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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LNET
38 #include "../../include/linux/lnet/lib-lnet.h"
39 #include <linux/log2.h>
40 #include <linux/ktime.h>
42 #define D_LNI D_CONSOLE
44 lnet_t the_lnet; /* THE state of the network */
45 EXPORT_SYMBOL(the_lnet);
48 static char *ip2nets = "";
49 module_param(ip2nets, charp, 0444);
50 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
52 static char *networks = "";
53 module_param(networks, charp, 0444);
54 MODULE_PARM_DESC(networks, "local networks");
56 static char *routes = "";
57 module_param(routes, charp, 0444);
58 MODULE_PARM_DESC(routes, "routes to non-local networks");
60 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
61 module_param(rnet_htable_size, int, 0444);
62 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
71 lnet_get_networks(void)
76 if (*networks != 0 && *ip2nets != 0) {
77 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or 'ip2nets' but not both at once\n");
82 rc = lnet_parse_ip2nets(&nets, ip2nets);
83 return (rc == 0) ? nets : NULL;
95 spin_lock_init(&the_lnet.ln_eq_wait_lock);
96 init_waitqueue_head(&the_lnet.ln_eq_waitq);
97 mutex_init(&the_lnet.ln_lnd_mutex);
98 mutex_init(&the_lnet.ln_api_mutex);
102 lnet_fini_locks(void)
108 lnet_create_remote_nets_table(void)
111 struct list_head *hash;
113 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
114 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
115 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
117 CERROR("Failed to create remote nets hash table\n");
121 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
122 INIT_LIST_HEAD(&hash[i]);
123 the_lnet.ln_remote_nets_hash = hash;
128 lnet_destroy_remote_nets_table(void)
132 if (the_lnet.ln_remote_nets_hash == NULL)
135 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
136 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
138 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
139 LNET_REMOTE_NETS_HASH_SIZE *
140 sizeof(the_lnet.ln_remote_nets_hash[0]));
141 the_lnet.ln_remote_nets_hash = NULL;
145 lnet_destroy_locks(void)
147 if (the_lnet.ln_res_lock != NULL) {
148 cfs_percpt_lock_free(the_lnet.ln_res_lock);
149 the_lnet.ln_res_lock = NULL;
152 if (the_lnet.ln_net_lock != NULL) {
153 cfs_percpt_lock_free(the_lnet.ln_net_lock);
154 the_lnet.ln_net_lock = NULL;
161 lnet_create_locks(void)
165 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
166 if (the_lnet.ln_res_lock == NULL)
169 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
170 if (the_lnet.ln_net_lock == NULL)
176 lnet_destroy_locks();
180 static void lnet_assert_wire_constants(void)
182 /* Wire protocol assertions generated by 'wirecheck'
183 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
184 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
185 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
188 CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
189 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
190 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
191 CLASSERT(LNET_MSG_ACK == 0);
192 CLASSERT(LNET_MSG_PUT == 1);
193 CLASSERT(LNET_MSG_GET == 2);
194 CLASSERT(LNET_MSG_REPLY == 3);
195 CLASSERT(LNET_MSG_HELLO == 4);
197 /* Checks for struct ptl_handle_wire_t */
198 CLASSERT((int)sizeof(lnet_handle_wire_t) == 16);
199 CLASSERT((int)offsetof(lnet_handle_wire_t, wh_interface_cookie) == 0);
200 CLASSERT((int)sizeof(((lnet_handle_wire_t *)0)->wh_interface_cookie) == 8);
201 CLASSERT((int)offsetof(lnet_handle_wire_t, wh_object_cookie) == 8);
202 CLASSERT((int)sizeof(((lnet_handle_wire_t *)0)->wh_object_cookie) == 8);
204 /* Checks for struct lnet_magicversion_t */
205 CLASSERT((int)sizeof(lnet_magicversion_t) == 8);
206 CLASSERT((int)offsetof(lnet_magicversion_t, magic) == 0);
207 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->magic) == 4);
208 CLASSERT((int)offsetof(lnet_magicversion_t, version_major) == 4);
209 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_major) == 2);
210 CLASSERT((int)offsetof(lnet_magicversion_t, version_minor) == 6);
211 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_minor) == 2);
213 /* Checks for struct lnet_hdr_t */
214 CLASSERT((int)sizeof(lnet_hdr_t) == 72);
215 CLASSERT((int)offsetof(lnet_hdr_t, dest_nid) == 0);
216 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_nid) == 8);
217 CLASSERT((int)offsetof(lnet_hdr_t, src_nid) == 8);
218 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_nid) == 8);
219 CLASSERT((int)offsetof(lnet_hdr_t, dest_pid) == 16);
220 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_pid) == 4);
221 CLASSERT((int)offsetof(lnet_hdr_t, src_pid) == 20);
222 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_pid) == 4);
223 CLASSERT((int)offsetof(lnet_hdr_t, type) == 24);
224 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->type) == 4);
225 CLASSERT((int)offsetof(lnet_hdr_t, payload_length) == 28);
226 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->payload_length) == 4);
227 CLASSERT((int)offsetof(lnet_hdr_t, msg) == 32);
228 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg) == 40);
231 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.dst_wmd) == 32);
232 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.dst_wmd) == 16);
233 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.match_bits) == 48);
234 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.match_bits) == 8);
