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) 2007, 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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
35 #define DEBUG_SUBSYSTEM S_LNET
37 #include <linux/sched.h>
38 #ifdef HAVE_SCHED_HEADERS
39 #include <linux/sched/signal.h>
41 #include <linux/uaccess.h>
43 #include <lnet/lib-lnet.h>
44 #include <uapi/linux/lnet/lnet-dlc.h>
46 /* Value indicating that recovery needs to re-check a peer immediately. */
47 #define LNET_REDISCOVER_PEER (1)
49 static int lnet_peer_queue_for_discovery(struct lnet_peer *lp);
52 lnet_peer_remove_from_remote_list(struct lnet_peer_ni *lpni)
54 if (!list_empty(&lpni->lpni_on_remote_peer_ni_list)) {
55 list_del_init(&lpni->lpni_on_remote_peer_ni_list);
56 lnet_peer_ni_decref_locked(lpni);
61 lnet_peer_net_added(struct lnet_net *net)
63 struct lnet_peer_ni *lpni, *tmp;
65 list_for_each_entry_safe(lpni, tmp, &the_lnet.ln_remote_peer_ni_list,
66 lpni_on_remote_peer_ni_list) {
68 if (LNET_NIDNET(lpni->lpni_nid) == net->net_id) {
71 spin_lock(&lpni->lpni_lock);
72 lpni->lpni_txcredits =
73 lpni->lpni_net->net_tunables.lct_peer_tx_credits;
74 lpni->lpni_mintxcredits = lpni->lpni_txcredits;
75 lpni->lpni_rtrcredits =
76 lnet_peer_buffer_credits(lpni->lpni_net);
77 lpni->lpni_minrtrcredits = lpni->lpni_rtrcredits;
78 spin_unlock(&lpni->lpni_lock);
80 lnet_peer_remove_from_remote_list(lpni);
86 lnet_peer_tables_destroy(void)
88 struct lnet_peer_table *ptable;
89 struct list_head *hash;
93 if (!the_lnet.ln_peer_tables)
96 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
97 hash = ptable->pt_hash;
98 if (!hash) /* not intialized */
101 LASSERT(list_empty(&ptable->pt_zombie_list));
103 ptable->pt_hash = NULL;
104 for (j = 0; j < LNET_PEER_HASH_SIZE; j++)
105 LASSERT(list_empty(&hash[j]));
107 LIBCFS_FREE(hash, LNET_PEER_HASH_SIZE * sizeof(*hash));
110 cfs_percpt_free(the_lnet.ln_peer_tables);
111 the_lnet.ln_peer_tables = NULL;
115 lnet_peer_tables_create(void)
117 struct lnet_peer_table *ptable;
118 struct list_head *hash;
122 the_lnet.ln_peer_tables = cfs_percpt_alloc(lnet_cpt_table(),
124 if (the_lnet.ln_peer_tables == NULL) {
125 CERROR("Failed to allocate cpu-partition peer tables\n");
129 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
130 LIBCFS_CPT_ALLOC(hash, lnet_cpt_table(), i,
131 LNET_PEER_HASH_SIZE * sizeof(*hash));
133 CERROR("Failed to create peer hash table\n");
134 lnet_peer_tables_destroy();
138 spin_lock_init(&ptable->pt_zombie_lock);
139 INIT_LIST_HEAD(&ptable->pt_zombie_list);
141 INIT_LIST_HEAD(&ptable->pt_peer_list);
143 for (j = 0; j < LNET_PEER_HASH_SIZE; j++)
144 INIT_LIST_HEAD(&hash[j]);
145 ptable->pt_hash = hash; /* sign of initialization */
151 static struct lnet_peer_ni *
152 lnet_peer_ni_alloc(lnet_nid_t nid)
154 struct lnet_peer_ni *lpni;
155 struct lnet_net *net;
158 cpt = lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
160 LIBCFS_CPT_ALLOC(lpni, lnet_cpt_table(), cpt, sizeof(*lpni));
164 INIT_LIST_HEAD(&lpni->lpni_txq);
165 INIT_LIST_HEAD(&lpni->lpni_hashlist);
166 INIT_LIST_HEAD(&lpni->lpni_peer_nis);
167 INIT_LIST_HEAD(&lpni->lpni_recovery);
168 INIT_LIST_HEAD(&lpni->lpni_on_remote_peer_ni_list);
169 LNetInvalidateMDHandle(&lpni->lpni_recovery_ping_mdh);
171 spin_lock_init(&lpni->lpni_lock);
173 if (lnet_peers_start_down())
174 lpni->lpni_ns_status = LNET_NI_STATUS_DOWN;
176 lpni->lpni_ns_status = LNET_NI_STATUS_UP;
177 lpni->lpni_ping_feats = LNET_PING_FEAT_INVAL;
178 lpni->lpni_nid = nid;
179 lpni->lpni_cpt = cpt;
180 atomic_set(&lpni->lpni_healthv, LNET_MAX_HEALTH_VALUE);
182 net = lnet_get_net_locked(LNET_NIDNET(nid));
183 lpni->lpni_net = net;
185 lpni->lpni_txcredits = net->net_tunables.lct_peer_tx_credits;
186 lpni->lpni_mintxcredits = lpni->lpni_txcredits;
187 lpni->lpni_rtrcredits = lnet_peer_buffer_credits(net);
188 lpni->lpni_minrtrcredits = lpni->lpni_rtrcredits;
191 * This peer_ni is not on a local network, so we
192 * cannot add the credits here. In case the net is
193 * added later, add the peer_ni to the remote peer ni
194 * list so it can be easily found and revisited.
196 /* FIXME: per-net implementation instead? */
197 atomic_inc(&lpni->lpni_refcount);
198 list_add_tail(&lpni->lpni_on_remote_peer_ni_list,
199 &the_lnet.ln_remote_peer_ni_list);
202 CDEBUG(D_NET, "%p nid %s\n", lpni, libcfs_nid2str(lpni->lpni_nid));
207 static struct lnet_peer_net *
208 lnet_peer_net_alloc(__u32 net_id)
210 struct lnet_peer_net *lpn;
212 LIBCFS_CPT_ALLOC(lpn, lnet_cpt_table(), CFS_CPT_ANY, sizeof(*lpn));
216 INIT_LIST_HEAD(&lpn->lpn_peer_nets);
217 INIT_LIST_HEAD(&lpn->lpn_peer_nis);
218 lpn->lpn_net_id = net_id;
220 CDEBUG(D_NET, "%p net %s\n", lpn, libcfs_net2str(lpn->lpn_net_id));
226 lnet_destroy_peer_net_locked(struct lnet_peer_net *lpn)
228 struct lnet_peer *lp;
230 CDEBUG(D_NET, "%p net %s\n", lpn, libcfs_net2str(lpn->lpn_net_id));
232 LASSERT(atomic_read(&lpn->lpn_refcount) == 0);
233 LASSERT(list_empty(&lpn->lpn_peer_nis));
234 LASSERT(list_empty(&lpn->lpn_peer_nets));
236 lpn->lpn_peer = NULL;
237 LIBCFS_FREE(lpn, sizeof(*lpn));
239 lnet_peer_decref_locked(lp);
242 static struct lnet_peer *
243 lnet_peer_alloc(lnet_nid_t nid)
245 struct lnet_peer *lp;
247 LIBCFS_CPT_ALLOC(lp, lnet_cpt_table(), CFS_CPT_ANY, sizeof(*lp));
251 INIT_LIST_HEAD(&lp->lp_rtrq);
252 INIT_LIST_HEAD(&lp->lp_routes);
253 INIT_LIST_HEAD(&lp->lp_peer_list);
254 INIT_LIST_HEAD(&lp->lp_peer_nets);
255 INIT_LIST_HEAD(&lp->lp_dc_list);
256 INIT_LIST_HEAD(&lp->lp_dc_pendq);
257 INIT_LIST_HEAD(&lp->lp_rtr_list);
258 init_waitqueue_head(&lp->lp_dc_waitq);
259 spin_lock_init(&lp->lp_lock);
260 lp->lp_primary_nid = nid;
262 * Turn off discovery for loopback peer. If you're creating a peer
263 * for the loopback interface then that was initiated when we
264 * attempted to send a message over the loopback. There is no need
265 * to ever use a different interface when sending messages to
268 if (LNET_NETTYP(LNET_NIDNET(nid)) == LOLND)
269 lp->lp_state = LNET_PEER_NO_DISCOVERY;
270 lp->lp_cpt = lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
272 CDEBUG(D_NET, "%p nid %s\n", lp, libcfs_nid2str(lp->lp_primary_nid));
278 lnet_destroy_peer_locked(struct lnet_peer *lp)
280 CDEBUG(D_NET, "%p nid %s\n", lp, libcfs_nid2str(lp->lp_primary_nid));
282 LASSERT(atomic_read(&lp->lp_refcount) == 0);
283 LASSERT(lp->lp_rtr_refcount == 0);
284 LASSERT(list_empty(&lp->lp_peer_nets));
285 LASSERT(list_empty(&lp->lp_peer_list));
286 LASSERT(list_empty(&lp->lp_dc_list));
289 lnet_ping_buffer_decref(lp->lp_data);
292 * if there are messages still on the pending queue, then make
293 * sure to queue them on the ln_msg_resend list so they can be
294 * resent at a later point if the discovery thread is still
296 * If the discovery thread has stopped, then the wakeup will be a
297 * no-op, and it is expected the lnet_shutdown_lndnets() will
298 * eventually be called, which will traverse this list and
299 * finalize the messages on the list.
300 * We can not resend them now because we're holding the cpt lock.
301 * Releasing the lock can cause an inconsistent state
303 spin_lock(&the_lnet.ln_msg_resend_lock);
304 spin_lock(&lp->lp_lock);
305 list_splice(&lp->lp_dc_pendq, &the_lnet.ln_msg_resend);
306 spin_unlock(&lp->lp_lock);
307 spin_unlock(&the_lnet.ln_msg_resend_lock);
308 wake_up(&the_lnet.ln_dc_waitq);
310 LIBCFS_FREE(lp, sizeof(*lp));
314 * Detach a peer_ni from its peer_net. If this was the last peer_ni on
315 * that peer_net, detach the peer_net from the peer.
317 * Call with lnet_net_lock/EX held
320 lnet_peer_detach_peer_ni_locked(struct lnet_peer_ni *lpni)
322 struct lnet_peer_table *ptable;
323 struct lnet_peer_net *lpn;
324 struct lnet_peer *lp;
327 * Belts and suspenders: gracefully handle teardown of a
328 * partially connected peer_ni.
330 lpn = lpni->lpni_peer_net;
332 list_del_init(&lpni->lpni_peer_nis);
334 * If there are no lpni's left, we detach lpn from
335 * lp_peer_nets, so it cannot be found anymore.
337 if (list_empty(&lpn->lpn_peer_nis))
338 list_del_init(&lpn->lpn_peer_nets);
340 /* Update peer NID count. */
345 * If there are no more peer nets, make the peer unfindable
346 * via the peer_tables.
348 * Otherwise, if the peer is DISCOVERED, tell discovery to
349 * take another look at it. This is a no-op if discovery for
350 * this peer did the detaching.
352 if (list_empty(&lp->lp_peer_nets)) {
353 list_del_init(&lp->lp_peer_list);
354 ptable = the_lnet.ln_peer_tables[lp->lp_cpt];
356 } else if (the_lnet.ln_dc_state != LNET_DC_STATE_RUNNING) {
357 /* Discovery isn't running, nothing to do here. */
358 } else if (lp->lp_state & LNET_PEER_DISCOVERED) {
359 lnet_peer_queue_for_discovery(lp);
360 wake_up(&the_lnet.ln_dc_waitq);
362 CDEBUG(D_NET, "peer %s NID %s\n",
363 libcfs_nid2str(lp->lp_primary_nid),
364 libcfs_nid2str(lpni->lpni_nid));
367 /* called with lnet_net_lock LNET_LOCK_EX held */
369 lnet_peer_ni_del_locked(struct lnet_peer_ni *lpni, bool force)
371 struct lnet_peer_table *ptable = NULL;
373 /* don't remove a peer_ni if it's also a gateway */
374 if (lnet_isrouter(lpni) && !force) {
375 CERROR("Peer NI %s is a gateway. Can not delete it\n",
376 libcfs_nid2str(lpni->lpni_nid));
380 lnet_peer_remove_from_remote_list(lpni);
382 /* remove peer ni from the hash list. */
383 list_del_init(&lpni->lpni_hashlist);
386 * indicate the peer is being deleted so the monitor thread can
387 * remove it from the recovery queue.
389 spin_lock(&lpni->lpni_lock);
390 lpni->lpni_state |= LNET_PEER_NI_DELETING;
391 spin_unlock(&lpni->lpni_lock);
393 /* decrement the ref count on the peer table */
394 ptable = the_lnet.ln_peer_tables[lpni->lpni_cpt];
395 LASSERT(ptable->pt_number > 0);
399 * The peer_ni can no longer be found with a lookup. But there
400 * can be current users, so keep track of it on the zombie
401 * list until the reference count has gone to zero.
403 * The last reference may be lost in a place where the
404 * lnet_net_lock locks only a single cpt, and that cpt may not
405 * be lpni->lpni_cpt. So the zombie list of lnet_peer_table
408 spin_lock(&ptable->pt_zombie_lock);
409 list_add(&lpni->lpni_hashlist, &ptable->pt_zombie_list);
410 ptable->pt_zombies++;
411 spin_unlock(&ptable->pt_zombie_lock);
413 /* no need to keep this peer_ni on the hierarchy anymore */
414 lnet_peer_detach_peer_ni_locked(lpni);
416 /* remove hashlist reference on peer_ni */
417 lnet_peer_ni_decref_locked(lpni);
422 void lnet_peer_uninit(void)
424 struct lnet_peer_ni *lpni, *tmp;
426 lnet_net_lock(LNET_LOCK_EX);
428 /* remove all peer_nis from the remote peer and the hash list */
429 list_for_each_entry_safe(lpni, tmp, &the_lnet.ln_remote_peer_ni_list,
430 lpni_on_remote_peer_ni_list)
431 lnet_peer_ni_del_locked(lpni, false);
433 lnet_peer_tables_destroy();
435 lnet_net_unlock(LNET_LOCK_EX);
439 lnet_peer_del_locked(struct lnet_peer *peer)
441 struct lnet_peer_ni *lpni = NULL, *lpni2;
444 CDEBUG(D_NET, "peer %s\n", libcfs_nid2str(peer->lp_primary_nid));
446 lpni = lnet_get_next_peer_ni_locked(peer, NULL, lpni);
447 while (lpni != NULL) {
448 lpni2 = lnet_get_next_peer_ni_locked(peer, NULL, lpni);
449 rc = lnet_peer_ni_del_locked(lpni, false);
459 lnet_peer_del(struct lnet_peer *peer)
461 lnet_net_lock(LNET_LOCK_EX);
462 lnet_peer_del_locked(peer);
463 lnet_net_unlock(LNET_LOCK_EX);
469 * Delete a NID from a peer. Call with ln_api_mutex held.
472 * -EPERM: Non-DLC deletion from DLC-configured peer.
473 * -ENOENT: No lnet_peer_ni corresponding to the nid.
474 * -ECHILD: The lnet_peer_ni isn't connected to the peer.
475 * -EBUSY: The lnet_peer_ni is the primary, and not the only peer_ni.
