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 this 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_table_cleanup_locked(struct lnet_net *net,
343 struct lnet_peer_table *ptable)
346 struct lnet_peer_ni *next;
347 struct lnet_peer_ni *lpni;
348 struct lnet_peer *peer;
350 for (i = 0; i < LNET_PEER_HASH_SIZE; i++) {
351 list_for_each_entry_safe(lpni, next, &ptable->pt_hash[i],
353 if (net != NULL && net != lpni->lpni_net)
356 peer = lpni->lpni_peer_net->lpn_peer;
357 if (peer->lp_primary_nid != lpni->lpni_nid) {
358 lnet_peer_ni_del_locked(lpni);
362 * Removing the primary NID implies removing
363 * the entire peer. Advance next beyond any
364 * peer_ni that belongs to the same peer.
366 list_for_each_entry_from(next, &ptable->pt_hash[i],
368 if (next->lpni_peer_net->lpn_peer != peer)
371 lnet_peer_del_locked(peer);
377 lnet_peer_ni_finalize_wait(struct lnet_peer_table *ptable)
381 spin_lock(&ptable->pt_zombie_lock);
382 while (ptable->pt_zombies) {
383 spin_unlock(&ptable->pt_zombie_lock);
385 if (is_power_of_2(i)) {
387 "Waiting for %d zombies on peer table\n",
390 set_current_state(TASK_UNINTERRUPTIBLE);
391 schedule_timeout(cfs_time_seconds(1) >> 1);
392 spin_lock(&ptable->pt_zombie_lock);
394 spin_unlock(&ptable->pt_zombie_lock);
398 lnet_peer_table_del_rtrs_locked(struct lnet_net *net,
399 struct lnet_peer_table *ptable)
401 struct lnet_peer_ni *lp;
402 struct lnet_peer_ni *tmp;
406 for (i = 0; i < LNET_PEER_HASH_SIZE; i++) {
407 list_for_each_entry_safe(lp, tmp, &ptable->pt_hash[i],
409 if (net != lp->lpni_net)
412 if (lp->lpni_rtr_refcount == 0)
415 lpni_nid = lp->lpni_nid;
417 lnet_net_unlock(LNET_LOCK_EX);
418 lnet_del_route(LNET_NIDNET(LNET_NID_ANY), lpni_nid);
419 lnet_net_lock(LNET_LOCK_EX);
425 lnet_peer_tables_cleanup(struct lnet_net *net)
428 struct lnet_peer_table *ptable;
430 LASSERT(the_lnet.ln_state != LNET_STATE_SHUTDOWN || net != NULL);
431 /* If just deleting the peers for a NI, get rid of any routes these
432 * peers are gateways for. */
433 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
434 lnet_net_lock(LNET_LOCK_EX);
435 lnet_peer_table_del_rtrs_locked(net, ptable);
436 lnet_net_unlock(LNET_LOCK_EX);
439 /* Start the cleanup process */
440 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
441 lnet_net_lock(LNET_LOCK_EX);
442 lnet_peer_table_cleanup_locked(net, ptable);
443 lnet_net_unlock(LNET_LOCK_EX);
446 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables)
447 lnet_peer_ni_finalize_wait(ptable);
450 static struct lnet_peer_ni *
451 lnet_get_peer_ni_locked(struct lnet_peer_table *ptable, lnet_nid_t nid)
453 struct list_head *peers;
454 struct lnet_peer_ni *lp;
456 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
458 peers = &ptable->pt_hash[lnet_nid2peerhash(nid)];
459 list_for_each_entry(lp, peers, lpni_hashlist) {
460 if (lp->lpni_nid == nid) {
461 lnet_peer_ni_addref_locked(lp);
469 struct lnet_peer_ni *
470 lnet_find_peer_ni_locked(lnet_nid_t nid)
472 struct lnet_peer_ni *lpni;
473 struct lnet_peer_table *ptable;
476 cpt = lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
478 ptable = the_lnet.ln_peer_tables[cpt];
479 lpni = lnet_get_peer_ni_locked(ptable, nid);
485 lnet_find_or_create_peer_locked(lnet_nid_t dst_nid, int cpt)
487 struct lnet_peer_ni *lpni;
488 struct lnet_peer *lp;
490 lpni = lnet_find_peer_ni_locked(dst_nid);
492 lpni = lnet_nid2peerni_locked(dst_nid, cpt);
494 return ERR_CAST(lpni);
