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/
32 #define DEBUG_SUBSYSTEM S_LNET
34 #include <linux/ctype.h>
35 #include <linux/log2.h>
36 #include <linux/ktime.h>
37 #include <linux/moduleparam.h>
38 #include <linux/uaccess.h>
39 #ifdef HAVE_SCHED_HEADERS
40 #include <linux/sched/signal.h>
43 #include <lnet/udsp.h>
44 #include <lnet/lib-lnet.h>
46 #define D_LNI D_CONSOLE
49 * initialize ln_api_mutex statically, since it needs to be used in
50 * discovery_set callback. That module parameter callback can be called
51 * before module init completes. The mutex needs to be ready for use then.
53 struct lnet the_lnet = {
54 .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
55 }; /* THE state of the network */
56 EXPORT_SYMBOL(the_lnet);
58 static char *ip2nets = "";
59 module_param(ip2nets, charp, 0444);
60 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
62 static char *networks = "";
63 module_param(networks, charp, 0444);
64 MODULE_PARM_DESC(networks, "local networks");
66 static char *routes = "";
67 module_param(routes, charp, 0444);
68 MODULE_PARM_DESC(routes, "routes to non-local networks");
70 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
71 module_param(rnet_htable_size, int, 0444);
72 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
74 static int use_tcp_bonding;
75 module_param(use_tcp_bonding, int, 0444);
76 MODULE_PARM_DESC(use_tcp_bonding,
77 "use_tcp_bonding parameter has been removed");
79 unsigned int lnet_numa_range = 0;
80 module_param(lnet_numa_range, uint, 0444);
81 MODULE_PARM_DESC(lnet_numa_range,
82 "NUMA range to consider during Multi-Rail selection");
85 * lnet_health_sensitivity determines by how much we decrement the health
86 * value on sending error. The value defaults to 100, which means health
87 * interface health is decremented by 100 points every failure.
89 unsigned int lnet_health_sensitivity = 100;
90 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
91 #ifdef HAVE_KERNEL_PARAM_OPS
92 static struct kernel_param_ops param_ops_health_sensitivity = {
93 .set = sensitivity_set,
96 #define param_check_health_sensitivity(name, p) \
97 __param_check(name, p, int)
98 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
100 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
101 &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
103 MODULE_PARM_DESC(lnet_health_sensitivity,
104 "Value to decrement the health value by on error");
107 * lnet_recovery_interval determines how often we should perform recovery
108 * on unhealthy interfaces.
110 unsigned int lnet_recovery_interval = 1;
111 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
112 #ifdef HAVE_KERNEL_PARAM_OPS
113 static struct kernel_param_ops param_ops_recovery_interval = {
114 .set = recovery_interval_set,
115 .get = param_get_int,
117 #define param_check_recovery_interval(name, p) \
118 __param_check(name, p, int)
119 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
121 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
122 &lnet_recovery_interval, S_IRUGO|S_IWUSR);
124 MODULE_PARM_DESC(lnet_recovery_interval,
125 "Interval to recover unhealthy interfaces in seconds");
127 unsigned int lnet_recovery_limit;
128 module_param(lnet_recovery_limit, uint, 0644);
129 MODULE_PARM_DESC(lnet_recovery_limit,
130 "How long to attempt recovery of unhealthy peer interfaces in seconds. Set to 0 to allow indefinite recovery");
132 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
133 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
135 static struct kernel_param_ops param_ops_interfaces_max = {
137 .get = param_get_int,
140 #define param_check_interfaces_max(name, p) \
141 __param_check(name, p, int)
143 #ifdef HAVE_KERNEL_PARAM_OPS
144 module_param(lnet_interfaces_max, interfaces_max, 0644);
146 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
147 ¶m_ops_interfaces_max, 0644);
149 MODULE_PARM_DESC(lnet_interfaces_max,
150 "Maximum number of interfaces in a node.");
152 unsigned lnet_peer_discovery_disabled = 0;
153 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
155 static struct kernel_param_ops param_ops_discovery_disabled = {
156 .set = discovery_set,
157 .get = param_get_int,
160 #define param_check_discovery_disabled(name, p) \
161 __param_check(name, p, int)
162 #ifdef HAVE_KERNEL_PARAM_OPS
163 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
165 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
166 ¶m_ops_discovery_disabled, 0644);
168 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
169 "Set to 1 to disable peer discovery on this node.");
171 unsigned int lnet_drop_asym_route;
172 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
174 static struct kernel_param_ops param_ops_drop_asym_route = {
175 .set = drop_asym_route_set,
176 .get = param_get_int,
179 #define param_check_drop_asym_route(name, p) \
180 __param_check(name, p, int)
181 #ifdef HAVE_KERNEL_PARAM_OPS
182 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
184 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
185 ¶m_ops_drop_asym_route, 0644);
187 MODULE_PARM_DESC(lnet_drop_asym_route,
188 "Set to 1 to drop asymmetrical route messages.");
190 #define LNET_TRANSACTION_TIMEOUT_DEFAULT 50
191 unsigned int lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_DEFAULT;
192 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
193 #ifdef HAVE_KERNEL_PARAM_OPS
194 static struct kernel_param_ops param_ops_transaction_timeout = {
195 .set = transaction_to_set,
196 .get = param_get_int,
199 #define param_check_transaction_timeout(name, p) \
200 __param_check(name, p, int)
201 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
203 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
204 &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
206 MODULE_PARM_DESC(lnet_transaction_timeout,
207 "Maximum number of seconds to wait for a peer response.");
209 #define LNET_RETRY_COUNT_DEFAULT 2
210 unsigned int lnet_retry_count = LNET_RETRY_COUNT_DEFAULT;
211 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
212 #ifdef HAVE_KERNEL_PARAM_OPS
213 static struct kernel_param_ops param_ops_retry_count = {
214 .set = retry_count_set,
215 .get = param_get_int,
218 #define param_check_retry_count(name, p) \
219 __param_check(name, p, int)
220 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
222 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
223 &lnet_retry_count, S_IRUGO|S_IWUSR);
225 MODULE_PARM_DESC(lnet_retry_count,
226 "Maximum number of times to retry transmitting a message");
228 unsigned int lnet_response_tracking = 3;
229 static int response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp);
231 #ifdef HAVE_KERNEL_PARAM_OPS
232 static struct kernel_param_ops param_ops_response_tracking = {
233 .set = response_tracking_set,
234 .get = param_get_int,
237 #define param_check_response_tracking(name, p) \
238 __param_check(name, p, int)
239 module_param(lnet_response_tracking, response_tracking, 0644);
241 module_param_call(lnet_response_tracking, response_tracking_set, param_get_int,
242 &lnet_response_tracking, 0644);
244 MODULE_PARM_DESC(lnet_response_tracking,
245 "(0|1|2|3) LNet Internal Only|GET Reply only|PUT ACK only|Full Tracking (default)");
247 #define LNET_LND_TIMEOUT_DEFAULT ((LNET_TRANSACTION_TIMEOUT_DEFAULT - 1) / \
248 (LNET_RETRY_COUNT_DEFAULT + 1))
249 unsigned int lnet_lnd_timeout = LNET_LND_TIMEOUT_DEFAULT;
250 static void lnet_set_lnd_timeout(void)
252 lnet_lnd_timeout = (lnet_transaction_timeout - 1) /
253 (lnet_retry_count + 1);
256 unsigned int lnet_current_net_count;
259 * This sequence number keeps track of how many times DLC was used to
260 * update the local NIs. It is incremented when a NI is added or
261 * removed and checked when sending a message to determine if there is
262 * a need to re-run the selection algorithm. See lnet_select_pathway()
263 * for more details on its usage.
265 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
267 static int lnet_ping(struct lnet_process_id id, signed long timeout,
268 struct lnet_process_id __user *ids, int n_ids);
270 static int lnet_discover(struct lnet_process_id id, __u32 force,
271 struct lnet_process_id __user *ids, int n_ids);
274 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
277 unsigned *sensitivity = (unsigned *)kp->arg;
280 rc = kstrtoul(val, 0, &value);
282 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
287 * The purpose of locking the api_mutex here is to ensure that
288 * the correct value ends up stored properly.
290 mutex_lock(&the_lnet.ln_api_mutex);
292 if (value > LNET_MAX_HEALTH_VALUE) {
293 mutex_unlock(&the_lnet.ln_api_mutex);
294 CERROR("Invalid health value. Maximum: %d value = %lu\n",
295 LNET_MAX_HEALTH_VALUE, value);
299 if (*sensitivity != 0 && value == 0 && lnet_retry_count != 0) {
300 lnet_retry_count = 0;
301 lnet_set_lnd_timeout();
304 *sensitivity = value;
306 mutex_unlock(&the_lnet.ln_api_mutex);
312 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
315 unsigned *interval = (unsigned *)kp->arg;
318 rc = kstrtoul(val, 0, &value);
320 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
325 CERROR("lnet_recovery_interval must be at least 1 second\n");
330 * The purpose of locking the api_mutex here is to ensure that
331 * the correct value ends up stored properly.
333 mutex_lock(&the_lnet.ln_api_mutex);
337 mutex_unlock(&the_lnet.ln_api_mutex);
343 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
346 unsigned *discovery_off = (unsigned *)kp->arg;
348 struct lnet_ping_buffer *pbuf;
350 rc = kstrtoul(val, 0, &value);
352 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
356 value = (value) ? 1 : 0;
359 * The purpose of locking the api_mutex here is to ensure that
360 * the correct value ends up stored properly.
362 mutex_lock(&the_lnet.ln_api_mutex);
364 if (value == *discovery_off) {
365 mutex_unlock(&the_lnet.ln_api_mutex);
370 * We still want to set the discovery value even when LNet is not
371 * running. This is the case when LNet is being loaded and we want
372 * the module parameters to take effect. Otherwise if we're
373 * changing the value dynamically, we want to set it after
376 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
377 *discovery_off = value;
378 mutex_unlock(&the_lnet.ln_api_mutex);
382 /* tell peers that discovery setting has changed */
383 lnet_net_lock(LNET_LOCK_EX);
384 pbuf = the_lnet.ln_ping_target;
386 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
388 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
389 lnet_net_unlock(LNET_LOCK_EX);
391 /* only send a push when we're turning off discovery */
392 if (*discovery_off <= 0 && value > 0)
393 lnet_push_update_to_peers(1);
394 *discovery_off = value;
396 mutex_unlock(&the_lnet.ln_api_mutex);
402 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
405 unsigned int *drop_asym_route = (unsigned int *)kp->arg;
408 rc = kstrtoul(val, 0, &value);
410 CERROR("Invalid module parameter value for "
411 "'lnet_drop_asym_route'\n");
416 * The purpose of locking the api_mutex here is to ensure that
417 * the correct value ends up stored properly.
419 mutex_lock(&the_lnet.ln_api_mutex);
421 if (value == *drop_asym_route) {
422 mutex_unlock(&the_lnet.ln_api_mutex);
426 *drop_asym_route = value;
428 mutex_unlock(&the_lnet.ln_api_mutex);
434 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
437 unsigned *transaction_to = (unsigned *)kp->arg;
440 rc = kstrtoul(val, 0, &value);
442 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
447 * The purpose of locking the api_mutex here is to ensure that
448 * the correct value ends up stored properly.
450 mutex_lock(&the_lnet.ln_api_mutex);
452 if (value <= lnet_retry_count || value == 0) {
453 mutex_unlock(&the_lnet.ln_api_mutex);
454 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
455 "Has to be greater than lnet_retry_count (%u)\n",
456 value, lnet_retry_count);
460 if (value == *transaction_to) {
461 mutex_unlock(&the_lnet.ln_api_mutex);
465 *transaction_to = value;
466 /* Update the lnet_lnd_timeout now that we've modified the
467 * transaction timeout
469 lnet_set_lnd_timeout();
471 mutex_unlock(&the_lnet.ln_api_mutex);
477 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
480 unsigned *retry_count = (unsigned *)kp->arg;
483 rc = kstrtoul(val, 0, &value);
485 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
490 * The purpose of locking the api_mutex here is to ensure that
491 * the correct value ends up stored properly.