478 lnet_peer_del_nid(struct lnet_peer *lp, lnet_nid_t nid, unsigned flags)
480 struct lnet_peer_ni *lpni;
481 lnet_nid_t primary_nid = lp->lp_primary_nid;
483 bool force = (flags & LNET_PEER_RTR_NI_FORCE_DEL) ? true : false;
485 if (!(flags & LNET_PEER_CONFIGURED)) {
486 if (lp->lp_state & LNET_PEER_CONFIGURED) {
491 lpni = lnet_find_peer_ni_locked(nid);
496 lnet_peer_ni_decref_locked(lpni);
497 if (lp != lpni->lpni_peer_net->lpn_peer) {
503 * This function only allows deletion of the primary NID if it
506 if (nid == lp->lp_primary_nid && lp->lp_nnis != 1 && !force) {
511 lnet_net_lock(LNET_LOCK_EX);
513 if (nid == lp->lp_primary_nid && lp->lp_nnis != 1 && force) {
514 struct lnet_peer_ni *lpni2;
515 /* assign the next peer_ni to be the primary */
516 lpni2 = lnet_get_next_peer_ni_locked(lp, NULL, lpni);
518 lp->lp_primary_nid = lpni->lpni_nid;
520 rc = lnet_peer_ni_del_locked(lpni, force);
522 lnet_net_unlock(LNET_LOCK_EX);
525 CDEBUG(D_NET, "peer %s NID %s flags %#x: %d\n",
526 libcfs_nid2str(primary_nid), libcfs_nid2str(nid), flags, rc);
532 lnet_peer_table_cleanup_locked(struct lnet_net *net,
533 struct lnet_peer_table *ptable)
536 struct lnet_peer_ni *next;
537 struct lnet_peer_ni *lpni;
538 struct lnet_peer *peer;
540 for (i = 0; i < LNET_PEER_HASH_SIZE; i++) {
541 list_for_each_entry_safe(lpni, next, &ptable->pt_hash[i],
543 if (net != NULL && net != lpni->lpni_net)
546 peer = lpni->lpni_peer_net->lpn_peer;
547 if (peer->lp_primary_nid != lpni->lpni_nid) {
548 lnet_peer_ni_del_locked(lpni, false);
552 * Removing the primary NID implies removing
553 * the entire peer. Advance next beyond any
554 * peer_ni that belongs to the same peer.
556 list_for_each_entry_from(next, &ptable->pt_hash[i],
558 if (next->lpni_peer_net->lpn_peer != peer)
561 lnet_peer_del_locked(peer);
567 lnet_peer_ni_finalize_wait(struct lnet_peer_table *ptable)
571 spin_lock(&ptable->pt_zombie_lock);
572 while (ptable->pt_zombies) {
573 spin_unlock(&ptable->pt_zombie_lock);
575 if (is_power_of_2(i)) {
577 "Waiting for %d zombies on peer table\n",
580 set_current_state(TASK_UNINTERRUPTIBLE);
581 schedule_timeout(cfs_time_seconds(1) >> 1);
582 spin_lock(&ptable->pt_zombie_lock);
584 spin_unlock(&ptable->pt_zombie_lock);
588 lnet_peer_table_del_rtrs_locked(struct lnet_net *net,
589 struct lnet_peer_table *ptable)
591 struct lnet_peer_ni *lp;
592 struct lnet_peer_ni *tmp;
596 for (i = 0; i < LNET_PEER_HASH_SIZE; i++) {
597 list_for_each_entry_safe(lp, tmp, &ptable->pt_hash[i],
599 if (net != lp->lpni_net)
602 if (!lnet_isrouter(lp))
605 gw_nid = lp->lpni_peer_net->lpn_peer->lp_primary_nid;
607 lnet_net_unlock(LNET_LOCK_EX);
608 lnet_del_route(LNET_NIDNET(LNET_NID_ANY), gw_nid);
609 lnet_net_lock(LNET_LOCK_EX);
615 lnet_peer_tables_cleanup(struct lnet_net *net)
618 struct lnet_peer_table *ptable;
620 LASSERT(the_lnet.ln_state != LNET_STATE_SHUTDOWN || net != NULL);
621 /* If just deleting the peers for a NI, get rid of any routes these
622 * peers are gateways for. */
623 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
624 lnet_net_lock(LNET_LOCK_EX);
625 lnet_peer_table_del_rtrs_locked(net, ptable);
626 lnet_net_unlock(LNET_LOCK_EX);
629 /* Start the cleanup process */
630 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
631 lnet_net_lock(LNET_LOCK_EX);
632 lnet_peer_table_cleanup_locked(net, ptable);
633 lnet_net_unlock(LNET_LOCK_EX);
636 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables)
637 lnet_peer_ni_finalize_wait(ptable);
640 static struct lnet_peer_ni *
641 lnet_get_peer_ni_locked(struct lnet_peer_table *ptable, lnet_nid_t nid)
643 struct list_head *peers;
644 struct lnet_peer_ni *lp;
646 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
648 peers = &ptable->pt_hash[lnet_nid2peerhash(nid)];
649 list_for_each_entry(lp, peers, lpni_hashlist) {
650 if (lp->lpni_nid == nid) {
651 lnet_peer_ni_addref_locked(lp);
659 struct lnet_peer_ni *
660 lnet_find_peer_ni_locked(lnet_nid_t nid)
662 struct lnet_peer_ni *lpni;
663 struct lnet_peer_table *ptable;
666 cpt = lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
668 ptable = the_lnet.ln_peer_tables[cpt];
669 lpni = lnet_get_peer_ni_locked(ptable, nid);
674 struct lnet_peer_ni *
675 lnet_peer_get_ni_locked(struct lnet_peer *lp, lnet_nid_t nid)
677 struct lnet_peer_net *lpn;
678 struct lnet_peer_ni *lpni;
680 lpn = lnet_peer_get_net_locked(lp, LNET_NIDNET(nid));
684 list_for_each_entry(lpni, &lpn->lpn_peer_nis, lpni_peer_nis) {
685 if (lpni->lpni_nid == nid)
693 lnet_find_peer(lnet_nid_t nid)
695 struct lnet_peer_ni *lpni;
696 struct lnet_peer *lp = NULL;
699 cpt = lnet_net_lock_current();
700 lpni = lnet_find_peer_ni_locked(nid);
702 lp = lpni->lpni_peer_net->lpn_peer;
703 lnet_peer_addref_locked(lp);
704 lnet_peer_ni_decref_locked(lpni);
706 lnet_net_unlock(cpt);
711 struct lnet_peer_ni *
712 lnet_get_next_peer_ni_locked(struct lnet_peer *peer,
713 struct lnet_peer_net *peer_net,
714 struct lnet_peer_ni *prev)
716 struct lnet_peer_ni *lpni;
717 struct lnet_peer_net *net = peer_net;
721 if (list_empty(&peer->lp_peer_nets))
724 net = list_entry(peer->lp_peer_nets.next,
725 struct lnet_peer_net,
728 lpni = list_entry(net->lpn_peer_nis.next, struct lnet_peer_ni,
734 if (prev->lpni_peer_nis.next == &prev->lpni_peer_net->lpn_peer_nis) {
736 * if you reached the end of the peer ni list and the peer
737 * net is specified then there are no more peer nis in that
744 * we reached the end of this net ni list. move to the
747 if (prev->lpni_peer_net->lpn_peer_nets.next ==
749 /* no more nets and no more NIs. */
752 /* get the next net */
753 net = list_entry(prev->lpni_peer_net->lpn_peer_nets.next,
754 struct lnet_peer_net,
756 /* get the ni on it */
757 lpni = list_entry(net->lpn_peer_nis.next, struct lnet_peer_ni,
763 /* there are more nis left */
764 lpni = list_entry(prev->lpni_peer_nis.next,
765 struct lnet_peer_ni, lpni_peer_nis);
770 /* Call with the ln_api_mutex held */
771 int lnet_get_peer_list(u32 *countp, u32 *sizep, struct lnet_process_id __user *ids)
773 struct lnet_process_id id;
774 struct lnet_peer_table *ptable;
775 struct lnet_peer *lp;
784 if (the_lnet.ln_state != LNET_STATE_RUNNING)
787 lncpt = cfs_percpt_number(the_lnet.ln_peer_tables);
790 * Count the number of peers, and return E2BIG if the buffer
791 * is too small. We'll also return the desired size.
794 for (cpt = 0; cpt < lncpt; cpt++) {
795 ptable = the_lnet.ln_peer_tables[cpt];
796 count += ptable->pt_peers;
798 size = count * sizeof(*ids);
803 * Walk the peer lists and copy out the primary nids.
804 * This is safe because the peer lists are only modified
805 * while the ln_api_mutex is held. So we don't need to
806 * hold the lnet_net_lock as well, and can therefore
807 * directly call copy_to_user().
810 memset(&id, 0, sizeof(id));
811 id.pid = LNET_PID_LUSTRE;
813 for (cpt = 0; cpt < lncpt; cpt++) {
814 ptable = the_lnet.ln_peer_tables[cpt];
815 list_for_each_entry(lp, &ptable->pt_peer_list, lp_peer_list) {
818 id.nid = lp->lp_primary_nid;
819 if (copy_to_user(&ids[i], &id, sizeof(id)))
832 * Start pushes to peers that need to be updated for a configuration
833 * change on this node.
836 lnet_push_update_to_peers(int force)
838 struct lnet_peer_table *ptable;
839 struct lnet_peer *lp;
843 lnet_net_lock(LNET_LOCK_EX);
844 lncpt = cfs_percpt_number(the_lnet.ln_peer_tables);
845 for (cpt = 0; cpt < lncpt; cpt++) {
846 ptable = the_lnet.ln_peer_tables[cpt];
847 list_for_each_entry(lp, &ptable->pt_peer_list, lp_peer_list) {
849 spin_lock(&lp->lp_lock);
850 if (lp->lp_state & LNET_PEER_MULTI_RAIL)
851 lp->lp_state |= LNET_PEER_FORCE_PUSH;
852 spin_unlock(&lp->lp_lock);
854 if (lnet_peer_needs_push(lp))
855 lnet_peer_queue_for_discovery(lp);
858 lnet_net_unlock(LNET_LOCK_EX);
859 wake_up(&the_lnet.ln_dc_waitq);
863 * Test whether a ni is a preferred ni for this peer_ni, e.g, whether
864 * this is a preferred point-to-point path. Call with lnet_net_lock in
868 lnet_peer_is_pref_nid_locked(struct lnet_peer_ni *lpni, lnet_nid_t nid)
872 if (lpni->lpni_pref_nnids == 0)
874 if (lpni->lpni_pref_nnids == 1)
875 return lpni->lpni_pref.nid == nid;
876 for (i = 0; i < lpni->lpni_pref_nnids; i++) {
877 if (lpni->lpni_pref.nids[i] == nid)
884 * Set a single ni as preferred, provided no preferred ni is already
885 * defined. Only to be used for non-multi-rail peer_ni.
888 lnet_peer_ni_set_non_mr_pref_nid(struct lnet_peer_ni *lpni, lnet_nid_t nid)
892 spin_lock(&lpni->lpni_lock);
893 if (nid == LNET_NID_ANY) {
895 } else if (lpni->lpni_pref_nnids > 0) {
897 } else if (lpni->lpni_pref_nnids == 0) {
898 lpni->lpni_pref.nid = nid;
899 lpni->lpni_pref_nnids = 1;
900 lpni->lpni_state |= LNET_PEER_NI_NON_MR_PREF;
902 spin_unlock(&lpni->lpni_lock);
904 CDEBUG(D_NET, "peer %s nid %s: %d\n",
905 libcfs_nid2str(lpni->lpni_nid), libcfs_nid2str(nid), rc);
910 * Clear the preferred NID from a non-multi-rail peer_ni, provided
911 * this preference was set by lnet_peer_ni_set_non_mr_pref_nid().
914 lnet_peer_ni_clr_non_mr_pref_nid(struct lnet_peer_ni *lpni)
918 spin_lock(&lpni->lpni_lock);
919 if (lpni->lpni_state & LNET_PEER_NI_NON_MR_PREF) {
920 lpni->lpni_pref_nnids = 0;
921 lpni->lpni_state &= ~LNET_PEER_NI_NON_MR_PREF;
922 } else if (lpni->lpni_pref_nnids == 0) {
927 spin_unlock(&lpni->lpni_lock);
929 CDEBUG(D_NET, "peer %s: %d\n",
930 libcfs_nid2str(lpni->lpni_nid), rc);
935 * Clear the preferred NIDs from a non-multi-rail peer.
