_head lh_hash_chain;
191 #define lh_entry(ptr, type, member) \
192 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
195 struct list_head me_list;
197 struct lnet_processid me_match_id;
198 unsigned int me_portal;
199 unsigned int me_pos; /* hash offset in mt_hash */
201 __u64 me_ignore_bits;
202 enum lnet_unlink me_unlink;
203 struct lnet_libmd *me_md;
207 struct list_head md_list;
208 struct lnet_libhandle md_lh;
209 struct lnet_me *md_me;
211 unsigned int md_offset;
212 unsigned int md_length;
213 unsigned int md_max_size;
216 unsigned int md_options;
217 unsigned int md_flags;
218 unsigned int md_niov; /* # frags at end of struct */
220 struct lnet_rsp_tracker *md_rspt_ptr;
221 lnet_handler_t md_handler;
222 struct lnet_handle_md md_bulk_handle;
223 struct bio_vec md_kiov[LNET_MAX_IOV];
226 #define LNET_MD_FLAG_ZOMBIE BIT(0)
227 #define LNET_MD_FLAG_AUTO_UNLINK BIT(1)
228 #define LNET_MD_FLAG_ABORTED BIT(2)
229 /* LNET_MD_FLAG_HANDLING is set when a non-unlink event handler
230 * is being called for an event relating to the md.
231 * It ensures only one such handler runs at a time.
232 * The final "unlink" event is only called once the
233 * md_refcount has reached zero, and this flag has been cleared,
234 * ensuring that it doesn't race with any other event handler
237 #define LNET_MD_FLAG_HANDLING BIT(3)
238 #define LNET_MD_FLAG_DISCARD BIT(4)
239 #define LNET_MD_FLAG_GPU BIT(5) /**< Special mapping needs */
241 struct lnet_test_peer {
242 /* info about peers we are trying to fail */
243 struct list_head tp_list; /* ln_test_peers */
244 struct lnet_nid tp_nid; /* matching nid */
245 unsigned int tp_threshold; /* # failures to simulate */
248 #define LNET_COOKIE_TYPE_MD 1
249 #define LNET_COOKIE_TYPE_ME 2
250 #define LNET_COOKIE_TYPE_EQ 3
251 #define LNET_COOKIE_TYPE_BITS 2
252 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
258 void (*nf_addr2str)(u32 addr, char *str, size_t size);
259 void (*nf_addr2str_size)(const __be32 *addr, size_t asize,
260 char *str, size_t size);
261 int (*nf_str2addr)(const char *str, int nob, u32 *addr);
262 int (*nf_str2addr_size)(const char *str, int nob,
263 __be32 *addr, size_t *asize);
264 int (*nf_parse_addrlist)(char *str, int len,
265 struct list_head *list);
266 int (*nf_print_addrlist)(char *buffer, int count,
267 struct list_head *list);
268 int (*nf_match_addr)(u32 addr, struct list_head *list);
269 int (*nf_min_max)(struct list_head *nidlist, u32 *min_nid,
273 struct lnet_ni; /* forward ref */
277 /* fields initialized by the LND */
280 int (*lnd_startup)(struct lnet_ni *ni);
281 void (*lnd_shutdown)(struct lnet_ni *ni);
282 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
284 /* In data movement APIs below, payload buffers are described as a set
285 * of 'niov' fragments which are in pages.
286 * The LND may NOT overwrite these fragment descriptors.
287 * An 'offset' and may specify a byte offset within the set of
288 * fragments to start from
291 /* Start sending a preformatted message. 'private' is NULL for PUT and
292 * GET messages; otherwise this is a response to an incoming message
293 * and 'private' is the 'private' passed to lnet_parse(). Return
294 * non-zero for immediate failure, otherwise complete later with
296 int (*lnd_send)(struct lnet_ni *ni, void *private,
297 struct lnet_msg *msg);
299 /* Start receiving 'mlen' bytes of payload data, skipping the following
300 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
301 * lnet_parse(). Return non-zero for immedaite failure, otherwise
302 * complete later with lnet_finalize(). This also gives back a receive
303 * credit if the LND does flow control. */
304 int (*lnd_recv)(struct lnet_ni *ni, void *private, struct lnet_msg *msg,
305 int delayed, unsigned int niov,
306 struct bio_vec *kiov,
307 unsigned int offset, unsigned int mlen, unsigned int rlen);
309 /* lnet_parse() has had to delay processing of this message
310 * (e.g. waiting for a forwarding buffer or send credits). Give the
311 * LND a chance to free urgently needed resources. If called, return 0
312 * for success and do NOT give back a receive credit; that has to wait
313 * until lnd_recv() gets called. On failure return < 0 and
314 * release resources; lnd_recv() will not be called. */
315 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private,
316 struct lnet_msg *msg, void **new_privatep);
318 /* notification of peer down */
319 void (*lnd_notify_peer_down)(struct lnet_nid *peer);
321 /* accept a new connection */
322 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock);
324 /* get dma_dev priority */
325 unsigned int (*lnd_get_dev_prio)(struct lnet_ni *ni,
326 unsigned int dev_idx);
328 /* Handle LND specific Netlink handling */
329 int (*lnd_nl_set)(int cmd, struct nlattr *attr, int type, void *data);
331 const struct ln_key_list *lnd_keys;
334 struct lnet_tx_queue {
335 int tq_credits; /* # tx credits free */