493 mutex_lock(&the_lnet.ln_api_mutex);
495 if (lnet_health_sensitivity == 0 && value > 0) {
496 mutex_unlock(&the_lnet.ln_api_mutex);
497 CERROR("Can not set lnet_retry_count when health feature is turned off\n");
501 if (value > lnet_transaction_timeout) {
502 mutex_unlock(&the_lnet.ln_api_mutex);
503 CERROR("Invalid value for lnet_retry_count (%lu). "
504 "Has to be smaller than lnet_transaction_timeout (%u)\n",
505 value, lnet_transaction_timeout);
509 *retry_count = value;
511 /* Update the lnet_lnd_timeout now that we've modified the
514 lnet_set_lnd_timeout();
516 mutex_unlock(&the_lnet.ln_api_mutex);
522 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
526 rc = kstrtoint(val, 0, &value);
528 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
532 if (value < LNET_INTERFACES_MIN) {
533 CWARN("max interfaces provided are too small, setting to %d\n",
534 LNET_INTERFACES_MAX_DEFAULT);
535 value = LNET_INTERFACES_MAX_DEFAULT;
538 *(int *)kp->arg = value;
544 response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp)
547 unsigned long new_value;
549 rc = kstrtoul(val, 0, &new_value);
551 CERROR("Invalid value for 'lnet_response_tracking'\n");
555 if (new_value < 0 || new_value > 3) {
556 CWARN("Invalid value (%lu) for 'lnet_response_tracking'\n",
561 lnet_response_tracking = new_value;
567 lnet_get_routes(void)
573 lnet_get_networks(void)
578 if (*networks != 0 && *ip2nets != 0) {
579 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
580 "'ip2nets' but not both at once\n");
585 rc = lnet_parse_ip2nets(&nets, ip2nets);
586 return (rc == 0) ? nets : NULL;
596 lnet_init_locks(void)
598 spin_lock_init(&the_lnet.ln_eq_wait_lock);
599 spin_lock_init(&the_lnet.ln_msg_resend_lock);
600 init_completion(&the_lnet.ln_mt_wait_complete);
601 mutex_init(&the_lnet.ln_lnd_mutex);
604 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
605 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
607 struct kmem_cache *lnet_udsp_cachep; /* udsp cache */
608 struct kmem_cache *lnet_rspt_cachep; /* response tracker cache */
609 struct kmem_cache *lnet_msg_cachep;
612 lnet_slab_setup(void)
614 /* create specific kmem_cache for MEs and small MDs (i.e., originally
615 * allocated in <size-xxx> kmem_cache).
617 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
619 if (!lnet_mes_cachep)
622 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
623 LNET_SMALL_MD_SIZE, 0, 0,
625 if (!lnet_small_mds_cachep)
628 lnet_udsp_cachep = kmem_cache_create("lnet_udsp",
629 sizeof(struct lnet_udsp),
631 if (!lnet_udsp_cachep)
634 lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
636 if (!lnet_rspt_cachep)
639 lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
641 if (!lnet_msg_cachep)
648 lnet_slab_cleanup(void)
650 if (lnet_msg_cachep) {
651 kmem_cache_destroy(lnet_msg_cachep);
652 lnet_msg_cachep = NULL;
655 if (lnet_rspt_cachep) {
656 kmem_cache_destroy(lnet_rspt_cachep);
657 lnet_rspt_cachep = NULL;
660 if (lnet_udsp_cachep) {
661 kmem_cache_destroy(lnet_udsp_cachep);
662 lnet_udsp_cachep = NULL;
665 if (lnet_small_mds_cachep) {
666 kmem_cache_destroy(lnet_small_mds_cachep);
667 lnet_small_mds_cachep = NULL;
670 if (lnet_mes_cachep) {
671 kmem_cache_destroy(lnet_mes_cachep);
672 lnet_mes_cachep = NULL;
677 lnet_create_remote_nets_table(void)
680 struct list_head *hash;
682 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
683 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
684 CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
686 CERROR("Failed to create remote nets hash table\n");
690 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
691 INIT_LIST_HEAD(&hash[i]);
692 the_lnet.ln_remote_nets_hash = hash;
697 lnet_destroy_remote_nets_table(void)
701 if (the_lnet.ln_remote_nets_hash == NULL)
704 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
705 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
707 CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
708 LNET_REMOTE_NETS_HASH_SIZE);
709 the_lnet.ln_remote_nets_hash = NULL;
713 lnet_destroy_locks(void)
715 if (the_lnet.ln_res_lock != NULL) {
716 cfs_percpt_lock_free(the_lnet.ln_res_lock);
717 the_lnet.ln_res_lock = NULL;
720 if (the_lnet.ln_net_lock != NULL) {
721 cfs_percpt_lock_free(the_lnet.ln_net_lock);
722 the_lnet.ln_net_lock = NULL;
727 lnet_create_locks(void)
731 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
732 if (the_lnet.ln_res_lock == NULL)
735 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
736 if (the_lnet.ln_net_lock == NULL)
742 lnet_destroy_locks();
746 static void lnet_assert_wire_constants(void)
748 /* Wire protocol assertions generated by 'wirecheck'
749 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
750 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
751 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
755 BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
756 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
757 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
758 BUILD_BUG_ON(LNET_MSG_ACK != 0);
759 BUILD_BUG_ON(LNET_MSG_PUT != 1);
760 BUILD_BUG_ON(LNET_MSG_GET != 2);
761 BUILD_BUG_ON(LNET_MSG_REPLY != 3);
762 BUILD_BUG_ON(LNET_MSG_HELLO != 4);
764 BUILD_BUG_ON((int)sizeof(lnet_nid_t) != 8);
765 BUILD_BUG_ON((int)sizeof(lnet_pid_t) != 4);
767 /* Checks for struct lnet_process_id_packed */
768 BUILD_BUG_ON((int)sizeof(struct lnet_process_id_packed) != 12);
769 BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, nid) != 0);
770 BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->nid) != 8);
771 BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, pid) != 8);
772 BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->pid) != 4);
774 /* Checks for struct lnet_handle_wire */
775 BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
776 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
777 wh_interface_cookie) != 0);
778 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
779 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
780 wh_object_cookie) != 8);
781 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
783 /* Checks for struct struct lnet_magicversion */
784 BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
785 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
786 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
787 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
788 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
789 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
790 version_minor) != 6);
791 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
793 /* Checks for struct struct lnet_hdr */
794 BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
795 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
796 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
797 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
798 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
799 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
800 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
801 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
802 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
803 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
804 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
805 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
806 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
807 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
808 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
811 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
812 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
813 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
814 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
815 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
816 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
819 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
820 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
821 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
822 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
823 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
824 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
825 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
826 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
827 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
828 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
831 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
832 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
833 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
834 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
835 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
836 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
837 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
838 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
839 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
840 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
843 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
844 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
847 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
848 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
849 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
850 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
852 /* Checks for struct lnet_ni_status and related constants */
853 BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
854 BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
855 BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
857 /* Checks for struct lnet_ni_status */
858 BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
859 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
860 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
861 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
862 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
863 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
864 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
866 /* Checks for struct lnet_ping_info and related constants */
867 BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
868 BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
869 BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
870 BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
871 BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
872 BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
873 BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
874 BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
876 /* Checks for struct lnet_ping_info */
877 BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
878 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
879 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
880 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
881 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
882 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
883 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
884 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
885 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
886 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
887 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
889 /* Acceptor connection request */
890 BUILD_BUG_ON(LNET_PROTO_ACCEPTOR_VERSION != 1);
892 /* Checks for struct lnet_acceptor_connreq */
893 BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq) != 16);
894 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_magic) != 0);
895 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_magic) != 4);
896 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_version) != 4);
897 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_version) != 4);
898 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_nid) != 8);
899 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_nid) != 8);
901 /* Checks for struct lnet_counters_common */
902 BUILD_BUG_ON((int)sizeof(struct lnet_counters_common) != 60);
903 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_alloc) != 0);
904 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_alloc) != 4);
905 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_max) != 4);
906 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_max) != 4);
907 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_errors) != 8);
908 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_errors) != 4);
909 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_count) != 12);
910 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_count) != 4);
911 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_count) != 16);
912 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_count) != 4);
913 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_count) != 20);
914 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_count) != 4);
915 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_count) != 24);
916 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_count) != 4);
917 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_length) != 28);
918 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_length) != 8);
919 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_length) != 36);
920 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_length) != 8);
921 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_length) != 44);
922 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_length) != 8);
923 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_length) != 52);
924 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_length) != 8);
927 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
929 const struct lnet_lnd *lnd;
931 /* holding lnd mutex */
932 if (type >= NUM_LNDS)
934 lnd = the_lnet.ln_lnds[type];
935 LASSERT(!lnd || lnd->lnd_type == type);
941 lnet_get_lnd_timeout(void)
943 return lnet_lnd_timeout;
945 EXPORT_SYMBOL(lnet_get_lnd_timeout);
948 lnet_register_lnd(const struct lnet_lnd *lnd)
950 mutex_lock(&the_lnet.ln_lnd_mutex);
952 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
953 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
955 the_lnet.ln_lnds[lnd->lnd_type] = lnd;
957 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
959 mutex_unlock(&the_lnet.ln_lnd_mutex);
961 EXPORT_SYMBOL(lnet_register_lnd);
964 lnet_unregister_lnd(const struct lnet_lnd *lnd)
966 mutex_lock(&the_lnet.ln_lnd_mutex);
968 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
970 the_lnet.ln_lnds[lnd->lnd_type] = NULL;
971 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
973 mutex_unlock(&the_lnet.ln_lnd_mutex);
975 EXPORT_SYMBOL(lnet_unregister_lnd);
978 lnet_counters_get_common_locked(struct lnet_counters_common *common)
980 struct lnet_counters *ctr;
983 /* FIXME !!! Their is no assert_lnet_net_locked() to ensure this
984 * actually called under the protection of the lnet_net_lock.
986 memset(common, 0, sizeof(*common));
988 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
989 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
990 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
991 common->lcc_errors += ctr->lct_common.lcc_errors;
992 common->lcc_send_count += ctr->lct_common.lcc_send_count;
993 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
994 common->lcc_route_count += ctr->lct_common.lcc_route_count;
995 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
996 common->lcc_send_length += ctr->lct_common.lcc_send_length;
997 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
998 common->lcc_route_length += ctr->lct_common.lcc_route_length;
999 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
1004 lnet_counters_get_common(struct lnet_counters_common *common)
1006 lnet_net_lock(LNET_LOCK_EX);
1007 lnet_counters_get_common_locked(common);
1008 lnet_net_unlock(LNET_LOCK_EX);
1010 EXPORT_SYMBOL(lnet_counters_get_common);
1013 lnet_counters_get(struct lnet_counters *counters)
1015 struct lnet_counters *ctr;
1016 struct lnet_counters_health *health = &counters->lct_health;
1019 memset(counters, 0, sizeof(*counters));
1021 lnet_net_lock(LNET_LOCK_EX);
1023 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1024 GOTO(out_unlock, rc = -ENODEV);
1026 lnet_counters_get_common_locked(&counters->lct_common);
1028 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
1029 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
1030 health->lch_resend_count += ctr->lct_health.lch_resend_count;
1031 health->lch_response_timeout_count +=
1032 ctr->lct_health.lch_response_timeout_count;
1033 health->lch_local_interrupt_count +=
1034 ctr->lct_health.lch_local_interrupt_count;
1035 health->lch_local_dropped_count +=
1036 ctr->lct_health.lch_local_dropped_count;
1037 health->lch_local_aborted_count +=
1038 ctr->lct_health.lch_local_aborted_count;
1039 health->lch_local_no_route_count +=
1040 ctr->lct_health.lch_local_no_route_count;
1041 health->lch_local_timeout_count +=
1042 ctr->lct_health.lch_local_timeout_count;
1043 health->lch_local_error_count +=
1044 ctr->lct_health.lch_local_error_count;
1045 health->lch_remote_dropped_count +=
1046 ctr->lct_health.lch_remote_dropped_count;
1047 health->lch_remote_error_count +=
1048 ctr->lct_health.lch_remote_error_count;
1049 health->lch_remote_timeout_count +=
1050 ctr->lct_health.lch_remote_timeout_count;
1051 health->lch_network_timeout_count +=
1052 ctr->lct_health.lch_network_timeout_count;
1055 lnet_net_unlock(LNET_LOCK_EX);
1058 EXPORT_SYMBOL(lnet_counters_get);
1061 lnet_counters_reset(void)
1063 struct lnet_counters *counters;
1066 lnet_net_lock(LNET_LOCK_EX);
1068 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1071 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
1072 memset(counters, 0, sizeof(struct lnet_counters));
1074 lnet_net_unlock(LNET_LOCK_EX);
1078 lnet_res_type2str(int type)
1083 case LNET_COOKIE_TYPE_MD:
1085 case LNET_COOKIE_TYPE_ME:
1087 case LNET_COOKIE_TYPE_EQ:
1093 lnet_res_container_cleanup(struct lnet_res_container *rec)
1097 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
1100 while (!list_empty(&rec->rec_active)) {
1101 struct list_head *e = rec->rec_active.next;
1104 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1105 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1107 } else { /* NB: Active MEs should be attached on portals */
1114 /* Found alive MD/ME/EQ, user really should unlink/free
1115 * all of them before finalize LNet, but if someone didn't,
1116 * we have to recycle garbage for him */
1117 CERROR("%d active elements on exit of %s container\n",
1118 count, lnet_res_type2str(rec->rec_type));
1121 if (rec->rec_lh_hash != NULL) {
1122 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1123 rec->rec_lh_hash = NULL;
1126 rec->rec_type = 0; /* mark it as finalized */
1130 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1135 LASSERT(rec->rec_type == 0);
1137 rec->rec_type = type;
1138 INIT_LIST_HEAD(&rec->rec_active);
1140 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1142 /* Arbitrary choice of hash table size */
1143 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1144 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1145 if (rec->rec_lh_hash == NULL) {
1150 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1151 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1156 CERROR("Failed to setup %s resource container\n",
1157 lnet_res_type2str(type));
1158 lnet_res_container_cleanup(rec);
1163 lnet_res_containers_destroy(struct lnet_res_container **recs)
1165 struct lnet_res_container *rec;
1168 cfs_percpt_for_each(rec, i, recs)
1169 lnet_res_container_cleanup(rec);
1171 cfs_percpt_free(recs);
1174 static struct lnet_res_container **
1175 lnet_res_containers_create(int type)
1177 struct lnet_res_container **recs;
1178 struct lnet_res_container *rec;
1182 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1184 CERROR("Failed to allocate %s resource containers\n",
1185 lnet_res_type2str(type));
1189 cfs_percpt_for_each(rec, i, recs) {
1190 rc = lnet_res_container_setup(rec, i, type);
1192 lnet_res_containers_destroy(recs);
1200 struct lnet_libhandle *
1201 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1203 /* ALWAYS called with lnet_res_lock held */
1204 struct list_head *head;
1205 struct lnet_libhandle *lh;
1208 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1211 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1212 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1214 list_for_each_entry(lh, head, lh_hash_chain) {
1215 if (lh->lh_cookie == cookie)
1223 lnet_res_lh_initialize(struct lnet_res_container *rec,
1224 struct lnet_libhandle *lh)
1226 /* ALWAYS called with lnet_res_lock held */
1227 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1230 lh->lh_cookie = rec->rec_lh_cookie;
1231 rec->rec_lh_cookie += 1 << ibits;
1233 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1235 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1239 lnet_create_array_of_queues(void)
1241 struct list_head **qs;
1242 struct list_head *q;
1245 qs = cfs_percpt_alloc(lnet_cpt_table(),
1246 sizeof(struct list_head));
1248 CERROR("Failed to allocate queues\n");
1252 cfs_percpt_for_each(q, i, qs)
1258 static int lnet_unprepare(void);
1261 lnet_prepare(lnet_pid_t requested_pid)
1263 /* Prepare to bring up the network */
1264 struct lnet_res_container **recs;
1267 if (requested_pid == LNET_PID_ANY) {
1268 /* Don't instantiate LNET just for me */
1272 LASSERT(the_lnet.ln_refcount == 0);
1274 the_lnet.ln_routing = 0;
1276 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1277 the_lnet.ln_pid = requested_pid;
1279 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1280 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1281 INIT_LIST_HEAD(&the_lnet.ln_nets);
1282 INIT_LIST_HEAD(&the_lnet.ln_routers);
1283 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1284 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1285 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1286 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1287 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1288 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1289 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1290 INIT_LIST_HEAD(&the_lnet.ln_udsp_list);
1291 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1292 the_lnet.ln_mt_handler = NULL;
1293 init_completion(&the_lnet.ln_started);
1295 rc = lnet_slab_setup();
1299 rc = lnet_create_remote_nets_table();
1304 * NB the interface cookie in wire handles guards against delayed
1305 * replies and ACKs appearing valid after reboot.
