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) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_LNET
34 #include <linux/log2.h>
35 #include <linux/ktime.h>
37 #include <lnet/lib-lnet.h>
39 #define D_LNI D_CONSOLE
41 lnet_t the_lnet; /* THE state of the network */
42 EXPORT_SYMBOL(the_lnet);
44 static char *ip2nets = "";
45 module_param(ip2nets, charp, 0444);
46 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
48 static char *networks = "";
49 module_param(networks, charp, 0444);
50 MODULE_PARM_DESC(networks, "local networks");
52 static char *routes = "";
53 module_param(routes, charp, 0444);
54 MODULE_PARM_DESC(routes, "routes to non-local networks");
56 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
57 module_param(rnet_htable_size, int, 0444);
58 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
60 static int use_tcp_bonding = false;
61 module_param(use_tcp_bonding, int, 0444);
62 MODULE_PARM_DESC(use_tcp_bonding,
63 "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
66 * This sequence number keeps track of how many times DLC was used to
67 * update the configuration. It is incremented on any DLC update and
68 * checked when sending a message to determine if there is a need to
69 * re-run the selection algorithm to handle configuration change.
70 * Look at lnet_select_pathway() for more details on its usage.
72 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
74 static int lnet_ping(lnet_process_id_t id, signed long timeout,
75 lnet_process_id_t __user *ids, int n_ids);
84 lnet_get_networks(void)
89 if (*networks != 0 && *ip2nets != 0) {
90 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
91 "'ip2nets' but not both at once\n");
96 rc = lnet_parse_ip2nets(&nets, ip2nets);
97 return (rc == 0) ? nets : NULL;
107 lnet_init_locks(void)
109 spin_lock_init(&the_lnet.ln_eq_wait_lock);
110 init_waitqueue_head(&the_lnet.ln_eq_waitq);
111 init_waitqueue_head(&the_lnet.ln_rc_waitq);
112 mutex_init(&the_lnet.ln_lnd_mutex);
113 mutex_init(&the_lnet.ln_api_mutex);
117 lnet_fini_locks(void)
121 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
122 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
126 lnet_descriptor_setup(void)
128 /* create specific kmem_cache for MEs and small MDs (i.e., originally
129 * allocated in <size-xxx> kmem_cache).
131 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(lnet_me_t),
133 if (!lnet_mes_cachep)
136 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
137 LNET_SMALL_MD_SIZE, 0, 0,
139 if (!lnet_small_mds_cachep)
146 lnet_descriptor_cleanup(void)
149 if (lnet_small_mds_cachep) {
150 kmem_cache_destroy(lnet_small_mds_cachep);
151 lnet_small_mds_cachep = NULL;
154 if (lnet_mes_cachep) {
155 kmem_cache_destroy(lnet_mes_cachep);
156 lnet_mes_cachep = NULL;
161 lnet_create_remote_nets_table(void)
164 struct list_head *hash;
166 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
167 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
168 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
170 CERROR("Failed to create remote nets hash table\n");
174 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
175 INIT_LIST_HEAD(&hash[i]);
176 the_lnet.ln_remote_nets_hash = hash;
181 lnet_destroy_remote_nets_table(void)
185 if (the_lnet.ln_remote_nets_hash == NULL)
188 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
189 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
191 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
192 LNET_REMOTE_NETS_HASH_SIZE *
193 sizeof(the_lnet.ln_remote_nets_hash[0]));
194 the_lnet.ln_remote_nets_hash = NULL;
198 lnet_destroy_locks(void)
200 if (the_lnet.ln_res_lock != NULL) {
201 cfs_percpt_lock_free(the_lnet.ln_res_lock);
202 the_lnet.ln_res_lock = NULL;
205 if (the_lnet.ln_net_lock != NULL) {
206 cfs_percpt_lock_free(the_lnet.ln_net_lock);
207 the_lnet.ln_net_lock = NULL;
214 lnet_create_locks(void)
218 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
219 if (the_lnet.ln_res_lock == NULL)
222 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
223 if (the_lnet.ln_net_lock == NULL)
229 lnet_destroy_locks();
233 static void lnet_assert_wire_constants(void)
235 /* Wire protocol assertions generated by 'wirecheck'
236 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
237 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
238 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
241 CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
242 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
243 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
244 CLASSERT(LNET_MSG_ACK == 0);
245 CLASSERT(LNET_MSG_PUT == 1);
246 CLASSERT(LNET_MSG_GET == 2);
247 CLASSERT(LNET_MSG_REPLY == 3);
248 CLASSERT(LNET_MSG_HELLO == 4);
250 /* Checks for struct lnet_handle_wire */
251 CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
252 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
253 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
254 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
255 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
257 /* Checks for struct lnet_magicversion_t */
258 CLASSERT((int)sizeof(lnet_magicversion_t) == 8);
259 CLASSERT((int)offsetof(lnet_magicversion_t, magic) == 0);
260 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->magic) == 4);
261 CLASSERT((int)offsetof(lnet_magicversion_t, version_major) == 4);
262 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_major) == 2);
263 CLASSERT((int)offsetof(lnet_magicversion_t, version_minor) == 6);
264 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_minor) == 2);
266 /* Checks for struct lnet_hdr_t */
267 CLASSERT((int)sizeof(lnet_hdr_t) == 72);
268 CLASSERT((int)offsetof(lnet_hdr_t, dest_nid) == 0);
269 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_nid) == 8);
270 CLASSERT((int)offsetof(lnet_hdr_t, src_nid) == 8);
271 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_nid) == 8);
272 CLASSERT((int)offsetof(lnet_hdr_t, dest_pid) == 16);
273 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_pid) == 4);
274 CLASSERT((int)offsetof(lnet_hdr_t, src_pid) == 20);
275 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_pid) == 4);
276 CLASSERT((int)offsetof(lnet_hdr_t, type) == 24);
277 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->type) == 4);
278 CLASSERT((int)offsetof(lnet_hdr_t, payload_length) == 28);
279 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->payload_length) == 4);
280 CLASSERT((int)offsetof(lnet_hdr_t, msg) == 32);
281 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg) == 40);
284 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.dst_wmd) == 32);
285 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.dst_wmd) == 16);
286 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.match_bits) == 48);
287 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.match_bits) == 8);
288 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.mlength) == 56);
289 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.mlength) == 4);
292 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ack_wmd) == 32);
293 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ack_wmd) == 16);
294 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.match_bits) == 48);
295 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.match_bits) == 8);
296 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.hdr_data) == 56);
297 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.hdr_data) == 8);
298 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ptl_index) == 64);