938 lnet_peer_clr_non_mr_pref_nids(struct lnet_peer *lp)
940 struct lnet_peer_ni *lpni = NULL;
942 while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL)
943 lnet_peer_ni_clr_non_mr_pref_nid(lpni);
947 lnet_peer_add_pref_nid(struct lnet_peer_ni *lpni, lnet_nid_t nid)
949 lnet_nid_t *nids = NULL;
950 lnet_nid_t *oldnids = NULL;
951 struct lnet_peer *lp = lpni->lpni_peer_net->lpn_peer;
956 if (nid == LNET_NID_ANY) {
961 if (lpni->lpni_pref_nnids == 1 && lpni->lpni_pref.nid == nid) {
966 /* A non-MR node may have only one preferred NI per peer_ni */
967 if (lpni->lpni_pref_nnids > 0) {
968 if (!(lp->lp_state & LNET_PEER_MULTI_RAIL)) {
974 if (lpni->lpni_pref_nnids != 0) {
975 size = sizeof(*nids) * (lpni->lpni_pref_nnids + 1);
976 LIBCFS_CPT_ALLOC(nids, lnet_cpt_table(), lpni->lpni_cpt, size);
981 for (i = 0; i < lpni->lpni_pref_nnids; i++) {
982 if (lpni->lpni_pref.nids[i] == nid) {
983 LIBCFS_FREE(nids, size);
987 nids[i] = lpni->lpni_pref.nids[i];
992 lnet_net_lock(LNET_LOCK_EX);
993 spin_lock(&lpni->lpni_lock);
994 if (lpni->lpni_pref_nnids == 0) {
995 lpni->lpni_pref.nid = nid;
997 oldnids = lpni->lpni_pref.nids;
998 lpni->lpni_pref.nids = nids;
1000 lpni->lpni_pref_nnids++;
1001 lpni->lpni_state &= ~LNET_PEER_NI_NON_MR_PREF;
1002 spin_unlock(&lpni->lpni_lock);
1003 lnet_net_unlock(LNET_LOCK_EX);
1006 size = sizeof(*nids) * (lpni->lpni_pref_nnids - 1);
1007 LIBCFS_FREE(oldnids, sizeof(*oldnids) * size);
1010 if (rc == -EEXIST && (lpni->lpni_state & LNET_PEER_NI_NON_MR_PREF)) {
1011 spin_lock(&lpni->lpni_lock);
1012 lpni->lpni_state &= ~LNET_PEER_NI_NON_MR_PREF;
1013 spin_unlock(&lpni->lpni_lock);
1015 CDEBUG(D_NET, "peer %s nid %s: %d\n",
1016 libcfs_nid2str(lp->lp_primary_nid), libcfs_nid2str(nid), rc);
1021 lnet_peer_del_pref_nid(struct lnet_peer_ni *lpni, lnet_nid_t nid)
1023 lnet_nid_t *nids = NULL;
1024 lnet_nid_t *oldnids = NULL;
1025 struct lnet_peer *lp = lpni->lpni_peer_net->lpn_peer;
1030 if (lpni->lpni_pref_nnids == 0) {
1035 if (lpni->lpni_pref_nnids == 1) {
1036 if (lpni->lpni_pref.nid != nid) {
1040 } else if (lpni->lpni_pref_nnids == 2) {
1041 if (lpni->lpni_pref.nids[0] != nid &&
1042 lpni->lpni_pref.nids[1] != nid) {
1047 size = sizeof(*nids) * (lpni->lpni_pref_nnids - 1);
1048 LIBCFS_CPT_ALLOC(nids, lnet_cpt_table(), lpni->lpni_cpt, size);
1053 for (i = 0, j = 0; i < lpni->lpni_pref_nnids; i++) {
1054 if (lpni->lpni_pref.nids[i] != nid)
1056 nids[j++] = lpni->lpni_pref.nids[i];
1058 /* Check if we actually removed a nid. */
1059 if (j == lpni->lpni_pref_nnids) {
1060 LIBCFS_FREE(nids, size);
1066 lnet_net_lock(LNET_LOCK_EX);
1067 spin_lock(&lpni->lpni_lock);
1068 if (lpni->lpni_pref_nnids == 1) {
1069 lpni->lpni_pref.nid = LNET_NID_ANY;
1070 } else if (lpni->lpni_pref_nnids == 2) {
1071 oldnids = lpni->lpni_pref.nids;
1072 if (oldnids[0] == nid)
1073 lpni->lpni_pref.nid = oldnids[1];
1075 lpni->lpni_pref.nid = oldnids[2];
1077 oldnids = lpni->lpni_pref.nids;
1078 lpni->lpni_pref.nids = nids;
1080 lpni->lpni_pref_nnids--;
1081 lpni->lpni_state &= ~LNET_PEER_NI_NON_MR_PREF;
1082 spin_unlock(&lpni->lpni_lock);
1083 lnet_net_unlock(LNET_LOCK_EX);
1086 size = sizeof(*nids) * (lpni->lpni_pref_nnids + 1);
1087 LIBCFS_FREE(oldnids, sizeof(*oldnids) * size);
1090 CDEBUG(D_NET, "peer %s nid %s: %d\n",
1091 libcfs_nid2str(lp->lp_primary_nid), libcfs_nid2str(nid), rc);
1096 lnet_peer_primary_nid_locked(lnet_nid_t nid)
1098 struct lnet_peer_ni *lpni;
1099 lnet_nid_t primary_nid = nid;
1101 lpni = lnet_find_peer_ni_locked(nid);
1103 primary_nid = lpni->lpni_peer_net->lpn_peer->lp_primary_nid;
1104 lnet_peer_ni_decref_locked(lpni);
1111 LNetPrimaryNID(lnet_nid_t nid)
1113 struct lnet_peer *lp;
1114 struct lnet_peer_ni *lpni;
1115 lnet_nid_t primary_nid = nid;
1119 cpt = lnet_net_lock_current();
1120 lpni = lnet_nid2peerni_locked(nid, LNET_NID_ANY, cpt);
1125 lp = lpni->lpni_peer_net->lpn_peer;
1126 while (!lnet_peer_is_uptodate(lp)) {
1127 rc = lnet_discover_peer_locked(lpni, cpt, true);
1130 lp = lpni->lpni_peer_net->lpn_peer;
1132 primary_nid = lp->lp_primary_nid;
1134 lnet_peer_ni_decref_locked(lpni);
1136 lnet_net_unlock(cpt);
1138 CDEBUG(D_NET, "NID %s primary NID %s rc %d\n", libcfs_nid2str(nid),
1139 libcfs_nid2str(primary_nid), rc);
1142 EXPORT_SYMBOL(LNetPrimaryNID);
1144 struct lnet_peer_net *
1145 lnet_peer_get_net_locked(struct lnet_peer *peer, __u32 net_id)
1147 struct lnet_peer_net *peer_net;
1148 list_for_each_entry(peer_net, &peer->lp_peer_nets, lpn_peer_nets) {
1149 if (peer_net->lpn_net_id == net_id)
1156 * Attach a peer_ni to a peer_net and peer. This function assumes
1157 * peer_ni is not already attached to the peer_net/peer. The peer_ni
1158 * may be attached to a different peer, in which case it will be
1159 * properly detached first. The whole operation is done atomically.
1161 * Always returns 0. This is the last function called from functions
1162 * that do return an int, so returning 0 here allows the compiler to
1166 lnet_peer_attach_peer_ni(struct lnet_peer *lp,
1167 struct lnet_peer_net *lpn,
1168 struct lnet_peer_ni *lpni,
1171 struct lnet_peer_table *ptable;
1173 /* Install the new peer_ni */
1174 lnet_net_lock(LNET_LOCK_EX);
1175 /* Add peer_ni to global peer table hash, if necessary. */
1176 if (list_empty(&lpni->lpni_hashlist)) {
1177 int hash = lnet_nid2peerhash(lpni->lpni_nid);
1179 ptable = the_lnet.ln_peer_tables[lpni->lpni_cpt];
1180 list_add_tail(&lpni->lpni_hashlist, &ptable->pt_hash[hash]);
1181 ptable->pt_version++;
1182 ptable->pt_number++;
1183 /* This is the 1st refcount on lpni. */
1184 atomic_inc(&lpni->lpni_refcount);
1187 /* Detach the peer_ni from an existing peer, if necessary. */
1188 if (lpni->lpni_peer_net) {
1189 LASSERT(lpni->lpni_peer_net != lpn);
1190 LASSERT(lpni->lpni_peer_net->lpn_peer != lp);
1191 lnet_peer_detach_peer_ni_locked(lpni);
1192 lnet_peer_net_decref_locked(lpni->lpni_peer_net);
1193 lpni->lpni_peer_net = NULL;
1196 /* Add peer_ni to peer_net */
1197 lpni->lpni_peer_net = lpn;
1198 list_add_tail(&lpni->lpni_peer_nis, &lpn->lpn_peer_nis);
1199 lnet_peer_net_addref_locked(lpn);
1201 /* Add peer_net to peer */
1202 if (!lpn->lpn_peer) {
1204 list_add_tail(&lpn->lpn_peer_nets, &lp->lp_peer_nets);
1205 lnet_peer_addref_locked(lp);
1208 /* Add peer to global peer list, if necessary */
1209 ptable = the_lnet.ln_peer_tables[lp->lp_cpt];
1210 if (list_empty(&lp->lp_peer_list)) {
1211 list_add_tail(&lp->lp_peer_list, &ptable->pt_peer_list);
1216 /* Update peer state */
1217 spin_lock(&lp->lp_lock);
1218 if (flags & LNET_PEER_CONFIGURED) {
1219 if (!(lp->lp_state & LNET_PEER_CONFIGURED))
1220 lp->lp_state |= LNET_PEER_CONFIGURED;
1222 if (flags & LNET_PEER_MULTI_RAIL) {
1223 if (!(lp->lp_state & LNET_PEER_MULTI_RAIL)) {
1224 lp->lp_state |= LNET_PEER_MULTI_RAIL;
1225 lnet_peer_clr_non_mr_pref_nids(lp);
1228 spin_unlock(&lp->lp_lock);
1231 lnet_net_unlock(LNET_LOCK_EX);
1233 CDEBUG(D_NET, "peer %s NID %s flags %#x\n",
1234 libcfs_nid2str(lp->lp_primary_nid),
1235 libcfs_nid2str(lpni->lpni_nid), flags);
1241 * Create a new peer, with nid as its primary nid.
1243 * Call with the lnet_api_mutex held.
1246 lnet_peer_add(lnet_nid_t nid, unsigned flags)
1248 struct lnet_peer *lp;
1249 struct lnet_peer_net *lpn;
1250 struct lnet_peer_ni *lpni;
1253 LASSERT(nid != LNET_NID_ANY);
1256 * No need for the lnet_net_lock here, because the
1257 * lnet_api_mutex is held.
1259 lpni = lnet_find_peer_ni_locked(nid);
1261 /* A peer with this NID already exists. */
1262 lp = lpni->lpni_peer_net->lpn_peer;
1263 lnet_peer_ni_decref_locked(lpni);
1265 * This is an error if the peer was configured and the
1266 * primary NID differs or an attempt is made to change
1267 * the Multi-Rail flag. Otherwise the assumption is
1268 * that an existing peer is being modified.
1270 if (lp->lp_state & LNET_PEER_CONFIGURED) {
1271 if (lp->lp_primary_nid != nid)
1273 else if ((lp->lp_state ^ flags) & LNET_PEER_MULTI_RAIL)
1277 /* Delete and recreate as a configured peer. */
1281 /* Create peer, peer_net, and peer_ni. */
1283 lp = lnet_peer_alloc(nid);
1286 lpn = lnet_peer_net_alloc(LNET_NIDNET(nid));
1289 lpni = lnet_peer_ni_alloc(nid);
1293 return lnet_peer_attach_peer_ni(lp, lpn, lpni, flags);
1296 LIBCFS_FREE(lpn, sizeof(*lpn));
1298 LIBCFS_FREE(lp, sizeof(*lp));
1300 CDEBUG(D_NET, "peer %s NID flags %#x: %d\n",
1301 libcfs_nid2str(nid), flags, rc);
1306 * Add a NID to a peer. Call with ln_api_mutex held.
1309 * -EPERM: Non-DLC addition to a DLC-configured peer.
1310 * -EEXIST: The NID was configured by DLC for a different peer.
1311 * -ENOMEM: Out of memory.
1312 * -ENOTUNIQ: Adding a second peer NID on a single network on a
1313 * non-multi-rail peer.
1316 lnet_peer_add_nid(struct lnet_peer *lp, lnet_nid_t nid, unsigned flags)
1318 struct lnet_peer_net *lpn;
1319 struct lnet_peer_ni *lpni;
1323 LASSERT(nid != LNET_NID_ANY);
1325 /* A configured peer can only be updated through configuration. */
1326 if (!(flags & LNET_PEER_CONFIGURED)) {
1327 if (lp->lp_state & LNET_PEER_CONFIGURED) {
1334 * The MULTI_RAIL flag can be set but not cleared, because
1335 * that would leave the peer struct in an invalid state.
1337 if (flags & LNET_PEER_MULTI_RAIL) {
1338 spin_lock(&lp->lp_lock);
1339 if (!(lp->lp_state & LNET_PEER_MULTI_RAIL)) {
1340 lp->lp_state |= LNET_PEER_MULTI_RAIL;
1341 lnet_peer_clr_non_mr_pref_nids(lp);
1343 spin_unlock(&lp->lp_lock);
1344 } else if (lp->lp_state & LNET_PEER_MULTI_RAIL) {
1349 lpni = lnet_find_peer_ni_locked(nid);
1352 * A peer_ni already exists. This is only a problem if
1353 * it is not connected to this peer and was configured
1356 lnet_peer_ni_decref_locked(lpni);
1357 if (lpni->lpni_peer_net->lpn_peer == lp)
1359 if (lnet_peer_ni_is_configured(lpni)) {
1363 /* If this is the primary NID, destroy the peer. */
1364 if (lnet_peer_ni_is_primary(lpni)) {
1365 struct lnet_peer *rtr_lp =
1366 lpni->lpni_peer_net->lpn_peer;
1367 int rtr_refcount = rtr_lp->lp_rtr_refcount;
1369 * if we're trying to delete a router it means
1370 * we're moving this peer NI to a new peer so must
1371 * transfer router properties to the new peer
1373 if (rtr_refcount > 0) {
1374 flags |= LNET_PEER_RTR_NI_FORCE_DEL;
1375 lnet_rtr_transfer_to_peer(rtr_lp, lp);
1377 lnet_peer_del(lpni->lpni_peer_net->lpn_peer);
1378 lpni = lnet_peer_ni_alloc(nid);
1385 lpni = lnet_peer_ni_alloc(nid);
1393 * Get the peer_net. Check that we're not adding a second
1394 * peer_ni on a peer_net of a non-multi-rail peer.
1396 lpn = lnet_peer_get_net_locked(lp, LNET_NIDNET(nid));
1398 lpn = lnet_peer_net_alloc(LNET_NIDNET(nid));
1403 } else if (!(lp->lp_state & LNET_PEER_MULTI_RAIL)) {
1408 return lnet_peer_attach_peer_ni(lp, lpn, lpni, flags);
1411 /* If the peer_ni was allocated above its peer_net pointer is NULL */
1412 if (!lpni->lpni_peer_net)
1413 LIBCFS_FREE(lpni, sizeof(*lpni));
1415 CDEBUG(D_NET, "peer %s NID %s flags %#x: %d\n",
1416 libcfs_nid2str(lp->lp_primary_nid), libcfs_nid2str(nid),
1422 * Update the primary NID of a peer, if possible.
1424 * Call with the lnet_api_mutex held.
1427 lnet_peer_set_primary_nid(struct lnet_peer *lp, lnet_nid_t nid, unsigned flags)
1429 lnet_nid_t old = lp->lp_primary_nid;
1432 if (lp->lp_primary_nid == nid)
1434 rc = lnet_peer_add_nid(lp, nid, flags);
1437 lp->lp_primary_nid = nid;
1439 CDEBUG(D_NET, "peer %s NID %s: %d\n",
1440 libcfs_nid2str(old), libcfs_nid2str(nid), rc);
1445 * lpni creation initiated due to traffic either sending or receiving.
1448 lnet_peer_ni_traffic_add(lnet_nid_t nid, lnet_nid_t pref)
1450 struct lnet_peer *lp;
1451 struct lnet_peer_net *lpn;
1452 struct lnet_peer_ni *lpni;
1456 if (nid == LNET_NID_ANY) {
1461 /* lnet_net_lock is not needed here because ln_api_lock is held */
1462 lpni = lnet_find_peer_ni_locked(nid);
1465 * We must have raced with another thread. Since we
1466 * know next to nothing about a peer_ni created by
1467 * traffic, we just assume everything is ok and
1470 lnet_peer_ni_decref_locked(lpni);
1474 /* Create peer, peer_net, and peer_ni. */
1476 lp = lnet_peer_alloc(nid);
1479 lpn = lnet_peer_net_alloc(LNET_NIDNET(nid));
1482 lpni = lnet_peer_ni_alloc(nid);
1485 if (pref != LNET_NID_ANY)
1486 lnet_peer_ni_set_non_mr_pref_nid(lpni, pref);
1488 return lnet_peer_attach_peer_ni(lp, lpn, lpni, flags);
1491 LIBCFS_FREE(lpn, sizeof(*lpn));
1493 LIBCFS_FREE(lp, sizeof(*lp));
1495 CDEBUG(D_NET, "peer %s: %d\n", libcfs_nid2str(nid), rc);
1500 * Implementation of IOC_LIBCFS_ADD_PEER_NI.
1502 * This API handles the following combinations:
1503 * Create a peer with its primary NI if only the prim_nid is provided
1504 * Add a NID to a peer identified by the prim_nid. The peer identified
1505 * by the prim_nid must already exist.
1506 * The peer being created may be non-MR.
1508 * The caller must hold ln_api_mutex. This prevents the peer from
1509 * being created/modified/deleted by a different thread.
1512 lnet_add_peer_ni(lnet_nid_t prim_nid, lnet_nid_t nid, bool mr)
1514 struct lnet_peer *lp = NULL;
1515 struct lnet_peer_ni *lpni;
1518 /* The prim_nid must always be specified */
1519 if (prim_nid == LNET_NID_ANY)
1522 flags = LNET_PEER_CONFIGURED;
1524 flags |= LNET_PEER_MULTI_RAIL;
1527 * If nid isn't specified, we must create a new peer with
1528 * prim_nid as its primary nid.
