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, 2017, 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
35 #include <linux/ctype.h>
36 #include <linux/log2.h>
37 #include <linux/ktime.h>
38 #include <linux/moduleparam.h>
39 #include <linux/uaccess.h>
41 #include <lnet/lib-lnet.h>
43 #define D_LNI D_CONSOLE
46 * initialize ln_api_mutex statically, since it needs to be used in
47 * discovery_set callback. That module parameter callback can be called
48 * before module init completes. The mutex needs to be ready for use then.
50 struct lnet the_lnet = {
51 .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
52 }; /* THE state of the network */
53 EXPORT_SYMBOL(the_lnet);
55 static char *ip2nets = "";
56 module_param(ip2nets, charp, 0444);
57 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
59 static char *networks = "";
60 module_param(networks, charp, 0444);
61 MODULE_PARM_DESC(networks, "local networks");
63 static char *routes = "";
64 module_param(routes, charp, 0444);
65 MODULE_PARM_DESC(routes, "routes to non-local networks");
67 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
68 module_param(rnet_htable_size, int, 0444);
69 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
71 static int use_tcp_bonding = false;
72 module_param(use_tcp_bonding, int, 0444);
73 MODULE_PARM_DESC(use_tcp_bonding,
74 "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
76 unsigned int lnet_numa_range = 0;
77 module_param(lnet_numa_range, uint, 0444);
78 MODULE_PARM_DESC(lnet_numa_range,
79 "NUMA range to consider during Multi-Rail selection");
82 * lnet_health_sensitivity determines by how much we decrement the health
83 * value on sending error. The value defaults to 100, which means health
84 * interface health is decremented by 100 points every failure.
86 unsigned int lnet_health_sensitivity = 100;
87 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
88 #ifdef HAVE_KERNEL_PARAM_OPS
89 static struct kernel_param_ops param_ops_health_sensitivity = {
90 .set = sensitivity_set,
93 #define param_check_health_sensitivity(name, p) \
94 __param_check(name, p, int)
95 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
97 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
98 &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
100 MODULE_PARM_DESC(lnet_health_sensitivity,
101 "Value to decrement the health value by on error");
104 * lnet_recovery_interval determines how often we should perform recovery
105 * on unhealthy interfaces.
107 unsigned int lnet_recovery_interval = 1;
108 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
109 #ifdef HAVE_KERNEL_PARAM_OPS
110 static struct kernel_param_ops param_ops_recovery_interval = {
111 .set = recovery_interval_set,
112 .get = param_get_int,
114 #define param_check_recovery_interval(name, p) \
115 __param_check(name, p, int)
116 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
118 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
119 &lnet_recovery_interval, S_IRUGO|S_IWUSR);
121 MODULE_PARM_DESC(lnet_recovery_interval,
122 "Interval to recover unhealthy interfaces in seconds");
124 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
125 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
127 #ifdef HAVE_KERNEL_PARAM_OPS
128 static struct kernel_param_ops param_ops_interfaces_max = {
130 .get = param_get_int,
133 #define param_check_interfaces_max(name, p) \
134 __param_check(name, p, int)
136 module_param(lnet_interfaces_max, interfaces_max, 0644);
138 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
139 &lnet_interfaces_max, 0644);
141 MODULE_PARM_DESC(lnet_interfaces_max,
142 "Maximum number of interfaces in a node.");
144 unsigned lnet_peer_discovery_disabled = 0;
145 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
147 #ifdef HAVE_KERNEL_PARAM_OPS
148 static struct kernel_param_ops param_ops_discovery_disabled = {
149 .set = discovery_set,
150 .get = param_get_int,
153 #define param_check_discovery_disabled(name, p) \
154 __param_check(name, p, int)
155 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
157 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
158 &lnet_peer_discovery_disabled, 0644);
160 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
161 "Set to 1 to disable peer discovery on this node.");
163 unsigned int lnet_drop_asym_route;
164 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
166 #ifdef HAVE_KERNEL_PARAM_OPS
167 static struct kernel_param_ops param_ops_drop_asym_route = {
168 .set = drop_asym_route_set,
169 .get = param_get_int,
172 #define param_check_drop_asym_route(name, p) \
173 __param_check(name, p, int)
174 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
176 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
177 &lnet_drop_asym_route, 0644);
179 MODULE_PARM_DESC(lnet_drop_asym_route,
180 "Set to 1 to drop asymmetrical route messages.");
182 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
183 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 50
185 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
186 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
187 #ifdef HAVE_KERNEL_PARAM_OPS
188 static struct kernel_param_ops param_ops_transaction_timeout = {
189 .set = transaction_to_set,
190 .get = param_get_int,
193 #define param_check_transaction_timeout(name, p) \
194 __param_check(name, p, int)
195 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
197 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
198 &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
200 MODULE_PARM_DESC(lnet_transaction_timeout,
201 "Maximum number of seconds to wait for a peer response.");
203 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 2
204 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
205 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
206 #ifdef HAVE_KERNEL_PARAM_OPS
207 static struct kernel_param_ops param_ops_retry_count = {
208 .set = retry_count_set,
209 .get = param_get_int,
212 #define param_check_retry_count(name, p) \
213 __param_check(name, p, int)
214 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
216 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
217 &lnet_retry_count, S_IRUGO|S_IWUSR);
219 MODULE_PARM_DESC(lnet_retry_count,
220 "Maximum number of times to retry transmitting a message");
223 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
226 * This sequence number keeps track of how many times DLC was used to
227 * update the local NIs. It is incremented when a NI is added or
228 * removed and checked when sending a message to determine if there is
229 * a need to re-run the selection algorithm. See lnet_select_pathway()
230 * for more details on its usage.
232 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
234 static int lnet_ping(struct lnet_process_id id, signed long timeout,
235 struct lnet_process_id __user *ids, int n_ids);
237 static int lnet_discover(struct lnet_process_id id, __u32 force,
238 struct lnet_process_id __user *ids, int n_ids);
241 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
244 unsigned *sensitivity = (unsigned *)kp->arg;
247 rc = kstrtoul(val, 0, &value);
249 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
254 * The purpose of locking the api_mutex here is to ensure that
255 * the correct value ends up stored properly.
257 mutex_lock(&the_lnet.ln_api_mutex);
259 if (value > LNET_MAX_HEALTH_VALUE) {
260 mutex_unlock(&the_lnet.ln_api_mutex);
261 CERROR("Invalid health value. Maximum: %d value = %lu\n",
262 LNET_MAX_HEALTH_VALUE, value);
267 * if we're turning on health then use the health timeout
270 if (*sensitivity == 0 && value != 0) {
271 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
272 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
274 * if we're turning off health then use the no health timeout
277 } else if (*sensitivity != 0 && value == 0) {
278 lnet_transaction_timeout =
279 LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
280 lnet_retry_count = 0;
283 *sensitivity = value;
285 mutex_unlock(&the_lnet.ln_api_mutex);
291 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
294 unsigned *interval = (unsigned *)kp->arg;
297 rc = kstrtoul(val, 0, &value);
299 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
304 CERROR("lnet_recovery_interval must be at least 1 second\n");
309 * The purpose of locking the api_mutex here is to ensure that
310 * the correct value ends up stored properly.
312 mutex_lock(&the_lnet.ln_api_mutex);
316 mutex_unlock(&the_lnet.ln_api_mutex);
322 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
325 unsigned *discovery = (unsigned *)kp->arg;
327 struct lnet_ping_buffer *pbuf;
329 rc = kstrtoul(val, 0, &value);
331 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
335 value = (value) ? 1 : 0;
338 * The purpose of locking the api_mutex here is to ensure that
339 * the correct value ends up stored properly.
341 mutex_lock(&the_lnet.ln_api_mutex);
343 if (value == *discovery) {
344 mutex_unlock(&the_lnet.ln_api_mutex);
350 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
351 mutex_unlock(&the_lnet.ln_api_mutex);
355 /* tell peers that discovery setting has changed */
356 lnet_net_lock(LNET_LOCK_EX);
357 pbuf = the_lnet.ln_ping_target;
359 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
361 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
362 lnet_net_unlock(LNET_LOCK_EX);
364 lnet_push_update_to_peers(1);
366 mutex_unlock(&the_lnet.ln_api_mutex);
372 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
375 unsigned int *drop_asym_route = (unsigned int *)kp->arg;
378 rc = kstrtoul(val, 0, &value);
380 CERROR("Invalid module parameter value for "
381 "'lnet_drop_asym_route'\n");
386 * The purpose of locking the api_mutex here is to ensure that
387 * the correct value ends up stored properly.
389 mutex_lock(&the_lnet.ln_api_mutex);
391 if (value == *drop_asym_route) {
392 mutex_unlock(&the_lnet.ln_api_mutex);
396 *drop_asym_route = value;
398 mutex_unlock(&the_lnet.ln_api_mutex);
404 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
407 unsigned *transaction_to = (unsigned *)kp->arg;
410 rc = kstrtoul(val, 0, &value);
412 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
417 * The purpose of locking the api_mutex here is to ensure that
418 * the correct value ends up stored properly.
420 mutex_lock(&the_lnet.ln_api_mutex);
422 if (value < lnet_retry_count || value == 0) {
423 mutex_unlock(&the_lnet.ln_api_mutex);
424 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
425 "Has to be greater than lnet_retry_count (%u)\n",
426 value, lnet_retry_count);
430 if (value == *transaction_to) {
431 mutex_unlock(&the_lnet.ln_api_mutex);
435 *transaction_to = value;
436 if (lnet_retry_count == 0)
437 lnet_lnd_timeout = value;
439 lnet_lnd_timeout = value / lnet_retry_count;
441 mutex_unlock(&the_lnet.ln_api_mutex);
447 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
450 unsigned *retry_count = (unsigned *)kp->arg;
453 rc = kstrtoul(val, 0, &value);
455 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
460 * The purpose of locking the api_mutex here is to ensure that
461 * the correct value ends up stored properly.
463 mutex_lock(&the_lnet.ln_api_mutex);
465 if (lnet_health_sensitivity == 0) {
466 mutex_unlock(&the_lnet.ln_api_mutex);
467 CERROR("Can not set retry_count when health feature is turned off\n");
471 if (value > lnet_transaction_timeout) {
472 mutex_unlock(&the_lnet.ln_api_mutex);
473 CERROR("Invalid value for lnet_retry_count (%lu). "
474 "Has to be smaller than lnet_transaction_timeout (%u)\n",
475 value, lnet_transaction_timeout);
479 *retry_count = value;
482 lnet_lnd_timeout = lnet_transaction_timeout;
484 lnet_lnd_timeout = lnet_transaction_timeout / value;
486 mutex_unlock(&the_lnet.ln_api_mutex);
492 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
496 rc = kstrtoint(val, 0, &value);
498 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
502 if (value < LNET_INTERFACES_MIN) {
503 CWARN("max interfaces provided are too small, setting to %d\n",
504 LNET_INTERFACES_MAX_DEFAULT);
505 value = LNET_INTERFACES_MAX_DEFAULT;
508 *(int *)kp->arg = value;
514 lnet_get_routes(void)
520 lnet_get_networks(void)
525 if (*networks != 0 && *ip2nets != 0) {
526 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
527 "'ip2nets' but not both at once\n");
532 rc = lnet_parse_ip2nets(&nets, ip2nets);
533 return (rc == 0) ? nets : NULL;
543 lnet_init_locks(void)
545 spin_lock_init(&the_lnet.ln_eq_wait_lock);
546 spin_lock_init(&the_lnet.ln_msg_resend_lock);
547 init_waitqueue_head(&the_lnet.ln_eq_waitq);
548 init_waitqueue_head(&the_lnet.ln_mt_waitq);
549 mutex_init(&the_lnet.ln_lnd_mutex);
553 lnet_fini_locks(void)
557 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
558 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
562 lnet_descriptor_setup(void)
564 /* create specific kmem_cache for MEs and small MDs (i.e., originally
565 * allocated in <size-xxx> kmem_cache).