299 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ptl_index) == 4);
300 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.offset) == 68);
301 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.offset) == 4);
304 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.return_wmd) == 32);
305 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.return_wmd) == 16);
306 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.match_bits) == 48);
307 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.match_bits) == 8);
308 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.ptl_index) == 56);
309 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.ptl_index) == 4);
310 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.src_offset) == 60);
311 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.src_offset) == 4);
312 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.sink_length) == 64);
313 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.sink_length) == 4);
316 CLASSERT((int)offsetof(lnet_hdr_t, msg.reply.dst_wmd) == 32);
317 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.reply.dst_wmd) == 16);
320 CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.incarnation) == 32);
321 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.incarnation) == 8);
322 CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.type) == 40);
323 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.type) == 4);
326 static lnd_t *lnet_find_lnd_by_type(__u32 type)
329 struct list_head *tmp;
331 /* holding lnd mutex */
332 list_for_each(tmp, &the_lnet.ln_lnds) {
333 lnd = list_entry(tmp, lnd_t, lnd_list);
335 if (lnd->lnd_type == type)
342 lnet_register_lnd (lnd_t *lnd)
344 mutex_lock(&the_lnet.ln_lnd_mutex);
346 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
347 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
349 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
350 lnd->lnd_refcount = 0;
352 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
354 mutex_unlock(&the_lnet.ln_lnd_mutex);
356 EXPORT_SYMBOL(lnet_register_lnd);
359 lnet_unregister_lnd (lnd_t *lnd)
361 mutex_lock(&the_lnet.ln_lnd_mutex);
363 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
364 LASSERT(lnd->lnd_refcount == 0);
366 list_del(&lnd->lnd_list);
367 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
369 mutex_unlock(&the_lnet.ln_lnd_mutex);
371 EXPORT_SYMBOL(lnet_unregister_lnd);
374 lnet_counters_get(lnet_counters_t *counters)
376 lnet_counters_t *ctr;
379 memset(counters, 0, sizeof(*counters));
381 lnet_net_lock(LNET_LOCK_EX);
383 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
384 counters->msgs_max += ctr->msgs_max;
385 counters->msgs_alloc += ctr->msgs_alloc;
386 counters->errors += ctr->errors;
387 counters->send_count += ctr->send_count;
388 counters->recv_count += ctr->recv_count;
389 counters->route_count += ctr->route_count;
390 counters->drop_count += ctr->drop_count;
391 counters->send_length += ctr->send_length;
392 counters->recv_length += ctr->recv_length;
393 counters->route_length += ctr->route_length;
394 counters->drop_length += ctr->drop_length;
397 lnet_net_unlock(LNET_LOCK_EX);
399 EXPORT_SYMBOL(lnet_counters_get);
402 lnet_counters_reset(void)
404 lnet_counters_t *counters;
407 lnet_net_lock(LNET_LOCK_EX);
409 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
410 memset(counters, 0, sizeof(lnet_counters_t));
412 lnet_net_unlock(LNET_LOCK_EX);
416 lnet_res_type2str(int type)
421 case LNET_COOKIE_TYPE_MD:
423 case LNET_COOKIE_TYPE_ME:
425 case LNET_COOKIE_TYPE_EQ:
431 lnet_res_container_cleanup(struct lnet_res_container *rec)
435 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
438 while (!list_empty(&rec->rec_active)) {
439 struct list_head *e = rec->rec_active.next;
442 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
443 lnet_eq_free(list_entry(e, lnet_eq_t, eq_list));
445 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
446 lnet_md_free(list_entry(e, lnet_libmd_t, md_list));
448 } else { /* NB: Active MEs should be attached on portals */
455 /* Found alive MD/ME/EQ, user really should unlink/free
456 * all of them before finalize LNet, but if someone didn't,
457 * we have to recycle garbage for him */
458 CERROR("%d active elements on exit of %s container\n",
459 count, lnet_res_type2str(rec->rec_type));
462 if (rec->rec_lh_hash != NULL) {
463 LIBCFS_FREE(rec->rec_lh_hash,
464 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
465 rec->rec_lh_hash = NULL;
468 rec->rec_type = 0; /* mark it as finalized */
472 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
477 LASSERT(rec->rec_type == 0);
479 rec->rec_type = type;
480 INIT_LIST_HEAD(&rec->rec_active);
482 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
484 /* Arbitrary choice of hash table size */
485 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
486 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
487 if (rec->rec_lh_hash == NULL) {
492 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
493 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
498 CERROR("Failed to setup %s resource container\n",
499 lnet_res_type2str(type));
500 lnet_res_container_cleanup(rec);
505 lnet_res_containers_destroy(struct lnet_res_container **recs)
507 struct lnet_res_container *rec;
510 cfs_percpt_for_each(rec, i, recs)
511 lnet_res_container_cleanup(rec);
513 cfs_percpt_free(recs);
516 static struct lnet_res_container **
517 lnet_res_containers_create(int type)
519 struct lnet_res_container **recs;
520 struct lnet_res_container *rec;
524 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
526 CERROR("Failed to allocate %s resource containers\n",
527 lnet_res_type2str(type));
531 cfs_percpt_for_each(rec, i, recs) {
532 rc = lnet_res_container_setup(rec, i, type);
534 lnet_res_containers_destroy(recs);
543 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
545 /* ALWAYS called with lnet_res_lock held */
546 struct list_head *head;
547 lnet_libhandle_t *lh;
550 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
553 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
554 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
556 list_for_each_entry(lh, head, lh_hash_chain) {
557 if (lh->lh_cookie == cookie)
565 lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh)
567 /* ALWAYS called with lnet_res_lock held */
568 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
571 lh->lh_cookie = rec->rec_lh_cookie;
572 rec->rec_lh_cookie += 1 << ibits;
574 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
576 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
579 static int lnet_unprepare(void);
582 lnet_prepare(lnet_pid_t requested_pid)
584 /* Prepare to bring up the network */
585 struct lnet_res_container **recs;
588 if (requested_pid == LNET_PID_ANY) {
589 /* Don't instantiate LNET just for me */
593 LASSERT(the_lnet.ln_refcount == 0);
595 the_lnet.ln_routing = 0;
597 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
598 the_lnet.ln_pid = requested_pid;
600 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
601 INIT_LIST_HEAD(&the_lnet.ln_peers);
602 INIT_LIST_HEAD(&the_lnet.ln_nets);
603 INIT_LIST_HEAD(&the_lnet.ln_routers);
604 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
605 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
607 rc = lnet_descriptor_setup();
611 rc = lnet_create_remote_nets_table();
616 * NB the interface cookie in wire handles guards against delayed
617 * replies and ACKs appearing valid after reboot.
