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_get)(int cmd, struct sk_buff *msg, int type, void *data);
330 int (*lnd_nl_set)(int cmd, struct nlattr *attr, int type, void *data);
332 const struct ln_key_list *lnd_keys;
335 struct lnet_tx_queue {
336 int tq_credits; /* # tx credits free */
337 int tq_credits_min; /* lowest it's been */
338 int tq_credits_max; /* total # tx credits */
339 struct list_head tq_delayed; /* delayed TXs */
342 enum lnet_net_state {
343 /* set when net block is allocated */
344 LNET_NET_STATE_INIT = 0,
345 /* set when NIs in net are started successfully */
346 LNET_NET_STATE_ACTIVE,
347 /* set if all NIs in net are in FAILED state */
348 LNET_NET_STATE_INACTIVE,
349 /* set when shutting down a NET */
350 LNET_NET_STATE_DELETING
354 /* initial state when NI is created */
355 LNET_NI_STATE_INIT = 0,
356 /* set when NI is brought up */
357 LNET_NI_STATE_ACTIVE,
358 /* set when NI is being shutdown */
359 LNET_NI_STATE_DELETING,
362 #define LNET_NI_RECOVERY_PENDING BIT(0)
363 #define LNET_NI_RECOVERY_FAILED BIT(1)
365 enum lnet_stats_type {
366 LNET_STATS_TYPE_SEND = 0,
367 LNET_STATS_TYPE_RECV,
371 struct lnet_comm_count {
372 atomic_t co_get_count;
373 atomic_t co_put_count;
374 atomic_t co_reply_count;
375 atomic_t co_ack_count;
376 atomic_t co_hello_count;
379 struct lnet_element_stats {
380 struct lnet_comm_count el_send_stats;
381 struct lnet_comm_count el_recv_stats;
382 struct lnet_comm_count el_drop_stats;
385 struct lnet_health_local_stats {
386 atomic_t hlt_local_interrupt;
387 atomic_t hlt_local_dropped;
388 atomic_t hlt_local_aborted;
389 atomic_t hlt_local_no_route;
390 atomic_t hlt_local_timeout;
391 atomic_t hlt_local_error;
394 struct lnet_health_remote_stats {
395 atomic_t hlt_remote_dropped;
396 atomic_t hlt_remote_timeout;
397 atomic_t hlt_remote_error;
398 atomic_t hlt_network_timeout;
402 /* chain on the ln_nets */
403 struct list_head net_list;
405 /* net ID, which is composed of
406 * (net_type << 16) | net_num.
407 * net_type can be one of the enumerated types defined in
408 * lnet/include/lnet/nidstr.h */
411 /* round robin selection */
414 /* total number of CPTs in the array */
417 /* cumulative CPTs of all NIs in this net */
420 /* relative net selection priority */
421 __u32 net_sel_priority;
423 /* network tunables */
424 struct lnet_ioctl_config_lnd_cmn_tunables net_tunables;
427 * boolean to indicate that the tunables have been set and
430 bool net_tunables_set;
432 /* procedural interface */
433 const struct lnet_lnd *net_lnd;
435 /* list of NIs on this net */
436 struct list_head net_ni_list;
438 /* list of NIs being added, but not started yet */
439 struct list_head net_ni_added;
441 /* dying LND instances */
442 struct list_head net_ni_zombie;
444 /* when I was last alive */
445 time64_t net_last_alive;
447 /* protects access to net_last_alive */
450 /* list of router nids preferred for this network */
451 struct list_head net_rtr_pref_nids;
454 /* Normally Netlink atttributes are defined in UAPI headers but Lustre is
455 * different in that the ABI is in a constant state of change unlike other
456 * Netlink interfaces. LNet sends a special header to help user land handle
460 /** enum lnet_net_attrs - LNet NI netlink properties
461 * attributes that describe LNet 'NI'
462 * These values are used to piece together
463 * messages for sending and receiving.
465 * @LNET_NET_ATTR_UNSPEC: unspecified attribute to catch errors
467 * @LNET_NET_ATTR_HDR: grouping for LNet net data (NLA_NUL_STRING)
468 * @LNET_NET_ATTR_TYPE: LNet net this NI belongs to (NLA_STRING)
469 * @LNET_NET_ATTR_LOCAL: Local NI information (NLA_NESTED)
471 enum lnet_net_attrs {
472 LNET_NET_ATTR_UNSPEC = 0,
478 __LNET_NET_ATTR_MAX_PLUS_ONE,
481 #define LNET_NET_ATTR_MAX (__LNET_NET_ATTR_MAX_PLUS_ONE - 1)
483 /** enum lnet_net_local_ni_attrs - LNet local NI netlink properties
484 * attributes that describe local
487 * @LNET_NET_LOCAL_NI_ATTR_UNSPEC: unspecified attribute to catch
490 * @LNET_NET_LOCAL_NI_ATTR_NID: NID that represents this NI
492 * @LNET_NET_LOCAL_NI_ATTR_STATUS: State of this NI (NLA_STRING)
493 * @LNET_NET_LOCAL_NI_ATTR_INTERFACE: Defines physical devices. used
494 * to be many devices but no longer
497 * @LNET_NET_LOCAL_NI_ATTR_STATS: NI general msg stats (NLA_NESTED)
498 * @LNET_NET_LOCAL_NI_ATTR_UDSP_INFO: NI UDSP state (NLA_NESTED)
499 * @LNET_NET_LOCAL_NI_ATTR_SEND_STATS: NI send stats (NLA_NESTED)
500 * @LNET_NET_LOCAL_NI_ATTR_RECV_STATS: NI recieved stats (NLA_NESTED)
501 * @LNET_NET_LOCAL_NI_ATTR_DROPPED_STATS: NI dropped stats (NLA_NESTED)
502 * @LNET_NET_LOCAL_NI_ATTR_HEALTH_STATS: NI health stats (NLA_NESTED)
503 * @LNET_NET_LOCAL_NI_ATTR_TUNABLES: NI tunables (NLA_NESTED)
504 * @LNET_NET_LOCAL_NI_ATTR_LND_TUNABLES: NI LND tunables (NLA_NESTED)
505 * @LNET_NET_LOCAL_NI_ATTR_DEV_CPT: NI CPT interface bound to
507 * @LNET_NET_LOCAL_NI_ATTR_CPTS: CPT core used by this NI
510 enum lnet_net_local_ni_attrs {
511 LNET_NET_LOCAL_NI_ATTR_UNSPEC = 0,
513 LNET_NET_LOCAL_NI_ATTR_NID,
514 LNET_NET_LOCAL_NI_ATTR_STATUS,
515 LNET_NET_LOCAL_NI_ATTR_INTERFACE,
517 LNET_NET_LOCAL_NI_ATTR_STATS,
518 LNET_NET_LOCAL_NI_ATTR_UDSP_INFO,
519 LNET_NET_LOCAL_NI_ATTR_SEND_STATS,
520 LNET_NET_LOCAL_NI_ATTR_RECV_STATS,
521 LNET_NET_LOCAL_NI_ATTR_DROPPED_STATS,
522 LNET_NET_LOCAL_NI_ATTR_HEALTH_STATS,
523 LNET_NET_LOCAL_NI_ATTR_TUNABLES,
524 LNET_NET_LOCAL_NI_ATTR_LND_TUNABLES,
525 LNET_NET_LOCAL_NI_DEV_CPT,
526 LNET_NET_LOCAL_NI_CPTS,
528 __LNET_NET_LOCAL_NI_ATTR_MAX_PLUS_ONE,
531 #define LNET_NET_LOCAL_NI_ATTR_MAX (__LNET_NET_LOCAL_NI_ATTR_MAX_PLUS_ONE - 1)
533 /** enum lnet_net_local_ni_intf_attrs - LNet NI device netlink properties
534 * attribute that reports the device
537 * @LNET_NET_LOCAL_NI_INTF_ATTR_UNSPEC: unspecified attribute to catch errors
539 * @LNET_NET_LOCAL_NI_INTF_ATTR_TYPE: Physcial device interface (NLA_STRING)
541 enum lnet_net_local_ni_intf_attrs {
542 LNET_NET_LOCAL_NI_INTF_ATTR_UNSPEC = 0,
544 LNET_NET_LOCAL_NI_INTF_ATTR_TYPE,
546 __LNET_NET_LOCAL_NI_INTF_ATTR_MAX_PLUS_ONE,
549 #define LNET_NET_LOCAL_NI_INTF_ATTR_MAX (__LNET_NET_LOCAL_NI_INTF_ATTR_MAX_PLUS_ONE - 1)
551 /** enum lnet_net_local_ni_stats_attrs - LNet NI netlink properties