567 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
569 if (!lnet_mes_cachep)
572 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
573 LNET_SMALL_MD_SIZE, 0, 0,
575 if (!lnet_small_mds_cachep)
582 lnet_descriptor_cleanup(void)
585 if (lnet_small_mds_cachep) {
586 kmem_cache_destroy(lnet_small_mds_cachep);
587 lnet_small_mds_cachep = NULL;
590 if (lnet_mes_cachep) {
591 kmem_cache_destroy(lnet_mes_cachep);
592 lnet_mes_cachep = NULL;
597 lnet_create_remote_nets_table(void)
600 struct list_head *hash;
602 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
603 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
604 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
606 CERROR("Failed to create remote nets hash table\n");
610 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
611 INIT_LIST_HEAD(&hash[i]);
612 the_lnet.ln_remote_nets_hash = hash;
617 lnet_destroy_remote_nets_table(void)
621 if (the_lnet.ln_remote_nets_hash == NULL)
624 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
625 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
627 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
628 LNET_REMOTE_NETS_HASH_SIZE *
629 sizeof(the_lnet.ln_remote_nets_hash[0]));
630 the_lnet.ln_remote_nets_hash = NULL;
634 lnet_destroy_locks(void)
636 if (the_lnet.ln_res_lock != NULL) {
637 cfs_percpt_lock_free(the_lnet.ln_res_lock);
638 the_lnet.ln_res_lock = NULL;
641 if (the_lnet.ln_net_lock != NULL) {
642 cfs_percpt_lock_free(the_lnet.ln_net_lock);
643 the_lnet.ln_net_lock = NULL;
650 lnet_create_locks(void)
654 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
655 if (the_lnet.ln_res_lock == NULL)
658 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
659 if (the_lnet.ln_net_lock == NULL)
665 lnet_destroy_locks();
669 static void lnet_assert_wire_constants(void)
671 /* Wire protocol assertions generated by 'wirecheck'
672 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
673 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
674 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
677 CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
678 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
679 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
680 CLASSERT(LNET_MSG_ACK == 0);
681 CLASSERT(LNET_MSG_PUT == 1);
682 CLASSERT(LNET_MSG_GET == 2);
683 CLASSERT(LNET_MSG_REPLY == 3);
684 CLASSERT(LNET_MSG_HELLO == 4);
686 /* Checks for struct lnet_handle_wire */
687 CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
688 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
689 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
690 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
691 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
693 /* Checks for struct struct lnet_magicversion */
694 CLASSERT((int)sizeof(struct lnet_magicversion) == 8);
695 CLASSERT((int)offsetof(struct lnet_magicversion, magic) == 0);
696 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->magic) == 4);
697 CLASSERT((int)offsetof(struct lnet_magicversion, version_major) == 4);
698 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_major) == 2);
699 CLASSERT((int)offsetof(struct lnet_magicversion, version_minor) == 6);
700 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_minor) == 2);
702 /* Checks for struct struct lnet_hdr */
703 CLASSERT((int)sizeof(struct lnet_hdr) == 72);
704 CLASSERT((int)offsetof(struct lnet_hdr, dest_nid) == 0);
705 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_nid) == 8);
706 CLASSERT((int)offsetof(struct lnet_hdr, src_nid) == 8);
707 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_nid) == 8);
708 CLASSERT((int)offsetof(struct lnet_hdr, dest_pid) == 16);
709 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_pid) == 4);
710 CLASSERT((int)offsetof(struct lnet_hdr, src_pid) == 20);
711 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_pid) == 4);
712 CLASSERT((int)offsetof(struct lnet_hdr, type) == 24);
713 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->type) == 4);
714 CLASSERT((int)offsetof(struct lnet_hdr, payload_length) == 28);
715 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->payload_length) == 4);
716 CLASSERT((int)offsetof(struct lnet_hdr, msg) == 32);
717 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg) == 40);
720 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) == 32);
721 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) == 16);
722 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.match_bits) == 48);
723 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) == 8);
724 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.mlength) == 56);
725 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) == 4);
728 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) == 32);
729 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) == 16);
730 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.match_bits) == 48);
731 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) == 8);
732 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.hdr_data) == 56);
733 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) == 8);
734 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ptl_index) == 64);
735 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) == 4);
736 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.offset) == 68);
737 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) == 4);
740 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.return_wmd) == 32);
741 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) == 16);
742 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.match_bits) == 48);
743 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) == 8);
744 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.ptl_index) == 56);
745 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) == 4);
746 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.src_offset) == 60);
747 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) == 4);
748 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.sink_length) == 64);
749 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) == 4);
752 CLASSERT((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) == 32);
753 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) == 16);
756 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.incarnation) == 32);
757 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) == 8);
758 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.type) == 40);
759 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) == 4);
761 /* Checks for struct lnet_ni_status and related constants */
762 CLASSERT(LNET_NI_STATUS_INVALID == 0x00000000);
763 CLASSERT(LNET_NI_STATUS_UP == 0x15aac0de);
764 CLASSERT(LNET_NI_STATUS_DOWN == 0xdeadface);
766 /* Checks for struct lnet_ni_status */
767 CLASSERT((int)sizeof(struct lnet_ni_status) == 16);
768 CLASSERT((int)offsetof(struct lnet_ni_status, ns_nid) == 0);
769 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) == 8);
770 CLASSERT((int)offsetof(struct lnet_ni_status, ns_status) == 8);
771 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_status) == 4);
772 CLASSERT((int)offsetof(struct lnet_ni_status, ns_unused) == 12);
773 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) == 4);
775 /* Checks for struct lnet_ping_info and related constants */
776 CLASSERT(LNET_PROTO_PING_MAGIC == 0x70696E67);
777 CLASSERT(LNET_PING_FEAT_INVAL == 0);
778 CLASSERT(LNET_PING_FEAT_BASE == 1);
779 CLASSERT(LNET_PING_FEAT_NI_STATUS == 2);
780 CLASSERT(LNET_PING_FEAT_RTE_DISABLED == 4);
781 CLASSERT(LNET_PING_FEAT_MULTI_RAIL == 8);
782 CLASSERT(LNET_PING_FEAT_DISCOVERY == 16);
783 CLASSERT(LNET_PING_FEAT_BITS == 31);
785 /* Checks for struct lnet_ping_info */
786 CLASSERT((int)sizeof(struct lnet_ping_info) == 16);
787 CLASSERT((int)offsetof(struct lnet_ping_info, pi_magic) == 0);
788 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) == 4);
789 CLASSERT((int)offsetof(struct lnet_ping_info, pi_features) == 4);
790 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_features) == 4);
791 CLASSERT((int)offsetof(struct lnet_ping_info, pi_pid) == 8);
792 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) == 4);
793 CLASSERT((int)offsetof(struct lnet_ping_info, pi_nnis) == 12);
794 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) == 4);
795 CLASSERT((int)offsetof(struct lnet_ping_info, pi_ni) == 16);
796 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) == 0);
799 static struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
801 struct lnet_lnd *lnd;
802 struct list_head *tmp;
804 /* holding lnd mutex */
805 list_for_each(tmp, &the_lnet.ln_lnds) {
806 lnd = list_entry(tmp, struct lnet_lnd, lnd_list);
808 if (lnd->lnd_type == type)
815 lnet_get_lnd_timeout(void)
817 return lnet_lnd_timeout;
819 EXPORT_SYMBOL(lnet_get_lnd_timeout);
822 lnet_register_lnd(struct lnet_lnd *lnd)
824 mutex_lock(&the_lnet.ln_lnd_mutex);
826 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
827 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
829 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
830 lnd->lnd_refcount = 0;
832 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
834 mutex_unlock(&the_lnet.ln_lnd_mutex);
836 EXPORT_SYMBOL(lnet_register_lnd);
839 lnet_unregister_lnd(struct lnet_lnd *lnd)
841 mutex_lock(&the_lnet.ln_lnd_mutex);
843 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
844 LASSERT(lnd->lnd_refcount == 0);
846 list_del(&lnd->lnd_list);
847 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
849 mutex_unlock(&the_lnet.ln_lnd_mutex);
851 EXPORT_SYMBOL(lnet_unregister_lnd);
854 lnet_counters_get_common(struct lnet_counters_common *common)
856 struct lnet_counters *ctr;
859 memset(common, 0, sizeof(*common));
861 lnet_net_lock(LNET_LOCK_EX);
863 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
864 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
865 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
866 common->lcc_errors += ctr->lct_common.lcc_errors;
867 common->lcc_send_count += ctr->lct_common.lcc_send_count;
868 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
869 common->lcc_route_count += ctr->lct_common.lcc_route_count;
870 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
871 common->lcc_send_length += ctr->lct_common.lcc_send_length;
872 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
873 common->lcc_route_length += ctr->lct_common.lcc_route_length;
874 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
876 lnet_net_unlock(LNET_LOCK_EX);
878 EXPORT_SYMBOL(lnet_counters_get_common);
881 lnet_counters_get(struct lnet_counters *counters)
883 struct lnet_counters *ctr;
884 struct lnet_counters_health *health = &counters->lct_health;
887 memset(counters, 0, sizeof(*counters));
889 lnet_counters_get_common(&counters->lct_common);
891 lnet_net_lock(LNET_LOCK_EX);
893 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
894 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
895 health->lch_resend_count += ctr->lct_health.lch_resend_count;
896 health->lch_response_timeout_count +=
897 ctr->lct_health.lch_response_timeout_count;
898 health->lch_local_interrupt_count +=
899 ctr->lct_health.lch_local_interrupt_count;
900 health->lch_local_dropped_count +=
901 ctr->lct_health.lch_local_dropped_count;
902 health->lch_local_aborted_count +=
903 ctr->lct_health.lch_local_aborted_count;
904 health->lch_local_no_route_count +=
905 ctr->lct_health.lch_local_no_route_count;
906 health->lch_local_timeout_count +=
907 ctr->lct_health.lch_local_timeout_count;
908 health->lch_local_error_count +=
909 ctr->lct_health.lch_local_error_count;
910 health->lch_remote_dropped_count +=
911 ctr->lct_health.lch_remote_dropped_count;
912 health->lch_remote_error_count +=
913 ctr->lct_health.lch_remote_error_count;
914 health->lch_remote_timeout_count +=
915 ctr->lct_health.lch_remote_timeout_count;
916 health->lch_network_timeout_count +=
917 ctr->lct_health.lch_network_timeout_count;
919 lnet_net_unlock(LNET_LOCK_EX);
921 EXPORT_SYMBOL(lnet_counters_get);
924 lnet_counters_reset(void)
926 struct lnet_counters *counters;
929 lnet_net_lock(LNET_LOCK_EX);
931 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
932 memset(counters, 0, sizeof(struct lnet_counters));
934 lnet_net_unlock(LNET_LOCK_EX);
938 lnet_res_type2str(int type)
943 case LNET_COOKIE_TYPE_MD:
945 case LNET_COOKIE_TYPE_ME:
947 case LNET_COOKIE_TYPE_EQ:
953 lnet_res_container_cleanup(struct lnet_res_container *rec)
957 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
960 while (!list_empty(&rec->rec_active)) {
961 struct list_head *e = rec->rec_active.next;
964 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
965 lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
967 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
968 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
970 } else { /* NB: Active MEs should be attached on portals */
977 /* Found alive MD/ME/EQ, user really should unlink/free
978 * all of them before finalize LNet, but if someone didn't,
979 * we have to recycle garbage for him */
980 CERROR("%d active elements on exit of %s container\n",
981 count, lnet_res_type2str(rec->rec_type));
984 if (rec->rec_lh_hash != NULL) {
985 LIBCFS_FREE(rec->rec_lh_hash,
986 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
987 rec->rec_lh_hash = NULL;
990 rec->rec_type = 0; /* mark it as finalized */
994 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
999 LASSERT(rec->rec_type == 0);
1001 rec->rec_type = type;
1002 INIT_LIST_HEAD(&rec->rec_active);
1004 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1006 /* Arbitrary choice of hash table size */
1007 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1008 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1009 if (rec->rec_lh_hash == NULL) {
1014 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1015 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1020 CERROR("Failed to setup %s resource container\n",
1021 lnet_res_type2str(type));
1022 lnet_res_container_cleanup(rec);
1027 lnet_res_containers_destroy(struct lnet_res_container **recs)
1029 struct lnet_res_container *rec;
1032 cfs_percpt_for_each(rec, i, recs)
1033 lnet_res_container_cleanup(rec);
1035 cfs_percpt_free(recs);
1038 static struct lnet_res_container **
1039 lnet_res_containers_create(int type)
1041 struct lnet_res_container **recs;
1042 struct lnet_res_container *rec;
1046 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1048 CERROR("Failed to allocate %s resource containers\n",
1049 lnet_res_type2str(type));
1053 cfs_percpt_for_each(rec, i, recs) {
1054 rc = lnet_res_container_setup(rec, i, type);
1056 lnet_res_containers_destroy(recs);
1064 struct lnet_libhandle *
1065 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1067 /* ALWAYS called with lnet_res_lock held */
1068 struct list_head *head;
1069 struct lnet_libhandle *lh;
1072 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1075 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1076 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1078 list_for_each_entry(lh, head, lh_hash_chain) {
1079 if (lh->lh_cookie == cookie)
1087 lnet_res_lh_initialize(struct lnet_res_container *rec,
1088 struct lnet_libhandle *lh)
1090 /* ALWAYS called with lnet_res_lock held */
1091 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1094 lh->lh_cookie = rec->rec_lh_cookie;
1095 rec->rec_lh_cookie += 1 << ibits;
1097 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1099 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1103 lnet_create_array_of_queues(void)
1105 struct list_head **qs;
1106 struct list_head *q;
1109 qs = cfs_percpt_alloc(lnet_cpt_table(),
1110 sizeof(struct list_head));
1112 CERROR("Failed to allocate queues\n");
1116 cfs_percpt_for_each(q, i, qs)
1122 static int lnet_unprepare(void);
1125 lnet_prepare(lnet_pid_t requested_pid)
1127 /* Prepare to bring up the network */
1128 struct lnet_res_container **recs;
1131 if (requested_pid == LNET_PID_ANY) {
1132 /* Don't instantiate LNET just for me */
1136 LASSERT(the_lnet.ln_refcount == 0);
1138 the_lnet.ln_routing = 0;
1140 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1141 the_lnet.ln_pid = requested_pid;
1143 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1144 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1145 INIT_LIST_HEAD(&the_lnet.ln_nets);
1146 INIT_LIST_HEAD(&the_lnet.ln_routers);
1147 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1148 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1149 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1150 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1151 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1152 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1153 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1154 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1155 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1157 rc = lnet_descriptor_setup();
1161 rc = lnet_create_remote_nets_table();
1166 * NB the interface cookie in wire handles guards against delayed
1167 * replies and ACKs appearing valid after reboot.