497 lp = lpni->lpni_peer_net->lpn_peer;
498 lnet_peer_ni_decref_locked(lpni);
503 struct lnet_peer_ni *
504 lnet_get_peer_ni_idx_locked(int idx, struct lnet_peer_net **lpn,
505 struct lnet_peer **lp)
507 struct lnet_peer_ni *lpni;
509 list_for_each_entry((*lp), &the_lnet.ln_peers, lp_on_lnet_peer_list) {
510 list_for_each_entry((*lpn), &((*lp)->lp_peer_nets), lpn_on_peer_list) {
511 list_for_each_entry(lpni, &((*lpn)->lpn_peer_nis),
512 lpni_on_peer_net_list)
521 struct lnet_peer_ni *
522 lnet_get_next_peer_ni_locked(struct lnet_peer *peer,
523 struct lnet_peer_net *peer_net,
524 struct lnet_peer_ni *prev)
526 struct lnet_peer_ni *lpni;
527 struct lnet_peer_net *net = peer_net;
531 net = list_entry(peer->lp_peer_nets.next,
532 struct lnet_peer_net,
534 lpni = list_entry(net->lpn_peer_nis.next, struct lnet_peer_ni,
535 lpni_on_peer_net_list);
540 if (prev->lpni_on_peer_net_list.next ==
541 &prev->lpni_peer_net->lpn_peer_nis) {
543 * if you reached the end of the peer ni list and the peer
544 * net is specified then there are no more peer nis in that
551 * we reached the end of this net ni list. move to the
554 if (prev->lpni_peer_net->lpn_on_peer_list.next ==
556 /* no more nets and no more NIs. */
559 /* get the next net */
560 net = list_entry(prev->lpni_peer_net->lpn_on_peer_list.next,
561 struct lnet_peer_net,
563 /* get the ni on it */
564 lpni = list_entry(net->lpn_peer_nis.next, struct lnet_peer_ni,
565 lpni_on_peer_net_list);
570 /* there are more nis left */
571 lpni = list_entry(prev->lpni_on_peer_net_list.next,
572 struct lnet_peer_ni, lpni_on_peer_net_list);
578 lnet_peer_is_ni_pref_locked(struct lnet_peer_ni *lpni, struct lnet_ni *ni)
582 for (i = 0; i < lpni->lpni_pref_nnids; i++) {
583 if (lpni->lpni_pref_nids[i] == ni->ni_nid)
590 lnet_peer_primary_nid_locked(lnet_nid_t nid)
592 struct lnet_peer_ni *lpni;
593 lnet_nid_t primary_nid = nid;
595 lpni = lnet_find_peer_ni_locked(nid);
597 primary_nid = lpni->lpni_peer_net->lpn_peer->lp_primary_nid;
598 lnet_peer_ni_decref_locked(lpni);
605 LNetPrimaryNID(lnet_nid_t nid)
607 struct lnet_peer_ni *lpni;
608 lnet_nid_t primary_nid = nid;
612 cpt = lnet_net_lock_current();
613 lpni = lnet_nid2peerni_locked(nid, cpt);
618 primary_nid = lpni->lpni_peer_net->lpn_peer->lp_primary_nid;
619 lnet_peer_ni_decref_locked(lpni);
621 lnet_net_unlock(cpt);
623 CDEBUG(D_NET, "NID %s primary NID %s rc %d\n", libcfs_nid2str(nid),
624 libcfs_nid2str(primary_nid), rc);
627 EXPORT_SYMBOL(LNetPrimaryNID);
629 struct lnet_peer_net *
630 lnet_peer_get_net_locked(struct lnet_peer *peer, __u32 net_id)
632 struct lnet_peer_net *peer_net;
633 list_for_each_entry(peer_net, &peer->lp_peer_nets, lpn_on_peer_list) {
634 if (peer_net->lpn_net_id == net_id)
641 lnet_peer_setup_hierarchy(struct lnet_peer *lp, struct lnet_peer_ni *lpni,
644 struct lnet_peer_net *lpn = NULL;
645 struct lnet_peer_table *ptable;
646 __u32 net_id = LNET_NIDNET(nid);
649 * Create the peer_ni, peer_net, and peer if they don't exist
653 lpn = lnet_peer_get_net_locked(lp, net_id);
655 lp = lnet_peer_alloc(nid);
661 lpn = lnet_peer_net_alloc(net_id);
663 goto out_maybe_free_lp;
667 lpni = lnet_peer_ni_alloc(nid);
669 goto out_maybe_free_lpn;
672 /* Install the new peer_ni */
673 lnet_net_lock(LNET_LOCK_EX);
674 /* Add peer_ni to global peer table hash, if necessary. */
675 if (list_empty(&lpni->lpni_hashlist)) {
676 ptable = the_lnet.ln_peer_tables[lpni->lpni_cpt];
