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, 2014, 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 <lnet/lib-lnet.h>
38 #include <lnet/lib-dlc.h>
41 lnet_peer_remove_from_remote_list(struct lnet_peer_ni *lpni)
43 if (!list_empty(&lpni->lpni_on_remote_peer_ni_list)) {
44 list_del_init(&lpni->lpni_on_remote_peer_ni_list);
45 lnet_peer_ni_decref_locked(lpni);
50 lnet_peer_net_added(struct lnet_net *net)
52 struct lnet_peer_ni *lpni, *tmp;
54 list_for_each_entry_safe(lpni, tmp, &the_lnet.ln_remote_peer_ni_list,
55 lpni_on_remote_peer_ni_list) {
57 if (LNET_NIDNET(lpni->lpni_nid) == net->net_id) {
60 spin_lock(&lpni->lpni_lock);
61 lpni->lpni_txcredits =
62 lpni->lpni_net->net_tunables.lct_peer_tx_credits;
63 lpni->lpni_mintxcredits = lpni->lpni_txcredits;
64 lpni->lpni_rtrcredits =
65 lnet_peer_buffer_credits(lpni->lpni_net);
66 lpni->lpni_minrtrcredits = lpni->lpni_rtrcredits;
67 spin_unlock(&lpni->lpni_lock);
69 lnet_peer_remove_from_remote_list(lpni);
75 lnet_peer_tables_destroy(void)
77 struct lnet_peer_table *ptable;
78 struct list_head *hash;
82 if (!the_lnet.ln_peer_tables)
85 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
86 hash = ptable->pt_hash;
87 if (!hash) /* not intialized */
90 LASSERT(list_empty(&ptable->pt_zombie_list));
92 ptable->pt_hash = NULL;
93 for (j = 0; j < LNET_PEER_HASH_SIZE; j++)
94 LASSERT(list_empty(&hash[j]));
96 LIBCFS_FREE(hash, LNET_PEER_HASH_SIZE * sizeof(*hash));
99 cfs_percpt_free(the_lnet.ln_peer_tables);
100 the_lnet.ln_peer_tables = NULL;
104 lnet_peer_tables_create(void)
106 struct lnet_peer_table *ptable;
107 struct list_head *hash;
111 the_lnet.ln_peer_tables = cfs_percpt_alloc(lnet_cpt_table(),
113 if (the_lnet.ln_peer_tables == NULL) {
114 CERROR("Failed to allocate cpu-partition peer tables\n");
118 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
119 LIBCFS_CPT_ALLOC(hash, lnet_cpt_table(), i,
120 LNET_PEER_HASH_SIZE * sizeof(*hash));
122 CERROR("Failed to create peer hash table\n");
123 lnet_peer_tables_destroy();
127 spin_lock_init(&ptable->pt_zombie_lock);
128 INIT_LIST_HEAD(&ptable->pt_zombie_list);
130 for (j = 0; j < LNET_PEER_HASH_SIZE; j++)
131 INIT_LIST_HEAD(&hash[j]);
132 ptable->pt_hash = hash; /* sign of initialization */
138 static struct lnet_peer_ni *
139 lnet_peer_ni_alloc(lnet_nid_t nid)
141 struct lnet_peer_ni *lpni;
142 struct lnet_net *net;
145 cpt = lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
147 LIBCFS_CPT_ALLOC(lpni, lnet_cpt_table(), cpt, sizeof(*lpni));
151 INIT_LIST_HEAD(&lpni->lpni_txq);
152 INIT_LIST_HEAD(&lpni->lpni_rtrq);
153 INIT_LIST_HEAD(&lpni->lpni_routes);
154 INIT_LIST_HEAD(&lpni->lpni_hashlist);
155 INIT_LIST_HEAD(&lpni->lpni_on_peer_net_list);
156 INIT_LIST_HEAD(&lpni->lpni_on_remote_peer_ni_list);
158 spin_lock_init(&lpni->lpni_lock);
160 lpni->lpni_alive = !lnet_peers_start_down(); /* 1 bit!! */
161 lpni->lpni_last_alive = cfs_time_current(); /* assumes alive */
162 lpni->lpni_ping_feats = LNET_PING_FEAT_INVAL;
163 lpni->lpni_nid = nid;
164 lpni->lpni_cpt = cpt;
165 lnet_set_peer_ni_health_locked(lpni, true);
167 net = lnet_get_net_locked(LNET_NIDNET(nid));
168 lpni->lpni_net = net;
170 lpni->lpni_txcredits = net->net_tunables.lct_peer_tx_credits;
171 lpni->lpni_mintxcredits = lpni->lpni_txcredits;
172 lpni->lpni_rtrcredits = lnet_peer_buffer_credits(net);
173 lpni->lpni_minrtrcredits = lpni->lpni_rtrcredits;
176 * This peer_ni is not on a local network, so we
177 * cannot add the credits here. In case the net is
178 * added later, add the peer_ni to the remote peer ni
179 * list so it can be easily found and revisited.