552 * attributes that reports the
553 * network traffic stats
555 * @LNET_NET_LOCAL_NI_STATS_ATTR_UNSPEC: unspecified attribute to catch
558 * @LNET_NET_LOCAL_NI_STATS_ATTR_SEND_COUNT: Number of sent messages
560 * @LNET_NET_LOCAL_NI_STATS_ATTR_RECV_COUNT: Number of recieved messages
562 * @LNET_NET_LOCAL_NI_STATS_ATTR_DROP_COUNT: Number of dropped messages
565 enum lnet_net_local_ni_stats_attrs {
566 LNET_NET_LOCAL_NI_STATS_ATTR_UNSPEC = 0,
568 LNET_NET_LOCAL_NI_STATS_ATTR_SEND_COUNT,
569 LNET_NET_LOCAL_NI_STATS_ATTR_RECV_COUNT,
570 LNET_NET_LOCAL_NI_STATS_ATTR_DROP_COUNT,
571 __LNET_NET_LOCAL_NI_STATS_ATTR_MAX_PLUS_ONE,
574 #define LNET_NET_LOCAL_NI_STATS_ATTR_MAX (__LNET_NET_LOCAL_NI_STATS_ATTR_MAX_PLUS_ONE - 1)
576 /** enum lnet_net_local_ni_msg_stats_attrs - LNet NI netlink
577 * properties attributes
578 * that reports the message
581 * @LNET_NET_LOCAL_NI_MSG_STATS_ATTR_UNSPEC: unspecified attribute
584 * @LNET_NET_LOCAL_NI_MSG_STATS_ATTR_PUT_COUNT: Number of PUT messages
586 * @LNET_NET_LOCAL_NI_MSG_STATS_ATTR_GET_COUNT: Number of GET messages
588 * @LNET_NET_LOCAL_NI_MSG_STATS_ATTR_REPLY_COUNT: Number of REPLY messages
590 * @LNET_NET_LOCAL_NI_MSG_STATS_ATTR_ACK_COUNT: Number of ACK messages
592 * @LNET_NET_LOCAL_NI_MSG_STATS_ATTR_HELLO_COUNT: Number of HELLO messages
595 enum lnet_net_local_ni_msg_stats_attrs {
596 LNET_NET_LOCAL_NI_MSG_STATS_ATTR_UNSPEC = 0,
598 LNET_NET_LOCAL_NI_MSG_STATS_ATTR_PUT_COUNT,
599 LNET_NET_LOCAL_NI_MSG_STATS_ATTR_GET_COUNT,
600 LNET_NET_LOCAL_NI_MSG_STATS_ATTR_REPLY_COUNT,
601 LNET_NET_LOCAL_NI_MSG_STATS_ATTR_ACK_COUNT,
602 LNET_NET_LOCAL_NI_MSG_STATS_ATTR_HELLO_COUNT,
603 __LNET_NET_LOCAL_NI_MSG_STATS_ATTR_MAX_PLUS_ONE,
606 #define LNET_NET_LOCAL_NI_MSG_STATS_ATTR_MAX (__LNET_NET_LOCAL_NI_MSG_STATS_ATTR_MAX_PLUS_ONE - 1)
608 /** enum lnet_net_local_ni_health_stats_attrs - LNet NI netlink
609 * properties attributes
613 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_UNSPEC: unspecified attribute
616 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_FATAL_ERRORS: How many fatal errors
618 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_LEVEL: How healthly is NI
620 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_INTERRUPTS: How many interrupts
622 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_DROPPED: How much traffic has
623 * been dropped (NLA_U32)
624 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_ABORTED: How many aborts
626 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_NO_ROUTE: How often routing broke
628 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_TIMEOUTS: How often timeouts
630 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_ERROR: The number of errors
632 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_PING_COUNT: Number of successful
634 * @LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_NEXT_PING: Number of next pings
637 enum lnet_net_local_ni_health_stats_attrs {
638 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_UNSPEC = 0,
639 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_PAD = LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_UNSPEC,
641 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_FATAL_ERRORS,
642 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_LEVEL,
643 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_INTERRUPTS,
644 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_DROPPED,
645 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_ABORTED,
646 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_NO_ROUTE,
647 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_TIMEOUTS,
648 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_ERROR,
649 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_PING_COUNT,
650 LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_NEXT_PING,
651 __LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_MAX_PLUS_ONE,
653 #define LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_MAX (__LNET_NET_LOCAL_NI_HEALTH_STATS_ATTR_MAX_PLUS_ONE - 1)
655 /** enum lnet_net_local_ni_tunables_attrs - LNet NI tunables
656 * netlink properties.
657 * Performance options
660 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_UNSPEC: unspecified attribute
663 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_TIMEOUT: Timeout for LNet peer.
665 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_CREDITS: Credits for LNet peer.
667 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_BUFFER_CREDITS: Buffer credits for
668 * LNet peer. (NLA_S32)
669 * @LNET_NET_LOCAL_NI_TUNABLES_ATTR_CREDITS: Credits for LNet peer
672 enum lnet_net_local_ni_tunables_attr {
673 LNET_NET_LOCAL_NI_TUNABLES_ATTR_UNSPEC = 0,
675 LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_TIMEOUT,
676 LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_CREDITS,
677 LNET_NET_LOCAL_NI_TUNABLES_ATTR_PEER_BUFFER_CREDITS,
678 LNET_NET_LOCAL_NI_TUNABLES_ATTR_CREDITS,
679 __LNET_NET_LOCAL_NI_TUNABLES_ATTR_MAX_PLUS_ONE,
682 #define LNET_NET_LOCAL_NI_TUNABLES_ATTR_MAX (__LNET_NET_LOCAL_NI_TUNABLES_ATTR_MAX_PLUS_ONE - 1)
684 /** enum lnet_route_attrs - LNet route netlink
685 * attributes that describe
688 * @LNET_ROUTE_ATTR_UNSPEC: unspecified attribute to
691 * @LNET_ROUTE_ATTR_HDR: grouping for LNet route data
693 * @LNET_ROUTE_ATTR_NET: LNet remote network reached
694 * by the route (NLA_STRING)
695 * @LNET_ROUTE_ATTR_GATEWAY: gateway for the route
697 * @LNET_ROUTE_ATTR_HOP: route hop count (NLA_S32)
699 * @LNET_ROUTE_ATTR_PRIORITY: rank of this network path
701 * @LNET_ROUTE_ATTR_HEALTH_SENSITIVITY: rate of health value change
702 * for the route (NLA_U32)
703 * @LNET_ROUTE_ATTR_STATE: state of route (NLA_STRING)
705 * @LNET_ROUTE_ATTR_TYPE: Report if we support multi-hop
708 enum lnet_route_attrs {
709 LNET_ROUTE_ATTR_UNSPEC = 0,
713 LNET_ROUTE_ATTR_GATEWAY,
715 LNET_ROUTE_ATTR_PRIORITY,
716 LNET_ROUTE_ATTR_HEALTH_SENSITIVITY,
717 LNET_ROUTE_ATTR_STATE,
718 LNET_ROUTE_ATTR_TYPE,
719 __LNET_ROUTE_ATTR_MAX_PLUS_ONE,
722 #define LNET_ROUTE_ATTR_MAX (__LNET_ROUTE_ATTR_MAX_PLUS_ONE - 1)
724 /** enum lnet_peer_ni_attrs - LNet peer NI netlink properties
725 * attributes that describe LNet peer 'NI'.