1307 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1309 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1310 sizeof(struct lnet_counters));
1311 if (the_lnet.ln_counters == NULL) {
1312 CERROR("Failed to allocate counters for LNet\n");
1317 rc = lnet_peer_tables_create();
1321 rc = lnet_msg_containers_create();
1325 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1326 LNET_COOKIE_TYPE_EQ);
1330 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1336 the_lnet.ln_md_containers = recs;
1338 rc = lnet_portals_create();
1340 CERROR("Failed to create portals for LNet: %d\n", rc);
1344 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1345 if (!the_lnet.ln_mt_zombie_rstqs) {
1358 lnet_unprepare (void)
1360 /* NB no LNET_LOCK since this is the last reference. All LND instances
1361 * have shut down already, so it is safe to unlink and free all
1362 * descriptors, even those that appear committed to a network op (eg MD
1363 * with non-zero pending count) */
1365 lnet_fail_nid(LNET_NID_ANY, 0);
1367 LASSERT(the_lnet.ln_refcount == 0);
1368 LASSERT(list_empty(&the_lnet.ln_test_peers));
1369 LASSERT(list_empty(&the_lnet.ln_nets));
1371 if (the_lnet.ln_mt_zombie_rstqs) {
1372 lnet_clean_zombie_rstqs();
1373 the_lnet.ln_mt_zombie_rstqs = NULL;
1376 lnet_assert_handler_unused(the_lnet.ln_mt_handler);
1377 the_lnet.ln_mt_handler = NULL;
1379 lnet_portals_destroy();
1381 if (the_lnet.ln_md_containers != NULL) {
1382 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1383 the_lnet.ln_md_containers = NULL;
1386 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1388 lnet_msg_containers_destroy();
1390 lnet_rtrpools_free(0);
1392 if (the_lnet.ln_counters != NULL) {
1393 cfs_percpt_free(the_lnet.ln_counters);
1394 the_lnet.ln_counters = NULL;
1396 lnet_destroy_remote_nets_table();
1397 lnet_udsp_destroy(true);
1398 lnet_slab_cleanup();
1404 lnet_net2ni_locked(__u32 net_id, int cpt)
1407 struct lnet_net *net;
1409 LASSERT(cpt != LNET_LOCK_EX);
1411 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1412 if (net->net_id == net_id) {
1413 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1423 lnet_net2ni_addref(__u32 net)
1428 ni = lnet_net2ni_locked(net, 0);
1430 lnet_ni_addref_locked(ni, 0);
1435 EXPORT_SYMBOL(lnet_net2ni_addref);
1438 lnet_get_net_locked(__u32 net_id)
1440 struct lnet_net *net;
1442 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1443 if (net->net_id == net_id)
1451 lnet_net_clr_pref_rtrs(struct lnet_net *net)
1453 struct list_head zombies;
1454 struct lnet_nid_list *ne;
1455 struct lnet_nid_list *tmp;
1457 INIT_LIST_HEAD(&zombies);
1459 lnet_net_lock(LNET_LOCK_EX);
1460 list_splice_init(&net->net_rtr_pref_nids, &zombies);
1461 lnet_net_unlock(LNET_LOCK_EX);
1463 list_for_each_entry_safe(ne, tmp, &zombies, nl_list) {
1464 list_del_init(&ne->nl_list);
1465 LIBCFS_FREE(ne, sizeof(*ne));
1470 lnet_net_add_pref_rtr(struct lnet_net *net,
1472 __must_hold(&the_lnet.ln_api_mutex)
1474 struct lnet_nid_list *ne;
1476 /* This function is called with api_mutex held. When the api_mutex
1477 * is held the list can not be modified, as it is only modified as
1478 * a result of applying a UDSP and that happens under api_mutex
1481 list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1482 if (ne->nl_nid == gw_nid)
1486 LIBCFS_ALLOC(ne, sizeof(*ne));
1490 ne->nl_nid = gw_nid;
1492 /* Lock the cpt to protect against addition and checks in the
1493 * selection algorithm
1495 lnet_net_lock(LNET_LOCK_EX);
1496 list_add(&ne->nl_list, &net->net_rtr_pref_nids);
1497 lnet_net_unlock(LNET_LOCK_EX);
1503 lnet_net_is_pref_rtr_locked(struct lnet_net *net, lnet_nid_t rtr_nid)
1505 struct lnet_nid_list *ne;
1507 CDEBUG(D_NET, "%s: rtr pref emtpy: %d\n",
1508 libcfs_net2str(net->net_id),
1509 list_empty(&net->net_rtr_pref_nids));
1511 if (list_empty(&net->net_rtr_pref_nids))
1514 list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1515 CDEBUG(D_NET, "Comparing pref %s with gw %s\n",
1516 libcfs_nid2str(ne->nl_nid),
1517 libcfs_nid2str(rtr_nid));
1518 if (rtr_nid == ne->nl_nid)
1526 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1531 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1536 val = hash_long(key, LNET_CPT_BITS);
1537 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1541 return (unsigned int)(key + val + (val >> 1)) % number;
1545 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1547 struct lnet_net *net;
1549 /* must called with hold of lnet_net_lock */
1550 if (LNET_CPT_NUMBER == 1)
1551 return 0; /* the only one */
1554 * If NI is provided then use the CPT identified in the NI cpt
1555 * list if one exists. If one doesn't exist, then that NI is
1556 * associated with all CPTs and it follows that the net it belongs
1557 * to is implicitly associated with all CPTs, so just hash the nid
1561 if (ni->ni_cpts != NULL)
1562 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1565 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1568 /* no NI provided so look at the net */
1569 net = lnet_get_net_locked(LNET_NIDNET(nid));
1571 if (net != NULL && net->net_cpts != NULL) {
1572 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1575 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1579 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1584 if (LNET_CPT_NUMBER == 1)
1585 return 0; /* the only one */
1587 cpt = lnet_net_lock_current();
1589 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1591 lnet_net_unlock(cpt);
1595 EXPORT_SYMBOL(lnet_cpt_of_nid);
1598 lnet_islocalnet_locked(__u32 net_id)
1600 struct lnet_net *net;
1603 net = lnet_get_net_locked(net_id);
1605 local = net != NULL;
1611 lnet_islocalnet(__u32 net_id)
1616 cpt = lnet_net_lock_current();
1618 local = lnet_islocalnet_locked(net_id);
1620 lnet_net_unlock(cpt);
1626 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1628 struct lnet_net *net;
1631 LASSERT(cpt != LNET_LOCK_EX);
1633 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1634 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1635 if (ni->ni_nid == nid)
1644 lnet_nid2ni_addref(lnet_nid_t nid)
1649 ni = lnet_nid2ni_locked(nid, 0);
1651 lnet_ni_addref_locked(ni, 0);
1656 EXPORT_SYMBOL(lnet_nid2ni_addref);
1659 lnet_islocalnid(lnet_nid_t nid)
1664 cpt = lnet_net_lock_current();
1665 ni = lnet_nid2ni_locked(nid, cpt);
1666 lnet_net_unlock(cpt);
1672 lnet_count_acceptor_nets(void)
1674 /* Return the # of NIs that need the acceptor. */
1676 struct lnet_net *net;
1679 cpt = lnet_net_lock_current();
1680 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1681 /* all socklnd type networks should have the acceptor
1683 if (net->net_lnd->lnd_accept != NULL)
1687 lnet_net_unlock(cpt);
1692 struct lnet_ping_buffer *
1693 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1695 struct lnet_ping_buffer *pbuf;
1697 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1699 pbuf->pb_nnis = nnis;
1700 pbuf->pb_needs_post = false;
1701 atomic_set(&pbuf->pb_refcnt, 1);
1708 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1710 LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1711 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1714 static struct lnet_ping_buffer *
1715 lnet_ping_target_create(int nnis)
1717 struct lnet_ping_buffer *pbuf;
1719 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1721 CERROR("Can't allocate ping source [%d]\n", nnis);
1725 pbuf->pb_info.pi_nnis = nnis;
1726 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1727 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1728 pbuf->pb_info.pi_features =
1729 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1735 lnet_get_net_ni_count_locked(struct lnet_net *net)
1740 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1747 lnet_get_net_ni_count_pre(struct lnet_net *net)
1752 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1759 lnet_get_ni_count(void)
1762 struct lnet_net *net;
1767 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1768 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1778 lnet_get_net_count(void)
1780 struct lnet_net *net;
1785 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1795 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1797 struct lnet_ni_status *stat;
1801 __swab32s(&pbuf->pb_info.pi_magic);
1802 __swab32s(&pbuf->pb_info.pi_features);
1803 __swab32s(&pbuf->pb_info.pi_pid);
1804 __swab32s(&pbuf->pb_info.pi_nnis);
1805 nnis = pbuf->pb_info.pi_nnis;
1806 if (nnis > pbuf->pb_nnis)
1807 nnis = pbuf->pb_nnis;
1808 for (i = 0; i < nnis; i++) {
1809 stat = &pbuf->pb_info.pi_ni[i];
1810 __swab64s(&stat->ns_nid);
1811 __swab32s(&stat->ns_status);
1816 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1820 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1822 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1824 /* Loopback is guaranteed to be present */
1825 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1827 if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1833 lnet_ping_target_destroy(void)
1835 struct lnet_net *net;
1838 lnet_net_lock(LNET_LOCK_EX);
1840 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1841 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1843 ni->ni_status = NULL;
1848 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1849 the_lnet.ln_ping_target = NULL;
1851 lnet_net_unlock(LNET_LOCK_EX);
1855 lnet_ping_target_event_handler(struct lnet_event *event)
1857 struct lnet_ping_buffer *pbuf = event->md_user_ptr;
1859 if (event->unlinked)
1860 lnet_ping_buffer_decref(pbuf);
1864 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1865 struct lnet_handle_md *ping_mdh,
1866 int ni_count, bool set_eq)
1868 struct lnet_process_id id = {
1869 .nid = LNET_NID_ANY,
1873 struct lnet_md md = { NULL };
1877 the_lnet.ln_ping_target_handler =
1878 lnet_ping_target_event_handler;
1880 *ppbuf = lnet_ping_target_create(ni_count);
1881 if (*ppbuf == NULL) {
1886 /* Ping target ME/MD */
1887 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1888 LNET_PROTO_PING_MATCHBITS, 0,
1889 LNET_UNLINK, LNET_INS_AFTER);
1892 CERROR("Can't create ping target ME: %d\n", rc);
1893 goto fail_decref_ping_buffer;
1896 /* initialize md content */
1897 md.start = &(*ppbuf)->pb_info;
1898 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1899 md.threshold = LNET_MD_THRESH_INF;
1901 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1902 LNET_MD_MANAGE_REMOTE;
1903 md.handler = the_lnet.ln_ping_target_handler;
1904 md.user_ptr = *ppbuf;
1906 rc = LNetMDAttach(me, &md, LNET_RETAIN, ping_mdh);
1908 CERROR("Can't attach ping target MD: %d\n", rc);
1909 goto fail_decref_ping_buffer;
1911 lnet_ping_buffer_addref(*ppbuf);
1915 fail_decref_ping_buffer:
1916 LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1917 lnet_ping_buffer_decref(*ppbuf);
1924 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1925 struct lnet_handle_md *ping_mdh)
1927 LNetMDUnlink(*ping_mdh);
1928 LNetInvalidateMDHandle(ping_mdh);
1930 /* NB the MD could be busy; this just starts the unlink */
1931 wait_var_event_warning(&pbuf->pb_refcnt,
1932 atomic_read(&pbuf->pb_refcnt) <= 1,
1933 "Still waiting for ping data MD to unlink\n");
1937 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1940 struct lnet_net *net;
1941 struct lnet_ni_status *ns;
1946 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1947 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1948 LASSERT(i < pbuf->pb_nnis);
1950 ns = &pbuf->pb_info.pi_ni[i];
1952 ns->ns_nid = ni->ni_nid;
1955 ns->ns_status = (ni->ni_status != NULL) ?