181 /* FIXME: per-net implementation instead? */
182 atomic_inc(&lpni->lpni_refcount);
183 list_add_tail(&lpni->lpni_on_remote_peer_ni_list,
184 &the_lnet.ln_remote_peer_ni_list);
187 /* TODO: update flags */
192 static struct lnet_peer_net *
193 lnet_peer_net_alloc(__u32 net_id)
195 struct lnet_peer_net *lpn;
197 LIBCFS_CPT_ALLOC(lpn, lnet_cpt_table(), CFS_CPT_ANY, sizeof(*lpn));
201 INIT_LIST_HEAD(&lpn->lpn_on_peer_list);
202 INIT_LIST_HEAD(&lpn->lpn_peer_nis);
203 lpn->lpn_net_id = net_id;
208 static struct lnet_peer *
209 lnet_peer_alloc(lnet_nid_t nid)
211 struct lnet_peer *lp;
213 LIBCFS_CPT_ALLOC(lp, lnet_cpt_table(), CFS_CPT_ANY, sizeof(*lp));
217 INIT_LIST_HEAD(&lp->lp_on_lnet_peer_list);
218 INIT_LIST_HEAD(&lp->lp_peer_nets);
219 lp->lp_primary_nid = nid;
221 /* TODO: update flags */
228 lnet_try_destroy_peer_hierarchy_locked(struct lnet_peer_ni *lpni)
230 struct lnet_peer_net *peer_net;
231 struct lnet_peer *peer;
233 /* TODO: could the below situation happen? accessing an already
235 if (lpni->lpni_peer_net == NULL ||
236 lpni->lpni_peer_net->lpn_peer == NULL)
239 peer_net = lpni->lpni_peer_net;
240 peer = lpni->lpni_peer_net->lpn_peer;
242 list_del_init(&lpni->lpni_on_peer_net_list);
243 lpni->lpni_peer_net = NULL;
245 /* if peer_net is empty, then remove it from the peer */
246 if (list_empty(&peer_net->lpn_peer_nis)) {
247 list_del_init(&peer_net->lpn_on_peer_list);
248 peer_net->lpn_peer = NULL;
249 LIBCFS_FREE(peer_net, sizeof(*peer_net));
251 /* if the peer is empty then remove it from the
252 * the_lnet.ln_peers */
253 if (list_empty(&peer->lp_peer_nets)) {
254 list_del_init(&peer->lp_on_lnet_peer_list);
255 LIBCFS_FREE(peer, sizeof(*peer));
260 /* called with lnet_net_lock LNET_LOCK_EX held */
262 lnet_peer_ni_del_locked(struct lnet_peer_ni *lpni)
264 struct lnet_peer_table *ptable = NULL;
266 /* don't remove a peer_ni if it's also a gateway */
267 if (lpni->lpni_rtr_refcount > 0) {
268 CERROR("Peer NI %s is a gateway. Can not delete it\n",
269 libcfs_nid2str(lpni->lpni_nid));
273 lnet_peer_remove_from_remote_list(lpni);
275 /* remove peer ni from the hash list. */
276 list_del_init(&lpni->lpni_hashlist);
278 /* decrement the ref count on the peer table */
279 ptable = the_lnet.ln_peer_tables[lpni->lpni_cpt];
280 LASSERT(atomic_read(&ptable->pt_number) > 0);
281 atomic_dec(&ptable->pt_number);
284 * The peer_ni can no longer be found with a lookup. But there
285 * can be current users, so keep track of it on the zombie
286 * list until the reference count has gone to zero.
288 * The last reference may be lost in a place where the
289 * lnet_net_lock locks only a single cpt, and that cpt may not
290 * be lpni->lpni_cpt. So the zombie list of lnet_peer_table
293 spin_lock(&ptable->pt_zombie_lock);
294 list_add(&lpni->lpni_hashlist, &ptable->pt_zombie_list);
295 ptable->pt_zombies++;
296 spin_unlock(&ptable->pt_zombie_lock);
298 /* no need to keep this peer on the hierarchy anymore */
299 lnet_try_destroy_peer_hierarchy_locked(lpni);
301 /* decrement reference on peer */
302 lnet_peer_ni_decref_locked(lpni);
307 void lnet_peer_uninit(void)
309 struct lnet_peer_ni *lpni, *tmp;
311 lnet_net_lock(LNET_LOCK_EX);
313 /* remove all peer_nis from the remote peer and the hash list */
314 list_for_each_entry_safe(lpni, tmp, &the_lnet.ln_remote_peer_ni_list,
315 lpni_on_remote_peer_ni_list)
316 lnet_peer_ni_del_locked(lpni);
318 lnet_peer_tables_destroy();
320 lnet_net_unlock(LNET_LOCK_EX);
324 lnet_peer_del_locked(struct lnet_peer *peer)
326 struct lnet_peer_ni *lpni = NULL, *lpni2;
329 lpni = lnet_get_next_peer_ni_locked(peer, NULL, lpni);
330 while (lpni != NULL) {
331 lpni2 = lnet_get_next_peer_ni_locked(peer, NULL, lpni);
332 rc = lnet_peer_ni_del_locked(lpni);
342 lnet_peer_del(struct lnet_peer *peer)
344 lnet_net_lock(LNET_LOCK_EX);
345 lnet_peer_del_locked(peer);
346 lnet_net_unlock(LNET_LOCK_EX);
352 * Delete a NID from a peer.