726 * These values are used to piece together
727 * messages for sending and receiving.
729 * @LNET_PEER_NI_ATTR_UNSPEC: unspecified attribute to catch errors
731 * @LNET_PEER_NI_ATTR_HDR: grouping for LNet peer data
733 * @LNET_PEER_NI_ATTR_PRIMARY_NID: primary NID of this peer (NLA_STRING)
734 * @LNET_PEER_NI_ATTR_MULTIRAIL: Do we support MultiRail ? (NLA_FLAG)
735 * @LNET_PEER_NI_ATTR_STATE: Bitfields of the peer state (NLA_U32)
736 * @LNET_PEER_NI_ATTR_PEER_NI_LIST: List of remote peers we can reach
739 enum lnet_peer_ni_attrs {
740 LNET_PEER_NI_ATTR_UNSPEC = 0,
742 LNET_PEER_NI_ATTR_HDR,
743 LNET_PEER_NI_ATTR_PRIMARY_NID,
744 LNET_PEER_NI_ATTR_MULTIRAIL,
745 LNET_PEER_NI_ATTR_STATE,
746 LNET_PEER_NI_ATTR_PEER_NI_LIST,
747 __LNET_PEER_NI_ATTR_MAX_PLUS_ONE,
750 #define LNET_PEER_NI_ATTR_MAX (__LNET_PEER_NI_ATTR_MAX_PLUS_ONE - 1)
752 /** enum lnet_peer_ni_list_attrs - LNet remote peer netlink
753 * properties attributes that
754 * describe remote LNet peer 'NI'.
755 * These values are used to piece
756 * together messages for sending
759 * @LNET_PEER_NI_LIST_ATTR_UNSPEC: unspecified attribute to catch
762 * @LNET_PEER_NI_LIST_ATTR_NID: remote peer's NID (NLA_STRING)
763 * @LNET_PEER_NI_LIST_ATTR_UDSP_INFO: remote peer's UDSP info
765 * @LNET_PEER_NI_LIST_ATTR_STATE: state of remote peer
768 * @LNET_PEER_NI_LIST_ATTR_MAX_TX_CREDITS: Maximum TX credits for remote
770 * @LNET_PEER_NI_LIST_ATTR_CUR_TX_CREDITS: Current TX credits for remote
772 * @LNET_PEER_NI_LIST_ATTR_MIN_TX_CREDITS: Minimum TX credits for remote
774 * @LNET_PEER_NI_LIST_ATTR_QUEUE_BUF_COUNT: Size of TX queue buffer
776 * @LNET_PEER_NI_LIST_ATTR_CUR_RTR_CREDITS: Current router credits for
777 * remote peer (NLA_U32)
778 * @LNET_PEER_NI_LIST_ATTR_MIN_RTR_CREDITS: Minimum router credits for
779 * remote peer (NLA_U32)
780 * @LNET_PEER_NI_LIST_ATTR_REFCOUNT: Remote peer reference count
782 * @LNET_PEER_NI_LIST_ATTR_STATS_COUNT: Remote peer general stats,
783 * reports sent, recieved, and
784 * dropped packets. (NLA_NESTED)
786 * @LNET_PEER_NI_LIST_ATTR_SENT_STATS: Remote peer sent stats,
787 * reports gets, puts, acks, and
788 * hello packets. (NLA_NESTED)
789 * @LNET_PEER_NI_LIST_ATTR_RECV_STATS: Remote peer recieved stats,
790 * reports gets, puts, acks, and
791 * hello packets. (NLA_NESTED)
792 * @LNET_PEER_NI_LIST_ATTR_DROP_STATS: Remote peer dropped stats,
793 * reports gets, puts, acks, and
794 * hello packets. (NLA_NESTED)
795 * @LNET_PEER_NI_LIST_ATTR_HEALTH_STATS: Report the stats about the
796 * health of the remote peer.
799 enum lnet_peer_ni_list_attr {
800 LNET_PEER_NI_LIST_ATTR_UNSPEC = 0,
802 LNET_PEER_NI_LIST_ATTR_NID,
803 LNET_PEER_NI_LIST_ATTR_UDSP_INFO,
804 LNET_PEER_NI_LIST_ATTR_STATE,
806 LNET_PEER_NI_LIST_ATTR_MAX_TX_CREDITS,
807 LNET_PEER_NI_LIST_ATTR_CUR_TX_CREDITS,
808 LNET_PEER_NI_LIST_ATTR_MIN_TX_CREDITS,
809 LNET_PEER_NI_LIST_ATTR_QUEUE_BUF_COUNT,
810 LNET_PEER_NI_LIST_ATTR_CUR_RTR_CREDITS,
811 LNET_PEER_NI_LIST_ATTR_MIN_RTR_CREDITS,
812 LNET_PEER_NI_LIST_ATTR_REFCOUNT,
813 LNET_PEER_NI_LIST_ATTR_STATS_COUNT,
815 LNET_PEER_NI_LIST_ATTR_SENT_STATS,
816 LNET_PEER_NI_LIST_ATTR_RECV_STATS,
817 LNET_PEER_NI_LIST_ATTR_DROP_STATS,
818 LNET_PEER_NI_LIST_ATTR_HEALTH_STATS,
820 __LNET_PEER_NI_LIST_ATTR_MAX_PLUS_ONE,
823 #define LNET_PEER_NI_LIST_ATTR_MAX (__LNET_PEER_NI_LIST_ATTR_MAX_PLUS_ONE - 1)
825 /** enum lnet_peer_ni_list_stats_count - LNet remote peer traffic
826 * stats netlink properties
827 * attributes that provide
828 * traffic stats on the
829 * remote LNet peer 'NI'.
830 * These values are used to
831 * piece together messages
832 * for sending and receiving.
834 * @LNET_PEER_NI_LIST_STATS_COUNT_ATTR_UNSPEC: unspecified attribute to
837 * @LNET_PEER_NI_LIST_STATS_COUNT_ATTR_SEND_COUNT: Number of sent packets for
838 * remote peer (NLA_U32)
839 * @LNET_PEER_NI_LIST_STATS_COUNT_ATTR_RECV_COUNT: Numebr of recieved packets
840 * for remote peer (NLA_U32)
841 * @LNET_PEER_NI_LIST_STATS_COUNT_ATTR_DROP_COUNT: Number of dropped packets
842 * for remote peer (NLA_U32)
844 enum lnet_peer_ni_list_stats_count {
845 LNET_PEER_NI_LIST_STATS_COUNT_ATTR_UNSPEC = 0,
847 LNET_PEER_NI_LIST_STATS_COUNT_ATTR_SEND_COUNT,
848 LNET_PEER_NI_LIST_STATS_COUNT_ATTR_RECV_COUNT,
849 LNET_PEER_NI_LIST_STATS_COUNT_ATTR_DROP_COUNT,
850 __LNET_PEER_NI_LIST_STATS_COUNT_ATTR_MAX_PLUS_ONE,
853 #define LNET_PEER_NI_LIST_STATS_COUNT_ATTR_MAX (__LNET_PEER_NI_LIST_STATS_COUNT_ATTR_MAX_PLUS_ONE - 1)
855 /** enum lnet_peer_ni_list_stats - LNet remote peer stats netlink
856 * properties attributes that
857 * provide stats on the remote
858 * LNet peer 'NI'. These values are
859 * used to piece together messages
860 * for sending and receiving.
