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 struct lnet 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");
65 unsigned int lnet_numa_range = 0;
66 module_param(lnet_numa_range, uint, 0444);
67 MODULE_PARM_DESC(lnet_numa_range,
68 "NUMA range to consider during Multi-Rail selection");
71 * This sequence number keeps track of how many times DLC was used to
72 * update the local NIs. It is incremented when a NI is added or
73 * removed and checked when sending a message to determine if there is
74 * a need to re-run the selection algorithm. See lnet_select_pathway()
75 * for more details on its usage.
77 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
79 static int lnet_ping(struct lnet_process_id id, signed long timeout,
80 struct lnet_process_id __user *ids, int n_ids);
89 lnet_get_networks(void)
94 if (*networks != 0 && *ip2nets != 0) {
95 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
96 "'ip2nets' but not both at once\n");
101 rc = lnet_parse_ip2nets(&nets, ip2nets);
102 return (rc == 0) ? nets : NULL;
112 lnet_init_locks(void)
114 spin_lock_init(&the_lnet.ln_eq_wait_lock);
115 init_waitqueue_head(&the_lnet.ln_eq_waitq);
116 init_waitqueue_head(&the_lnet.ln_rc_waitq);
117 mutex_init(&the_lnet.ln_lnd_mutex);
118 mutex_init(&the_lnet.ln_api_mutex);
122 lnet_fini_locks(void)
126 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
127 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
131 lnet_descriptor_setup(void)
133 /* create specific kmem_cache for MEs and small MDs (i.e., originally
134 * allocated in <size-xxx> kmem_cache).
136 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
138 if (!lnet_mes_cachep)
141 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
142 LNET_SMALL_MD_SIZE, 0, 0,
144 if (!lnet_small_mds_cachep)
151 lnet_descriptor_cleanup(void)
154 if (lnet_small_mds_cachep) {
155 kmem_cache_destroy(lnet_small_mds_cachep);
156 lnet_small_mds_cachep = NULL;
159 if (lnet_mes_cachep) {
160 kmem_cache_destroy(lnet_mes_cachep);
161 lnet_mes_cachep = NULL;
166 lnet_create_remote_nets_table(void)
169 struct list_head *hash;
171 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
172 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
173 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
175 CERROR("Failed to create remote nets hash table\n");
179 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
180 INIT_LIST_HEAD(&hash[i]);
181 the_lnet.ln_remote_nets_hash = hash;
186 lnet_destroy_remote_nets_table(void)
190 if (the_lnet.ln_remote_nets_hash == NULL)
193 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
194 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
196 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
197 LNET_REMOTE_NETS_HASH_SIZE *
198 sizeof(the_lnet.ln_remote_nets_hash[0]));
199 the_lnet.ln_remote_nets_hash = NULL;
203 lnet_destroy_locks(void)
205 if (the_lnet.ln_res_lock != NULL) {
206 cfs_percpt_lock_free(the_lnet.ln_res_lock);
207 the_lnet.ln_res_lock = NULL;
210 if (the_lnet.ln_net_lock != NULL) {
211 cfs_percpt_lock_free(the_lnet.ln_net_lock);
212 the_lnet.ln_net_lock = NULL;
219 lnet_create_locks(void)
223 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
224 if (the_lnet.ln_res_lock == NULL)
227 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
228 if (the_lnet.ln_net_lock == NULL)
234 lnet_destroy_locks();
238 static void lnet_assert_wire_constants(void)
240 /* Wire protocol assertions generated by 'wirecheck'
241 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
242 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
243 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
246 CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
247 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
248 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
249 CLASSERT(LNET_MSG_ACK == 0);
250 CLASSERT(LNET_MSG_PUT == 1);
251 CLASSERT(LNET_MSG_GET == 2);
252 CLASSERT(LNET_MSG_REPLY == 3);
253 CLASSERT(LNET_MSG_HELLO == 4);
255 /* Checks for struct lnet_handle_wire */
256 CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
257 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
258 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
259 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
260 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
262 /* Checks for struct struct lnet_magicversion */
263 CLASSERT((int)sizeof(struct lnet_magicversion) == 8);
264 CLASSERT((int)offsetof(struct lnet_magicversion, magic) == 0);
265 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->magic) == 4);
266 CLASSERT((int)offsetof(struct lnet_magicversion, version_major) == 4);
267 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_major) == 2);
268 CLASSERT((int)offsetof(struct lnet_magicversion, version_minor) == 6);
269 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_minor) == 2);
271 /* Checks for struct struct lnet_hdr */
272 CLASSERT((int)sizeof(struct lnet_hdr) == 72);
273 CLASSERT((int)offsetof(struct lnet_hdr, dest_nid) == 0);
274 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_nid) == 8);
275 CLASSERT((int)offsetof(struct lnet_hdr, src_nid) == 8);
276 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_nid) == 8);
277 CLASSERT((int)offsetof(struct lnet_hdr, dest_pid) == 16);
278 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_pid) == 4);
279 CLASSERT((int)offsetof(struct lnet_hdr, src_pid) == 20);
280 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_pid) == 4);
281 CLASSERT((int)offsetof(struct lnet_hdr, type) == 24);
282 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->type) == 4);
283 CLASSERT((int)offsetof(struct lnet_hdr, payload_length) == 28);
284 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->payload_length) == 4);
285 CLASSERT((int)offsetof(struct lnet_hdr, msg) == 32);
286 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg) == 40);
289 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) == 32);
290 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) == 16);
291 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.match_bits) == 48);
292 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) == 8);
293 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.mlength) == 56);
294 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) == 4);
297 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) == 32);
298 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) == 16);
299 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.match_bits) == 48);
300 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) == 8);
301 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.hdr_data) == 56);
302 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) == 8);
303 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ptl_index) == 64);
304 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) == 4);
305 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.offset) == 68);
306 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) == 4);
309 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.return_wmd) == 32);
310 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) == 16);
311 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.match_bits) == 48);
312 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) == 8);
313 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.ptl_index) == 56);
314 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) == 4);
315 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.src_offset) == 60);
316 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) == 4);
317 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.sink_length) == 64);
318 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) == 4);
321 CLASSERT((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) == 32);
322 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) == 16);
325 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.incarnation) == 32);
326 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) == 8);
327 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.type) == 40);
328 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) == 4);
331 static struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
333 struct lnet_lnd *lnd;
334 struct list_head *tmp;
336 /* holding lnd mutex */
337 list_for_each(tmp, &the_lnet.ln_lnds) {
338 lnd = list_entry(tmp, struct lnet_lnd, lnd_list);
340 if (lnd->lnd_type == type)
347 lnet_register_lnd(struct lnet_lnd *lnd)
349 mutex_lock(&the_lnet.ln_lnd_mutex);
351 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
352 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
354 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
355 lnd->lnd_refcount = 0;
357 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
359 mutex_unlock(&the_lnet.ln_lnd_mutex);