1956 ni->ni_status->ns_status :
1965 * We (ab)use the ns_status of the loopback interface to
1966 * transmit the sequence number. The first interface listed
1967 * must be the loopback interface.
1969 rc = lnet_ping_info_validate(&pbuf->pb_info);
1971 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1974 LNET_PING_BUFFER_SEQNO(pbuf) =
1975 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1979 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1980 struct lnet_handle_md ping_mdh)
1982 struct lnet_ping_buffer *old_pbuf = NULL;
1983 struct lnet_handle_md old_ping_md;
1985 /* switch the NIs to point to the new ping info created */
1986 lnet_net_lock(LNET_LOCK_EX);
1988 if (!the_lnet.ln_routing)
1989 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1990 if (!lnet_peer_discovery_disabled)
1991 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1993 /* Ensure only known feature bits have been set. */
1994 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1995 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1997 lnet_ping_target_install_locked(pbuf);
1999 if (the_lnet.ln_ping_target) {
2000 old_pbuf = the_lnet.ln_ping_target;
2001 old_ping_md = the_lnet.ln_ping_target_md;
2003 the_lnet.ln_ping_target_md = ping_mdh;
2004 the_lnet.ln_ping_target = pbuf;
2006 lnet_net_unlock(LNET_LOCK_EX);
2009 /* unlink and free the old ping info */
2010 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
2011 lnet_ping_buffer_decref(old_pbuf);
2014 lnet_push_update_to_peers(0);
2018 lnet_ping_target_fini(void)
2020 lnet_ping_md_unlink(the_lnet.ln_ping_target,
2021 &the_lnet.ln_ping_target_md);
2023 lnet_assert_handler_unused(the_lnet.ln_ping_target_handler);
2024 lnet_ping_target_destroy();
2027 /* Resize the push target. */
2028 int lnet_push_target_resize(void)
2030 struct lnet_handle_md mdh;
2031 struct lnet_handle_md old_mdh;
2032 struct lnet_ping_buffer *pbuf;
2033 struct lnet_ping_buffer *old_pbuf;
2038 nnis = the_lnet.ln_push_target_nnis;
2040 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
2044 /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
2045 * dropped when we need to resize again (see "old_pbuf" below) or when
2046 * LNet is shutdown (see lnet_push_target_fini())
2048 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
2050 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
2054 rc = lnet_push_target_post(pbuf, &mdh);
2056 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
2057 lnet_ping_buffer_decref(pbuf);
2061 lnet_net_lock(LNET_LOCK_EX);
2062 old_pbuf = the_lnet.ln_push_target;
2063 old_mdh = the_lnet.ln_push_target_md;
2064 the_lnet.ln_push_target = pbuf;
2065 the_lnet.ln_push_target_md = mdh;
2066 lnet_net_unlock(LNET_LOCK_EX);
2069 LNetMDUnlink(old_mdh);
2070 /* Drop ref set by lnet_ping_buffer_alloc() */
2071 lnet_ping_buffer_decref(old_pbuf);
2074 /* Received another push or reply that requires a larger buffer */
2075 if (nnis < the_lnet.ln_push_target_nnis)
2078 CDEBUG(D_NET, "nnis %d success\n", nnis);
2082 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
2083 struct lnet_handle_md *mdhp)
2085 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
2086 struct lnet_md md = { NULL };
2090 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
2091 LNET_PROTO_PING_MATCHBITS, 0,
2092 LNET_UNLINK, LNET_INS_AFTER);
2095 CERROR("Can't create push target ME: %d\n", rc);
2099 pbuf->pb_needs_post = false;
2101 /* This reference is dropped by lnet_push_target_event_handler() */
2102 lnet_ping_buffer_addref(pbuf);
2104 /* initialize md content */
2105 md.start = &pbuf->pb_info;
2106 md.length = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2109 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
2111 md.handler = the_lnet.ln_push_target_handler;
2113 rc = LNetMDAttach(me, &md, LNET_UNLINK, mdhp);
2115 CERROR("Can't attach push MD: %d\n", rc);
2116 lnet_ping_buffer_decref(pbuf);
2117 pbuf->pb_needs_post = true;
2121 CDEBUG(D_NET, "posted push target %p\n", pbuf);
2126 static void lnet_push_target_event_handler(struct lnet_event *ev)
2128 struct lnet_ping_buffer *pbuf = ev->md_user_ptr;
2130 CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
2133 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2134 lnet_swap_pinginfo(pbuf);
2136 if (ev->type == LNET_EVENT_UNLINK) {
2137 /* Drop ref added by lnet_push_target_post() */
2138 lnet_ping_buffer_decref(pbuf);
2142 lnet_peer_push_event(ev);
2144 /* Drop ref added by lnet_push_target_post */
2145 lnet_ping_buffer_decref(pbuf);
2148 /* Initialize the push target. */
2149 static int lnet_push_target_init(void)
2153 if (the_lnet.ln_push_target)
2156 the_lnet.ln_push_target_handler =
2157 lnet_push_target_event_handler;
2159 rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2162 /* Start at the required minimum, we'll enlarge if required. */
2163 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2165 rc = lnet_push_target_resize();
2168 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2169 the_lnet.ln_push_target_handler = NULL;
2175 /* Clean up the push target. */
2176 static void lnet_push_target_fini(void)
2178 if (!the_lnet.ln_push_target)
2181 /* Unlink and invalidate to prevent new references. */
2182 LNetMDUnlink(the_lnet.ln_push_target_md);
2183 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2185 /* Wait for the unlink to complete. */
2186 wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2187 atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2188 "Still waiting for ping data MD to unlink\n");
2190 /* Drop ref set by lnet_ping_buffer_alloc() */
2191 lnet_ping_buffer_decref(the_lnet.ln_push_target);
2192 the_lnet.ln_push_target = NULL;
2193 the_lnet.ln_push_target_nnis = 0;
2195 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2196 lnet_assert_handler_unused(the_lnet.ln_push_target_handler);
2197 the_lnet.ln_push_target_handler = NULL;
2201 lnet_ni_tq_credits(struct lnet_ni *ni)
2205 LASSERT(ni->ni_ncpts >= 1);
2207 if (ni->ni_ncpts == 1)
2208 return ni->ni_net->net_tunables.lct_max_tx_credits;
2210 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2211 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2212 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2218 lnet_ni_unlink_locked(struct lnet_ni *ni)
2220 /* move it to zombie list and nobody can find it anymore */
2221 LASSERT(!list_empty(&ni->ni_netlist));
2222 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2223 lnet_ni_decref_locked(ni, 0);
2227 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2232 struct list_head *zombie_list = &net->net_ni_zombie;
2235 * Now wait for the NIs I just nuked to show up on the zombie
2236 * list and shut them down in guaranteed thread context
2239 while (!list_empty(zombie_list)) {
2243 ni = list_entry(zombie_list->next,
2244 struct lnet_ni, ni_netlist);
2245 list_del_init(&ni->ni_netlist);
2246 /* the ni should be in deleting state. If it's not it's
2248 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2249 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2252 /* still busy, add it back to zombie list */
2253 list_add(&ni->ni_netlist, zombie_list);
2257 if (!list_empty(&ni->ni_netlist)) {
2258 /* Unlock mutex while waiting to allow other
2259 * threads to read the LNet state and fall through
2262 lnet_net_unlock(LNET_LOCK_EX);
2263 mutex_unlock(&the_lnet.ln_api_mutex);
2266 if ((i & (-i)) == i) {
2268 "Waiting for zombie LNI %s\n",
2269 libcfs_nid2str(ni->ni_nid));
2271 schedule_timeout_uninterruptible(cfs_time_seconds(1));
2273 mutex_lock(&the_lnet.ln_api_mutex);
2274 lnet_net_lock(LNET_LOCK_EX);
2278 lnet_net_unlock(LNET_LOCK_EX);
2280 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2282 LASSERT(!in_interrupt());
2283 /* Holding the mutex makes it safe for lnd_shutdown
2284 * to call module_put(). Module unload cannot finish
2285 * until lnet_unregister_lnd() completes, and that
2286 * requires the mutex.
2288 mutex_lock(&the_lnet.ln_lnd_mutex);
2289 (net->net_lnd->lnd_shutdown)(ni);
2290 mutex_unlock(&the_lnet.ln_lnd_mutex);
2293 CDEBUG(D_LNI, "Removed LNI %s\n",
2294 libcfs_nid2str(ni->ni_nid));
2298 lnet_net_lock(LNET_LOCK_EX);
2302 /* shutdown down the NI and release refcount */
2304 lnet_shutdown_lndni(struct lnet_ni *ni)
2307 struct lnet_net *net = ni->ni_net;
2309 lnet_net_lock(LNET_LOCK_EX);
2311 ni->ni_state = LNET_NI_STATE_DELETING;
2313 lnet_ni_unlink_locked(ni);
2314 lnet_incr_dlc_seq();
2315 lnet_net_unlock(LNET_LOCK_EX);
2317 /* clear messages for this NI on the lazy portal */
2318 for (i = 0; i < the_lnet.ln_nportals; i++)
2319 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2321 lnet_net_lock(LNET_LOCK_EX);
2322 lnet_clear_zombies_nis_locked(net);
2323 lnet_net_unlock(LNET_LOCK_EX);
2327 lnet_shutdown_lndnet(struct lnet_net *net)
2331 lnet_net_lock(LNET_LOCK_EX);
2333 list_del_init(&net->net_list);
2335 while (!list_empty(&net->net_ni_list)) {
2336 ni = list_entry(net->net_ni_list.next,
2337 struct lnet_ni, ni_netlist);
2338 lnet_net_unlock(LNET_LOCK_EX);
2339 lnet_shutdown_lndni(ni);
2340 lnet_net_lock(LNET_LOCK_EX);
2343 lnet_net_unlock(LNET_LOCK_EX);
2345 /* Do peer table cleanup for this net */
2346 lnet_peer_tables_cleanup(net);
2352 lnet_shutdown_lndnets(void)
2354 struct lnet_net *net;
2356 struct lnet_msg *msg, *tmp;
2358 /* NB called holding the global mutex */
2360 /* All quiet on the API front */
2361 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2362 LASSERT(the_lnet.ln_refcount == 0);
2364 lnet_net_lock(LNET_LOCK_EX);
2365 the_lnet.ln_state = LNET_STATE_STOPPING;
2368 * move the nets to the zombie list to avoid them being
2369 * picked up for new work. LONET is also included in the
2370 * Nets that will be moved to the zombie list
2372 list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2374 /* Drop the cached loopback Net. */
2375 if (the_lnet.ln_loni != NULL) {
2376 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2377 the_lnet.ln_loni = NULL;
2379 lnet_net_unlock(LNET_LOCK_EX);
2381 /* iterate through the net zombie list and delete each net */
2382 while (!list_empty(&the_lnet.ln_net_zombie)) {
2383 net = list_entry(the_lnet.ln_net_zombie.next,
2384 struct lnet_net, net_list);
2385 lnet_shutdown_lndnet(net);
2388 spin_lock(&the_lnet.ln_msg_resend_lock);
2389 list_splice(&the_lnet.ln_msg_resend, &resend);
2390 spin_unlock(&the_lnet.ln_msg_resend_lock);
2392 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2393 list_del_init(&msg->msg_list);
2394 msg->msg_no_resend = true;
2395 lnet_finalize(msg, -ECANCELED);
2398 lnet_net_lock(LNET_LOCK_EX);
2399 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2400 lnet_net_unlock(LNET_LOCK_EX);
2404 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2407 struct lnet_tx_queue *tq;
2409 struct lnet_net *net = ni->ni_net;
2411 mutex_lock(&the_lnet.ln_lnd_mutex);
2414 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2415 ni->ni_lnd_tunables_set = true;
2418 rc = (net->net_lnd->lnd_startup)(ni);
2420 mutex_unlock(&the_lnet.ln_lnd_mutex);
2423 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2424 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2429 ni->ni_state = LNET_NI_STATE_ACTIVE;
2432 /* We keep a reference on the loopback net through the loopback NI */
2433 if (net->net_lnd->lnd_type == LOLND) {
2435 LASSERT(the_lnet.ln_loni == NULL);
2436 the_lnet.ln_loni = ni;
2437 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2438 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2439 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2440 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2444 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2445 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2446 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2447 libcfs_lnd2str(net->net_lnd->lnd_type),
2448 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2450 /* shutdown the NI since if we get here then it must've already
2453 lnet_shutdown_lndni(ni);
2457 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2458 tq->tq_credits_min =
2459 tq->tq_credits_max =
2460 tq->tq_credits = lnet_ni_tq_credits(ni);
2463 atomic_set(&ni->ni_tx_credits,
2464 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2465 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2467 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2468 libcfs_nid2str(ni->ni_nid),
2469 ni->ni_net->net_tunables.lct_peer_tx_credits,
2470 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2471 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2472 ni->ni_net->net_tunables.lct_peer_timeout);
2481 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2484 struct lnet_net *net_l = NULL;
2485 LIST_HEAD(local_ni_list);
2489 const struct lnet_lnd *lnd;
2491 net->net_tunables.lct_peer_timeout;
2493 net->net_tunables.lct_max_tx_credits;
2494 int peerrtrcredits =
2495 net->net_tunables.lct_peer_rtr_credits;
2498 * make sure that this net is unique. If it isn't then
2499 * we are adding interfaces to an already existing network, and
2500 * 'net' is just a convenient way to pass in the list.