619 the_lnet.ln_interface_cookie = ktime_get_real_ns();
621 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
622 sizeof(lnet_counters_t));
623 if (the_lnet.ln_counters == NULL) {
624 CERROR("Failed to allocate counters for LNet\n");
629 rc = lnet_peer_tables_create();
633 rc = lnet_msg_containers_create();
637 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
638 LNET_COOKIE_TYPE_EQ);
642 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
648 the_lnet.ln_me_containers = recs;
650 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
656 the_lnet.ln_md_containers = recs;
658 rc = lnet_portals_create();
660 CERROR("Failed to create portals for LNet: %d\n", rc);
672 lnet_unprepare (void)
674 /* NB no LNET_LOCK since this is the last reference. All LND instances
675 * have shut down already, so it is safe to unlink and free all
676 * descriptors, even those that appear committed to a network op (eg MD
677 * with non-zero pending count) */
679 lnet_fail_nid(LNET_NID_ANY, 0);
681 LASSERT(the_lnet.ln_refcount == 0);
682 LASSERT(list_empty(&the_lnet.ln_test_peers));
683 LASSERT(list_empty(&the_lnet.ln_nets));
685 lnet_portals_destroy();
687 if (the_lnet.ln_md_containers != NULL) {
688 lnet_res_containers_destroy(the_lnet.ln_md_containers);
689 the_lnet.ln_md_containers = NULL;
692 if (the_lnet.ln_me_containers != NULL) {
693 lnet_res_containers_destroy(the_lnet.ln_me_containers);
694 the_lnet.ln_me_containers = NULL;
697 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
699 lnet_msg_containers_destroy();
700 lnet_peer_tables_destroy();
701 lnet_rtrpools_free(0);
703 if (the_lnet.ln_counters != NULL) {
704 cfs_percpt_free(the_lnet.ln_counters);
705 the_lnet.ln_counters = NULL;
707 lnet_destroy_remote_nets_table();
708 lnet_descriptor_cleanup();
714 lnet_net2ni_locked(__u32 net_id, int cpt)
717 struct lnet_net *net;
719 LASSERT(cpt != LNET_LOCK_EX);
721 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
722 if (net->net_id == net_id) {
723 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
733 lnet_net2ni(__u32 net)
738 ni = lnet_net2ni_locked(net, 0);
743 EXPORT_SYMBOL(lnet_net2ni);
746 lnet_get_net_locked(__u32 net_id)
748 struct lnet_net *net;
750 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
751 if (net->net_id == net_id)
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, struct lnet_ni *ni)
780 struct lnet_net *net;
782 /* must called with hold of lnet_net_lock */
783 if (LNET_CPT_NUMBER == 1)
784 return 0; /* the only one */
787 * If NI is provided then use the CPT identified in the NI cpt
788 * list if one exists. If one doesn't exist, then that NI is
789 * associated with all CPTs and it follows that the net it belongs
790 * to is implicitly associated with all CPTs, so just hash the nid
794 if (ni->ni_cpts != NULL)
795 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
798 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
801 /* no NI provided so look at the net */
802 net = lnet_get_net_locked(LNET_NIDNET(nid));
804 if (net != NULL && net->net_cpts != NULL) {
805 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
808 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
812 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
817 if (LNET_CPT_NUMBER == 1)
818 return 0; /* the only one */
820 cpt = lnet_net_lock_current();
822 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
824 lnet_net_unlock(cpt);
828 EXPORT_SYMBOL(lnet_cpt_of_nid);
831 lnet_islocalnet(__u32 net_id)
833 struct lnet_net *net;
837 cpt = lnet_net_lock_current();
839 net = lnet_get_net_locked(net_id);
843 lnet_net_unlock(cpt);
849 lnet_is_ni_healthy_locked(struct lnet_ni *ni)
851 if (ni->ni_state == LNET_NI_STATE_ACTIVE ||
852 ni->ni_state == LNET_NI_STATE_DEGRADED)
859 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
861 struct lnet_net *net;
864 LASSERT(cpt != LNET_LOCK_EX);
866 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
867 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
868 if (ni->ni_nid == nid)
877 lnet_nid2ni_addref(lnet_nid_t nid)
882 ni = lnet_nid2ni_locked(nid, 0);
884 lnet_ni_addref_locked(ni, 0);
889 EXPORT_SYMBOL(lnet_nid2ni_addref);
892 lnet_islocalnid(lnet_nid_t nid)
897 cpt = lnet_net_lock_current();
898 ni = lnet_nid2ni_locked(nid, cpt);
899 lnet_net_unlock(cpt);
905 lnet_count_acceptor_nets(void)
907 /* Return the # of NIs that need the acceptor. */
909 struct lnet_net *net;
912 cpt = lnet_net_lock_current();
913 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
914 /* all socklnd type networks should have the acceptor
916 if (net->net_lnd->lnd_accept != NULL)
920 lnet_net_unlock(cpt);
925 static struct lnet_ping_info *
926 lnet_ping_info_create(int num_ni)
928 struct lnet_ping_info *ping_info;
931 infosz = offsetof(struct lnet_ping_info, pi_ni[num_ni]);
932 LIBCFS_ALLOC(ping_info, infosz);
933 if (ping_info == NULL) {
934 CERROR("Can't allocate ping info[%d]\n", num_ni);
938 ping_info->pi_nnis = num_ni;
939 ping_info->pi_pid = the_lnet.ln_pid;
940 ping_info->pi_magic = LNET_PROTO_PING_MAGIC;
941 ping_info->pi_features = LNET_PING_FEAT_NI_STATUS;
947 lnet_get_net_ni_count_locked(struct lnet_net *net)
952 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
959 lnet_get_ni_count(void)
962 struct lnet_net *net;
967 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
968 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
978 lnet_ping_info_free(struct lnet_ping_info *pinfo)
981 offsetof(struct lnet_ping_info,
982 pi_ni[pinfo->pi_nnis]));
986 lnet_ping_info_destroy(void)
988 struct lnet_net *net;
991 lnet_net_lock(LNET_LOCK_EX);
993 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
994 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
996 ni->ni_status = NULL;
1001 lnet_ping_info_free(the_lnet.ln_ping_info);
1002 the_lnet.ln_ping_info = NULL;
1004 lnet_net_unlock(LNET_LOCK_EX);
1008 lnet_ping_event_handler(lnet_event_t *event)
1010 struct lnet_ping_info *pinfo = event->md.user_ptr;
1012 if (event->unlinked)
1013 pinfo->pi_features = LNET_PING_FEAT_INVAL;
1017 lnet_ping_info_setup(struct lnet_ping_info **ppinfo, lnet_handle_md_t *md_handle,
1018 int ni_count, bool set_eq)
1020 lnet_handle_me_t me_handle;
1021 lnet_process_id_t id = {LNET_NID_ANY, LNET_PID_ANY};
1022 lnet_md_t md = {NULL};
1026 rc = LNetEQAlloc(0, lnet_ping_event_handler,
1027 &the_lnet.ln_ping_target_eq);
1029 CERROR("Can't allocate ping EQ: %d\n", rc);
1034 *ppinfo = lnet_ping_info_create(ni_count);
1035 if (*ppinfo == NULL) {
1040 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1041 LNET_PROTO_PING_MATCHBITS, 0,
1042 LNET_UNLINK, LNET_INS_AFTER,
1045 CERROR("Can't create ping ME: %d\n", rc);
1049 /* initialize md content */
1051 md.length = offsetof(struct lnet_ping_info,
1052 pi_ni[(*ppinfo)->pi_nnis]);
1053 md.threshold = LNET_MD_THRESH_INF;
1055 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1056 LNET_MD_MANAGE_REMOTE;
1058 md.eq_handle = the_lnet.ln_ping_target_eq;
1059 md.user_ptr = *ppinfo;
1061 rc = LNetMDAttach(me_handle, md, LNET_RETAIN, md_handle);
1063 CERROR("Can't attach ping MD: %d\n", rc);
1070 rc2 = LNetMEUnlink(me_handle);
1073 lnet_ping_info_free(*ppinfo);
1077 LNetEQFree(the_lnet.ln_ping_target_eq);
1082 lnet_ping_md_unlink(struct lnet_ping_info *pinfo, lnet_handle_md_t *md_handle)
1084 sigset_t blocked = cfs_block_allsigs();
1086 LNetMDUnlink(*md_handle);
1087 LNetInvalidateHandle(md_handle);
1089 /* NB md could be busy; this just starts the unlink */
1090 while (pinfo->pi_features != LNET_PING_FEAT_INVAL) {
1091 CDEBUG(D_NET, "Still waiting for ping MD to unlink\n");
1092 set_current_state(TASK_UNINTERRUPTIBLE);
1093 schedule_timeout(cfs_time_seconds(1));
1096 cfs_restore_sigs(blocked);
1100 lnet_ping_info_install_locked(struct lnet_ping_info *ping_info)
1104 struct lnet_net *net;
1105 struct lnet_ni_status *ns;
1108 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1109 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1110 LASSERT(i < ping_info->pi_nnis);
1112 ns = &ping_info->pi_ni[i];
1114 ns->ns_nid = ni->ni_nid;
1117 ns->ns_status = (ni->ni_status != NULL) ?