862 * @LNET_PEER_NI_LIST_STATS_ATTR_UNSPEC: unspecified attribute to catch
865 * @LNET_PEER_NI_LIST_STATS_ATTR_PUT: PUT message count for remote
867 * @LNET_PEER_NI_LIST_STATS_ATTR_GET: GET message count for remote
869 * @LNET_PEER_NI_LIST_STATS_ATTR_REPLY: REPLY message count for remote
871 * @LNET_PEER_NI_LIST_STATS_ATTR_ACK: ACK message count for remote
873 * @LNET_PEER_NI_LIST_STATS_ATTR_HEALTH: HELLO message count for remote
876 enum lnet_peer_ni_list_stats {
877 LNET_PEER_NI_LIST_STATS_ATTR_UNSPEC = 0,
879 LNET_PEER_NI_LIST_STATS_ATTR_PUT,
880 LNET_PEER_NI_LIST_STATS_ATTR_GET,
881 LNET_PEER_NI_LIST_STATS_ATTR_REPLY,
882 LNET_PEER_NI_LIST_STATS_ATTR_ACK,
883 LNET_PEER_NI_LIST_STATS_ATTR_HELLO,
884 __LNET_PEER_NI_LIST_STATS_ATTR_MAX_PLUS_ONE,
887 #define LNET_PEER_NI_LIST_STATS_ATTR_MAX (__LNET_PEER_NI_LIST_STATS_ATTR_MAX_PLUS_ONE - 1)
889 /** enum lnet_peer_ni_list_health_stats - LNet remote peer health
890 * stats netlink properties
891 * attributes that provide
892 * stats on the health of a
893 * remote LNet peer 'NI'.
894 * These values are used to
895 * piece together messages
896 * for sending and receiving.
898 * @LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_UNSPEC: unspecified attribute to
901 * @LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_VALUE: Health level of remote
903 * @LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_DROP: drop message state for
904 * remote peer (NLA_U32)
905 * @LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_TIMEOUT: timeout set for remote
907 * @LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_ERROR: total errors for remote
909 * @LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_NETWORK_TIMEOUT: network timeout for
910 * remote peer (NLA_U32)
911 * @LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_PING_COUNT: number of pings for
912 * remote peer (NLA_U32)
913 * @LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_NEXT_PING: timestamp for next ping
914 * sent by remote peer
917 enum lnet_peer_ni_list_health_stats {
918 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_UNSPEC = 0,
919 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_PAD = LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_UNSPEC,
921 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_VALUE,
922 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_DROPPED,
923 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_TIMEOUT,
924 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_ERROR,
925 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_NETWORK_TIMEOUT,
926 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_PING_COUNT,
927 LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_NEXT_PING,
929 __LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_MAX_PLUS_ONE,
932 #define LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_MAX (__LNET_PEER_NI_LIST_HEALTH_STATS_ATTR_MAX_PLUS_ONE - 1)
934 /** LNet netlink ping API */
936 /** enum lnet_ping_attr - LNet ping netlink properties
937 * attributes to describe ping format
938 * These values are used to piece together
939 * messages for sending and receiving.
941 * @LNET_PING_ATTR_UNSPEC: unspecified attribute to catch errors
943 * @LNET_PING_ATTR_HDR: grouping for LNet ping data (NLA_NUL_STRING)
944 * @LNET_PING_ATTR_PRIMARY_NID: Source NID for ping request (NLA_STRING)
945 * @LNET_PING_ATTR_ERRNO: error code if we fail to ping (NLA_S16)
946 * @LNET_PING_ATTR_MULTIRAIL: Report if MR is supported (NLA_FLAG)
947 * @LNET_PING_ATTR_PEER_NI_LIST: List of peer NI's (NLA_NESTED)
949 enum lnet_ping_attr {
950 LNET_PING_ATTR_UNSPEC = 0,
953 LNET_PING_ATTR_PRIMARY_NID,
954 LNET_PING_ATTR_ERRNO,
955 LNET_PING_ATTR_MULTIRAIL,
956 LNET_PING_ATTR_PEER_NI_LIST,
957 __LNET_PING_ATTR_MAX_PLUS_ONE,
960 #define LNET_PING_ATTR_MAX (__LNET_PING_ATTR_MAX_PLUS_ONE - 1)
962 /** enum lnet_ping_peer_ni_attr - LNet peer ni information reported by
963 * ping command. A list of these are
964 * returned with a ping request.
966 * @LNET_PING_PEER_NI_ATTR_UNSPEC: unspecified attribute to catch errrors
968 * @LNET_PING_PEER_NI_ATTR_NID: NID address of peer NI. (NLA_STRING)
970 enum lnet_ping_peer_ni_attr {
971 LNET_PING_PEER_NI_ATTR_UNSPEC = 0,
973 LNET_PING_PEER_NI_ATTR_NID,
974 __LNET_PING_PEER_NI_ATTR_MAX_PLUS_ONE,
977 #define LNET_PING_PEER_NI_ATTR_MAX (__LNET_PING_PEER_NI_ATTR_MAX_PLUS_ONE - 1)
979 /** enum lnet_udsp_info_attr - LNet UDSP information reported for
980 * some subsystem that tracks it.
982 * @LNET_UDSP_INFO_ATTR_UNSPEC: unspecified attribute to catch errors
984 * @LNET_UDSP_INFO_ATTR_NET_PRIORITY, LNet net priority in selection.
986 * @LNET_UDSP_INFO_ATTR_NID_PRIORITY, NID's priority in selection.
988 * @LNET_UDSP_INFO_ATTR_PREF_RTR_NIDS_LIST: Which gateway's are preferred.
990 * @LNET_UDSP_INFO_ATTR_PREF_NIDS_LIST: Which NIDs are preferred.
993 enum lnet_udsp_info_attr {
994 LNET_UDSP_INFO_ATTR_UNSPEC = 0,
996 LNET_UDSP_INFO_ATTR_NET_PRIORITY,
997 LNET_UDSP_INFO_ATTR_NID_PRIORITY,
998 LNET_UDSP_INFO_ATTR_PREF_RTR_NIDS_LIST,
999 LNET_UDSP_INFO_ATTR_PREF_NIDS_LIST,
1000 __LNET_UDSP_INFO_ATTR_MAX_PLUS_ONE,
1003 #define LNET_UDSP_INFO_ATTR_MAX (__LNET_UDSP_INFO_ATTR_MAX_PLUS_ONE - 1)
1005 /** enum lnet_udsp_info_pref_nids_attr - LNet UDSP information reported for
1006 * some subsystem that tracks it.