1530 if (nid == LNET_NID_ANY)
1531 return lnet_peer_add(prim_nid, flags);
1533 /* Look up the prim_nid, which must exist. */
1534 lpni = lnet_find_peer_ni_locked(prim_nid);
1537 lnet_peer_ni_decref_locked(lpni);
1538 lp = lpni->lpni_peer_net->lpn_peer;
1540 /* Peer must have been configured. */
1541 if (!(lp->lp_state & LNET_PEER_CONFIGURED)) {
1542 CDEBUG(D_NET, "peer %s was not configured\n",
1543 libcfs_nid2str(prim_nid));
1547 /* Primary NID must match */
1548 if (lp->lp_primary_nid != prim_nid) {
1549 CDEBUG(D_NET, "prim_nid %s is not primary for peer %s\n",
1550 libcfs_nid2str(prim_nid),
1551 libcfs_nid2str(lp->lp_primary_nid));
1555 /* Multi-Rail flag must match. */
1556 if ((lp->lp_state ^ flags) & LNET_PEER_MULTI_RAIL) {
1557 CDEBUG(D_NET, "multi-rail state mismatch for peer %s\n",
1558 libcfs_nid2str(prim_nid));
1562 return lnet_peer_add_nid(lp, nid, flags);
1566 * Implementation of IOC_LIBCFS_DEL_PEER_NI.
1568 * This API handles the following combinations:
1569 * Delete a NI from a peer if both prim_nid and nid are provided.
1570 * Delete a peer if only prim_nid is provided.
1571 * Delete a peer if its primary nid is provided.
1573 * The caller must hold ln_api_mutex. This prevents the peer from
1574 * being modified/deleted by a different thread.
1577 lnet_del_peer_ni(lnet_nid_t prim_nid, lnet_nid_t nid)
1579 struct lnet_peer *lp;
1580 struct lnet_peer_ni *lpni;
1583 if (prim_nid == LNET_NID_ANY)
1586 lpni = lnet_find_peer_ni_locked(prim_nid);
1589 lnet_peer_ni_decref_locked(lpni);
1590 lp = lpni->lpni_peer_net->lpn_peer;
1592 if (prim_nid != lp->lp_primary_nid) {
1593 CDEBUG(D_NET, "prim_nid %s is not primary for peer %s\n",
1594 libcfs_nid2str(prim_nid),
1595 libcfs_nid2str(lp->lp_primary_nid));
1599 lnet_net_lock(LNET_LOCK_EX);
1600 if (lp->lp_rtr_refcount > 0) {
1601 lnet_net_unlock(LNET_LOCK_EX);
1602 CERROR("%s is a router. Can not be deleted\n",
1603 libcfs_nid2str(prim_nid));
1606 lnet_net_unlock(LNET_LOCK_EX);
1608 if (nid == LNET_NID_ANY || nid == lp->lp_primary_nid)
1609 return lnet_peer_del(lp);
1611 flags = LNET_PEER_CONFIGURED;
1612 if (lp->lp_state & LNET_PEER_MULTI_RAIL)
1613 flags |= LNET_PEER_MULTI_RAIL;
1615 return lnet_peer_del_nid(lp, nid, flags);
1619 lnet_destroy_peer_ni_locked(struct lnet_peer_ni *lpni)
1621 struct lnet_peer_table *ptable;
1622 struct lnet_peer_net *lpn;
1624 CDEBUG(D_NET, "%p nid %s\n", lpni, libcfs_nid2str(lpni->lpni_nid));
1626 LASSERT(atomic_read(&lpni->lpni_refcount) == 0);
1627 LASSERT(list_empty(&lpni->lpni_txq));
1628 LASSERT(lpni->lpni_txqnob == 0);
1629 LASSERT(list_empty(&lpni->lpni_peer_nis));
1630 LASSERT(list_empty(&lpni->lpni_on_remote_peer_ni_list));
1632 lpn = lpni->lpni_peer_net;
1633 lpni->lpni_peer_net = NULL;
1634 lpni->lpni_net = NULL;
1636 /* remove the peer ni from the zombie list */
1637 ptable = the_lnet.ln_peer_tables[lpni->lpni_cpt];
1638 spin_lock(&ptable->pt_zombie_lock);
1639 list_del_init(&lpni->lpni_hashlist);
1640 ptable->pt_zombies--;
1641 spin_unlock(&ptable->pt_zombie_lock);
1643 if (lpni->lpni_pref_nnids > 1) {
1644 LIBCFS_FREE(lpni->lpni_pref.nids,
1645 sizeof(*lpni->lpni_pref.nids) * lpni->lpni_pref_nnids);
1647 LIBCFS_FREE(lpni, sizeof(*lpni));
1649 lnet_peer_net_decref_locked(lpn);
1652 struct lnet_peer_ni *
1653 lnet_nid2peerni_ex(lnet_nid_t nid, int cpt)
1655 struct lnet_peer_ni *lpni = NULL;
1658 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1659 return ERR_PTR(-ESHUTDOWN);
1662 * find if a peer_ni already exists.
1663 * If so then just return that.
1665 lpni = lnet_find_peer_ni_locked(nid);
1669 lnet_net_unlock(cpt);
1671 rc = lnet_peer_ni_traffic_add(nid, LNET_NID_ANY);
1674 goto out_net_relock;
1677 lpni = lnet_find_peer_ni_locked(nid);
1687 * Get a peer_ni for the given nid, create it if necessary. Takes a
1688 * hold on the peer_ni.
1690 struct lnet_peer_ni *
1691 lnet_nid2peerni_locked(lnet_nid_t nid, lnet_nid_t pref, int cpt)
1693 struct lnet_peer_ni *lpni = NULL;
1696 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1697 return ERR_PTR(-ESHUTDOWN);
1700 * find if a peer_ni already exists.
1701 * If so then just return that.
1703 lpni = lnet_find_peer_ni_locked(nid);
1709 * use the lnet_api_mutex to serialize the creation of the peer_ni
1710 * and the creation/deletion of the local ni/net. When a local ni is
1711 * created, if there exists a set of peer_nis on that network,
1712 * they need to be traversed and updated. When a local NI is
1713 * deleted, which could result in a network being deleted, then
1714 * all peer nis on that network need to be removed as well.
1716 * Creation through traffic should also be serialized with
1717 * creation through DLC.
1719 lnet_net_unlock(cpt);
1720 mutex_lock(&the_lnet.ln_api_mutex);
1722 * Shutdown is only set under the ln_api_lock, so a single
1723 * check here is sufficent.
1725 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
1726 lpni = ERR_PTR(-ESHUTDOWN);
1727 goto out_mutex_unlock;
1730 rc = lnet_peer_ni_traffic_add(nid, pref);
1733 goto out_mutex_unlock;
1736 lpni = lnet_find_peer_ni_locked(nid);
1740 mutex_unlock(&the_lnet.ln_api_mutex);
1743 /* Lock has been dropped, check again for shutdown. */
1744 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
1746 lnet_peer_ni_decref_locked(lpni);
1747 lpni = ERR_PTR(-ESHUTDOWN);
1758 lnet_peer_gw_discovery(struct lnet_peer *lp)
1762 spin_lock(&lp->lp_lock);
1763 if (lp->lp_state & LNET_PEER_RTR_DISCOVERY)
1765 spin_unlock(&lp->lp_lock);
1771 * Is a peer uptodate from the point of view of discovery?
1773 * If it is currently being processed, obviously not.
1774 * A forced Ping or Push is also handled by the discovery thread.
1776 * Otherwise look at whether the peer needs rediscovering.
1779 lnet_peer_is_uptodate(struct lnet_peer *lp)
1783 spin_lock(&lp->lp_lock);
1784 if (lp->lp_state & (LNET_PEER_DISCOVERING |
1785 LNET_PEER_FORCE_PING |
1786 LNET_PEER_FORCE_PUSH)) {
1788 } else if (lp->lp_state & LNET_PEER_NO_DISCOVERY) {
1790 } else if (lp->lp_state & LNET_PEER_REDISCOVER) {
1791 if (lnet_peer_discovery_disabled)
1795 } else if (lnet_peer_needs_push(lp)) {
1797 } else if (lp->lp_state & LNET_PEER_DISCOVERED) {
1798 if (lp->lp_state & LNET_PEER_NIDS_UPTODATE)
1805 spin_unlock(&lp->lp_lock);
1811 * Queue a peer for the attention of the discovery thread. Call with
1812 * lnet_net_lock/EX held. Returns 0 if the peer was queued, and
1813 * -EALREADY if the peer was already queued.
1815 static int lnet_peer_queue_for_discovery(struct lnet_peer *lp)
1819 spin_lock(&lp->lp_lock);
1820 if (!(lp->lp_state & LNET_PEER_DISCOVERING))
1821 lp->lp_state |= LNET_PEER_DISCOVERING;
1822 spin_unlock(&lp->lp_lock);
1823 if (list_empty(&lp->lp_dc_list)) {
1824 lnet_peer_addref_locked(lp);
1825 list_add_tail(&lp->lp_dc_list, &the_lnet.ln_dc_request);
1826 wake_up(&the_lnet.ln_dc_waitq);
1832 CDEBUG(D_NET, "Queue peer %s: %d\n",
1833 libcfs_nid2str(lp->lp_primary_nid), rc);
1839 * Discovery of a peer is complete. Wake all waiters on the peer.
1840 * Call with lnet_net_lock/EX held.
1842 static void lnet_peer_discovery_complete(struct lnet_peer *lp)
1844 struct lnet_msg *msg, *tmp;
1846 struct list_head pending_msgs;
1848 INIT_LIST_HEAD(&pending_msgs);
1850 CDEBUG(D_NET, "Discovery complete. Dequeue peer %s\n",
1851 libcfs_nid2str(lp->lp_primary_nid));
1853 list_del_init(&lp->lp_dc_list);
1854 spin_lock(&lp->lp_lock);
1855 list_splice_init(&lp->lp_dc_pendq, &pending_msgs);
1856 spin_unlock(&lp->lp_lock);
1857 wake_up_all(&lp->lp_dc_waitq);
1859 if (lp->lp_rtr_refcount > 0)
1860 lnet_router_discovery_complete(lp);
1862 lnet_net_unlock(LNET_LOCK_EX);
1864 /* iterate through all pending messages and send them again */
1865 list_for_each_entry_safe(msg, tmp, &pending_msgs, msg_list) {
1866 list_del_init(&msg->msg_list);
1867 if (lp->lp_dc_error) {
1868 lnet_finalize(msg, lp->lp_dc_error);
1872 CDEBUG(D_NET, "sending pending message %s to target %s\n",
1873 lnet_msgtyp2str(msg->msg_type),
1874 libcfs_id2str(msg->msg_target));
1875 rc = lnet_send(msg->msg_src_nid_param, msg,
1876 msg->msg_rtr_nid_param);
1878 CNETERR("Error sending %s to %s: %d\n",
1879 lnet_msgtyp2str(msg->msg_type),
1880 libcfs_id2str(msg->msg_target), rc);
1881 lnet_finalize(msg, rc);
1884 lnet_net_lock(LNET_LOCK_EX);
1885 lnet_peer_decref_locked(lp);
1889 * Handle inbound push.
1890 * Like any event handler, called with lnet_res_lock/CPT held.
1892 void lnet_peer_push_event(struct lnet_event *ev)
1894 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1895 struct lnet_peer *lp;
1897 /* lnet_find_peer() adds a refcount */
1898 lp = lnet_find_peer(ev->source.nid);
1900 CDEBUG(D_NET, "Push Put from unknown %s (source %s). Ignoring...\n",
1901 libcfs_nid2str(ev->initiator.nid),
1902 libcfs_nid2str(ev->source.nid));
1906 /* Ensure peer state remains consistent while we modify it. */
1907 spin_lock(&lp->lp_lock);
1910 * If some kind of error happened the contents of the message
1911 * cannot be used. Clear the NIDS_UPTODATE and set the
1912 * FORCE_PING flag to trigger a ping.
1915 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
1916 lp->lp_state |= LNET_PEER_FORCE_PING;
1917 CDEBUG(D_NET, "Push Put error %d from %s (source %s)\n",
1919 libcfs_nid2str(lp->lp_primary_nid),
1920 libcfs_nid2str(ev->source.nid));
1925 * A push with invalid or corrupted info. Clear the UPTODATE
1926 * flag to trigger a ping.
1928 if (lnet_ping_info_validate(&pbuf->pb_info)) {
1929 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
1930 lp->lp_state |= LNET_PEER_FORCE_PING;
1931 CDEBUG(D_NET, "Corrupted Push from %s\n",
1932 libcfs_nid2str(lp->lp_primary_nid));
1937 * Make sure we'll allocate the correct size ping buffer when
1940 if (lp->lp_data_nnis < pbuf->pb_info.pi_nnis)
1941 lp->lp_data_nnis = pbuf->pb_info.pi_nnis;
1944 * A non-Multi-Rail peer is not supposed to be capable of
1947 if (!(pbuf->pb_info.pi_features & LNET_PING_FEAT_MULTI_RAIL)) {
1948 CERROR("Push from non-Multi-Rail peer %s dropped\n",
1949 libcfs_nid2str(lp->lp_primary_nid));
1954 * Check the MULTIRAIL flag. Complain if the peer was DLC
1955 * configured without it.
1957 if (!(lp->lp_state & LNET_PEER_MULTI_RAIL)) {
1958 if (lp->lp_state & LNET_PEER_CONFIGURED) {
1959 CERROR("Push says %s is Multi-Rail, DLC says not\n",
1960 libcfs_nid2str(lp->lp_primary_nid));
1962 lp->lp_state |= LNET_PEER_MULTI_RAIL;
1963 lnet_peer_clr_non_mr_pref_nids(lp);
1968 * The peer may have discovery disabled at its end. Set
1969 * NO_DISCOVERY as appropriate.
1971 if (!(pbuf->pb_info.pi_features & LNET_PING_FEAT_DISCOVERY)) {
1972 CDEBUG(D_NET, "Peer %s has discovery disabled\n",
1973 libcfs_nid2str(lp->lp_primary_nid));
1974 lp->lp_state |= LNET_PEER_NO_DISCOVERY;
1975 } else if (lp->lp_state & LNET_PEER_NO_DISCOVERY) {
1976 CDEBUG(D_NET, "Peer %s has discovery enabled\n",
1977 libcfs_nid2str(lp->lp_primary_nid));
1978 lp->lp_state &= ~LNET_PEER_NO_DISCOVERY;
1982 * Check for truncation of the Put message. Clear the
1983 * NIDS_UPTODATE flag and set FORCE_PING to trigger a ping,
1984 * and tell discovery to allocate a bigger buffer.
1986 if (pbuf->pb_nnis < pbuf->pb_info.pi_nnis) {
1987 if (the_lnet.ln_push_target_nnis < pbuf->pb_info.pi_nnis)
1988 the_lnet.ln_push_target_nnis = pbuf->pb_info.pi_nnis;
1989 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
1990 lp->lp_state |= LNET_PEER_FORCE_PING;
1991 CDEBUG(D_NET, "Truncated Push from %s (%d nids)\n",
1992 libcfs_nid2str(lp->lp_primary_nid),
1993 pbuf->pb_info.pi_nnis);
1998 * Check whether the Put data is stale. Stale data can just be
2001 if (pbuf->pb_info.pi_nnis > 1 &&
2002 lp->lp_primary_nid == pbuf->pb_info.pi_ni[1].ns_nid &&
2003 LNET_PING_BUFFER_SEQNO(pbuf) < lp->lp_peer_seqno) {
2004 CDEBUG(D_NET, "Stale Push from %s: got %u have %u\n",
2005 libcfs_nid2str(lp->lp_primary_nid),
2006 LNET_PING_BUFFER_SEQNO(pbuf),
2012 * Check whether the Put data is new, in which case we clear
2013 * the UPTODATE flag and prepare to process it.