1118 ni->ni_status->ns_status :
1130 lnet_ping_target_update(struct lnet_ping_info *pinfo, lnet_handle_md_t md_handle)
1132 struct lnet_ping_info *old_pinfo = NULL;
1133 lnet_handle_md_t old_md;
1135 /* switch the NIs to point to the new ping info created */
1136 lnet_net_lock(LNET_LOCK_EX);
1138 if (!the_lnet.ln_routing)
1139 pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1140 lnet_ping_info_install_locked(pinfo);
1142 if (the_lnet.ln_ping_info != NULL) {
1143 old_pinfo = the_lnet.ln_ping_info;
1144 old_md = the_lnet.ln_ping_target_md;
1146 the_lnet.ln_ping_target_md = md_handle;
1147 the_lnet.ln_ping_info = pinfo;
1149 lnet_net_unlock(LNET_LOCK_EX);
1151 if (old_pinfo != NULL) {
1152 /* unlink the old ping info */
1153 lnet_ping_md_unlink(old_pinfo, &old_md);
1154 lnet_ping_info_free(old_pinfo);
1159 lnet_ping_target_fini(void)
1163 lnet_ping_md_unlink(the_lnet.ln_ping_info,
1164 &the_lnet.ln_ping_target_md);
1166 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1169 lnet_ping_info_destroy();
1173 lnet_ni_tq_credits(lnet_ni_t *ni)
1177 LASSERT(ni->ni_ncpts >= 1);
1179 if (ni->ni_ncpts == 1)
1180 return ni->ni_net->net_tunables.lct_max_tx_credits;
1182 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
1183 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
1184 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
1190 lnet_ni_unlink_locked(lnet_ni_t *ni)
1192 if (!list_empty(&ni->ni_cptlist)) {
1193 list_del_init(&ni->ni_cptlist);
1194 lnet_ni_decref_locked(ni, 0);
1197 /* move it to zombie list and nobody can find it anymore */
1198 LASSERT(!list_empty(&ni->ni_netlist));
1199 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
1200 lnet_ni_decref_locked(ni, 0);
1204 lnet_clear_zombies_nis_locked(struct lnet_net *net)
1209 struct list_head *zombie_list = &net->net_ni_zombie;
1212 * Now wait for the NIs I just nuked to show up on the zombie
1213 * list and shut them down in guaranteed thread context
1216 while (!list_empty(zombie_list)) {
1220 ni = list_entry(zombie_list->next,
1221 lnet_ni_t, ni_netlist);
1222 list_del_init(&ni->ni_netlist);
1223 /* the ni should be in deleting state. If it's not it's
1225 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
1226 cfs_percpt_for_each(ref, j, ni->ni_refs) {
1229 /* still busy, add it back to zombie list */
1230 list_add(&ni->ni_netlist, zombie_list);
1234 if (!list_empty(&ni->ni_netlist)) {
1235 lnet_net_unlock(LNET_LOCK_EX);
1237 if ((i & (-i)) == i) {
1239 "Waiting for zombie LNI %s\n",
1240 libcfs_nid2str(ni->ni_nid));
1242 set_current_state(TASK_UNINTERRUPTIBLE);
1243 schedule_timeout(cfs_time_seconds(1));
1244 lnet_net_lock(LNET_LOCK_EX);
1248 lnet_net_unlock(LNET_LOCK_EX);
1250 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
1252 LASSERT(!in_interrupt());
1253 (net->net_lnd->lnd_shutdown)(ni);
1256 CDEBUG(D_LNI, "Removed LNI %s\n",
1257 libcfs_nid2str(ni->ni_nid));
1261 lnet_net_lock(LNET_LOCK_EX);
1265 /* shutdown down the NI and release refcount */
1267 lnet_shutdown_lndni(struct lnet_ni *ni)
1270 struct lnet_net *net = ni->ni_net;
1272 lnet_net_lock(LNET_LOCK_EX);
1273 ni->ni_state = LNET_NI_STATE_DELETING;
1274 lnet_ni_unlink_locked(ni);
1275 lnet_net_unlock(LNET_LOCK_EX);
1277 /* clear messages for this NI on the lazy portal */
1278 for (i = 0; i < the_lnet.ln_nportals; i++)
1279 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
1281 /* Do peer table cleanup for this ni */
1282 lnet_peer_tables_cleanup(ni);
1284 lnet_net_lock(LNET_LOCK_EX);
1285 lnet_clear_zombies_nis_locked(net);
1286 lnet_net_unlock(LNET_LOCK_EX);
1290 lnet_shutdown_lndnet(struct lnet_net *net)
1294 lnet_net_lock(LNET_LOCK_EX);
1296 net->net_state = LNET_NET_STATE_DELETING;
1298 list_del_init(&net->net_list);
1300 while (!list_empty(&net->net_ni_list)) {
1301 ni = list_entry(net->net_ni_list.next,
1302 lnet_ni_t, ni_netlist);
1303 lnet_net_unlock(LNET_LOCK_EX);
1304 lnet_shutdown_lndni(ni);
1305 lnet_net_lock(LNET_LOCK_EX);
1309 * decrement ref count on lnd only when the entire network goes
1312 net->net_lnd->lnd_refcount--;
1314 lnet_net_unlock(LNET_LOCK_EX);
1320 lnet_shutdown_lndnets(void)
1322 struct lnet_net *net;
1324 /* NB called holding the global mutex */
1326 /* All quiet on the API front */
1327 LASSERT(!the_lnet.ln_shutdown);
1328 LASSERT(the_lnet.ln_refcount == 0);
1330 lnet_net_lock(LNET_LOCK_EX);
1331 the_lnet.ln_shutdown = 1; /* flag shutdown */
1333 while (!list_empty(&the_lnet.ln_nets)) {
1335 * move the nets to the zombie list to avoid them being
1336 * picked up for new work. LONET is also included in the
1337 * Nets that will be moved to the zombie list
1339 net = list_entry(the_lnet.ln_nets.next,
1340 struct lnet_net, net_list);
1341 list_move(&net->net_list, &the_lnet.ln_net_zombie);
1344 /* Drop the cached loopback Net. */
1345 if (the_lnet.ln_loni != NULL) {
1346 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
1347 the_lnet.ln_loni = NULL;
1349 lnet_net_unlock(LNET_LOCK_EX);
1351 /* iterate through the net zombie list and delete each net */
1352 while (!list_empty(&the_lnet.ln_net_zombie)) {
1353 net = list_entry(the_lnet.ln_net_zombie.next,
1354 struct lnet_net, net_list);
1355 lnet_shutdown_lndnet(net);
1358 lnet_net_lock(LNET_LOCK_EX);
1359 the_lnet.ln_shutdown = 0;
1360 lnet_net_unlock(LNET_LOCK_EX);
1364 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
1367 struct lnet_tx_queue *tq;
1369 struct lnet_net *net = ni->ni_net;
1371 mutex_lock(&the_lnet.ln_lnd_mutex);
1374 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
1375 ni->ni_lnd_tunables_set = true;
1378 rc = (net->net_lnd->lnd_startup)(ni);
1380 mutex_unlock(&the_lnet.ln_lnd_mutex);
1383 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1384 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
1385 lnet_net_lock(LNET_LOCK_EX);
1386 net->net_lnd->lnd_refcount--;
1387 lnet_net_unlock(LNET_LOCK_EX);
1391 ni->ni_state = LNET_NI_STATE_ACTIVE;
1393 /* We keep a reference on the loopback net through the loopback NI */
1394 if (net->net_lnd->lnd_type == LOLND) {
1396 LASSERT(the_lnet.ln_loni == NULL);
1397 the_lnet.ln_loni = ni;
1398 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
1399 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
1400 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
1401 ni->ni_net->net_tunables.lct_peer_timeout = 0;
1405 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
1406 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
1407 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1408 libcfs_lnd2str(net->net_lnd->lnd_type),
1409 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
1411 /* shutdown the NI since if we get here then it must've already
1414 lnet_shutdown_lndni(ni);
1418 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
1419 tq->tq_credits_min =
1420 tq->tq_credits_max =
1421 tq->tq_credits = lnet_ni_tq_credits(ni);
1424 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
1425 libcfs_nid2str(ni->ni_nid),
1426 ni->ni_net->net_tunables.lct_peer_tx_credits,
1427 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
1428 ni->ni_net->net_tunables.lct_peer_rtr_credits,
1429 ni->ni_net->net_tunables.lct_peer_timeout);
1438 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
1441 struct lnet_net *net_l = NULL;
1442 struct list_head local_ni_list;
1448 net->net_tunables.lct_peer_timeout;
1450 net->net_tunables.lct_max_tx_credits;
1451 int peerrtrcredits =
1452 net->net_tunables.lct_peer_rtr_credits;
1454 INIT_LIST_HEAD(&local_ni_list);
1457 * make sure that this net is unique. If it isn't then
1458 * we are adding interfaces to an already existing network, and
1459 * 'net' is just a convenient way to pass in the list.