336 int tq_credits_min; /* lowest it's been */
337 int tq_credits_max; /* total # tx credits */
338 struct list_head tq_delayed; /* delayed TXs */
341 enum lnet_net_state {
342 /* set when net block is allocated */
343 LNET_NET_STATE_INIT = 0,
344 /* set when NIs in net are started successfully */
345 LNET_NET_STATE_ACTIVE,
346 /* set if all NIs in net are in FAILED state */
347 LNET_NET_STATE_INACTIVE,
348 /* set when shutting down a NET */
349 LNET_NET_STATE_DELETING
353 /* initial state when NI is created */
354 LNET_NI_STATE_INIT = 0,
355 /* set when NI is brought up */
356 LNET_NI_STATE_ACTIVE,
357 /* set when NI is being shutdown */
358 LNET_NI_STATE_DELETING,
361 #define LNET_NI_RECOVERY_PENDING BIT(0)
362 #define LNET_NI_RECOVERY_FAILED BIT(1)
364 enum lnet_stats_type {
365 LNET_STATS_TYPE_SEND = 0,
366 LNET_STATS_TYPE_RECV,
370 struct lnet_comm_count {
371 atomic_t co_get_count;
372 atomic_t co_put_count;
373 atomic_t co_reply_count;
374 atomic_t co_ack_count;
375 atomic_t co_hello_count;
378 struct lnet_element_stats {
379 struct lnet_comm_count el_send_stats;
380 struct lnet_comm_count el_recv_stats;
381 struct lnet_comm_count el_drop_stats;
384 struct lnet_health_local_stats {
385 atomic_t hlt_local_interrupt;
386 atomic_t hlt_local_dropped;
387 atomic_t hlt_local_aborted;
388 atomic_t hlt_local_no_route;
389 atomic_t hlt_local_timeout;
390 atomic_t hlt_local_error;
393 struct lnet_health_remote_stats {
394 atomic_t hlt_remote_dropped;
395 atomic_t hlt_remote_timeout;
396 atomic_t hlt_remote_error;
397 atomic_t hlt_network_timeout;
401 /* chain on the ln_nets */
402 struct list_head net_list;
404 /* net ID, which is composed of
405 * (net_type << 16) | net_num.
406 * net_type can be one of the enumerated types defined in
407 * lnet/include/lnet/nidstr.h */
410 /* round robin selection */
413 /* total number of CPTs in the array */
416 /* cumulative CPTs of all NIs in this net */
419 /* relative net selection priority */
420 __u32 net_sel_priority;
422 /* network tunables */
423 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
426 * boolean to indicate that the tunables have been set and
429 bool net_tunables_set;
431 /* procedural interface */
432 const struct lnet_lnd *net_lnd;
434 /* list of NIs on this net */
435 struct list_head net_ni_list;
437 /* list of NIs being added, but not started yet */
438 struct list_head net_ni_added;
440 /* dying LND instances */
441 struct list_head net_ni_zombie;
443 /* when I was last alive */
444 time64_t net_last_alive;
446 /* protects access to net_last_alive */
449 /* list of router nids preferred for this network */
450 struct list_head net_rtr_pref_nids;
453 /* Normally Netlink atttributes are defined in UAPI headers but Lustre is
454 * different in that the ABI is in a constant state of change unlike other
455 * Netlink interfaces. LNet sends a special header to help user land handle
459 /** enum lnet_net_attrs - LNet NI netlink properties
460 * attributes that describe LNet 'NI'
461 * These values are used to piece together
462 * messages for sending and receiving.
464 * @LNET_NET_ATTR_UNSPEC: unspecified attribute to catch errors
466 * @LNET_NET_ATTR_HDR: grouping for LNet net data (NLA_NESTED)
467 * @LNET_NET_ATTR_TYPE: LNet net this NI belongs to (NLA_STRING)
468 * @LNET_NET_ATTR_LOCAL: Local NI information (NLA_NESTED)
470 enum lnet_net_attrs {
471 LNET_NET_ATTR_UNSPEC = 0,
477 __LNET_NET_ATTR_MAX_PLUS_ONE,
480 #define LNET_NET_ATTR_MAX (__LNET_NET_ATTR_MAX_PLUS_ONE - 1)
482 /** enum lnet_net_local_ni_attrs - LNet local NI netlink properties
483 * attributes that describe local NI
485 * @LNET_NET_LOCAL_NI_ATTR_UNSPEC: unspecified attribute to catch errors
487 * @LNET_NET_LOCAL_NI_ATTR_NID: NID that represents this NI (NLA_STRING)
488 * @LNET_NET_LOCAL_NI_ATTR_STATUS: State of this NI (NLA_STRING)
489 * @LNET_NET_LOCAL_NI_ATTR_INTERFACE: Defines physical devices (NLA_NESTED)
490 * Used to be many devices but no longer.
492 enum lnet_net_local_ni_attrs {
493 LNET_NET_LOCAL_NI_ATTR_UNSPEC = 0,
495 LNET_NET_LOCAL_NI_ATTR_NID,
496 LNET_NET_LOCAL_NI_ATTR_STATUS,
497 LNET_NET_LOCAL_NI_ATTR_INTERFACE,
499 __LNET_NET_LOCAL_NI_ATTR_MAX_PLUS_ONE,
502 #define LNET_NET_LOCAL_NI_ATTR_MAX (__LNET_NET_LOCAL_NI_ATTR_MAX_PLUS_ONE - 1)
504 /** enum lnet_net_local_ni_intf_attrs - LNet NI device netlink properties
505 * attribute that reports the device
508 * @LNET_NET_LOCAL_NI_INTF_ATTR_UNSPEC: unspecified attribute to catch errors
510 * @LNET_NET_LOCAL_NI_INTF_ATTR_TYPE: Physcial device interface (NLA_STRING)
512 enum lnet_net_local_ni_intf_attrs {
513 LNET_NET_LOCAL_NI_INTF_ATTR_UNSPEC = 0,
515 LNET_NET_LOCAL_NI_INTF_ATTR_TYPE,
517 __LNET_NET_LOCAL_NI_INTF_ATTR_MAX_PLUS_ONE,
520 #define LNET_NET_LOCAL_NI_INTF_ATTR_MAX (__LNET_NET_LOCAL_NI_INTF_ATTR_MAX_PLUS_ONE - 1)
522 /** enum lnet_net_local_ni_tunables_attrs - LNet NI tunables
523 * netlink properties.