2015 * If the Put data is current, and the peer is UPTODATE then
2016 * we assome everything is all right and drop the data as
2019 if (LNET_PING_BUFFER_SEQNO(pbuf) > lp->lp_peer_seqno) {
2020 lp->lp_peer_seqno = LNET_PING_BUFFER_SEQNO(pbuf);
2021 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
2022 } else if (lp->lp_state & LNET_PEER_NIDS_UPTODATE) {
2023 CDEBUG(D_NET, "Stale Push from %s: got %u have %u\n",
2024 libcfs_nid2str(lp->lp_primary_nid),
2025 LNET_PING_BUFFER_SEQNO(pbuf),
2031 * If there is data present that hasn't been processed yet,
2032 * we'll replace it if the Put contained newer data and it
2033 * fits. We're racing with a Ping or earlier Push in this
2036 if (lp->lp_state & LNET_PEER_DATA_PRESENT) {
2037 if (LNET_PING_BUFFER_SEQNO(pbuf) >
2038 LNET_PING_BUFFER_SEQNO(lp->lp_data) &&
2039 pbuf->pb_info.pi_nnis <= lp->lp_data->pb_nnis) {
2040 memcpy(&lp->lp_data->pb_info, &pbuf->pb_info,
2041 LNET_PING_INFO_SIZE(pbuf->pb_info.pi_nnis));
2042 CDEBUG(D_NET, "Ping/Push race from %s: %u vs %u\n",
2043 libcfs_nid2str(lp->lp_primary_nid),
2044 LNET_PING_BUFFER_SEQNO(pbuf),
2045 LNET_PING_BUFFER_SEQNO(lp->lp_data));
2051 * Allocate a buffer to copy the data. On a failure we drop
2052 * the Push and set FORCE_PING to force the discovery
2053 * thread to fix the problem by pinging the peer.
2055 lp->lp_data = lnet_ping_buffer_alloc(lp->lp_data_nnis, GFP_ATOMIC);
2057 lp->lp_state |= LNET_PEER_FORCE_PING;
2058 CDEBUG(D_NET, "Cannot allocate Push buffer for %s %u\n",
2059 libcfs_nid2str(lp->lp_primary_nid),
2060 LNET_PING_BUFFER_SEQNO(pbuf));
2065 memcpy(&lp->lp_data->pb_info, &pbuf->pb_info,
2066 LNET_PING_INFO_SIZE(pbuf->pb_info.pi_nnis));
2067 lp->lp_state |= LNET_PEER_DATA_PRESENT;
2068 CDEBUG(D_NET, "Received Push %s %u\n",
2069 libcfs_nid2str(lp->lp_primary_nid),
2070 LNET_PING_BUFFER_SEQNO(pbuf));
2074 * Queue the peer for discovery if not done, force it on the request
2075 * queue and wake the discovery thread if the peer was already queued,
2076 * because its status changed.
2078 spin_unlock(&lp->lp_lock);
2079 lnet_net_lock(LNET_LOCK_EX);
2080 if (!lnet_peer_is_uptodate(lp) && lnet_peer_queue_for_discovery(lp)) {
2081 list_move(&lp->lp_dc_list, &the_lnet.ln_dc_request);
2082 wake_up(&the_lnet.ln_dc_waitq);
2084 /* Drop refcount from lookup */
2085 lnet_peer_decref_locked(lp);
2086 lnet_net_unlock(LNET_LOCK_EX);
2090 * Clear the discovery error state, unless we're already discovering
2091 * this peer, in which case the error is current.
2093 static void lnet_peer_clear_discovery_error(struct lnet_peer *lp)
2095 spin_lock(&lp->lp_lock);
2096 if (!(lp->lp_state & LNET_PEER_DISCOVERING))
2097 lp->lp_dc_error = 0;
2098 spin_unlock(&lp->lp_lock);
2102 * Peer discovery slow path. The ln_api_mutex is held on entry, and
2103 * dropped/retaken within this function. An lnet_peer_ni is passed in
2104 * because discovery could tear down an lnet_peer.
2107 lnet_discover_peer_locked(struct lnet_peer_ni *lpni, int cpt, bool block)
2110 struct lnet_peer *lp;
2114 lnet_net_unlock(cpt);
2115 lnet_net_lock(LNET_LOCK_EX);
2116 lp = lpni->lpni_peer_net->lpn_peer;
2117 lnet_peer_clear_discovery_error(lp);
2120 * We're willing to be interrupted. The lpni can become a
2121 * zombie if we race with DLC, so we must check for that.
2124 prepare_to_wait(&lp->lp_dc_waitq, &wait, TASK_INTERRUPTIBLE);
2125 if (signal_pending(current))
2127 if (the_lnet.ln_dc_state != LNET_DC_STATE_RUNNING)
2129 if (lp->lp_dc_error)
2131 if (lnet_peer_is_uptodate(lp))
2133 lnet_peer_queue_for_discovery(lp);
2135 * if caller requested a non-blocking operation then
2136 * return immediately. Once discovery is complete then the
2137 * peer ref will be decremented and any pending messages
2138 * that were stopped due to discovery will be transmitted.
2143 lnet_peer_addref_locked(lp);
2144 lnet_net_unlock(LNET_LOCK_EX);
2146 finish_wait(&lp->lp_dc_waitq, &wait);
2147 lnet_net_lock(LNET_LOCK_EX);
2148 lnet_peer_decref_locked(lp);
2149 /* Peer may have changed */
2150 lp = lpni->lpni_peer_net->lpn_peer;
2152 finish_wait(&lp->lp_dc_waitq, &wait);
2154 lnet_net_unlock(LNET_LOCK_EX);
2158 * If the peer has changed after we've discovered the older peer,
2159 * then we need to discovery the new peer to make sure the
2160 * interface information is up to date
2162 if (lp != lpni->lpni_peer_net->lpn_peer)
2165 if (signal_pending(current))
2167 else if (the_lnet.ln_dc_state != LNET_DC_STATE_RUNNING)
2169 else if (lp->lp_dc_error)
2170 rc = lp->lp_dc_error;
2172 CDEBUG(D_NET, "non-blocking discovery\n");
2173 else if (!lnet_peer_is_uptodate(lp))
2176 CDEBUG(D_NET, "peer %s NID %s: %d. %s\n",
2177 (lp ? libcfs_nid2str(lp->lp_primary_nid) : "(none)"),
2178 libcfs_nid2str(lpni->lpni_nid), rc,
2179 (!block) ? "pending discovery" : "discovery complete");
2184 /* Handle an incoming ack for a push. */
2186 lnet_discovery_event_ack(struct lnet_peer *lp, struct lnet_event *ev)
2188 struct lnet_ping_buffer *pbuf;
2190 pbuf = LNET_PING_INFO_TO_BUFFER(ev->md.start);
2191 spin_lock(&lp->lp_lock);
2192 lp->lp_state &= ~LNET_PEER_PUSH_SENT;
2193 lp->lp_push_error = ev->status;
2195 lp->lp_state |= LNET_PEER_PUSH_FAILED;
2197 lp->lp_node_seqno = LNET_PING_BUFFER_SEQNO(pbuf);
2198 spin_unlock(&lp->lp_lock);
2200 CDEBUG(D_NET, "peer %s ev->status %d\n",
2201 libcfs_nid2str(lp->lp_primary_nid), ev->status);
2204 /* Handle a Reply message. This is the reply to a Ping message. */
2206 lnet_discovery_event_reply(struct lnet_peer *lp, struct lnet_event *ev)
2208 struct lnet_ping_buffer *pbuf;
2211 spin_lock(&lp->lp_lock);
2214 * If some kind of error happened the contents of message
2215 * cannot be used. Set PING_FAILED to trigger a retry.
2218 lp->lp_state |= LNET_PEER_PING_FAILED;
2219 lp->lp_ping_error = ev->status;
2220 CDEBUG(D_NET, "Ping Reply error %d from %s (source %s)\n",
2222 libcfs_nid2str(lp->lp_primary_nid),
2223 libcfs_nid2str(ev->source.nid));
2227 pbuf = LNET_PING_INFO_TO_BUFFER(ev->md.start);
2228 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2229 lnet_swap_pinginfo(pbuf);
2232 * A reply with invalid or corrupted info. Set PING_FAILED to
2235 rc = lnet_ping_info_validate(&pbuf->pb_info);
2237 lp->lp_state |= LNET_PEER_PING_FAILED;
2238 lp->lp_ping_error = 0;
2239 CDEBUG(D_NET, "Corrupted Ping Reply from %s: %d\n",
2240 libcfs_nid2str(lp->lp_primary_nid), rc);
2245 * Update the MULTI_RAIL flag based on the reply. If the peer
2246 * was configured with DLC then the setting should match what
2249 if (pbuf->pb_info.pi_features & LNET_PING_FEAT_MULTI_RAIL) {
2250 if (lp->lp_state & LNET_PEER_MULTI_RAIL) {
2251 /* Everything's fine */
2252 } else if (lp->lp_state & LNET_PEER_CONFIGURED) {
2253 CWARN("Reply says %s is Multi-Rail, DLC says not\n",
2254 libcfs_nid2str(lp->lp_primary_nid));
2256 lp->lp_state |= LNET_PEER_MULTI_RAIL;
2257 lnet_peer_clr_non_mr_pref_nids(lp);
2259 } else if (lp->lp_state & LNET_PEER_MULTI_RAIL) {
2260 if (lp->lp_state & LNET_PEER_CONFIGURED) {
2261 CWARN("DLC says %s is Multi-Rail, Reply says not\n",
2262 libcfs_nid2str(lp->lp_primary_nid));
2264 CERROR("Multi-Rail state vanished from %s\n",
2265 libcfs_nid2str(lp->lp_primary_nid));
2266 lp->lp_state &= ~LNET_PEER_MULTI_RAIL;
2271 * Make sure we'll allocate the correct size ping buffer when
2274 if (lp->lp_data_nnis < pbuf->pb_info.pi_nnis)
2275 lp->lp_data_nnis = pbuf->pb_info.pi_nnis;
2278 * The peer may have discovery disabled at its end. Set
2279 * NO_DISCOVERY as appropriate.
2281 if (!(pbuf->pb_info.pi_features & LNET_PING_FEAT_DISCOVERY)) {
2282 CDEBUG(D_NET, "Peer %s has discovery disabled\n",
2283 libcfs_nid2str(lp->lp_primary_nid));
2284 lp->lp_state |= LNET_PEER_NO_DISCOVERY;
2285 } else if (lp->lp_state & LNET_PEER_NO_DISCOVERY) {
2286 CDEBUG(D_NET, "Peer %s has discovery enabled\n",
2287 libcfs_nid2str(lp->lp_primary_nid));
2288 lp->lp_state &= ~LNET_PEER_NO_DISCOVERY;
2292 * Check for truncation of the Reply. Clear PING_SENT and set
2293 * PING_FAILED to trigger a retry.
2295 if (pbuf->pb_nnis < pbuf->pb_info.pi_nnis) {
2296 if (the_lnet.ln_push_target_nnis < pbuf->pb_info.pi_nnis)
2297 the_lnet.ln_push_target_nnis = pbuf->pb_info.pi_nnis;
2298 lp->lp_state |= LNET_PEER_PING_FAILED;
2299 lp->lp_ping_error = 0;
2300 CDEBUG(D_NET, "Truncated Reply from %s (%d nids)\n",
2301 libcfs_nid2str(lp->lp_primary_nid),
2302 pbuf->pb_info.pi_nnis);
2307 * Check the sequence numbers in the reply. These are only
2308 * available if the reply came from a Multi-Rail peer.
2310 if (pbuf->pb_info.pi_features & LNET_PING_FEAT_MULTI_RAIL &&
2311 pbuf->pb_info.pi_nnis > 1 &&
2312 lp->lp_primary_nid == pbuf->pb_info.pi_ni[1].ns_nid) {
2313 if (LNET_PING_BUFFER_SEQNO(pbuf) < lp->lp_peer_seqno) {
2314 CDEBUG(D_NET, "Stale Reply from %s: got %u have %u\n",
2315 libcfs_nid2str(lp->lp_primary_nid),
2316 LNET_PING_BUFFER_SEQNO(pbuf),
2321 if (LNET_PING_BUFFER_SEQNO(pbuf) > lp->lp_peer_seqno)
2322 lp->lp_peer_seqno = LNET_PING_BUFFER_SEQNO(pbuf);
2325 /* We're happy with the state of the data in the buffer. */
2326 CDEBUG(D_NET, "peer %s data present %u\n",
2327 libcfs_nid2str(lp->lp_primary_nid), lp->lp_peer_seqno);
2328 if (lp->lp_state & LNET_PEER_DATA_PRESENT)
2329 lnet_ping_buffer_decref(lp->lp_data);
2331 lp->lp_state |= LNET_PEER_DATA_PRESENT;
2332 lnet_ping_buffer_addref(pbuf);
2335 lp->lp_state &= ~LNET_PEER_PING_SENT;
2336 spin_unlock(&lp->lp_lock);
2340 * Send event handling. Only matters for error cases, where we clean
2341 * up state on the peer and peer_ni that would otherwise be updated in
2342 * the REPLY event handler for a successful Ping, and the ACK event
2343 * handler for a successful Push.
2346 lnet_discovery_event_send(struct lnet_peer *lp, struct lnet_event *ev)
2353 spin_lock(&lp->lp_lock);
2354 if (ev->msg_type == LNET_MSG_GET) {
2355 lp->lp_state &= ~LNET_PEER_PING_SENT;
2356 lp->lp_state |= LNET_PEER_PING_FAILED;
2357 lp->lp_ping_error = ev->status;
2358 } else { /* ev->msg_type == LNET_MSG_PUT */
2359 lp->lp_state &= ~LNET_PEER_PUSH_SENT;
2360 lp->lp_state |= LNET_PEER_PUSH_FAILED;
2361 lp->lp_push_error = ev->status;
2363 spin_unlock(&lp->lp_lock);
2364 rc = LNET_REDISCOVER_PEER;
2366 CDEBUG(D_NET, "%s Send to %s: %d\n",
2367 (ev->msg_type == LNET_MSG_GET ? "Ping" : "Push"),
2368 libcfs_nid2str(ev->target.nid), rc);
2373 * Unlink event handling. This event is only seen if a call to
2374 * LNetMDUnlink() caused the event to be unlinked. If this call was
2375 * made after the event was set up in LNetGet() or LNetPut() then we
2376 * assume the Ping or Push timed out.