353 * Implements a few sanity checks.
354 * Call with ln_api_mutex held.
357 lnet_peer_del_nid(struct lnet_peer *lp, lnet_nid_t nid)
359 struct lnet_peer *lp2;
360 struct lnet_peer_ni *lpni;
362 lpni = lnet_find_peer_ni_locked(nid);
364 CERROR("Cannot remove unknown nid %s from peer %s\n",
366 libcfs_nid2str(lp->lp_primary_nid));
369 lnet_peer_ni_decref_locked(lpni);
370 lp2 = lpni->lpni_peer_net->lpn_peer;
372 CERROR("Nid %s is attached to peer %s, not peer %s\n",
374 libcfs_nid2str(lp2->lp_primary_nid),
375 libcfs_nid2str(lp->lp_primary_nid));
380 * This function only allows deletion of the primary NID if it
383 if (nid == lp->lp_primary_nid && lnet_get_num_peer_nis(lp) != 1) {
384 CERROR("Cannot delete primary NID %s from multi-NID peer\n",
385 libcfs_nid2str(nid));
389 lnet_net_lock(LNET_LOCK_EX);
390 lnet_peer_ni_del_locked(lpni);
391 lnet_net_unlock(LNET_LOCK_EX);
397 lnet_peer_table_cleanup_locked(struct lnet_net *net,
398 struct lnet_peer_table *ptable)
401 struct lnet_peer_ni *next;
402 struct lnet_peer_ni *lpni;
403 struct lnet_peer *peer;
405 for (i = 0; i < LNET_PEER_HASH_SIZE; i++) {
406 list_for_each_entry_safe(lpni, next, &ptable->pt_hash[i],
408 if (net != NULL && net != lpni->lpni_net)
411 peer = lpni->lpni_peer_net->lpn_peer;
412 if (peer->lp_primary_nid != lpni->lpni_nid) {
413 lnet_peer_ni_del_locked(lpni);
417 * Removing the primary NID implies removing
418 * the entire peer. Advance next beyond any
419 * peer_ni that belongs to the same peer.
421 list_for_each_entry_from(next, &ptable->pt_hash[i],
423 if (next->lpni_peer_net->lpn_peer != peer)
426 lnet_peer_del_locked(peer);
432 lnet_peer_ni_finalize_wait(struct lnet_peer_table *ptable)
436 spin_lock(&ptable->pt_zombie_lock);
437 while (ptable->pt_zombies) {
438 spin_unlock(&ptable->pt_zombie_lock);
442 "Waiting for %d zombies on peer table\n",
445 set_current_state(TASK_UNINTERRUPTIBLE);
446 schedule_timeout(cfs_time_seconds(1) >> 1);
447 spin_lock(&ptable->pt_zombie_lock);
449 spin_unlock(&ptable->pt_zombie_lock);
453 lnet_peer_table_del_rtrs_locked(struct lnet_net *net,
454 struct lnet_peer_table *ptable)
456 struct lnet_peer_ni *lp;
457 struct lnet_peer_ni *tmp;
461 for (i = 0; i < LNET_PEER_HASH_SIZE; i++) {
462 list_for_each_entry_safe(lp, tmp, &ptable->pt_hash[i],
464 if (net != lp->lpni_net)
467 if (lp->lpni_rtr_refcount == 0)
470 lpni_nid = lp->lpni_nid;
472 lnet_net_unlock(LNET_LOCK_EX);
473 lnet_del_route(LNET_NIDNET(LNET_NID_ANY), lpni_nid);
474 lnet_net_lock(LNET_LOCK_EX);
480 lnet_peer_tables_cleanup(struct lnet_net *net)
483 struct lnet_peer_table *ptable;
485 LASSERT(the_lnet.ln_state != LNET_STATE_SHUTDOWN || net != NULL);
486 /* If just deleting the peers for a NI, get rid of any routes these
487 * peers are gateways for. */
488 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
489 lnet_net_lock(LNET_LOCK_EX);
490 lnet_peer_table_del_rtrs_locked(net, ptable);
491 lnet_net_unlock(LNET_LOCK_EX);
494 /* Start the cleanup process */