2501 * if it is unique we need to find the LND and load it if
2504 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2505 lnd_type = LNET_NETTYP(net->net_id);
2507 mutex_lock(&the_lnet.ln_lnd_mutex);
2508 lnd = lnet_find_lnd_by_type(lnd_type);
2511 mutex_unlock(&the_lnet.ln_lnd_mutex);
2512 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2513 mutex_lock(&the_lnet.ln_lnd_mutex);
2515 lnd = lnet_find_lnd_by_type(lnd_type);
2517 mutex_unlock(&the_lnet.ln_lnd_mutex);
2518 CERROR("Can't load LND %s, module %s, rc=%d\n",
2519 libcfs_lnd2str(lnd_type),
2520 libcfs_lnd2modname(lnd_type), rc);
2521 #ifndef HAVE_MODULE_LOADING_SUPPORT
2522 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2523 "compiled with kernel module "
2524 "loading support.");
2533 mutex_unlock(&the_lnet.ln_lnd_mutex);
2539 * net_l: if the network being added is unique then net_l
2540 * will point to that network
2541 * if the network being added is not unique then
2542 * net_l points to the existing network.
2544 * When we enter the loop below, we'll pick NIs off he
2545 * network beign added and start them up, then add them to
2546 * a local ni list. Once we've successfully started all
2547 * the NIs then we join the local NI list (of started up
2548 * networks) with the net_l->net_ni_list, which should
2549 * point to the correct network to add the new ni list to
2551 * If any of the new NIs fail to start up, then we want to
2552 * iterate through the local ni list, which should include
2553 * any NIs which were successfully started up, and shut
2556 * After than we want to delete the network being added,
2557 * to avoid a memory leak.
2559 while (!list_empty(&net->net_ni_added)) {
2560 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2562 list_del_init(&ni->ni_netlist);
2564 /* make sure that the the NI we're about to start
2565 * up is actually unique. if it's not fail. */
2566 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2567 ni->ni_interface)) {
2572 /* adjust the pointer the parent network, just in case it
2573 * the net is a duplicate */
2576 rc = lnet_startup_lndni(ni, tun);
2582 list_add_tail(&ni->ni_netlist, &local_ni_list);
2587 lnet_net_lock(LNET_LOCK_EX);
2588 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2589 lnet_incr_dlc_seq();
2590 lnet_net_unlock(LNET_LOCK_EX);
2592 /* if the network is not unique then we don't want to keep
2593 * it around after we're done. Free it. Otherwise add that
2594 * net to the global the_lnet.ln_nets */
2595 if (net_l != net && net_l != NULL) {
2597 * TODO - note. currently the tunables can not be updated
2603 * restore tunables after it has been overwitten by the
2606 if (peer_timeout != -1)
2607 net->net_tunables.lct_peer_timeout = peer_timeout;
2608 if (maxtxcredits != -1)
2609 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2610 if (peerrtrcredits != -1)
2611 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2613 lnet_net_lock(LNET_LOCK_EX);
2614 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2615 lnet_net_unlock(LNET_LOCK_EX);
2618 /* update net count */
2619 lnet_current_net_count = lnet_get_net_count();
2625 * shutdown the new NIs that are being started up
2626 * free the NET being started
2628 while (!list_empty(&local_ni_list)) {
2629 ni = list_entry(local_ni_list.next, struct lnet_ni,
2632 lnet_shutdown_lndni(ni);
2642 lnet_startup_lndnets(struct list_head *netlist)
2644 struct lnet_net *net;
2649 * Change to running state before bringing up the LNDs. This
2650 * allows lnet_shutdown_lndnets() to assert that we've passed
2653 lnet_net_lock(LNET_LOCK_EX);
2654 the_lnet.ln_state = LNET_STATE_RUNNING;
2655 lnet_net_unlock(LNET_LOCK_EX);
2657 while (!list_empty(netlist)) {
2658 net = list_entry(netlist->next, struct lnet_net, net_list);
2659 list_del_init(&net->net_list);
2661 rc = lnet_startup_lndnet(net, NULL);
2671 lnet_shutdown_lndnets();
2677 * Initialize LNet library.
2679 * Automatically called at module loading time. Caller has to call
2680 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2681 * latter returned 0. It must be called exactly once.
2683 * \retval 0 on success
2684 * \retval -ve on failures.
2686 int lnet_lib_init(void)
2690 lnet_assert_wire_constants();
2692 /* refer to global cfs_cpt_table for now */
2693 the_lnet.ln_cpt_table = cfs_cpt_tab;
2694 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2696 LASSERT(the_lnet.ln_cpt_number > 0);
2697 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2698 /* we are under risk of consuming all lh_cookie */
2699 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2700 "please change setting of CPT-table and retry\n",
2701 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2705 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2706 the_lnet.ln_cpt_bits++;
2708 rc = lnet_create_locks();
2710 CERROR("Can't create LNet global locks: %d\n", rc);
2714 the_lnet.ln_refcount = 0;
2715 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2716 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2718 /* The hash table size is the number of bits it takes to express the set
2719 * ln_num_routes, minus 1 (better to under estimate than over so we
2720 * don't waste memory). */
2721 if (rnet_htable_size <= 0)
2722 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2723 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2724 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2725 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2726 order_base_2(rnet_htable_size) - 1);
2728 /* All LNDs apart from the LOLND are in separate modules. They
2729 * register themselves when their module loads, and unregister
2730 * themselves when their module is unloaded. */
2731 lnet_register_lnd(&the_lolnd);
2736 * Finalize LNet library.
2738 * \pre lnet_lib_init() called with success.
2739 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2741 * As this happens at module-unload, all lnds must already be unloaded,
2742 * so they must already be unregistered.
2744 void lnet_lib_exit(void)
2748 LASSERT(the_lnet.ln_refcount == 0);
2749 lnet_unregister_lnd(&the_lolnd);
2750 for (i = 0; i < NUM_LNDS; i++)
2751 LASSERT(!the_lnet.ln_lnds[i]);
2752 lnet_destroy_locks();
2756 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2758 * Users must call this function at least once before any other functions.
2759 * For each successful call there must be a corresponding call to
2760 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2763 * The PID used by LNet may be different from the one requested.
2766 * \param requested_pid PID requested by the caller.
2768 * \return >= 0 on success, and < 0 error code on failures.
2771 LNetNIInit(lnet_pid_t requested_pid)
2773 int im_a_router = 0;
2776 struct lnet_ping_buffer *pbuf;
2777 struct lnet_handle_md ping_mdh;
2778 LIST_HEAD(net_head);
2779 struct lnet_net *net;
2781 mutex_lock(&the_lnet.ln_api_mutex);
2783 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2785 if (the_lnet.ln_refcount > 0) {
2786 rc = the_lnet.ln_refcount++;
2787 mutex_unlock(&the_lnet.ln_api_mutex);
2791 rc = lnet_prepare(requested_pid);
2793 mutex_unlock(&the_lnet.ln_api_mutex);
2797 /* create a network for Loopback network */
2798 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2801 goto err_empty_list;
2804 /* Add in the loopback NI */
2805 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2807 goto err_empty_list;
2810 if (use_tcp_bonding)
2811 CWARN("use_tcp_bonding has been removed. Use Multi-Rail and Dynamic Discovery instead, see LU-13641\n");
2813 /* If LNet is being initialized via DLC it is possible
2814 * that the user requests not to load module parameters (ones which
2815 * are supported by DLC) on initialization. Therefore, make sure not
2816 * to load networks, routes and forwarding from module parameters
2817 * in this case. On cleanup in case of failure only clean up
2818 * routes if it has been loaded */
2819 if (!the_lnet.ln_nis_from_mod_params) {
2820 rc = lnet_parse_networks(&net_head, lnet_get_networks());
2822 goto err_empty_list;
2825 ni_count = lnet_startup_lndnets(&net_head);
2828 goto err_empty_list;
2831 if (!the_lnet.ln_nis_from_mod_params) {
2832 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2834 goto err_shutdown_lndnis;
2836 rc = lnet_rtrpools_alloc(im_a_router);
2838 goto err_destroy_routes;
2841 rc = lnet_acceptor_start();
2843 goto err_destroy_routes;
2845 the_lnet.ln_refcount = 1;
2846 /* Now I may use my own API functions... */
2848 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2850 goto err_acceptor_stop;
2852 lnet_ping_target_update(pbuf, ping_mdh);
2854 the_lnet.ln_mt_handler = lnet_mt_event_handler;
2856 rc = lnet_push_target_init();
2860 rc = lnet_peer_discovery_start();
2862 goto err_destroy_push_target;
2864 rc = lnet_monitor_thr_start();
2866 goto err_stop_discovery_thr;
2869 lnet_router_debugfs_init();
2871 mutex_unlock(&the_lnet.ln_api_mutex);
2873 complete_all(&the_lnet.ln_started);
2875 /* wait for all routers to start */
2876 lnet_wait_router_start();
2880 err_stop_discovery_thr:
2881 lnet_peer_discovery_stop();
2882 err_destroy_push_target:
2883 lnet_push_target_fini();
2885 lnet_ping_target_fini();
2887 the_lnet.ln_refcount = 0;
2888 lnet_acceptor_stop();
2890 if (!the_lnet.ln_nis_from_mod_params)
2891 lnet_destroy_routes();
2892 err_shutdown_lndnis:
2893 lnet_shutdown_lndnets();
2897 mutex_unlock(&the_lnet.ln_api_mutex);
2898 while (!list_empty(&net_head)) {
2899 struct lnet_net *net;
2901 net = list_entry(net_head.next, struct lnet_net, net_list);
2902 list_del_init(&net->net_list);
2907 EXPORT_SYMBOL(LNetNIInit);
2910 * Stop LNet interfaces, routing, and forwarding.
2912 * Users must call this function once for each successful call to LNetNIInit().
2913 * Once the LNetNIFini() operation has been started, the results of pending
2914 * API operations are undefined.
2916 * \return always 0 for current implementation.