1008 * @LNET_UDSP_INFO_PREF_NIDS_ATTR_UNSPEC: unspecified attribute to catch errors
1010 * @LNET_UDSP_INFO_PREF_NIDS_ATTR_INDEX, UDSP prority NIDs label
1012 * @LNET_UDSP_INFO_PREF_NIDS_ATTR_NID, UDSP prority NID (NLA_STRING)
1014 enum lnet_udsp_info_pref_nids_attr {
1015 LNET_UDSP_INFO_PREF_NIDS_ATTR_UNSPEC = 0,
1017 LNET_UDSP_INFO_PREF_NIDS_ATTR_INDEX,
1018 LNET_UDSP_INFO_PREF_NIDS_ATTR_NID,
1019 __LNET_UDSP_INFO_PREF_NIDS_ATTR_MAX_PLUS_ONE,
1022 #define LNET_UDSP_INFO_PREF_NIDS_ATTR_MAX (__LNET_UDSP_INFO_PREF_NIDS_ATTR_MAX_PLUS_ONE - 1)
1025 /* chain on the lnet_net structure */
1026 struct list_head ni_netlist;
1028 /* chain on the recovery queue */
1029 struct list_head ni_recovery;
1031 /* MD handle for recovery ping */
1032 struct lnet_handle_md ni_ping_mdh;
1036 /* number of CPTs */
1039 /* bond NI on some CPTs */
1042 /* interface's NID */
1043 struct lnet_nid ni_nid;
1045 /* instance-specific data */
1048 /* per ni credits */
1049 atomic_t ni_tx_credits;
1051 /* percpt TX queues */
1052 struct lnet_tx_queue **ni_tx_queues;
1054 /* percpt reference count */
1057 /* pointer to parent network */
1058 struct lnet_net *ni_net;
1060 /* my health status */
1063 /* NI FSM. Protected by lnet_ni_lock() */
1064 enum lnet_ni_state ni_state;
1066 /* Recovery state. Protected by lnet_ni_lock() */
1067 __u32 ni_recovery_state;
1069 /* When to send the next recovery ping */
1070 time64_t ni_next_ping;
1071 /* How many pings sent during current recovery period did not receive
1072 * a reply. NB: reset whenever _any_ message arrives on this NI
1074 unsigned int ni_ping_count;
1076 /* per NI LND tunables */
1077 struct lnet_lnd_tunables ni_lnd_tunables;
1079 /* lnd tunables set explicitly */
1080 bool ni_lnd_tunables_set;
1083 struct lnet_element_stats ni_stats;
1084 struct lnet_health_local_stats ni_hstats;
1086 /* physical device CPT */
1089 /* sequence number used to round robin over nis within a net */
1094 * initialized to LNET_MAX_HEALTH_VALUE
1095 * Value is decremented every time we fail to send a message over
1096 * this NI because of a NI specific failure.
1097 * Value is incremented if we successfully send a message.
1099 atomic_t ni_healthv;
1102 * Set to 1 by the LND when it receives an event telling it the device
1103 * has gone into a fatal state. Set to 0 when the LND receives an
1104 * even telling it the device is back online.
1106 atomic_t ni_fatal_error_on;
1108 /* the relative selection priority of this NI */
1109 __u32 ni_sel_priority;
1112 * equivalent interface to use
1115 struct net *ni_net_ns; /* original net namespace */
1118 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL
1121 * Descriptor of a ping info buffer: keep a separate indicator of the
1122 * size and a reference count. The type is used both as a source and
1123 * sink of data, so we need to keep some information outside of the
1124 * area that may be overwritten by network data.
1126 struct lnet_ping_buffer {
1127 int pb_nbytes; /* sizeof pb_info */
1130 struct lnet_ping_info pb_info;
1133 #define LNET_PING_BUFFER_SIZE(bytes) \
1134 (offsetof(struct lnet_ping_buffer, pb_info) + bytes)
1135 #define LNET_PING_BUFFER_LONI(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_nid)
1136 #define LNET_PING_BUFFER_SEQNO(PBUF) ((PBUF)->pb_info.pi_ni[0].ns_status)
1138 #define LNET_PING_INFO_TO_BUFFER(PINFO) \
1139 container_of((PINFO), struct lnet_ping_buffer, pb_info)
1142 lnet_ping_sts_size(const struct lnet_nid *nid)
1146 if (nid_is_nid4(nid))
1147 return sizeof(struct lnet_ni_status);
1149 size = offsetof(struct lnet_ni_large_status, ns_nid) +
1152 return round_up(size, 4);
1155 static inline struct lnet_ni_large_status *
1156 lnet_ping_sts_next(const struct lnet_ni_large_status *nis)
1158 return (void *)nis + lnet_ping_sts_size(&nis->ns_nid);
1162 lnet_ping_at_least_two_entries(const struct lnet_ping_info *pi)
1164 /* Return true if we have at lease two entries. There is always a
1165 * least one, a 4-byte lo0 interface.
1167 struct lnet_ni_large_status *lns;
1169 if ((pi->pi_features & LNET_PING_FEAT_LARGE_ADDR) == 0)
1170 return pi->pi_nnis <= 2;
1171 /* There is at least 1 large-address entry */
1172 if (pi->pi_nnis != 1)
1174 lns = (void *)&pi->pi_ni[1];
1175 lns = lnet_ping_sts_next(lns);
1177 return ((void *)pi + lnet_ping_info_size(pi) <= (void *)lns);
1180 struct lnet_nid_list {
1181 struct list_head nl_list;
1182 struct lnet_nid nl_nid;
1185 struct lnet_peer_ni {
1186 /* chain on lpn_peer_nis */
1187 struct list_head lpni_peer_nis;
1188 /* chain on remote peer list */
1189 struct list_head lpni_on_remote_peer_ni_list;
1190 /* chain on recovery queue */
1191 struct list_head lpni_recovery;
1192 /* chain on peer hash */
1193 struct list_head lpni_hashlist;
1194 /* messages blocking for tx credits */
1195 struct list_head lpni_txq;
1196 /* pointer to peer net I'm part of */
1197 struct lnet_peer_net *lpni_peer_net;
1198 /* statistics kept on each peer NI */
1199 struct lnet_element_stats lpni_stats;
1200 struct lnet_health_remote_stats lpni_hstats;
1201 /* spin lock protecting credits and lpni_txq */
1202 spinlock_t lpni_lock;
1203 /* # tx credits available */
1205 /* low water mark */
1206 int lpni_mintxcredits;
1208 * Each peer_ni in a gateway maintains its own credits. This
1209 * allows more traffic to gateways that have multiple interfaces.
1211 /* # router credits */
1212 int lpni_rtrcredits;
1213 /* low water mark */
1214 int lpni_minrtrcredits;
1215 /* bytes queued for sending */
1217 /* network peer is on */
1218 struct lnet_net *lpni_net;
1220 struct lnet_nid lpni_nid;
1222 struct kref lpni_kref;
1223 /* health value for the peer */
1224 atomic_t lpni_healthv;
1225 /* recovery ping mdh */
1226 struct lnet_handle_md lpni_recovery_ping_mdh;
1227 /* When to send the next recovery ping */
1228 time64_t lpni_next_ping;
1229 /* How many pings sent during current recovery period did not receive
1230 * a reply. NB: reset whenever _any_ message arrives from this peer NI
1232 unsigned int lpni_ping_count;
1233 /* CPT this peer attached on */
1235 /* state flags -- protected by lpni_lock */
1236 unsigned lpni_state;
1237 /* status of the peer NI as reported by the peer */
1238 __u32 lpni_ns_status;
1239 /* sequence number used to round robin over peer nis within a net */
1241 /* sequence number used to round robin over gateways */
1243 /* returned RC ping features. Protected with lpni_lock */
1244 unsigned int lpni_ping_feats;
1245 /* time last message was received from the peer */
1246 time64_t lpni_last_alive;
1247 /* preferred local nids: if only one, use lpni_pref.nid */
1249 struct lnet_nid nid;
1250 struct list_head nids;
1252 /* list of router nids preferred for this peer NI */
1253 struct list_head lpni_rtr_pref_nids;
1254 /* The relative selection priority of this peer NI */
1255 __u32 lpni_sel_priority;
1256 /* number of preferred NIDs in lnpi_pref_nids */
1257 __u32 lpni_pref_nnids;
1260 /* Preferred path added due to traffic on non-MR peer_ni */
1261 #define LNET_PEER_NI_NON_MR_PREF BIT(0)
1262 /* peer is being recovered. */
1263 #define LNET_PEER_NI_RECOVERY_PENDING BIT(1)
1264 /* recovery ping failed */
1265 #define LNET_PEER_NI_RECOVERY_FAILED BIT(2)
1266 /* peer is being deleted */
1267 #define LNET_PEER_NI_DELETING BIT(3)
1270 /* chain on pt_peer_list */
1271 struct list_head lp_peer_list;
1273 /* list of peer nets */
1274 struct list_head lp_peer_nets;
1276 /* list of messages pending discovery*/
1277 struct list_head lp_dc_pendq;
1279 /* chain on router list */
1280 struct list_head lp_rtr_list;
1282 /* primary NID of the peer */
1283 struct lnet_nid lp_primary_nid;
1285 /* source NID to use during discovery */
1286 struct lnet_nid lp_disc_src_nid;
1287 /* destination NID to use during discovery */
1288 struct lnet_nid lp_disc_dst_nid;
1290 /* net to perform discovery on */
1291 __u32 lp_disc_net_id;
1293 /* CPT of peer_table */
1296 /* number of NIDs on this peer */
1299 /* # refs from lnet_route::lr_gateway */
1300 int lp_rtr_refcount;
1303 * peer specific health sensitivity value to decrement peer nis in
1304 * this peer with if set to something other than 0
1306 __u32 lp_health_sensitivity;
1308 /* messages blocking for router credits */
1309 struct list_head lp_rtrq;
1311 /* routes on this peer */
1312 struct list_head lp_routes;
1314 /* reference count */
1315 atomic_t lp_refcount;
1317 /* lock protecting peer state flags and lpni_rtrq */
1320 /* peer state flags */
1323 /* buffer for data pushed by peer */
1324 struct lnet_ping_buffer *lp_data;
1326 /* MD handle for ping in progress */
1327 struct lnet_handle_md lp_ping_mdh;
1329 /* MD handle for push in progress */
1330 struct lnet_handle_md lp_push_mdh;
1332 /* number of bytes for sizing pb_info in push data */
1335 /* NI config sequence number of peer */
1336 __u32 lp_peer_seqno;
1338 /* Local NI config sequence number acked by peer */
1339 __u32 lp_node_seqno;
1341 /* Local NI config sequence number sent to peer */
1342 __u32 lp_node_seqno_sent;
1344 /* Ping error encountered during discovery. */
1347 /* Push error encountered during discovery. */
1350 /* Error encountered during discovery. */
1353 /* time it was put on the ln_dc_working queue */
1354 time64_t lp_last_queued;
1356 /* link on discovery-related lists */
1357 struct list_head lp_dc_list;
1359 /* tasks waiting on discovery of this peer */
1360 wait_queue_head_t lp_dc_waitq;
1362 /* cached peer aliveness */
1365 /* sequence number used to round robin traffic to this peer's
1370 /* timestamp of primary nid lock */
1371 __u64 lp_prim_lock_ts;
1375 * The status flags in lp_state. Their semantics have chosen so that
1376 * lp_state can be zero-initialized.