677 list_add_tail(&lpni->lpni_hashlist,
678 &ptable->pt_hash[lnet_nid2peerhash(nid)]);
679 ptable->pt_version++;
680 atomic_inc(&ptable->pt_number);
681 atomic_inc(&lpni->lpni_refcount);
684 /* Detach the peer_ni from an existing peer, if necessary. */
685 if (lpni->lpni_peer_net && lpni->lpni_peer_net->lpn_peer != lp)
686 lnet_try_destroy_peer_hierarchy_locked(lpni);
688 /* Add peer_ni to peer_net */
689 lpni->lpni_peer_net = lpn;
690 list_add_tail(&lpni->lpni_on_peer_net_list, &lpn->lpn_peer_nis);
692 /* Add peer_net to peer */
693 if (!lpn->lpn_peer) {
695 list_add_tail(&lpn->lpn_on_peer_list, &lp->lp_peer_nets);
698 /* Add peer to global peer list */
699 if (list_empty(&lp->lp_on_lnet_peer_list))
700 list_add_tail(&lp->lp_on_lnet_peer_list, &the_lnet.ln_peers);
701 lnet_net_unlock(LNET_LOCK_EX);
706 if (list_empty(&lpn->lpn_on_peer_list))
707 LIBCFS_FREE(lpn, sizeof(*lpn));
709 if (list_empty(&lp->lp_on_lnet_peer_list))
710 LIBCFS_FREE(lp, sizeof(*lp));
716 lnet_add_prim_lpni(lnet_nid_t nid)
719 struct lnet_peer *peer;
720 struct lnet_peer_ni *lpni;
722 LASSERT(nid != LNET_NID_ANY);
725 * lookup the NID and its peer
726 * if the peer doesn't exist, create it.
727 * if this is a non-MR peer then change its state to MR and exit.
728 * if this is an MR peer and it's a primary NI: NO-OP.
729 * if this is an MR peer and it's not a primary NI. Operation not
732 * The adding and deleting of peer nis is being serialized through
733 * the api_mutex. So we can look up peers with the mutex locked
734 * safely. Only when we need to change the ptable, do we need to
735 * exclusively lock the lnet_net_lock()
737 lpni = lnet_find_peer_ni_locked(nid);
739 rc = lnet_peer_setup_hierarchy(NULL, NULL, nid);
742 lpni = lnet_find_peer_ni_locked(nid);
747 lnet_peer_ni_decref_locked(lpni);
749 peer = lpni->lpni_peer_net->lpn_peer;
752 * If we found a lpni with the same nid as the NID we're trying to
753 * create, then we're trying to create an already existing lpni
754 * that belongs to a different peer
756 if (peer->lp_primary_nid != nid)
760 * if we found an lpni that is not a multi-rail, which could occur
761 * if lpni is already created as a non-mr lpni or we just created
762 * it, then make sure you indicate that this lpni is a primary mr
765 * TODO: update flags if necessary
767 if (!peer->lp_multi_rail && peer->lp_primary_nid == nid)
768 peer->lp_multi_rail = true;
774 lnet_add_peer_ni_to_prim_lpni(lnet_nid_t prim_nid, lnet_nid_t nid)
776 struct lnet_peer *peer, *primary_peer;
777 struct lnet_peer_ni *lpni = NULL, *klpni = NULL;
779 LASSERT(prim_nid != LNET_NID_ANY && nid != LNET_NID_ANY);
782 * key nid must be created by this point. If not then this
783 * operation is not permitted
785 klpni = lnet_find_peer_ni_locked(prim_nid);
789 lnet_peer_ni_decref_locked(klpni);
791 primary_peer = klpni->lpni_peer_net->lpn_peer;
793 lpni = lnet_find_peer_ni_locked(nid);
795 lnet_peer_ni_decref_locked(lpni);
797 peer = lpni->lpni_peer_net->lpn_peer;
799 * lpni already exists in the system but it belongs to
800 * a different peer. We can't re-added it
802 if (peer->lp_primary_nid != prim_nid && peer->lp_multi_rail) {
803 CERROR("Cannot add NID %s owned by peer %s to peer %s\n",
804 libcfs_nid2str(lpni->lpni_nid),
805 libcfs_nid2str(peer->lp_primary_nid),
806 libcfs_nid2str(prim_nid));
808 } else if (peer->lp_primary_nid == prim_nid) {
810 * found a peer_ni that is already part of the
811 * peer. This is a no-op operation.