2921 mutex_lock(&the_lnet.ln_api_mutex);
2923 LASSERT(the_lnet.ln_refcount > 0);
2925 if (the_lnet.ln_refcount != 1) {
2926 the_lnet.ln_refcount--;
2928 LASSERT(!the_lnet.ln_niinit_self);
2932 lnet_router_debugfs_fini();
2933 lnet_monitor_thr_stop();
2934 lnet_peer_discovery_stop();
2935 lnet_push_target_fini();
2936 lnet_ping_target_fini();
2938 /* Teardown fns that use my own API functions BEFORE here */
2939 the_lnet.ln_refcount = 0;
2941 lnet_acceptor_stop();
2942 lnet_destroy_routes();
2943 lnet_shutdown_lndnets();
2947 mutex_unlock(&the_lnet.ln_api_mutex);
2950 EXPORT_SYMBOL(LNetNIFini);
2953 * Grabs the ni data from the ni structure and fills the out
2956 * \param[in] ni network interface structure
2957 * \param[out] cfg_ni NI config information
2958 * \param[out] tun network and LND tunables
2961 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2962 struct lnet_ioctl_config_lnd_tunables *tun,
2963 struct lnet_ioctl_element_stats *stats,
2966 size_t min_size = 0;
2969 if (!ni || !cfg_ni || !tun)
2972 if (ni->ni_interface != NULL) {
2973 strncpy(cfg_ni->lic_ni_intf,
2975 sizeof(cfg_ni->lic_ni_intf));
2978 cfg_ni->lic_nid = ni->ni_nid;
2979 if (ni->ni_nid == LNET_NID_LO_0)
2980 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2982 cfg_ni->lic_status = ni->ni_status->ns_status;
2983 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2985 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2988 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2989 LNET_STATS_TYPE_SEND);
2990 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2991 LNET_STATS_TYPE_RECV);
2992 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2993 LNET_STATS_TYPE_DROP);
2997 * tun->lt_tun will always be present, but in order to be
2998 * backwards compatible, we need to deal with the cases when
2999 * tun->lt_tun is smaller than what the kernel has, because it
3000 * comes from an older version of a userspace program, then we'll
3001 * need to copy as much information as we have available space.
3003 min_size = tun_size - sizeof(tun->lt_cmn);
3004 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
3006 /* copy over the cpts */
3007 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
3008 ni->ni_cpts == NULL) {
3009 for (i = 0; i < ni->ni_ncpts; i++)
3010 cfg_ni->lic_cpts[i] = i;
3013 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
3014 i < LNET_MAX_SHOW_NUM_CPT;
3016 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
3018 cfg_ni->lic_ncpts = ni->ni_ncpts;
3022 * NOTE: This is a legacy function left in the code to be backwards
3023 * compatible with older userspace programs. It should eventually be
3026 * Grabs the ni data from the ni structure and fills the out
3029 * \param[in] ni network interface structure
3030 * \param[out] config config information
3033 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
3034 struct lnet_ioctl_config_data *config)
3036 struct lnet_ioctl_net_config *net_config;
3037 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
3038 size_t min_size, tunable_size = 0;
3044 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
3048 if (!ni->ni_interface)
3051 strncpy(net_config->ni_interface,
3053 sizeof(net_config->ni_interface));
3055 config->cfg_nid = ni->ni_nid;
3056 config->cfg_config_u.cfg_net.net_peer_timeout =
3057 ni->ni_net->net_tunables.lct_peer_timeout;
3058 config->cfg_config_u.cfg_net.net_max_tx_credits =
3059 ni->ni_net->net_tunables.lct_max_tx_credits;
3060 config->cfg_config_u.cfg_net.net_peer_tx_credits =
3061 ni->ni_net->net_tunables.lct_peer_tx_credits;
3062 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
3063 ni->ni_net->net_tunables.lct_peer_rtr_credits;
3065 if (ni->ni_nid == LNET_NID_LO_0)
3066 net_config->ni_status = LNET_NI_STATUS_UP;
3068 net_config->ni_status = ni->ni_status->ns_status;
3071 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
3073 for (i = 0; i < num_cpts; i++)
3074 net_config->ni_cpts[i] = ni->ni_cpts[i];
3076 config->cfg_ncpts = num_cpts;
3080 * See if user land tools sent in a newer and larger version
3081 * of struct lnet_tunables than what the kernel uses.
3083 min_size = sizeof(*config) + sizeof(*net_config);
3085 if (config->cfg_hdr.ioc_len > min_size)
3086 tunable_size = config->cfg_hdr.ioc_len - min_size;
3088 /* Don't copy too much data to user space */
3089 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
3090 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
3092 if (lnd_cfg && min_size) {
3093 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
3094 config->cfg_config_u.cfg_net.net_interface_count = 1;
3096 /* Tell user land that kernel side has less data */
3097 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
3098 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
3099 config->cfg_hdr.ioc_len -= min_size;
3105 lnet_get_ni_idx_locked(int idx)
3108 struct lnet_net *net;
3110 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3111 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3120 int lnet_get_net_healthv_locked(struct lnet_net *net)
3123 int best_healthv = 0;
3126 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3127 healthv = atomic_read(&ni->ni_healthv);
3128 if (healthv > best_healthv)
3129 best_healthv = healthv;
3132 return best_healthv;
3136 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
3139 struct lnet_net *net = mynet;
3142 * It is possible that the net has been cleaned out while there is
3143 * a message being sent. This function accessed the net without
3144 * checking if the list is empty
3148 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3150 if (list_empty(&net->net_ni_list))
3152 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3158 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3159 /* if you reached the end of the ni list and the net is
3160 * specified, then there are no more nis in that net */
3164 /* we reached the end of this net ni list. move to the
3166 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3167 /* no more nets and no more NIs. */
3170 /* get the next net */
3171 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3173 if (list_empty(&net->net_ni_list))
3175 /* get the ni on it */
3176 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3182 if (list_empty(&prev->ni_netlist))
3185 /* there are more nis left */
3186 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3192 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3197 int idx = config->cfg_count;
3199 cpt = lnet_net_lock_current();
3201 ni = lnet_get_ni_idx_locked(idx);
3206 lnet_fill_ni_info_legacy(ni, config);
3210 lnet_net_unlock(cpt);
3215 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3216 struct lnet_ioctl_config_lnd_tunables *tun,
3217 struct lnet_ioctl_element_stats *stats,
3224 if (!cfg_ni || !tun || !stats)
3227 cpt = lnet_net_lock_current();
3229 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3234 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3238 lnet_net_unlock(cpt);
3242 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3251 cpt = lnet_net_lock_current();
3253 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3256 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3260 lnet_net_unlock(cpt);
3265 static int lnet_add_net_common(struct lnet_net *net,
3266 struct lnet_ioctl_config_lnd_tunables *tun)
3268 struct lnet_handle_md ping_mdh;
3269 struct lnet_ping_buffer *pbuf;
3270 struct lnet_remotenet *rnet;
3276 lnet_net_lock(LNET_LOCK_EX);
3277 rnet = lnet_find_rnet_locked(net->net_id);
3278 lnet_net_unlock(LNET_LOCK_EX);
3280 * make sure that the net added doesn't invalidate the current
3281 * configuration LNet is keeping
3284 CERROR("Adding net %s will invalidate routing configuration\n",
3285 libcfs_net2str(net->net_id));
3291 * make sure you calculate the correct number of slots in the ping
3292 * buffer. Since the ping info is a flattened list of all the NIs,
3293 * we should allocate enough slots to accomodate the number of NIs
3294 * which will be added.
3296 * since ni hasn't been configured yet, use
3297 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3299 net_ni_count = lnet_get_net_ni_count_pre(net);
3301 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3302 net_ni_count + lnet_get_ni_count(),
3310 memcpy(&net->net_tunables,
3311 &tun->lt_cmn, sizeof(net->net_tunables));
3313 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3315 net_id = net->net_id;
3317 rc = lnet_startup_lndnet(net,
3318 (tun) ? &tun->lt_tun : NULL);
3322 lnet_net_lock(LNET_LOCK_EX);
3323 net = lnet_get_net_locked(net_id);
3326 /* apply the UDSPs */
3327 rc = lnet_udsp_apply_policies_on_net(net);
3329 CERROR("Failed to apply UDSPs on local net %s\n",
3330 libcfs_net2str(net->net_id));
3332 /* At this point we lost track of which NI was just added, so we
3333 * just re-apply the policies on all of the NIs on this net
3335 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3336 rc = lnet_udsp_apply_policies_on_ni(ni);
3338 CERROR("Failed to apply UDSPs on ni %s\n",
3339 libcfs_nid2str(ni->ni_nid));
3341 lnet_net_unlock(LNET_LOCK_EX);
3344 * Start the acceptor thread if this is the first network
3345 * being added that requires the thread.
3347 if (net->net_lnd->lnd_accept) {
3348 rc = lnet_acceptor_start();
3350 /* shutdown the net that we just started */
3351 CERROR("Failed to start up acceptor thread\n");
3352 lnet_shutdown_lndnet(net);
3357 lnet_net_lock(LNET_LOCK_EX);
3358 lnet_peer_net_added(net);
3359 lnet_net_unlock(LNET_LOCK_EX);
3361 lnet_ping_target_update(pbuf, ping_mdh);
3366 lnet_ping_md_unlink(pbuf, &ping_mdh);
3367 lnet_ping_buffer_decref(pbuf);
3372 lnet_set_tune_defaults(struct lnet_ioctl_config_lnd_tunables *tun)
3375 if (!tun->lt_cmn.lct_peer_timeout)
3376 tun->lt_cmn.lct_peer_timeout = DEFAULT_PEER_TIMEOUT;
3377 if (!tun->lt_cmn.lct_peer_tx_credits)
3378 tun->lt_cmn.lct_peer_tx_credits = DEFAULT_PEER_CREDITS;
3379 if (!tun->lt_cmn.lct_max_tx_credits)
3380 tun->lt_cmn.lct_max_tx_credits = DEFAULT_CREDITS;
3384 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3385 struct lnet_ioctl_config_lnd_tunables *tun)
3387 struct lnet_net *net;
3390 LIST_HEAD(net_head);
3392 rc = lnet_parse_ip2nets(&nets, ip2nets);
3396 rc = lnet_parse_networks(&net_head, nets);
3400 lnet_set_tune_defaults(tun);
3402 mutex_lock(&the_lnet.ln_api_mutex);
3403 while (!list_empty(&net_head)) {
3404 net = list_entry(net_head.next, struct lnet_net, net_list);
3405 list_del_init(&net->net_list);
3406 rc = lnet_add_net_common(net, tun);
3412 mutex_unlock(&the_lnet.ln_api_mutex);
3414 while (!list_empty(&net_head)) {
3415 net = list_entry(net_head.next, struct lnet_net, net_list);
3416 list_del_init(&net->net_list);
3422 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3424 struct lnet_net *net;
3426 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3428 __u32 net_id, lnd_type;
3430 /* get the tunables if they are available */
3431 if (conf->lic_cfg_hdr.ioc_len >=
3432 sizeof(*conf) + sizeof(*tun))
3433 tun = (struct lnet_ioctl_config_lnd_tunables *)
3436 /* handle legacy ip2nets from DLC */
3437 if (conf->lic_legacy_ip2nets[0] != '\0')
3438 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3441 net_id = LNET_NIDNET(conf->lic_nid);
3442 lnd_type = LNET_NETTYP(net_id);
3444 if (!libcfs_isknown_lnd(lnd_type)) {
3445 CERROR("No valid net and lnd information provided\n");
3449 net = lnet_net_alloc(net_id, NULL);
3453 for (i = 0; i < conf->lic_ncpts; i++) {
3454 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3458 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3463 lnet_set_tune_defaults(tun);
3465 mutex_lock(&the_lnet.ln_api_mutex);
3467 rc = lnet_add_net_common(net, tun);
3469 mutex_unlock(&the_lnet.ln_api_mutex);
3474 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3476 struct lnet_net *net;
3478 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3479 struct lnet_ping_buffer *pbuf;
3480 struct lnet_handle_md ping_mdh;
3485 /* don't allow userspace to shutdown the LOLND */
3486 if (LNET_NETTYP(net_id) == LOLND)
3489 mutex_lock(&the_lnet.ln_api_mutex);
3493 net = lnet_get_net_locked(net_id);
3495 CERROR("net %s not found\n",
3496 libcfs_net2str(net_id));
3501 addr = LNET_NIDADDR(conf->lic_nid);
3503 /* remove the entire net */
3504 net_count = lnet_get_net_ni_count_locked(net);
3508 /* create and link a new ping info, before removing the old one */
3509 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3510 lnet_get_ni_count() - net_count,
3513 goto unlock_api_mutex;
3515 lnet_shutdown_lndnet(net);
3517 lnet_acceptor_stop();
3519 lnet_ping_target_update(pbuf, ping_mdh);
3521 goto unlock_api_mutex;
3524 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3526 CERROR("nid %s not found\n",
3527 libcfs_nid2str(conf->lic_nid));
3532 net_count = lnet_get_net_ni_count_locked(net);
3536 /* create and link a new ping info, before removing the old one */
3537 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3538 lnet_get_ni_count() - 1, false);
3540 goto unlock_api_mutex;
3542 lnet_shutdown_lndni(ni);
3544 lnet_acceptor_stop();
3546 lnet_ping_target_update(pbuf, ping_mdh);
3548 /* check if the net is empty and remove it if it is */
3550 lnet_shutdown_lndnet(net);
3552 goto unlock_api_mutex;
3557 mutex_unlock(&the_lnet.ln_api_mutex);
3563 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3564 * They are only expected to be called for unique networks.