361 EXPORT_SYMBOL(lnet_register_lnd);
364 lnet_unregister_lnd(struct lnet_lnd *lnd)
366 mutex_lock(&the_lnet.ln_lnd_mutex);
368 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
369 LASSERT(lnd->lnd_refcount == 0);
371 list_del(&lnd->lnd_list);
372 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
374 mutex_unlock(&the_lnet.ln_lnd_mutex);
376 EXPORT_SYMBOL(lnet_unregister_lnd);
379 lnet_counters_get(struct lnet_counters *counters)
381 struct lnet_counters *ctr;
384 memset(counters, 0, sizeof(*counters));
386 lnet_net_lock(LNET_LOCK_EX);
388 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
389 counters->msgs_max += ctr->msgs_max;
390 counters->msgs_alloc += ctr->msgs_alloc;
391 counters->errors += ctr->errors;
392 counters->send_count += ctr->send_count;
393 counters->recv_count += ctr->recv_count;
394 counters->route_count += ctr->route_count;
395 counters->drop_count += ctr->drop_count;
396 counters->send_length += ctr->send_length;
397 counters->recv_length += ctr->recv_length;
398 counters->route_length += ctr->route_length;
399 counters->drop_length += ctr->drop_length;
402 lnet_net_unlock(LNET_LOCK_EX);
404 EXPORT_SYMBOL(lnet_counters_get);
407 lnet_counters_reset(void)
409 struct lnet_counters *counters;
412 lnet_net_lock(LNET_LOCK_EX);
414 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
415 memset(counters, 0, sizeof(struct lnet_counters));
417 lnet_net_unlock(LNET_LOCK_EX);
421 lnet_res_type2str(int type)
426 case LNET_COOKIE_TYPE_MD:
428 case LNET_COOKIE_TYPE_ME:
430 case LNET_COOKIE_TYPE_EQ:
436 lnet_res_container_cleanup(struct lnet_res_container *rec)
440 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
443 while (!list_empty(&rec->rec_active)) {
444 struct list_head *e = rec->rec_active.next;
447 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
448 lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
450 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
451 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
453 } else { /* NB: Active MEs should be attached on portals */
460 /* Found alive MD/ME/EQ, user really should unlink/free
461 * all of them before finalize LNet, but if someone didn't,
462 * we have to recycle garbage for him */
463 CERROR("%d active elements on exit of %s container\n",
464 count, lnet_res_type2str(rec->rec_type));
467 if (rec->rec_lh_hash != NULL) {
468 LIBCFS_FREE(rec->rec_lh_hash,
469 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
470 rec->rec_lh_hash = NULL;
473 rec->rec_type = 0; /* mark it as finalized */
477 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
482 LASSERT(rec->rec_type == 0);
484 rec->rec_type = type;
485 INIT_LIST_HEAD(&rec->rec_active);
487 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
489 /* Arbitrary choice of hash table size */
490 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
491 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
492 if (rec->rec_lh_hash == NULL) {
497 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
498 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
503 CERROR("Failed to setup %s resource container\n",
504 lnet_res_type2str(type));
505 lnet_res_container_cleanup(rec);
510 lnet_res_containers_destroy(struct lnet_res_container **recs)
512 struct lnet_res_container *rec;
515 cfs_percpt_for_each(rec, i, recs)
516 lnet_res_container_cleanup(rec);
518 cfs_percpt_free(recs);
521 static struct lnet_res_container **
522 lnet_res_containers_create(int type)
524 struct lnet_res_container **recs;
525 struct lnet_res_container *rec;
529 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
531 CERROR("Failed to allocate %s resource containers\n",
532 lnet_res_type2str(type));
536 cfs_percpt_for_each(rec, i, recs) {
537 rc = lnet_res_container_setup(rec, i, type);
539 lnet_res_containers_destroy(recs);
547 struct lnet_libhandle *
548 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
550 /* ALWAYS called with lnet_res_lock held */
551 struct list_head *head;
552 struct lnet_libhandle *lh;
555 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
558 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
559 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
561 list_for_each_entry(lh, head, lh_hash_chain) {
562 if (lh->lh_cookie == cookie)
570 lnet_res_lh_initialize(struct lnet_res_container *rec,
571 struct lnet_libhandle *lh)
573 /* ALWAYS called with lnet_res_lock held */
574 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
577 lh->lh_cookie = rec->rec_lh_cookie;
578 rec->rec_lh_cookie += 1 << ibits;
580 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
582 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
585 static int lnet_unprepare(void);
588 lnet_prepare(lnet_pid_t requested_pid)
590 /* Prepare to bring up the network */
591 struct lnet_res_container **recs;
594 if (requested_pid == LNET_PID_ANY) {
595 /* Don't instantiate LNET just for me */
599 LASSERT(the_lnet.ln_refcount == 0);
601 the_lnet.ln_routing = 0;
603 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
604 the_lnet.ln_pid = requested_pid;
606 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
607 INIT_LIST_HEAD(&the_lnet.ln_peers);
608 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
609 INIT_LIST_HEAD(&the_lnet.ln_nets);
610 INIT_LIST_HEAD(&the_lnet.ln_routers);
611 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
612 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
614 rc = lnet_descriptor_setup();
618 rc = lnet_create_remote_nets_table();
623 * NB the interface cookie in wire handles guards against delayed
624 * replies and ACKs appearing valid after reboot.
626 the_lnet.ln_interface_cookie = ktime_get_real_ns();
628 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
629 sizeof(struct lnet_counters));
630 if (the_lnet.ln_counters == NULL) {
631 CERROR("Failed to allocate counters for LNet\n");
636 rc = lnet_peer_tables_create();
640 rc = lnet_msg_containers_create();
644 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
645 LNET_COOKIE_TYPE_EQ);
649 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
655 the_lnet.ln_me_containers = recs;
657 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
663 the_lnet.ln_md_containers = recs;
665 rc = lnet_portals_create();
667 CERROR("Failed to create portals for LNet: %d\n", rc);
679 lnet_unprepare (void)
681 /* NB no LNET_LOCK since this is the last reference. All LND instances
682 * have shut down already, so it is safe to unlink and free all
683 * descriptors, even those that appear committed to a network op (eg MD
684 * with non-zero pending count) */
686 lnet_fail_nid(LNET_NID_ANY, 0);
688 LASSERT(the_lnet.ln_refcount == 0);
689 LASSERT(list_empty(&the_lnet.ln_test_peers));
690 LASSERT(list_empty(&the_lnet.ln_nets));
692 lnet_portals_destroy();
694 if (the_lnet.ln_md_containers != NULL) {
695 lnet_res_containers_destroy(the_lnet.ln_md_containers);
696 the_lnet.ln_md_containers = NULL;
699 if (the_lnet.ln_me_containers != NULL) {
700 lnet_res_containers_destroy(the_lnet.ln_me_containers);
701 the_lnet.ln_me_containers = NULL;
704 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
706 lnet_msg_containers_destroy();
708 lnet_rtrpools_free(0);
710 if (the_lnet.ln_counters != NULL) {
711 cfs_percpt_free(the_lnet.ln_counters);
712 the_lnet.ln_counters = NULL;
714 lnet_destroy_remote_nets_table();
715 lnet_descriptor_cleanup();
721 lnet_net2ni_locked(__u32 net_id, int cpt)
724 struct lnet_net *net;
726 LASSERT(cpt != LNET_LOCK_EX);
728 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
729 if (net->net_id == net_id) {
730 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
740 lnet_net2ni_addref(__u32 net)
745 ni = lnet_net2ni_locked(net, 0);
747 lnet_ni_addref_locked(ni, 0);
752 EXPORT_SYMBOL(lnet_net2ni_addref);
755 lnet_get_net_locked(__u32 net_id)
757 struct lnet_net *net;
759 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
760 if (net->net_id == net_id)
768 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
773 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
778 val = hash_long(key, LNET_CPT_BITS);
779 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
783 return (unsigned int)(key + val + (val >> 1)) % number;
787 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
789 struct lnet_net *net;
791 /* must called with hold of lnet_net_lock */
792 if (LNET_CPT_NUMBER == 1)
793 return 0; /* the only one */
796 * If NI is provided then use the CPT identified in the NI cpt
797 * list if one exists. If one doesn't exist, then that NI is
798 * associated with all CPTs and it follows that the net it belongs
799 * to is implicitly associated with all CPTs, so just hash the nid
803 if (ni->ni_cpts != NULL)
804 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
807 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
810 /* no NI provided so look at the net */