524 * Performance options
527 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_UNSPEC: unspecified attribute
530 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_TIMEOUT: Timeout for LNet peer.
532 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_CREDITS: Credits for LNet peer.
534 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_BUFFER_CREDITS: Buffer credits for
535 * LNet peer. (NLA_S32)
536 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_CREDITS: Credits for LNet peer
539 enum lnet_net_local_ni_tunables_attr {
540 LNET_NET_LOCAL_NI_TUNABLES_ATTR_UNSPEC = 0,
542 LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_TIMEOUT,
543 LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_CREDITS,
544 LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_BUFFER_CREDITS,
545 LNET_NET_LOCAL_NI_TUNABLES_ATTR_CREDITS,
546 __LNET_NET_LOCAL_NI_TUNABLES_ATTR_MAX_PLUS_ONE,
549 #define LNET_NET_LOCAL_NI_TUNABLES_ATTR_MAX (__LNET_NET_LOCAL_NI_TUNABLES_ATTR_MAX_PLUS_ONE - 1)
551 /** enum lnet_route_attrs - LNet route netlink
552 * attributes that describe
555 * @LNET_ROUTE_ATTR_UNSPEC: unspecified attribute to
558 * @LNET_ROUTE_ATTR_HDR: grouping for LNet route data
560 * @LNET_ROUTE_ATTR_NET: LNet remote network reached
561 * by the route (NLA_STRING)
562 * @LNET_ROUTE_ATTR_GATEWAY: gateway for the route
564 * @LNET_ROUTE_ATTR_HOP: route hop count (NLA_S32)
566 * @LNET_ROUTE_ATTR_PRIORITY: rank of this network path
568 * @LNET_ROUTE_ATTR_HEALTH_SENSITIVITY: rate of health value change
569 * for the route (NLA_U32)
570 * @LNET_ROUTE_ATTR_STATE: state of route (NLA_STRING)
572 * @LNET_ROUTE_ATTR_TYPE: Report if we support multi-hop
575 enum lnet_route_attrs {
576 LNET_ROUTE_ATTR_UNSPEC = 0,
580 LNET_ROUTE_ATTR_GATEWAY,
582 LNET_ROUTE_ATTR_PRIORITY,
583 LNET_ROUTE_ATTR_HEALTH_SENSITIVITY,
584 LNET_ROUTE_ATTR_STATE,
585 LNET_ROUTE_ATTR_TYPE,
586 __LNET_ROUTE_ATTR_MAX_PLUS_ONE,
589 #define LNET_ROUTE_ATTR_MAX (__LNET_ROUTE_ATTR_MAX_PLUS_ONE - 1)
591 /** LNet netlink ping API */
593 /** enum lnet_ping_atts - LNet ping netlink properties
594 * attributes to describe ping format
595 * These values are used to piece together
596 * messages for sending and receiving.
598 * @LNET_PING_ATTR_UNSPEC: unspecified attribute to catch errors
600 * @LNET_PING_ATTR_HDR: grouping for LNet ping data (NLA_NUL_STRING)
601 * @LNET_PING_ATTR_PRIMARY_NID: Source NID for ping request (NLA_STRING)
602 * @LNET_PING_ATTR_ERRNO: error code if we fail to ping (NLA_S16)
603 * @LNET_PING_ATTR_MULTIRAIL: Report if MR is supported (NLA_FLAG)
604 * @LNET_PING_ATTR_PEER_NI_LIST: List of peer NI's (NLA_NESTED)
606 enum lnet_ping_attr {
607 LNET_PING_ATTR_UNSPEC = 0,
610 LNET_PING_ATTR_PRIMARY_NID,
611 LNET_PING_ATTR_ERRNO,
612 LNET_PING_ATTR_MULTIRAIL,
613 LNET_PING_ATTR_PEER_NI_LIST,
614 __LNET_PING_ATTR_MAX_PLUS_ONE,
617 #define LNET_PING_ATTR_MAX (__LNET_PING_ATTR_MAX_PLUS_ONE - 1)
619 /** enium lnet_ping_peer_ni_attr - LNet peer ni information reported by
620 * ping command. A list of these are
621 * returned with a ping request.