3565 * That can be as a result of older DLC library
3566 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3569 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3571 struct lnet_net *net;
3572 LIST_HEAD(net_head);
3574 struct lnet_ioctl_config_lnd_tunables tun;
3575 const char *nets = conf->cfg_config_u.cfg_net.net_intf;
3577 /* Create a net/ni structures for the network string */
3578 rc = lnet_parse_networks(&net_head, nets);
3580 return rc == 0 ? -EINVAL : rc;
3582 mutex_lock(&the_lnet.ln_api_mutex);
3585 rc = -EINVAL; /* only add one network per call */
3586 goto out_unlock_clean;
3589 net = list_entry(net_head.next, struct lnet_net, net_list);
3590 list_del_init(&net->net_list);
3592 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3594 memset(&tun, 0, sizeof(tun));
3596 tun.lt_cmn.lct_peer_timeout =
3597 (!conf->cfg_config_u.cfg_net.net_peer_timeout) ? DEFAULT_PEER_TIMEOUT :
3598 conf->cfg_config_u.cfg_net.net_peer_timeout;
3599 tun.lt_cmn.lct_peer_tx_credits =
3600 (!conf->cfg_config_u.cfg_net.net_peer_tx_credits) ? DEFAULT_PEER_CREDITS :
3601 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3602 tun.lt_cmn.lct_peer_rtr_credits =
3603 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3604 tun.lt_cmn.lct_max_tx_credits =
3605 (!conf->cfg_config_u.cfg_net.net_max_tx_credits) ? DEFAULT_CREDITS :
3606 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3608 rc = lnet_add_net_common(net, &tun);
3611 mutex_unlock(&the_lnet.ln_api_mutex);
3612 while (!list_empty(&net_head)) {
3613 /* net_head list is empty in success case */
3614 net = list_entry(net_head.next, struct lnet_net, net_list);
3615 list_del_init(&net->net_list);
3622 lnet_dyn_del_net(__u32 net_id)
3624 struct lnet_net *net;
3625 struct lnet_ping_buffer *pbuf;
3626 struct lnet_handle_md ping_mdh;
3630 /* don't allow userspace to shutdown the LOLND */
3631 if (LNET_NETTYP(net_id) == LOLND)
3634 mutex_lock(&the_lnet.ln_api_mutex);
3638 net = lnet_get_net_locked(net_id);
3645 net_ni_count = lnet_get_net_ni_count_locked(net);
3649 /* create and link a new ping info, before removing the old one */
3650 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3651 lnet_get_ni_count() - net_ni_count, false);
3655 lnet_shutdown_lndnet(net);
3657 lnet_acceptor_stop();
3659 lnet_ping_target_update(pbuf, ping_mdh);
3662 mutex_unlock(&the_lnet.ln_api_mutex);
3667 void lnet_incr_dlc_seq(void)
3669 atomic_inc(&lnet_dlc_seq_no);
3672 __u32 lnet_get_dlc_seq_locked(void)
3674 return atomic_read(&lnet_dlc_seq_no);
3678 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3680 struct lnet_net *net;
3683 lnet_net_lock(LNET_LOCK_EX);
3684 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3685 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3686 if (ni->ni_nid == nid || all) {
3687 atomic_set(&ni->ni_healthv, value);
3688 if (list_empty(&ni->ni_recovery) &&
3689 value < LNET_MAX_HEALTH_VALUE) {
3690 CERROR("manually adding local NI %s to recovery\n",
3691 libcfs_nid2str(ni->ni_nid));
3692 list_add_tail(&ni->ni_recovery,
3693 &the_lnet.ln_mt_localNIRecovq);
3694 lnet_ni_addref_locked(ni, 0);
3697 lnet_net_unlock(LNET_LOCK_EX);
3703 lnet_net_unlock(LNET_LOCK_EX);
3707 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3711 lnet_nid_t nid = stats->hlni_nid;
3713 cpt = lnet_net_lock_current();
3714 ni = lnet_nid2ni_locked(nid, cpt);
3721 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3722 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3723 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3724 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3725 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3726 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3727 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3730 lnet_net_unlock(cpt);
3736 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3741 lnet_net_lock(LNET_LOCK_EX);
3742 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3743 list->rlst_nid_array[i] = ni->ni_nid;
3745 if (i >= LNET_MAX_SHOW_NUM_NID)
3748 lnet_net_unlock(LNET_LOCK_EX);
3749 list->rlst_num_nids = i;
3755 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3757 struct lnet_peer_ni *lpni;
3760 lnet_net_lock(LNET_LOCK_EX);
3761 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3762 list->rlst_nid_array[i] = lpni->lpni_nid;
3764 if (i >= LNET_MAX_SHOW_NUM_NID)
3767 lnet_net_unlock(LNET_LOCK_EX);
3768 list->rlst_num_nids = i;
3774 * LNet ioctl handler.
3778 LNetCtl(unsigned int cmd, void *arg)
3780 struct libcfs_ioctl_data *data = arg;
3781 struct lnet_ioctl_config_data *config;
3782 struct lnet_process_id id = {0};
3786 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3787 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3790 case IOC_LIBCFS_GET_NI:
3791 rc = LNetGetId(data->ioc_count, &id);
3792 data->ioc_nid = id.nid;
3795 case IOC_LIBCFS_FAIL_NID:
3796 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3798 case IOC_LIBCFS_ADD_ROUTE: {
3799 /* default router sensitivity to 1 */
3800 unsigned int sensitivity = 1;
3803 if (config->cfg_hdr.ioc_len < sizeof(*config))
3806 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3808 config->cfg_config_u.cfg_route.rtr_sensitivity;
3811 mutex_lock(&the_lnet.ln_api_mutex);
3812 rc = lnet_add_route(config->cfg_net,
3813 config->cfg_config_u.cfg_route.rtr_hop,
3815 config->cfg_config_u.cfg_route.
3816 rtr_priority, sensitivity);
3817 mutex_unlock(&the_lnet.ln_api_mutex);
3821 case IOC_LIBCFS_DEL_ROUTE:
3824 if (config->cfg_hdr.ioc_len < sizeof(*config))
3827 mutex_lock(&the_lnet.ln_api_mutex);
3828 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3829 mutex_unlock(&the_lnet.ln_api_mutex);
3832 case IOC_LIBCFS_GET_ROUTE:
3835 if (config->cfg_hdr.ioc_len < sizeof(*config))
3838 mutex_lock(&the_lnet.ln_api_mutex);
3839 rc = lnet_get_route(config->cfg_count,
3841 &config->cfg_config_u.cfg_route.rtr_hop,
3843 &config->cfg_config_u.cfg_route.rtr_flags,
3844 &config->cfg_config_u.cfg_route.
3846 &config->cfg_config_u.cfg_route.
3848 mutex_unlock(&the_lnet.ln_api_mutex);
3851 case IOC_LIBCFS_GET_LOCAL_NI: {
3852 struct lnet_ioctl_config_ni *cfg_ni;
3853 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3854 struct lnet_ioctl_element_stats *stats;
3859 /* get the tunables if they are available */
3860 if (cfg_ni->lic_cfg_hdr.ioc_len <
3861 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3864 stats = (struct lnet_ioctl_element_stats *)
3866 tun = (struct lnet_ioctl_config_lnd_tunables *)
3867 (cfg_ni->lic_bulk + sizeof(*stats));
3869 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3872 mutex_lock(&the_lnet.ln_api_mutex);
3873 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3874 mutex_unlock(&the_lnet.ln_api_mutex);
3878 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3879 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3881 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3884 mutex_lock(&the_lnet.ln_api_mutex);
3885 rc = lnet_get_ni_stats(msg_stats);
3886 mutex_unlock(&the_lnet.ln_api_mutex);
3891 case IOC_LIBCFS_GET_NET: {
3892 size_t total = sizeof(*config) +
3893 sizeof(struct lnet_ioctl_net_config);
3896 if (config->cfg_hdr.ioc_len < total)
3899 mutex_lock(&the_lnet.ln_api_mutex);
3900 rc = lnet_get_net_config(config);
3901 mutex_unlock(&the_lnet.ln_api_mutex);
3905 case IOC_LIBCFS_GET_LNET_STATS:
3907 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3909 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3912 mutex_lock(&the_lnet.ln_api_mutex);
3913 rc = lnet_counters_get(&lnet_stats->st_cntrs);
3914 mutex_unlock(&the_lnet.ln_api_mutex);
3918 case IOC_LIBCFS_CONFIG_RTR:
3921 if (config->cfg_hdr.ioc_len < sizeof(*config))
3924 mutex_lock(&the_lnet.ln_api_mutex);
3925 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3926 rc = lnet_rtrpools_enable();
3927 mutex_unlock(&the_lnet.ln_api_mutex);
3930 lnet_rtrpools_disable();
3931 mutex_unlock(&the_lnet.ln_api_mutex);
3934 case IOC_LIBCFS_ADD_BUF:
3937 if (config->cfg_hdr.ioc_len < sizeof(*config))
3940 mutex_lock(&the_lnet.ln_api_mutex);
3941 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3943 config->cfg_config_u.cfg_buffers.
3945 config->cfg_config_u.cfg_buffers.
3947 mutex_unlock(&the_lnet.ln_api_mutex);
3950 case IOC_LIBCFS_SET_NUMA_RANGE: {
3951 struct lnet_ioctl_set_value *numa;
3953 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3955 lnet_net_lock(LNET_LOCK_EX);
3956 lnet_numa_range = numa->sv_value;
3957 lnet_net_unlock(LNET_LOCK_EX);
3961 case IOC_LIBCFS_GET_NUMA_RANGE: {
3962 struct lnet_ioctl_set_value *numa;
3964 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3966 numa->sv_value = lnet_numa_range;
3970 case IOC_LIBCFS_GET_BUF: {
3971 struct lnet_ioctl_pool_cfg *pool_cfg;
3972 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3976 if (config->cfg_hdr.ioc_len < total)
3979 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3981 mutex_lock(&the_lnet.ln_api_mutex);
3982 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3983 mutex_unlock(&the_lnet.ln_api_mutex);
3987 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3988 struct lnet_ioctl_local_ni_hstats *stats = arg;
3990 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3993 mutex_lock(&the_lnet.ln_api_mutex);
3994 rc = lnet_get_local_ni_hstats(stats);
3995 mutex_unlock(&the_lnet.ln_api_mutex);
4000 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
4001 struct lnet_ioctl_recovery_list *list = arg;
4002 if (list->rlst_hdr.ioc_len < sizeof(*list))
4005 mutex_lock(&the_lnet.ln_api_mutex);
4006 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
4007 rc = lnet_get_local_ni_recovery_list(list);
4009 rc = lnet_get_peer_ni_recovery_list(list);
4010 mutex_unlock(&the_lnet.ln_api_mutex);
4014 case IOC_LIBCFS_ADD_PEER_NI: {
4015 struct lnet_ioctl_peer_cfg *cfg = arg;
4017 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4020 mutex_lock(&the_lnet.ln_api_mutex);
4021 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
4024 mutex_unlock(&the_lnet.ln_api_mutex);
4028 case IOC_LIBCFS_DEL_PEER_NI: {
4029 struct lnet_ioctl_peer_cfg *cfg = arg;
4031 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4034 mutex_lock(&the_lnet.ln_api_mutex);
4035 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
4036 cfg->prcfg_cfg_nid);
4037 mutex_unlock(&the_lnet.ln_api_mutex);
4041 case IOC_LIBCFS_GET_PEER_INFO: {
4042 struct lnet_ioctl_peer *peer_info = arg;
4044 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
4047 mutex_lock(&the_lnet.ln_api_mutex);
4048 rc = lnet_get_peer_ni_info(
4049 peer_info->pr_count,
4051 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
4052 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
4053 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
4054 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
4055 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
4056 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
4057 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
4058 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
4059 mutex_unlock(&the_lnet.ln_api_mutex);
4063 case IOC_LIBCFS_GET_PEER_NI: {
4064 struct lnet_ioctl_peer_cfg *cfg = arg;
4066 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4069 mutex_lock(&the_lnet.ln_api_mutex);
4070 rc = lnet_get_peer_info(cfg,
4071 (void __user *)cfg->prcfg_bulk);
4072 mutex_unlock(&the_lnet.ln_api_mutex);
4076 case IOC_LIBCFS_GET_PEER_LIST: {
4077 struct lnet_ioctl_peer_cfg *cfg = arg;
4079 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4082 mutex_lock(&the_lnet.ln_api_mutex);
4083 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
4084 (struct lnet_process_id __user *)cfg->prcfg_bulk);
4085 mutex_unlock(&the_lnet.ln_api_mutex);
4089 case IOC_LIBCFS_SET_HEALHV: {
4090 struct lnet_ioctl_reset_health_cfg *cfg = arg;
4092 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
4094 if (cfg->rh_value < 0 ||
4095 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
4096 value = LNET_MAX_HEALTH_VALUE;
4098 value = cfg->rh_value;
4099 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
4100 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
4101 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
4102 mutex_lock(&the_lnet.ln_api_mutex);
4103 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
4104 lnet_ni_set_healthv(cfg->rh_nid, value,
4107 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
4109 mutex_unlock(&the_lnet.ln_api_mutex);
4113 case IOC_LIBCFS_NOTIFY_ROUTER: {
4114 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
4116 /* The deadline passed in by the user should be some time in
4117 * seconds in the future since the UNIX epoch. We have to map
4118 * that deadline to the wall clock.