1378 * A peer is marked MULTI_RAIL in two cases: it was configured using DLC
1379 * as multi-rail aware, or the LNET_PING_FEAT_MULTI_RAIL bit was set.
1381 * A peer is marked NO_DISCOVERY if the LNET_PING_FEAT_DISCOVERY bit was
1382 * NOT set when the peer was pinged by discovery.
1384 * A peer is marked ROUTER if it indicates so in the feature bit.
1386 #define LNET_PEER_MULTI_RAIL BIT(0) /* Multi-rail aware */
1387 #define LNET_PEER_NO_DISCOVERY BIT(1) /* Peer disabled discovery */
1388 #define LNET_PEER_ROUTER_ENABLED BIT(2) /* router feature enabled */
1391 * A peer is marked CONFIGURED if it was configured by DLC.
1393 * In addition, a peer is marked DISCOVERED if it has fully passed
1394 * through Peer Discovery.
1396 * When Peer Discovery is disabled, the discovery thread will mark
1397 * peers REDISCOVER to indicate that they should be re-examined if
1398 * discovery is (re)enabled on the node.
1400 * A peer that was created as the result of inbound traffic will not
1403 #define LNET_PEER_CONFIGURED BIT(3) /* Configured via DLC */
1404 #define LNET_PEER_DISCOVERED BIT(4) /* Peer was discovered */
1405 #define LNET_PEER_REDISCOVER BIT(5) /* Discovery was disabled */
1407 * A peer is marked DISCOVERING when discovery is in progress.
1408 * The other flags below correspond to stages of discovery.
1410 #define LNET_PEER_DISCOVERING BIT(6) /* Discovering */
1411 #define LNET_PEER_DATA_PRESENT BIT(7) /* Remote peer data present */
1412 #define LNET_PEER_NIDS_UPTODATE BIT(8) /* Remote peer info uptodate */
1413 #define LNET_PEER_PING_SENT BIT(9) /* Waiting for REPLY to Ping */
1414 #define LNET_PEER_PUSH_SENT BIT(10) /* Waiting for ACK of Push */
1415 #define LNET_PEER_PING_FAILED BIT(11) /* Ping send failure */
1416 #define LNET_PEER_PUSH_FAILED BIT(12) /* Push send failure */
1418 * A ping can be forced as a way to fix up state, or as a manual
1419 * intervention by an admin.
1420 * A push can be forced in circumstances that would normally not
1421 * allow for one to happen.
1423 #define LNET_PEER_FORCE_PING BIT(13) /* Forced Ping */
1424 #define LNET_PEER_FORCE_PUSH BIT(14) /* Forced Push */
1426 /* force delete even if router */
1427 #define LNET_PEER_RTR_NI_FORCE_DEL BIT(15)
1429 /* gw undergoing alive discovery */
1430 #define LNET_PEER_RTR_DISCOVERY BIT(16)
1431 /* gw has undergone discovery (does not indicate success or failure) */
1432 #define LNET_PEER_RTR_DISCOVERED BIT(17)
1434 /* peer is marked for deletion */
1435 #define LNET_PEER_MARK_DELETION BIT(18)
1436 /* lnet_peer_del()/lnet_peer_del_locked() has been called on the peer */
1437 #define LNET_PEER_MARK_DELETED BIT(19)
1438 /* lock primary NID to what's requested by ULP */
1439 #define LNET_PEER_LOCK_PRIMARY BIT(20)
1440 /* this is for informational purposes only. It is set if a peer gets
1441 * configured from Lustre with a primary NID which belongs to another peer
1442 * which is also configured by Lustre as the primary NID.
