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, 2013, 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 <lnet/lib-lnet.h>
39 #include <lnet/lib-dlc.h>
41 #include <linux/log2.h>
45 #define D_LNI D_CONSOLE
47 #define D_LNI D_CONFIG
50 lnet_t the_lnet; /* THE state of the network */
51 EXPORT_SYMBOL(the_lnet);
55 static char *ip2nets = "";
56 CFS_MODULE_PARM(ip2nets, "s", charp, 0444,
57 "LNET network <- IP table");
59 static char *networks = "";
60 CFS_MODULE_PARM(networks, "s", charp, 0444,
63 static char *routes = "";
64 CFS_MODULE_PARM(routes, "s", charp, 0444,
65 "routes to non-local networks");
67 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
68 CFS_MODULE_PARM(rnet_htable_size, "i", int, 0444,
69 "size of remote network hash table");
78 lnet_get_networks(void)
83 if (*networks != 0 && *ip2nets != 0) {
84 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
85 "'ip2nets' but not both at once\n");
90 rc = lnet_parse_ip2nets(&nets, ip2nets);
91 return (rc == 0) ? nets : NULL;
101 lnet_init_locks(void)
103 spin_lock_init(&the_lnet.ln_eq_wait_lock);
104 init_waitqueue_head(&the_lnet.ln_eq_waitq);
105 mutex_init(&the_lnet.ln_lnd_mutex);
106 mutex_init(&the_lnet.ln_api_mutex);
110 lnet_fini_locks(void)
117 lnet_get_routes(void)
119 char *str = getenv("LNET_ROUTES");
121 return (str == NULL) ? "" : str;
125 lnet_get_networks (void)
127 static char default_networks[256];
128 char *networks = getenv("LNET_NETWORKS");
133 struct list_head *tmp;
135 if (networks != NULL)
138 /* In userland, the default 'networks=' is the list of known net types */
139 len = sizeof(default_networks);
140 str = default_networks;
144 list_for_each(tmp, &the_lnet.ln_lnds) {
145 lnd_t *lnd = list_entry(tmp, lnd_t, lnd_list);
147 nob = snprintf(str, len, "%s%s", sep,
148 libcfs_lnd2str(lnd->lnd_type));
150 /* overflowed the string; leave it where it was */
159 return default_networks;
162 # ifndef HAVE_LIBPTHREAD
164 void lnet_init_locks(void)
166 the_lnet.ln_eq_wait_lock = 0;
167 the_lnet.ln_lnd_mutex = 0;
168 the_lnet.ln_api_mutex = 0;
171 void lnet_fini_locks(void)
173 LASSERT(the_lnet.ln_api_mutex == 0);
174 LASSERT(the_lnet.ln_lnd_mutex == 0);
175 LASSERT(the_lnet.ln_eq_wait_lock == 0);
180 void lnet_init_locks(void)
182 pthread_cond_init(&the_lnet.ln_eq_cond, NULL);
183 pthread_mutex_init(&the_lnet.ln_eq_wait_lock, NULL);
184 pthread_mutex_init(&the_lnet.ln_lnd_mutex, NULL);
185 pthread_mutex_init(&the_lnet.ln_api_mutex, NULL);
188 void lnet_fini_locks(void)
190 pthread_mutex_destroy(&the_lnet.ln_api_mutex);
191 pthread_mutex_destroy(&the_lnet.ln_lnd_mutex);
192 pthread_mutex_destroy(&the_lnet.ln_eq_wait_lock);
193 pthread_cond_destroy(&the_lnet.ln_eq_cond);
200 lnet_create_remote_nets_table(void)
203 struct list_head *hash;
205 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
206 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
207 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
209 CERROR("Failed to create remote nets hash table\n");
213 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
214 INIT_LIST_HEAD(&hash[i]);
215 the_lnet.ln_remote_nets_hash = hash;
220 lnet_destroy_remote_nets_table(void)
224 if (the_lnet.ln_remote_nets_hash == NULL)
227 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
228 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
230 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
231 LNET_REMOTE_NETS_HASH_SIZE *
232 sizeof(the_lnet.ln_remote_nets_hash[0]));
233 the_lnet.ln_remote_nets_hash = NULL;
237 lnet_destroy_locks(void)
239 if (the_lnet.ln_res_lock != NULL) {
240 cfs_percpt_lock_free(the_lnet.ln_res_lock);
241 the_lnet.ln_res_lock = NULL;
244 if (the_lnet.ln_net_lock != NULL) {
245 cfs_percpt_lock_free(the_lnet.ln_net_lock);
246 the_lnet.ln_net_lock = NULL;
253 lnet_create_locks(void)
257 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
258 if (the_lnet.ln_res_lock == NULL)
261 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
262 if (the_lnet.ln_net_lock == NULL)
268 lnet_destroy_locks();
272 void lnet_assert_wire_constants (void)
274 /* Wire protocol assertions generated by 'wirecheck'
275 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
276 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
277 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
280 CLASSERT (LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
281 CLASSERT (LNET_PROTO_TCP_VERSION_MAJOR == 1);
282 CLASSERT (LNET_PROTO_TCP_VERSION_MINOR == 0);
283 CLASSERT (LNET_MSG_ACK == 0);
284 CLASSERT (LNET_MSG_PUT == 1);
285 CLASSERT (LNET_MSG_GET == 2);
286 CLASSERT (LNET_MSG_REPLY == 3);
287 CLASSERT (LNET_MSG_HELLO == 4);
289 /* Checks for struct ptl_handle_wire_t */
290 CLASSERT ((int)sizeof(lnet_handle_wire_t) == 16);
291 CLASSERT ((int)offsetof(lnet_handle_wire_t, wh_interface_cookie) == 0);
292 CLASSERT ((int)sizeof(((lnet_handle_wire_t *)0)->wh_interface_cookie) == 8);
293 CLASSERT ((int)offsetof(lnet_handle_wire_t, wh_object_cookie) == 8);
294 CLASSERT ((int)sizeof(((lnet_handle_wire_t *)0)->wh_object_cookie) == 8);
296 /* Checks for struct lnet_magicversion_t */
297 CLASSERT ((int)sizeof(lnet_magicversion_t) == 8);
298 CLASSERT ((int)offsetof(lnet_magicversion_t, magic) == 0);
299 CLASSERT ((int)sizeof(((lnet_magicversion_t *)0)->magic) == 4);
300 CLASSERT ((int)offsetof(lnet_magicversion_t, version_major) == 4);
301 CLASSERT ((int)sizeof(((lnet_magicversion_t *)0)->version_major) == 2);
302 CLASSERT ((int)offsetof(lnet_magicversion_t, version_minor) == 6);
303 CLASSERT ((int)sizeof(((lnet_magicversion_t *)0)->version_minor) == 2);
305 /* Checks for struct lnet_hdr_t */
306 CLASSERT ((int)sizeof(lnet_hdr_t) == 72);
307 CLASSERT ((int)offsetof(lnet_hdr_t, dest_nid) == 0);
308 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->dest_nid) == 8);
309 CLASSERT ((int)offsetof(lnet_hdr_t, src_nid) == 8);
310 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->src_nid) == 8);
311 CLASSERT ((int)offsetof(lnet_hdr_t, dest_pid) == 16);
312 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->dest_pid) == 4);
313 CLASSERT ((int)offsetof(lnet_hdr_t, src_pid) == 20);
314 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->src_pid) == 4);
315 CLASSERT ((int)offsetof(lnet_hdr_t, type) == 24);
316 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->type) == 4);
317 CLASSERT ((int)offsetof(lnet_hdr_t, payload_length) == 28);
318 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->payload_length) == 4);
319 CLASSERT ((int)offsetof(lnet_hdr_t, msg) == 32);
320 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg) == 40);
323 CLASSERT ((int)offsetof(lnet_hdr_t, msg.ack.dst_wmd) == 32);
324 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.ack.dst_wmd) == 16);
325 CLASSERT ((int)offsetof(lnet_hdr_t, msg.ack.match_bits) == 48);
326 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.ack.match_bits) == 8);
327 CLASSERT ((int)offsetof(lnet_hdr_t, msg.ack.mlength) == 56);
328 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.ack.mlength) == 4);
331 CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.ack_wmd) == 32);
332 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.ack_wmd) == 16);
333 CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.match_bits) == 48);
334 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.match_bits) == 8);
335 CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.hdr_data) == 56);
336 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.hdr_data) == 8);
337 CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.ptl_index) == 64);
338 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.ptl_index) == 4);
339 CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.offset) == 68);
340 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.offset) == 4);
343 CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.return_wmd) == 32);