811 net = lnet_get_net_locked(LNET_NIDNET(nid));
813 if (net != NULL && net->net_cpts != NULL) {
814 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
817 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
821 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
826 if (LNET_CPT_NUMBER == 1)
827 return 0; /* the only one */
829 cpt = lnet_net_lock_current();
831 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
833 lnet_net_unlock(cpt);
837 EXPORT_SYMBOL(lnet_cpt_of_nid);
840 lnet_islocalnet(__u32 net_id)
842 struct lnet_net *net;
846 cpt = lnet_net_lock_current();
848 net = lnet_get_net_locked(net_id);
852 lnet_net_unlock(cpt);
858 lnet_is_ni_healthy_locked(struct lnet_ni *ni)
860 if (ni->ni_state == LNET_NI_STATE_ACTIVE ||
861 ni->ni_state == LNET_NI_STATE_DEGRADED)
868 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
870 struct lnet_net *net;
873 LASSERT(cpt != LNET_LOCK_EX);
875 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
876 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
877 if (ni->ni_nid == nid)
886 lnet_nid2ni_addref(lnet_nid_t nid)
891 ni = lnet_nid2ni_locked(nid, 0);
893 lnet_ni_addref_locked(ni, 0);
898 EXPORT_SYMBOL(lnet_nid2ni_addref);
901 lnet_islocalnid(lnet_nid_t nid)
906 cpt = lnet_net_lock_current();
907 ni = lnet_nid2ni_locked(nid, cpt);
908 lnet_net_unlock(cpt);
914 lnet_count_acceptor_nets(void)
916 /* Return the # of NIs that need the acceptor. */
918 struct lnet_net *net;
921 cpt = lnet_net_lock_current();
922 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
923 /* all socklnd type networks should have the acceptor
925 if (net->net_lnd->lnd_accept != NULL)
929 lnet_net_unlock(cpt);
934 static struct lnet_ping_info *
935 lnet_ping_info_create(int num_ni)
937 struct lnet_ping_info *ping_info;
940 infosz = offsetof(struct lnet_ping_info, pi_ni[num_ni]);
941 LIBCFS_ALLOC(ping_info, infosz);
942 if (ping_info == NULL) {
943 CERROR("Can't allocate ping info[%d]\n", num_ni);
947 ping_info->pi_nnis = num_ni;
948 ping_info->pi_pid = the_lnet.ln_pid;
949 ping_info->pi_magic = LNET_PROTO_PING_MAGIC;
950 ping_info->pi_features = LNET_PING_FEAT_NI_STATUS;
956 lnet_get_net_ni_count_locked(struct lnet_net *net)
961 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
968 lnet_get_net_ni_count_pre(struct lnet_net *net)
973 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
980 lnet_get_ni_count(void)
983 struct lnet_net *net;
988 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
989 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
999 lnet_ping_info_free(struct lnet_ping_info *pinfo)
1002 offsetof(struct lnet_ping_info,
1003 pi_ni[pinfo->pi_nnis]));
1007 lnet_ping_info_destroy(void)
1009 struct lnet_net *net;
1012 lnet_net_lock(LNET_LOCK_EX);
1014 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1015 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1017 ni->ni_status = NULL;
1022 lnet_ping_info_free(the_lnet.ln_ping_info);
1023 the_lnet.ln_ping_info = NULL;
1025 lnet_net_unlock(LNET_LOCK_EX);
1029 lnet_ping_event_handler(struct lnet_event *event)
1031 struct lnet_ping_info *pinfo = event->md.user_ptr;
1033 if (event->unlinked)
1034 pinfo->pi_features = LNET_PING_FEAT_INVAL;
1038 lnet_ping_info_setup(struct lnet_ping_info **ppinfo,
1039 struct lnet_handle_md *md_handle,
1040 int ni_count, bool set_eq)
1042 struct lnet_process_id id = {
1043 .nid = LNET_NID_ANY,
1046 struct lnet_handle_me me_handle;
1047 struct lnet_md md = { NULL };
1051 rc = LNetEQAlloc(0, lnet_ping_event_handler,
1052 &the_lnet.ln_ping_target_eq);
1054 CERROR("Can't allocate ping EQ: %d\n", rc);
1059 *ppinfo = lnet_ping_info_create(ni_count);
1060 if (*ppinfo == NULL) {
1065 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1066 LNET_PROTO_PING_MATCHBITS, 0,
1067 LNET_UNLINK, LNET_INS_AFTER,
1070 CERROR("Can't create ping ME: %d\n", rc);
1074 /* initialize md content */
1076 md.length = offsetof(struct lnet_ping_info,
1077 pi_ni[(*ppinfo)->pi_nnis]);
1078 md.threshold = LNET_MD_THRESH_INF;
1080 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1081 LNET_MD_MANAGE_REMOTE;
1083 md.eq_handle = the_lnet.ln_ping_target_eq;
1084 md.user_ptr = *ppinfo;
1086 rc = LNetMDAttach(me_handle, md, LNET_RETAIN, md_handle);
1088 CERROR("Can't attach ping MD: %d\n", rc);
1095 rc2 = LNetMEUnlink(me_handle);
1098 lnet_ping_info_free(*ppinfo);
1102 LNetEQFree(the_lnet.ln_ping_target_eq);
1107 lnet_ping_md_unlink(struct lnet_ping_info *pinfo, struct lnet_handle_md *md_handle)
1109 sigset_t blocked = cfs_block_allsigs();
1111 LNetMDUnlink(*md_handle);
1112 LNetInvalidateMDHandle(md_handle);
1114 /* NB md could be busy; this just starts the unlink */
1115 while (pinfo->pi_features != LNET_PING_FEAT_INVAL) {
1116 CDEBUG(D_NET, "Still waiting for ping MD to unlink\n");
1117 set_current_state(TASK_UNINTERRUPTIBLE);
1118 schedule_timeout(cfs_time_seconds(1));
1121 cfs_restore_sigs(blocked);
1125 lnet_ping_info_install_locked(struct lnet_ping_info *ping_info)
1129 struct lnet_net *net;
1130 struct lnet_ni_status *ns;
1133 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1134 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1135 LASSERT(i < ping_info->pi_nnis);
1137 ns = &ping_info->pi_ni[i];
1139 ns->ns_nid = ni->ni_nid;
1142 ns->ns_status = (ni->ni_status != NULL) ?
1143 ni->ni_status->ns_status :
1155 lnet_ping_target_update(struct lnet_ping_info *pinfo,
1156 struct lnet_handle_md md_handle)
1158 struct lnet_ping_info *old_pinfo = NULL;
1159 struct lnet_handle_md old_md;
1161 /* switch the NIs to point to the new ping info created */
1162 lnet_net_lock(LNET_LOCK_EX);
1164 if (!the_lnet.ln_routing)
1165 pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1166 lnet_ping_info_install_locked(pinfo);
1168 if (the_lnet.ln_ping_info != NULL) {
1169 old_pinfo = the_lnet.ln_ping_info;
1170 old_md = the_lnet.ln_ping_target_md;
1172 the_lnet.ln_ping_target_md = md_handle;
1173 the_lnet.ln_ping_info = pinfo;
1175 lnet_net_unlock(LNET_LOCK_EX);
1177 if (old_pinfo != NULL) {
1178 /* unlink the old ping info */
1179 lnet_ping_md_unlink(old_pinfo, &old_md);
1180 lnet_ping_info_free(old_pinfo);
1185 lnet_ping_target_fini(void)
1189 lnet_ping_md_unlink(the_lnet.ln_ping_info,
1190 &the_lnet.ln_ping_target_md);
1192 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1195 lnet_ping_info_destroy();
1199 lnet_ni_tq_credits(struct lnet_ni *ni)
1203 LASSERT(ni->ni_ncpts >= 1);
1205 if (ni->ni_ncpts == 1)
1206 return ni->ni_net->net_tunables.lct_max_tx_credits;
1208 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
1209 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
1210 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
1216 lnet_ni_unlink_locked(struct lnet_ni *ni)
1218 if (!list_empty(&ni->ni_cptlist)) {
1219 list_del_init(&ni->ni_cptlist);
1220 lnet_ni_decref_locked(ni, 0);
1223 /* move it to zombie list and nobody can find it anymore */
1224 LASSERT(!list_empty(&ni->ni_netlist));
1225 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
1226 lnet_ni_decref_locked(ni, 0);
1230 lnet_clear_zombies_nis_locked(struct lnet_net *net)
1235 struct list_head *zombie_list = &net->net_ni_zombie;
1238 * Now wait for the NIs I just nuked to show up on the zombie
1239 * list and shut them down in guaranteed thread context
1242 while (!list_empty(zombie_list)) {
1246 ni = list_entry(zombie_list->next,
1247 struct lnet_ni, ni_netlist);
1248 list_del_init(&ni->ni_netlist);
1249 /* the ni should be in deleting state. If it's not it's
1251 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
1252 cfs_percpt_for_each(ref, j, ni->ni_refs) {
1255 /* still busy, add it back to zombie list */
1256 list_add(&ni->ni_netlist, zombie_list);
1260 if (!list_empty(&ni->ni_netlist)) {
1261 lnet_net_unlock(LNET_LOCK_EX);
1263 if ((i & (-i)) == i) {
1265 "Waiting for zombie LNI %s\n",
1266 libcfs_nid2str(ni->ni_nid));
1268 set_current_state(TASK_UNINTERRUPTIBLE);
1269 schedule_timeout(cfs_time_seconds(1));
1270 lnet_net_lock(LNET_LOCK_EX);
1274 lnet_net_unlock(LNET_LOCK_EX);
1276 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
1278 LASSERT(!in_interrupt());
1279 (net->net_lnd->lnd_shutdown)(ni);
1282 CDEBUG(D_LNI, "Removed LNI %s\n",
1283 libcfs_nid2str(ni->ni_nid));
1287 lnet_net_lock(LNET_LOCK_EX);
1291 /* shutdown down the NI and release refcount */
1293 lnet_shutdown_lndni(struct lnet_ni *ni)
1296 struct lnet_net *net = ni->ni_net;
1298 lnet_net_lock(LNET_LOCK_EX);
1299 ni->ni_state = LNET_NI_STATE_DELETING;
1300 lnet_ni_unlink_locked(ni);
1301 lnet_incr_dlc_seq();
1302 lnet_net_unlock(LNET_LOCK_EX);
1304 /* clear messages for this NI on the lazy portal */
1305 for (i = 0; i < the_lnet.ln_nportals; i++)
1306 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
1308 lnet_net_lock(LNET_LOCK_EX);
1309 lnet_clear_zombies_nis_locked(net);
1310 lnet_net_unlock(LNET_LOCK_EX);