1444 #define LNET_PEER_BAD_CONFIG BIT(21)
1446 struct lnet_peer_net {
1447 /* chain on lp_peer_nets */
1448 struct list_head lpn_peer_nets;
1450 /* list of peer_nis on this network */
1451 struct list_head lpn_peer_nis;
1453 /* pointer to the peer I'm part of */
1454 struct lnet_peer *lpn_peer;
1459 /* peer net health */
1462 /* time of next router ping on this net */
1463 time64_t lpn_next_ping;
1465 /* selection sequence number */
1468 /* relative peer net selection priority */
1469 __u32 lpn_sel_priority;
1471 /* reference count */
1472 atomic_t lpn_refcount;
1475 /* peer hash size */
1476 #define LNET_PEER_HASH_BITS 9
1477 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS)
1480 * peer hash table - one per CPT
1482 * protected by lnet_net_lock/EX for update
1487 * protected by pt_zombie_lock:
1491 * pt_zombie lock nests inside lnet_net_lock
1493 struct lnet_peer_table {
1494 int pt_version; /* /proc validity stamp */
1495 struct list_head *pt_hash; /* NID->peer hash */
1496 struct list_head pt_peer_list; /* peers */
1497 int pt_peers; /* # peers */
1498 struct list_head pt_zombie_list; /* zombie peer_ni */
1499 int pt_zombies; /* # zombie peers_ni */
1500 spinlock_t pt_zombie_lock; /* protect list and count */
1503 /* peer aliveness is enabled only on routers for peers in a network where the
1504 * struct lnet_ni::ni_peertimeout has been set to a positive value
1506 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
1507 ((lp)->lpni_net) && \
1508 (lp)->lpni_net->net_tunables.lct_peer_timeout > 0)
1511 struct list_head lr_list; /* chain on net */
1512 struct list_head lr_gwlist; /* chain on gateway */
1513 struct lnet_peer *lr_gateway; /* router node */
1514 struct lnet_nid lr_nid; /* NID used to add route */
1515 __u32 lr_net; /* remote network number */
1516 __u32 lr_lnet; /* local network number */
1517 int lr_seq; /* sequence for round-robin */
1518 __u32 lr_hops; /* how far I am */
1519 unsigned int lr_priority; /* route priority */
1520 atomic_t lr_alive; /* cached route aliveness */
1521 bool lr_single_hop; /* this route is single-hop */
1524 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7)
1525 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16)
1526 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits)
1528 struct lnet_remotenet {
1529 /* chain on ln_remote_nets_hash */
1530 struct list_head lrn_list;
1532 struct list_head lrn_routes;
1537 /** lnet message has credit and can be submitted to lnd for send/receive */
1538 #define LNET_CREDIT_OK 0
1539 /** lnet message is waiting for credit */
1540 #define LNET_CREDIT_WAIT 1
1541 /** lnet message is waiting for discovery */
1542 #define LNET_DC_WAIT 2
1544 struct lnet_rtrbufpool {
1545 /* my free buffer pool */
1546 struct list_head rbp_bufs;
1547 /* messages blocking for a buffer */
1548 struct list_head rbp_msgs;
1549 /* # pages in each buffer */
1551 /* requested number of buffers */
1552 int rbp_req_nbuffers;
1553 /* # buffers actually allocated */
1555 /* # free buffers / blocked messages */
1557 /* low water mark */
1561 struct lnet_rtrbuf {
1562 struct list_head rb_list; /* chain on rbp_bufs */
1563 struct lnet_rtrbufpool *rb_pool; /* owning pool */
1564 struct bio_vec rb_kiov[0]; /* the buffer space */
1567 #define LNET_PEER_HASHSIZE 503 /* prime! */
1569 enum lnet_match_flags {
1570 /* Didn't match anything */
1571 LNET_MATCHMD_NONE = BIT(0),
1573 LNET_MATCHMD_OK = BIT(1),
1574 /* Must be discarded */
1575 LNET_MATCHMD_DROP = BIT(2),
1576 /* match and buffer is exhausted */
1577 LNET_MATCHMD_EXHAUSTED = BIT(3),
1579 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
1582 /* Options for struct lnet_portal::ptl_options */
1583 #define LNET_PTL_LAZY BIT(0)
1584 #define LNET_PTL_MATCH_UNIQUE BIT(1) /* unique match, for RDMA */
1585 #define LNET_PTL_MATCH_WILDCARD BIT(2) /* wildcard match, request portal */
1587 /* parameter for matching operations (GET, PUT) */
1588 struct lnet_match_info {
1590 struct lnet_processid mi_id;
1591 unsigned int mi_cpt;
1592 unsigned int mi_opc;
1593 unsigned int mi_portal;
1594 unsigned int mi_rlength;
1595 unsigned int mi_roffset;
1598 /* ME hash of RDMA portal */
1599 #define LNET_MT_HASH_BITS 8
1600 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS)
1601 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1)
1602 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
1603 * the last entry is reserved for MEs with ignore-bits */
1604 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE
1605 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
1606 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
1607 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
1608 #define LNET_MT_BITS_U64 6 /* 2^6 bits */
1609 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
1610 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
1612 /* portal match table */
1613 struct lnet_match_table {
1614 /* reserved for upcoming patches, CPU partition ID */
1615 unsigned int mt_cpt;
1616 unsigned int mt_portal; /* portal index */
1617 /* match table is set as "enabled" if there's non-exhausted MD
1618 * attached on mt_mhash, it's only valid for wildcard portal */
1619 unsigned int mt_enabled;
1620 /* bitmap to flag whether MEs on mt_hash are exhausted or not */
1621 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
1622 struct list_head *mt_mhash; /* matching hash */
1625 /* these are only useful for wildcard portal */
1626 /* Turn off message rotor for wildcard portals */
1627 #define LNET_PTL_ROTOR_OFF 0
1628 /* round-robin dispatch all PUT messages for wildcard portals */
1629 #define LNET_PTL_ROTOR_ON 1
1630 /* round-robin dispatch routed PUT message for wildcard portals */
1631 #define LNET_PTL_ROTOR_RR_RT 2
1632 /* dispatch routed PUT message by hashing source NID for wildcard portals */
1633 #define LNET_PTL_ROTOR_HASH_RT 3
1635 struct lnet_portal {
1636 spinlock_t ptl_lock;
1637 unsigned int ptl_index; /* portal ID, reserved */
1638 /* flags on this portal: lazy, unique... */
1639 unsigned int ptl_options;
1640 /* list of messages which are stealing buffer */
1641 struct list_head ptl_msg_stealing;
1642 /* messages blocking for MD */
1643 struct list_head ptl_msg_delayed;
1644 /* Match table for each CPT */
1645 struct lnet_match_table **ptl_mtables;
1646 /* spread rotor of incoming "PUT" */
1647 unsigned int ptl_rotor;
1648 /* # active entries for this portal */
1650 /* array of active entries' cpu-partition-id */
1654 #define LNET_LH_HASH_BITS 12
1655 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS)
1656 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1)
1658 /* resource container (ME, MD, EQ) */
1659 struct lnet_res_container {
1660 unsigned int rec_type; /* container type */
1661 __u64 rec_lh_cookie; /* cookie generator */
1662 struct list_head rec_active; /* active resource list */
1663 struct list_head *rec_lh_hash; /* handle hash */
1666 /* message container */
1667 struct lnet_msg_container {
1668 int msc_init; /* initialized or not */
1669 /* max # threads finalizing */
1670 int msc_nfinalizers;
1671 /* msgs waiting to complete finalizing */
1672 struct list_head msc_finalizing;
1673 /* msgs waiting to be resent */
1674 struct list_head msc_resending;
1675 struct list_head msc_active; /* active message list */
1676 /* threads doing finalization */
1677 void **msc_finalizers;
1678 /* threads doing resends */
1679 void **msc_resenders;
1682 /* This UDSP structures need to match the user space liblnetconfig structures
1683 * in order for the marshall and unmarshall functions to be common.