495 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
496 lnet_net_lock(LNET_LOCK_EX);
497 lnet_peer_table_cleanup_locked(net, ptable);
498 lnet_net_unlock(LNET_LOCK_EX);
501 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables)
502 lnet_peer_ni_finalize_wait(ptable);
505 static struct lnet_peer_ni *
506 lnet_get_peer_ni_locked(struct lnet_peer_table *ptable, lnet_nid_t nid)
508 struct list_head *peers;
509 struct lnet_peer_ni *lp;
511 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
513 peers = &ptable->pt_hash[lnet_nid2peerhash(nid)];
514 list_for_each_entry(lp, peers, lpni_hashlist) {
515 if (lp->lpni_nid == nid) {
516 lnet_peer_ni_addref_locked(lp);
524 struct lnet_peer_ni *
525 lnet_find_peer_ni_locked(lnet_nid_t nid)
527 struct lnet_peer_ni *lpni;
528 struct lnet_peer_table *ptable;
531 cpt = lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
533 ptable = the_lnet.ln_peer_tables[cpt];
534 lpni = lnet_get_peer_ni_locked(ptable, nid);
540 lnet_find_or_create_peer_locked(lnet_nid_t dst_nid, int cpt)
542 struct lnet_peer_ni *lpni;
543 struct lnet_peer *lp;
545 lpni = lnet_find_peer_ni_locked(dst_nid);
547 lpni = lnet_nid2peerni_locked(dst_nid, cpt);
549 return ERR_CAST(lpni);
552 lp = lpni->lpni_peer_net->lpn_peer;
553 lnet_peer_ni_decref_locked(lpni);
558 struct lnet_peer_ni *
559 lnet_get_peer_ni_idx_locked(int idx, struct lnet_peer_net **lpn,
560 struct lnet_peer **lp)
562 struct lnet_peer_ni *lpni;
564 list_for_each_entry((*lp), &the_lnet.ln_peers, lp_on_lnet_peer_list) {
565 list_for_each_entry((*lpn), &((*lp)->lp_peer_nets), lpn_on_peer_list) {
566 list_for_each_entry(lpni, &((*lpn)->lpn_peer_nis),
567 lpni_on_peer_net_list)
576 struct lnet_peer_ni *
577 lnet_get_next_peer_ni_locked(struct lnet_peer *peer,
578 struct lnet_peer_net *peer_net,
579 struct lnet_peer_ni *prev)
581 struct lnet_peer_ni *lpni;
582 struct lnet_peer_net *net = peer_net;
586 net = list_entry(peer->lp_peer_nets.next,
587 struct lnet_peer_net,
589 lpni = list_entry(net->lpn_peer_nis.next, struct lnet_peer_ni,
590 lpni_on_peer_net_list);
595 if (prev->lpni_on_peer_net_list.next ==
596 &prev->lpni_peer_net->lpn_peer_nis) {
598 * if you reached the end of the peer ni list and the peer
599 * net is specified then there are no more peer nis in that
606 * we reached the end of this net ni list. move to the
609 if (prev->lpni_peer_net->lpn_on_peer_list.next ==
611 /* no more nets and no more NIs. */
614 /* get the next net */
615 net = list_entry(prev->lpni_peer_net->lpn_on_peer_list.next,
616 struct lnet_peer_net,
618 /* get the ni on it */
619 lpni = list_entry(net->lpn_peer_nis.next, struct lnet_peer_ni,
620 lpni_on_peer_net_list);
625 /* there are more nis left */
626 lpni = list_entry(prev->lpni_on_peer_net_list.next,
627 struct lnet_peer_ni, lpni_on_peer_net_list);
633 lnet_peer_is_ni_pref_locked(struct lnet_peer_ni *lpni, struct lnet_ni *ni)
637 for (i = 0; i < lpni->lpni_pref_nnids; i++) {
638 if (lpni->lpni_pref_nids[i] == ni->ni_nid)
645 lnet_peer_primary_nid(lnet_nid_t nid)