817 * TODO: else if (peer->lp_primary_nid != prim_nid &&
818 * !peer->lp_multi_rail)
819 * peer is not an MR peer and it will be moved in the next
820 * step to klpni, so update its flags accordingly.
821 * lnet_move_peer_ni()
825 * TODO: call lnet_update_peer() from here to update the
826 * flags. This is the case when the lpni you're trying to
827 * add is already part of the peer. This could've been
828 * added by the DD previously, so go ahead and do any
829 * updates to the state if necessary
835 * When we get here we either have found an existing lpni, which
836 * we can switch to the new peer. Or we need to create one and
837 * add it to the new peer
839 return lnet_peer_setup_hierarchy(primary_peer, lpni, nid);
843 * lpni creation initiated due to traffic either sending or receiving.
846 lnet_peer_ni_traffic_add(lnet_nid_t nid)
848 struct lnet_peer_ni *lpni;
851 if (nid == LNET_NID_ANY)
854 /* lnet_net_lock is not needed here because ln_api_lock is held */
855 lpni = lnet_find_peer_ni_locked(nid);
858 * TODO: lnet_update_primary_nid() but not all of it
859 * only indicate if we're converting this to MR capable
860 * Can happen due to DD
862 lnet_peer_ni_decref_locked(lpni);
864 rc = lnet_peer_setup_hierarchy(NULL, NULL, nid);
872 lnet_peer_ni_add_non_mr(lnet_nid_t nid)
874 struct lnet_peer_ni *lpni;
876 lpni = lnet_find_peer_ni_locked(nid);
878 CERROR("Cannot add %s as non-mr when it already exists\n",
879 libcfs_nid2str(nid));
880 lnet_peer_ni_decref_locked(lpni);
884 return lnet_peer_setup_hierarchy(NULL, NULL, nid);
888 * This API handles the following combinations:
889 * Create a primary NI if only the prim_nid is provided
890 * Create or add an lpni to a primary NI. Primary NI must've already
892 * Create a non-MR peer.
895 lnet_add_peer_ni_to_peer(lnet_nid_t prim_nid, lnet_nid_t nid, bool mr)
898 * Caller trying to setup an MR like peer hierarchy but
899 * specifying it to be non-MR. This is not allowed.
901 if (prim_nid != LNET_NID_ANY &&
902 nid != LNET_NID_ANY && !mr)
905 /* Add the primary NID of a peer */
906 if (prim_nid != LNET_NID_ANY &&
907 nid == LNET_NID_ANY && mr)
908 return lnet_add_prim_lpni(prim_nid);
910 /* Add a NID to an existing peer */
911 if (prim_nid != LNET_NID_ANY &&
912 nid != LNET_NID_ANY && mr)
913 return lnet_add_peer_ni_to_prim_lpni(prim_nid, nid);
915 /* Add a non-MR peer NI */
916 if (((prim_nid != LNET_NID_ANY &&
917 nid == LNET_NID_ANY) ||
918 (prim_nid == LNET_NID_ANY &&
919 nid != LNET_NID_ANY)) && !mr)
920 return lnet_peer_ni_add_non_mr(prim_nid != LNET_NID_ANY ?