2379 lnet_discovery_event_unlink(struct lnet_peer *lp, struct lnet_event *ev)
2381 spin_lock(&lp->lp_lock);
2382 /* We've passed through LNetGet() */
2383 if (lp->lp_state & LNET_PEER_PING_SENT) {
2384 lp->lp_state &= ~LNET_PEER_PING_SENT;
2385 lp->lp_state |= LNET_PEER_PING_FAILED;
2386 lp->lp_ping_error = -ETIMEDOUT;
2387 CDEBUG(D_NET, "Ping Unlink for message to peer %s\n",
2388 libcfs_nid2str(lp->lp_primary_nid));
2390 /* We've passed through LNetPut() */
2391 if (lp->lp_state & LNET_PEER_PUSH_SENT) {
2392 lp->lp_state &= ~LNET_PEER_PUSH_SENT;
2393 lp->lp_state |= LNET_PEER_PUSH_FAILED;
2394 lp->lp_push_error = -ETIMEDOUT;
2395 CDEBUG(D_NET, "Push Unlink for message to peer %s\n",
2396 libcfs_nid2str(lp->lp_primary_nid));
2398 spin_unlock(&lp->lp_lock);
2402 * Event handler for the discovery EQ.
2404 * Called with lnet_res_lock(cpt) held. The cpt is the
2405 * lnet_cpt_of_cookie() of the md handle cookie.
2407 static void lnet_discovery_event_handler(struct lnet_event *event)
2409 struct lnet_peer *lp = event->md.user_ptr;
2410 struct lnet_ping_buffer *pbuf;
2413 /* discovery needs to take another look */
2414 rc = LNET_REDISCOVER_PEER;
2416 CDEBUG(D_NET, "Received event: %d\n", event->type);
2418 switch (event->type) {
2419 case LNET_EVENT_ACK:
2420 lnet_discovery_event_ack(lp, event);
2422 case LNET_EVENT_REPLY:
2423 lnet_discovery_event_reply(lp, event);
2425 case LNET_EVENT_SEND:
2426 /* Only send failure triggers a retry. */
2427 rc = lnet_discovery_event_send(lp, event);
2429 case LNET_EVENT_UNLINK:
2430 /* LNetMDUnlink() was called */
2431 lnet_discovery_event_unlink(lp, event);
2434 /* Invalid events. */
2437 lnet_net_lock(LNET_LOCK_EX);
2438 if (event->unlinked) {
2439 pbuf = LNET_PING_INFO_TO_BUFFER(event->md.start);
2440 lnet_ping_buffer_decref(pbuf);
2441 lnet_peer_decref_locked(lp);
2444 /* put peer back at end of request queue, if discovery not already
2446 if (rc == LNET_REDISCOVER_PEER && !lnet_peer_is_uptodate(lp)) {
2447 list_move_tail(&lp->lp_dc_list, &the_lnet.ln_dc_request);
2448 wake_up(&the_lnet.ln_dc_waitq);
2450 lnet_net_unlock(LNET_LOCK_EX);
2454 * Build a peer from incoming data.
2456 * The NIDs in the incoming data are supposed to be structured as follows:
2459 * - other NIDs in same net
2460 * - NIDs in second net
2461 * - NIDs in third net
2463 * This due to the way the list of NIDs in the data is created.
2465 * Note that this function will mark the peer uptodate unless an
2466 * ENOMEM is encontered. All other errors are due to a conflict
2467 * between the DLC configuration and what discovery sees. We treat DLC
2468 * as binding, and therefore set the NIDS_UPTODATE flag to prevent the
2469 * peer from becoming stuck in discovery.
2471 static int lnet_peer_merge_data(struct lnet_peer *lp,
2472 struct lnet_ping_buffer *pbuf)
2474 struct lnet_peer_ni *lpni;
2475 lnet_nid_t *curnis = NULL;
2476 struct lnet_ni_status *addnis = NULL;
2477 lnet_nid_t *delnis = NULL;
2487 flags = LNET_PEER_DISCOVERED;
2488 if (pbuf->pb_info.pi_features & LNET_PING_FEAT_MULTI_RAIL)
2489 flags |= LNET_PEER_MULTI_RAIL;
2492 * Cache the routing feature for the peer; whether it is enabled
2493 * for disabled as reported by the remote peer.
2495 spin_lock(&lp->lp_lock);
2496 if (!(pbuf->pb_info.pi_features & LNET_PING_FEAT_RTE_DISABLED))
2497 lp->lp_state |= LNET_PEER_ROUTER_ENABLED;
2499 lp->lp_state &= ~LNET_PEER_ROUTER_ENABLED;
2500 spin_unlock(&lp->lp_lock);
2502 nnis = MAX(lp->lp_nnis, pbuf->pb_info.pi_nnis);
2503 LIBCFS_ALLOC(curnis, nnis * sizeof(*curnis));
2504 LIBCFS_ALLOC(addnis, nnis * sizeof(*addnis));
2505 LIBCFS_ALLOC(delnis, nnis * sizeof(*delnis));
2506 if (!curnis || !addnis || !delnis) {
2514 /* Construct the list of NIDs present in peer. */
2516 while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL)
2517 curnis[ncurnis++] = lpni->lpni_nid;
2520 * Check for NIDs in pbuf not present in curnis[].
2521 * The loop starts at 1 to skip the loopback NID.
2523 for (i = 1; i < pbuf->pb_info.pi_nnis; i++) {
2524 for (j = 0; j < ncurnis; j++)
2525 if (pbuf->pb_info.pi_ni[i].ns_nid == curnis[j])
2528 addnis[naddnis++] = pbuf->pb_info.pi_ni[i];
2531 * Check for NIDs in curnis[] not present in pbuf.
2532 * The nested loop starts at 1 to skip the loopback NID.
2534 * But never add the loopback NID to delnis[]: if it is
2535 * present in curnis[] then this peer is for this node.
2537 for (i = 0; i < ncurnis; i++) {
2538 if (LNET_NETTYP(LNET_NIDNET(curnis[i])) == LOLND)
2540 for (j = 1; j < pbuf->pb_info.pi_nnis; j++) {
2541 if (curnis[i] == pbuf->pb_info.pi_ni[j].ns_nid) {
2543 * update the information we cache for the
2544 * peer with the latest information we
2547 lpni = lnet_find_peer_ni_locked(curnis[i]);
2549 lpni->lpni_ns_status = pbuf->pb_info.pi_ni[j].ns_status;
2550 lnet_peer_ni_decref_locked(lpni);
2555 if (j == pbuf->pb_info.pi_nnis)
2556 delnis[ndelnis++] = curnis[i];
2559 for (i = 0; i < naddnis; i++) {
2560 rc = lnet_peer_add_nid(lp, addnis[i].ns_nid, flags);
2562 CERROR("Error adding NID %s to peer %s: %d\n",
2563 libcfs_nid2str(addnis[i].ns_nid),
2564 libcfs_nid2str(lp->lp_primary_nid), rc);
2568 lpni = lnet_find_peer_ni_locked(addnis[i].ns_nid);
2570 lpni->lpni_ns_status = addnis[i].ns_status;
2571 lnet_peer_ni_decref_locked(lpni);
2575 for (i = 0; i < ndelnis; i++) {
2577 * for routers it's okay to delete the primary_nid because
2578 * the upper layers don't really rely on it. So if we're
2579 * being told that the router changed its primary_nid
2580 * then it's okay to delete it.
2582 if (lp->lp_rtr_refcount > 0)
2583 flags |= LNET_PEER_RTR_NI_FORCE_DEL;
2584 rc = lnet_peer_del_nid(lp, delnis[i], flags);
2586 CERROR("Error deleting NID %s from peer %s: %d\n",
2587 libcfs_nid2str(delnis[i]),
2588 libcfs_nid2str(lp->lp_primary_nid), rc);
2594 * Errors other than -ENOMEM are due to peers having been
2595 * configured with DLC. Ignore these because DLC overrides
2600 LIBCFS_FREE(curnis, nnis * sizeof(*curnis));
2601 LIBCFS_FREE(addnis, nnis * sizeof(*addnis));
2602 LIBCFS_FREE(delnis, nnis * sizeof(*delnis));
2603 lnet_ping_buffer_decref(pbuf);
2604 CDEBUG(D_NET, "peer %s: %d\n", libcfs_nid2str(lp->lp_primary_nid), rc);
2607 spin_lock(&lp->lp_lock);
2608 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
2609 lp->lp_state |= LNET_PEER_FORCE_PING;
2610 spin_unlock(&lp->lp_lock);
2616 * The data in pbuf says lp is its primary peer, but the data was
2617 * received by a different peer. Try to update lp with the data.
2620 lnet_peer_set_primary_data(struct lnet_peer *lp, struct lnet_ping_buffer *pbuf)
2622 struct lnet_handle_md mdh;
2624 /* Queue lp for discovery, and force it on the request queue. */
2625 lnet_net_lock(LNET_LOCK_EX);
2626 if (lnet_peer_queue_for_discovery(lp))
2627 list_move(&lp->lp_dc_list, &the_lnet.ln_dc_request);
2628 lnet_net_unlock(LNET_LOCK_EX);
2630 LNetInvalidateMDHandle(&mdh);
2633 * Decide whether we can move the peer to the DATA_PRESENT state.
2635 * We replace stale data for a multi-rail peer, repair PING_FAILED
2636 * status, and preempt FORCE_PING.
2638 * If after that we have DATA_PRESENT, we merge it into this peer.
2640 spin_lock(&lp->lp_lock);
2641 if (lp->lp_state & LNET_PEER_MULTI_RAIL) {
2642 if (lp->lp_peer_seqno < LNET_PING_BUFFER_SEQNO(pbuf)) {
2643 lp->lp_peer_seqno = LNET_PING_BUFFER_SEQNO(pbuf);
2644 } else if (lp->lp_state & LNET_PEER_DATA_PRESENT) {
2645 lp->lp_state &= ~LNET_PEER_DATA_PRESENT;
2646 lnet_ping_buffer_decref(pbuf);
2651 if (lp->lp_state & LNET_PEER_DATA_PRESENT) {
2652 lnet_ping_buffer_decref(lp->lp_data);
2654 lp->lp_state &= ~LNET_PEER_DATA_PRESENT;
2656 if (lp->lp_state & LNET_PEER_PING_FAILED) {
2657 mdh = lp->lp_ping_mdh;
2658 LNetInvalidateMDHandle(&lp->lp_ping_mdh);
2659 lp->lp_state &= ~LNET_PEER_PING_FAILED;
2660 lp->lp_ping_error = 0;
2662 if (lp->lp_state & LNET_PEER_FORCE_PING)
2663 lp->lp_state &= ~LNET_PEER_FORCE_PING;
2664 lp->lp_state |= LNET_PEER_NIDS_UPTODATE;
2665 spin_unlock(&lp->lp_lock);
2667 if (!LNetMDHandleIsInvalid(mdh))
2671 return lnet_peer_merge_data(lp, pbuf);
2673 CDEBUG(D_NET, "peer %s\n", libcfs_nid2str(lp->lp_primary_nid));
2678 * Update a peer using the data received.
2680 static int lnet_peer_data_present(struct lnet_peer *lp)
2681 __must_hold(&lp->lp_lock)
2683 struct lnet_ping_buffer *pbuf;
2684 struct lnet_peer_ni *lpni;
2685 lnet_nid_t nid = LNET_NID_ANY;
2691 lp->lp_state &= ~LNET_PEER_DATA_PRESENT;
2692 lp->lp_state |= LNET_PEER_NIDS_UPTODATE;
2693 spin_unlock(&lp->lp_lock);
2696 * Modifications of peer structures are done while holding the
2697 * ln_api_mutex. A global lock is required because we may be
2698 * modifying multiple peer structures, and a mutex greatly
2699 * simplifies memory management.
2701 * The actual changes to the data structures must also protect
2702 * against concurrent lookups, for which the lnet_net_lock in
2703 * LNET_LOCK_EX mode is used.
2705 mutex_lock(&the_lnet.ln_api_mutex);
2706 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
2712 * If this peer is not on the peer list then it is being torn
2713 * down, and our reference count may be all that is keeping it
2714 * alive. Don't do any work on it.
2716 if (list_empty(&lp->lp_peer_list))
2719 flags = LNET_PEER_DISCOVERED;
2720 if (pbuf->pb_info.pi_features & LNET_PING_FEAT_MULTI_RAIL)
2721 flags |= LNET_PEER_MULTI_RAIL;
2724 * Check whether the primary NID in the message matches the
2725 * primary NID of the peer. If it does, update the peer, if
2726 * it it does not, check whether there is already a peer with
2727 * that primary NID. If no such peer exists, try to update
2728 * the primary NID of the current peer (allowed if it was
2729 * created due to message traffic) and complete the update.
2730 * If the peer did exist, hand off the data to it.
2732 * The peer for the loopback interface is a special case: this
2733 * is the peer for the local node, and we want to set its
2734 * primary NID to the correct value here. Moreover, this peer
2735 * can show up with only the loopback NID in the ping buffer.
2737 if (pbuf->pb_info.pi_nnis <= 1)
2739 nid = pbuf->pb_info.pi_ni[1].ns_nid;
2740 if (LNET_NETTYP(LNET_NIDNET(lp->lp_primary_nid)) == LOLND) {
2741 rc = lnet_peer_set_primary_nid(lp, nid, flags);
2743 rc = lnet_peer_merge_data(lp, pbuf);
2744 } else if (lp->lp_primary_nid == nid) {
2745 rc = lnet_peer_merge_data(lp, pbuf);
2747 lpni = lnet_find_peer_ni_locked(nid);
2749 rc = lnet_peer_set_primary_nid(lp, nid, flags);
2751 CERROR("Primary NID error %s versus %s: %d\n",
2752 libcfs_nid2str(lp->lp_primary_nid),
2753 libcfs_nid2str(nid), rc);
2755 rc = lnet_peer_merge_data(lp, pbuf);
2758 struct lnet_peer *new_lp;
2759 new_lp = lpni->lpni_peer_net->lpn_peer;
2760 rc = lnet_peer_set_primary_data(new_lp, pbuf);
2761 lnet_consolidate_routes_locked(lp, new_lp);
2762 lnet_peer_ni_decref_locked(lpni);
2766 CDEBUG(D_NET, "peer %s: %d\n", libcfs_nid2str(lp->lp_primary_nid), rc);
2767 mutex_unlock(&the_lnet.ln_api_mutex);
2769 spin_lock(&lp->lp_lock);
2770 /* Tell discovery to re-check the peer immediately. */
2772 rc = LNET_REDISCOVER_PEER;
2777 * A ping failed. Clear the PING_FAILED state and set the
2778 * FORCE_PING state, to ensure a retry even if discovery is
2779 * disabled. This avoids being left with incorrect state.
2781 static int lnet_peer_ping_failed(struct lnet_peer *lp)
2782 __must_hold(&lp->lp_lock)
2784 struct lnet_handle_md mdh;
2787 mdh = lp->lp_ping_mdh;
2788 LNetInvalidateMDHandle(&lp->lp_ping_mdh);
2789 lp->lp_state &= ~LNET_PEER_PING_FAILED;
2790 lp->lp_state |= LNET_PEER_FORCE_PING;
2791 rc = lp->lp_ping_error;
2792 lp->lp_ping_error = 0;
2793 spin_unlock(&lp->lp_lock);
2795 if (!LNetMDHandleIsInvalid(mdh))
2798 CDEBUG(D_NET, "peer %s:%d\n",
2799 libcfs_nid2str(lp->lp_primary_nid), rc);
2801 spin_lock(&lp->lp_lock);
2802 return rc ? rc : LNET_REDISCOVER_PEER;
2806 * Select NID to send a Ping or Push to.