1169 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1171 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1172 sizeof(struct lnet_counters));
1173 if (the_lnet.ln_counters == NULL) {
1174 CERROR("Failed to allocate counters for LNet\n");
1179 rc = lnet_peer_tables_create();
1183 rc = lnet_msg_containers_create();
1187 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1188 LNET_COOKIE_TYPE_EQ);
1192 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
1198 the_lnet.ln_me_containers = recs;
1200 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1206 the_lnet.ln_md_containers = recs;
1208 rc = lnet_portals_create();
1210 CERROR("Failed to create portals for LNet: %d\n", rc);
1214 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1215 if (!the_lnet.ln_mt_zombie_rstqs) {
1228 lnet_unprepare (void)
1232 /* NB no LNET_LOCK since this is the last reference. All LND instances
1233 * have shut down already, so it is safe to unlink and free all
1234 * descriptors, even those that appear committed to a network op (eg MD
1235 * with non-zero pending count) */
1237 lnet_fail_nid(LNET_NID_ANY, 0);
1239 LASSERT(the_lnet.ln_refcount == 0);
1240 LASSERT(list_empty(&the_lnet.ln_test_peers));
1241 LASSERT(list_empty(&the_lnet.ln_nets));
1243 if (the_lnet.ln_mt_zombie_rstqs) {
1244 lnet_clean_zombie_rstqs();
1245 the_lnet.ln_mt_zombie_rstqs = NULL;
1248 if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1249 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1250 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1254 lnet_portals_destroy();
1256 if (the_lnet.ln_md_containers != NULL) {
1257 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1258 the_lnet.ln_md_containers = NULL;
1261 if (the_lnet.ln_me_containers != NULL) {
1262 lnet_res_containers_destroy(the_lnet.ln_me_containers);
1263 the_lnet.ln_me_containers = NULL;
1266 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1268 lnet_msg_containers_destroy();
1270 lnet_rtrpools_free(0);
1272 if (the_lnet.ln_counters != NULL) {
1273 cfs_percpt_free(the_lnet.ln_counters);
1274 the_lnet.ln_counters = NULL;
1276 lnet_destroy_remote_nets_table();
1277 lnet_descriptor_cleanup();
1283 lnet_net2ni_locked(__u32 net_id, int cpt)
1286 struct lnet_net *net;
1288 LASSERT(cpt != LNET_LOCK_EX);
1290 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1291 if (net->net_id == net_id) {
1292 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1302 lnet_net2ni_addref(__u32 net)
1307 ni = lnet_net2ni_locked(net, 0);
1309 lnet_ni_addref_locked(ni, 0);
1314 EXPORT_SYMBOL(lnet_net2ni_addref);
1317 lnet_get_net_locked(__u32 net_id)
1319 struct lnet_net *net;
1321 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1322 if (net->net_id == net_id)
1330 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1335 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1340 val = hash_long(key, LNET_CPT_BITS);
1341 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1345 return (unsigned int)(key + val + (val >> 1)) % number;
1349 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1351 struct lnet_net *net;
1353 /* must called with hold of lnet_net_lock */
1354 if (LNET_CPT_NUMBER == 1)
1355 return 0; /* the only one */
1358 * If NI is provided then use the CPT identified in the NI cpt
1359 * list if one exists. If one doesn't exist, then that NI is
1360 * associated with all CPTs and it follows that the net it belongs
1361 * to is implicitly associated with all CPTs, so just hash the nid
1365 if (ni->ni_cpts != NULL)
1366 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1369 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1372 /* no NI provided so look at the net */
1373 net = lnet_get_net_locked(LNET_NIDNET(nid));
1375 if (net != NULL && net->net_cpts != NULL) {
1376 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1379 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1383 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1388 if (LNET_CPT_NUMBER == 1)
1389 return 0; /* the only one */
1391 cpt = lnet_net_lock_current();
1393 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1395 lnet_net_unlock(cpt);
1399 EXPORT_SYMBOL(lnet_cpt_of_nid);
1402 lnet_islocalnet(__u32 net_id)
1404 struct lnet_net *net;
1408 cpt = lnet_net_lock_current();
1410 net = lnet_get_net_locked(net_id);
1412 local = net != NULL;
1414 lnet_net_unlock(cpt);
1420 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1422 struct lnet_net *net;
1425 LASSERT(cpt != LNET_LOCK_EX);
1427 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1428 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1429 if (ni->ni_nid == nid)
1438 lnet_nid2ni_addref(lnet_nid_t nid)
1443 ni = lnet_nid2ni_locked(nid, 0);
1445 lnet_ni_addref_locked(ni, 0);
1450 EXPORT_SYMBOL(lnet_nid2ni_addref);
1453 lnet_islocalnid(lnet_nid_t nid)
1458 cpt = lnet_net_lock_current();
1459 ni = lnet_nid2ni_locked(nid, cpt);
1460 lnet_net_unlock(cpt);
1466 lnet_count_acceptor_nets(void)
1468 /* Return the # of NIs that need the acceptor. */
1470 struct lnet_net *net;
1473 cpt = lnet_net_lock_current();
1474 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1475 /* all socklnd type networks should have the acceptor
1477 if (net->net_lnd->lnd_accept != NULL)
1481 lnet_net_unlock(cpt);
1486 struct lnet_ping_buffer *
1487 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1489 struct lnet_ping_buffer *pbuf;
1491 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1493 pbuf->pb_nnis = nnis;
1494 atomic_set(&pbuf->pb_refcnt, 1);
1501 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1503 LASSERT(lnet_ping_buffer_numref(pbuf) == 0);
1504 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1507 static struct lnet_ping_buffer *
1508 lnet_ping_target_create(int nnis)
1510 struct lnet_ping_buffer *pbuf;
1512 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1514 CERROR("Can't allocate ping source [%d]\n", nnis);
1518 pbuf->pb_info.pi_nnis = nnis;
1519 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1520 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1521 pbuf->pb_info.pi_features =
1522 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1528 lnet_get_net_ni_count_locked(struct lnet_net *net)
1533 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1540 lnet_get_net_ni_count_pre(struct lnet_net *net)
1545 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1552 lnet_get_ni_count(void)
1555 struct lnet_net *net;
1560 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1561 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1571 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1575 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1577 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1579 /* Loopback is guaranteed to be present */
1580 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1582 if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1588 lnet_ping_target_destroy(void)
1590 struct lnet_net *net;
1593 lnet_net_lock(LNET_LOCK_EX);
1595 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1596 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1598 ni->ni_status = NULL;
1603 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1604 the_lnet.ln_ping_target = NULL;
1606 lnet_net_unlock(LNET_LOCK_EX);
1610 lnet_ping_target_event_handler(struct lnet_event *event)
1612 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1614 if (event->unlinked)
1615 lnet_ping_buffer_decref(pbuf);
1619 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1620 struct lnet_handle_md *ping_mdh,
1621 int ni_count, bool set_eq)
1623 struct lnet_process_id id = {
1624 .nid = LNET_NID_ANY,
1627 struct lnet_handle_me me_handle;
1628 struct lnet_md md = { NULL };
1632 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1633 &the_lnet.ln_ping_target_eq);
1635 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1640 *ppbuf = lnet_ping_target_create(ni_count);
1641 if (*ppbuf == NULL) {
1646 /* Ping target ME/MD */
1647 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1648 LNET_PROTO_PING_MATCHBITS, 0,
1649 LNET_UNLINK, LNET_INS_AFTER,
1652 CERROR("Can't create ping target ME: %d\n", rc);
1653 goto fail_decref_ping_buffer;
1656 /* initialize md content */
1657 md.start = &(*ppbuf)->pb_info;
1658 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1659 md.threshold = LNET_MD_THRESH_INF;
1661 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1662 LNET_MD_MANAGE_REMOTE;
1663 md.eq_handle = the_lnet.ln_ping_target_eq;
1664 md.user_ptr = *ppbuf;
1666 rc = LNetMDAttach(me_handle, md, LNET_RETAIN, ping_mdh);
1668 CERROR("Can't attach ping target MD: %d\n", rc);
1669 goto fail_unlink_ping_me;
1671 lnet_ping_buffer_addref(*ppbuf);
1675 fail_unlink_ping_me:
1676 rc2 = LNetMEUnlink(me_handle);
1678 fail_decref_ping_buffer:
1679 LASSERT(lnet_ping_buffer_numref(*ppbuf) == 1);
1680 lnet_ping_buffer_decref(*ppbuf);
1684 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1691 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1692 struct lnet_handle_md *ping_mdh)
1694 sigset_t blocked = cfs_block_allsigs();
1696 LNetMDUnlink(*ping_mdh);
1697 LNetInvalidateMDHandle(ping_mdh);
1699 /* NB the MD could be busy; this just starts the unlink */
1700 while (lnet_ping_buffer_numref(pbuf) > 1) {
1701 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1702 set_current_state(TASK_UNINTERRUPTIBLE);
1703 schedule_timeout(cfs_time_seconds(1));
1706 cfs_restore_sigs(blocked);
1710 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1713 struct lnet_net *net;
1714 struct lnet_ni_status *ns;
1719 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1720 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1721 LASSERT(i < pbuf->pb_nnis);
1723 ns = &pbuf->pb_info.pi_ni[i];
1725 ns->ns_nid = ni->ni_nid;
1728 ns->ns_status = (ni->ni_status != NULL) ?