344 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.return_wmd) == 16);
345 CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.match_bits) == 48);
346 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.match_bits) == 8);
347 CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.ptl_index) == 56);
348 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.ptl_index) == 4);
349 CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.src_offset) == 60);
350 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.src_offset) == 4);
351 CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.sink_length) == 64);
352 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.sink_length) == 4);
355 CLASSERT ((int)offsetof(lnet_hdr_t, msg.reply.dst_wmd) == 32);
356 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.reply.dst_wmd) == 16);
359 CLASSERT ((int)offsetof(lnet_hdr_t, msg.hello.incarnation) == 32);
360 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.hello.incarnation) == 8);
361 CLASSERT ((int)offsetof(lnet_hdr_t, msg.hello.type) == 40);
362 CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.hello.type) == 4);
366 lnet_find_lnd_by_type (int type)
369 struct list_head *tmp;
371 /* holding lnd mutex */
372 list_for_each(tmp, &the_lnet.ln_lnds) {
373 lnd = list_entry(tmp, lnd_t, lnd_list);
375 if ((int)lnd->lnd_type == type)
382 lnet_register_lnd (lnd_t *lnd)
384 LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
386 LASSERT(the_lnet.ln_init);
387 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
388 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
390 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
391 lnd->lnd_refcount = 0;
393 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
395 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
397 EXPORT_SYMBOL(lnet_register_lnd);
400 lnet_unregister_lnd (lnd_t *lnd)
402 LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
404 LASSERT(the_lnet.ln_init);
405 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
406 LASSERT(lnd->lnd_refcount == 0);
408 list_del(&lnd->lnd_list);
409 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
411 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
413 EXPORT_SYMBOL(lnet_unregister_lnd);
416 lnet_counters_get(lnet_counters_t *counters)
418 lnet_counters_t *ctr;
421 memset(counters, 0, sizeof(*counters));
423 lnet_net_lock(LNET_LOCK_EX);
425 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
426 counters->msgs_max += ctr->msgs_max;
427 counters->msgs_alloc += ctr->msgs_alloc;
428 counters->errors += ctr->errors;
429 counters->send_count += ctr->send_count;
430 counters->recv_count += ctr->recv_count;
431 counters->route_count += ctr->route_count;
432 counters->drop_count += ctr->drop_count;
433 counters->send_length += ctr->send_length;
434 counters->recv_length += ctr->recv_length;
435 counters->route_length += ctr->route_length;
436 counters->drop_length += ctr->drop_length;
439 lnet_net_unlock(LNET_LOCK_EX);
441 EXPORT_SYMBOL(lnet_counters_get);
444 lnet_counters_reset(void)
446 lnet_counters_t *counters;
449 lnet_net_lock(LNET_LOCK_EX);
451 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
452 memset(counters, 0, sizeof(lnet_counters_t));
454 lnet_net_unlock(LNET_LOCK_EX);
456 EXPORT_SYMBOL(lnet_counters_reset);
458 #ifdef LNET_USE_LIB_FREELIST
461 lnet_freelist_init(lnet_freelist_t *fl, int n, int size)
467 size += offsetof (lnet_freeobj_t, fo_contents);
469 LIBCFS_ALLOC(space, n * size);
473 INIT_LIST_HEAD(&fl->fl_list);
476 fl->fl_objsize = size;
479 list_add((struct list_head *)space, &fl->fl_list);
487 lnet_freelist_fini(lnet_freelist_t *fl)
489 struct list_head *el;
492 if (fl->fl_nobjs == 0)
496 for (el = fl->fl_list.next; el != &fl->fl_list; el = el->next)
499 LASSERT (count == fl->fl_nobjs);
501 LIBCFS_FREE(fl->fl_objs, fl->fl_nobjs * fl->fl_objsize);
502 memset (fl, 0, sizeof (*fl));
505 #endif /* LNET_USE_LIB_FREELIST */
507 __u64 lnet_create_interface_cookie (void)
509 /* NB the interface cookie in wire handles guards against delayed
510 * replies and ACKs appearing valid after reboot. Initialisation time,
511 * even if it's only implemented to millisecond resolution is probably
512 * easily good enough. */
516 int rc = gettimeofday (&tv, NULL);
519 do_gettimeofday(&tv);
523 cookie += tv.tv_usec;
528 lnet_res_type2str(int type)
533 case LNET_COOKIE_TYPE_MD:
535 case LNET_COOKIE_TYPE_ME:
537 case LNET_COOKIE_TYPE_EQ:
543 lnet_res_container_cleanup(struct lnet_res_container *rec)
547 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
550 while (!list_empty(&rec->rec_active)) {
551 struct list_head *e = rec->rec_active.next;
554 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
555 lnet_eq_free(list_entry(e, lnet_eq_t, eq_list));
557 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
558 lnet_md_free(list_entry(e, lnet_libmd_t, md_list));
560 } else { /* NB: Active MEs should be attached on portals */
567 /* Found alive MD/ME/EQ, user really should unlink/free
568 * all of them before finalize LNet, but if someone didn't,
569 * we have to recycle garbage for him */
570 CERROR("%d active elements on exit of %s container\n",
571 count, lnet_res_type2str(rec->rec_type));
574 #ifdef LNET_USE_LIB_FREELIST
575 lnet_freelist_fini(&rec->rec_freelist);
577 if (rec->rec_lh_hash != NULL) {
578 LIBCFS_FREE(rec->rec_lh_hash,
579 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
580 rec->rec_lh_hash = NULL;
583 rec->rec_type = 0; /* mark it as finalized */
587 lnet_res_container_setup(struct lnet_res_container *rec,
588 int cpt, int type, int objnum, int objsz)
593 LASSERT(rec->rec_type == 0);
595 rec->rec_type = type;
596 INIT_LIST_HEAD(&rec->rec_active);
598 #ifdef LNET_USE_LIB_FREELIST
599 memset(&rec->rec_freelist, 0, sizeof(rec->rec_freelist));
600 rc = lnet_freelist_init(&rec->rec_freelist, objnum, objsz);
604 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
606 /* Arbitrary choice of hash table size */
607 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
608 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
609 if (rec->rec_lh_hash == NULL) {
614 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
615 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
620 CERROR("Failed to setup %s resource container\n",
621 lnet_res_type2str(type));
622 lnet_res_container_cleanup(rec);
627 lnet_res_containers_destroy(struct lnet_res_container **recs)
629 struct lnet_res_container *rec;
632 cfs_percpt_for_each(rec, i, recs)
633 lnet_res_container_cleanup(rec);
635 cfs_percpt_free(recs);
638 static struct lnet_res_container **
639 lnet_res_containers_create(int type, int objnum, int objsz)
641 struct lnet_res_container **recs;
642 struct lnet_res_container *rec;
646 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
648 CERROR("Failed to allocate %s resource containers\n",
649 lnet_res_type2str(type));
653 cfs_percpt_for_each(rec, i, recs) {
654 rc = lnet_res_container_setup(rec, i, type, objnum, objsz);
656 lnet_res_containers_destroy(recs);
665 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
667 /* ALWAYS called with lnet_res_lock held */
668 struct list_head *head;
669 lnet_libhandle_t *lh;
672 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
675 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
676 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
678 list_for_each_entry(lh, head, lh_hash_chain) {
679 if (lh->lh_cookie == cookie)
687 lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh)
689 /* ALWAYS called with lnet_res_lock held */
690 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
693 lh->lh_cookie = rec->rec_lh_cookie;
694 rec->rec_lh_cookie += 1 << ibits;
696 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
698 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
703 * Reserved API - do not use.