647 struct lnet_peer_ni *lpni;
648 lnet_nid_t primary_nid = nid;
651 cpt = lnet_net_lock_current();
652 lpni = lnet_find_peer_ni_locked(nid);
654 primary_nid = lpni->lpni_peer_net->lpn_peer->lp_primary_nid;
655 lnet_peer_ni_decref_locked(lpni);
657 lnet_net_unlock(cpt);
663 LNetPrimaryNID(lnet_nid_t nid)
665 struct lnet_peer_ni *lpni;
666 lnet_nid_t primary_nid = nid;
670 cpt = lnet_net_lock_current();
671 lpni = lnet_nid2peerni_locked(nid, cpt);
676 primary_nid = lpni->lpni_peer_net->lpn_peer->lp_primary_nid;
677 lnet_peer_ni_decref_locked(lpni);
679 lnet_net_unlock(cpt);
681 CDEBUG(D_NET, "NID %s primary NID %s rc %d\n", libcfs_nid2str(nid),
682 libcfs_nid2str(primary_nid), rc);
685 EXPORT_SYMBOL(LNetPrimaryNID);
687 struct lnet_peer_net *
688 lnet_peer_get_net_locked(struct lnet_peer *peer, __u32 net_id)
690 struct lnet_peer_net *peer_net;
691 list_for_each_entry(peer_net, &peer->lp_peer_nets, lpn_on_peer_list) {
692 if (peer_net->lpn_net_id == net_id)
699 lnet_peer_setup_hierarchy(struct lnet_peer *lp, struct lnet_peer_ni *lpni,
702 struct lnet_peer_net *lpn = NULL;
703 struct lnet_peer_table *ptable;
704 __u32 net_id = LNET_NIDNET(nid);
707 * Create the peer_ni, peer_net, and peer if they don't exist
711 lpn = lnet_peer_get_net_locked(lp, net_id);
713 lp = lnet_peer_alloc(nid);
719 lpn = lnet_peer_net_alloc(net_id);
721 goto out_maybe_free_lp;
725 lpni = lnet_peer_ni_alloc(nid);
727 goto out_maybe_free_lpn;
730 /* Install the new peer_ni */
731 lnet_net_lock(LNET_LOCK_EX);
732 /* Add peer_ni to global peer table hash, if necessary. */
733 if (list_empty(&lpni->lpni_hashlist)) {
734 ptable = the_lnet.ln_peer_tables[lpni->lpni_cpt];
735 list_add_tail(&lpni->lpni_hashlist,
736 &ptable->pt_hash[lnet_nid2peerhash(nid)]);
737 ptable->pt_version++;
738 atomic_inc(&ptable->pt_number);
739 atomic_inc(&lpni->lpni_refcount);
742 /* Detach the peer_ni from an existing peer, if necessary. */
743 if (lpni->lpni_peer_net && lpni->lpni_peer_net->lpn_peer != lp)
744 lnet_try_destroy_peer_hierarchy_locked(lpni);
746 /* Add peer_ni to peer_net */
747 lpni->lpni_peer_net = lpn;
748 list_add_tail(&lpni->lpni_on_peer_net_list, &lpn->lpn_peer_nis);
750 /* Add peer_net to peer */
751 if (!lpn->lpn_peer) {
753 list_add_tail(&lpn->lpn_on_peer_list, &lp->lp_peer_nets);
756 /* Add peer to global peer list */
757 if (list_empty(&lp->lp_on_lnet_peer_list))
758 list_add_tail(&lp->lp_on_lnet_peer_list, &the_lnet.ln_peers);
759 lnet_net_unlock(LNET_LOCK_EX);
764 if (list_empty(&lpn->lpn_on_peer_list))
765 LIBCFS_FREE(lpn, sizeof(*lpn));
767 if (list_empty(&lp->lp_on_lnet_peer_list))
768 LIBCFS_FREE(lp, sizeof(*lp));
774 lnet_add_prim_lpni(lnet_nid_t nid)
777 struct lnet_peer *peer;
778 struct lnet_peer_ni *lpni;
780 LASSERT(nid != LNET_NID_ANY);
783 * lookup the NID and its peer
784 * if the peer doesn't exist, create it.
785 * if this is a non-MR peer then change its state to MR and exit.
786 * if this is an MR peer and it's a primary NI: NO-OP.