1729 ni->ni_status->ns_status :
1738 * We (ab)use the ns_status of the loopback interface to
1739 * transmit the sequence number. The first interface listed
1740 * must be the loopback interface.
1742 rc = lnet_ping_info_validate(&pbuf->pb_info);
1744 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1747 LNET_PING_BUFFER_SEQNO(pbuf) =
1748 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1752 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1753 struct lnet_handle_md ping_mdh)
1755 struct lnet_ping_buffer *old_pbuf = NULL;
1756 struct lnet_handle_md old_ping_md;
1758 /* switch the NIs to point to the new ping info created */
1759 lnet_net_lock(LNET_LOCK_EX);
1761 if (!the_lnet.ln_routing)
1762 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1763 if (!lnet_peer_discovery_disabled)
1764 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1766 /* Ensure only known feature bits have been set. */
1767 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1768 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1770 lnet_ping_target_install_locked(pbuf);
1772 if (the_lnet.ln_ping_target) {
1773 old_pbuf = the_lnet.ln_ping_target;
1774 old_ping_md = the_lnet.ln_ping_target_md;
1776 the_lnet.ln_ping_target_md = ping_mdh;
1777 the_lnet.ln_ping_target = pbuf;
1779 lnet_net_unlock(LNET_LOCK_EX);
1782 /* unlink and free the old ping info */
1783 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1784 lnet_ping_buffer_decref(old_pbuf);
1787 lnet_push_update_to_peers(0);
1791 lnet_ping_target_fini(void)
1795 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1796 &the_lnet.ln_ping_target_md);
1798 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1801 lnet_ping_target_destroy();
1804 /* Resize the push target. */
1805 int lnet_push_target_resize(void)
1807 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1808 struct lnet_md md = { NULL };
1809 struct lnet_handle_me meh;
1810 struct lnet_handle_md mdh;
1811 struct lnet_handle_md old_mdh;
1812 struct lnet_ping_buffer *pbuf;
1813 struct lnet_ping_buffer *old_pbuf;
1814 int nnis = the_lnet.ln_push_target_nnis;
1822 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1828 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1829 LNET_PROTO_PING_MATCHBITS, 0,
1830 LNET_UNLINK, LNET_INS_AFTER,
1833 CERROR("Can't create push target ME: %d\n", rc);
1834 goto fail_decref_pbuf;
1837 /* initialize md content */
1838 md.start = &pbuf->pb_info;
1839 md.length = LNET_PING_INFO_SIZE(nnis);
1840 md.threshold = LNET_MD_THRESH_INF;
1842 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1843 LNET_MD_MANAGE_REMOTE;
1845 md.eq_handle = the_lnet.ln_push_target_eq;
1847 rc = LNetMDAttach(meh, md, LNET_RETAIN, &mdh);
1849 CERROR("Can't attach push MD: %d\n", rc);
1850 goto fail_unlink_meh;
1852 lnet_ping_buffer_addref(pbuf);
1854 lnet_net_lock(LNET_LOCK_EX);
1855 old_pbuf = the_lnet.ln_push_target;
1856 old_mdh = the_lnet.ln_push_target_md;
1857 the_lnet.ln_push_target = pbuf;
1858 the_lnet.ln_push_target_md = mdh;
1859 lnet_net_unlock(LNET_LOCK_EX);
1862 LNetMDUnlink(old_mdh);
1863 lnet_ping_buffer_decref(old_pbuf);
1866 if (nnis < the_lnet.ln_push_target_nnis)
1869 CDEBUG(D_NET, "nnis %d success\n", nnis);
1876 lnet_ping_buffer_decref(pbuf);
1878 CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1882 static void lnet_push_target_event_handler(struct lnet_event *ev)
1884 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1886 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1887 lnet_swap_pinginfo(pbuf);
1889 lnet_peer_push_event(ev);
1891 lnet_ping_buffer_decref(pbuf);
1894 /* Initialize the push target. */
1895 static int lnet_push_target_init(void)
1899 if (the_lnet.ln_push_target)
1902 rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1903 &the_lnet.ln_push_target_eq);
1905 CERROR("Can't allocated push target EQ: %d\n", rc);
1909 /* Start at the required minimum, we'll enlarge if required. */
1910 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1912 rc = lnet_push_target_resize();
1915 LNetEQFree(the_lnet.ln_push_target_eq);
1916 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1922 /* Clean up the push target. */
1923 static void lnet_push_target_fini(void)
1925 if (!the_lnet.ln_push_target)
1928 /* Unlink and invalidate to prevent new references. */
1929 LNetMDUnlink(the_lnet.ln_push_target_md);
1930 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1932 /* Wait for the unlink to complete. */
1933 while (lnet_ping_buffer_numref(the_lnet.ln_push_target) > 1) {
1934 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1935 set_current_state(TASK_UNINTERRUPTIBLE);
1936 schedule_timeout(cfs_time_seconds(1));
1939 lnet_ping_buffer_decref(the_lnet.ln_push_target);
1940 the_lnet.ln_push_target = NULL;
1941 the_lnet.ln_push_target_nnis = 0;
1943 LNetEQFree(the_lnet.ln_push_target_eq);
1944 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1948 lnet_ni_tq_credits(struct lnet_ni *ni)
1952 LASSERT(ni->ni_ncpts >= 1);
1954 if (ni->ni_ncpts == 1)
1955 return ni->ni_net->net_tunables.lct_max_tx_credits;
1957 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
1958 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
1959 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
1965 lnet_ni_unlink_locked(struct lnet_ni *ni)
1967 /* move it to zombie list and nobody can find it anymore */
1968 LASSERT(!list_empty(&ni->ni_netlist));
1969 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
1970 lnet_ni_decref_locked(ni, 0);
1974 lnet_clear_zombies_nis_locked(struct lnet_net *net)
1979 struct list_head *zombie_list = &net->net_ni_zombie;
1982 * Now wait for the NIs I just nuked to show up on the zombie
1983 * list and shut them down in guaranteed thread context
1986 while (!list_empty(zombie_list)) {
1990 ni = list_entry(zombie_list->next,
1991 struct lnet_ni, ni_netlist);
1992 list_del_init(&ni->ni_netlist);
1993 /* the ni should be in deleting state. If it's not it's
1995 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
1996 cfs_percpt_for_each(ref, j, ni->ni_refs) {
1999 /* still busy, add it back to zombie list */
2000 list_add(&ni->ni_netlist, zombie_list);
2004 if (!list_empty(&ni->ni_netlist)) {
2005 /* Unlock mutex while waiting to allow other
2006 * threads to read the LNet state and fall through
2009 lnet_net_unlock(LNET_LOCK_EX);
2010 mutex_unlock(&the_lnet.ln_api_mutex);
2013 if ((i & (-i)) == i) {
2015 "Waiting for zombie LNI %s\n",
2016 libcfs_nid2str(ni->ni_nid));
2018 set_current_state(TASK_UNINTERRUPTIBLE);
2019 schedule_timeout(cfs_time_seconds(1));
2021 mutex_lock(&the_lnet.ln_api_mutex);
2022 lnet_net_lock(LNET_LOCK_EX);
2026 lnet_net_unlock(LNET_LOCK_EX);
2028 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2030 LASSERT(!in_interrupt());
2031 (net->net_lnd->lnd_shutdown)(ni);
2034 CDEBUG(D_LNI, "Removed LNI %s\n",
2035 libcfs_nid2str(ni->ni_nid));
2039 lnet_net_lock(LNET_LOCK_EX);
2043 /* shutdown down the NI and release refcount */
2045 lnet_shutdown_lndni(struct lnet_ni *ni)
2048 struct lnet_net *net = ni->ni_net;
2050 lnet_net_lock(LNET_LOCK_EX);
2052 ni->ni_state = LNET_NI_STATE_DELETING;
2054 lnet_ni_unlink_locked(ni);
2055 lnet_incr_dlc_seq();
2056 lnet_net_unlock(LNET_LOCK_EX);
2058 /* clear messages for this NI on the lazy portal */
2059 for (i = 0; i < the_lnet.ln_nportals; i++)
2060 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2062 lnet_net_lock(LNET_LOCK_EX);
2063 lnet_clear_zombies_nis_locked(net);
2064 lnet_net_unlock(LNET_LOCK_EX);
2068 lnet_shutdown_lndnet(struct lnet_net *net)
2072 lnet_net_lock(LNET_LOCK_EX);
2074 net->net_state = LNET_NET_STATE_DELETING;
2076 list_del_init(&net->net_list);
2078 while (!list_empty(&net->net_ni_list)) {
2079 ni = list_entry(net->net_ni_list.next,
2080 struct lnet_ni, ni_netlist);
2081 lnet_net_unlock(LNET_LOCK_EX);
2082 lnet_shutdown_lndni(ni);
2083 lnet_net_lock(LNET_LOCK_EX);
2086 lnet_net_unlock(LNET_LOCK_EX);
2088 /* Do peer table cleanup for this net */
2089 lnet_peer_tables_cleanup(net);
2091 lnet_net_lock(LNET_LOCK_EX);
2093 * decrement ref count on lnd only when the entire network goes
2096 net->net_lnd->lnd_refcount--;
2098 lnet_net_unlock(LNET_LOCK_EX);
2104 lnet_shutdown_lndnets(void)
2106 struct lnet_net *net;
2107 struct list_head resend;
2108 struct lnet_msg *msg, *tmp;
2110 INIT_LIST_HEAD(&resend);
2112 /* NB called holding the global mutex */
2114 /* All quiet on the API front */
2115 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2116 LASSERT(the_lnet.ln_refcount == 0);
2118 lnet_net_lock(LNET_LOCK_EX);
2119 the_lnet.ln_state = LNET_STATE_STOPPING;
2121 while (!list_empty(&the_lnet.ln_nets)) {
2123 * move the nets to the zombie list to avoid them being
2124 * picked up for new work. LONET is also included in the
2125 * Nets that will be moved to the zombie list
2127 net = list_entry(the_lnet.ln_nets.next,
2128 struct lnet_net, net_list);
2129 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2132 /* Drop the cached loopback Net. */
2133 if (the_lnet.ln_loni != NULL) {
2134 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2135 the_lnet.ln_loni = NULL;
2137 lnet_net_unlock(LNET_LOCK_EX);
2139 /* iterate through the net zombie list and delete each net */
2140 while (!list_empty(&the_lnet.ln_net_zombie)) {
2141 net = list_entry(the_lnet.ln_net_zombie.next,
2142 struct lnet_net, net_list);
2143 lnet_shutdown_lndnet(net);
2146 spin_lock(&the_lnet.ln_msg_resend_lock);
2147 list_splice(&the_lnet.ln_msg_resend, &resend);
2148 spin_unlock(&the_lnet.ln_msg_resend_lock);
2150 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2151 list_del_init(&msg->msg_list);
2152 msg->msg_no_resend = true;
2153 lnet_finalize(msg, -ECANCELED);
2156 lnet_net_lock(LNET_LOCK_EX);
2157 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2158 lnet_net_unlock(LNET_LOCK_EX);
2162 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2165 struct lnet_tx_queue *tq;
2167 struct lnet_net *net = ni->ni_net;
2169 mutex_lock(&the_lnet.ln_lnd_mutex);
2172 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2173 ni->ni_lnd_tunables_set = true;
2176 rc = (net->net_lnd->lnd_startup)(ni);
2178 mutex_unlock(&the_lnet.ln_lnd_mutex);
2181 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2182 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2183 lnet_net_lock(LNET_LOCK_EX);
2184 net->net_lnd->lnd_refcount--;
2185 lnet_net_unlock(LNET_LOCK_EX);
2190 ni->ni_state = LNET_NI_STATE_ACTIVE;
2193 /* We keep a reference on the loopback net through the loopback NI */
2194 if (net->net_lnd->lnd_type == LOLND) {
2196 LASSERT(the_lnet.ln_loni == NULL);
2197 the_lnet.ln_loni = ni;
2198 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2199 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2200 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2201 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2205 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2206 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2207 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2208 libcfs_lnd2str(net->net_lnd->lnd_type),
2209 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2211 /* shutdown the NI since if we get here then it must've already
2214 lnet_shutdown_lndni(ni);
2218 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2219 tq->tq_credits_min =
2220 tq->tq_credits_max =
2221 tq->tq_credits = lnet_ni_tq_credits(ni);
2224 atomic_set(&ni->ni_tx_credits,
2225 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2226 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2228 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2229 libcfs_nid2str(ni->ni_nid),
2230 ni->ni_net->net_tunables.lct_peer_tx_credits,
2231 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2232 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2233 ni->ni_net->net_tunables.lct_peer_timeout);
2242 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2245 struct lnet_net *net_l = NULL;
2246 struct list_head local_ni_list;
2250 struct lnet_lnd *lnd;
2252 net->net_tunables.lct_peer_timeout;
2254 net->net_tunables.lct_max_tx_credits;
2255 int peerrtrcredits =
2256 net->net_tunables.lct_peer_rtr_credits;
2258 INIT_LIST_HEAD(&local_ni_list);
2261 * make sure that this net is unique. If it isn't then
2262 * we are adding interfaces to an already existing network, and
2263 * 'net' is just a convenient way to pass in the list.
