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>
42 #include <lnet/udsp.h>
43 #include <lnet/lib-lnet.h>
45 #define D_LNI D_CONSOLE
48 * initialize ln_api_mutex statically, since it needs to be used in
49 * discovery_set callback. That module parameter callback can be called
50 * before module init completes. The mutex needs to be ready for use then.
52 struct lnet the_lnet = {
53 .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
54 }; /* THE state of the network */
55 EXPORT_SYMBOL(the_lnet);
57 static char *ip2nets = "";
58 module_param(ip2nets, charp, 0444);
59 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
61 static char *networks = "";
62 module_param(networks, charp, 0444);
63 MODULE_PARM_DESC(networks, "local networks");
65 static char *routes = "";
66 module_param(routes, charp, 0444);
67 MODULE_PARM_DESC(routes, "routes to non-local networks");
69 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
70 module_param(rnet_htable_size, int, 0444);
71 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
73 static int use_tcp_bonding;
74 module_param(use_tcp_bonding, int, 0444);
75 MODULE_PARM_DESC(use_tcp_bonding,
76 "use_tcp_bonding parameter has been removed");
78 unsigned int lnet_numa_range = 0;
79 module_param(lnet_numa_range, uint, 0444);
80 MODULE_PARM_DESC(lnet_numa_range,
81 "NUMA range to consider during Multi-Rail selection");
84 * lnet_health_sensitivity determines by how much we decrement the health
85 * value on sending error. The value defaults to 100, which means health
86 * interface health is decremented by 100 points every failure.
88 unsigned int lnet_health_sensitivity = 100;
89 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
90 #ifdef HAVE_KERNEL_PARAM_OPS
91 static struct kernel_param_ops param_ops_health_sensitivity = {
92 .set = sensitivity_set,
95 #define param_check_health_sensitivity(name, p) \
96 __param_check(name, p, int)
97 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
99 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
100 &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
102 MODULE_PARM_DESC(lnet_health_sensitivity,
103 "Value to decrement the health value by on error");
106 * lnet_recovery_interval determines how often we should perform recovery
107 * on unhealthy interfaces.
109 unsigned int lnet_recovery_interval = 1;
110 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
111 #ifdef HAVE_KERNEL_PARAM_OPS
112 static struct kernel_param_ops param_ops_recovery_interval = {
113 .set = recovery_interval_set,
114 .get = param_get_int,
116 #define param_check_recovery_interval(name, p) \
117 __param_check(name, p, int)
118 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
120 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
121 &lnet_recovery_interval, S_IRUGO|S_IWUSR);
123 MODULE_PARM_DESC(lnet_recovery_interval,
124 "DEPRECATED - Interval to recover unhealthy interfaces in seconds");
126 unsigned int lnet_recovery_limit;
127 module_param(lnet_recovery_limit, uint, 0644);
128 MODULE_PARM_DESC(lnet_recovery_limit,
129 "How long to attempt recovery of unhealthy peer interfaces in seconds. Set to 0 to allow indefinite recovery");
131 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
132 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
134 static struct kernel_param_ops param_ops_interfaces_max = {
136 .get = param_get_int,
139 #define param_check_interfaces_max(name, p) \
140 __param_check(name, p, int)
142 #ifdef HAVE_KERNEL_PARAM_OPS
143 module_param(lnet_interfaces_max, interfaces_max, 0644);
145 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
146 ¶m_ops_interfaces_max, 0644);
148 MODULE_PARM_DESC(lnet_interfaces_max,
149 "Maximum number of interfaces in a node.");
151 unsigned lnet_peer_discovery_disabled = 0;
152 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
154 static struct kernel_param_ops param_ops_discovery_disabled = {
155 .set = discovery_set,
156 .get = param_get_int,
159 #define param_check_discovery_disabled(name, p) \
160 __param_check(name, p, int)
161 #ifdef HAVE_KERNEL_PARAM_OPS
162 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
164 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
165 ¶m_ops_discovery_disabled, 0644);
167 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
168 "Set to 1 to disable peer discovery on this node.");
170 unsigned int lnet_drop_asym_route;
171 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
173 static struct kernel_param_ops param_ops_drop_asym_route = {
174 .set = drop_asym_route_set,
175 .get = param_get_int,
178 #define param_check_drop_asym_route(name, p) \
179 __param_check(name, p, int)
180 #ifdef HAVE_KERNEL_PARAM_OPS
181 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