235 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.mlength) == 56);
236 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.mlength) == 4);
239 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ack_wmd) == 32);
240 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ack_wmd) == 16);
241 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.match_bits) == 48);
242 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.match_bits) == 8);
243 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.hdr_data) == 56);
244 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.hdr_data) == 8);
245 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ptl_index) == 64);
246 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ptl_index) == 4);
247 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.offset) == 68);
248 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.offset) == 4);
251 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.return_wmd) == 32);
252 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.return_wmd) == 16);
253 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.match_bits) == 48);
254 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.match_bits) == 8);
255 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.ptl_index) == 56);
256 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.ptl_index) == 4);
257 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.src_offset) == 60);
258 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.src_offset) == 4);
259 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.sink_length) == 64);
260 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.sink_length) == 4);
263 CLASSERT((int)offsetof(lnet_hdr_t, msg.reply.dst_wmd) == 32);
264 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.reply.dst_wmd) == 16);
267 CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.incarnation) == 32);
268 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.incarnation) == 8);
269 CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.type) == 40);
270 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.type) == 4);
274 lnet_find_lnd_by_type(int type)
277 struct list_head *tmp;
279 /* holding lnd mutex */
280 list_for_each(tmp, &the_lnet.ln_lnds) {
281 lnd = list_entry(tmp, lnd_t, lnd_list);
283 if ((int)lnd->lnd_type == type)
291 lnet_register_lnd(lnd_t *lnd)
293 LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
295 LASSERT(the_lnet.ln_init);
296 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
297 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
299 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
300 lnd->lnd_refcount = 0;
302 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
304 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
306 EXPORT_SYMBOL(lnet_register_lnd);
309 lnet_unregister_lnd(lnd_t *lnd)
311 LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
313 LASSERT(the_lnet.ln_init);
314 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
315 LASSERT(lnd->lnd_refcount == 0);
317 list_del(&lnd->lnd_list);
318 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
320 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
322 EXPORT_SYMBOL(lnet_unregister_lnd);
325 lnet_counters_get(lnet_counters_t *counters)
327 lnet_counters_t *ctr;
330 memset(counters, 0, sizeof(*counters));
332 lnet_net_lock(LNET_LOCK_EX);
334 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
335 counters->msgs_max += ctr->msgs_max;
336 counters->msgs_alloc += ctr->msgs_alloc;
337 counters->errors += ctr->errors;
338 counters->send_count += ctr->send_count;
339 counters->recv_count += ctr->recv_count;
340 counters->route_count += ctr->route_count;
341 counters->drop_count += ctr->drop_count;
342 counters->send_length += ctr->send_length;
343 counters->recv_length += ctr->recv_length;
344 counters->route_length += ctr->route_length;
345 counters->drop_length += ctr->drop_length;
348 lnet_net_unlock(LNET_LOCK_EX);
350 EXPORT_SYMBOL(lnet_counters_get);
353 lnet_counters_reset(void)
355 lnet_counters_t *counters;
358 lnet_net_lock(LNET_LOCK_EX);
360 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
361 memset(counters, 0, sizeof(lnet_counters_t));
363 lnet_net_unlock(LNET_LOCK_EX);
365 EXPORT_SYMBOL(lnet_counters_reset);
367 #ifdef LNET_USE_LIB_FREELIST
370 lnet_freelist_init(lnet_freelist_t *fl, int n, int size)
376 size += offsetof(lnet_freeobj_t, fo_contents);
378 LIBCFS_ALLOC(space, n * size);
382 INIT_LIST_HEAD(&fl->fl_list);
385 fl->fl_objsize = size;
388 memset(space, 0, size);
389 list_add((struct list_head *)space, &fl->fl_list);
397 lnet_freelist_fini(lnet_freelist_t *fl)
399 struct list_head *el;
402 if (fl->fl_nobjs == 0)
406 for (el = fl->fl_list.next; el != &fl->fl_list; el = el->next)
409 LASSERT(count == fl->fl_nobjs);
411 LIBCFS_FREE(fl->fl_objs, fl->fl_nobjs * fl->fl_objsize);
412 memset(fl, 0, sizeof(*fl));
415 #endif /* LNET_USE_LIB_FREELIST */
418 lnet_create_interface_cookie(void)
420 /* NB the interface cookie in wire handles guards against delayed
421 * replies and ACKs appearing valid after reboot.
423 return ktime_get_ns();
427 lnet_res_type2str(int type)
432 case LNET_COOKIE_TYPE_MD:
434 case LNET_COOKIE_TYPE_ME:
436 case LNET_COOKIE_TYPE_EQ:
442 lnet_res_container_cleanup(struct lnet_res_container *rec)
446 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