2264 * if it is unique we need to find the LND and load it if
2267 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2268 lnd_type = LNET_NETTYP(net->net_id);
2270 mutex_lock(&the_lnet.ln_lnd_mutex);
2271 lnd = lnet_find_lnd_by_type(lnd_type);
2274 mutex_unlock(&the_lnet.ln_lnd_mutex);
2275 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2276 mutex_lock(&the_lnet.ln_lnd_mutex);
2278 lnd = lnet_find_lnd_by_type(lnd_type);
2280 mutex_unlock(&the_lnet.ln_lnd_mutex);
2281 CERROR("Can't load LND %s, module %s, rc=%d\n",
2282 libcfs_lnd2str(lnd_type),
2283 libcfs_lnd2modname(lnd_type), rc);
2284 #ifndef HAVE_MODULE_LOADING_SUPPORT
2285 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2286 "compiled with kernel module "
2287 "loading support.");
2294 lnet_net_lock(LNET_LOCK_EX);
2295 lnd->lnd_refcount++;
2296 lnet_net_unlock(LNET_LOCK_EX);
2300 mutex_unlock(&the_lnet.ln_lnd_mutex);
2306 * net_l: if the network being added is unique then net_l
2307 * will point to that network
2308 * if the network being added is not unique then
2309 * net_l points to the existing network.
2311 * When we enter the loop below, we'll pick NIs off he
2312 * network beign added and start them up, then add them to
2313 * a local ni list. Once we've successfully started all
2314 * the NIs then we join the local NI list (of started up
2315 * networks) with the net_l->net_ni_list, which should
2316 * point to the correct network to add the new ni list to
2318 * If any of the new NIs fail to start up, then we want to
2319 * iterate through the local ni list, which should include
2320 * any NIs which were successfully started up, and shut
2323 * After than we want to delete the network being added,
2324 * to avoid a memory leak.
2328 * When a network uses TCP bonding then all its interfaces
2329 * must be specified when the network is first defined: the
2330 * TCP bonding code doesn't allow for interfaces to be added
2333 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2334 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2339 while (!list_empty(&net->net_ni_added)) {
2340 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2342 list_del_init(&ni->ni_netlist);
2344 /* make sure that the the NI we're about to start
2345 * up is actually unique. if it's not fail. */
2346 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2347 ni->ni_interfaces[0])) {
2352 /* adjust the pointer the parent network, just in case it
2353 * the net is a duplicate */
2356 rc = lnet_startup_lndni(ni, tun);
2358 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
2359 ni->ni_net->net_lnd->lnd_query != NULL);
2365 list_add_tail(&ni->ni_netlist, &local_ni_list);
2370 lnet_net_lock(LNET_LOCK_EX);
2371 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2372 lnet_incr_dlc_seq();
2373 lnet_net_unlock(LNET_LOCK_EX);
2375 /* if the network is not unique then we don't want to keep
2376 * it around after we're done. Free it. Otherwise add that
2377 * net to the global the_lnet.ln_nets */
2378 if (net_l != net && net_l != NULL) {
2380 * TODO - note. currently the tunables can not be updated
2385 net->net_state = LNET_NET_STATE_ACTIVE;
2387 * restore tunables after it has been overwitten by the
2390 if (peer_timeout != -1)
2391 net->net_tunables.lct_peer_timeout = peer_timeout;
2392 if (maxtxcredits != -1)
2393 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2394 if (peerrtrcredits != -1)
2395 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2397 lnet_net_lock(LNET_LOCK_EX);
2398 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2399 lnet_net_unlock(LNET_LOCK_EX);
2406 * shutdown the new NIs that are being started up
2407 * free the NET being started
2409 while (!list_empty(&local_ni_list)) {
2410 ni = list_entry(local_ni_list.next, struct lnet_ni,
2413 lnet_shutdown_lndni(ni);
2423 lnet_startup_lndnets(struct list_head *netlist)
2425 struct lnet_net *net;
2430 * Change to running state before bringing up the LNDs. This
2431 * allows lnet_shutdown_lndnets() to assert that we've passed
2434 lnet_net_lock(LNET_LOCK_EX);
2435 the_lnet.ln_state = LNET_STATE_RUNNING;
2436 lnet_net_unlock(LNET_LOCK_EX);
2438 while (!list_empty(netlist)) {
2439 net = list_entry(netlist->next, struct lnet_net, net_list);
2440 list_del_init(&net->net_list);
2442 rc = lnet_startup_lndnet(net, NULL);
2452 lnet_shutdown_lndnets();
2458 * Initialize LNet library.
2460 * Automatically called at module loading time. Caller has to call
2461 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2462 * latter returned 0. It must be called exactly once.
2464 * \retval 0 on success
2465 * \retval -ve on failures.
2467 int lnet_lib_init(void)
2471 lnet_assert_wire_constants();
2473 /* refer to global cfs_cpt_table for now */
2474 the_lnet.ln_cpt_table = cfs_cpt_table;
2475 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
2477 LASSERT(the_lnet.ln_cpt_number > 0);
2478 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2479 /* we are under risk of consuming all lh_cookie */
2480 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2481 "please change setting of CPT-table and retry\n",
2482 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2486 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2487 the_lnet.ln_cpt_bits++;
2489 rc = lnet_create_locks();
2491 CERROR("Can't create LNet global locks: %d\n", rc);
2495 the_lnet.ln_refcount = 0;
2496 LNetInvalidateEQHandle(&the_lnet.ln_rc_eqh);
2497 INIT_LIST_HEAD(&the_lnet.ln_lnds);
2498 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2499 INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
2500 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2501 INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
2503 /* The hash table size is the number of bits it takes to express the set
2504 * ln_num_routes, minus 1 (better to under estimate than over so we
2505 * don't waste memory). */
2506 if (rnet_htable_size <= 0)
2507 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2508 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2509 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2510 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2511 order_base_2(rnet_htable_size) - 1);
2513 /* All LNDs apart from the LOLND are in separate modules. They
2514 * register themselves when their module loads, and unregister
2515 * themselves when their module is unloaded. */
2516 lnet_register_lnd(&the_lolnd);
2521 * Finalize LNet library.
2523 * \pre lnet_lib_init() called with success.
2524 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2526 void lnet_lib_exit(void)
2528 LASSERT(the_lnet.ln_refcount == 0);
2530 while (!list_empty(&the_lnet.ln_lnds))
2531 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
2532 struct lnet_lnd, lnd_list));
2533 lnet_destroy_locks();
2537 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2539 * Users must call this function at least once before any other functions.
2540 * For each successful call there must be a corresponding call to
2541 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2544 * The PID used by LNet may be different from the one requested.
2547 * \param requested_pid PID requested by the caller.
2549 * \return >= 0 on success, and < 0 error code on failures.
2552 LNetNIInit(lnet_pid_t requested_pid)
2554 int im_a_router = 0;
2557 struct lnet_ping_buffer *pbuf;
2558 struct lnet_handle_md ping_mdh;
2559 struct list_head net_head;
2560 struct lnet_net *net;
2562 INIT_LIST_HEAD(&net_head);
2564 mutex_lock(&the_lnet.ln_api_mutex);
2566 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2568 if (the_lnet.ln_refcount > 0) {
2569 rc = the_lnet.ln_refcount++;
2570 mutex_unlock(&the_lnet.ln_api_mutex);
2574 rc = lnet_prepare(requested_pid);
2576 mutex_unlock(&the_lnet.ln_api_mutex);
2580 /* create a network for Loopback network */
2581 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2584 goto err_empty_list;
2587 /* Add in the loopback NI */
2588 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2590 goto err_empty_list;
2593 /* If LNet is being initialized via DLC it is possible
2594 * that the user requests not to load module parameters (ones which
2595 * are supported by DLC) on initialization. Therefore, make sure not
2596 * to load networks, routes and forwarding from module parameters
2597 * in this case. On cleanup in case of failure only clean up
2598 * routes if it has been loaded */
2599 if (!the_lnet.ln_nis_from_mod_params) {
2600 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2603 goto err_empty_list;
2606 ni_count = lnet_startup_lndnets(&net_head);
2609 goto err_empty_list;
2612 if (!the_lnet.ln_nis_from_mod_params) {
2613 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2615 goto err_shutdown_lndnis;
2617 rc = lnet_check_routes();
2619 goto err_destroy_routes;
2621 rc = lnet_rtrpools_alloc(im_a_router);
2623 goto err_destroy_routes;
2626 rc = lnet_acceptor_start();
2628 goto err_destroy_routes;
2630 the_lnet.ln_refcount = 1;
2631 /* Now I may use my own API functions... */
2633 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2635 goto err_acceptor_stop;
2637 lnet_ping_target_update(pbuf, ping_mdh);
2639 rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2641 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2645 rc = lnet_monitor_thr_start();
2649 rc = lnet_push_target_init();
2651 goto err_stop_monitor_thr;
2653 rc = lnet_peer_discovery_start();
2655 goto err_destroy_push_target;
2658 lnet_router_debugfs_init();
2660 mutex_unlock(&the_lnet.ln_api_mutex);
2664 err_destroy_push_target:
2665 lnet_push_target_fini();
2666 err_stop_monitor_thr:
2667 lnet_monitor_thr_stop();
2669 lnet_ping_target_fini();
2671 the_lnet.ln_refcount = 0;
2672 lnet_acceptor_stop();
2674 if (!the_lnet.ln_nis_from_mod_params)
2675 lnet_destroy_routes();
2676 err_shutdown_lndnis:
2677 lnet_shutdown_lndnets();
2681 mutex_unlock(&the_lnet.ln_api_mutex);
2682 while (!list_empty(&net_head)) {
2683 struct lnet_net *net;
2685 net = list_entry(net_head.next, struct lnet_net, net_list);
2686 list_del_init(&net->net_list);
2691 EXPORT_SYMBOL(LNetNIInit);
2694 * Stop LNet interfaces, routing, and forwarding.
2696 * Users must call this function once for each successful call to LNetNIInit().
2697 * Once the LNetNIFini() operation has been started, the results of pending
2698 * API operations are undefined.
2700 * \return always 0 for current implementation.