787 * if this is an MR peer and it's not a primary NI. Operation not
790 * The adding and deleting of peer nis is being serialized through
791 * the api_mutex. So we can look up peers with the mutex locked
792 * safely. Only when we need to change the ptable, do we need to
793 * exclusively lock the lnet_net_lock()
795 lpni = lnet_find_peer_ni_locked(nid);
797 rc = lnet_peer_setup_hierarchy(NULL, NULL, nid);
800 lpni = lnet_find_peer_ni_locked(nid);
805 lnet_peer_ni_decref_locked(lpni);
807 peer = lpni->lpni_peer_net->lpn_peer;
810 * If we found a lpni with the same nid as the NID we're trying to
811 * create, then we're trying to create an already existing lpni
812 * that belongs to a different peer
814 if (peer->lp_primary_nid != nid)
818 * if we found an lpni that is not a multi-rail, which could occur
819 * if lpni is already created as a non-mr lpni or we just created
820 * it, then make sure you indicate that this lpni is a primary mr
823 * TODO: update flags if necessary
825 if (!peer->lp_multi_rail && peer->lp_primary_nid == nid)
826 peer->lp_multi_rail = true;
832 lnet_add_peer_ni_to_prim_lpni(lnet_nid_t prim_nid, lnet_nid_t nid)
834 struct lnet_peer *peer, *primary_peer;
835 struct lnet_peer_ni *lpni = NULL, *klpni = NULL;
837 LASSERT(prim_nid != LNET_NID_ANY && nid != LNET_NID_ANY);
840 * key nid must be created by this point. If not then this
841 * operation is not permitted
843 klpni = lnet_find_peer_ni_locked(prim_nid);
847 lnet_peer_ni_decref_locked(klpni);
849 primary_peer = klpni->lpni_peer_net->lpn_peer;
851 lpni = lnet_find_peer_ni_locked(nid);
853 lnet_peer_ni_decref_locked(lpni);
855 peer = lpni->lpni_peer_net->lpn_peer;
857 * lpni already exists in the system but it belongs to
858 * a different peer. We can't re-added it
860 if (peer->lp_primary_nid != prim_nid && peer->lp_multi_rail) {
861 CERROR("Cannot add NID %s owned by peer %s to peer %s\n",
862 libcfs_nid2str(lpni->lpni_nid),
863 libcfs_nid2str(peer->lp_primary_nid),
864 libcfs_nid2str(prim_nid));
866 } else if (peer->lp_primary_nid == prim_nid) {
868 * found a peer_ni that is already part of the
869 * peer. This is a no-op operation.
875 * TODO: else if (peer->lp_primary_nid != prim_nid &&
876 * !peer->lp_multi_rail)
877 * peer is not an MR peer and it will be moved in the next
878 * step to klpni, so update its flags accordingly.
879 * lnet_move_peer_ni()
883 * TODO: call lnet_update_peer() from here to update the
884 * flags. This is the case when the lpni you're trying to
885 * add is already part of the peer. This could've been
886 * added by the DD previously, so go ahead and do any
887 * updates to the state if necessary
893 * When we get here we either have found an existing lpni, which
894 * we can switch to the new peer. Or we need to create one and
895 * add it to the new peer
897 return lnet_peer_setup_hierarchy(primary_peer, lpni, nid);
901 * lpni creation initiated due to traffic either sending or receiving.
904 lnet_peer_ni_traffic_add(lnet_nid_t nid)
906 struct lnet_peer_ni *lpni;
909 if (nid == LNET_NID_ANY)
912 /* lnet_net_lock is not needed here because ln_api_lock is held */
913 lpni = lnet_find_peer_ni_locked(nid);
916 * TODO: lnet_update_primary_nid() but not all of it
917 * only indicate if we're converting this to MR capable
918 * Can happen due to DD
920 lnet_peer_ni_decref_locked(lpni);
922 rc = lnet_peer_setup_hierarchy(NULL, NULL, nid);
930 lnet_peer_ni_add_non_mr(lnet_nid_t nid)
932 struct lnet_peer_ni *lpni;
934 lpni = lnet_find_peer_ni_locked(nid);
936 CERROR("Cannot add %s as non-mr when it already exists\n",
937 libcfs_nid2str(nid));
938 lnet_peer_ni_decref_locked(lpni);
942 return lnet_peer_setup_hierarchy(NULL, NULL, nid);
946 * Implementation of IOC_LIBCFS_ADD_PEER_NI.
948 * This API handles the following combinations:
949 * Create a primary NI if only the prim_nid is provided
950 * Create or add an lpni to a primary NI. Primary NI must've already
952 * Create a non-MR peer.
955 lnet_add_peer_ni(lnet_nid_t prim_nid, lnet_nid_t nid, bool mr)
958 * Caller trying to setup an MR like peer hierarchy but
959 * specifying it to be non-MR. This is not allowed.
961 if (prim_nid != LNET_NID_ANY &&
962 nid != LNET_NID_ANY && !mr)
965 /* Add the primary NID of a peer */
966 if (prim_nid != LNET_NID_ANY &&
967 nid == LNET_NID_ANY && mr)
968 return lnet_add_prim_lpni(prim_nid);
970 /* Add a NID to an existing peer */
971 if (prim_nid != LNET_NID_ANY &&
972 nid != LNET_NID_ANY && mr)
973 return lnet_add_peer_ni_to_prim_lpni(prim_nid, nid);
975 /* Add a non-MR peer NI */
976 if (((prim_nid != LNET_NID_ANY &&
977 nid == LNET_NID_ANY) ||
978 (prim_nid == LNET_NID_ANY &&
979 nid != LNET_NID_ANY)) && !mr)
980 return lnet_peer_ni_add_non_mr(prim_nid != LNET_NID_ANY ?
987 * Implementation of IOC_LIBCFS_DEL_PEER_NI.
989 * This API handles the following combinations:
990 * Delete a NI from a peer if both prim_nid and nid are provided.
991 * Delete a peer if only prim_nid is provided.
992 * Delete a peer if its primary nid is provided.