2808 static lnet_nid_t lnet_peer_select_nid(struct lnet_peer *lp)
2810 struct lnet_peer_ni *lpni;
2812 /* Look for a direct-connected NID for this peer. */
2814 while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL) {
2815 if (!lnet_get_net_locked(lpni->lpni_peer_net->lpn_net_id))
2820 return lpni->lpni_nid;
2822 /* Look for a routed-connected NID for this peer. */
2824 while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL) {
2825 if (!lnet_find_rnet_locked(lpni->lpni_peer_net->lpn_net_id))
2830 return lpni->lpni_nid;
2832 return LNET_NID_ANY;
2835 /* Active side of ping. */
2836 static int lnet_peer_send_ping(struct lnet_peer *lp)
2837 __must_hold(&lp->lp_lock)
2844 lp->lp_state |= LNET_PEER_PING_SENT;
2845 lp->lp_state &= ~LNET_PEER_FORCE_PING;
2846 spin_unlock(&lp->lp_lock);
2848 cpt = lnet_net_lock_current();
2849 /* Refcount for MD. */
2850 lnet_peer_addref_locked(lp);
2851 pnid = lnet_peer_select_nid(lp);
2852 lnet_net_unlock(cpt);
2854 nnis = MAX(lp->lp_data_nnis, LNET_INTERFACES_MIN);
2856 rc = lnet_send_ping(pnid, &lp->lp_ping_mdh, nnis, lp,
2857 the_lnet.ln_dc_eqh, false);
2860 * if LNetMDBind in lnet_send_ping fails we need to decrement the
2861 * refcount on the peer, otherwise LNetMDUnlink will be called
2862 * which will eventually do that.
2866 lnet_peer_decref_locked(lp);
2867 lnet_net_unlock(cpt);
2868 rc = -rc; /* change the rc to negative value */
2870 } else if (rc < 0) {
2874 CDEBUG(D_NET, "peer %s\n", libcfs_nid2str(lp->lp_primary_nid));
2876 spin_lock(&lp->lp_lock);
2880 CDEBUG(D_NET, "peer %s: %d\n", libcfs_nid2str(lp->lp_primary_nid), rc);
2882 * The errors that get us here are considered hard errors and
2883 * cause Discovery to terminate. So we clear PING_SENT, but do
2884 * not set either PING_FAILED or FORCE_PING. In fact we need
2885 * to clear PING_FAILED, because the unlink event handler will
2886 * have set it if we called LNetMDUnlink() above.
2888 spin_lock(&lp->lp_lock);
2889 lp->lp_state &= ~(LNET_PEER_PING_SENT | LNET_PEER_PING_FAILED);
2894 * This function exists because you cannot call LNetMDUnlink() from an
2897 static int lnet_peer_push_failed(struct lnet_peer *lp)
2898 __must_hold(&lp->lp_lock)
2900 struct lnet_handle_md mdh;
2903 mdh = lp->lp_push_mdh;
2904 LNetInvalidateMDHandle(&lp->lp_push_mdh);
2905 lp->lp_state &= ~LNET_PEER_PUSH_FAILED;
2906 rc = lp->lp_push_error;
2907 lp->lp_push_error = 0;
2908 spin_unlock(&lp->lp_lock);
2910 if (!LNetMDHandleIsInvalid(mdh))
2913 CDEBUG(D_NET, "peer %s\n", libcfs_nid2str(lp->lp_primary_nid));
2914 spin_lock(&lp->lp_lock);
2915 return rc ? rc : LNET_REDISCOVER_PEER;
2918 /* Active side of push. */
2919 static int lnet_peer_send_push(struct lnet_peer *lp)
2920 __must_hold(&lp->lp_lock)
2922 struct lnet_ping_buffer *pbuf;
2923 struct lnet_process_id id;
2928 /* Don't push to a non-multi-rail peer. */
2929 if (!(lp->lp_state & LNET_PEER_MULTI_RAIL)) {
2930 lp->lp_state &= ~LNET_PEER_FORCE_PUSH;
2934 lp->lp_state |= LNET_PEER_PUSH_SENT;
2935 lp->lp_state &= ~LNET_PEER_FORCE_PUSH;
2936 spin_unlock(&lp->lp_lock);
2938 cpt = lnet_net_lock_current();
2939 pbuf = the_lnet.ln_ping_target;
2940 lnet_ping_buffer_addref(pbuf);
2941 lnet_net_unlock(cpt);
2943 /* Push source MD */
2944 md.start = &pbuf->pb_info;
2945 md.length = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2946 md.threshold = 2; /* Put/Ack */
2949 md.eq_handle = the_lnet.ln_dc_eqh;
2952 rc = LNetMDBind(md, LNET_UNLINK, &lp->lp_push_mdh);
2954 lnet_ping_buffer_decref(pbuf);
2955 CERROR("Can't bind push source MD: %d\n", rc);
2958 cpt = lnet_net_lock_current();
2959 /* Refcount for MD. */
2960 lnet_peer_addref_locked(lp);
2961 id.pid = LNET_PID_LUSTRE;
2962 id.nid = lnet_peer_select_nid(lp);
2963 lnet_net_unlock(cpt);
2965 if (id.nid == LNET_NID_ANY) {
2970 rc = LNetPut(LNET_NID_ANY, lp->lp_push_mdh,
2971 LNET_ACK_REQ, id, LNET_RESERVED_PORTAL,
2972 LNET_PROTO_PING_MATCHBITS, 0, 0);
2977 CDEBUG(D_NET, "peer %s\n", libcfs_nid2str(lp->lp_primary_nid));
2979 spin_lock(&lp->lp_lock);
2983 LNetMDUnlink(lp->lp_push_mdh);
2984 LNetInvalidateMDHandle(&lp->lp_push_mdh);
2986 CDEBUG(D_NET, "peer %s: %d\n", libcfs_nid2str(lp->lp_primary_nid), rc);
2988 * The errors that get us here are considered hard errors and
2989 * cause Discovery to terminate. So we clear PUSH_SENT, but do
2990 * not set PUSH_FAILED. In fact we need to clear PUSH_FAILED,
2991 * because the unlink event handler will have set it if we
2992 * called LNetMDUnlink() above.
2994 spin_lock(&lp->lp_lock);
2995 lp->lp_state &= ~(LNET_PEER_PUSH_SENT | LNET_PEER_PUSH_FAILED);
3000 * An unrecoverable error was encountered during discovery.
3001 * Set error status in peer and abort discovery.
3003 static void lnet_peer_discovery_error(struct lnet_peer *lp, int error)
3005 CDEBUG(D_NET, "Discovery error %s: %d\n",
3006 libcfs_nid2str(lp->lp_primary_nid), error);
3008 spin_lock(&lp->lp_lock);
3009 lp->lp_dc_error = error;
3010 lp->lp_state &= ~LNET_PEER_DISCOVERING;
3011 lp->lp_state |= LNET_PEER_REDISCOVER;
3012 spin_unlock(&lp->lp_lock);
3016 * Mark the peer as discovered.
3018 static int lnet_peer_discovered(struct lnet_peer *lp)
3019 __must_hold(&lp->lp_lock)
3021 lp->lp_state |= LNET_PEER_DISCOVERED;
3022 lp->lp_state &= ~(LNET_PEER_DISCOVERING |
3023 LNET_PEER_REDISCOVER);
3025 CDEBUG(D_NET, "peer %s\n", libcfs_nid2str(lp->lp_primary_nid));
3031 * Mark the peer as to be rediscovered.
3033 static int lnet_peer_rediscover(struct lnet_peer *lp)
3034 __must_hold(&lp->lp_lock)
3036 lp->lp_state |= LNET_PEER_REDISCOVER;
3037 lp->lp_state &= ~LNET_PEER_DISCOVERING;
3039 CDEBUG(D_NET, "peer %s\n", libcfs_nid2str(lp->lp_primary_nid));
3045 * Discovering this peer is taking too long. Cancel any Ping or Push
3046 * that discovery is waiting on by unlinking the relevant MDs. The
3047 * lnet_discovery_event_handler() will proceed from here and complete
3050 static void lnet_peer_cancel_discovery(struct lnet_peer *lp)
3052 struct lnet_handle_md ping_mdh;
3053 struct lnet_handle_md push_mdh;
3055 LNetInvalidateMDHandle(&ping_mdh);
3056 LNetInvalidateMDHandle(&push_mdh);
3058 spin_lock(&lp->lp_lock);
3059 if (lp->lp_state & LNET_PEER_PING_SENT) {
3060 ping_mdh = lp->lp_ping_mdh;
3061 LNetInvalidateMDHandle(&lp->lp_ping_mdh);
3063 if (lp->lp_state & LNET_PEER_PUSH_SENT) {
3064 push_mdh = lp->lp_push_mdh;
3065 LNetInvalidateMDHandle(&lp->lp_push_mdh);
3067 spin_unlock(&lp->lp_lock);
3069 if (!LNetMDHandleIsInvalid(ping_mdh))
3070 LNetMDUnlink(ping_mdh);
3071 if (!LNetMDHandleIsInvalid(push_mdh))
3072 LNetMDUnlink(push_mdh);
3076 * Wait for work to be queued or some other change that must be
3077 * attended to. Returns non-zero if the discovery thread should shut
3080 static int lnet_peer_discovery_wait_for_work(void)
3087 cpt = lnet_net_lock_current();
3089 prepare_to_wait(&the_lnet.ln_dc_waitq, &wait,
3090 TASK_INTERRUPTIBLE);
3091 if (the_lnet.ln_dc_state == LNET_DC_STATE_STOPPING)
3093 if (lnet_push_target_resize_needed())
3095 if (!list_empty(&the_lnet.ln_dc_request))
3097 if (!list_empty(&the_lnet.ln_msg_resend))
3099 lnet_net_unlock(cpt);
3102 * wakeup max every second to check if there are peers that
3103 * have been stuck on the working queue for greater than
3106 schedule_timeout(cfs_time_seconds(1));
3107 finish_wait(&the_lnet.ln_dc_waitq, &wait);
3108 cpt = lnet_net_lock_current();
3110 finish_wait(&the_lnet.ln_dc_waitq, &wait);
3112 if (the_lnet.ln_dc_state == LNET_DC_STATE_STOPPING)
3115 lnet_net_unlock(cpt);
3117 CDEBUG(D_NET, "woken: %d\n", rc);
3123 * Messages that were pending on a destroyed peer will be put on a global
3124 * resend list. The message resend list will be checked by
3125 * the discovery thread when it wakes up, and will resend messages. These
3126 * messages can still be sendable in the case the lpni which was the initial
3127 * cause of the message re-queue was transfered to another peer.
3129 * It is possible that LNet could be shutdown while we're iterating
3130 * through the list. lnet_shudown_lndnets() will attempt to access the
3131 * resend list, but will have to wait until the spinlock is released, by
3132 * which time there shouldn't be any more messages on the resend list.
3133 * During shutdown lnet_send() will fail and lnet_finalize() will be called
3134 * for the messages so they can be released. The other case is that
3135 * lnet_shudown_lndnets() can finalize all the messages before this
3136 * function can visit the resend list, in which case this function will be
3139 static void lnet_resend_msgs(void)
3141 struct lnet_msg *msg, *tmp;
3142 struct list_head resend;
3145 INIT_LIST_HEAD(&resend);
3147 spin_lock(&the_lnet.ln_msg_resend_lock);
3148 list_splice(&the_lnet.ln_msg_resend, &resend);
3149 spin_unlock(&the_lnet.ln_msg_resend_lock);
3151 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
3152 list_del_init(&msg->msg_list);
3153 rc = lnet_send(msg->msg_src_nid_param, msg,
3154 msg->msg_rtr_nid_param);
3156 CNETERR("Error sending %s to %s: %d\n",
3157 lnet_msgtyp2str(msg->msg_type),
3158 libcfs_id2str(msg->msg_target), rc);
3159 lnet_finalize(msg, rc);
3164 /* The discovery thread. */
3165 static int lnet_peer_discovery(void *arg)
3167 struct lnet_peer *lp;
3170 CDEBUG(D_NET, "started\n");
3171 cfs_block_allsigs();
3174 if (lnet_peer_discovery_wait_for_work())
3179 if (lnet_push_target_resize_needed())
3180 lnet_push_target_resize();
3182 lnet_net_lock(LNET_LOCK_EX);
3183 if (the_lnet.ln_dc_state == LNET_DC_STATE_STOPPING)
3187 * Process all incoming discovery work requests. When
3188 * discovery must wait on a peer to change state, it
3189 * is added to the tail of the ln_dc_working queue. A
3190 * timestamp keeps track of when the peer was added,
3191 * so we can time out discovery requests that take too
3194 while (!list_empty(&the_lnet.ln_dc_request)) {
3195 lp = list_first_entry(&the_lnet.ln_dc_request,
3196 struct lnet_peer, lp_dc_list);
3197 list_move(&lp->lp_dc_list, &the_lnet.ln_dc_working);
3199 * set the time the peer was put on the dc_working
3200 * queue. It shouldn't remain on the queue
3201 * forever, in case the GET message (for ping)
3202 * doesn't get a REPLY or the PUT message (for
3203 * push) doesn't get an ACK.
3205 lp->lp_last_queued = ktime_get_real_seconds();
3206 lnet_net_unlock(LNET_LOCK_EX);
3209 * Select an action depending on the state of
3210 * the peer and whether discovery is disabled.
3211 * The check whether discovery is disabled is
3212 * done after the code that handles processing
3213 * for arrived data, cleanup for failures, and
3214 * forcing a Ping or Push.
3216 spin_lock(&lp->lp_lock);
3217 CDEBUG(D_NET, "peer %s state %#x\n",
3218 libcfs_nid2str(lp->lp_primary_nid),
3220 if (lp->lp_state & LNET_PEER_DATA_PRESENT)
3221 rc = lnet_peer_data_present(lp);
3222 else if (lp->lp_state & LNET_PEER_PING_FAILED)
3223 rc = lnet_peer_ping_failed(lp);
3224 else if (lp->lp_state & LNET_PEER_PUSH_FAILED)
3225 rc = lnet_peer_push_failed(lp);
3226 else if (lp->lp_state & LNET_PEER_FORCE_PING)
3227 rc = lnet_peer_send_ping(lp);
3228 else if (lp->lp_state & LNET_PEER_FORCE_PUSH)
3229 rc = lnet_peer_send_push(lp);
3230 else if (lnet_peer_discovery_disabled)
3231 rc = lnet_peer_rediscover(lp);
3232 else if (!(lp->lp_state & LNET_PEER_NIDS_UPTODATE))
3233 rc = lnet_peer_send_ping(lp);
3234 else if (lnet_peer_needs_push(lp))
3235 rc = lnet_peer_send_push(lp);
3237 rc = lnet_peer_discovered(lp);
3238 CDEBUG(D_NET, "peer %s state %#x rc %d\n",
3239 libcfs_nid2str(lp->lp_primary_nid),
3241 spin_unlock(&lp->lp_lock);
3243 lnet_net_lock(LNET_LOCK_EX);
3244 if (rc == LNET_REDISCOVER_PEER) {
3245 list_move(&lp->lp_dc_list,
3246 &the_lnet.ln_dc_request);
3248 lnet_peer_discovery_error(lp, rc);
3250 if (!(lp->lp_state & LNET_PEER_DISCOVERING))
3251 lnet_peer_discovery_complete(lp);
3252 if (the_lnet.ln_dc_state == LNET_DC_STATE_STOPPING)
3256 lnet_net_unlock(LNET_LOCK_EX);
3259 CDEBUG(D_NET, "stopping\n");
3261 * Clean up before telling lnet_peer_discovery_stop() that
3262 * we're done. Use wake_up() below to somewhat reduce the
3263 * size of the thundering herd if there are multiple threads
3264 * waiting on discovery of a single peer.