2705 mutex_lock(&the_lnet.ln_api_mutex);
2707 LASSERT(the_lnet.ln_refcount > 0);
2709 if (the_lnet.ln_refcount != 1) {
2710 the_lnet.ln_refcount--;
2712 LASSERT(!the_lnet.ln_niinit_self);
2716 lnet_router_debugfs_fini();
2717 lnet_peer_discovery_stop();
2718 lnet_push_target_fini();
2719 lnet_monitor_thr_stop();
2720 lnet_ping_target_fini();
2722 /* Teardown fns that use my own API functions BEFORE here */
2723 the_lnet.ln_refcount = 0;
2725 lnet_acceptor_stop();
2726 lnet_destroy_routes();
2727 lnet_shutdown_lndnets();
2731 mutex_unlock(&the_lnet.ln_api_mutex);
2734 EXPORT_SYMBOL(LNetNIFini);
2737 * Grabs the ni data from the ni structure and fills the out
2740 * \param[in] ni network interface structure
2741 * \param[out] cfg_ni NI config information
2742 * \param[out] tun network and LND tunables
2745 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2746 struct lnet_ioctl_config_lnd_tunables *tun,
2747 struct lnet_ioctl_element_stats *stats,
2750 size_t min_size = 0;
2753 if (!ni || !cfg_ni || !tun)
2756 if (ni->ni_interfaces[0] != NULL) {
2757 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2758 if (ni->ni_interfaces[i] != NULL) {
2759 strncpy(cfg_ni->lic_ni_intf[i],
2760 ni->ni_interfaces[i],
2761 sizeof(cfg_ni->lic_ni_intf[i]));
2766 cfg_ni->lic_nid = ni->ni_nid;
2767 if (ni->ni_nid == LNET_NID_LO_0)
2768 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2770 cfg_ni->lic_status = ni->ni_status->ns_status;
2771 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2772 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2774 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2777 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2778 LNET_STATS_TYPE_SEND);
2779 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2780 LNET_STATS_TYPE_RECV);
2781 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2782 LNET_STATS_TYPE_DROP);
2786 * tun->lt_tun will always be present, but in order to be
2787 * backwards compatible, we need to deal with the cases when
2788 * tun->lt_tun is smaller than what the kernel has, because it
2789 * comes from an older version of a userspace program, then we'll
2790 * need to copy as much information as we have available space.
2792 min_size = tun_size - sizeof(tun->lt_cmn);
2793 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2795 /* copy over the cpts */
2796 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2797 ni->ni_cpts == NULL) {
2798 for (i = 0; i < ni->ni_ncpts; i++)
2799 cfg_ni->lic_cpts[i] = i;
2802 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2803 i < LNET_MAX_SHOW_NUM_CPT;
2805 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2807 cfg_ni->lic_ncpts = ni->ni_ncpts;
2811 * NOTE: This is a legacy function left in the code to be backwards
2812 * compatible with older userspace programs. It should eventually be
2815 * Grabs the ni data from the ni structure and fills the out
2818 * \param[in] ni network interface structure
2819 * \param[out] config config information
2822 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2823 struct lnet_ioctl_config_data *config)
2825 struct lnet_ioctl_net_config *net_config;
2826 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2827 size_t min_size, tunable_size = 0;
2833 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2837 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2838 ARRAY_SIZE(net_config->ni_interfaces));
2840 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2841 if (!ni->ni_interfaces[i])
2844 strncpy(net_config->ni_interfaces[i],
2845 ni->ni_interfaces[i],
2846 sizeof(net_config->ni_interfaces[i]));
2849 config->cfg_nid = ni->ni_nid;
2850 config->cfg_config_u.cfg_net.net_peer_timeout =
2851 ni->ni_net->net_tunables.lct_peer_timeout;
2852 config->cfg_config_u.cfg_net.net_max_tx_credits =
2853 ni->ni_net->net_tunables.lct_max_tx_credits;
2854 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2855 ni->ni_net->net_tunables.lct_peer_tx_credits;
2856 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2857 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2859 if (ni->ni_nid == LNET_NID_LO_0)
2860 net_config->ni_status = LNET_NI_STATUS_UP;
2862 net_config->ni_status = ni->ni_status->ns_status;
2865 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2867 for (i = 0; i < num_cpts; i++)
2868 net_config->ni_cpts[i] = ni->ni_cpts[i];
2870 config->cfg_ncpts = num_cpts;
2874 * See if user land tools sent in a newer and larger version
2875 * of struct lnet_tunables than what the kernel uses.
2877 min_size = sizeof(*config) + sizeof(*net_config);
2879 if (config->cfg_hdr.ioc_len > min_size)
2880 tunable_size = config->cfg_hdr.ioc_len - min_size;
2882 /* Don't copy too much data to user space */
2883 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2884 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2886 if (lnd_cfg && min_size) {
2887 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2888 config->cfg_config_u.cfg_net.net_interface_count = 1;
2890 /* Tell user land that kernel side has less data */
2891 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2892 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2893 config->cfg_hdr.ioc_len -= min_size;
2899 lnet_get_ni_idx_locked(int idx)
2902 struct lnet_net *net;
2904 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2905 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2915 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2918 struct lnet_net *net = mynet;
2921 * It is possible that the net has been cleaned out while there is
2922 * a message being sent. This function accessed the net without
2923 * checking if the list is empty
2927 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2929 if (list_empty(&net->net_ni_list))
2931 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2937 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2938 /* if you reached the end of the ni list and the net is
2939 * specified, then there are no more nis in that net */
2943 /* we reached the end of this net ni list. move to the
2945 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2946 /* no more nets and no more NIs. */
2949 /* get the next net */
2950 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2952 if (list_empty(&net->net_ni_list))
2954 /* get the ni on it */
2955 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2961 if (list_empty(&prev->ni_netlist))
2964 /* there are more nis left */
2965 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2971 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2976 int idx = config->cfg_count;
2978 cpt = lnet_net_lock_current();
2980 ni = lnet_get_ni_idx_locked(idx);
2985 lnet_fill_ni_info_legacy(ni, config);
2989 lnet_net_unlock(cpt);
2994 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
2995 struct lnet_ioctl_config_lnd_tunables *tun,
2996 struct lnet_ioctl_element_stats *stats,
3003 if (!cfg_ni || !tun || !stats)
3006 cpt = lnet_net_lock_current();
3008 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3013 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3017 lnet_net_unlock(cpt);
3021 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3030 cpt = lnet_net_lock_current();
3032 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3035 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3039 lnet_net_unlock(cpt);
3044 static int lnet_add_net_common(struct lnet_net *net,
3045 struct lnet_ioctl_config_lnd_tunables *tun)
3048 struct lnet_ping_buffer *pbuf;
3049 struct lnet_handle_md ping_mdh;
3051 struct lnet_remotenet *rnet;
3053 int num_acceptor_nets;
3055 lnet_net_lock(LNET_LOCK_EX);
3056 rnet = lnet_find_rnet_locked(net->net_id);
3057 lnet_net_unlock(LNET_LOCK_EX);
3059 * make sure that the net added doesn't invalidate the current
3060 * configuration LNet is keeping
3063 CERROR("Adding net %s will invalidate routing configuration\n",
3064 libcfs_net2str(net->net_id));
3070 * make sure you calculate the correct number of slots in the ping
3071 * buffer. Since the ping info is a flattened list of all the NIs,
3072 * we should allocate enough slots to accomodate the number of NIs
3073 * which will be added.
3075 * since ni hasn't been configured yet, use
3076 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3078 net_ni_count = lnet_get_net_ni_count_pre(net);
3080 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3081 net_ni_count + lnet_get_ni_count(),
3089 memcpy(&net->net_tunables,
3090 &tun->lt_cmn, sizeof(net->net_tunables));
3092 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3095 * before starting this network get a count of the current TCP
3096 * networks which require the acceptor thread running. If that
3097 * count is == 0 before we start up this network, then we'd want to
3098 * start up the acceptor thread after starting up this network
3100 num_acceptor_nets = lnet_count_acceptor_nets();
3102 net_id = net->net_id;
3104 rc = lnet_startup_lndnet(net,
3105 (tun) ? &tun->lt_tun : NULL);
3109 lnet_net_lock(LNET_LOCK_EX);
3110 net = lnet_get_net_locked(net_id);
3111 lnet_net_unlock(LNET_LOCK_EX);
3116 * Start the acceptor thread if this is the first network
3117 * being added that requires the thread.
3119 if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3120 rc = lnet_acceptor_start();
3122 /* shutdown the net that we just started */
3123 CERROR("Failed to start up acceptor thread\n");
3124 lnet_shutdown_lndnet(net);
3129 lnet_net_lock(LNET_LOCK_EX);
3130 lnet_peer_net_added(net);
3131 lnet_net_unlock(LNET_LOCK_EX);
3133 lnet_ping_target_update(pbuf, ping_mdh);
3138 lnet_ping_md_unlink(pbuf, &ping_mdh);
3139 lnet_ping_buffer_decref(pbuf);
3143 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3144 struct lnet_ioctl_config_lnd_tunables *tun)
3146 struct lnet_net *net;
3149 struct list_head net_head;
3151 INIT_LIST_HEAD(&net_head);
3153 rc = lnet_parse_ip2nets(&nets, ip2nets);
3157 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3161 mutex_lock(&the_lnet.ln_api_mutex);
3162 while (!list_empty(&net_head)) {
3163 net = list_entry(net_head.next, struct lnet_net, net_list);
3164 list_del_init(&net->net_list);
3165 rc = lnet_add_net_common(net, tun);
3171 mutex_unlock(&the_lnet.ln_api_mutex);
3173 while (!list_empty(&net_head)) {
3174 net = list_entry(net_head.next, struct lnet_net, net_list);
3175 list_del_init(&net->net_list);
3181 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3183 struct lnet_net *net;
3185 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3187 __u32 net_id, lnd_type;
3189 /* get the tunables if they are available */
3190 if (conf->lic_cfg_hdr.ioc_len >=
3191 sizeof(*conf) + sizeof(*tun))
3192 tun = (struct lnet_ioctl_config_lnd_tunables *)
3195 /* handle legacy ip2nets from DLC */
3196 if (conf->lic_legacy_ip2nets[0] != '\0')
3197 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3200 net_id = LNET_NIDNET(conf->lic_nid);
3201 lnd_type = LNET_NETTYP(net_id);
3203 if (!libcfs_isknown_lnd(lnd_type)) {
3204 CERROR("No valid net and lnd information provided\n");
3208 net = lnet_net_alloc(net_id, NULL);
3212 for (i = 0; i < conf->lic_ncpts; i++) {
3213 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3217 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3218 conf->lic_ni_intf[0]);
3222 mutex_lock(&the_lnet.ln_api_mutex);
3224 rc = lnet_add_net_common(net, tun);
3226 mutex_unlock(&the_lnet.ln_api_mutex);
3231 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3233 struct lnet_net *net;
3235 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3236 struct lnet_ping_buffer *pbuf;
3237 struct lnet_handle_md ping_mdh;
3242 /* don't allow userspace to shutdown the LOLND */
3243 if (LNET_NETTYP(net_id) == LOLND)
3246 mutex_lock(&the_lnet.ln_api_mutex);
3250 net = lnet_get_net_locked(net_id);
3252 CERROR("net %s not found\n",
3253 libcfs_net2str(net_id));
3258 addr = LNET_NIDADDR(conf->lic_nid);
3260 /* remove the entire net */
3261 net_count = lnet_get_net_ni_count_locked(net);
3265 /* create and link a new ping info, before removing the old one */
3266 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3267 lnet_get_ni_count() - net_count,
3270 goto unlock_api_mutex;
3272 lnet_shutdown_lndnet(net);
3274 if (lnet_count_acceptor_nets() == 0)
3275 lnet_acceptor_stop();
3277 lnet_ping_target_update(pbuf, ping_mdh);
3279 goto unlock_api_mutex;
3282 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3284 CERROR("nid %s not found\n",
3285 libcfs_nid2str(conf->lic_nid));
3290 net_count = lnet_get_net_ni_count_locked(net);
3294 /* create and link a new ping info, before removing the old one */
3295 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3296 lnet_get_ni_count() - 1, false);
3298 goto unlock_api_mutex;
3300 lnet_shutdown_lndni(ni);
3302 if (lnet_count_acceptor_nets() == 0)
3303 lnet_acceptor_stop();
3305 lnet_ping_target_update(pbuf, ping_mdh);
3307 /* check if the net is empty and remove it if it is */
3309 lnet_shutdown_lndnet(net);
3311 goto unlock_api_mutex;
3316 mutex_unlock(&the_lnet.ln_api_mutex);
3322 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3323 * They are only expected to be called for unique networks.