1460 * if it is unique we need to find the LND and load it if
1463 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
1464 lnd_type = LNET_NETTYP(net->net_id);
1466 LASSERT(libcfs_isknown_lnd(lnd_type));
1468 if (lnd_type == CIBLND || lnd_type == OPENIBLND ||
1469 lnd_type == IIBLND || lnd_type == VIBLND) {
1470 CERROR("LND %s obsoleted\n", libcfs_lnd2str(lnd_type));
1475 mutex_lock(&the_lnet.ln_lnd_mutex);
1476 lnd = lnet_find_lnd_by_type(lnd_type);
1479 mutex_unlock(&the_lnet.ln_lnd_mutex);
1480 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
1481 mutex_lock(&the_lnet.ln_lnd_mutex);
1483 lnd = lnet_find_lnd_by_type(lnd_type);
1485 mutex_unlock(&the_lnet.ln_lnd_mutex);
1486 CERROR("Can't load LND %s, module %s, rc=%d\n",
1487 libcfs_lnd2str(lnd_type),
1488 libcfs_lnd2modname(lnd_type), rc);
1489 #ifndef HAVE_MODULE_LOADING_SUPPORT
1490 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
1491 "compiled with kernel module "
1492 "loading support.");
1499 lnet_net_lock(LNET_LOCK_EX);
1500 lnd->lnd_refcount++;
1501 lnet_net_unlock(LNET_LOCK_EX);
1505 mutex_unlock(&the_lnet.ln_lnd_mutex);
1511 * net_l: if the network being added is unique then net_l
1512 * will point to that network
1513 * if the network being added is not unique then
1514 * net_l points to the existing network.
1516 * When we enter the loop below, we'll pick NIs off he
1517 * network beign added and start them up, then add them to
1518 * a local ni list. Once we've successfully started all
1519 * the NIs then we join the local NI list (of started up
1520 * networks) with the net_l->net_ni_list, which should
1521 * point to the correct network to add the new ni list to
1523 * If any of the new NIs fail to start up, then we want to
1524 * iterate through the local ni list, which should include
1525 * any NIs which were successfully started up, and shut
1528 * After than we want to delete the network being added,
1529 * to avoid a memory leak.
1533 * When a network uses TCP bonding then all its interfaces
1534 * must be specified when the network is first defined: the
1535 * TCP bonding code doesn't allow for interfaces to be added
1538 if (net_l != net && net_l != NULL && use_tcp_bonding &&
1539 LNET_NETTYP(net_l->net_id) == SOCKLND) {
1544 while (!list_empty(&net->net_ni_added)) {
1545 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
1547 list_del_init(&ni->ni_netlist);
1549 /* make sure that the the NI we're about to start
1550 * up is actually unique. if it's not fail. */
1551 if (!lnet_ni_unique_net(&net_l->net_ni_list,
1552 ni->ni_interfaces[0])) {
1557 /* adjust the pointer the parent network, just in case it
1558 * the net is a duplicate */
1561 rc = lnet_startup_lndni(ni, tun);
1563 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
1564 ni->ni_net->net_lnd->lnd_query != NULL);
1570 list_add_tail(&ni->ni_netlist, &local_ni_list);
1575 lnet_net_lock(LNET_LOCK_EX);
1576 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
1577 lnet_incr_dlc_seq();
1578 lnet_net_unlock(LNET_LOCK_EX);
1580 /* if the network is not unique then we don't want to keep
1581 * it around after we're done. Free it. Otherwise add that
1582 * net to the global the_lnet.ln_nets */
1583 if (net_l != net && net_l != NULL) {
1585 * TODO - note. currently the tunables can not be updated
1590 net->net_state = LNET_NET_STATE_ACTIVE;
1592 * restore tunables after it has been overwitten by the
1595 if (peer_timeout != -1)
1596 net->net_tunables.lct_peer_timeout = peer_timeout;
1597 if (maxtxcredits != -1)
1598 net->net_tunables.lct_max_tx_credits = maxtxcredits;
1599 if (peerrtrcredits != -1)
1600 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
1602 lnet_net_lock(LNET_LOCK_EX);
1603 list_add_tail(&net->net_list, &the_lnet.ln_nets);
1604 lnet_net_unlock(LNET_LOCK_EX);
1611 * shutdown the new NIs that are being started up
1612 * free the NET being started
1614 while (!list_empty(&local_ni_list)) {
1615 ni = list_entry(local_ni_list.next, struct lnet_ni,
1618 lnet_shutdown_lndni(ni);
1628 lnet_startup_lndnets(struct list_head *netlist)
1630 struct lnet_net *net;
1634 while (!list_empty(netlist)) {
1635 net = list_entry(netlist->next, struct lnet_net, net_list);
1636 list_del_init(&net->net_list);
1638 rc = lnet_startup_lndnet(net, NULL);
1648 lnet_shutdown_lndnets();
1654 * Initialize LNet library.
1656 * Automatically called at module loading time. Caller has to call
1657 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
1658 * latter returned 0. It must be called exactly once.
1660 * \retval 0 on success
1661 * \retval -ve on failures.
1663 int lnet_lib_init(void)
1667 lnet_assert_wire_constants();
1669 memset(&the_lnet, 0, sizeof(the_lnet));
1671 /* refer to global cfs_cpt_table for now */
1672 the_lnet.ln_cpt_table = cfs_cpt_table;
1673 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
1675 LASSERT(the_lnet.ln_cpt_number > 0);
1676 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
1677 /* we are under risk of consuming all lh_cookie */
1678 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
1679 "please change setting of CPT-table and retry\n",
1680 the_lnet.ln_cpt_number, LNET_CPT_MAX);
1684 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
1685 the_lnet.ln_cpt_bits++;
1687 rc = lnet_create_locks();
1689 CERROR("Can't create LNet global locks: %d\n", rc);
1693 the_lnet.ln_refcount = 0;
1694 LNetInvalidateHandle(&the_lnet.ln_rc_eqh);
1695 INIT_LIST_HEAD(&the_lnet.ln_lnds);
1696 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
1697 INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
1698 INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
1700 /* The hash table size is the number of bits it takes to express the set
1701 * ln_num_routes, minus 1 (better to under estimate than over so we
1702 * don't waste memory). */
1703 if (rnet_htable_size <= 0)
1704 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
1705 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
1706 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
1707 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
1708 order_base_2(rnet_htable_size) - 1);
1710 /* All LNDs apart from the LOLND are in separate modules. They
1711 * register themselves when their module loads, and unregister
1712 * themselves when their module is unloaded. */
1713 lnet_register_lnd(&the_lolnd);
1718 * Finalize LNet library.
1720 * \pre lnet_lib_init() called with success.
1721 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1723 void lnet_lib_exit(void)
1725 LASSERT(the_lnet.ln_refcount == 0);
1727 while (!list_empty(&the_lnet.ln_lnds))
1728 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
1730 lnet_destroy_locks();
1734 * Set LNet PID and start LNet interfaces, routing, and forwarding.
1736 * Users must call this function at least once before any other functions.