994 * The caller must hold ln_api_mutex. This prevents the peer from
995 * being modified/deleted by a different thread.
998 lnet_del_peer_ni(lnet_nid_t prim_nid, lnet_nid_t nid)
1000 struct lnet_peer *lp;
1001 struct lnet_peer_ni *lpni;
1003 if (prim_nid == LNET_NID_ANY)
1006 lpni = lnet_find_peer_ni_locked(prim_nid);
1009 lnet_peer_ni_decref_locked(lpni);
1010 lp = lpni->lpni_peer_net->lpn_peer;
1012 if (prim_nid != lp->lp_primary_nid) {
1013 CDEBUG(D_NET, "prim_nid %s is not primary for peer %s\n",
1014 libcfs_nid2str(prim_nid),
1015 libcfs_nid2str(lp->lp_primary_nid));
1019 if (nid == LNET_NID_ANY || nid == lp->lp_primary_nid)
1020 return lnet_peer_del(lp);
1022 return lnet_peer_del_nid(lp, nid);
1026 lnet_destroy_peer_ni_locked(struct lnet_peer_ni *lpni)
1028 struct lnet_peer_table *ptable;
1030 LASSERT(atomic_read(&lpni->lpni_refcount) == 0);
1031 LASSERT(lpni->lpni_rtr_refcount == 0);
1032 LASSERT(list_empty(&lpni->lpni_txq));
1033 LASSERT(lpni->lpni_txqnob == 0);
1035 lpni->lpni_net = NULL;
1037 /* remove the peer ni from the zombie list */
1038 ptable = the_lnet.ln_peer_tables[lpni->lpni_cpt];
1039 spin_lock(&ptable->pt_zombie_lock);
1040 list_del_init(&lpni->lpni_hashlist);
1041 ptable->pt_zombies--;
1042 spin_unlock(&ptable->pt_zombie_lock);
1044 LIBCFS_FREE(lpni, sizeof(*lpni));
1047 struct lnet_peer_ni *
1048 lnet_nid2peerni_ex(lnet_nid_t nid, int cpt)
1050 struct lnet_peer_ni *lpni = NULL;
1053 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1054 return ERR_PTR(-ESHUTDOWN);
1057 * find if a peer_ni already exists.
1058 * If so then just return that.
1060 lpni = lnet_find_peer_ni_locked(nid);
1064 lnet_net_unlock(cpt);
1066 rc = lnet_peer_ni_traffic_add(nid);
1069 goto out_net_relock;
1072 lpni = lnet_find_peer_ni_locked(nid);
1081 struct lnet_peer_ni *
1082 lnet_nid2peerni_locked(lnet_nid_t nid, int cpt)
1084 struct lnet_peer_ni *lpni = NULL;
1087 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1088 return ERR_PTR(-ESHUTDOWN);
1091 * find if a peer_ni already exists.
1092 * If so then just return that.
1094 lpni = lnet_find_peer_ni_locked(nid);
1100 * use the lnet_api_mutex to serialize the creation of the peer_ni
1101 * and the creation/deletion of the local ni/net. When a local ni is
1102 * created, if there exists a set of peer_nis on that network,
1103 * they need to be traversed and updated. When a local NI is
1104 * deleted, which could result in a network being deleted, then
1105 * all peer nis on that network need to be removed as well.
1107 * Creation through traffic should also be serialized with
1108 * creation through DLC.
1110 lnet_net_unlock(cpt);
1111 mutex_lock(&the_lnet.ln_api_mutex);
1113 * Shutdown is only set under the ln_api_lock, so a single
1114 * check here is sufficent.