449 while (!list_empty(&rec->rec_active)) {
450 struct list_head *e = rec->rec_active.next;
453 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
454 lnet_eq_free(list_entry(e, lnet_eq_t, eq_list));
456 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
457 lnet_md_free(list_entry(e, lnet_libmd_t, md_list));
459 } else { /* NB: Active MEs should be attached on portals */
466 /* Found alive MD/ME/EQ, user really should unlink/free
467 * all of them before finalize LNet, but if someone didn't,
468 * we have to recycle garbage for him */
469 CERROR("%d active elements on exit of %s container\n",
470 count, lnet_res_type2str(rec->rec_type));
473 #ifdef LNET_USE_LIB_FREELIST
474 lnet_freelist_fini(&rec->rec_freelist);
476 if (rec->rec_lh_hash != NULL) {
477 LIBCFS_FREE(rec->rec_lh_hash,
478 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
479 rec->rec_lh_hash = NULL;
482 rec->rec_type = 0; /* mark it as finalized */
486 lnet_res_container_setup(struct lnet_res_container *rec,
487 int cpt, int type, int objnum, int objsz)
492 LASSERT(rec->rec_type == 0);
494 rec->rec_type = type;
495 INIT_LIST_HEAD(&rec->rec_active);
497 #ifdef LNET_USE_LIB_FREELIST
498 memset(&rec->rec_freelist, 0, sizeof(rec->rec_freelist));
499 rc = lnet_freelist_init(&rec->rec_freelist, objnum, objsz);
503 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
505 /* Arbitrary choice of hash table size */
506 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
507 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
508 if (rec->rec_lh_hash == NULL) {
513 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
514 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
519 CERROR("Failed to setup %s resource container\n",
520 lnet_res_type2str(type));
521 lnet_res_container_cleanup(rec);
526 lnet_res_containers_destroy(struct lnet_res_container **recs)
528 struct lnet_res_container *rec;
531 cfs_percpt_for_each(rec, i, recs)
532 lnet_res_container_cleanup(rec);
534 cfs_percpt_free(recs);
537 static struct lnet_res_container **
538 lnet_res_containers_create(int type, int objnum, int objsz)
540 struct lnet_res_container **recs;
541 struct lnet_res_container *rec;
545 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
547 CERROR("Failed to allocate %s resource containers\n",
548 lnet_res_type2str(type));
552 cfs_percpt_for_each(rec, i, recs) {
553 rc = lnet_res_container_setup(rec, i, type, objnum, objsz);
555 lnet_res_containers_destroy(recs);
564 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
566 /* ALWAYS called with lnet_res_lock held */
567 struct list_head *head;
568 lnet_libhandle_t *lh;
571 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
574 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
575 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
577 list_for_each_entry(lh, head, lh_hash_chain) {
578 if (lh->lh_cookie == cookie)
586 lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh)
588 /* ALWAYS called with lnet_res_lock held */
589 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
592 lh->lh_cookie = rec->rec_lh_cookie;
593 rec->rec_lh_cookie += 1 << ibits;
595 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
597 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
601 int lnet_unprepare(void);
604 lnet_prepare(lnet_pid_t requested_pid)
606 /* Prepare to bring up the network */
607 struct lnet_res_container **recs;
610 LASSERT(the_lnet.ln_refcount == 0);
612 the_lnet.ln_routing = 0;
614 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
615 the_lnet.ln_pid = requested_pid;
617 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
618 INIT_LIST_HEAD(&the_lnet.ln_nis);
619 INIT_LIST_HEAD(&the_lnet.ln_nis_cpt);
620 INIT_LIST_HEAD(&the_lnet.ln_nis_zombie);
621 INIT_LIST_HEAD(&the_lnet.ln_routers);
623 rc = lnet_create_remote_nets_table();
627 the_lnet.ln_interface_cookie = lnet_create_interface_cookie();
629 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
630 sizeof(lnet_counters_t));
631 if (the_lnet.ln_counters == NULL) {
632 CERROR("Failed to allocate counters for LNet\n");
637 rc = lnet_peer_tables_create();
641 rc = lnet_msg_containers_create();
645 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
646 LNET_COOKIE_TYPE_EQ, LNET_FL_MAX_EQS,
651 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME, LNET_FL_MAX_MES,
658 the_lnet.ln_me_containers = recs;
660 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD, LNET_FL_MAX_MDS,
661 sizeof(lnet_libmd_t));
667 the_lnet.ln_md_containers = recs;
669 rc = lnet_portals_create();
671 CERROR("Failed to create portals for LNet: %d\n", rc);
685 /* NB no LNET_LOCK since this is the last reference. All LND instances
686 * have shut down already, so it is safe to unlink and free all
687 * descriptors, even those that appear committed to a network op (eg MD
688 * with non-zero pending count) */
690 lnet_fail_nid(LNET_NID_ANY, 0);
692 LASSERT(the_lnet.ln_refcount == 0);
693 LASSERT(list_empty(&the_lnet.ln_test_peers));
694 LASSERT(list_empty(&the_lnet.ln_nis));
695 LASSERT(list_empty(&the_lnet.ln_nis_cpt));
696 LASSERT(list_empty(&the_lnet.ln_nis_zombie));
698 lnet_portals_destroy();
700 if (the_lnet.ln_md_containers != NULL) {
701 lnet_res_containers_destroy(the_lnet.ln_md_containers);
702 the_lnet.ln_md_containers = NULL;
705 if (the_lnet.ln_me_containers != NULL) {
706 lnet_res_containers_destroy(the_lnet.ln_me_containers);
707 the_lnet.ln_me_containers = NULL;