4120 deadline += ktime_get_seconds();
4121 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
4125 case IOC_LIBCFS_LNET_DIST:
4126 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
4127 if (rc < 0 && rc != -EHOSTUNREACH)
4130 data->ioc_u32[0] = rc;
4133 case IOC_LIBCFS_TESTPROTOCOMPAT:
4134 the_lnet.ln_testprotocompat = data->ioc_flags;
4137 case IOC_LIBCFS_LNET_FAULT:
4138 return lnet_fault_ctl(data->ioc_flags, data);
4140 case IOC_LIBCFS_PING: {
4141 signed long timeout;
4143 id.nid = data->ioc_nid;
4144 id.pid = data->ioc_u32[0];
4146 /* If timeout is negative then set default of 3 minutes */
4147 if (((s32)data->ioc_u32[1] <= 0) ||
4148 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4149 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4151 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
4153 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
4154 data->ioc_plen1 / sizeof(struct lnet_process_id));
4159 data->ioc_count = rc;
4163 case IOC_LIBCFS_PING_PEER: {
4164 struct lnet_ioctl_ping_data *ping = arg;
4165 struct lnet_peer *lp;
4166 signed long timeout;
4168 /* If timeout is negative then set default of 3 minutes */
4169 if (((s32)ping->op_param) <= 0 ||
4170 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4171 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4173 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
4175 rc = lnet_ping(ping->ping_id, timeout,
4181 mutex_lock(&the_lnet.ln_api_mutex);
4182 lp = lnet_find_peer(ping->ping_id.nid);
4184 ping->ping_id.nid = lp->lp_primary_nid;
4185 ping->mr_info = lnet_peer_is_multi_rail(lp);
4186 lnet_peer_decref_locked(lp);
4188 mutex_unlock(&the_lnet.ln_api_mutex);
4190 ping->ping_count = rc;
4194 case IOC_LIBCFS_DISCOVER: {
4195 struct lnet_ioctl_ping_data *discover = arg;
4196 struct lnet_peer *lp;
4198 rc = lnet_discover(discover->ping_id, discover->op_param,
4200 discover->ping_count);
4204 mutex_lock(&the_lnet.ln_api_mutex);
4205 lp = lnet_find_peer(discover->ping_id.nid);
4207 discover->ping_id.nid = lp->lp_primary_nid;
4208 discover->mr_info = lnet_peer_is_multi_rail(lp);
4209 lnet_peer_decref_locked(lp);
4211 mutex_unlock(&the_lnet.ln_api_mutex);
4213 discover->ping_count = rc;
4217 case IOC_LIBCFS_ADD_UDSP: {
4218 struct lnet_ioctl_udsp *ioc_udsp = arg;
4219 __u32 bulk_size = ioc_udsp->iou_hdr.ioc_len;
4221 mutex_lock(&the_lnet.ln_api_mutex);
4222 rc = lnet_udsp_demarshal_add(arg, bulk_size);
4224 rc = lnet_udsp_apply_policies(NULL, false);
4225 CDEBUG(D_NET, "policy application returned %d\n", rc);
4228 mutex_unlock(&the_lnet.ln_api_mutex);
4233 case IOC_LIBCFS_DEL_UDSP: {
4234 struct lnet_ioctl_udsp *ioc_udsp = arg;
4235 int idx = ioc_udsp->iou_idx;
4237 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4240 mutex_lock(&the_lnet.ln_api_mutex);
4241 rc = lnet_udsp_del_policy(idx);
4243 rc = lnet_udsp_apply_policies(NULL, false);
4244 CDEBUG(D_NET, "policy re-application returned %d\n",
4248 mutex_unlock(&the_lnet.ln_api_mutex);
4253 case IOC_LIBCFS_GET_UDSP_SIZE: {
4254 struct lnet_ioctl_udsp *ioc_udsp = arg;
4255 struct lnet_udsp *udsp;
4257 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4262 mutex_lock(&the_lnet.ln_api_mutex);
4263 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4267 /* coming in iou_idx will hold the idx of the udsp
4268 * to get the size of. going out the iou_idx will
4269 * hold the size of the UDSP found at the passed
4272 ioc_udsp->iou_idx = lnet_get_udsp_size(udsp);
4273 if (ioc_udsp->iou_idx < 0)
4276 mutex_unlock(&the_lnet.ln_api_mutex);
4281 case IOC_LIBCFS_GET_UDSP: {
4282 struct lnet_ioctl_udsp *ioc_udsp = arg;
4283 struct lnet_udsp *udsp;
4285 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4290 mutex_lock(&the_lnet.ln_api_mutex);
4291 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4295 rc = lnet_udsp_marshal(udsp, ioc_udsp);
4296 mutex_unlock(&the_lnet.ln_api_mutex);
4301 case IOC_LIBCFS_GET_CONST_UDSP_INFO: {
4302 struct lnet_ioctl_construct_udsp_info *info = arg;
4304 if (info->cud_hdr.ioc_len < sizeof(*info))
4307 CDEBUG(D_NET, "GET_UDSP_INFO for %s\n",
4308 libcfs_nid2str(info->cud_nid));
4310 mutex_lock(&the_lnet.ln_api_mutex);
4311 lnet_udsp_get_construct_info(info);
4312 mutex_unlock(&the_lnet.ln_api_mutex);
4318 ni = lnet_net2ni_addref(data->ioc_net);
4322 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4325 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4332 EXPORT_SYMBOL(LNetCtl);
4334 void LNetDebugPeer(struct lnet_process_id id)
4336 lnet_debug_peer(id.nid);
4338 EXPORT_SYMBOL(LNetDebugPeer);
4341 * Determine if the specified peer \a nid is on the local node.
4343 * \param nid peer nid to check
4345 * \retval true If peer NID is on the local node.
4346 * \retval false If peer NID is not on the local node.
4348 bool LNetIsPeerLocal(lnet_nid_t nid)
4350 struct lnet_net *net;
4354 cpt = lnet_net_lock_current();
4355 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4356 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4357 if (ni->ni_nid == nid) {
4358 lnet_net_unlock(cpt);
4363 lnet_net_unlock(cpt);
4367 EXPORT_SYMBOL(LNetIsPeerLocal);
4370 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4371 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4373 * \param index Index of the interface to look up.
4374 * \param id On successful return, this location will hold the
4375 * struct lnet_process_id ID of the interface.
4377 * \retval 0 If an interface exists at \a index.
4378 * \retval -ENOENT If no interface has been found.
4381 LNetGetId(unsigned int index, struct lnet_process_id *id)
4384 struct lnet_net *net;
4388 LASSERT(the_lnet.ln_refcount > 0);
4390 cpt = lnet_net_lock_current();
4392 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4393 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4397 id->nid = ni->ni_nid;
4398 id->pid = the_lnet.ln_pid;
4404 lnet_net_unlock(cpt);
4407 EXPORT_SYMBOL(LNetGetId);
4412 struct lnet_handle_md mdh;
4413 struct completion completion;
4417 lnet_ping_event_handler(struct lnet_event *event)
4419 struct ping_data *pd = event->md_user_ptr;
4421 CDEBUG(D_NET, "ping event (%d %d)%s\n",
4422 event->type, event->status,
4423 event->unlinked ? " unlinked" : "");
4425 if (event->status) {
4427 pd->rc = event->status;
4428 } else if (event->type == LNET_EVENT_REPLY) {
4430 pd->rc = event->mlength;
4432 if (event->unlinked)
4433 complete(&pd->completion);
4436 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4437 struct lnet_process_id __user *ids, int n_ids)
4439 struct lnet_md md = { NULL };
4440 struct ping_data pd = { 0 };
4441 struct lnet_ping_buffer *pbuf;
4442 struct lnet_process_id tmpid;
4448 /* n_ids limit is arbitrary */
4449 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4453 * if the user buffer has more space than the lnet_interfaces_max
4454 * then only fill it up to lnet_interfaces_max
4456 if (n_ids > lnet_interfaces_max)
4457 n_ids = lnet_interfaces_max;
4459 if (id.pid == LNET_PID_ANY)
4460 id.pid = LNET_PID_LUSTRE;
4462 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4466 /* initialize md content */
4467 md.start = &pbuf->pb_info;
4468 md.length = LNET_PING_INFO_SIZE(n_ids);
4469 md.threshold = 2; /* GET/REPLY */
4471 md.options = LNET_MD_TRUNCATE;
4473 md.handler = lnet_ping_event_handler;
4475 init_completion(&pd.completion);
4477 rc = LNetMDBind(&md, LNET_UNLINK, &pd.mdh);
4479 CERROR("Can't bind MD: %d\n", rc);
4480 goto fail_ping_buffer_decref;
4483 rc = LNetGet(LNET_NID_ANY, pd.mdh, id,
4484 LNET_RESERVED_PORTAL,
4485 LNET_PROTO_PING_MATCHBITS, 0, false);
4488 /* Don't CERROR; this could be deliberate! */
4489 rc2 = LNetMDUnlink(pd.mdh);
4492 /* NB must wait for the UNLINK event below... */
4495 if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4496 /* Ensure completion in finite time... */
4497 LNetMDUnlink(pd.mdh);
4498 wait_for_completion(&pd.completion);
4502 goto fail_ping_buffer_decref;
4506 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4508 rc = -EPROTO; /* if I can't parse... */
4511 CERROR("%s: ping info too short %d\n",
4512 libcfs_id2str(id), nob);
4513 goto fail_ping_buffer_decref;
4516 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4517 lnet_swap_pinginfo(pbuf);
4518 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4519 CERROR("%s: Unexpected magic %08x\n",
4520 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4521 goto fail_ping_buffer_decref;
4524 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4525 CERROR("%s: ping w/o NI status: 0x%x\n",
4526 libcfs_id2str(id), pbuf->pb_info.pi_features);
4527 goto fail_ping_buffer_decref;
4530 if (nob < LNET_PING_INFO_SIZE(0)) {
4531 CERROR("%s: Short reply %d(%d min)\n",
4533 nob, (int)LNET_PING_INFO_SIZE(0));
4534 goto fail_ping_buffer_decref;
4537 if (pbuf->pb_info.pi_nnis < n_ids)
4538 n_ids = pbuf->pb_info.pi_nnis;
4540 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4541 CERROR("%s: Short reply %d(%d expected)\n",
4543 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4544 goto fail_ping_buffer_decref;
4547 rc = -EFAULT; /* if I segv in copy_to_user()... */
4549 memset(&tmpid, 0, sizeof(tmpid));
4550 for (i = 0; i < n_ids; i++) {
4551 tmpid.pid = pbuf->pb_info.pi_pid;
4552 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4553 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4554 goto fail_ping_buffer_decref;
4556 rc = pbuf->pb_info.pi_nnis;
4558 fail_ping_buffer_decref:
4559 lnet_ping_buffer_decref(pbuf);
4564 lnet_discover(struct lnet_process_id id, __u32 force,
4565 struct lnet_process_id __user *ids, int n_ids)
4567 struct lnet_peer_ni *lpni;
4568 struct lnet_peer_ni *p;
4569 struct lnet_peer *lp;
4570 struct lnet_process_id *buf;
4576 id.nid == LNET_NID_ANY)
4579 if (id.pid == LNET_PID_ANY)
4580 id.pid = LNET_PID_LUSTRE;
4583 * If the user buffer has more space than the lnet_interfaces_max,
4584 * then only fill it up to lnet_interfaces_max.
4586 if (n_ids > lnet_interfaces_max)
4587 n_ids = lnet_interfaces_max;
4589 CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4593 cpt = lnet_net_lock_current();
4594 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4601 * Clearing the NIDS_UPTODATE flag ensures the peer will
4602 * be discovered, provided discovery has not been disabled.
4604 lp = lpni->lpni_peer_net->lpn_peer;
4605 spin_lock(&lp->lp_lock);
4606 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4607 /* If the force flag is set, force a PING and PUSH as well. */
4609 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4610 spin_unlock(&lp->lp_lock);
4611 rc = lnet_discover_peer_locked(lpni, cpt, true);
4615 /* The lpni (or lp) for this NID may have changed and our ref is
4616 * the only thing keeping the old one around. Release the ref
4617 * and lookup the lpni again
4619 lnet_peer_ni_decref_locked(lpni);
4620 lpni = lnet_find_peer_ni_locked(id.nid);
4625 lp = lpni->lpni_peer_net->lpn_peer;
4629 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4630 buf[i].pid = id.pid;
4631 buf[i].nid = p->lpni_nid;
4638 lnet_peer_ni_decref_locked(lpni);
4640 lnet_net_unlock(cpt);
4643 if (copy_to_user(ids, buf, rc * sizeof(*buf)))
4645 CFS_FREE_PTR_ARRAY(buf, n_ids);
4651 * Retrieve peer discovery status.
4653 * \retval 1 if lnet_peer_discovery_disabled is 0
4654 * \retval 0 if lnet_peer_discovery_disabled is 1
4657 LNetGetPeerDiscoveryStatus(void)
4659 return !lnet_peer_discovery_disabled;
4661 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);