1116 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
1117 lpni = ERR_PTR(-ESHUTDOWN);
1118 goto out_mutex_unlock;
1121 rc = lnet_peer_ni_traffic_add(nid);
1124 goto out_mutex_unlock;
1127 lpni = lnet_find_peer_ni_locked(nid);
1131 mutex_unlock(&the_lnet.ln_api_mutex);
1138 lnet_debug_peer(lnet_nid_t nid)
1140 char *aliveness = "NA";
1141 struct lnet_peer_ni *lp;
1144 cpt = lnet_cpt_of_nid(nid, NULL);
1147 lp = lnet_nid2peerni_locked(nid, cpt);
1149 lnet_net_unlock(cpt);
1150 CDEBUG(D_WARNING, "No peer %s\n", libcfs_nid2str(nid));
1154 if (lnet_isrouter(lp) || lnet_peer_aliveness_enabled(lp))
1155 aliveness = lp->lpni_alive ? "up" : "down";
1157 CDEBUG(D_WARNING, "%-24s %4d %5s %5d %5d %5d %5d %5d %ld\n",
1158 libcfs_nid2str(lp->lpni_nid), atomic_read(&lp->lpni_refcount),
1159 aliveness, lp->lpni_net->net_tunables.lct_peer_tx_credits,
1160 lp->lpni_rtrcredits, lp->lpni_minrtrcredits,
1161 lp->lpni_txcredits, lp->lpni_mintxcredits, lp->lpni_txqnob);
1163 lnet_peer_ni_decref_locked(lp);
1165 lnet_net_unlock(cpt);
1168 int lnet_get_peer_ni_info(__u32 peer_index, __u64 *nid,
1169 char aliveness[LNET_MAX_STR_LEN],
1170 __u32 *cpt_iter, __u32 *refcount,
1171 __u32 *ni_peer_tx_credits, __u32 *peer_tx_credits,
1172 __u32 *peer_rtr_credits, __u32 *peer_min_rtr_credits,
1173 __u32 *peer_tx_qnob)
1175 struct lnet_peer_table *peer_table;
1176 struct lnet_peer_ni *lp;
1181 /* get the number of CPTs */
1182 lncpt = cfs_percpt_number(the_lnet.ln_peer_tables);
1184 /* if the cpt number to be examined is >= the number of cpts in
1185 * the system then indicate that there are no more cpts to examin
1187 if (*cpt_iter >= lncpt)
1190 /* get the current table */
1191 peer_table = the_lnet.ln_peer_tables[*cpt_iter];
1192 /* if the ptable is NULL then there are no more cpts to examine */
1193 if (peer_table == NULL)
1196 lnet_net_lock(*cpt_iter);
1198 for (j = 0; j < LNET_PEER_HASH_SIZE && !found; j++) {
1199 struct list_head *peers = &peer_table->pt_hash[j];
1201 list_for_each_entry(lp, peers, lpni_hashlist) {
1202 if (peer_index-- > 0)
1205 snprintf(aliveness, LNET_MAX_STR_LEN, "NA");
1206 if (lnet_isrouter(lp) ||
1207 lnet_peer_aliveness_enabled(lp))
1208 snprintf(aliveness, LNET_MAX_STR_LEN,
1209 lp->lpni_alive ? "up" : "down");
1211 *nid = lp->lpni_nid;
1212 *refcount = atomic_read(&lp->lpni_refcount);
1213 *ni_peer_tx_credits =
1214 lp->lpni_net->net_tunables.lct_peer_tx_credits;
1215 *peer_tx_credits = lp->lpni_txcredits;
1216 *peer_rtr_credits = lp->lpni_rtrcredits;
1217 *peer_min_rtr_credits = lp->lpni_mintxcredits;
1218 *peer_tx_qnob = lp->lpni_txqnob;
1224 lnet_net_unlock(*cpt_iter);
1228 return found ? 0 : -ENOENT;
1231 int lnet_get_peer_info(__u32 idx, lnet_nid_t *primary_nid, lnet_nid_t *nid,
1232 bool *mr, struct lnet_peer_ni_credit_info *peer_ni_info,
1233 struct lnet_ioctl_element_stats *peer_ni_stats)
1235 struct lnet_peer_ni *lpni = NULL;
1236 struct lnet_peer_net *lpn = NULL;
1237 struct lnet_peer *lp = NULL;
1239 lpni = lnet_get_peer_ni_idx_locked(idx, &lpn, &lp);
1244 *primary_nid = lp->lp_primary_nid;
1245 *mr = lp->lp_multi_rail;
1246 *nid = lpni->lpni_nid;
1247 snprintf(peer_ni_info->cr_aliveness, LNET_MAX_STR_LEN, "NA");
1248 if (lnet_isrouter(lpni) ||
1249 lnet_peer_aliveness_enabled(lpni))
1250 snprintf(peer_ni_info->cr_aliveness, LNET_MAX_STR_LEN,
1251 lpni->lpni_alive ? "up" : "down");
1253 peer_ni_info->cr_refcount = atomic_read(&lpni->lpni_refcount);
1254 peer_ni_info->cr_ni_peer_tx_credits = (lpni->lpni_net != NULL) ?
1255 lpni->lpni_net->net_tunables.lct_peer_tx_credits : 0;
1256 peer_ni_info->cr_peer_tx_credits = lpni->lpni_txcredits;
1257 peer_ni_info->cr_peer_rtr_credits = lpni->lpni_rtrcredits;
1258 peer_ni_info->cr_peer_min_rtr_credits = lpni->lpni_minrtrcredits;
1259 peer_ni_info->cr_peer_min_tx_credits = lpni->lpni_mintxcredits;
1260 peer_ni_info->cr_peer_tx_qnob = lpni->lpni_txqnob;
1262 peer_ni_stats->send_count = atomic_read(&lpni->lpni_stats.send_count);
1263 peer_ni_stats->recv_count = atomic_read(&lpni->lpni_stats.recv_count);
1264 peer_ni_stats->drop_count = atomic_read(&lpni->lpni_stats.drop_count);