710 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
712 lnet_msg_containers_destroy();
713 lnet_peer_tables_destroy();
714 lnet_rtrpools_free();
716 if (the_lnet.ln_counters != NULL) {
717 cfs_percpt_free(the_lnet.ln_counters);
718 the_lnet.ln_counters = NULL;
720 lnet_destroy_remote_nets_table();
726 lnet_net2ni_locked(__u32 net, int cpt)
728 struct list_head *tmp;
731 LASSERT(cpt != LNET_LOCK_EX);
733 list_for_each(tmp, &the_lnet.ln_nis) {
734 ni = list_entry(tmp, lnet_ni_t, ni_list);
736 if (LNET_NIDNET(ni->ni_nid) == net) {
737 lnet_ni_addref_locked(ni, cpt);
746 lnet_net2ni(__u32 net)
751 ni = lnet_net2ni_locked(net, 0);
756 EXPORT_SYMBOL(lnet_net2ni);
759 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
764 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
769 val = hash_long(key, LNET_CPT_BITS);
770 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
774 return (unsigned int)(key + val + (val >> 1)) % number;
778 lnet_cpt_of_nid_locked(lnet_nid_t nid)
782 /* must called with hold of lnet_net_lock */
783 if (LNET_CPT_NUMBER == 1)
784 return 0; /* the only one */
786 /* take lnet_net_lock(any) would be OK */
787 if (!list_empty(&the_lnet.ln_nis_cpt)) {
788 list_for_each_entry(ni, &the_lnet.ln_nis_cpt, ni_cptlist) {
789 if (LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid))
792 LASSERT(ni->ni_cpts != NULL);
793 return ni->ni_cpts[lnet_nid_cpt_hash
794 (nid, ni->ni_ncpts)];
798 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
802 lnet_cpt_of_nid(lnet_nid_t nid)
807 if (LNET_CPT_NUMBER == 1)
808 return 0; /* the only one */
810 if (list_empty(&the_lnet.ln_nis_cpt))
811 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
813 cpt = lnet_net_lock_current();
814 cpt2 = lnet_cpt_of_nid_locked(nid);
815 lnet_net_unlock(cpt);
819 EXPORT_SYMBOL(lnet_cpt_of_nid);
822 lnet_islocalnet(__u32 net)
827 cpt = lnet_net_lock_current();
829 ni = lnet_net2ni_locked(net, cpt);
831 lnet_ni_decref_locked(ni, cpt);
833 lnet_net_unlock(cpt);
839 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
842 struct list_head *tmp;
844 LASSERT(cpt != LNET_LOCK_EX);
846 list_for_each(tmp, &the_lnet.ln_nis) {
847 ni = list_entry(tmp, lnet_ni_t, ni_list);
849 if (ni->ni_nid == nid) {
850 lnet_ni_addref_locked(ni, cpt);
859 lnet_islocalnid(lnet_nid_t nid)
864 cpt = lnet_net_lock_current();
865 ni = lnet_nid2ni_locked(nid, cpt);
867 lnet_ni_decref_locked(ni, cpt);
868 lnet_net_unlock(cpt);
874 lnet_count_acceptor_nis(void)
876 /* Return the # of NIs that need the acceptor. */
878 struct list_head *tmp;
882 cpt = lnet_net_lock_current();
883 list_for_each(tmp, &the_lnet.ln_nis) {
884 ni = list_entry(tmp, lnet_ni_t, ni_list);
886 if (ni->ni_lnd->lnd_accept != NULL)
890 lnet_net_unlock(cpt);
896 lnet_ni_tq_credits(lnet_ni_t *ni)
900 LASSERT(ni->ni_ncpts >= 1);
902 if (ni->ni_ncpts == 1)
903 return ni->ni_maxtxcredits;
905 credits = ni->ni_maxtxcredits / ni->ni_ncpts;
906 credits = max(credits, 8 * ni->ni_peertxcredits);
907 credits = min(credits, ni->ni_maxtxcredits);
913 lnet_shutdown_lndnis(void)
919 /* NB called holding the global mutex */
921 /* All quiet on the API front */
922 LASSERT(!the_lnet.ln_shutdown);
923 LASSERT(the_lnet.ln_refcount == 0);
924 LASSERT(list_empty(&the_lnet.ln_nis_zombie));
926 lnet_net_lock(LNET_LOCK_EX);
927 the_lnet.ln_shutdown = 1; /* flag shutdown */
929 /* Unlink NIs from the global table */
930 while (!list_empty(&the_lnet.ln_nis)) {
931 ni = list_entry(the_lnet.ln_nis.next,
933 /* move it to zombie list and nobody can find it anymore */
934 list_move(&ni->ni_list, &the_lnet.ln_nis_zombie);
935 lnet_ni_decref_locked(ni, 0); /* drop ln_nis' ref */
937 if (!list_empty(&ni->ni_cptlist)) {
938 list_del_init(&ni->ni_cptlist);
939 lnet_ni_decref_locked(ni, 0);
943 /* Drop the cached eqwait NI. */
944 if (the_lnet.ln_eq_waitni != NULL) {
945 lnet_ni_decref_locked(the_lnet.ln_eq_waitni, 0);
946 the_lnet.ln_eq_waitni = NULL;
949 /* Drop the cached loopback NI. */
950 if (the_lnet.ln_loni != NULL) {
951 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
952 the_lnet.ln_loni = NULL;
955 lnet_net_unlock(LNET_LOCK_EX);
957 /* Clear lazy portals and drop delayed messages which hold refs
958 * on their lnet_msg_t::msg_rxpeer */
959 for (i = 0; i < the_lnet.ln_nportals; i++)
960 LNetClearLazyPortal(i);
962 /* Clear the peer table and wait for all peers to go (they hold refs on
964 lnet_peer_tables_cleanup();
966 lnet_net_lock(LNET_LOCK_EX);
967 /* Now wait for the NI's I just nuked to show up on ln_zombie_nis
968 * and shut them down in guaranteed thread context */
970 while (!list_empty(&the_lnet.ln_nis_zombie)) {
974 ni = list_entry(the_lnet.ln_nis_zombie.next,
976 list_del_init(&ni->ni_list);
977 cfs_percpt_for_each(ref, j, ni->ni_refs) {
980 /* still busy, add it back to zombie list */
981 list_add(&ni->ni_list, &the_lnet.ln_nis_zombie);
985 if (!list_empty(&ni->ni_list)) {
986 lnet_net_unlock(LNET_LOCK_EX);
988 if ((i & (-i)) == i) {
989 CDEBUG(D_WARNING, "Waiting for zombie LNI %s\n",
990 libcfs_nid2str(ni->ni_nid));
992 set_current_state(TASK_UNINTERRUPTIBLE);
993 schedule_timeout(cfs_time_seconds(1));
994 lnet_net_lock(LNET_LOCK_EX);
998 ni->ni_lnd->lnd_refcount--;
999 lnet_net_unlock(LNET_LOCK_EX);
1001 islo = ni->ni_lnd->lnd_type == LOLND;
1003 LASSERT(!in_interrupt());
1004 (ni->ni_lnd->lnd_shutdown)(ni);
1006 /* can't deref lnd anymore now; it might have unregistered
1010 CDEBUG(D_LNI, "Removed LNI %s\n",