704 * Temporary workaround to allow uOSS and test programs force server
705 * mode in userspace. See comments near ln_server_mode_flag in
706 * lnet/lib-types.h */
710 the_lnet.ln_server_mode_flag = 1;
714 int lnet_unprepare(void);
717 lnet_prepare(lnet_pid_t requested_pid)
719 /* Prepare to bring up the network */
720 struct lnet_res_container **recs;
723 if (requested_pid == LNET_PID_ANY) {
724 /* Don't instantiate LNET just for me */
728 LASSERT (the_lnet.ln_refcount == 0);
730 the_lnet.ln_routing = 0;
733 LASSERT ((requested_pid & LNET_PID_USERFLAG) == 0);
734 the_lnet.ln_pid = requested_pid;
736 if (the_lnet.ln_server_mode_flag) {/* server case (uOSS) */
737 LASSERT ((requested_pid & LNET_PID_USERFLAG) == 0);
739 if (current_uid() != 0) /* Only root can run user-space server */
741 the_lnet.ln_pid = requested_pid;
743 } else {/* client case (liblustre) */
745 /* My PID must be unique on this node and flag I'm userspace */
746 the_lnet.ln_pid = getpid() | LNET_PID_USERFLAG;
750 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
751 INIT_LIST_HEAD(&the_lnet.ln_nis);
752 INIT_LIST_HEAD(&the_lnet.ln_nis_cpt);
753 INIT_LIST_HEAD(&the_lnet.ln_nis_zombie);
754 INIT_LIST_HEAD(&the_lnet.ln_routers);
756 rc = lnet_create_remote_nets_table();
760 the_lnet.ln_interface_cookie = lnet_create_interface_cookie();
762 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
763 sizeof(lnet_counters_t));
764 if (the_lnet.ln_counters == NULL) {
765 CERROR("Failed to allocate counters for LNet\n");
770 rc = lnet_peer_tables_create();
774 rc = lnet_msg_containers_create();
778 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
779 LNET_COOKIE_TYPE_EQ, LNET_FL_MAX_EQS,
784 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME, LNET_FL_MAX_MES,
789 the_lnet.ln_me_containers = recs;
791 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD, LNET_FL_MAX_MDS,
792 sizeof(lnet_libmd_t));
796 the_lnet.ln_md_containers = recs;
798 rc = lnet_portals_create();
800 CERROR("Failed to create portals for LNet: %d\n", rc);
812 lnet_unprepare (void)
814 /* NB no LNET_LOCK since this is the last reference. All LND instances
815 * have shut down already, so it is safe to unlink and free all
816 * descriptors, even those that appear committed to a network op (eg MD
817 * with non-zero pending count) */
819 lnet_fail_nid(LNET_NID_ANY, 0);
821 LASSERT(the_lnet.ln_refcount == 0);
822 LASSERT(list_empty(&the_lnet.ln_test_peers));
823 LASSERT(list_empty(&the_lnet.ln_nis));
824 LASSERT(list_empty(&the_lnet.ln_nis_cpt));
825 LASSERT(list_empty(&the_lnet.ln_nis_zombie));
827 lnet_portals_destroy();
829 if (the_lnet.ln_md_containers != NULL) {
830 lnet_res_containers_destroy(the_lnet.ln_md_containers);
831 the_lnet.ln_md_containers = NULL;
834 if (the_lnet.ln_me_containers != NULL) {
835 lnet_res_containers_destroy(the_lnet.ln_me_containers);
836 the_lnet.ln_me_containers = NULL;
839 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
841 lnet_msg_containers_destroy();
842 lnet_peer_tables_destroy();
843 lnet_rtrpools_free(0);
845 if (the_lnet.ln_counters != NULL) {
846 cfs_percpt_free(the_lnet.ln_counters);
847 the_lnet.ln_counters = NULL;
849 lnet_destroy_remote_nets_table();
855 lnet_net2ni_locked(__u32 net, int cpt)
857 struct list_head *tmp;
860 LASSERT(cpt != LNET_LOCK_EX);
862 list_for_each(tmp, &the_lnet.ln_nis) {
863 ni = list_entry(tmp, lnet_ni_t, ni_list);
865 if (LNET_NIDNET(ni->ni_nid) == net) {
866 lnet_ni_addref_locked(ni, cpt);
875 lnet_net2ni(__u32 net)
880 ni = lnet_net2ni_locked(net, 0);
885 EXPORT_SYMBOL(lnet_net2ni);
888 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
893 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
898 val = hash_long(key, LNET_CPT_BITS);
899 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
903 return (unsigned int)(key + val + (val >> 1)) % number;
907 lnet_cpt_of_nid_locked(lnet_nid_t nid)
911 /* must called with hold of lnet_net_lock */
912 if (LNET_CPT_NUMBER == 1)
913 return 0; /* the only one */
915 /* take lnet_net_lock(any) would be OK */
916 if (!list_empty(&the_lnet.ln_nis_cpt)) {
917 list_for_each_entry(ni, &the_lnet.ln_nis_cpt, ni_cptlist) {
918 if (LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid))
921 LASSERT(ni->ni_cpts != NULL);
922 return ni->ni_cpts[lnet_nid_cpt_hash
923 (nid, ni->ni_ncpts)];
927 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