1314 lnet_shutdown_lndnet(struct lnet_net *net)
1318 lnet_net_lock(LNET_LOCK_EX);
1320 net->net_state = LNET_NET_STATE_DELETING;
1322 list_del_init(&net->net_list);
1324 while (!list_empty(&net->net_ni_list)) {
1325 ni = list_entry(net->net_ni_list.next,
1326 struct lnet_ni, ni_netlist);
1327 lnet_net_unlock(LNET_LOCK_EX);
1328 lnet_shutdown_lndni(ni);
1329 lnet_net_lock(LNET_LOCK_EX);
1332 lnet_net_unlock(LNET_LOCK_EX);
1334 /* Do peer table cleanup for this net */
1335 lnet_peer_tables_cleanup(net);
1337 lnet_net_lock(LNET_LOCK_EX);
1339 * decrement ref count on lnd only when the entire network goes
1342 net->net_lnd->lnd_refcount--;
1344 lnet_net_unlock(LNET_LOCK_EX);
1350 lnet_shutdown_lndnets(void)
1352 struct lnet_net *net;
1354 /* NB called holding the global mutex */
1356 /* All quiet on the API front */
1357 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
1358 LASSERT(the_lnet.ln_refcount == 0);
1360 lnet_net_lock(LNET_LOCK_EX);
1361 the_lnet.ln_state = LNET_STATE_STOPPING;
1363 while (!list_empty(&the_lnet.ln_nets)) {
1365 * move the nets to the zombie list to avoid them being
1366 * picked up for new work. LONET is also included in the
1367 * Nets that will be moved to the zombie list
1369 net = list_entry(the_lnet.ln_nets.next,
1370 struct lnet_net, net_list);
1371 list_move(&net->net_list, &the_lnet.ln_net_zombie);
1374 /* Drop the cached loopback Net. */
1375 if (the_lnet.ln_loni != NULL) {
1376 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
1377 the_lnet.ln_loni = NULL;
1379 lnet_net_unlock(LNET_LOCK_EX);
1381 /* iterate through the net zombie list and delete each net */
1382 while (!list_empty(&the_lnet.ln_net_zombie)) {
1383 net = list_entry(the_lnet.ln_net_zombie.next,
1384 struct lnet_net, net_list);
1385 lnet_shutdown_lndnet(net);
1388 lnet_net_lock(LNET_LOCK_EX);
1389 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
1390 lnet_net_unlock(LNET_LOCK_EX);
1394 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
1397 struct lnet_tx_queue *tq;
1399 struct lnet_net *net = ni->ni_net;
1401 mutex_lock(&the_lnet.ln_lnd_mutex);
1404 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
1405 ni->ni_lnd_tunables_set = true;
1408 rc = (net->net_lnd->lnd_startup)(ni);
1410 mutex_unlock(&the_lnet.ln_lnd_mutex);
1413 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1414 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
1415 lnet_net_lock(LNET_LOCK_EX);
1416 net->net_lnd->lnd_refcount--;
1417 lnet_net_unlock(LNET_LOCK_EX);
1421 ni->ni_state = LNET_NI_STATE_ACTIVE;
1423 /* We keep a reference on the loopback net through the loopback NI */
1424 if (net->net_lnd->lnd_type == LOLND) {
1426 LASSERT(the_lnet.ln_loni == NULL);
1427 the_lnet.ln_loni = ni;
1428 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
1429 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
1430 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
1431 ni->ni_net->net_tunables.lct_peer_timeout = 0;
1435 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
1436 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
1437 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1438 libcfs_lnd2str(net->net_lnd->lnd_type),
1439 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
1441 /* shutdown the NI since if we get here then it must've already
1444 lnet_shutdown_lndni(ni);
1448 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
1449 tq->tq_credits_min =
1450 tq->tq_credits_max =
1451 tq->tq_credits = lnet_ni_tq_credits(ni);
1454 atomic_set(&ni->ni_tx_credits,
1455 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
1457 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
1458 libcfs_nid2str(ni->ni_nid),
1459 ni->ni_net->net_tunables.lct_peer_tx_credits,
1460 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
1461 ni->ni_net->net_tunables.lct_peer_rtr_credits,
1462 ni->ni_net->net_tunables.lct_peer_timeout);
1471 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
1474 struct lnet_net *net_l = NULL;
1475 struct list_head local_ni_list;
1479 struct lnet_lnd *lnd;
1481 net->net_tunables.lct_peer_timeout;
1483 net->net_tunables.lct_max_tx_credits;
1484 int peerrtrcredits =
1485 net->net_tunables.lct_peer_rtr_credits;
1487 INIT_LIST_HEAD(&local_ni_list);
1490 * make sure that this net is unique. If it isn't then
1491 * we are adding interfaces to an already existing network, and
1492 * 'net' is just a convenient way to pass in the list.
1493 * if it is unique we need to find the LND and load it if
1496 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
1497 lnd_type = LNET_NETTYP(net->net_id);
1499 LASSERT(libcfs_isknown_lnd(lnd_type));
1501 mutex_lock(&the_lnet.ln_lnd_mutex);
1502 lnd = lnet_find_lnd_by_type(lnd_type);
1505 mutex_unlock(&the_lnet.ln_lnd_mutex);
1506 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
1507 mutex_lock(&the_lnet.ln_lnd_mutex);
1509 lnd = lnet_find_lnd_by_type(lnd_type);
1511 mutex_unlock(&the_lnet.ln_lnd_mutex);
1512 CERROR("Can't load LND %s, module %s, rc=%d\n",
1513 libcfs_lnd2str(lnd_type),
1514 libcfs_lnd2modname(lnd_type), rc);
1515 #ifndef HAVE_MODULE_LOADING_SUPPORT
1516 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
1517 "compiled with kernel module "
1518 "loading support.");
1525 lnet_net_lock(LNET_LOCK_EX);
1526 lnd->lnd_refcount++;
1527 lnet_net_unlock(LNET_LOCK_EX);
1531 mutex_unlock(&the_lnet.ln_lnd_mutex);
1537 * net_l: if the network being added is unique then net_l
1538 * will point to that network
1539 * if the network being added is not unique then
1540 * net_l points to the existing network.
1542 * When we enter the loop below, we'll pick NIs off he
1543 * network beign added and start them up, then add them to
1544 * a local ni list. Once we've successfully started all
1545 * the NIs then we join the local NI list (of started up
1546 * networks) with the net_l->net_ni_list, which should
1547 * point to the correct network to add the new ni list to
1549 * If any of the new NIs fail to start up, then we want to
1550 * iterate through the local ni list, which should include
1551 * any NIs which were successfully started up, and shut
1554 * After than we want to delete the network being added,
1555 * to avoid a memory leak.
1559 * When a network uses TCP bonding then all its interfaces
1560 * must be specified when the network is first defined: the
1561 * TCP bonding code doesn't allow for interfaces to be added
1564 if (net_l != net && net_l != NULL && use_tcp_bonding &&
1565 LNET_NETTYP(net_l->net_id) == SOCKLND) {
1570 while (!list_empty(&net->net_ni_added)) {
1571 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
1573 list_del_init(&ni->ni_netlist);
1575 /* make sure that the the NI we're about to start
1576 * up is actually unique. if it's not fail. */
1577 if (!lnet_ni_unique_net(&net_l->net_ni_list,
1578 ni->ni_interfaces[0])) {
1583 /* adjust the pointer the parent network, just in case it
1584 * the net is a duplicate */
1587 rc = lnet_startup_lndni(ni, tun);
1589 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
1590 ni->ni_net->net_lnd->lnd_query != NULL);
1596 list_add_tail(&ni->ni_netlist, &local_ni_list);
1601 lnet_net_lock(LNET_LOCK_EX);
1602 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
1603 lnet_incr_dlc_seq();
1604 lnet_net_unlock(LNET_LOCK_EX);
1606 /* if the network is not unique then we don't want to keep
1607 * it around after we're done. Free it. Otherwise add that
1608 * net to the global the_lnet.ln_nets */
1609 if (net_l != net && net_l != NULL) {
1611 * TODO - note. currently the tunables can not be updated
1616 net->net_state = LNET_NET_STATE_ACTIVE;
1618 * restore tunables after it has been overwitten by the
1621 if (peer_timeout != -1)
1622 net->net_tunables.lct_peer_timeout = peer_timeout;
1623 if (maxtxcredits != -1)
1624 net->net_tunables.lct_max_tx_credits = maxtxcredits;
1625 if (peerrtrcredits != -1)
1626 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
1628 lnet_net_lock(LNET_LOCK_EX);
1629 list_add_tail(&net->net_list, &the_lnet.ln_nets);
1630 lnet_net_unlock(LNET_LOCK_EX);
1637 * shutdown the new NIs that are being started up
1638 * free the NET being started
1640 while (!list_empty(&local_ni_list)) {
1641 ni = list_entry(local_ni_list.next, struct lnet_ni,
1644 lnet_shutdown_lndni(ni);
1654 lnet_startup_lndnets(struct list_head *netlist)
1656 struct lnet_net *net;
1661 * Change to running state before bringing up the LNDs. This
1662 * allows lnet_shutdown_lndnets() to assert that we've passed
1665 lnet_net_lock(LNET_LOCK_EX);
1666 the_lnet.ln_state = LNET_STATE_RUNNING;
1667 lnet_net_unlock(LNET_LOCK_EX);
1669 while (!list_empty(netlist)) {
1670 net = list_entry(netlist->next, struct lnet_net, net_list);
1671 list_del_init(&net->net_list);
1673 rc = lnet_startup_lndnet(net, NULL);
1683 lnet_shutdown_lndnets();
1689 * Initialize LNet library.