1737 * For each successful call there must be a corresponding call to
1738 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1741 * The PID used by LNet may be different from the one requested.
1744 * \param requested_pid PID requested by the caller.
1746 * \return >= 0 on success, and < 0 error code on failures.
1749 LNetNIInit(lnet_pid_t requested_pid)
1751 int im_a_router = 0;
1754 struct lnet_ping_info *pinfo;
1755 lnet_handle_md_t md_handle;
1756 struct list_head net_head;
1757 struct lnet_net *net;
1759 INIT_LIST_HEAD(&net_head);
1761 mutex_lock(&the_lnet.ln_api_mutex);
1763 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
1765 if (the_lnet.ln_refcount > 0) {
1766 rc = the_lnet.ln_refcount++;
1767 mutex_unlock(&the_lnet.ln_api_mutex);
1771 rc = lnet_prepare(requested_pid);
1773 mutex_unlock(&the_lnet.ln_api_mutex);
1777 /* create a network for Loopback network */
1778 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
1781 goto err_empty_list;
1784 /* Add in the loopback NI */
1785 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
1787 goto err_empty_list;
1790 /* If LNet is being initialized via DLC it is possible
1791 * that the user requests not to load module parameters (ones which
1792 * are supported by DLC) on initialization. Therefore, make sure not
1793 * to load networks, routes and forwarding from module parameters
1794 * in this case. On cleanup in case of failure only clean up
1795 * routes if it has been loaded */
1796 if (!the_lnet.ln_nis_from_mod_params) {
1797 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
1800 goto err_empty_list;
1803 ni_count = lnet_startup_lndnets(&net_head);
1806 goto err_empty_list;
1809 if (!the_lnet.ln_nis_from_mod_params) {
1810 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
1812 goto err_shutdown_lndnis;
1814 rc = lnet_check_routes();
1816 goto err_destroy_routes;
1818 rc = lnet_rtrpools_alloc(im_a_router);
1820 goto err_destroy_routes;
1823 rc = lnet_acceptor_start();
1825 goto err_destroy_routes;
1827 the_lnet.ln_refcount = 1;
1828 /* Now I may use my own API functions... */
1830 rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count, true);
1832 goto err_acceptor_stop;
1834 lnet_ping_target_update(pinfo, md_handle);
1836 rc = lnet_router_checker_start();
1843 mutex_unlock(&the_lnet.ln_api_mutex);
1848 lnet_ping_target_fini();
1850 the_lnet.ln_refcount = 0;
1851 lnet_acceptor_stop();
1853 if (!the_lnet.ln_nis_from_mod_params)
1854 lnet_destroy_routes();
1855 err_shutdown_lndnis:
1856 lnet_shutdown_lndnets();
1860 mutex_unlock(&the_lnet.ln_api_mutex);
1861 while (!list_empty(&net_head)) {
1862 struct lnet_net *net;
1864 net = list_entry(net_head.next, struct lnet_net, net_list);
1865 list_del_init(&net->net_list);
1870 EXPORT_SYMBOL(LNetNIInit);
1873 * Stop LNet interfaces, routing, and forwarding.
1875 * Users must call this function once for each successful call to LNetNIInit().
1876 * Once the LNetNIFini() operation has been started, the results of pending
1877 * API operations are undefined.
1879 * \return always 0 for current implementation.
1884 mutex_lock(&the_lnet.ln_api_mutex);
1886 LASSERT(the_lnet.ln_refcount > 0);
1888 if (the_lnet.ln_refcount != 1) {
1889 the_lnet.ln_refcount--;
1891 LASSERT(!the_lnet.ln_niinit_self);
1896 lnet_router_checker_stop();
1897 lnet_ping_target_fini();
1899 /* Teardown fns that use my own API functions BEFORE here */
1900 the_lnet.ln_refcount = 0;
1902 lnet_acceptor_stop();
1903 lnet_destroy_routes();
1904 lnet_shutdown_lndnets();
1908 mutex_unlock(&the_lnet.ln_api_mutex);
1911 EXPORT_SYMBOL(LNetNIFini);
1914 * Grabs the ni data from the ni structure and fills the out
1917 * \param[in] ni network interface structure
1918 * \param[out] cpt_count the number of cpts the ni is on
1919 * \param[out] nid Network Interface ID
1920 * \param[out] peer_timeout NI peer timeout
1921 * \param[out] peer_tx_crdits NI peer transmit credits
1922 * \param[out] peer_rtr_credits NI peer router credits
1923 * \param[out] max_tx_credits NI max transmit credit
1924 * \param[out] net_config Network configuration
1927 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_data *config)
1929 struct lnet_ioctl_net_config *net_config;
1930 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
1931 size_t min_size, tunable_size = 0;
1937 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
1941 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
1942 ARRAY_SIZE(net_config->ni_interfaces));
1944 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
1945 if (!ni->ni_interfaces[i])
1948 strncpy(net_config->ni_interfaces[i],
1949 ni->ni_interfaces[i],
1950 sizeof(net_config->ni_interfaces[i]));
1953 config->cfg_nid = ni->ni_nid;
1954 config->cfg_config_u.cfg_net.net_peer_timeout =
1955 ni->ni_net->net_tunables.lct_peer_timeout;
1956 config->cfg_config_u.cfg_net.net_max_tx_credits =
1957 ni->ni_net->net_tunables.lct_max_tx_credits;
1958 config->cfg_config_u.cfg_net.net_peer_tx_credits =
1959 ni->ni_net->net_tunables.lct_peer_tx_credits;
1960 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
1961 ni->ni_net->net_tunables.lct_peer_rtr_credits;
1963 net_config->ni_status = ni->ni_status->ns_status;
1966 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
1968 for (i = 0; i < num_cpts; i++)
1969 net_config->ni_cpts[i] = ni->ni_cpts[i];
1971 config->cfg_ncpts = num_cpts;
1975 * See if user land tools sent in a newer and larger version
1976 * of struct lnet_tunables than what the kernel uses.
1978 min_size = sizeof(*config) + sizeof(*net_config);
1980 if (config->cfg_hdr.ioc_len > min_size)
1981 tunable_size = config->cfg_hdr.ioc_len - min_size;
1983 /* Don't copy too much data to user space */
1984 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
1985 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
1987 if (lnd_cfg && min_size) {
1988 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
1989 config->cfg_config_u.cfg_net.net_interface_count = 1;
1991 /* Tell user land that kernel side has less data */
1992 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
1993 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
1994 config->cfg_hdr.ioc_len -= min_size;
2000 lnet_get_ni_idx_locked(int idx)
2003 struct lnet_net *net;
2005 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2006 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2016 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2019 struct lnet_net *net = mynet;
2023 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2025 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2031 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2032 /* if you reached the end of the ni list and the net is
2033 * specified, then there are no more nis in that net */
2037 /* we reached the end of this net ni list. move to the
2039 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2040 /* no more nets and no more NIs. */
2043 /* get the next net */
2044 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2046 /* get the ni on it */
2047 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2053 /* there are more nis left */
2054 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2060 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2065 int idx = config->cfg_count;
2067 cpt = lnet_net_lock_current();
2069 ni = lnet_get_ni_idx_locked(idx);
2074 lnet_fill_ni_info(ni, config);
2078 lnet_net_unlock(cpt);
2083 lnet_dyn_add_ni(lnet_pid_t requested_pid, struct lnet_ioctl_config_data *conf)
2085 char *nets = conf->cfg_config_u.cfg_net.net_intf;
2086 struct lnet_ping_info *pinfo;
2087 lnet_handle_md_t md_handle;
2088 struct lnet_net *net;
2089 struct list_head net_head;
2091 lnet_remotenet_t *rnet;
2093 int num_acceptor_nets;
2095 struct lnet_ioctl_config_lnd_tunables *lnd_tunables = NULL;
2097 INIT_LIST_HEAD(&net_head);
2099 if (conf && conf->cfg_hdr.ioc_len > sizeof(*conf))
2100 lnd_tunables = (struct lnet_ioctl_config_lnd_tunables *)conf->cfg_bulk;
2102 /* Create a net/ni structures for the network string */
2103 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
2105 return rc == 0 ? -EINVAL : rc;
2107 mutex_lock(&the_lnet.ln_api_mutex);
2110 rc = -EINVAL; /* only add one network per call */
2114 net = list_entry(net_head.next, struct lnet_net, net_list);
2116 lnet_net_lock(LNET_LOCK_EX);
2117 rnet = lnet_find_rnet_locked(net->net_id);
2118 lnet_net_unlock(LNET_LOCK_EX);
2119 /* make sure that the net added doesn't invalidate the current
2120 * configuration LNet is keeping */
2122 CERROR("Adding net %s will invalidate routing configuration\n",
2129 * make sure you calculate the correct number of slots in the ping
2130 * info. Since the ping info is a flattened list of all the NIs,
2131 * we should allocate enough slots to accomodate the number of NIs
2132 * which will be added.