3267 /* Queue cleanup 1: stop all pending pings and pushes. */
3268 lnet_net_lock(LNET_LOCK_EX);
3269 while (!list_empty(&the_lnet.ln_dc_working)) {
3270 lp = list_first_entry(&the_lnet.ln_dc_working,
3271 struct lnet_peer, lp_dc_list);
3272 list_move(&lp->lp_dc_list, &the_lnet.ln_dc_expired);
3273 lnet_net_unlock(LNET_LOCK_EX);
3274 lnet_peer_cancel_discovery(lp);
3275 lnet_net_lock(LNET_LOCK_EX);
3277 lnet_net_unlock(LNET_LOCK_EX);
3279 /* Queue cleanup 2: wait for the expired queue to clear. */
3280 while (!list_empty(&the_lnet.ln_dc_expired))
3281 schedule_timeout(cfs_time_seconds(1));
3283 /* Queue cleanup 3: clear the request queue. */
3284 lnet_net_lock(LNET_LOCK_EX);
3285 while (!list_empty(&the_lnet.ln_dc_request)) {
3286 lp = list_first_entry(&the_lnet.ln_dc_request,
3287 struct lnet_peer, lp_dc_list);
3288 lnet_peer_discovery_error(lp, -ESHUTDOWN);
3289 lnet_peer_discovery_complete(lp);
3291 lnet_net_unlock(LNET_LOCK_EX);
3293 LNetEQFree(the_lnet.ln_dc_eqh);
3294 LNetInvalidateEQHandle(&the_lnet.ln_dc_eqh);
3296 the_lnet.ln_dc_state = LNET_DC_STATE_SHUTDOWN;
3297 wake_up(&the_lnet.ln_dc_waitq);
3299 CDEBUG(D_NET, "stopped\n");
3304 /* ln_api_mutex is held on entry. */
3305 int lnet_peer_discovery_start(void)
3307 struct task_struct *task;
3310 if (the_lnet.ln_dc_state != LNET_DC_STATE_SHUTDOWN)
3313 rc = LNetEQAlloc(0, lnet_discovery_event_handler, &the_lnet.ln_dc_eqh);
3315 CERROR("Can't allocate discovery EQ: %d\n", rc);
3319 the_lnet.ln_dc_state = LNET_DC_STATE_RUNNING;
3320 task = kthread_run(lnet_peer_discovery, NULL, "lnet_discovery");
3323 CERROR("Can't start peer discovery thread: %d\n", rc);
3325 LNetEQFree(the_lnet.ln_dc_eqh);
3326 LNetInvalidateEQHandle(&the_lnet.ln_dc_eqh);
3328 the_lnet.ln_dc_state = LNET_DC_STATE_SHUTDOWN;
3331 CDEBUG(D_NET, "discovery start: %d\n", rc);
3336 /* ln_api_mutex is held on entry. */
3337 void lnet_peer_discovery_stop(void)
3339 if (the_lnet.ln_dc_state == LNET_DC_STATE_SHUTDOWN)
3342 LASSERT(the_lnet.ln_dc_state == LNET_DC_STATE_RUNNING);
3343 the_lnet.ln_dc_state = LNET_DC_STATE_STOPPING;
3344 wake_up(&the_lnet.ln_dc_waitq);
3346 wait_event(the_lnet.ln_dc_waitq,
3347 the_lnet.ln_dc_state == LNET_DC_STATE_SHUTDOWN);
3349 LASSERT(list_empty(&the_lnet.ln_dc_request));
3350 LASSERT(list_empty(&the_lnet.ln_dc_working));
3351 LASSERT(list_empty(&the_lnet.ln_dc_expired));
3353 CDEBUG(D_NET, "discovery stopped\n");
3359 lnet_debug_peer(lnet_nid_t nid)
3361 char *aliveness = "NA";
3362 struct lnet_peer_ni *lp;
3365 cpt = lnet_cpt_of_nid(nid, NULL);
3368 lp = lnet_nid2peerni_locked(nid, LNET_NID_ANY, cpt);
3370 lnet_net_unlock(cpt);
3371 CDEBUG(D_WARNING, "No peer %s\n", libcfs_nid2str(nid));
3375 if (lnet_isrouter(lp) || lnet_peer_aliveness_enabled(lp))
3376 aliveness = (lnet_is_peer_ni_alive(lp)) ? "up" : "down";
3378 CDEBUG(D_WARNING, "%-24s %4d %5s %5d %5d %5d %5d %5d %ld\n",
3379 libcfs_nid2str(lp->lpni_nid), atomic_read(&lp->lpni_refcount),
3380 aliveness, lp->lpni_net->net_tunables.lct_peer_tx_credits,
3381 lp->lpni_rtrcredits, lp->lpni_minrtrcredits,
3382 lp->lpni_txcredits, lp->lpni_mintxcredits, lp->lpni_txqnob);
3384 lnet_peer_ni_decref_locked(lp);
3386 lnet_net_unlock(cpt);
3389 /* Gathering information for userspace. */
3391 int lnet_get_peer_ni_info(__u32 peer_index, __u64 *nid,
3392 char aliveness[LNET_MAX_STR_LEN],
3393 __u32 *cpt_iter, __u32 *refcount,
3394 __u32 *ni_peer_tx_credits, __u32 *peer_tx_credits,
3395 __u32 *peer_rtr_credits, __u32 *peer_min_rtr_credits,
3396 __u32 *peer_tx_qnob)
3398 struct lnet_peer_table *peer_table;
3399 struct lnet_peer_ni *lp;
3404 /* get the number of CPTs */
3405 lncpt = cfs_percpt_number(the_lnet.ln_peer_tables);
3407 /* if the cpt number to be examined is >= the number of cpts in
3408 * the system then indicate that there are no more cpts to examin
3410 if (*cpt_iter >= lncpt)
3413 /* get the current table */
3414 peer_table = the_lnet.ln_peer_tables[*cpt_iter];
3415 /* if the ptable is NULL then there are no more cpts to examine */
3416 if (peer_table == NULL)
3419 lnet_net_lock(*cpt_iter);
3421 for (j = 0; j < LNET_PEER_HASH_SIZE && !found; j++) {
3422 struct list_head *peers = &peer_table->pt_hash[j];
3424 list_for_each_entry(lp, peers, lpni_hashlist) {
3425 if (peer_index-- > 0)
3428 snprintf(aliveness, LNET_MAX_STR_LEN, "NA");
3429 if (lnet_isrouter(lp) ||
3430 lnet_peer_aliveness_enabled(lp))
3431 snprintf(aliveness, LNET_MAX_STR_LEN,
3432 lnet_is_peer_ni_alive(lp) ? "up" : "down");
3434 *nid = lp->lpni_nid;
3435 *refcount = atomic_read(&lp->lpni_refcount);
3436 *ni_peer_tx_credits =
3437 lp->lpni_net->net_tunables.lct_peer_tx_credits;
3438 *peer_tx_credits = lp->lpni_txcredits;
3439 *peer_rtr_credits = lp->lpni_rtrcredits;
3440 *peer_min_rtr_credits = lp->lpni_mintxcredits;
3441 *peer_tx_qnob = lp->lpni_txqnob;
3447 lnet_net_unlock(*cpt_iter);
3451 return found ? 0 : -ENOENT;
3454 /* ln_api_mutex is held, which keeps the peer list stable */
3455 int lnet_get_peer_info(struct lnet_ioctl_peer_cfg *cfg, void __user *bulk)
3457 struct lnet_ioctl_element_stats *lpni_stats;
3458 struct lnet_ioctl_element_msg_stats *lpni_msg_stats;
3459 struct lnet_ioctl_peer_ni_hstats *lpni_hstats;
3460 struct lnet_peer_ni_credit_info *lpni_info;
3461 struct lnet_peer_ni *lpni;
3462 struct lnet_peer *lp;
3467 lp = lnet_find_peer(cfg->prcfg_prim_nid);
3474 size = sizeof(nid) + sizeof(*lpni_info) + sizeof(*lpni_stats)
3475 + sizeof(*lpni_msg_stats) + sizeof(*lpni_hstats);
3476 size *= lp->lp_nnis;
3477 if (size > cfg->prcfg_size) {
3478 cfg->prcfg_size = size;
3483 cfg->prcfg_prim_nid = lp->lp_primary_nid;
3484 cfg->prcfg_mr = lnet_peer_is_multi_rail(lp);
3485 cfg->prcfg_cfg_nid = lp->lp_primary_nid;
3486 cfg->prcfg_count = lp->lp_nnis;
3487 cfg->prcfg_size = size;
3488 cfg->prcfg_state = lp->lp_state;
3490 /* Allocate helper buffers. */
3492 LIBCFS_ALLOC(lpni_info, sizeof(*lpni_info));
3495 LIBCFS_ALLOC(lpni_stats, sizeof(*lpni_stats));
3498 LIBCFS_ALLOC(lpni_msg_stats, sizeof(*lpni_msg_stats));
3499 if (!lpni_msg_stats)
3500 goto out_free_stats;
3501 LIBCFS_ALLOC(lpni_hstats, sizeof(*lpni_hstats));
3503 goto out_free_msg_stats;
3508 while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL) {
3509 nid = lpni->lpni_nid;
3510 if (copy_to_user(bulk, &nid, sizeof(nid)))
3511 goto out_free_hstats;
3512 bulk += sizeof(nid);
3514 memset(lpni_info, 0, sizeof(*lpni_info));
3515 snprintf(lpni_info->cr_aliveness, LNET_MAX_STR_LEN, "NA");
3516 if (lnet_isrouter(lpni) ||
3517 lnet_peer_aliveness_enabled(lpni))
3518 snprintf(lpni_info->cr_aliveness, LNET_MAX_STR_LEN,
3519 lnet_is_peer_ni_alive(lpni) ? "up" : "down");
3521 lpni_info->cr_refcount = atomic_read(&lpni->lpni_refcount);
3522 lpni_info->cr_ni_peer_tx_credits = (lpni->lpni_net != NULL) ?
3523 lpni->lpni_net->net_tunables.lct_peer_tx_credits : 0;
3524 lpni_info->cr_peer_tx_credits = lpni->lpni_txcredits;
3525 lpni_info->cr_peer_rtr_credits = lpni->lpni_rtrcredits;
3526 lpni_info->cr_peer_min_rtr_credits = lpni->lpni_minrtrcredits;
3527 lpni_info->cr_peer_min_tx_credits = lpni->lpni_mintxcredits;
3528 lpni_info->cr_peer_tx_qnob = lpni->lpni_txqnob;
3529 if (copy_to_user(bulk, lpni_info, sizeof(*lpni_info)))
3530 goto out_free_hstats;
3531 bulk += sizeof(*lpni_info);
3533 memset(lpni_stats, 0, sizeof(*lpni_stats));
3534 lpni_stats->iel_send_count = lnet_sum_stats(&lpni->lpni_stats,
3535 LNET_STATS_TYPE_SEND);
3536 lpni_stats->iel_recv_count = lnet_sum_stats(&lpni->lpni_stats,
3537 LNET_STATS_TYPE_RECV);
3538 lpni_stats->iel_drop_count = lnet_sum_stats(&lpni->lpni_stats,
3539 LNET_STATS_TYPE_DROP);
3540 if (copy_to_user(bulk, lpni_stats, sizeof(*lpni_stats)))
3541 goto out_free_hstats;
3542 bulk += sizeof(*lpni_stats);
3543 lnet_usr_translate_stats(lpni_msg_stats, &lpni->lpni_stats);
3544 if (copy_to_user(bulk, lpni_msg_stats, sizeof(*lpni_msg_stats)))
3545 goto out_free_hstats;
3546 bulk += sizeof(*lpni_msg_stats);
3547 lpni_hstats->hlpni_network_timeout =
3548 atomic_read(&lpni->lpni_hstats.hlt_network_timeout);
3549 lpni_hstats->hlpni_remote_dropped =
3550 atomic_read(&lpni->lpni_hstats.hlt_remote_dropped);
3551 lpni_hstats->hlpni_remote_timeout =
3552 atomic_read(&lpni->lpni_hstats.hlt_remote_timeout);
3553 lpni_hstats->hlpni_remote_error =
3554 atomic_read(&lpni->lpni_hstats.hlt_remote_error);
3555 lpni_hstats->hlpni_health_value =
3556 atomic_read(&lpni->lpni_healthv);
3557 if (copy_to_user(bulk, lpni_hstats, sizeof(*lpni_hstats)))
3558 goto out_free_hstats;
3559 bulk += sizeof(*lpni_hstats);
3564 LIBCFS_FREE(lpni_hstats, sizeof(*lpni_hstats));
3566 LIBCFS_FREE(lpni_msg_stats, sizeof(*lpni_msg_stats));
3568 LIBCFS_FREE(lpni_stats, sizeof(*lpni_stats));
3570 LIBCFS_FREE(lpni_info, sizeof(*lpni_info));
3572 lnet_peer_decref_locked(lp);
3578 lnet_peer_ni_add_to_recoveryq_locked(struct lnet_peer_ni *lpni)
3580 /* the mt could've shutdown and cleaned up the queues */
3581 if (the_lnet.ln_mt_state != LNET_MT_STATE_RUNNING)
3584 if (list_empty(&lpni->lpni_recovery) &&
3585 atomic_read(&lpni->lpni_healthv) < LNET_MAX_HEALTH_VALUE) {
3586 CERROR("lpni %s added to recovery queue. Health = %d\n",
3587 libcfs_nid2str(lpni->lpni_nid),
3588 atomic_read(&lpni->lpni_healthv));
3589 list_add_tail(&lpni->lpni_recovery, &the_lnet.ln_mt_peerNIRecovq);
3590 lnet_peer_ni_addref_locked(lpni);
3594 /* Call with the ln_api_mutex held */
3596 lnet_peer_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3598 struct lnet_peer_table *ptable;
3599 struct lnet_peer *lp;
3600 struct lnet_peer_net *lpn;
3601 struct lnet_peer_ni *lpni;
3605 if (the_lnet.ln_state != LNET_STATE_RUNNING)
3609 lnet_net_lock(LNET_LOCK_EX);
3610 lpni = lnet_find_peer_ni_locked(nid);
3612 lnet_net_unlock(LNET_LOCK_EX);
3615 atomic_set(&lpni->lpni_healthv, value);
3616 lnet_peer_ni_add_to_recoveryq_locked(lpni);
3617 lnet_peer_ni_decref_locked(lpni);
3618 lnet_net_unlock(LNET_LOCK_EX);
3622 lncpt = cfs_percpt_number(the_lnet.ln_peer_tables);
3625 * Walk all the peers and reset the healhv for each one to the
3628 lnet_net_lock(LNET_LOCK_EX);
3629 for (cpt = 0; cpt < lncpt; cpt++) {
3630 ptable = the_lnet.ln_peer_tables[cpt];
3631 list_for_each_entry(lp, &ptable->pt_peer_list, lp_peer_list) {
3632 list_for_each_entry(lpn, &lp->lp_peer_nets, lpn_peer_nets) {
3633 list_for_each_entry(lpni, &lpn->lpn_peer_nis,
3635 atomic_set(&lpni->lpni_healthv, value);
3636 lnet_peer_ni_add_to_recoveryq_locked(lpni);
3641 lnet_net_unlock(LNET_LOCK_EX);