1691 * Automatically called at module loading time. Caller has to call
1692 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
1693 * latter returned 0. It must be called exactly once.
1695 * \retval 0 on success
1696 * \retval -ve on failures.
1698 int lnet_lib_init(void)
1702 lnet_assert_wire_constants();
1704 memset(&the_lnet, 0, sizeof(the_lnet));
1706 /* refer to global cfs_cpt_table for now */
1707 the_lnet.ln_cpt_table = cfs_cpt_table;
1708 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
1710 LASSERT(the_lnet.ln_cpt_number > 0);
1711 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
1712 /* we are under risk of consuming all lh_cookie */
1713 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
1714 "please change setting of CPT-table and retry\n",
1715 the_lnet.ln_cpt_number, LNET_CPT_MAX);
1719 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
1720 the_lnet.ln_cpt_bits++;
1722 rc = lnet_create_locks();
1724 CERROR("Can't create LNet global locks: %d\n", rc);
1728 the_lnet.ln_refcount = 0;
1729 LNetInvalidateEQHandle(&the_lnet.ln_rc_eqh);
1730 INIT_LIST_HEAD(&the_lnet.ln_lnds);
1731 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
1732 INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
1733 INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
1735 /* The hash table size is the number of bits it takes to express the set
1736 * ln_num_routes, minus 1 (better to under estimate than over so we
1737 * don't waste memory). */
1738 if (rnet_htable_size <= 0)
1739 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
1740 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
1741 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
1742 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
1743 order_base_2(rnet_htable_size) - 1);
1745 /* All LNDs apart from the LOLND are in separate modules. They
1746 * register themselves when their module loads, and unregister
1747 * themselves when their module is unloaded. */
1748 lnet_register_lnd(&the_lolnd);
1753 * Finalize LNet library.
1755 * \pre lnet_lib_init() called with success.
1756 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1758 void lnet_lib_exit(void)
1760 LASSERT(the_lnet.ln_refcount == 0);
1762 while (!list_empty(&the_lnet.ln_lnds))
1763 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
1764 struct lnet_lnd, lnd_list));
1765 lnet_destroy_locks();
1769 * Set LNet PID and start LNet interfaces, routing, and forwarding.
1771 * Users must call this function at least once before any other functions.
1772 * For each successful call there must be a corresponding call to
1773 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1776 * The PID used by LNet may be different from the one requested.
1779 * \param requested_pid PID requested by the caller.
1781 * \return >= 0 on success, and < 0 error code on failures.
1784 LNetNIInit(lnet_pid_t requested_pid)
1786 int im_a_router = 0;
1789 struct lnet_ping_info *pinfo;
1790 struct lnet_handle_md md_handle;
1791 struct list_head net_head;
1792 struct lnet_net *net;
1794 INIT_LIST_HEAD(&net_head);
1796 mutex_lock(&the_lnet.ln_api_mutex);
1798 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
1800 if (the_lnet.ln_refcount > 0) {
1801 rc = the_lnet.ln_refcount++;
1802 mutex_unlock(&the_lnet.ln_api_mutex);
1806 rc = lnet_prepare(requested_pid);
1808 mutex_unlock(&the_lnet.ln_api_mutex);
1812 /* create a network for Loopback network */
1813 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
1816 goto err_empty_list;
1819 /* Add in the loopback NI */
1820 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
1822 goto err_empty_list;
1825 /* If LNet is being initialized via DLC it is possible
1826 * that the user requests not to load module parameters (ones which
1827 * are supported by DLC) on initialization. Therefore, make sure not
1828 * to load networks, routes and forwarding from module parameters
1829 * in this case. On cleanup in case of failure only clean up
1830 * routes if it has been loaded */
1831 if (!the_lnet.ln_nis_from_mod_params) {
1832 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
1835 goto err_empty_list;
1838 ni_count = lnet_startup_lndnets(&net_head);
1841 goto err_empty_list;
1844 if (!the_lnet.ln_nis_from_mod_params) {
1845 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
1847 goto err_shutdown_lndnis;
1849 rc = lnet_check_routes();
1851 goto err_destroy_routes;
1853 rc = lnet_rtrpools_alloc(im_a_router);
1855 goto err_destroy_routes;
1858 rc = lnet_acceptor_start();
1860 goto err_destroy_routes;
1862 the_lnet.ln_refcount = 1;
1863 /* Now I may use my own API functions... */
1865 rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count, true);
1867 goto err_acceptor_stop;
1869 lnet_ping_target_update(pinfo, md_handle);
1871 rc = lnet_router_checker_start();
1878 mutex_unlock(&the_lnet.ln_api_mutex);
1883 lnet_ping_target_fini();
1885 the_lnet.ln_refcount = 0;
1886 lnet_acceptor_stop();
1888 if (!the_lnet.ln_nis_from_mod_params)
1889 lnet_destroy_routes();
1890 err_shutdown_lndnis:
1891 lnet_shutdown_lndnets();
1895 mutex_unlock(&the_lnet.ln_api_mutex);
1896 while (!list_empty(&net_head)) {
1897 struct lnet_net *net;
1899 net = list_entry(net_head.next, struct lnet_net, net_list);
1900 list_del_init(&net->net_list);
1905 EXPORT_SYMBOL(LNetNIInit);
1908 * Stop LNet interfaces, routing, and forwarding.
1910 * Users must call this function once for each successful call to LNetNIInit().
1911 * Once the LNetNIFini() operation has been started, the results of pending
1912 * API operations are undefined.
1914 * \return always 0 for current implementation.
1919 mutex_lock(&the_lnet.ln_api_mutex);
1921 LASSERT(the_lnet.ln_refcount > 0);
1923 if (the_lnet.ln_refcount != 1) {
1924 the_lnet.ln_refcount--;
1926 LASSERT(!the_lnet.ln_niinit_self);
1931 lnet_router_checker_stop();
1932 lnet_ping_target_fini();
1934 /* Teardown fns that use my own API functions BEFORE here */
1935 the_lnet.ln_refcount = 0;
1937 lnet_acceptor_stop();
1938 lnet_destroy_routes();
1939 lnet_shutdown_lndnets();
1943 mutex_unlock(&the_lnet.ln_api_mutex);
1946 EXPORT_SYMBOL(LNetNIFini);
1949 * Grabs the ni data from the ni structure and fills the out
1952 * \param[in] ni network interface structure
1953 * \param[out] cfg_ni NI config information
1954 * \param[out] tun network and LND tunables
1957 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
1958 struct lnet_ioctl_config_lnd_tunables *tun,
1959 struct lnet_ioctl_element_stats *stats,
1962 size_t min_size = 0;
1965 if (!ni || !cfg_ni || !tun)
1968 if (ni->ni_interfaces[0] != NULL) {
1969 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
1970 if (ni->ni_interfaces[i] != NULL) {
1971 strncpy(cfg_ni->lic_ni_intf[i],
1972 ni->ni_interfaces[i],
1973 sizeof(cfg_ni->lic_ni_intf[i]));
1978 cfg_ni->lic_nid = ni->ni_nid;
1979 cfg_ni->lic_status = ni->ni_status->ns_status;
1980 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
1981 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
1983 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
1986 stats->iel_send_count = atomic_read(&ni->ni_stats.send_count);
1987 stats->iel_recv_count = atomic_read(&ni->ni_stats.recv_count);
1991 * tun->lt_tun will always be present, but in order to be
1992 * backwards compatible, we need to deal with the cases when
1993 * tun->lt_tun is smaller than what the kernel has, because it
1994 * comes from an older version of a userspace program, then we'll
1995 * need to copy as much information as we have available space.