927 lnet_del_peer_ni_from_peer(lnet_nid_t prim_nid, lnet_nid_t nid)
929 lnet_nid_t local_nid;
930 struct lnet_peer *peer;
931 struct lnet_peer_ni *lpni;
934 if (prim_nid == LNET_NID_ANY)
937 local_nid = (nid != LNET_NID_ANY) ? nid : prim_nid;
939 lpni = lnet_find_peer_ni_locked(local_nid);
942 lnet_peer_ni_decref_locked(lpni);
944 peer = lpni->lpni_peer_net->lpn_peer;
945 LASSERT(peer != NULL);
947 if (peer->lp_primary_nid == lpni->lpni_nid) {
949 * deleting the primary ni is equivalent to deleting the
952 lnet_net_lock(LNET_LOCK_EX);
953 rc = lnet_peer_del_locked(peer);
954 lnet_net_unlock(LNET_LOCK_EX);
959 lnet_net_lock(LNET_LOCK_EX);
960 rc = lnet_peer_ni_del_locked(lpni);
961 lnet_net_unlock(LNET_LOCK_EX);
967 lnet_destroy_peer_ni_locked(struct lnet_peer_ni *lpni)
969 struct lnet_peer_table *ptable;
971 LASSERT(atomic_read(&lpni->lpni_refcount) == 0);
972 LASSERT(lpni->lpni_rtr_refcount == 0);
973 LASSERT(list_empty(&lpni->lpni_txq));
974 LASSERT(lpni->lpni_txqnob == 0);
976 lpni->lpni_net = NULL;
978 /* remove the peer ni from the zombie list */
979 ptable = the_lnet.ln_peer_tables[lpni->lpni_cpt];
980 spin_lock(&ptable->pt_zombie_lock);
981 list_del_init(&lpni->lpni_hashlist);
982 ptable->pt_zombies--;
983 spin_unlock(&ptable->pt_zombie_lock);
985 LIBCFS_FREE(lpni, sizeof(*lpni));
988 struct lnet_peer_ni *
989 lnet_nid2peerni_ex(lnet_nid_t nid, int cpt)
991 struct lnet_peer_ni *lpni = NULL;
994 if (the_lnet.ln_state != LNET_STATE_RUNNING)
995 return ERR_PTR(-ESHUTDOWN);
998 * find if a peer_ni already exists.
999 * If so then just return that.
1001 lpni = lnet_find_peer_ni_locked(nid);
1005 lnet_net_unlock(cpt);
1007 rc = lnet_peer_ni_traffic_add(nid);
1010 goto out_net_relock;
1013 lpni = lnet_find_peer_ni_locked(nid);
1022 struct lnet_peer_ni *
1023 lnet_nid2peerni_locked(lnet_nid_t nid, int cpt)
1025 struct lnet_peer_ni *lpni = NULL;
1028 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1029 return ERR_PTR(-ESHUTDOWN);
1032 * find if a peer_ni already exists.
1033 * If so then just return that.
1035 lpni = lnet_find_peer_ni_locked(nid);
1041 * use the lnet_api_mutex to serialize the creation of the peer_ni
1042 * and the creation/deletion of the local ni/net. When a local ni is
1043 * created, if there exists a set of peer_nis on that network,
1044 * they need to be traversed and updated. When a local NI is
1045 * deleted, which could result in a network being deleted, then
1046 * all peer nis on that network need to be removed as well.
1048 * Creation through traffic should also be serialized with
1049 * creation through DLC.
1051 lnet_net_unlock(cpt);
1052 mutex_lock(&the_lnet.ln_api_mutex);
1054 * Shutdown is only set under the ln_api_lock, so a single
1055 * check here is sufficent.
1057 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
1058 lpni = ERR_PTR(-ESHUTDOWN);
1059 goto out_mutex_unlock;
1062 rc = lnet_peer_ni_traffic_add(nid);
1065 goto out_mutex_unlock;
1068 lpni = lnet_find_peer_ni_locked(nid);
1072 mutex_unlock(&the_lnet.ln_api_mutex);
1079 lnet_debug_peer(lnet_nid_t nid)
1081 char *aliveness = "NA";
1082 struct lnet_peer_ni *lp;
1085 cpt = lnet_cpt_of_nid(nid, NULL);
1088 lp = lnet_nid2peerni_locked(nid, cpt);
1090 lnet_net_unlock(cpt);
1091 CDEBUG(D_WARNING, "No peer %s\n", libcfs_nid2str(nid));
1095 if (lnet_isrouter(lp) || lnet_peer_aliveness_enabled(lp))
1096 aliveness = lp->lpni_alive ? "up" : "down";
1098 CDEBUG(D_WARNING, "%-24s %4d %5s %5d %5d %5d %5d %5d %ld\n",
1099 libcfs_nid2str(lp->lpni_nid), atomic_read(&lp->lpni_refcount),