623 * @LNET_PING_PEER_NI_ATTR_UNSPEC: unspecified attribute to catch errrors
625 * @LNET_PING_PEER_NI_ATTR_NID: NID address of peer NI. (NLA_STRING)
627 enum lnet_ping_peer_ni_attr {
628 LNET_PING_PEER_NI_ATTR_UNSPEC = 0,
630 LNET_PING_PEER_NI_ATTR_NID,
631 __LNET_PING_PEER_NI_ATTR_MAX_PLUS_ONE,
634 #define LNET_PING_PEER_NI_ATTR_MAX (__LNET_PING_PEER_NI_ATTR_MAX_PLUS_ONE - 1)
637 /* chain on the lnet_net structure */
638 struct list_head ni_netlist;
640 /* chain on the recovery queue */
641 struct list_head ni_recovery;
643 /* MD handle for recovery ping */
644 struct lnet_handle_md ni_ping_mdh;
651 /* bond NI on some CPTs */
654 /* interface's NID */
655 struct lnet_nid ni_nid;
657 /* instance-specific data */
661 atomic_t ni_tx_credits;
663 /* percpt TX queues */
664 struct lnet_tx_queue **ni_tx_queues;
666 /* percpt reference count */
669 /* pointer to parent network */
670 struct lnet_net *ni_net;
672 /* my health status */
675 /* NI FSM. Protected by lnet_ni_lock() */
676 enum lnet_ni_state ni_state;
678 /* Recovery state. Protected by lnet_ni_lock() */
679 __u32 ni_recovery_state;
681 /* When to send the next recovery ping */
682 time64_t ni_next_ping;
683 /* How many pings sent during current recovery period did not receive
684 * a reply. NB: reset whenever _any_ message arrives on this NI
686 unsigned int ni_ping_count;
688 /* per NI LND tunables */
689 struct lnet_lnd_tunables ni_lnd_tunables;
691 /* lnd tunables set explicitly */
692 bool ni_lnd_tunables_set;
695 struct lnet_element_stats ni_stats;
696 struct lnet_health_local_stats ni_hstats;
698 /* physical device CPT */
701 /* sequence number used to round robin over nis within a net */
706 * initialized to LNET_MAX_HEALTH_VALUE
707 * Value is decremented every time we fail to send a message over
708 * this NI because of a NI specific failure.
709 * Value is incremented if we successfully send a message.
714 * Set to 1 by the LND when it receives an event telling it the device
715 * has gone into a fatal state. Set to 0 when the LND receives an
716 * even telling it the device is back online.
718 atomic_t ni_fatal_error_on;
720 /* the relative selection priority of this NI */
721 __u32 ni_sel_priority;
724 * equivalent interface to use
727 struct net *ni_net_ns; /* original net namespace */
730 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
733 * Descriptor of a ping info buffer: keep a separate indicator of the
734 * size and a reference count. The type is used both as a source and
735 * sink of data, so we need to keep some information outside of the
736 * area that may be overwritten by network data.
738 struct lnet_ping_buffer {
739 int pb_nbytes; /* sizeof pb_info */
742 struct lnet_ping_info pb_info;
745 #define LNET_PING_BUFFER_SIZE(bytes) \
746 (offsetof(struct lnet_ping_buffer, pb_info) + bytes)
747 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
748 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
750 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
751 container_of((PINFO), struct lnet_ping_buffer, pb_info)
754 lnet_ping_sts_size(const struct lnet_nid *nid)
758 if (nid_is_nid4(nid))
759 return sizeof(struct lnet_ni_status);
761 size = offsetof(struct lnet_ni_large_status, ns_nid) +
764 return round_up(size, 4);
767 static inline struct lnet_ni_large_status *
768 lnet_ping_sts_next(const struct lnet_ni_large_status *nis)
770 return (void *)nis + lnet_ping_sts_size(&nis->ns_nid);
774 lnet_ping_at_least_two_entries(const struct lnet_ping_info *pi)
776 /* Return true if we have at lease two entries. There is always a
777 * least one, a 4-byte lo0 interface.
779 struct lnet_ni_large_status *lns;
781 if ((pi->pi_features & LNET_PING_FEAT_LARGE_ADDR) == 0)
782 return pi->pi_nnis <= 2;
783 /* There is at least 1 large-address entry */
784 if (pi->pi_nnis != 1)
786 lns = (void *)&pi->pi_ni[1];
787 lns = lnet_ping_sts_next(lns);
789 return ((void *)pi + lnet_ping_info_size(pi) <= (void *)lns);
792 struct lnet_nid_list {
793 struct list_head nl_list;
794 struct lnet_nid nl_nid;
797 struct lnet_peer_ni {
798 /* chain on lpn_peer_nis */
799 struct list_head lpni_peer_nis;
800 /* chain on remote peer list */
801 struct list_head lpni_on_remote_peer_ni_list;
802 /* chain on recovery queue */
803 struct list_head lpni_recovery;
804 /* chain on peer hash */
805 struct list_head lpni_hashlist;
806 /* messages blocking for tx credits */
807 struct list_head lpni_txq;
808 /* pointer to peer net I'm part of */
809 struct lnet_peer_net *lpni_peer_net;
810 /* statistics kept on each peer NI */
811 struct lnet_element_stats lpni_stats;
812 struct lnet_health_remote_stats lpni_hstats;
813 /* spin lock protecting credits and lpni_txq */
814 spinlock_t lpni_lock;
815 /* # tx credits available */
818 int lpni_mintxcredits;
820 * Each peer_ni in a gateway maintains its own credits. This
821 * allows more traffic to gateways that have multiple interfaces.