1997 min_size = tun_size - sizeof(tun->lt_cmn);
1998 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2000 /* copy over the cpts */
2001 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2002 ni->ni_cpts == NULL) {
2003 for (i = 0; i < ni->ni_ncpts; i++)
2004 cfg_ni->lic_cpts[i] = i;
2007 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2008 i < LNET_MAX_SHOW_NUM_CPT;
2010 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2012 cfg_ni->lic_ncpts = ni->ni_ncpts;
2016 * NOTE: This is a legacy function left in the code to be backwards
2017 * compatible with older userspace programs. It should eventually be
2020 * Grabs the ni data from the ni structure and fills the out
2023 * \param[in] ni network interface structure
2024 * \param[out] config config information
2027 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2028 struct lnet_ioctl_config_data *config)
2030 struct lnet_ioctl_net_config *net_config;
2031 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2032 size_t min_size, tunable_size = 0;
2038 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2042 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2043 ARRAY_SIZE(net_config->ni_interfaces));
2045 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2046 if (!ni->ni_interfaces[i])
2049 strncpy(net_config->ni_interfaces[i],
2050 ni->ni_interfaces[i],
2051 sizeof(net_config->ni_interfaces[i]));
2054 config->cfg_nid = ni->ni_nid;
2055 config->cfg_config_u.cfg_net.net_peer_timeout =
2056 ni->ni_net->net_tunables.lct_peer_timeout;
2057 config->cfg_config_u.cfg_net.net_max_tx_credits =
2058 ni->ni_net->net_tunables.lct_max_tx_credits;
2059 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2060 ni->ni_net->net_tunables.lct_peer_tx_credits;
2061 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2062 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2064 net_config->ni_status = ni->ni_status->ns_status;
2067 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2069 for (i = 0; i < num_cpts; i++)
2070 net_config->ni_cpts[i] = ni->ni_cpts[i];
2072 config->cfg_ncpts = num_cpts;
2076 * See if user land tools sent in a newer and larger version
2077 * of struct lnet_tunables than what the kernel uses.
2079 min_size = sizeof(*config) + sizeof(*net_config);
2081 if (config->cfg_hdr.ioc_len > min_size)
2082 tunable_size = config->cfg_hdr.ioc_len - min_size;
2084 /* Don't copy too much data to user space */
2085 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2086 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2088 if (lnd_cfg && min_size) {
2089 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2090 config->cfg_config_u.cfg_net.net_interface_count = 1;
2092 /* Tell user land that kernel side has less data */
2093 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2094 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2095 config->cfg_hdr.ioc_len -= min_size;
2101 lnet_get_ni_idx_locked(int idx)
2104 struct lnet_net *net;
2106 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2107 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2117 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2120 struct lnet_net *net = mynet;
2124 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2126 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2132 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2133 /* if you reached the end of the ni list and the net is
2134 * specified, then there are no more nis in that net */
2138 /* we reached the end of this net ni list. move to the
2140 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2141 /* no more nets and no more NIs. */
2144 /* get the next net */
2145 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2147 /* get the ni on it */
2148 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2154 /* there are more nis left */
2155 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2161 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2166 int idx = config->cfg_count;
2168 cpt = lnet_net_lock_current();
2170 ni = lnet_get_ni_idx_locked(idx);
2175 lnet_fill_ni_info_legacy(ni, config);
2179 lnet_net_unlock(cpt);
2184 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
2185 struct lnet_ioctl_config_lnd_tunables *tun,
2186 struct lnet_ioctl_element_stats *stats,
2193 if (!cfg_ni || !tun || !stats)
2196 cpt = lnet_net_lock_current();
2198 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
2203 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
2207 lnet_net_unlock(cpt);
2211 static int lnet_add_net_common(struct lnet_net *net,
2212 struct lnet_ioctl_config_lnd_tunables *tun)
2215 struct lnet_ping_info *pinfo;
2216 struct lnet_handle_md md_handle;
2218 struct lnet_remotenet *rnet;
2220 int num_acceptor_nets;
2222 lnet_net_lock(LNET_LOCK_EX);
2223 rnet = lnet_find_rnet_locked(net->net_id);
2224 lnet_net_unlock(LNET_LOCK_EX);
2226 * make sure that the net added doesn't invalidate the current
2227 * configuration LNet is keeping
2230 CERROR("Adding net %s will invalidate routing configuration\n",
2231 libcfs_net2str(net->net_id));
2237 * make sure you calculate the correct number of slots in the ping
2238 * info. Since the ping info is a flattened list of all the NIs,
2239 * we should allocate enough slots to accomodate the number of NIs
2240 * which will be added.
2242 * since ni hasn't been configured yet, use
2243 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
2245 net_ni_count = lnet_get_net_ni_count_pre(net);
2247 rc = lnet_ping_info_setup(&pinfo, &md_handle,
2248 net_ni_count + lnet_get_ni_count(),
2256 memcpy(&net->net_tunables,
2257 &tun->lt_cmn, sizeof(net->net_tunables));
2259 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
2262 * before starting this network get a count of the current TCP
2263 * networks which require the acceptor thread running. If that
2264 * count is == 0 before we start up this network, then we'd want to
2265 * start up the acceptor thread after starting up this network
2267 num_acceptor_nets = lnet_count_acceptor_nets();
2269 net_id = net->net_id;
2271 rc = lnet_startup_lndnet(net,
2272 (tun) ? &tun->lt_tun : NULL);
2276 lnet_net_lock(LNET_LOCK_EX);
2277 net = lnet_get_net_locked(net_id);
2278 lnet_net_unlock(LNET_LOCK_EX);
2283 * Start the acceptor thread if this is the first network
2284 * being added that requires the thread.
2286 if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
2287 rc = lnet_acceptor_start();
2289 /* shutdown the net that we just started */
2290 CERROR("Failed to start up acceptor thread\n");
2291 lnet_shutdown_lndnet(net);
2296 lnet_net_lock(LNET_LOCK_EX);
2297 lnet_peer_net_added(net);
2298 lnet_net_unlock(LNET_LOCK_EX);
2300 lnet_ping_target_update(pinfo, md_handle);
2305 lnet_ping_md_unlink(pinfo, &md_handle);
2306 lnet_ping_info_free(pinfo);
2310 static int lnet_handle_legacy_ip2nets(char *ip2nets,
2311 struct lnet_ioctl_config_lnd_tunables *tun)
2313 struct lnet_net *net;
2316 struct list_head net_head;
2318 INIT_LIST_HEAD(&net_head);
2320 rc = lnet_parse_ip2nets(&nets, ip2nets);
2324 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
2328 mutex_lock(&the_lnet.ln_api_mutex);
2329 while (!list_empty(&net_head)) {
2330 net = list_entry(net_head.next, struct lnet_net, net_list);
2331 list_del_init(&net->net_list);
2332 rc = lnet_add_net_common(net, tun);
2338 mutex_unlock(&the_lnet.ln_api_mutex);
2340 while (!list_empty(&net_head)) {
2341 net = list_entry(net_head.next, struct lnet_net, net_list);
2342 list_del_init(&net->net_list);
2348 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
2350 struct lnet_net *net;
2352 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
2356 /* get the tunables if they are available */
2357 if (conf->lic_cfg_hdr.ioc_len >=
2358 sizeof(*conf) + sizeof(*tun))
2359 tun = (struct lnet_ioctl_config_lnd_tunables *)
2362 /* handle legacy ip2nets from DLC */
2363 if (conf->lic_legacy_ip2nets[0] != '\0')
2364 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
2367 net_id = LNET_NIDNET(conf->lic_nid);
2369 net = lnet_net_alloc(net_id, NULL);
2373 for (i = 0; i < conf->lic_ncpts; i++) {
2374 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
2378 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
2379 conf->lic_ni_intf[0]);
2383 mutex_lock(&the_lnet.ln_api_mutex);
2385 rc = lnet_add_net_common(net, tun);
2387 mutex_unlock(&the_lnet.ln_api_mutex);
2392 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
2394 struct lnet_net *net;
2396 __u32 net_id = LNET_NIDNET(conf->lic_nid);
2397 struct lnet_ping_info *pinfo;
2398 struct lnet_handle_md md_handle;
2403 /* don't allow userspace to shutdown the LOLND */
2404 if (LNET_NETTYP(net_id) == LOLND)
2407 mutex_lock(&the_lnet.ln_api_mutex);
2411 net = lnet_get_net_locked(net_id);
2413 CERROR("net %s not found\n",
2414 libcfs_net2str(net_id));
2419 addr = LNET_NIDADDR(conf->lic_nid);
2421 /* remove the entire net */
2422 net_count = lnet_get_net_ni_count_locked(net);
2426 /* create and link a new ping info, before removing the old one */
2427 rc = lnet_ping_info_setup(&pinfo, &md_handle,
2428 lnet_get_ni_count() - net_count,
2433 lnet_shutdown_lndnet(net);
2435 if (lnet_count_acceptor_nets() == 0)
2436 lnet_acceptor_stop();
2438 lnet_ping_target_update(pinfo, md_handle);
2443 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
2445 CERROR("nid %s not found \n",
2446 libcfs_nid2str(conf->lic_nid));
2451 net_count = lnet_get_net_ni_count_locked(net);
2455 /* create and link a new ping info, before removing the old one */
2456 rc = lnet_ping_info_setup(&pinfo, &md_handle,
2457 lnet_get_ni_count() - 1, false);
2461 lnet_shutdown_lndni(ni);
2463 if (lnet_count_acceptor_nets() == 0)
2464 lnet_acceptor_stop();
2466 lnet_ping_target_update(pinfo, md_handle);
2468 /* check if the net is empty and remove it if it is */
2470 lnet_shutdown_lndnet(net);
2477 mutex_unlock(&the_lnet.ln_api_mutex);
2483 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
2484 * They are only expected to be called for unique networks.