1011 libcfs_nid2str(ni->ni_nid));
1016 lnet_net_lock(LNET_LOCK_EX);
1019 the_lnet.ln_shutdown = 0;
1020 lnet_net_unlock(LNET_LOCK_EX);
1022 if (the_lnet.ln_network_tokens != NULL) {
1023 LIBCFS_FREE(the_lnet.ln_network_tokens,
1024 the_lnet.ln_network_tokens_nob);
1025 the_lnet.ln_network_tokens = NULL;
1030 lnet_startup_lndnis(void)
1034 struct lnet_tx_queue *tq;
1035 struct list_head nilist;
1040 char *nets = lnet_get_networks();
1042 INIT_LIST_HEAD(&nilist);
1047 rc = lnet_parse_networks(&nilist, nets);
1051 while (!list_empty(&nilist)) {
1052 ni = list_entry(nilist.next, lnet_ni_t, ni_list);
1053 lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));
1055 LASSERT(libcfs_isknown_lnd(lnd_type));
1057 if (lnd_type == CIBLND ||
1058 lnd_type == OPENIBLND ||
1059 lnd_type == IIBLND ||
1060 lnd_type == VIBLND) {
1061 CERROR("LND %s obsoleted\n",
1062 libcfs_lnd2str(lnd_type));
1066 LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
1067 lnd = lnet_find_lnd_by_type(lnd_type);
1070 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
1071 rc = request_module("%s",
1072 libcfs_lnd2modname(lnd_type));
1073 LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
1075 lnd = lnet_find_lnd_by_type(lnd_type);
1077 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
1078 CERROR("Can't load LND %s, module %s, rc=%d\n",
1079 libcfs_lnd2str(lnd_type),
1080 libcfs_lnd2modname(lnd_type), rc);
1085 lnet_net_lock(LNET_LOCK_EX);
1086 lnd->lnd_refcount++;
1087 lnet_net_unlock(LNET_LOCK_EX);
1091 rc = (lnd->lnd_startup)(ni);
1093 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
1096 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1097 rc, libcfs_lnd2str(lnd->lnd_type));
1098 lnet_net_lock(LNET_LOCK_EX);
1099 lnd->lnd_refcount--;
1100 lnet_net_unlock(LNET_LOCK_EX);
1104 LASSERT(ni->ni_peertimeout <= 0 || lnd->lnd_query != NULL);
1106 list_del(&ni->ni_list);
1108 lnet_net_lock(LNET_LOCK_EX);
1109 /* refcount for ln_nis */
1110 lnet_ni_addref_locked(ni, 0);
1111 list_add_tail(&ni->ni_list, &the_lnet.ln_nis);
1112 if (ni->ni_cpts != NULL) {
1113 list_add_tail(&ni->ni_cptlist,
1114 &the_lnet.ln_nis_cpt);
1115 lnet_ni_addref_locked(ni, 0);
1118 lnet_net_unlock(LNET_LOCK_EX);
1120 if (lnd->lnd_type == LOLND) {
1122 LASSERT(the_lnet.ln_loni == NULL);
1123 the_lnet.ln_loni = ni;
1127 if (ni->ni_peertxcredits == 0 ||
1128 ni->ni_maxtxcredits == 0) {
1129 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1130 libcfs_lnd2str(lnd->lnd_type),
1131 ni->ni_peertxcredits == 0 ?
1136 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
1137 tq->tq_credits_min =
1138 tq->tq_credits_max =
1139 tq->tq_credits = lnet_ni_tq_credits(ni);
1142 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
1143 libcfs_nid2str(ni->ni_nid), ni->ni_peertxcredits,
1144 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
1145 ni->ni_peerrtrcredits, ni->ni_peertimeout);
1150 if (the_lnet.ln_eq_waitni != NULL && nicount > 1) {
1151 lnd_type = the_lnet.ln_eq_waitni->ni_lnd->lnd_type;
1152 LCONSOLE_ERROR_MSG(0x109, "LND %s can only run single-network\n",
1153 libcfs_lnd2str(lnd_type));
1160 lnet_shutdown_lndnis();
1162 while (!list_empty(&nilist)) {
1163 ni = list_entry(nilist.next, lnet_ni_t, ni_list);
1164 list_del(&ni->ni_list);
1172 * Initialize LNet library.
1174 * Only userspace program needs to call this function - it's automatically
1175 * called in the kernel at module loading time. Caller has to call LNetFini()
1176 * after a call to LNetInit(), if and only if the latter returned 0. It must
1177 * be called exactly once.
1179 * \return 0 on success, and -ve on failures.
1186 lnet_assert_wire_constants();
1187 LASSERT(!the_lnet.ln_init);
1189 memset(&the_lnet, 0, sizeof(the_lnet));
1191 /* refer to global cfs_cpt_table for now */
1192 the_lnet.ln_cpt_table = cfs_cpt_table;
1193 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
1195 LASSERT(the_lnet.ln_cpt_number > 0);
1196 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
1197 /* we are under risk of consuming all lh_cookie */
1198 CERROR("Can't have %d CPTs for LNet (max allowed is %d), please change setting of CPT-table and retry\n",
1199 the_lnet.ln_cpt_number, LNET_CPT_MAX);
1203 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
1204 the_lnet.ln_cpt_bits++;
1206 rc = lnet_create_locks();
1208 CERROR("Can't create LNet global locks: %d\n", rc);
1212 the_lnet.ln_refcount = 0;
1213 the_lnet.ln_init = 1;
1214 LNetInvalidateHandle(&the_lnet.ln_rc_eqh);
1215 INIT_LIST_HEAD(&the_lnet.ln_lnds);
1216 INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
1217 INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
1219 /* The hash table size is the number of bits it takes to express the set
1220 * ln_num_routes, minus 1 (better to under estimate than over so we
1221 * don't waste memory). */
1222 if (rnet_htable_size <= 0)
1223 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
1224 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
1225 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
1226 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
1227 order_base_2(rnet_htable_size) - 1);
1229 /* All LNDs apart from the LOLND are in separate modules. They
1230 * register themselves when their module loads, and unregister
1231 * themselves when their module is unloaded. */
1232 lnet_register_lnd(&the_lolnd);
1235 EXPORT_SYMBOL(LNetInit);
1238 * Finalize LNet library.