931 lnet_cpt_of_nid(lnet_nid_t nid)
936 if (LNET_CPT_NUMBER == 1)
937 return 0; /* the only one */
939 if (list_empty(&the_lnet.ln_nis_cpt))
940 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
942 cpt = lnet_net_lock_current();
943 cpt2 = lnet_cpt_of_nid_locked(nid);
944 lnet_net_unlock(cpt);
948 EXPORT_SYMBOL(lnet_cpt_of_nid);
951 lnet_islocalnet(__u32 net)
956 cpt = lnet_net_lock_current();
958 ni = lnet_net2ni_locked(net, cpt);
960 lnet_ni_decref_locked(ni, cpt);
962 lnet_net_unlock(cpt);
968 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
971 struct list_head *tmp;
973 LASSERT(cpt != LNET_LOCK_EX);
975 list_for_each(tmp, &the_lnet.ln_nis) {
976 ni = list_entry(tmp, lnet_ni_t, ni_list);
978 if (ni->ni_nid == nid) {
979 lnet_ni_addref_locked(ni, cpt);
988 lnet_islocalnid(lnet_nid_t nid)
993 cpt = lnet_net_lock_current();
994 ni = lnet_nid2ni_locked(nid, cpt);
996 lnet_ni_decref_locked(ni, cpt);
997 lnet_net_unlock(cpt);
1003 lnet_count_acceptor_nis (void)
1005 /* Return the # of NIs that need the acceptor. */
1007 #if defined(__KERNEL__) || defined(HAVE_LIBPTHREAD)
1008 struct list_head *tmp;
1012 cpt = lnet_net_lock_current();
1013 list_for_each(tmp, &the_lnet.ln_nis) {
1014 ni = list_entry(tmp, lnet_ni_t, ni_list);
1016 if (ni->ni_lnd->lnd_accept != NULL)
1020 lnet_net_unlock(cpt);
1022 #endif /* defined(__KERNEL__) || defined(HAVE_LIBPTHREAD) */
1026 static lnet_ping_info_t *
1027 lnet_ping_info_create(int num_ni)
1029 lnet_ping_info_t *ping_info;
1030 unsigned int infosz;
1032 infosz = offsetof(lnet_ping_info_t, pi_ni[num_ni]);
1033 LIBCFS_ALLOC(ping_info, infosz);
1034 if (ping_info == NULL) {
1035 CERROR("Can't allocate ping info[%d]\n", num_ni);
1039 ping_info->pi_nnis = num_ni;
1040 ping_info->pi_pid = the_lnet.ln_pid;
1041 ping_info->pi_magic = LNET_PROTO_PING_MAGIC;
1042 ping_info->pi_features = LNET_PING_FEAT_NI_STATUS;
1048 lnet_get_ni_count(void)
1055 list_for_each_entry(ni, &the_lnet.ln_nis, ni_list)
1064 lnet_ping_info_free(lnet_ping_info_t *pinfo)
1067 offsetof(lnet_ping_info_t,
1068 pi_ni[pinfo->pi_nnis]));
1072 lnet_ping_info_destroy(void)
1076 lnet_net_lock(LNET_LOCK_EX);
1078 list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
1080 ni->ni_status = NULL;
1084 lnet_ping_info_free(the_lnet.ln_ping_info);
1085 the_lnet.ln_ping_info = NULL;
1087 lnet_net_unlock(LNET_LOCK_EX);
1091 lnet_ping_event_handler(lnet_event_t *event)
1093 lnet_ping_info_t *pinfo = event->md.user_ptr;
1095 if (event->unlinked)
1096 pinfo->pi_features = LNET_PING_FEAT_INVAL;
1100 lnet_ping_info_setup(lnet_ping_info_t **ppinfo, lnet_handle_md_t *md_handle,
1101 int ni_count, bool set_eq)
1103 lnet_handle_me_t me_handle;
1104 lnet_process_id_t id = {LNET_NID_ANY, LNET_PID_ANY};
1109 rc = LNetEQAlloc(0, lnet_ping_event_handler,
1110 &the_lnet.ln_ping_target_eq);
1112 CERROR("Can't allocate ping EQ: %d\n", rc);
1117 *ppinfo = lnet_ping_info_create(ni_count);
1118 if (*ppinfo == NULL) {
1123 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1124 LNET_PROTO_PING_MATCHBITS, 0,
1125 LNET_UNLINK, LNET_INS_AFTER,
1128 CERROR("Can't create ping ME: %d\n", rc);
1132 /* initialize md content */
1134 md.length = offsetof(lnet_ping_info_t,
1135 pi_ni[(*ppinfo)->pi_nnis]);
1136 md.threshold = LNET_MD_THRESH_INF;
1138 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1139 LNET_MD_MANAGE_REMOTE;
1141 md.eq_handle = the_lnet.ln_ping_target_eq;
1142 md.user_ptr = *ppinfo;
1144 rc = LNetMDAttach(me_handle, md, LNET_RETAIN, md_handle);
1146 CERROR("Can't attach ping MD: %d\n", rc);
1153 rc2 = LNetMEUnlink(me_handle);
1156 lnet_ping_info_free(*ppinfo);
1160 LNetEQFree(the_lnet.ln_ping_target_eq);
1165 lnet_ping_md_unlink(lnet_ping_info_t *pinfo, lnet_handle_md_t *md_handle)
1167 sigset_t blocked = cfs_block_allsigs();
1169 LNetMDUnlink(*md_handle);
1170 LNetInvalidateHandle(md_handle);
1172 /* NB md could be busy; this just starts the unlink */
1173 while (pinfo->pi_features != LNET_PING_FEAT_INVAL) {
1174 CDEBUG(D_NET, "Still waiting for ping MD to unlink\n");
1175 cfs_pause(cfs_time_seconds(1));
1178 cfs_restore_sigs(blocked);
1182 lnet_ping_info_install_locked(lnet_ping_info_t *ping_info)
1186 lnet_ni_status_t *ns;
1189 list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
1190 LASSERT(i < ping_info->pi_nnis);
1192 ns = &ping_info->pi_ni[i];
1194 ns->ns_nid = ni->ni_nid;
1197 ns->ns_status = (ni->ni_status != NULL) ?