1686 /* Net is described as a
1690 struct lnet_ud_net_descr {
1692 struct list_head udn_net_num_range;
1695 /* each NID range is defined as
1697 * 2. address range descriptor
1699 struct lnet_ud_nid_descr {
1700 struct lnet_ud_net_descr ud_net_id;
1701 struct list_head ud_addr_range;
1705 /* a UDSP rule can have up to three user defined NID descriptors
1706 * - src: defines the local NID range for the rule
1707 * - dst: defines the peer NID range for the rule
1708 * - rte: defines the router NID range for the rule
1710 * An action union defines the action to take when the rule
1714 struct list_head udsp_on_list;
1716 struct lnet_ud_nid_descr udsp_src;
1717 struct lnet_ud_nid_descr udsp_dst;
1718 struct lnet_ud_nid_descr udsp_rte;
1719 enum lnet_udsp_action_type udsp_action_type;
1721 __u32 udsp_priority;
1725 /* Peer Discovery states */
1726 #define LNET_DC_STATE_SHUTDOWN 0 /* not started */
1727 #define LNET_DC_STATE_RUNNING 1 /* started up OK */
1728 #define LNET_DC_STATE_STOPPING 2 /* telling thread to stop */
1730 /* Router Checker states */
1731 #define LNET_MT_STATE_SHUTDOWN 0 /* not started */
1732 #define LNET_MT_STATE_RUNNING 1 /* started up OK */
1733 #define LNET_MT_STATE_STOPPING 2 /* telling thread to stop */
1736 #define LNET_STATE_SHUTDOWN 0 /* not started */
1737 #define LNET_STATE_RUNNING 1 /* started up OK */
1738 #define LNET_STATE_STOPPING 2 /* telling thread to stop */
1741 /* CPU partition table of LNet */
1742 struct cfs_cpt_table *ln_cpt_table;
1743 /* number of CPTs in ln_cpt_table */
1744 unsigned int ln_cpt_number;
1745 unsigned int ln_cpt_bits;
1747 /* protect LNet resources (ME/MD/EQ) */
1748 struct cfs_percpt_lock *ln_res_lock;
1751 /* the vector of portals */
1752 struct lnet_portal **ln_portals;
1753 /* percpt MD container */
1754 struct lnet_res_container **ln_md_containers;
1756 /* Event Queue container */
1757 struct lnet_res_container ln_eq_container;
1758 spinlock_t ln_eq_wait_lock;
1760 unsigned int ln_remote_nets_hbits;
1762 /* protect NI, peer table, credits, routers, rtrbuf... */
1763 struct cfs_percpt_lock *ln_net_lock;
1764 /* percpt message containers for active/finalizing/freed message */
1765 struct lnet_msg_container **ln_msg_containers;
1766 struct lnet_counters **ln_counters;
1767 struct lnet_peer_table **ln_peer_tables;
1768 /* list of peer nis not on a local network */
1769 struct list_head ln_remote_peer_ni_list;
1770 /* failure simulation */
1771 struct list_head ln_test_peers;
1772 struct list_head ln_drop_rules;
1773 struct list_head ln_delay_rules;
1775 struct list_head ln_nets;
1776 /* Sequence number used to round robin sends across all nets */
1778 /* the loopback NI */
1779 struct lnet_ni *ln_loni;
1780 /* network zombie list */
1781 struct list_head ln_net_zombie;
1782 /* resend messages list */
1783 struct list_head ln_msg_resend;
1784 /* spin lock to protect the msg resend list */
1785 spinlock_t ln_msg_resend_lock;
1787 /* remote networks with routes to them */
1788 struct list_head *ln_remote_nets_hash;
1789 /* validity stamp */
1790 __u64 ln_remote_nets_version;
1791 /* list of all known routers */
1792 struct list_head ln_routers;
1793 /* validity stamp */
1794 __u64 ln_routers_version;
1795 /* percpt router buffer pools */
1796 struct lnet_rtrbufpool **ln_rtrpools;
1799 * Ping target / Push source
1801 * The ping target and push source share a single buffer. The
1802 * ln_ping_target is protected against concurrent updates by
1805 struct lnet_handle_md ln_ping_target_md;
1806 lnet_handler_t ln_ping_target_handler;
1807 struct lnet_ping_buffer *ln_ping_target;
1808 atomic_t ln_ping_target_seqno;
1813 * ln_push_nnis contains the desired size of the push target.
1814 * The lnet_net_lock is used to handle update races. The old
1815 * buffer may linger a while after it has been unlinked, in
1816 * which case the event handler cleans up.
1818 lnet_handler_t ln_push_target_handler;
1819 struct lnet_handle_md ln_push_target_md;
1820 struct lnet_ping_buffer *ln_push_target;
1821 /* bytes needed for pb_info to receive push */
1822 int ln_push_target_nbytes;
1824 /* discovery event queue handle */
1825 lnet_handler_t ln_dc_handler;
1826 /* discovery requests */
1827 struct list_head ln_dc_request;
1828 /* discovery working list */
1829 struct list_head ln_dc_working;
1830 /* discovery expired list */
1831 struct list_head ln_dc_expired;
1832 /* discovery thread wait queue */
1833 wait_queue_head_t ln_dc_waitq;
1834 /* discovery startup/shutdown state */
1837 /* monitor thread startup/shutdown state */
1839 /* serialise startup/shutdown */
1840 struct semaphore ln_mt_signal;
1842 struct mutex ln_api_mutex;
1843 struct mutex ln_lnd_mutex;
1844 /* Have I called LNetNIInit myself? */
1846 /* LNetNIInit/LNetNIFini counter */
1848 /* SHUTDOWN/RUNNING/STOPPING */
1851 int ln_routing; /* am I a router? */
1852 lnet_pid_t ln_pid; /* requested pid */
1853 /* uniquely identifies this ni in this epoch */
1854 __u64 ln_interface_cookie;
1855 /* registered LNDs */
1856 const struct lnet_lnd *ln_lnds[NUM_LNDS];
1858 /* test protocol compatibility flags */
1859 unsigned long ln_testprotocompat;
1861 /* 0 - load the NIs from the mod params
1862 * 1 - do not load the NIs from the mod params
1863 * Reverse logic to ensure that other calls to LNetNIInit
1866 bool ln_nis_from_mod_params;
1869 * completion for the monitor thread. The monitor thread takes care of
1870 * checking routes, timedout messages and resending messages.
1872 struct completion ln_mt_wait_complete;
1874 /* per-cpt resend queues */
1875 struct list_head **ln_mt_resendqs;
1876 /* local NIs to recover */
1877 struct list_head ln_mt_localNIRecovq;
1878 /* local NIs to recover */
1879 struct list_head ln_mt_peerNIRecovq;
1881 * An array of queues for GET/PUT waiting for REPLY/ACK respectively.
1882 * There are CPT number of queues. Since response trackers will be
1883 * added on the fast path we can't afford to grab the exclusive
1884 * net lock to protect these queues. The CPT will be calculated
1885 * based on the mdh cookie.
1887 struct list_head **ln_mt_rstq;
1889 * A response tracker becomes a zombie when the associated MD is queued
1890 * for unlink before the response tracker is detached from the MD. An
1891 * entry on a zombie list can be freed when either the remaining
1892 * operations on the MD complete or when LNet has shut down.
1894 struct list_head **ln_mt_zombie_rstqs;
1895 /* recovery handler */
1896 lnet_handler_t ln_mt_handler;
1899 * Completed when the discovery and monitor threads can enter their
1902 struct completion ln_started;
1904 struct list_head ln_udsp_list;
1906 /* Number of messages that have exceeded their message deadline */
1907 atomic_t ln_late_msg_count;
1908 /* Total amount of time past their deadline for all late ^ messages */
1909 atomic64_t ln_late_msg_nsecs;
1911 /* for LNDs to signal that ping buffer needs updating */
1912 atomic_t ln_update_ping_buf;
1914 /* workqueue for serving lnd ping buffer update requests */
1915 struct workqueue_struct *ln_pb_update_wq;
1916 struct work_struct ln_pb_update_work;
1918 atomic_t ln_pb_update_ready;
1921 struct genl_filter_list {
1922 struct list_head lp_list;
1927 static const struct nla_policy scalar_attr_policy[LN_SCALAR_MAX + 1] = {
1928 [LN_SCALAR_ATTR_LIST] = { .type = NLA_NESTED },
1929 [LN_SCALAR_ATTR_LIST_SIZE] = { .type = NLA_U16 },
1930 [LN_SCALAR_ATTR_INDEX] = { .type = NLA_U16 },
1931 [LN_SCALAR_ATTR_NLA_TYPE] = { .type = NLA_U16 },
1932 [LN_SCALAR_ATTR_VALUE] = { .type = NLA_STRING },
1933 [LN_SCALAR_ATTR_KEY_FORMAT] = { .type = NLA_U16 },
1936 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
1937 const struct genl_family *family, int flags,
1938 u8 cmd, const struct ln_key_list *data[]);
1940 /* Special workaround for pre-4.19 kernels to send error messages
1941 * from dumpit routines. Newer kernels will send message with
1942 * NL_SET_ERR_MSG information by default if NETLINK_EXT_ACK is set.
1944 static inline int lnet_nl_send_error(struct sk_buff *msg, int portid, int seq,
1947 #ifndef HAVE_NL_DUMP_WITH_EXT_ACK
1948 struct nlmsghdr *nlh;
1953 nlh = nlmsg_put(msg, portid, seq, NLMSG_ERROR, sizeof(error), 0);
1956 #ifdef HAVE_NL_PARSE_WITH_EXT_ACK
1957 netlink_ack(msg, nlh, error, NULL);
1959 netlink_ack(msg, nlh, error);
1961 return nlmsg_len(nlh);