1240 * Only userspace program needs to call this function. It can be called
1243 * \pre LNetInit() called with success.
1244 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1249 LASSERT(the_lnet.ln_init);
1250 LASSERT(the_lnet.ln_refcount == 0);
1252 while (!list_empty(&the_lnet.ln_lnds))
1253 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
1255 lnet_destroy_locks();
1257 the_lnet.ln_init = 0;
1259 EXPORT_SYMBOL(LNetFini);
1262 * Set LNet PID and start LNet interfaces, routing, and forwarding.
1264 * Userspace program should call this after a successful call to LNetInit().
1265 * Users must call this function at least once before any other functions.
1266 * For each successful call there must be a corresponding call to
1267 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1270 * The PID used by LNet may be different from the one requested.
1273 * \param requested_pid PID requested by the caller.
1275 * \return >= 0 on success, and < 0 error code on failures.
1278 LNetNIInit(lnet_pid_t requested_pid)
1280 int im_a_router = 0;
1283 LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
1285 LASSERT(the_lnet.ln_init);
1286 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
1288 if (the_lnet.ln_refcount > 0) {
1289 rc = the_lnet.ln_refcount++;
1293 lnet_get_tunables();
1295 if (requested_pid == LNET_PID_ANY) {
1296 /* Don't instantiate LNET just for me */
1301 rc = lnet_prepare(requested_pid);
1305 rc = lnet_startup_lndnis();
1309 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
1313 rc = lnet_check_routes();
1317 rc = lnet_rtrpools_alloc(im_a_router);
1321 rc = lnet_acceptor_start();
1325 the_lnet.ln_refcount = 1;
1326 /* Now I may use my own API functions... */
1328 /* NB router checker needs the_lnet.ln_ping_info in
1329 * lnet_router_checker -> lnet_update_ni_status_locked */
1330 rc = lnet_ping_target_init();
1334 rc = lnet_router_checker_start();
1342 lnet_ping_target_fini();
1344 the_lnet.ln_refcount = 0;
1345 lnet_acceptor_stop();
1347 lnet_destroy_routes();
1348 lnet_shutdown_lndnis();
1354 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1357 EXPORT_SYMBOL(LNetNIInit);
1360 * Stop LNet interfaces, routing, and forwarding.
1362 * Users must call this function once for each successful call to LNetNIInit().
1363 * Once the LNetNIFini() operation has been started, the results of pending
1364 * API operations are undefined.
1366 * \return always 0 for current implementation.
1371 LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
1373 LASSERT(the_lnet.ln_init);
1374 LASSERT(the_lnet.ln_refcount > 0);
1376 if (the_lnet.ln_refcount != 1) {
1377 the_lnet.ln_refcount--;
1379 LASSERT(!the_lnet.ln_niinit_self);
1382 lnet_router_checker_stop();
1383 lnet_ping_target_fini();
1385 /* Teardown fns that use my own API functions BEFORE here */
1386 the_lnet.ln_refcount = 0;
1388 lnet_acceptor_stop();
1389 lnet_destroy_routes();
1390 lnet_shutdown_lndnis();
1394 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1397 EXPORT_SYMBOL(LNetNIFini);
1400 * This is an ugly hack to export IOC_LIBCFS_DEBUG_PEER and
1401 * IOC_LIBCFS_PORTALS_COMPATIBILITY commands to users, by tweaking the LNet
1402 * internal ioctl handler.
1404 * IOC_LIBCFS_PORTALS_COMPATIBILITY is now deprecated, don't use it.
1406 * \param cmd IOC_LIBCFS_DEBUG_PEER to print debugging data about a peer.
1407 * The data will be printed to system console. Don't use it excessively.
1408 * \param arg A pointer to lnet_process_id_t, process ID of the peer.
1410 * \return Always return 0 when called by users directly (i.e., not via ioctl).