1198 ni->ni_status->ns_status : LNET_NI_STATUS_UP;
1207 lnet_ping_target_update(lnet_ping_info_t *pinfo, lnet_handle_md_t md_handle)
1209 lnet_ping_info_t *old_pinfo = NULL;
1210 lnet_handle_md_t old_md;
1212 /* switch the NIs to point to the new ping info created */
1213 lnet_net_lock(LNET_LOCK_EX);
1215 if (!the_lnet.ln_routing)
1216 pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1217 lnet_ping_info_install_locked(pinfo);
1219 if (the_lnet.ln_ping_info != NULL) {
1220 old_pinfo = the_lnet.ln_ping_info;
1221 old_md = the_lnet.ln_ping_target_md;
1223 the_lnet.ln_ping_target_md = md_handle;
1224 the_lnet.ln_ping_info = pinfo;
1226 lnet_net_unlock(LNET_LOCK_EX);
1228 if (old_pinfo != NULL) {
1229 /* unlink the old ping info */
1230 lnet_ping_md_unlink(old_pinfo, &old_md);
1231 lnet_ping_info_free(old_pinfo);
1236 lnet_ping_target_fini(void)
1240 lnet_ping_md_unlink(the_lnet.ln_ping_info,
1241 &the_lnet.ln_ping_target_md);
1243 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1246 lnet_ping_info_destroy();
1250 lnet_ni_tq_credits(lnet_ni_t *ni)
1254 LASSERT(ni->ni_ncpts >= 1);
1256 if (ni->ni_ncpts == 1)
1257 return ni->ni_maxtxcredits;
1259 credits = ni->ni_maxtxcredits / ni->ni_ncpts;
1260 credits = max(credits, 8 * ni->ni_peertxcredits);
1261 credits = min(credits, ni->ni_maxtxcredits);
1267 lnet_clear_zombies_nis_locked(void)
1273 /* Now wait for the NI's I just nuked to show up on ln_zombie_nis
1274 * and shut them down in guaranteed thread context */
1276 while (!list_empty(&the_lnet.ln_nis_zombie)) {
1280 ni = list_entry(the_lnet.ln_nis_zombie.next,
1281 lnet_ni_t, ni_list);
1282 list_del_init(&ni->ni_list);
1283 cfs_percpt_for_each(ref, j, ni->ni_refs) {
1286 /* still busy, add it back to zombie list */
1287 list_add(&ni->ni_list, &the_lnet.ln_nis_zombie);
1291 if (!list_empty(&ni->ni_list)) {
1292 lnet_net_unlock(LNET_LOCK_EX);
1294 if ((i & (-i)) == i) {
1296 "Waiting for zombie LNI %s\n",
1297 libcfs_nid2str(ni->ni_nid));
1299 cfs_pause(cfs_time_seconds(1));
1300 lnet_net_lock(LNET_LOCK_EX);
1304 ni->ni_lnd->lnd_refcount--;
1305 lnet_net_unlock(LNET_LOCK_EX);
1307 islo = ni->ni_lnd->lnd_type == LOLND;
1309 LASSERT(!in_interrupt());
1310 (ni->ni_lnd->lnd_shutdown)(ni);
1312 /* can't deref lnd anymore now; it might have unregistered
1316 CDEBUG(D_LNI, "Removed LNI %s\n",
1317 libcfs_nid2str(ni->ni_nid));
1321 lnet_net_lock(LNET_LOCK_EX);
1326 lnet_shutdown_lndnis(void)
1331 /* NB called holding the global mutex */
1333 /* All quiet on the API front */
1334 LASSERT(!the_lnet.ln_shutdown);
1335 LASSERT(the_lnet.ln_refcount == 0);
1336 LASSERT(list_empty(&the_lnet.ln_nis_zombie));
1338 lnet_net_lock(LNET_LOCK_EX);
1339 the_lnet.ln_shutdown = 1; /* flag shutdown */
1341 /* Unlink NIs from the global table */
1342 while (!list_empty(&the_lnet.ln_nis)) {
1343 ni = list_entry(the_lnet.ln_nis.next,
1344 lnet_ni_t, ni_list);
1345 /* move it to zombie list and nobody can find it anymore */
1346 list_move(&ni->ni_list, &the_lnet.ln_nis_zombie);
1347 lnet_ni_decref_locked(ni, 0); /* drop ln_nis' ref */
1349 if (!list_empty(&ni->ni_cptlist)) {
1350 list_del_init(&ni->ni_cptlist);
1351 lnet_ni_decref_locked(ni, 0);
1355 /* Drop the cached eqwait NI. */
1356 if (the_lnet.ln_eq_waitni != NULL) {
1357 lnet_ni_decref_locked(the_lnet.ln_eq_waitni, 0);
1358 the_lnet.ln_eq_waitni = NULL;
1361 /* Drop the cached loopback NI. */
1362 if (the_lnet.ln_loni != NULL) {
1363 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
1364 the_lnet.ln_loni = NULL;
1367 lnet_net_unlock(LNET_LOCK_EX);
1369 /* Clear lazy portals and drop delayed messages which hold refs
1370 * on their lnet_msg_t::msg_rxpeer */
1371 for (i = 0; i < the_lnet.ln_nportals; i++)
1372 LNetClearLazyPortal(i);
1374 /* Clear the peer table and wait for all peers to go (they hold refs on
1376 lnet_peer_tables_cleanup(NULL);
1378 lnet_net_lock(LNET_LOCK_EX);
1380 lnet_clear_zombies_nis_locked();
1381 the_lnet.ln_shutdown = 0;
1382 lnet_net_unlock(LNET_LOCK_EX);
1386 lnet_shutdown_lndni(__u32 net)
1388 lnet_ping_info_t *pinfo;
1389 lnet_handle_md_t md_handle;
1390 lnet_ni_t *found_ni = NULL;
1394 if (LNET_NETTYP(net) == LOLND)
1397 ni_count = lnet_get_ni_count();
1399 /* create and link a new ping info, before removing the old one */
1400 rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count - 1, false);
1404 /* proceed with shutting down the NI */
1405 lnet_net_lock(LNET_LOCK_EX);
1407 found_ni = lnet_net2ni_locked(net, 0);
1408 if (found_ni == NULL) {
1409 lnet_net_unlock(LNET_LOCK_EX);
1410 lnet_ping_md_unlink(pinfo, &md_handle);
1411 lnet_ping_info_free(pinfo);
1415 /* decrement the reference counter on found_ni which was
1416 * incremented when we called lnet_net2ni_locked() */
1417 lnet_ni_decref_locked(found_ni, 0);
1419 /* Move ni to zombie list so nobody can find it anymore */
1420 list_move(&found_ni->ni_list, &the_lnet.ln_nis_zombie);
1422 /* Drop the lock reference for the ln_nis ref. */
1423 lnet_ni_decref_locked(found_ni, 0);
1425 if (!list_empty(&found_ni->ni_cptlist)) {
1426 list_del_init(&found_ni->ni_cptlist);
1427 lnet_ni_decref_locked(found_ni, 0);
1430 lnet_net_unlock(LNET_LOCK_EX);
1432 /* Do peer table cleanup for this ni */
1433 lnet_peer_tables_cleanup(found_ni);
1435 lnet_net_lock(LNET_LOCK_EX);
1436 lnet_clear_zombies_nis_locked();
1437 lnet_net_unlock(LNET_LOCK_EX);
1439 lnet_ping_target_update(pinfo, md_handle);
1445 lnet_startup_lndnis(struct list_head *nilist, __s32 peer_timeout,
1446 __s32 peer_cr, __s32 peer_buf_cr, __s32 credits,
1453 struct lnet_tx_queue *tq;
1456 while (!list_empty(nilist)) {
1457 ni = list_entry(nilist->next, lnet_ni_t, ni_list);
1458 lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));
1460 if (!libcfs_isknown_lnd(lnd_type))
1463 if (lnd_type == CIBLND ||
1464 lnd_type == OPENIBLND ||
1465 lnd_type == IIBLND ||
1466 lnd_type == VIBLND) {
1467 CERROR("LND %s obsoleted\n",
1468 libcfs_lnd2str(lnd_type));
1472 /* Make sure this new NI is unique. */
1473 lnet_net_lock(LNET_LOCK_EX);
1474 if (!lnet_net_unique(LNET_NIDNET(ni->ni_nid),
1475 &the_lnet.ln_nis)) {
1476 if (lnd_type == LOLND) {
1477 lnet_net_unlock(LNET_LOCK_EX);
1478 list_del(&ni->ni_list);
1483 CERROR("Net %s is not unique\n",
1484 libcfs_net2str(LNET_NIDNET(ni->ni_nid)));
1485 lnet_net_unlock(LNET_LOCK_EX);
1488 lnet_net_unlock(LNET_LOCK_EX);
1490 LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
1491 lnd = lnet_find_lnd_by_type(lnd_type);
1495 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
1496 rc = request_module("%s",
1497 libcfs_lnd2modname(lnd_type));
1498 LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
1500 lnd = lnet_find_lnd_by_type(lnd_type);
1502 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
1503 CERROR("Can't load LND %s, module %s, rc=%d\n",
1504 libcfs_lnd2str(lnd_type),
1505 libcfs_lnd2modname(lnd_type), rc);
1506 #ifndef HAVE_MODULE_LOADING_SUPPORT
1507 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
1508 "compiled with kernel module "
1509 "loading support.");
1516 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
1517 CERROR("LND %s not supported\n",
1518 libcfs_lnd2str(lnd_type));
1523 lnet_net_lock(LNET_LOCK_EX);
1524 lnd->lnd_refcount++;
1525 lnet_net_unlock(LNET_LOCK_EX);
1529 rc = (lnd->lnd_startup)(ni);
1531 LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
1534 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s"
1536 rc, libcfs_lnd2str(lnd->lnd_type));
1537 lnet_net_lock(LNET_LOCK_EX);
1538 lnd->lnd_refcount--;
1539 lnet_net_unlock(LNET_LOCK_EX);
1543 /* If given some LND tunable parameters, parse those now to
1544 * override the values in the NI structure. */
1545 if (peer_buf_cr >= 0)
1546 ni->ni_peerrtrcredits = peer_buf_cr;
1547 if (peer_timeout >= 0)
1548 ni->ni_peertimeout = peer_timeout;
1551 * Note: For now, don't allow the user to change
1552 * peertxcredits as this number is used in the
1553 * IB LND to control queue depth.