823 /* # router credits */
826 int lpni_minrtrcredits;
827 /* bytes queued for sending */
829 /* network peer is on */
830 struct lnet_net *lpni_net;
832 struct lnet_nid lpni_nid;
834 struct kref lpni_kref;
835 /* health value for the peer */
836 atomic_t lpni_healthv;
837 /* recovery ping mdh */
838 struct lnet_handle_md lpni_recovery_ping_mdh;
839 /* When to send the next recovery ping */
840 time64_t lpni_next_ping;
841 /* How many pings sent during current recovery period did not receive
842 * a reply. NB: reset whenever _any_ message arrives from this peer NI
844 unsigned int lpni_ping_count;
845 /* CPT this peer attached on */
847 /* state flags -- protected by lpni_lock */
849 /* status of the peer NI as reported by the peer */
850 __u32 lpni_ns_status;
851 /* sequence number used to round robin over peer nis within a net */
853 /* sequence number used to round robin over gateways */
855 /* returned RC ping features. Protected with lpni_lock */
856 unsigned int lpni_ping_feats;
857 /* time last message was received from the peer */
858 time64_t lpni_last_alive;
859 /* preferred local nids: if only one, use lpni_pref.nid */
862 struct list_head nids;
864 /* list of router nids preferred for this peer NI */
865 struct list_head lpni_rtr_pref_nids;
866 /* The relative selection priority of this peer NI */
867 __u32 lpni_sel_priority;
868 /* number of preferred NIDs in lnpi_pref_nids */
869 __u32 lpni_pref_nnids;
872 /* Preferred path added due to traffic on non-MR peer_ni */
873 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
874 /* peer is being recovered. */
875 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
876 /* recovery ping failed */
877 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
878 /* peer is being deleted */
879 #define LNET_PEER_NI_DELETING BIT(3)
882 /* chain on pt_peer_list */
883 struct list_head lp_peer_list;
885 /* list of peer nets */
886 struct list_head lp_peer_nets;
888 /* list of messages pending discovery*/
889 struct list_head lp_dc_pendq;
891 /* chain on router list */
892 struct list_head lp_rtr_list;
894 /* primary NID of the peer */
895 struct lnet_nid lp_primary_nid;
897 /* source NID to use during discovery */
898 struct lnet_nid lp_disc_src_nid;
899 /* destination NID to use during discovery */
900 struct lnet_nid lp_disc_dst_nid;
902 /* net to perform discovery on */
903 __u32 lp_disc_net_id;
905 /* CPT of peer_table */
908 /* number of NIDs on this peer */
911 /* # refs from lnet_route::lr_gateway */
915 * peer specific health sensitivity value to decrement peer nis in
916 * this peer with if set to something other than 0
918 __u32 lp_health_sensitivity;
920 /* messages blocking for router credits */
921 struct list_head lp_rtrq;
923 /* routes on this peer */
924 struct list_head lp_routes;
926 /* reference count */
927 atomic_t lp_refcount;
929 /* lock protecting peer state flags and lpni_rtrq */
932 /* peer state flags */
935 /* buffer for data pushed by peer */
936 struct lnet_ping_buffer *lp_data;
938 /* MD handle for ping in progress */
939 struct lnet_handle_md lp_ping_mdh;
941 /* MD handle for push in progress */
942 struct lnet_handle_md lp_push_mdh;
944 /* number of bytes for sizing pb_info in push data */
947 /* NI config sequence number of peer */
950 /* Local NI config sequence number acked by peer */
953 /* Local NI config sequence number sent to peer */
954 __u32 lp_node_seqno_sent;
956 /* Ping error encountered during discovery. */
959 /* Push error encountered during discovery. */
962 /* Error encountered during discovery. */
965 /* time it was put on the ln_dc_working queue */
966 time64_t lp_last_queued;
968 /* link on discovery-related lists */
969 struct list_head lp_dc_list;
971 /* tasks waiting on discovery of this peer */
972 wait_queue_head_t lp_dc_waitq;
974 /* cached peer aliveness */
977 /* sequence number used to round robin traffic to this peer's
984 * The status flags in lp_state. Their semantics have chosen so that
985 * lp_state can be zero-initialized.
987 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
988 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
990 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
991 * NOT set when the peer was pinged by discovery.
993 * A peer is marked ROUTER if it indicates so in the feature bit.
995 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
996 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
997 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
1000 * A peer is marked CONFIGURED if it was configured by DLC.
1002 * In addition, a peer is marked DISCOVERED if it has fully passed
1003 * through Peer Discovery.
1005 * When Peer Discovery is disabled, the discovery thread will mark
1006 * peers REDISCOVER to indicate that they should be re-examined if
1007 * discovery is (re)enabled on the node.
1009 * A peer that was created as the result of inbound traffic will not
1012 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
1013 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
1014 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
1016 * A peer is marked DISCOVERING when discovery is in progress.
1017 * The other flags below correspond to stages of discovery.
1019 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
1020 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
1021 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
1022 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
1023 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
1024 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
1025 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
1027 * A ping can be forced as a way to fix up state, or as a manual
1028 * intervention by an admin.
1029 * A push can be forced in circumstances that would normally not
1030 * allow for one to happen.
1032 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
1033 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
1035 /* force delete even if router */
1036 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
1038 /* gw undergoing alive discovery */
1039 #define LNET_PEER_RTR_DISCOVERY BIT(16)
1040 /* gw has undergone discovery (does not indicate success or failure) */
1041 #define LNET_PEER_RTR_DISCOVERED BIT(17)
1043 /* peer is marked for deletion */
1044 #define LNET_PEER_MARK_DELETION BIT(18)
1045 /* lnet_peer_del()/lnet_peer_del_locked() has been called on the peer */
1046 #define LNET_PEER_MARK_DELETED BIT(19)
1047 /* lock primary NID to what's requested by ULP */
1048 #define LNET_PEER_LOCK_PRIMARY BIT(20)
1049 /* this is for informational purposes only. It is set if a peer gets
1050 * configured from Lustre with a primary NID which belongs to another peer
1051 * which is also configured by Lustre as the primary NID.