3324 * That can be as a result of older DLC library
3325 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3328 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3330 struct lnet_net *net;
3331 struct list_head net_head;
3333 struct lnet_ioctl_config_lnd_tunables tun;
3334 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3336 INIT_LIST_HEAD(&net_head);
3338 /* Create a net/ni structures for the network string */
3339 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3341 return rc == 0 ? -EINVAL : rc;
3343 mutex_lock(&the_lnet.ln_api_mutex);
3346 rc = -EINVAL; /* only add one network per call */
3347 goto out_unlock_clean;
3350 net = list_entry(net_head.next, struct lnet_net, net_list);
3351 list_del_init(&net->net_list);
3353 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3355 memset(&tun, 0, sizeof(tun));
3357 tun.lt_cmn.lct_peer_timeout =
3358 conf->cfg_config_u.cfg_net.net_peer_timeout;
3359 tun.lt_cmn.lct_peer_tx_credits =
3360 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3361 tun.lt_cmn.lct_peer_rtr_credits =
3362 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3363 tun.lt_cmn.lct_max_tx_credits =
3364 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3366 rc = lnet_add_net_common(net, &tun);
3369 mutex_unlock(&the_lnet.ln_api_mutex);
3370 while (!list_empty(&net_head)) {
3371 /* net_head list is empty in success case */
3372 net = list_entry(net_head.next, struct lnet_net, net_list);
3373 list_del_init(&net->net_list);
3380 lnet_dyn_del_net(__u32 net_id)
3382 struct lnet_net *net;
3383 struct lnet_ping_buffer *pbuf;
3384 struct lnet_handle_md ping_mdh;
3388 /* don't allow userspace to shutdown the LOLND */
3389 if (LNET_NETTYP(net_id) == LOLND)
3392 mutex_lock(&the_lnet.ln_api_mutex);
3396 net = lnet_get_net_locked(net_id);
3403 net_ni_count = lnet_get_net_ni_count_locked(net);
3407 /* create and link a new ping info, before removing the old one */
3408 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3409 lnet_get_ni_count() - net_ni_count, false);
3413 lnet_shutdown_lndnet(net);
3415 if (lnet_count_acceptor_nets() == 0)
3416 lnet_acceptor_stop();
3418 lnet_ping_target_update(pbuf, ping_mdh);
3421 mutex_unlock(&the_lnet.ln_api_mutex);
3426 void lnet_incr_dlc_seq(void)
3428 atomic_inc(&lnet_dlc_seq_no);
3431 __u32 lnet_get_dlc_seq_locked(void)
3433 return atomic_read(&lnet_dlc_seq_no);
3437 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3439 struct lnet_net *net;
3442 lnet_net_lock(LNET_LOCK_EX);
3443 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3444 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3445 if (ni->ni_nid == nid || all) {
3446 atomic_set(&ni->ni_healthv, value);
3447 if (list_empty(&ni->ni_recovery) &&
3448 value < LNET_MAX_HEALTH_VALUE) {
3449 CERROR("manually adding local NI %s to recovery\n",
3450 libcfs_nid2str(ni->ni_nid));
3451 list_add_tail(&ni->ni_recovery,
3452 &the_lnet.ln_mt_localNIRecovq);
3453 lnet_ni_addref_locked(ni, 0);
3456 lnet_net_unlock(LNET_LOCK_EX);
3462 lnet_net_unlock(LNET_LOCK_EX);
3466 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3470 lnet_nid_t nid = stats->hlni_nid;
3472 cpt = lnet_net_lock_current();
3473 ni = lnet_nid2ni_locked(nid, cpt);
3480 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3481 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3482 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3483 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3484 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3485 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3486 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3489 lnet_net_unlock(cpt);
3495 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3500 lnet_net_lock(LNET_LOCK_EX);
3501 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3502 list->rlst_nid_array[i] = ni->ni_nid;
3504 if (i >= LNET_MAX_SHOW_NUM_NID)
3507 lnet_net_unlock(LNET_LOCK_EX);
3508 list->rlst_num_nids = i;
3514 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3516 struct lnet_peer_ni *lpni;
3519 lnet_net_lock(LNET_LOCK_EX);
3520 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3521 list->rlst_nid_array[i] = lpni->lpni_nid;
3523 if (i >= LNET_MAX_SHOW_NUM_NID)
3526 lnet_net_unlock(LNET_LOCK_EX);
3527 list->rlst_num_nids = i;
3533 * LNet ioctl handler.
3537 LNetCtl(unsigned int cmd, void *arg)
3539 struct libcfs_ioctl_data *data = arg;
3540 struct lnet_ioctl_config_data *config;
3541 struct lnet_process_id id = {0};
3545 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3546 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3549 case IOC_LIBCFS_GET_NI:
3550 rc = LNetGetId(data->ioc_count, &id);
3551 data->ioc_nid = id.nid;
3554 case IOC_LIBCFS_FAIL_NID:
3555 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3557 case IOC_LIBCFS_ADD_ROUTE:
3560 if (config->cfg_hdr.ioc_len < sizeof(*config))
3563 mutex_lock(&the_lnet.ln_api_mutex);
3564 rc = lnet_add_route(config->cfg_net,
3565 config->cfg_config_u.cfg_route.rtr_hop,
3567 config->cfg_config_u.cfg_route.
3570 rc = lnet_check_routes();
3572 lnet_del_route(config->cfg_net,
3575 mutex_unlock(&the_lnet.ln_api_mutex);
3578 case IOC_LIBCFS_DEL_ROUTE:
3581 if (config->cfg_hdr.ioc_len < sizeof(*config))
3584 mutex_lock(&the_lnet.ln_api_mutex);
3585 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3586 mutex_unlock(&the_lnet.ln_api_mutex);
3589 case IOC_LIBCFS_GET_ROUTE:
3592 if (config->cfg_hdr.ioc_len < sizeof(*config))
3595 mutex_lock(&the_lnet.ln_api_mutex);
3596 rc = lnet_get_route(config->cfg_count,
3598 &config->cfg_config_u.cfg_route.rtr_hop,
3600 &config->cfg_config_u.cfg_route.rtr_flags,
3601 &config->cfg_config_u.cfg_route.
3603 mutex_unlock(&the_lnet.ln_api_mutex);
3606 case IOC_LIBCFS_GET_LOCAL_NI: {
3607 struct lnet_ioctl_config_ni *cfg_ni;
3608 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3609 struct lnet_ioctl_element_stats *stats;
3614 /* get the tunables if they are available */
3615 if (cfg_ni->lic_cfg_hdr.ioc_len <
3616 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3619 stats = (struct lnet_ioctl_element_stats *)
3621 tun = (struct lnet_ioctl_config_lnd_tunables *)
3622 (cfg_ni->lic_bulk + sizeof(*stats));
3624 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3627 mutex_lock(&the_lnet.ln_api_mutex);
3628 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3629 mutex_unlock(&the_lnet.ln_api_mutex);
3633 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3634 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3636 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3639 mutex_lock(&the_lnet.ln_api_mutex);
3640 rc = lnet_get_ni_stats(msg_stats);
3641 mutex_unlock(&the_lnet.ln_api_mutex);
3646 case IOC_LIBCFS_GET_NET: {
3647 size_t total = sizeof(*config) +
3648 sizeof(struct lnet_ioctl_net_config);
3651 if (config->cfg_hdr.ioc_len < total)
3654 mutex_lock(&the_lnet.ln_api_mutex);
3655 rc = lnet_get_net_config(config);
3656 mutex_unlock(&the_lnet.ln_api_mutex);
3660 case IOC_LIBCFS_GET_LNET_STATS:
3662 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3664 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3667 mutex_lock(&the_lnet.ln_api_mutex);
3668 lnet_counters_get(&lnet_stats->st_cntrs);
3669 mutex_unlock(&the_lnet.ln_api_mutex);
3673 case IOC_LIBCFS_CONFIG_RTR:
3676 if (config->cfg_hdr.ioc_len < sizeof(*config))
3679 mutex_lock(&the_lnet.ln_api_mutex);
3680 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3681 rc = lnet_rtrpools_enable();
3682 mutex_unlock(&the_lnet.ln_api_mutex);
3685 lnet_rtrpools_disable();
3686 mutex_unlock(&the_lnet.ln_api_mutex);
3689 case IOC_LIBCFS_ADD_BUF:
3692 if (config->cfg_hdr.ioc_len < sizeof(*config))
3695 mutex_lock(&the_lnet.ln_api_mutex);
3696 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3698 config->cfg_config_u.cfg_buffers.
3700 config->cfg_config_u.cfg_buffers.
3702 mutex_unlock(&the_lnet.ln_api_mutex);
3705 case IOC_LIBCFS_SET_NUMA_RANGE: {
3706 struct lnet_ioctl_set_value *numa;
3708 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3710 lnet_net_lock(LNET_LOCK_EX);
3711 lnet_numa_range = numa->sv_value;
3712 lnet_net_unlock(LNET_LOCK_EX);
3716 case IOC_LIBCFS_GET_NUMA_RANGE: {
3717 struct lnet_ioctl_set_value *numa;
3719 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3721 numa->sv_value = lnet_numa_range;
3725 case IOC_LIBCFS_GET_BUF: {
3726 struct lnet_ioctl_pool_cfg *pool_cfg;
3727 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3731 if (config->cfg_hdr.ioc_len < total)
3734 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3736 mutex_lock(&the_lnet.ln_api_mutex);
3737 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3738 mutex_unlock(&the_lnet.ln_api_mutex);
3742 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3743 struct lnet_ioctl_local_ni_hstats *stats = arg;
3745 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3748 mutex_lock(&the_lnet.ln_api_mutex);
3749 rc = lnet_get_local_ni_hstats(stats);
3750 mutex_unlock(&the_lnet.ln_api_mutex);
3755 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3756 struct lnet_ioctl_recovery_list *list = arg;
3757 if (list->rlst_hdr.ioc_len < sizeof(*list))
3760 mutex_lock(&the_lnet.ln_api_mutex);
3761 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3762 rc = lnet_get_local_ni_recovery_list(list);
3764 rc = lnet_get_peer_ni_recovery_list(list);
3765 mutex_unlock(&the_lnet.ln_api_mutex);
3769 case IOC_LIBCFS_ADD_PEER_NI: {
3770 struct lnet_ioctl_peer_cfg *cfg = arg;
3772 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3775 mutex_lock(&the_lnet.ln_api_mutex);
3776 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3779 mutex_unlock(&the_lnet.ln_api_mutex);
3783 case IOC_LIBCFS_DEL_PEER_NI: {
3784 struct lnet_ioctl_peer_cfg *cfg = arg;
3786 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3789 mutex_lock(&the_lnet.ln_api_mutex);
3790 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3791 cfg->prcfg_cfg_nid);
3792 mutex_unlock(&the_lnet.ln_api_mutex);
3796 case IOC_LIBCFS_GET_PEER_INFO: {
3797 struct lnet_ioctl_peer *peer_info = arg;
3799 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3802 mutex_lock(&the_lnet.ln_api_mutex);
3803 rc = lnet_get_peer_ni_info(
3804 peer_info->pr_count,
3806 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3807 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3808 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3809 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3810 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3811 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3812 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3813 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3814 mutex_unlock(&the_lnet.ln_api_mutex);
3818 case IOC_LIBCFS_GET_PEER_NI: {
3819 struct lnet_ioctl_peer_cfg *cfg = arg;
3821 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3824 mutex_lock(&the_lnet.ln_api_mutex);
3825 rc = lnet_get_peer_info(cfg,
3826 (void __user *)cfg->prcfg_bulk);
3827 mutex_unlock(&the_lnet.ln_api_mutex);
3831 case IOC_LIBCFS_GET_PEER_LIST: {
3832 struct lnet_ioctl_peer_cfg *cfg = arg;
3834 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3837 mutex_lock(&the_lnet.ln_api_mutex);
3838 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3839 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3840 mutex_unlock(&the_lnet.ln_api_mutex);
3844 case IOC_LIBCFS_SET_HEALHV: {
3845 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3847 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3849 if (cfg->rh_value < 0 ||
3850 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3851 value = LNET_MAX_HEALTH_VALUE;
3853 value = cfg->rh_value;
3854 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3855 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3856 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3857 mutex_lock(&the_lnet.ln_api_mutex);
3858 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3859 lnet_ni_set_healthv(cfg->rh_nid, value,
3862 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3864 mutex_unlock(&the_lnet.ln_api_mutex);
3868 case IOC_LIBCFS_NOTIFY_ROUTER: {
3869 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3871 /* The deadline passed in by the user should be some time in
3872 * seconds in the future since the UNIX epoch. We have to map
3873 * that deadline to the wall clock.