1554 * if (peer_cr != -1)
1555 * ni->ni_peertxcredits = peer_cr;
1558 ni->ni_maxtxcredits = credits;
1560 LASSERT(ni->ni_peertimeout <= 0 || lnd->lnd_query != NULL);
1562 list_del(&ni->ni_list);
1564 lnet_net_lock(LNET_LOCK_EX);
1565 /* refcount for ln_nis */
1566 lnet_ni_addref_locked(ni, 0);
1567 list_add_tail(&ni->ni_list, &the_lnet.ln_nis);
1568 if (ni->ni_cpts != NULL) {
1569 list_add_tail(&ni->ni_cptlist,
1570 &the_lnet.ln_nis_cpt);
1571 lnet_ni_addref_locked(ni, 0);
1574 lnet_net_unlock(LNET_LOCK_EX);
1576 /* increment the ni_count here to account for the LOLND as
1577 * well. If we increment past this point then the number
1578 * of count will be missing the LOLND, and then ping and
1579 * will not report the LOLND
1581 if (ni_count != NULL)
1584 if (lnd->lnd_type == LOLND) {
1586 LASSERT(the_lnet.ln_loni == NULL);
1587 the_lnet.ln_loni = ni;
1592 if (lnd->lnd_wait != NULL) {
1593 if (the_lnet.ln_eq_waitni == NULL) {
1595 the_lnet.ln_eq_waitni = ni;
1598 # ifndef HAVE_LIBPTHREAD
1599 LCONSOLE_ERROR_MSG(0x106, "LND %s not supported in a "
1600 "single-threaded runtime\n",
1601 libcfs_lnd2str(lnd_type));
1606 if (ni->ni_peertxcredits == 0 ||
1607 ni->ni_maxtxcredits == 0) {
1608 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1609 libcfs_lnd2str(lnd->lnd_type),
1610 ni->ni_peertxcredits == 0 ?
1615 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
1616 tq->tq_credits_min =
1617 tq->tq_credits_max =
1618 tq->tq_credits = lnet_ni_tq_credits(ni);
1621 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
1622 libcfs_nid2str(ni->ni_nid), ni->ni_peertxcredits,
1623 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
1624 ni->ni_peerrtrcredits, ni->ni_peertimeout);
1629 while (!list_empty(nilist)) {
1630 ni = list_entry(nilist->next, lnet_ni_t, ni_list);
1631 list_del(&ni->ni_list);
1638 * Initialize LNet library.
1640 * Only userspace program needs to call this function - it's automatically
1641 * called in the kernel at module loading time. Caller has to call LNetFini()
1642 * after a call to LNetInit(), if and only if the latter returned 0. It must
1643 * be called exactly once.
1645 * \return 0 on success, and -ve on failures.
1652 lnet_assert_wire_constants();
1653 LASSERT(!the_lnet.ln_init);
1655 memset(&the_lnet, 0, sizeof(the_lnet));
1657 /* refer to global cfs_cpt_table for now */
1658 the_lnet.ln_cpt_table = cfs_cpt_table;
1659 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
1661 LASSERT(the_lnet.ln_cpt_number > 0);
1662 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
1663 /* we are under risk of consuming all lh_cookie */
1664 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
1665 "please change setting of CPT-table and retry\n",
1666 the_lnet.ln_cpt_number, LNET_CPT_MAX);
1670 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
1671 the_lnet.ln_cpt_bits++;
1673 rc = lnet_create_locks();
1675 CERROR("Can't create LNet global locks: %d\n", rc);
1679 the_lnet.ln_refcount = 0;
1680 the_lnet.ln_init = 1;
1681 LNetInvalidateHandle(&the_lnet.ln_rc_eqh);
1682 INIT_LIST_HEAD(&the_lnet.ln_lnds);
1683 INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
1684 INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
1687 /* The hash table size is the number of bits it takes to express the set
1688 * ln_num_routes, minus 1 (better to under estimate than over so we
1689 * don't waste memory). */
1690 if (rnet_htable_size <= 0)
1691 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
1692 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
1693 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
1694 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
1695 order_base_2(rnet_htable_size) - 1);
1697 /* All LNDs apart from the LOLND are in separate modules. They
1698 * register themselves when their module loads, and unregister
1699 * themselves when their module is unloaded. */
1701 the_lnet.ln_remote_nets_hbits = 8;
1704 * NB the order here determines default 'networks=' order */
1705 # ifdef HAVE_LIBPTHREAD
1706 LNET_REGISTER_ULND(the_tcplnd);
1709 lnet_register_lnd(&the_lolnd);
1712 EXPORT_SYMBOL(LNetInit);
1715 * Finalize LNet library.
1717 * Only userspace program needs to call this function. It can be called
1720 * \pre LNetInit() called with success.
1721 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1726 LASSERT(the_lnet.ln_init);
1727 LASSERT(the_lnet.ln_refcount == 0);
1729 while (!list_empty(&the_lnet.ln_lnds))
1730 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
1732 lnet_destroy_locks();
1734 the_lnet.ln_init = 0;
1736 EXPORT_SYMBOL(LNetFini);
1739 * Set LNet PID and start LNet interfaces, routing, and forwarding.