2485 * That can be as a result of older DLC library
2486 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
2489 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
2491 struct lnet_net *net;
2492 struct list_head net_head;
2494 struct lnet_ioctl_config_lnd_tunables tun;
2495 char *nets = conf->cfg_config_u.cfg_net.net_intf;
2497 INIT_LIST_HEAD(&net_head);
2499 /* Create a net/ni structures for the network string */
2500 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
2502 return rc == 0 ? -EINVAL : rc;
2504 mutex_lock(&the_lnet.ln_api_mutex);
2507 rc = -EINVAL; /* only add one network per call */
2508 goto out_unlock_clean;
2511 net = list_entry(net_head.next, struct lnet_net, net_list);
2512 list_del_init(&net->net_list);
2514 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
2516 memset(&tun, 0, sizeof(tun));
2518 tun.lt_cmn.lct_peer_timeout =
2519 conf->cfg_config_u.cfg_net.net_peer_timeout;
2520 tun.lt_cmn.lct_peer_tx_credits =
2521 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
2522 tun.lt_cmn.lct_peer_rtr_credits =
2523 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
2524 tun.lt_cmn.lct_max_tx_credits =
2525 conf->cfg_config_u.cfg_net.net_max_tx_credits;
2527 rc = lnet_add_net_common(net, &tun);
2530 mutex_unlock(&the_lnet.ln_api_mutex);
2531 while (!list_empty(&net_head)) {
2532 /* net_head list is empty in success case */
2533 net = list_entry(net_head.next, struct lnet_net, net_list);
2534 list_del_init(&net->net_list);
2541 lnet_dyn_del_net(__u32 net_id)
2543 struct lnet_net *net;
2544 struct lnet_ping_info *pinfo;
2545 struct lnet_handle_md md_handle;
2549 /* don't allow userspace to shutdown the LOLND */
2550 if (LNET_NETTYP(net_id) == LOLND)
2553 mutex_lock(&the_lnet.ln_api_mutex);
2557 net = lnet_get_net_locked(net_id);
2564 net_ni_count = lnet_get_net_ni_count_locked(net);
2568 /* create and link a new ping info, before removing the old one */
2569 rc = lnet_ping_info_setup(&pinfo, &md_handle,
2570 lnet_get_ni_count() - net_ni_count, false);
2574 lnet_shutdown_lndnet(net);
2576 if (lnet_count_acceptor_nets() == 0)
2577 lnet_acceptor_stop();
2579 lnet_ping_target_update(pinfo, md_handle);
2582 mutex_unlock(&the_lnet.ln_api_mutex);
2587 void lnet_incr_dlc_seq(void)
2589 atomic_inc(&lnet_dlc_seq_no);
2592 __u32 lnet_get_dlc_seq_locked(void)
2594 return atomic_read(&lnet_dlc_seq_no);
2598 * LNet ioctl handler.
2602 LNetCtl(unsigned int cmd, void *arg)
2604 struct libcfs_ioctl_data *data = arg;
2605 struct lnet_ioctl_config_data *config;
2606 struct lnet_process_id id = {0};
2610 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
2611 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
2614 case IOC_LIBCFS_GET_NI:
2615 rc = LNetGetId(data->ioc_count, &id);
2616 data->ioc_nid = id.nid;
2619 case IOC_LIBCFS_FAIL_NID:
2620 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
2622 case IOC_LIBCFS_ADD_ROUTE:
2625 if (config->cfg_hdr.ioc_len < sizeof(*config))
2628 mutex_lock(&the_lnet.ln_api_mutex);
2629 rc = lnet_add_route(config->cfg_net,
2630 config->cfg_config_u.cfg_route.rtr_hop,
2632 config->cfg_config_u.cfg_route.
2635 rc = lnet_check_routes();
2637 lnet_del_route(config->cfg_net,
2640 mutex_unlock(&the_lnet.ln_api_mutex);
2643 case IOC_LIBCFS_DEL_ROUTE:
2646 if (config->cfg_hdr.ioc_len < sizeof(*config))
2649 mutex_lock(&the_lnet.ln_api_mutex);
2650 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
2651 mutex_unlock(&the_lnet.ln_api_mutex);
2654 case IOC_LIBCFS_GET_ROUTE:
2657 if (config->cfg_hdr.ioc_len < sizeof(*config))
2660 mutex_lock(&the_lnet.ln_api_mutex);
2661 rc = lnet_get_route(config->cfg_count,
2663 &config->cfg_config_u.cfg_route.rtr_hop,
2665 &config->cfg_config_u.cfg_route.rtr_flags,
2666 &config->cfg_config_u.cfg_route.
2668 mutex_unlock(&the_lnet.ln_api_mutex);
2671 case IOC_LIBCFS_GET_LOCAL_NI: {
2672 struct lnet_ioctl_config_ni *cfg_ni;
2673 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
2674 struct lnet_ioctl_element_stats *stats;
2678 /* get the tunables if they are available */
2679 if (cfg_ni->lic_cfg_hdr.ioc_len <
2680 sizeof(*cfg_ni) + sizeof(*stats)+ sizeof(*tun))
2683 stats = (struct lnet_ioctl_element_stats *)
2685 tun = (struct lnet_ioctl_config_lnd_tunables *)
2686 (cfg_ni->lic_bulk + sizeof(*stats));
2688 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
2691 mutex_lock(&the_lnet.ln_api_mutex);
2692 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
2693 mutex_unlock(&the_lnet.ln_api_mutex);
2697 case IOC_LIBCFS_GET_NET: {
2698 size_t total = sizeof(*config) +
2699 sizeof(struct lnet_ioctl_net_config);
2702 if (config->cfg_hdr.ioc_len < total)
2705 mutex_lock(&the_lnet.ln_api_mutex);
2706 rc = lnet_get_net_config(config);
2707 mutex_unlock(&the_lnet.ln_api_mutex);
2711 case IOC_LIBCFS_GET_LNET_STATS:
2713 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
2715 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
2718 mutex_lock(&the_lnet.ln_api_mutex);
2719 lnet_counters_get(&lnet_stats->st_cntrs);
2720 mutex_unlock(&the_lnet.ln_api_mutex);
2724 case IOC_LIBCFS_CONFIG_RTR:
2727 if (config->cfg_hdr.ioc_len < sizeof(*config))
2730 mutex_lock(&the_lnet.ln_api_mutex);
2731 if (config->cfg_config_u.cfg_buffers.buf_enable) {
2732 rc = lnet_rtrpools_enable();
2733 mutex_unlock(&the_lnet.ln_api_mutex);
2736 lnet_rtrpools_disable();
2737 mutex_unlock(&the_lnet.ln_api_mutex);
2740 case IOC_LIBCFS_ADD_BUF:
2743 if (config->cfg_hdr.ioc_len < sizeof(*config))
2746 mutex_lock(&the_lnet.ln_api_mutex);
2747 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
2749 config->cfg_config_u.cfg_buffers.
2751 config->cfg_config_u.cfg_buffers.