1413 LNetCtl(unsigned int cmd, void *arg)
1415 struct libcfs_ioctl_data *data = arg;
1416 lnet_process_id_t id = {0};
1420 LASSERT(the_lnet.ln_init);
1421 LASSERT(the_lnet.ln_refcount > 0);
1424 case IOC_LIBCFS_GET_NI:
1425 rc = LNetGetId(data->ioc_count, &id);
1426 data->ioc_nid = id.nid;
1429 case IOC_LIBCFS_FAIL_NID:
1430 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
1432 case IOC_LIBCFS_ADD_ROUTE:
1433 rc = lnet_add_route(data->ioc_net, data->ioc_count,
1434 data->ioc_nid, data->ioc_priority);
1435 return (rc != 0) ? rc : lnet_check_routes();
1437 case IOC_LIBCFS_DEL_ROUTE:
1438 return lnet_del_route(data->ioc_net, data->ioc_nid);
1440 case IOC_LIBCFS_GET_ROUTE:
1441 return lnet_get_route(data->ioc_count,
1442 &data->ioc_net, &data->ioc_count,
1443 &data->ioc_nid, &data->ioc_flags,
1444 &data->ioc_priority);
1445 case IOC_LIBCFS_NOTIFY_ROUTER:
1446 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
1447 cfs_time_current() -
1448 cfs_time_seconds(get_seconds() -
1449 (time_t)data->ioc_u64[0]));
1451 case IOC_LIBCFS_PORTALS_COMPATIBILITY:
1452 /* This can be removed once lustre stops calling it */
1455 case IOC_LIBCFS_LNET_DIST:
1456 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
1457 if (rc < 0 && rc != -EHOSTUNREACH)
1460 data->ioc_u32[0] = rc;
1463 case IOC_LIBCFS_TESTPROTOCOMPAT:
1464 lnet_net_lock(LNET_LOCK_EX);
1465 the_lnet.ln_testprotocompat = data->ioc_flags;
1466 lnet_net_unlock(LNET_LOCK_EX);
1469 case IOC_LIBCFS_PING:
1470 id.nid = data->ioc_nid;
1471 id.pid = data->ioc_u32[0];
1472 rc = lnet_ping(id, data->ioc_u32[1], /* timeout */
1473 (lnet_process_id_t *)data->ioc_pbuf1,
1474 data->ioc_plen1/sizeof(lnet_process_id_t));
1477 data->ioc_count = rc;
1480 case IOC_LIBCFS_DEBUG_PEER: {
1481 /* CAVEAT EMPTOR: this one designed for calling directly; not
1483 id = *((lnet_process_id_t *) arg);
1485 lnet_debug_peer(id.nid);
1487 ni = lnet_net2ni(LNET_NIDNET(id.nid));
1489 CDEBUG(D_WARNING, "No NI for %s\n", libcfs_id2str(id));
1491 if (ni->ni_lnd->lnd_ctl == NULL) {
1492 CDEBUG(D_WARNING, "No ctl for %s\n",
1495 (void)ni->ni_lnd->lnd_ctl(ni, cmd, arg);
1504 ni = lnet_net2ni(data->ioc_net);
1508 if (ni->ni_lnd->lnd_ctl == NULL)
1511 rc = ni->ni_lnd->lnd_ctl(ni, cmd, arg);
1518 EXPORT_SYMBOL(LNetCtl);
1521 * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
1522 * all interfaces share a same PID, as requested by LNetNIInit().
1524 * \param index Index of the interface to look up.
1525 * \param id On successful return, this location will hold the
1526 * lnet_process_id_t ID of the interface.
1528 * \retval 0 If an interface exists at \a index.
1529 * \retval -ENOENT If no interface has been found.
1532 LNetGetId(unsigned int index, lnet_process_id_t *id)
1535 struct list_head *tmp;
1539 LASSERT(the_lnet.ln_init);
1541 /* LNetNI initilization failed? */
1542 if (the_lnet.ln_refcount == 0)
1545 cpt = lnet_net_lock_current();
1547 list_for_each(tmp, &the_lnet.ln_nis) {
1551 ni = list_entry(tmp, lnet_ni_t, ni_list);
1553 id->nid = ni->ni_nid;
1554 id->pid = the_lnet.ln_pid;
1559 lnet_net_unlock(cpt);
1562 EXPORT_SYMBOL(LNetGetId);
1565 * Print a string representation of handle \a h into buffer \a str of
1569 LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
1571 snprintf(str, len, "%#llx", h.cookie);
1573 EXPORT_SYMBOL(LNetSnprintHandle);
1576 lnet_create_ping_info(void)
1581 unsigned int infosz;
1583 lnet_process_id_t id;
1584 lnet_ping_info_t *pinfo;
1586 for (n = 0; ; n++) {
1587 rc = LNetGetId(n, &id);
1594 infosz = offsetof(lnet_ping_info_t, pi_ni[n]);
1595 LIBCFS_ALLOC(pinfo, infosz);
1596 if (pinfo == NULL) {
1597 CERROR("Can't allocate ping info[%d]\n", n);
1602 pinfo->pi_pid = the_lnet.ln_pid;
1603 pinfo->pi_magic = LNET_PROTO_PING_MAGIC;
1604 pinfo->pi_features = LNET_PING_FEAT_NI_STATUS;
1606 for (i = 0; i < n; i++) {
1607 lnet_ni_status_t *ns = &pinfo->pi_ni[i];
1609 rc = LNetGetId(i, &id);
1612 ns->ns_nid = id.nid;
1613 ns->ns_status = LNET_NI_STATUS_UP;
1617 ni = lnet_nid2ni_locked(id.nid, 0);
1618 LASSERT(ni != NULL);
1621 LASSERT(ni->ni_status == NULL);
1625 lnet_ni_decref_locked(ni, 0);
1629 the_lnet.ln_ping_info = pinfo;
1634 lnet_destroy_ping_info(void)
1640 list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
1642 ni->ni_status = NULL;
1648 LIBCFS_FREE(the_lnet.ln_ping_info,
1649 offsetof(lnet_ping_info_t,
1650 pi_ni[the_lnet.ln_ping_info->pi_nnis]));
1651 the_lnet.ln_ping_info = NULL;
1655 lnet_ping_target_init(void)
1657 lnet_md_t md = { NULL };
1658 lnet_handle_me_t meh;
1659 lnet_process_id_t id;
1664 rc = lnet_create_ping_info();
1668 /* We can have a tiny EQ since we only need to see the unlink event on
1669 * teardown, which by definition is the last one! */
1670 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &the_lnet.ln_ping_target_eq);
1672 CERROR("Can't allocate ping EQ: %d\n", rc);
1676 memset(&id, 0, sizeof(lnet_process_id_t));