1741 * Userspace program should call this after a successful call to LNetInit().
1742 * Users must call this function at least once before any other functions.
1743 * For each successful call there must be a corresponding call to
1744 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1747 * The PID used by LNet may be different from the one requested.
1750 * \param requested_pid PID requested by the caller.
1752 * \return >= 0 on success, and < 0 error code on failures.
1755 LNetNIInit(lnet_pid_t requested_pid)
1757 int im_a_router = 0;
1762 lnet_ping_info_t *pinfo;
1763 lnet_handle_md_t md_handle;
1764 struct list_head net_head;
1767 INIT_LIST_HEAD(&net_head);
1769 LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
1771 LASSERT(the_lnet.ln_init);
1772 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
1774 if (the_lnet.ln_refcount > 0) {
1775 rc = the_lnet.ln_refcount++;
1776 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1780 nets = lnet_get_networks();
1782 rc = lnet_prepare(requested_pid);
1786 rc = lnet_parse_networks(&net_head, nets);
1790 rc = lnet_startup_lndnis(&net_head, -1, -1, -1, -1, &ni_count);
1794 if (the_lnet.ln_eq_waitni != NULL && ni_count > 1) {
1795 lnd_type = the_lnet.ln_eq_waitni->ni_lnd->lnd_type;
1796 LCONSOLE_ERROR_MSG(0x109, "LND %s can only run single-network"
1798 libcfs_lnd2str(lnd_type));
1802 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
1806 rc = lnet_check_routes();
1810 rc = lnet_rtrpools_alloc(im_a_router);
1814 rc = lnet_acceptor_start();
1817 the_lnet.ln_refcount = 1;
1818 /* Now I may use my own API functions... */
1820 rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count, true);
1824 lnet_ping_target_update(pinfo, md_handle);
1826 rc = lnet_router_checker_start();
1832 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1837 the_lnet.ln_refcount = 0;
1838 lnet_ping_md_unlink(pinfo, &md_handle);
1839 lnet_ping_info_free(pinfo);
1841 lnet_acceptor_stop();
1842 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1845 lnet_destroy_routes();
1846 lnet_shutdown_lndnis();
1851 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1852 while (!list_empty(&net_head)) {
1853 ni = list_entry(net_head.next, struct lnet_ni, ni_list);
1854 list_del_init(&ni->ni_list);
1859 EXPORT_SYMBOL(LNetNIInit);
1862 * Stop LNet interfaces, routing, and forwarding.
1864 * Users must call this function once for each successful call to LNetNIInit().
1865 * Once the LNetNIFini() operation has been started, the results of pending
1866 * API operations are undefined.
1868 * \return always 0 for current implementation.
1873 LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
1875 LASSERT (the_lnet.ln_init);
1876 LASSERT (the_lnet.ln_refcount > 0);
1878 if (the_lnet.ln_refcount != 1) {
1879 the_lnet.ln_refcount--;
1881 LASSERT (!the_lnet.ln_niinit_self);
1884 lnet_router_checker_stop();
1885 lnet_ping_target_fini();
1887 /* Teardown fns that use my own API functions BEFORE here */
1888 the_lnet.ln_refcount = 0;
1890 lnet_acceptor_stop();
1891 lnet_destroy_routes();
1892 lnet_shutdown_lndnis();
1896 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1899 EXPORT_SYMBOL(LNetNIFini);
1902 lnet_dyn_add_ni(lnet_pid_t requested_pid, char *nets,
1903 __s32 peer_timeout, __s32 peer_cr, __s32 peer_buf_cr,
1906 lnet_ping_info_t *pinfo;
1907 lnet_handle_md_t md_handle;
1909 struct list_head net_head;
1912 INIT_LIST_HEAD(&net_head);
1914 /* Create a ni structure for the network string */
1915 rc = lnet_parse_networks(&net_head, nets);
1919 LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
1922 rc = -EINVAL; /* only add one interface per call */
1926 rc = lnet_ping_info_setup(&pinfo, &md_handle, 1 + lnet_get_ni_count(),
1931 rc = lnet_startup_lndnis(&net_head, peer_timeout, peer_cr,
1932 peer_buf_cr, credits, NULL);
1936 lnet_ping_target_update(pinfo, md_handle);
1937 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1942 lnet_ping_md_unlink(pinfo, &md_handle);
1943 lnet_ping_info_free(pinfo);
1945 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1946 while (!list_empty(&net_head)) {
1947 ni = list_entry(net_head.next, struct lnet_ni, ni_list);
1948 list_del_init(&ni->ni_list);
1955 lnet_dyn_del_ni(__u32 net)
1959 LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
1960 rc = lnet_shutdown_lndni(net);
1961 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
1967 * This is an ugly hack to export IOC_LIBCFS_DEBUG_PEER and
1968 * IOC_LIBCFS_PORTALS_COMPATIBILITY commands to users, by tweaking the LNet
1969 * internal ioctl handler.
1971 * IOC_LIBCFS_PORTALS_COMPATIBILITY is now deprecated, don't use it.
1973 * \param cmd IOC_LIBCFS_DEBUG_PEER to print debugging data about a peer.
1974 * The data will be printed to system console. Don't use it excessively.
1975 * \param arg A pointer to lnet_process_id_t, process ID of the peer.
1977 * \return Always return 0 when called by users directly (i.e., not via ioctl).
1980 LNetCtl(unsigned int cmd, void *arg)
1982 struct libcfs_ioctl_data *data = arg;
1983 struct lnet_ioctl_config_data *config;
1984 lnet_process_id_t id = {0};
1988 LASSERT(the_lnet.ln_init);
1991 case IOC_LIBCFS_GET_NI:
1992 rc = LNetGetId(data->ioc_count, &id);
1993 data->ioc_nid = id.nid;
1996 case IOC_LIBCFS_FAIL_NID:
1997 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
1999 case IOC_LIBCFS_ADD_ROUTE:
2000 LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
2001 rc = lnet_add_route(data->ioc_net, data->ioc_count,
2002 data->ioc_nid, data->ioc_priority);
2003 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
2004 return (rc != 0) ? rc : lnet_check_routes();
2006 case IOC_LIBCFS_DEL_ROUTE:
2008 LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
2009 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
2010 LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
2013 case IOC_LIBCFS_GET_ROUTE:
2015 return lnet_get_route(config->cfg_count,
2017 &config->cfg_config_u.cfg_route.rtr_hop,
2019 &config->cfg_config_u.cfg_route.rtr_flags,
2020 &config->cfg_config_u.cfg_route.