1100 aliveness, lp->lpni_net->net_tunables.lct_peer_tx_credits,
1101 lp->lpni_rtrcredits, lp->lpni_minrtrcredits,
1102 lp->lpni_txcredits, lp->lpni_mintxcredits, lp->lpni_txqnob);
1104 lnet_peer_ni_decref_locked(lp);
1106 lnet_net_unlock(cpt);
1109 int lnet_get_peer_ni_info(__u32 peer_index, __u64 *nid,
1110 char aliveness[LNET_MAX_STR_LEN],
1111 __u32 *cpt_iter, __u32 *refcount,
1112 __u32 *ni_peer_tx_credits, __u32 *peer_tx_credits,
1113 __u32 *peer_rtr_credits, __u32 *peer_min_rtr_credits,
1114 __u32 *peer_tx_qnob)
1116 struct lnet_peer_table *peer_table;
1117 struct lnet_peer_ni *lp;
1122 /* get the number of CPTs */
1123 lncpt = cfs_percpt_number(the_lnet.ln_peer_tables);
1125 /* if the cpt number to be examined is >= the number of cpts in
1126 * the system then indicate that there are no more cpts to examin
1128 if (*cpt_iter >= lncpt)
1131 /* get the current table */
1132 peer_table = the_lnet.ln_peer_tables[*cpt_iter];
1133 /* if the ptable is NULL then there are no more cpts to examine */
1134 if (peer_table == NULL)
1137 lnet_net_lock(*cpt_iter);
1139 for (j = 0; j < LNET_PEER_HASH_SIZE && !found; j++) {
1140 struct list_head *peers = &peer_table->pt_hash[j];
1142 list_for_each_entry(lp, peers, lpni_hashlist) {
1143 if (peer_index-- > 0)
1146 snprintf(aliveness, LNET_MAX_STR_LEN, "NA");
1147 if (lnet_isrouter(lp) ||
1148 lnet_peer_aliveness_enabled(lp))
1149 snprintf(aliveness, LNET_MAX_STR_LEN,
1150 lp->lpni_alive ? "up" : "down");
1152 *nid = lp->lpni_nid;
1153 *refcount = atomic_read(&lp->lpni_refcount);
1154 *ni_peer_tx_credits =
1155 lp->lpni_net->net_tunables.lct_peer_tx_credits;
1156 *peer_tx_credits = lp->lpni_txcredits;
1157 *peer_rtr_credits = lp->lpni_rtrcredits;
1158 *peer_min_rtr_credits = lp->lpni_mintxcredits;
1159 *peer_tx_qnob = lp->lpni_txqnob;
1165 lnet_net_unlock(*cpt_iter);
1169 return found ? 0 : -ENOENT;
1172 int lnet_get_peer_info(__u32 idx, lnet_nid_t *primary_nid, lnet_nid_t *nid,
1174 struct lnet_peer_ni_credit_info __user *peer_ni_info,
1175 struct lnet_ioctl_element_stats __user *peer_ni_stats)
1177 struct lnet_peer_ni *lpni = NULL;
1178 struct lnet_peer_net *lpn = NULL;
1179 struct lnet_peer *lp = NULL;
1180 struct lnet_peer_ni_credit_info ni_info;
1181 struct lnet_ioctl_element_stats ni_stats;
1184 lpni = lnet_get_peer_ni_idx_locked(idx, &lpn, &lp);
1189 *primary_nid = lp->lp_primary_nid;
1190 *mr = lp->lp_multi_rail;
1191 *nid = lpni->lpni_nid;
1192 snprintf(ni_info.cr_aliveness, LNET_MAX_STR_LEN, "NA");
1193 if (lnet_isrouter(lpni) ||
1194 lnet_peer_aliveness_enabled(lpni))
1195 snprintf(ni_info.cr_aliveness, LNET_MAX_STR_LEN,
1196 lpni->lpni_alive ? "up" : "down");
1198 ni_info.cr_refcount = atomic_read(&lpni->lpni_refcount);
1199 ni_info.cr_ni_peer_tx_credits = (lpni->lpni_net != NULL) ?
1200 lpni->lpni_net->net_tunables.lct_peer_tx_credits : 0;
1201 ni_info.cr_peer_tx_credits = lpni->lpni_txcredits;
1202 ni_info.cr_peer_rtr_credits = lpni->lpni_rtrcredits;
1203 ni_info.cr_peer_min_rtr_credits = lpni->lpni_minrtrcredits;
1204 ni_info.cr_peer_min_tx_credits = lpni->lpni_mintxcredits;
1205 ni_info.cr_peer_tx_qnob = lpni->lpni_txqnob;
1206 ni_info.cr_ncpt = lpni->lpni_cpt;
1208 ni_stats.iel_send_count = atomic_read(&lpni->lpni_stats.send_count);
1209 ni_stats.iel_recv_count = atomic_read(&lpni->lpni_stats.recv_count);
1210 ni_stats.iel_drop_count = atomic_read(&lpni->lpni_stats.drop_count);
1212 /* If copy_to_user fails */
1214 if (copy_to_user(peer_ni_info, &ni_info, sizeof(ni_info)))
1217 if (copy_to_user(peer_ni_stats, &ni_stats, sizeof(ni_stats)))