3875 deadline += ktime_get_seconds();
3876 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
3880 case IOC_LIBCFS_LNET_DIST:
3881 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3882 if (rc < 0 && rc != -EHOSTUNREACH)
3885 data->ioc_u32[0] = rc;
3888 case IOC_LIBCFS_TESTPROTOCOMPAT:
3889 lnet_net_lock(LNET_LOCK_EX);
3890 the_lnet.ln_testprotocompat = data->ioc_flags;
3891 lnet_net_unlock(LNET_LOCK_EX);
3894 case IOC_LIBCFS_LNET_FAULT:
3895 return lnet_fault_ctl(data->ioc_flags, data);
3897 case IOC_LIBCFS_PING: {
3898 signed long timeout;
3900 id.nid = data->ioc_nid;
3901 id.pid = data->ioc_u32[0];
3903 /* If timeout is negative then set default of 3 minutes */
3904 if (((s32)data->ioc_u32[1] <= 0) ||
3905 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3906 timeout = msecs_to_jiffies(DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC);
3908 timeout = msecs_to_jiffies(data->ioc_u32[1]);
3910 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3911 data->ioc_plen1 / sizeof(struct lnet_process_id));
3916 data->ioc_count = rc;
3920 case IOC_LIBCFS_PING_PEER: {
3921 struct lnet_ioctl_ping_data *ping = arg;
3922 struct lnet_peer *lp;
3923 signed long timeout;
3925 /* If timeout is negative then set default of 3 minutes */
3926 if (((s32)ping->op_param) <= 0 ||
3927 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3928 timeout = msecs_to_jiffies(DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC);
3930 timeout = msecs_to_jiffies(ping->op_param);
3932 rc = lnet_ping(ping->ping_id, timeout,
3938 mutex_lock(&the_lnet.ln_api_mutex);
3939 lp = lnet_find_peer(ping->ping_id.nid);
3941 ping->ping_id.nid = lp->lp_primary_nid;
3942 ping->mr_info = lnet_peer_is_multi_rail(lp);
3943 lnet_peer_decref_locked(lp);
3945 mutex_unlock(&the_lnet.ln_api_mutex);
3947 ping->ping_count = rc;
3951 case IOC_LIBCFS_DISCOVER: {
3952 struct lnet_ioctl_ping_data *discover = arg;
3953 struct lnet_peer *lp;
3955 rc = lnet_discover(discover->ping_id, discover->op_param,
3957 discover->ping_count);
3961 mutex_lock(&the_lnet.ln_api_mutex);
3962 lp = lnet_find_peer(discover->ping_id.nid);
3964 discover->ping_id.nid = lp->lp_primary_nid;
3965 discover->mr_info = lnet_peer_is_multi_rail(lp);
3966 lnet_peer_decref_locked(lp);
3968 mutex_unlock(&the_lnet.ln_api_mutex);
3970 discover->ping_count = rc;
3975 ni = lnet_net2ni_addref(data->ioc_net);
3979 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
3982 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
3989 EXPORT_SYMBOL(LNetCtl);
3991 void LNetDebugPeer(struct lnet_process_id id)
3993 lnet_debug_peer(id.nid);
3995 EXPORT_SYMBOL(LNetDebugPeer);
3998 * Determine if the specified peer \a nid is on the local node.
4000 * \param nid peer nid to check
4002 * \retval true If peer NID is on the local node.
4003 * \retval false If peer NID is not on the local node.
4005 bool LNetIsPeerLocal(lnet_nid_t nid)
4007 struct lnet_net *net;
4011 cpt = lnet_net_lock_current();
4012 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4013 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4014 if (ni->ni_nid == nid) {
4015 lnet_net_unlock(cpt);
4020 lnet_net_unlock(cpt);
4024 EXPORT_SYMBOL(LNetIsPeerLocal);
4027 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4028 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4030 * \param index Index of the interface to look up.
4031 * \param id On successful return, this location will hold the
4032 * struct lnet_process_id ID of the interface.
4034 * \retval 0 If an interface exists at \a index.
4035 * \retval -ENOENT If no interface has been found.
4038 LNetGetId(unsigned int index, struct lnet_process_id *id)
4041 struct lnet_net *net;
4045 LASSERT(the_lnet.ln_refcount > 0);
4047 cpt = lnet_net_lock_current();
4049 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4050 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4054 id->nid = ni->ni_nid;
4055 id->pid = the_lnet.ln_pid;
4061 lnet_net_unlock(cpt);
4064 EXPORT_SYMBOL(LNetGetId);
4066 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4067 struct lnet_process_id __user *ids, int n_ids)
4069 struct lnet_handle_eq eqh;
4070 struct lnet_handle_md mdh;
4071 struct lnet_event event;
4072 struct lnet_md md = { NULL };
4076 const signed long a_long_time = msecs_to_jiffies(60 * MSEC_PER_SEC);
4077 struct lnet_ping_buffer *pbuf;
4078 struct lnet_process_id tmpid;
4085 /* n_ids limit is arbitrary */
4086 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4090 * if the user buffer has more space than the lnet_interfaces_max
4091 * then only fill it up to lnet_interfaces_max
4093 if (n_ids > lnet_interfaces_max)
4094 n_ids = lnet_interfaces_max;
4096 if (id.pid == LNET_PID_ANY)
4097 id.pid = LNET_PID_LUSTRE;
4099 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4103 /* NB 2 events max (including any unlink event) */
4104 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
4106 CERROR("Can't allocate EQ: %d\n", rc);
4107 goto fail_ping_buffer_decref;
4110 /* initialize md content */
4111 md.start = &pbuf->pb_info;
4112 md.length = LNET_PING_INFO_SIZE(n_ids);
4113 md.threshold = 2; /* GET/REPLY */
4115 md.options = LNET_MD_TRUNCATE;
4119 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4121 CERROR("Can't bind MD: %d\n", rc);
4125 rc = LNetGet(LNET_NID_ANY, mdh, id,
4126 LNET_RESERVED_PORTAL,
4127 LNET_PROTO_PING_MATCHBITS, 0, false);
4130 /* Don't CERROR; this could be deliberate! */
4131 rc2 = LNetMDUnlink(mdh);
4134 /* NB must wait for the UNLINK event below... */
4136 timeout = a_long_time;
4140 /* MUST block for unlink to complete */
4142 blocked = cfs_block_allsigs();
4144 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
4147 cfs_restore_sigs(blocked);
4149 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4150 (rc2 <= 0) ? -1 : event.type,
4151 (rc2 <= 0) ? -1 : event.status,
4152 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4154 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
4156 if (rc2 <= 0 || event.status != 0) {
4157 /* timeout or error */
4158 if (!replied && rc == 0)
4159 rc = (rc2 < 0) ? rc2 :
4160 (rc2 == 0) ? -ETIMEDOUT :
4164 /* Ensure completion in finite time... */
4166 /* No assertion (racing with network) */
4168 timeout = a_long_time;
4169 } else if (rc2 == 0) {
4170 /* timed out waiting for unlink */
4171 CWARN("ping %s: late network completion\n",
4174 } else if (event.type == LNET_EVENT_REPLY) {
4178 } while (rc2 <= 0 || !event.unlinked);
4182 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4189 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4191 rc = -EPROTO; /* if I can't parse... */
4194 CERROR("%s: ping info too short %d\n",
4195 libcfs_id2str(id), nob);
4199 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4200 lnet_swap_pinginfo(pbuf);
4201 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4202 CERROR("%s: Unexpected magic %08x\n",
4203 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4207 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4208 CERROR("%s: ping w/o NI status: 0x%x\n",
4209 libcfs_id2str(id), pbuf->pb_info.pi_features);
4213 if (nob < LNET_PING_INFO_SIZE(0)) {
4214 CERROR("%s: Short reply %d(%d min)\n",
4216 nob, (int)LNET_PING_INFO_SIZE(0));
4220 if (pbuf->pb_info.pi_nnis < n_ids)
4221 n_ids = pbuf->pb_info.pi_nnis;
4223 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4224 CERROR("%s: Short reply %d(%d expected)\n",
4226 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4230 rc = -EFAULT; /* if I segv in copy_to_user()... */
4232 memset(&tmpid, 0, sizeof(tmpid));
4233 for (i = 0; i < n_ids; i++) {
4234 tmpid.pid = pbuf->pb_info.pi_pid;
4235 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4236 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4239 rc = pbuf->pb_info.pi_nnis;
4242 rc2 = LNetEQFree(eqh);
4244 CERROR("rc2 %d\n", rc2);
4247 fail_ping_buffer_decref:
4248 lnet_ping_buffer_decref(pbuf);
4253 lnet_discover(struct lnet_process_id id, __u32 force,
4254 struct lnet_process_id __user *ids, int n_ids)
4256 struct lnet_peer_ni *lpni;
4257 struct lnet_peer_ni *p;
4258 struct lnet_peer *lp;
4259 struct lnet_process_id *buf;
4263 int max_intf = lnet_interfaces_max;
4267 id.nid == LNET_NID_ANY)
4270 if (id.pid == LNET_PID_ANY)
4271 id.pid = LNET_PID_LUSTRE;
4274 * if the user buffer has more space than the max_intf
4275 * then only fill it up to max_intf
4277 if (n_ids > max_intf)
4280 buf_size = n_ids * sizeof(*buf);
4282 LIBCFS_ALLOC(buf, buf_size);
4286 cpt = lnet_net_lock_current();
4287 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4294 * Clearing the NIDS_UPTODATE flag ensures the peer will
4295 * be discovered, provided discovery has not been disabled.
4297 lp = lpni->lpni_peer_net->lpn_peer;
4298 spin_lock(&lp->lp_lock);
4299 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4300 /* If the force flag is set, force a PING and PUSH as well. */
4302 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4303 spin_unlock(&lp->lp_lock);
4304 rc = lnet_discover_peer_locked(lpni, cpt, true);
4308 /* Peer may have changed. */
4309 lp = lpni->lpni_peer_net->lpn_peer;
4310 if (lp->lp_nnis < n_ids)
4311 n_ids = lp->lp_nnis;
4315 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4316 buf[i].pid = id.pid;
4317 buf[i].nid = p->lpni_nid;
4322 lnet_net_unlock(cpt);
4325 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4331 lnet_peer_ni_decref_locked(lpni);
4333 lnet_net_unlock(cpt);
4335 LIBCFS_FREE(buf, buf_size);
4341 * Retrieve peer discovery status.
4343 * \retval 1 if lnet_peer_discovery_disabled is 0
4344 * \retval 0 if lnet_peer_discovery_disabled is 1
4347 LNetGetPeerDiscoveryStatus(void)
4349 return !lnet_peer_discovery_disabled;
4351 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);