1053 #define LNET_PEER_BAD_CONFIG BIT(21)
1055 struct lnet_peer_net {
1056 /* chain on lp_peer_nets */
1057 struct list_head lpn_peer_nets;
1059 /* list of peer_nis on this network */
1060 struct list_head lpn_peer_nis;
1062 /* pointer to the peer I'm part of */
1063 struct lnet_peer *lpn_peer;
1068 /* peer net health */
1071 /* time of next router ping on this net */
1072 time64_t lpn_next_ping;
1074 /* selection sequence number */
1077 /* relative peer net selection priority */
1078 __u32 lpn_sel_priority;
1080 /* reference count */
1081 atomic_t lpn_refcount;
1084 /* peer hash size */
1085 #define LNET_PEER_HASH_BITS 9
1086 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
1089 * peer hash table - one per CPT
1091 * protected by lnet_net_lock/EX for update
1096 * protected by pt_zombie_lock:
1100 * pt_zombie lock nests inside lnet_net_lock
1102 struct lnet_peer_table {
1103 int pt_version; /* /proc validity stamp */
1104 struct list_head *pt_hash; /* NID->peer hash */
1105 struct list_head pt_peer_list; /* peers */
1106 int pt_peers; /* # peers */
1107 struct list_head pt_zombie_list; /* zombie peer_ni */
1108 int pt_zombies; /* # zombie peers_ni */
1109 spinlock_t pt_zombie_lock; /* protect list and count */
1112 /* peer aliveness is enabled only on routers for peers in a network where the
1113 * struct lnet_ni::ni_peertimeout has been set to a positive value
1115 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
1116 ((lp)->lpni_net) && \
1117 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
1120 struct list_head lr_list; /* chain on net */
1121 struct list_head lr_gwlist; /* chain on gateway */
1122 struct lnet_peer *lr_gateway; /* router node */
1123 struct lnet_nid lr_nid; /* NID used to add route */
1124 __u32 lr_net; /* remote network number */
1125 __u32 lr_lnet; /* local network number */
1126 int lr_seq; /* sequence for round-robin */
1127 __u32 lr_hops; /* how far I am */
1128 unsigned int lr_priority; /* route priority */
1129 atomic_t lr_alive; /* cached route aliveness */
1130 bool lr_single_hop; /* this route is single-hop */
1133 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
1134 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
1135 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
1137 struct lnet_remotenet {
1138 /* chain on ln_remote_nets_hash */
1139 struct list_head lrn_list;
1141 struct list_head lrn_routes;
1146 /** lnet message has credit and can be submitted to lnd for send/receive */
1147 #define LNET_CREDIT_OK 0
1148 /** lnet message is waiting for credit */
1149 #define LNET_CREDIT_WAIT 1
1150 /** lnet message is waiting for discovery */
1151 #define LNET_DC_WAIT 2
1153 struct lnet_rtrbufpool {
1154 /* my free buffer pool */
1155 struct list_head rbp_bufs;
1156 /* messages blocking for a buffer */
1157 struct list_head rbp_msgs;
1158 /* # pages in each buffer */
1160 /* requested number of buffers */
1161 int rbp_req_nbuffers;
1162 /* # buffers actually allocated */
1164 /* # free buffers / blocked messages */
1166 /* low water mark */
1170 struct lnet_rtrbuf {
1171 struct list_head rb_list; /* chain on rbp_bufs */
1172 struct lnet_rtrbufpool *rb_pool; /* owning pool */
1173 struct bio_vec rb_kiov[0]; /* the buffer space */
1176 #define LNET_PEER_HASHSIZE 503 /* prime! */
1178 enum lnet_match_flags {
1179 /* Didn't match anything */
1180 LNET_MATCHMD_NONE = BIT(0),
1182 LNET_MATCHMD_OK = BIT(1),
1183 /* Must be discarded */
1184 LNET_MATCHMD_DROP = BIT(2),
1185 /* match and buffer is exhausted */
1186 LNET_MATCHMD_EXHAUSTED = BIT(3),
1188 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
1191 /* Options for struct lnet_portal::ptl_options */
1192 #define LNET_PTL_LAZY BIT(0)
1193 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
1194 #define LNET_PTL_MATCH_WILDCARD BIT(2) /* wildcard match, request portal */
1196 /* parameter for matching operations (GET, PUT) */
1197 struct lnet_match_info {
1199 struct lnet_processid mi_id;
1200 unsigned int mi_cpt;
1201 unsigned int mi_opc;
1202 unsigned int mi_portal;
1203 unsigned int mi_rlength;
1204 unsigned int mi_roffset;
1207 /* ME hash of RDMA portal */
1208 #define LNET_MT_HASH_BITS 8
1209 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
1210 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
1211 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
1212 * the last entry is reserved for MEs with ignore-bits */
1213 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
1214 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
1215 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
1216 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
1217 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
1218 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
1219 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
1221 /* portal match table */
1222 struct lnet_match_table {
1223 /* reserved for upcoming patches, CPU partition ID */
1224 unsigned int mt_cpt;
1225 unsigned int mt_portal; /* portal index */
1226 /* match table is set as "enabled" if there's non-exhausted MD
1227 * attached on mt_mhash, it's only valid for wildcard portal */
1228 unsigned int mt_enabled;
1229 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
1230 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
1231 struct list_head *mt_mhash; /* matching hash */
1234 /* these are only useful for wildcard portal */
1235 /* Turn off message rotor for wildcard portals */
1236 #define LNET_PTL_ROTOR_OFF 0
1237 /* round-robin dispatch all PUT messages for wildcard portals */
1238 #define LNET_PTL_ROTOR_ON 1