1677 id.nid = LNET_NID_ANY;
1678 id.pid = LNET_PID_ANY;
1680 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1681 LNET_PROTO_PING_MATCHBITS, 0,
1682 LNET_UNLINK, LNET_INS_AFTER,
1685 CERROR("Can't create ping ME: %d\n", rc);
1689 /* initialize md content */
1690 infosz = offsetof(lnet_ping_info_t,
1691 pi_ni[the_lnet.ln_ping_info->pi_nnis]);
1692 md.start = the_lnet.ln_ping_info;
1694 md.threshold = LNET_MD_THRESH_INF;
1696 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1697 LNET_MD_MANAGE_REMOTE;
1699 md.eq_handle = the_lnet.ln_ping_target_eq;
1701 rc = LNetMDAttach(meh, md,
1703 &the_lnet.ln_ping_target_md);
1705 CERROR("Can't attach ping MD: %d\n", rc);
1712 rc2 = LNetMEUnlink(meh);
1715 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1718 lnet_destroy_ping_info();
1723 lnet_ping_target_fini(void)
1728 int timeout_ms = 1000;
1729 sigset_t blocked = cfs_block_allsigs();
1731 LNetMDUnlink(the_lnet.ln_ping_target_md);
1732 /* NB md could be busy; this just starts the unlink */
1735 rc = LNetEQPoll(&the_lnet.ln_ping_target_eq, 1,
1736 timeout_ms, &event, &which);
1738 /* I expect overflow... */
1739 LASSERT(rc >= 0 || rc == -EOVERFLOW);
1742 /* timed out: provide a diagnostic */
1743 CWARN("Still waiting for ping MD to unlink\n");
1748 /* Got a valid event */
1753 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1755 lnet_destroy_ping_info();
1756 cfs_restore_sigs(blocked);
1760 lnet_ping(lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_ids)
1762 lnet_handle_eq_t eqh;
1763 lnet_handle_md_t mdh;
1765 lnet_md_t md = { NULL };
1769 const int a_long_time = 60000; /* mS */
1770 int infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
1771 lnet_ping_info_t *info;
1772 lnet_process_id_t tmpid;
1780 id.nid == LNET_NID_ANY ||
1781 timeout_ms > 500000 || /* arbitrary limit! */
1782 n_ids > 20) /* arbitrary limit! */
1785 if (id.pid == LNET_PID_ANY)
1786 id.pid = LUSTRE_SRV_LNET_PID;
1788 LIBCFS_ALLOC(info, infosz);
1792 /* NB 2 events max (including any unlink event) */
1793 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
1795 CERROR("Can't allocate EQ: %d\n", rc);
1799 /* initialize md content */
1802 md.threshold = 2; /*GET/REPLY*/
1804 md.options = LNET_MD_TRUNCATE;
1808 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
1810 CERROR("Can't bind MD: %d\n", rc);
1814 rc = LNetGet(LNET_NID_ANY, mdh, id,
1815 LNET_RESERVED_PORTAL,
1816 LNET_PROTO_PING_MATCHBITS, 0);
1819 /* Don't CERROR; this could be deliberate! */
1821 rc2 = LNetMDUnlink(mdh);
1824 /* NB must wait for the UNLINK event below... */
1826 timeout_ms = a_long_time;
1830 /* MUST block for unlink to complete */
1832 blocked = cfs_block_allsigs();
1834 rc2 = LNetEQPoll(&eqh, 1, timeout_ms, &event, &which);
1837 cfs_restore_sigs(blocked);
1839 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
1840 (rc2 <= 0) ? -1 : event.type,
1841 (rc2 <= 0) ? -1 : event.status,
1842 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
1844 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
1846 if (rc2 <= 0 || event.status != 0) {
1847 /* timeout or error */
1848 if (!replied && rc == 0)
1849 rc = (rc2 < 0) ? rc2 :
1850 (rc2 == 0) ? -ETIMEDOUT :
1854 /* Ensure completion in finite time... */
1856 /* No assertion (racing with network) */
1858 timeout_ms = a_long_time;
1859 } else if (rc2 == 0) {
1860 /* timed out waiting for unlink */
1861 CWARN("ping %s: late network completion\n",
1864 } else if (event.type == LNET_EVENT_REPLY) {
1869 } while (rc2 <= 0 || !event.unlinked);
1873 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
1880 LASSERT(nob >= 0 && nob <= infosz);
1882 rc = -EPROTO; /* if I can't parse... */
1885 /* can't check magic/version */
1886 CERROR("%s: ping info too short %d\n",
1887 libcfs_id2str(id), nob);
1891 if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
1892 lnet_swap_pinginfo(info);
1893 } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
1894 CERROR("%s: Unexpected magic %08x\n",
1895 libcfs_id2str(id), info->pi_magic);
1899 if ((info->pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
1900 CERROR("%s: ping w/o NI status: 0x%x\n",
1901 libcfs_id2str(id), info->pi_features);
1905 if (nob < offsetof(lnet_ping_info_t, pi_ni[0])) {
1906 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
1907 nob, (int)offsetof(lnet_ping_info_t, pi_ni[0]));
1911 if (info->pi_nnis < n_ids)
1912 n_ids = info->pi_nnis;
1914 if (nob < offsetof(lnet_ping_info_t, pi_ni[n_ids])) {
1915 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
1916 nob, (int)offsetof(lnet_ping_info_t, pi_ni[n_ids]));
1920 rc = -EFAULT; /* If I SEGV... */
1922 memset(&tmpid, 0, sizeof(tmpid));
1923 for (i = 0; i < n_ids; i++) {
1924 tmpid.pid = info->pi_pid;
1925 tmpid.nid = info->pi_ni[i].ns_nid;
1926 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
1932 rc2 = LNetEQFree(eqh);
1934 CERROR("rc2 %d\n", rc2);
1938 LIBCFS_FREE(info, infosz);