2134 * We can use lnet_get_net_ni_count_locked() since the net is not
2135 * on a public list yet, so locking is not a problem
2137 net_ni_count = lnet_get_net_ni_count_locked(net);
2139 rc = lnet_ping_info_setup(&pinfo, &md_handle,
2140 net_ni_count + lnet_get_ni_count(),
2145 list_del_init(&net->net_list);
2148 memcpy(&net->net_tunables,
2149 &lnd_tunables->lt_cmn, sizeof(lnd_tunables->lt_cmn));
2152 * before starting this network get a count of the current TCP
2153 * networks which require the acceptor thread running. If that
2154 * count is == 0 before we start up this network, then we'd want to
2155 * start up the acceptor thread after starting up this network
2157 num_acceptor_nets = lnet_count_acceptor_nets();
2160 * lnd_startup_lndnet() can deallocate 'net' even if it it returns
2161 * success, because we endded up adding interfaces to an existing
2162 * network. So grab the net_type now
2164 net_type = LNET_NETTYP(net->net_id);
2166 rc = lnet_startup_lndnet(net,
2167 (lnd_tunables) ? &lnd_tunables->lt_tun : NULL);
2172 * Start the acceptor thread if this is the first network
2173 * being added that requires the thread.
2175 if (net_type == SOCKLND && num_acceptor_nets == 0)
2177 rc = lnet_acceptor_start();
2179 /* shutdown the net that we just started */
2180 CERROR("Failed to start up acceptor thread\n");
2182 * Note that if we needed to start the acceptor
2183 * thread, then 'net' must have been the first TCP
2184 * network, therefore was unique, and therefore
2185 * wasn't deallocated by lnet_startup_lndnet()
2187 lnet_shutdown_lndnet(net);
2192 lnet_ping_target_update(pinfo, md_handle);
2193 mutex_unlock(&the_lnet.ln_api_mutex);
2198 lnet_ping_md_unlink(pinfo, &md_handle);
2199 lnet_ping_info_free(pinfo);
2201 mutex_unlock(&the_lnet.ln_api_mutex);
2202 while (!list_empty(&net_head)) {
2203 net = list_entry(net_head.next, struct lnet_net, net_list);
2204 list_del_init(&net->net_list);
2211 lnet_dyn_del_ni(__u32 net_id)
2213 struct lnet_net *net;
2214 struct lnet_ping_info *pinfo;
2215 lnet_handle_md_t md_handle;
2219 /* don't allow userspace to shutdown the LOLND */
2220 if (LNET_NETTYP(net_id) == LOLND)
2223 mutex_lock(&the_lnet.ln_api_mutex);
2227 net = lnet_get_net_locked(net_id);
2233 net_ni_count = lnet_get_net_ni_count_locked(net);
2237 /* create and link a new ping info, before removing the old one */
2238 rc = lnet_ping_info_setup(&pinfo, &md_handle,
2239 lnet_get_ni_count() - net_ni_count, false);
2243 lnet_shutdown_lndnet(net);
2245 if (lnet_count_acceptor_nets() == 0)
2246 lnet_acceptor_stop();
2248 lnet_ping_target_update(pinfo, md_handle);
2251 mutex_unlock(&the_lnet.ln_api_mutex);
2256 void lnet_incr_dlc_seq(void)
2258 atomic_inc(&lnet_dlc_seq_no);
2261 __u32 lnet_get_dlc_seq_locked(void)
2263 return atomic_read(&lnet_dlc_seq_no);
2267 * LNet ioctl handler.
2271 LNetCtl(unsigned int cmd, void *arg)
2273 struct libcfs_ioctl_data *data = arg;
2274 struct lnet_ioctl_config_data *config;
2275 lnet_process_id_t id = {0};
2279 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
2280 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
2283 case IOC_LIBCFS_GET_NI:
2284 rc = LNetGetId(data->ioc_count, &id);
2285 data->ioc_nid = id.nid;
2288 case IOC_LIBCFS_FAIL_NID:
2289 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
2291 case IOC_LIBCFS_ADD_ROUTE:
2294 if (config->cfg_hdr.ioc_len < sizeof(*config))
2297 mutex_lock(&the_lnet.ln_api_mutex);
2298 rc = lnet_add_route(config->cfg_net,
2299 config->cfg_config_u.cfg_route.rtr_hop,
2301 config->cfg_config_u.cfg_route.
2304 rc = lnet_check_routes();
2306 lnet_del_route(config->cfg_net,
2309 mutex_unlock(&the_lnet.ln_api_mutex);
2312 case IOC_LIBCFS_DEL_ROUTE:
2315 if (config->cfg_hdr.ioc_len < sizeof(*config))
2318 mutex_lock(&the_lnet.ln_api_mutex);
2319 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
2320 mutex_unlock(&the_lnet.ln_api_mutex);
2323 case IOC_LIBCFS_GET_ROUTE:
2326 if (config->cfg_hdr.ioc_len < sizeof(*config))
2329 return lnet_get_route(config->cfg_count,
2331 &config->cfg_config_u.cfg_route.rtr_hop,
2333 &config->cfg_config_u.cfg_route.rtr_flags,
2334 &config->cfg_config_u.cfg_route.
2337 case IOC_LIBCFS_GET_NET: {
2338 size_t total = sizeof(*config) +
2339 sizeof(struct lnet_ioctl_net_config);
2342 if (config->cfg_hdr.ioc_len < total)
2345 return lnet_get_net_config(config);
2348 case IOC_LIBCFS_GET_LNET_STATS:
2350 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
2352 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
2355 lnet_counters_get(&lnet_stats->st_cntrs);
2359 case IOC_LIBCFS_CONFIG_RTR:
2362 if (config->cfg_hdr.ioc_len < sizeof(*config))
2365 mutex_lock(&the_lnet.ln_api_mutex);
2366 if (config->cfg_config_u.cfg_buffers.buf_enable) {
2367 rc = lnet_rtrpools_enable();
2368 mutex_unlock(&the_lnet.ln_api_mutex);
2371 lnet_rtrpools_disable();
2372 mutex_unlock(&the_lnet.ln_api_mutex);
2375 case IOC_LIBCFS_ADD_BUF:
2378 if (config->cfg_hdr.ioc_len < sizeof(*config))
2381 mutex_lock(&the_lnet.ln_api_mutex);
2382 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
2384 config->cfg_config_u.cfg_buffers.
2386 config->cfg_config_u.cfg_buffers.