1239 /* round-robin dispatch routed PUT message for wildcard portals */
1240 #define LNET_PTL_ROTOR_RR_RT 2
1241 /* dispatch routed PUT message by hashing source NID for wildcard portals */
1242 #define LNET_PTL_ROTOR_HASH_RT 3
1244 struct lnet_portal {
1245 spinlock_t ptl_lock;
1246 unsigned int ptl_index; /* portal ID, reserved */
1247 /* flags on this portal: lazy, unique... */
1248 unsigned int ptl_options;
1249 /* list of messages which are stealing buffer */
1250 struct list_head ptl_msg_stealing;
1251 /* messages blocking for MD */
1252 struct list_head ptl_msg_delayed;
1253 /* Match table for each CPT */
1254 struct lnet_match_table **ptl_mtables;
1255 /* spread rotor of incoming "PUT" */
1256 unsigned int ptl_rotor;
1257 /* # active entries for this portal */
1259 /* array of active entries' cpu-partition-id */
1263 #define LNET_LH_HASH_BITS 12
1264 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
1265 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
1267 /* resource container (ME, MD, EQ) */
1268 struct lnet_res_container {
1269 unsigned int rec_type; /* container type */
1270 __u64 rec_lh_cookie; /* cookie generator */
1271 struct list_head rec_active; /* active resource list */
1272 struct list_head *rec_lh_hash; /* handle hash */
1275 /* message container */
1276 struct lnet_msg_container {
1277 int msc_init; /* initialized or not */
1278 /* max # threads finalizing */
1279 int msc_nfinalizers;
1280 /* msgs waiting to complete finalizing */
1281 struct list_head msc_finalizing;
1282 /* msgs waiting to be resent */
1283 struct list_head msc_resending;
1284 struct list_head msc_active; /* active message list */
1285 /* threads doing finalization */
1286 void **msc_finalizers;
1287 /* threads doing resends */
1288 void **msc_resenders;
1291 /* This UDSP structures need to match the user space liblnetconfig structures
1292 * in order for the marshall and unmarshall functions to be common.
1295 /* Net is described as a
1299 struct lnet_ud_net_descr {
1301 struct list_head udn_net_num_range;
1304 /* each NID range is defined as
1306 * 2. address range descriptor
1308 struct lnet_ud_nid_descr {
1309 struct lnet_ud_net_descr ud_net_id;
1310 struct list_head ud_addr_range;
1314 /* a UDSP rule can have up to three user defined NID descriptors
1315 * - src: defines the local NID range for the rule
1316 * - dst: defines the peer NID range for the rule
1317 * - rte: defines the router NID range for the rule
1319 * An action union defines the action to take when the rule
1323 struct list_head udsp_on_list;
1325 struct lnet_ud_nid_descr udsp_src;
1326 struct lnet_ud_nid_descr udsp_dst;
1327 struct lnet_ud_nid_descr udsp_rte;
1328 enum lnet_udsp_action_type udsp_action_type;
1330 __u32 udsp_priority;
1334 /* Peer Discovery states */
1335 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
1336 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
1337 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
1339 /* Router Checker states */
1340 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
1341 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
1342 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
1345 #define LNET_STATE_SHUTDOWN 0 /* not started */
1346 #define LNET_STATE_RUNNING 1 /* started up OK */
1347 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
1350 /* CPU partition table of LNet */
1351 struct cfs_cpt_table *ln_cpt_table;
1352 /* number of CPTs in ln_cpt_table */
1353 unsigned int ln_cpt_number;
1354 unsigned int ln_cpt_bits;
1356 /* protect LNet resources (ME/MD/EQ) */
1357 struct cfs_percpt_lock *ln_res_lock;
1360 /* the vector of portals */
1361 struct lnet_portal **ln_portals;
1362 /* percpt MD container */
1363 struct lnet_res_container **ln_md_containers;
1365 /* Event Queue container */
1366 struct lnet_res_container ln_eq_container;
1367 spinlock_t ln_eq_wait_lock;
1369 unsigned int ln_remote_nets_hbits;
1371 /* protect NI, peer table, credits, routers, rtrbuf... */
1372 struct cfs_percpt_lock *ln_net_lock;
1373 /* percpt message containers for active/finalizing/freed message */
1374 struct lnet_msg_container **ln_msg_containers;
1375 struct lnet_counters **ln_counters;
1376 struct lnet_peer_table **ln_peer_tables;
1377 /* list of peer nis not on a local network */
1378 struct list_head ln_remote_peer_ni_list;
1379 /* failure simulation */
1380 struct list_head ln_test_peers;
1381 struct list_head ln_drop_rules;
1382 struct list_head ln_delay_rules;
1384 struct list_head ln_nets;
1385 /* Sequence number used to round robin sends across all nets */
1387 /* the loopback NI */
1388 struct lnet_ni *ln_loni;
1389 /* network zombie list */
1390 struct list_head ln_net_zombie;
1391 /* resend messages list */
1392 struct list_head ln_msg_resend;
1393 /* spin lock to protect the msg resend list */
1394 spinlock_t ln_msg_resend_lock;
1396 /* remote networks with routes to them */
1397 struct list_head *ln_remote_nets_hash;
1398 /* validity stamp */
1399 __u64 ln_remote_nets_version;
1400 /* list of all known routers */
1401 struct list_head ln_routers;
1402 /* validity stamp */
1403 __u64 ln_routers_version;
1404 /* percpt router buffer pools */
1405 struct lnet_rtrbufpool **ln_rtrpools;
1408 * Ping target / Push source
1410 * The ping target and push source share a single buffer. The
1411 * ln_ping_target is protected against concurrent updates by
1414 struct lnet_handle_md ln_ping_target_md;
1415 lnet_handler_t ln_ping_target_handler;
1416 struct lnet_ping_buffer *ln_ping_target;
1417 atomic_t ln_ping_target_seqno;
1422 * ln_push_nnis contains the desired size of the push target.