2023 case IOC_LIBCFS_ADD_NET:
2026 case IOC_LIBCFS_DEL_NET:
2029 case IOC_LIBCFS_GET_NET:
2032 case IOC_LIBCFS_GET_LNET_STATS:
2034 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
2036 lnet_counters_get(&lnet_stats->st_cntrs);
2040 #if defined(__KERNEL__) && defined(LNET_ROUTER)
2041 case IOC_LIBCFS_CONFIG_RTR:
2044 case IOC_LIBCFS_ADD_BUF:
2048 case IOC_LIBCFS_GET_BUF:
2051 case IOC_LIBCFS_GET_PEER_INFO:
2054 case IOC_LIBCFS_NOTIFY_ROUTER:
2055 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
2056 cfs_time_current() -
2057 cfs_time_seconds(cfs_time_current_sec() -
2058 (time_t)data->ioc_u64[0]));
2060 case IOC_LIBCFS_PORTALS_COMPATIBILITY:
2061 /* This can be removed once lustre stops calling it */
2064 case IOC_LIBCFS_LNET_DIST:
2065 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
2066 if (rc < 0 && rc != -EHOSTUNREACH)
2069 data->ioc_u32[0] = rc;
2072 case IOC_LIBCFS_TESTPROTOCOMPAT:
2073 lnet_net_lock(LNET_LOCK_EX);
2074 the_lnet.ln_testprotocompat = data->ioc_flags;
2075 lnet_net_unlock(LNET_LOCK_EX);
2078 case IOC_LIBCFS_PING:
2079 id.nid = data->ioc_nid;
2080 id.pid = data->ioc_u32[0];
2081 rc = lnet_ping(id, data->ioc_u32[1], /* timeout */
2082 (lnet_process_id_t *)data->ioc_pbuf1,
2083 data->ioc_plen1/sizeof(lnet_process_id_t));
2086 data->ioc_count = rc;
2089 case IOC_LIBCFS_DEBUG_PEER: {
2090 /* CAVEAT EMPTOR: this one designed for calling directly; not
2092 id = *((lnet_process_id_t *) arg);
2094 lnet_debug_peer(id.nid);
2096 ni = lnet_net2ni(LNET_NIDNET(id.nid));
2098 CDEBUG(D_WARNING, "No NI for %s\n", libcfs_id2str(id));
2100 if (ni->ni_lnd->lnd_ctl == NULL) {
2101 CDEBUG(D_WARNING, "No ctl for %s\n",
2104 (void)ni->ni_lnd->lnd_ctl(ni, cmd, arg);
2113 ni = lnet_net2ni(data->ioc_net);
2117 if (ni->ni_lnd->lnd_ctl == NULL)
2120 rc = ni->ni_lnd->lnd_ctl(ni, cmd, arg);
2127 EXPORT_SYMBOL(LNetCtl);
2130 * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
2131 * all interfaces share a same PID, as requested by LNetNIInit().
2133 * \param index Index of the interface to look up.
2134 * \param id On successful return, this location will hold the
2135 * lnet_process_id_t ID of the interface.
2137 * \retval 0 If an interface exists at \a index.
2138 * \retval -ENOENT If no interface has been found.
2141 LNetGetId(unsigned int index, lnet_process_id_t *id)
2144 struct list_head *tmp;
2148 LASSERT(the_lnet.ln_init);
2149 LASSERT(the_lnet.ln_refcount > 0);
2151 cpt = lnet_net_lock_current();
2153 list_for_each(tmp, &the_lnet.ln_nis) {
2157 ni = list_entry(tmp, lnet_ni_t, ni_list);
2159 id->nid = ni->ni_nid;
2160 id->pid = the_lnet.ln_pid;
2165 lnet_net_unlock(cpt);
2168 EXPORT_SYMBOL(LNetGetId);
2171 * Print a string representation of handle \a h into buffer \a str of
2175 LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
2177 snprintf(str, len, LPX64, h.cookie);
2179 EXPORT_SYMBOL(LNetSnprintHandle);
2182 lnet_ping(lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids,
2185 lnet_handle_eq_t eqh;
2186 lnet_handle_md_t mdh;
2192 const int a_long_time = 60000; /* mS */
2194 lnet_ping_info_t *info;
2195 lnet_process_id_t tmpid;
2202 infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
2205 id.nid == LNET_NID_ANY ||
2206 timeout_ms > 500000 || /* arbitrary limit! */
2207 n_ids > 20) /* arbitrary limit! */
2210 if (id.pid == LNET_PID_ANY)
2211 id.pid = LUSTRE_SRV_LNET_PID;
2213 LIBCFS_ALLOC(info, infosz);
2217 /* NB 2 events max (including any unlink event) */
2218 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
2220 CERROR("Can't allocate EQ: %d\n", rc);
2224 /* initialize md content */
2227 md.threshold = 2; /*GET/REPLY*/
2229 md.options = LNET_MD_TRUNCATE;
2233 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
2235 CERROR("Can't bind MD: %d\n", rc);
2239 rc = LNetGet(LNET_NID_ANY, mdh, id,
2240 LNET_RESERVED_PORTAL,
2241 LNET_PROTO_PING_MATCHBITS, 0);
2244 /* Don't CERROR; this could be deliberate! */
2246 rc2 = LNetMDUnlink(mdh);
2249 /* NB must wait for the UNLINK event below... */
2251 timeout_ms = a_long_time;
2255 /* MUST block for unlink to complete */
2257 blocked = cfs_block_allsigs();
2259 rc2 = LNetEQPoll(&eqh, 1, timeout_ms, &event, &which);
2262 cfs_restore_sigs(blocked);
2264 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
2265 (rc2 <= 0) ? -1 : event.type,
2266 (rc2 <= 0) ? -1 : event.status,
2267 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
2269 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
2271 if (rc2 <= 0 || event.status != 0) {
2272 /* timeout or error */
2273 if (!replied && rc == 0)
2274 rc = (rc2 < 0) ? rc2 :
2275 (rc2 == 0) ? -ETIMEDOUT :
2279 /* Ensure completion in finite time... */
2281 /* No assertion (racing with network) */
2283 timeout_ms = a_long_time;
2284 } else if (rc2 == 0) {
2285 /* timed out waiting for unlink */
2286 CWARN("ping %s: late network completion\n",
2289 } else if (event.type == LNET_EVENT_REPLY) {
2294 } while (rc2 <= 0 || !event.unlinked);
2298 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
2305 LASSERT(nob >= 0 && nob <= infosz);
2307 rc = -EPROTO; /* if I can't parse... */
2310 /* can't check magic/version */
2311 CERROR("%s: ping info too short %d\n",
2312 libcfs_id2str(id), nob);
2316 if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
2317 lnet_swap_pinginfo(info);
2318 } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
2319 CERROR("%s: Unexpected magic %08x\n",
2320 libcfs_id2str(id), info->pi_magic);
2324 if ((info->pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
2325 CERROR("%s: ping w/o NI status: 0x%x\n",
2326 libcfs_id2str(id), info->pi_features);
2330 if (nob < offsetof(lnet_ping_info_t, pi_ni[0])) {
2331 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
2332 nob, (int)offsetof(lnet_ping_info_t, pi_ni[0]));
2336 if (info->pi_nnis < n_ids)
2337 n_ids = info->pi_nnis;
2339 if (nob < offsetof(lnet_ping_info_t, pi_ni[n_ids])) {
2340 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
2341 nob, (int)offsetof(lnet_ping_info_t, pi_ni[n_ids]));
2345 rc = -EFAULT; /* If I SEGV... */
2347 for (i = 0; i < n_ids; i++) {
2348 tmpid.pid = info->pi_pid;
2349 tmpid.nid = info->pi_ni[i].ns_nid;
2350 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
2356 rc2 = LNetEQFree(eqh);
2358 CERROR("rc2 %d\n", rc2);
2362 LIBCFS_FREE(info, infosz);