2388 mutex_unlock(&the_lnet.ln_api_mutex);
2391 case IOC_LIBCFS_GET_BUF: {
2392 struct lnet_ioctl_pool_cfg *pool_cfg;
2393 size_t total = sizeof(*config) + sizeof(*pool_cfg);
2397 if (config->cfg_hdr.ioc_len < total)
2400 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
2401 return lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
2404 case IOC_LIBCFS_GET_PEER_INFO: {
2405 struct lnet_ioctl_peer *peer_info = arg;
2407 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
2410 return lnet_get_peer_info(
2411 peer_info->pr_count,
2413 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
2414 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
2415 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
2416 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
2417 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
2418 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
2419 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_rtr_credits,
2420 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
2423 case IOC_LIBCFS_NOTIFY_ROUTER: {
2424 unsigned long jiffies_passed;
2426 jiffies_passed = ktime_get_real_seconds() - data->ioc_u64[0];
2427 jiffies_passed = cfs_time_seconds(jiffies_passed);
2429 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
2430 jiffies - jiffies_passed);
2433 case IOC_LIBCFS_LNET_DIST:
2434 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
2435 if (rc < 0 && rc != -EHOSTUNREACH)
2438 data->ioc_u32[0] = rc;
2441 case IOC_LIBCFS_TESTPROTOCOMPAT:
2442 lnet_net_lock(LNET_LOCK_EX);
2443 the_lnet.ln_testprotocompat = data->ioc_flags;
2444 lnet_net_unlock(LNET_LOCK_EX);
2447 case IOC_LIBCFS_LNET_FAULT:
2448 return lnet_fault_ctl(data->ioc_flags, data);
2450 case IOC_LIBCFS_PING: {
2451 signed long timeout;
2453 id.nid = data->ioc_nid;
2454 id.pid = data->ioc_u32[0];
2456 /* Don't block longer than 2 minutes */
2457 if (data->ioc_u32[1] > 120 * MSEC_PER_SEC)
2460 /* If timestamp is negative then disable timeout */
2461 if ((s32)data->ioc_u32[1] < 0)
2462 timeout = MAX_SCHEDULE_TIMEOUT;
2464 timeout = msecs_to_jiffies(data->ioc_u32[1]);
2466 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
2467 data->ioc_plen1 / sizeof(lnet_process_id_t));
2470 data->ioc_count = rc;
2474 ni = lnet_net2ni(data->ioc_net);
2478 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
2481 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
2487 EXPORT_SYMBOL(LNetCtl);
2489 void LNetDebugPeer(lnet_process_id_t id)
2491 lnet_debug_peer(id.nid);
2493 EXPORT_SYMBOL(LNetDebugPeer);
2496 * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
2497 * all interfaces share a same PID, as requested by LNetNIInit().
2499 * \param index Index of the interface to look up.
2500 * \param id On successful return, this location will hold the
2501 * lnet_process_id_t ID of the interface.
2503 * \retval 0 If an interface exists at \a index.
2504 * \retval -ENOENT If no interface has been found.
2507 LNetGetId(unsigned int index, lnet_process_id_t *id)
2510 struct lnet_net *net;
2514 LASSERT(the_lnet.ln_refcount > 0);
2516 cpt = lnet_net_lock_current();
2518 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2519 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2523 id->nid = ni->ni_nid;
2524 id->pid = the_lnet.ln_pid;
2530 lnet_net_unlock(cpt);
2533 EXPORT_SYMBOL(LNetGetId);
2536 * Print a string representation of handle \a h into buffer \a str of
2540 LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
2542 snprintf(str, len, "%#llx", h.cookie);
2544 EXPORT_SYMBOL(LNetSnprintHandle);
2546 static int lnet_ping(lnet_process_id_t id, signed long timeout,
2547 lnet_process_id_t __user *ids, int n_ids)
2549 lnet_handle_eq_t eqh;
2550 lnet_handle_md_t mdh;
2552 lnet_md_t md = { NULL };
2556 const signed long a_long_time = msecs_to_jiffies(60 * MSEC_PER_SEC);
2558 struct lnet_ping_info *info;
2559 lnet_process_id_t tmpid;
2566 infosz = offsetof(struct lnet_ping_info, pi_ni[n_ids]);
2568 /* n_ids limit is arbitrary */
2569 if (n_ids <= 0 || n_ids > 20 || id.nid == LNET_NID_ANY)
2572 if (id.pid == LNET_PID_ANY)
2573 id.pid = LNET_PID_LUSTRE;
2575 LIBCFS_ALLOC(info, infosz);
2579 /* NB 2 events max (including any unlink event) */
2580 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
2582 CERROR("Can't allocate EQ: %d\n", rc);
2586 /* initialize md content */
2589 md.threshold = 2; /*GET/REPLY*/
2591 md.options = LNET_MD_TRUNCATE;
2595 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
2597 CERROR("Can't bind MD: %d\n", rc);
2601 rc = LNetGet(LNET_NID_ANY, mdh, id,
2602 LNET_RESERVED_PORTAL,
2603 LNET_PROTO_PING_MATCHBITS, 0);
2606 /* Don't CERROR; this could be deliberate! */
2608 rc2 = LNetMDUnlink(mdh);
2611 /* NB must wait for the UNLINK event below... */
2613 timeout = a_long_time;
2617 /* MUST block for unlink to complete */
2619 blocked = cfs_block_allsigs();
2621 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
2624 cfs_restore_sigs(blocked);
2626 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
2627 (rc2 <= 0) ? -1 : event.type,
2628 (rc2 <= 0) ? -1 : event.status,
2629 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
2631 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
2633 if (rc2 <= 0 || event.status != 0) {
2634 /* timeout or error */
2635 if (!replied && rc == 0)
2636 rc = (rc2 < 0) ? rc2 :
2637 (rc2 == 0) ? -ETIMEDOUT :
2641 /* Ensure completion in finite time... */
2643 /* No assertion (racing with network) */
2645 timeout = a_long_time;
2646 } else if (rc2 == 0) {
2647 /* timed out waiting for unlink */
2648 CWARN("ping %s: late network completion\n",
2651 } else if (event.type == LNET_EVENT_REPLY) {
2656 } while (rc2 <= 0 || !event.unlinked);
2660 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
2667 LASSERT(nob >= 0 && nob <= infosz);
2669 rc = -EPROTO; /* if I can't parse... */
2672 /* can't check magic/version */
2673 CERROR("%s: ping info too short %d\n",
2674 libcfs_id2str(id), nob);
2678 if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
2679 lnet_swap_pinginfo(info);
2680 } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
2681 CERROR("%s: Unexpected magic %08x\n",
2682 libcfs_id2str(id), info->pi_magic);
2686 if ((info->pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
2687 CERROR("%s: ping w/o NI status: 0x%x\n",
2688 libcfs_id2str(id), info->pi_features);
2692 if (nob < offsetof(struct lnet_ping_info, pi_ni[0])) {
2693 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
2694 nob, (int)offsetof(struct lnet_ping_info, pi_ni[0]));
2698 if (info->pi_nnis < n_ids)
2699 n_ids = info->pi_nnis;
2701 if (nob < offsetof(struct lnet_ping_info, pi_ni[n_ids])) {
2702 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
2703 nob, (int)offsetof(struct lnet_ping_info, pi_ni[n_ids]));
2707 rc = -EFAULT; /* If I SEGV... */
2709 memset(&tmpid, 0, sizeof(tmpid));
2710 for (i = 0; i < n_ids; i++) {
2711 tmpid.pid = info->pi_pid;
2712 tmpid.nid = info->pi_ni[i].ns_nid;
2713 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
2719 rc2 = LNetEQFree(eqh);
2721 CERROR("rc2 %d\n", rc2);
2725 LIBCFS_FREE(info, infosz);