1423 * The lnet_net_lock is used to handle update races. The old
1424 * buffer may linger a while after it has been unlinked, in
1425 * which case the event handler cleans up.
1427 lnet_handler_t ln_push_target_handler;
1428 struct lnet_handle_md ln_push_target_md;
1429 struct lnet_ping_buffer *ln_push_target;
1430 /* bytes needed for pb_info to receive push */
1431 int ln_push_target_nbytes;
1433 /* discovery event queue handle */
1434 lnet_handler_t ln_dc_handler;
1435 /* discovery requests */
1436 struct list_head ln_dc_request;
1437 /* discovery working list */
1438 struct list_head ln_dc_working;
1439 /* discovery expired list */
1440 struct list_head ln_dc_expired;
1441 /* discovery thread wait queue */
1442 wait_queue_head_t ln_dc_waitq;
1443 /* discovery startup/shutdown state */
1446 /* monitor thread startup/shutdown state */
1448 /* serialise startup/shutdown */
1449 struct semaphore ln_mt_signal;
1451 struct mutex ln_api_mutex;
1452 struct mutex ln_lnd_mutex;
1453 /* Have I called LNetNIInit myself? */
1455 /* LNetNIInit/LNetNIFini counter */
1457 /* SHUTDOWN/RUNNING/STOPPING */
1460 int ln_routing; /* am I a router? */
1461 lnet_pid_t ln_pid; /* requested pid */
1462 /* uniquely identifies this ni in this epoch */
1463 __u64 ln_interface_cookie;
1464 /* registered LNDs */
1465 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1467 /* test protocol compatibility flags */
1468 unsigned long ln_testprotocompat;
1470 /* 0 - load the NIs from the mod params
1471 * 1 - do not load the NIs from the mod params
1472 * Reverse logic to ensure that other calls to LNetNIInit
1475 bool ln_nis_from_mod_params;
1478 * completion for the monitor thread. The monitor thread takes care of
1479 * checking routes, timedout messages and resending messages.
1481 struct completion ln_mt_wait_complete;
1483 /* per-cpt resend queues */
1484 struct list_head **ln_mt_resendqs;
1485 /* local NIs to recover */
1486 struct list_head ln_mt_localNIRecovq;
1487 /* local NIs to recover */
1488 struct list_head ln_mt_peerNIRecovq;
1490 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1491 * There are CPT number of queues. Since response trackers will be
1492 * added on the fast path we can't afford to grab the exclusive
1493 * net lock to protect these queues. The CPT will be calculated
1494 * based on the mdh cookie.
1496 struct list_head **ln_mt_rstq;
1498 * A response tracker becomes a zombie when the associated MD is queued
1499 * for unlink before the response tracker is detached from the MD. An
1500 * entry on a zombie list can be freed when either the remaining
1501 * operations on the MD complete or when LNet has shut down.
1503 struct list_head **ln_mt_zombie_rstqs;
1504 /* recovery handler */
1505 lnet_handler_t ln_mt_handler;
1508 * Completed when the discovery and monitor threads can enter their
1511 struct completion ln_started;
1513 struct list_head ln_udsp_list;
1515 /* Number of messages that have exceeded their message deadline */
1516 atomic_t ln_late_msg_count;
1517 /* Total amount of time past their deadline for all late ^ messages */
1518 atomic64_t ln_late_msg_nsecs;
1521 struct genl_filter_list {
1522 struct list_head lp_list;
1527 static const struct nla_policy scalar_attr_policy[LN_SCALAR_MAX + 1] = {
1528 [LN_SCALAR_ATTR_LIST] = { .type = NLA_NESTED },
1529 [LN_SCALAR_ATTR_LIST_SIZE] = { .type = NLA_U16 },
1530 [LN_SCALAR_ATTR_INDEX] = { .type = NLA_U16 },
1531 [LN_SCALAR_ATTR_NLA_TYPE] = { .type = NLA_U16 },
1532 [LN_SCALAR_ATTR_VALUE] = { .type = NLA_STRING },
1533 [LN_SCALAR_ATTR_KEY_FORMAT] = { .type = NLA_U16 },
1536 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
1537 const struct genl_family *family, int flags,
1538 u8 cmd, const struct ln_key_list *data[]);
1540 /* Special workaround for pre-4.19 kernels to send error messages
1541 * from dumpit routines. Newer kernels will send message with
1542 * NL_SET_ERR_MSG information by default if NETLINK_EXT_ACK is set.
1544 static inline int lnet_nl_send_error(struct sk_buff *msg, int portid, int seq,
1547 #ifndef HAVE_NL_DUMP_WITH_EXT_ACK
1548 struct nlmsghdr *nlh;
1553 nlh = nlmsg_put(msg, portid, seq, NLMSG_ERROR, sizeof(error), 0);
1556 #ifdef HAVE_NL_PARSE_WITH_EXT_ACK
1557 netlink_ack(msg, nlh, error, NULL);
1559 netlink_ack(msg, nlh, error);
1561 return nlmsg_len(nlh);