2753 mutex_unlock(&the_lnet.ln_api_mutex);
2756 case IOC_LIBCFS_SET_NUMA_RANGE: {
2757 struct lnet_ioctl_numa_range *numa;
2759 if (numa->nr_hdr.ioc_len != sizeof(*numa))
2761 mutex_lock(&the_lnet.ln_api_mutex);
2762 lnet_numa_range = numa->nr_range;
2763 mutex_unlock(&the_lnet.ln_api_mutex);
2767 case IOC_LIBCFS_GET_NUMA_RANGE: {
2768 struct lnet_ioctl_numa_range *numa;
2770 if (numa->nr_hdr.ioc_len != sizeof(*numa))
2772 numa->nr_range = lnet_numa_range;
2776 case IOC_LIBCFS_GET_BUF: {
2777 struct lnet_ioctl_pool_cfg *pool_cfg;
2778 size_t total = sizeof(*config) + sizeof(*pool_cfg);
2782 if (config->cfg_hdr.ioc_len < total)
2785 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
2787 mutex_lock(&the_lnet.ln_api_mutex);
2788 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
2789 mutex_unlock(&the_lnet.ln_api_mutex);
2793 case IOC_LIBCFS_ADD_PEER_NI: {
2794 struct lnet_ioctl_peer_cfg *cfg = arg;
2796 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
2799 mutex_lock(&the_lnet.ln_api_mutex);
2800 rc = lnet_add_peer_ni_to_peer(cfg->prcfg_prim_nid,
2803 mutex_unlock(&the_lnet.ln_api_mutex);
2807 case IOC_LIBCFS_DEL_PEER_NI: {
2808 struct lnet_ioctl_peer_cfg *cfg = arg;
2810 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
2813 mutex_lock(&the_lnet.ln_api_mutex);
2814 rc = lnet_del_peer_ni_from_peer(cfg->prcfg_prim_nid,
2815 cfg->prcfg_cfg_nid);
2816 mutex_unlock(&the_lnet.ln_api_mutex);
2820 case IOC_LIBCFS_GET_PEER_INFO: {
2821 struct lnet_ioctl_peer *peer_info = arg;
2823 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
2826 mutex_lock(&the_lnet.ln_api_mutex);
2827 rc = lnet_get_peer_ni_info(
2828 peer_info->pr_count,
2830 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
2831 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
2832 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
2833 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
2834 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
2835 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
2836 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
2837 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
2838 mutex_unlock(&the_lnet.ln_api_mutex);
2842 case IOC_LIBCFS_GET_PEER_NI: {
2843 struct lnet_ioctl_peer_cfg *cfg = arg;
2844 struct lnet_peer_ni_credit_info __user *lpni_cri;
2845 struct lnet_ioctl_element_stats __user *lpni_stats;
2846 size_t usr_size = sizeof(*lpni_cri) + sizeof(*lpni_stats);
2848 if ((cfg->prcfg_hdr.ioc_len != sizeof(*cfg)) ||
2849 (cfg->prcfg_size != usr_size))
2852 lpni_cri = cfg->prcfg_bulk;
2853 lpni_stats = cfg->prcfg_bulk + sizeof(*lpni_cri);
2855 mutex_lock(&the_lnet.ln_api_mutex);
2856 rc = lnet_get_peer_info(cfg->prcfg_count, &cfg->prcfg_prim_nid,
2857 &cfg->prcfg_cfg_nid, &cfg->prcfg_mr,
2858 lpni_cri, lpni_stats);
2859 mutex_unlock(&the_lnet.ln_api_mutex);
2863 case IOC_LIBCFS_NOTIFY_ROUTER: {
2864 unsigned long jiffies_passed;
2866 jiffies_passed = ktime_get_real_seconds() - data->ioc_u64[0];
2867 jiffies_passed = cfs_time_seconds(jiffies_passed);
2869 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
2870 jiffies - jiffies_passed);
2873 case IOC_LIBCFS_LNET_DIST:
2874 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
2875 if (rc < 0 && rc != -EHOSTUNREACH)
2878 data->ioc_u32[0] = rc;
2881 case IOC_LIBCFS_TESTPROTOCOMPAT:
2882 lnet_net_lock(LNET_LOCK_EX);
2883 the_lnet.ln_testprotocompat = data->ioc_flags;
2884 lnet_net_unlock(LNET_LOCK_EX);
2887 case IOC_LIBCFS_LNET_FAULT:
2888 return lnet_fault_ctl(data->ioc_flags, data);
2890 case IOC_LIBCFS_PING: {
2891 signed long timeout;
2893 id.nid = data->ioc_nid;
2894 id.pid = data->ioc_u32[0];
2896 /* Don't block longer than 2 minutes */
2897 if (data->ioc_u32[1] > 120 * MSEC_PER_SEC)
2900 /* If timestamp is negative then disable timeout */
2901 if ((s32)data->ioc_u32[1] < 0)
2902 timeout = MAX_SCHEDULE_TIMEOUT;
2904 timeout = msecs_to_jiffies(data->ioc_u32[1]);
2906 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
2907 data->ioc_plen1 / sizeof(struct lnet_process_id));
2910 data->ioc_count = rc;
2915 ni = lnet_net2ni_addref(data->ioc_net);
2919 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
2922 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
2929 EXPORT_SYMBOL(LNetCtl);
2931 void LNetDebugPeer(struct lnet_process_id id)
2933 lnet_debug_peer(id.nid);
2935 EXPORT_SYMBOL(LNetDebugPeer);
2938 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
2939 * Note that all interfaces share a same PID, as requested by LNetNIInit().
2941 * \param index Index of the interface to look up.
2942 * \param id On successful return, this location will hold the
2943 * struct lnet_process_id ID of the interface.
2945 * \retval 0 If an interface exists at \a index.
2946 * \retval -ENOENT If no interface has been found.
2949 LNetGetId(unsigned int index, struct lnet_process_id *id)
2952 struct lnet_net *net;
2956 LASSERT(the_lnet.ln_refcount > 0);
2958 cpt = lnet_net_lock_current();
2960 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2961 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2965 id->nid = ni->ni_nid;
2966 id->pid = the_lnet.ln_pid;
2972 lnet_net_unlock(cpt);
2975 EXPORT_SYMBOL(LNetGetId);
2977 static int lnet_ping(struct lnet_process_id id, signed long timeout,
2978 struct lnet_process_id __user *ids, int n_ids)
2980 struct lnet_handle_eq eqh;
2981 struct lnet_handle_md mdh;
2982 struct lnet_event event;
2983 struct lnet_md md = { NULL };
2987 const signed long a_long_time = msecs_to_jiffies(60 * MSEC_PER_SEC);
2989 struct lnet_ping_info *info;
2990 struct lnet_process_id tmpid;
2997 infosz = offsetof(struct lnet_ping_info, pi_ni[n_ids]);
2999 /* n_ids limit is arbitrary */
3000 if (n_ids <= 0 || n_ids > 20 || id.nid == LNET_NID_ANY)
3003 if (id.pid == LNET_PID_ANY)
3004 id.pid = LNET_PID_LUSTRE;
3006 LIBCFS_ALLOC(info, infosz);
3010 /* NB 2 events max (including any unlink event) */
3011 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
3013 CERROR("Can't allocate EQ: %d\n", rc);
3017 /* initialize md content */
3020 md.threshold = 2; /*GET/REPLY*/
3022 md.options = LNET_MD_TRUNCATE;
3026 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
3028 CERROR("Can't bind MD: %d\n", rc);
3032 rc = LNetGet(LNET_NID_ANY, mdh, id,
3033 LNET_RESERVED_PORTAL,
3034 LNET_PROTO_PING_MATCHBITS, 0);
3037 /* Don't CERROR; this could be deliberate! */
3039 rc2 = LNetMDUnlink(mdh);
3042 /* NB must wait for the UNLINK event below... */
3044 timeout = a_long_time;
3048 /* MUST block for unlink to complete */
3050 blocked = cfs_block_allsigs();
3052 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
3055 cfs_restore_sigs(blocked);
3057 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
3058 (rc2 <= 0) ? -1 : event.type,
3059 (rc2 <= 0) ? -1 : event.status,
3060 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
3062 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
3064 if (rc2 <= 0 || event.status != 0) {
3065 /* timeout or error */
3066 if (!replied && rc == 0)
3067 rc = (rc2 < 0) ? rc2 :
3068 (rc2 == 0) ? -ETIMEDOUT :
3072 /* Ensure completion in finite time... */
3074 /* No assertion (racing with network) */
3076 timeout = a_long_time;
3077 } else if (rc2 == 0) {
3078 /* timed out waiting for unlink */
3079 CWARN("ping %s: late network completion\n",
3082 } else if (event.type == LNET_EVENT_REPLY) {
3087 } while (rc2 <= 0 || !event.unlinked);
3091 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
3098 LASSERT(nob >= 0 && nob <= infosz);
3100 rc = -EPROTO; /* if I can't parse... */
3103 /* can't check magic/version */
3104 CERROR("%s: ping info too short %d\n",
3105 libcfs_id2str(id), nob);
3109 if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
3110 lnet_swap_pinginfo(info);
3111 } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
3112 CERROR("%s: Unexpected magic %08x\n",
3113 libcfs_id2str(id), info->pi_magic);
3117 if ((info->pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
3118 CERROR("%s: ping w/o NI status: 0x%x\n",
3119 libcfs_id2str(id), info->pi_features);
3123 if (nob < offsetof(struct lnet_ping_info, pi_ni[0])) {
3124 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
3125 nob, (int)offsetof(struct lnet_ping_info, pi_ni[0]));
3129 if (info->pi_nnis < n_ids)
3130 n_ids = info->pi_nnis;
3132 if (nob < offsetof(struct lnet_ping_info, pi_ni[n_ids])) {
3133 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
3134 nob, (int)offsetof(struct lnet_ping_info, pi_ni[n_ids]));
3138 rc = -EFAULT; /* If I SEGV... */
3140 memset(&tmpid, 0, sizeof(tmpid));
3141 for (i = 0; i < n_ids; i++) {
3142 tmpid.pid = info->pi_pid;
3143 tmpid.nid = info->pi_ni[i].ns_nid;
3144 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
3150 rc2 = LNetEQFree(eqh);
3152 CERROR("rc2 %d\n", rc2);
3156 LIBCFS_FREE(info, infosz);