183 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
184 ¶m_ops_drop_asym_route, 0644);
186 MODULE_PARM_DESC(lnet_drop_asym_route,
187 "Set to 1 to drop asymmetrical route messages.");
189 #define LNET_TRANSACTION_TIMEOUT_DEFAULT 50
190 unsigned int lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_DEFAULT;
191 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
192 #ifdef HAVE_KERNEL_PARAM_OPS
193 static struct kernel_param_ops param_ops_transaction_timeout = {
194 .set = transaction_to_set,
195 .get = param_get_int,
198 #define param_check_transaction_timeout(name, p) \
199 __param_check(name, p, int)
200 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
202 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
203 &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
205 MODULE_PARM_DESC(lnet_transaction_timeout,
206 "Maximum number of seconds to wait for a peer response.");
208 #define LNET_RETRY_COUNT_DEFAULT 2
209 unsigned int lnet_retry_count = LNET_RETRY_COUNT_DEFAULT;
210 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
211 #ifdef HAVE_KERNEL_PARAM_OPS
212 static struct kernel_param_ops param_ops_retry_count = {
213 .set = retry_count_set,
214 .get = param_get_int,
217 #define param_check_retry_count(name, p) \
218 __param_check(name, p, int)
219 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
221 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
222 &lnet_retry_count, S_IRUGO|S_IWUSR);
224 MODULE_PARM_DESC(lnet_retry_count,
225 "Maximum number of times to retry transmitting a message");
227 unsigned int lnet_response_tracking = 3;
228 static int response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp);
230 #ifdef HAVE_KERNEL_PARAM_OPS
231 static struct kernel_param_ops param_ops_response_tracking = {
232 .set = response_tracking_set,
233 .get = param_get_int,
236 #define param_check_response_tracking(name, p) \
237 __param_check(name, p, int)
238 module_param(lnet_response_tracking, response_tracking, 0644);
240 module_param_call(lnet_response_tracking, response_tracking_set, param_get_int,
241 &lnet_response_tracking, 0644);
243 MODULE_PARM_DESC(lnet_response_tracking,
244 "(0|1|2|3) LNet Internal Only|GET Reply only|PUT ACK only|Full Tracking (default)");
246 #define LNET_LND_TIMEOUT_DEFAULT ((LNET_TRANSACTION_TIMEOUT_DEFAULT - 1) / \
247 (LNET_RETRY_COUNT_DEFAULT + 1))
248 unsigned int lnet_lnd_timeout = LNET_LND_TIMEOUT_DEFAULT;
249 static void lnet_set_lnd_timeout(void)
251 lnet_lnd_timeout = (lnet_transaction_timeout - 1) /
252 (lnet_retry_count + 1);
256 * This sequence number keeps track of how many times DLC was used to
257 * update the local NIs. It is incremented when a NI is added or
258 * removed and checked when sending a message to determine if there is
259 * a need to re-run the selection algorithm. See lnet_select_pathway()
260 * for more details on its usage.
262 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
264 static int lnet_ping(struct lnet_process_id id, struct lnet_nid *src_nid,
265 signed long timeout, struct lnet_process_id __user *ids,
268 static int lnet_discover(struct lnet_process_id id, __u32 force,
269 struct lnet_process_id __user *ids, int n_ids);
272 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
275 unsigned *sensitivity = (unsigned *)kp->arg;
278 rc = kstrtoul(val, 0, &value);
280 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
285 * The purpose of locking the api_mutex here is to ensure that
286 * the correct value ends up stored properly.
288 mutex_lock(&the_lnet.ln_api_mutex);
290 if (value > LNET_MAX_HEALTH_VALUE) {
291 mutex_unlock(&the_lnet.ln_api_mutex);
292 CERROR("Invalid health value. Maximum: %d value = %lu\n",
293 LNET_MAX_HEALTH_VALUE, value);
297 if (*sensitivity != 0 && value == 0 && lnet_retry_count != 0) {
298 lnet_retry_count = 0;
299 lnet_set_lnd_timeout();
302 *sensitivity = value;
304 mutex_unlock(&the_lnet.ln_api_mutex);
310 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
312 CWARN("'lnet_recovery_interval' has been deprecated\n");
318 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
321 unsigned *discovery_off = (unsigned *)kp->arg;
323 struct lnet_ping_buffer *pbuf;
325 rc = kstrtoul(val, 0, &value);
327 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
331 value = (value) ? 1 : 0;
334 * The purpose of locking the api_mutex here is to ensure that
335 * the correct value ends up stored properly.
337 mutex_lock(&the_lnet.ln_api_mutex);
339 if (value == *discovery_off) {
340 mutex_unlock(&the_lnet.ln_api_mutex);
345 * We still want to set the discovery value even when LNet is not
346 * running. This is the case when LNet is being loaded and we want
347 * the module parameters to take effect. Otherwise if we're
348 * changing the value dynamically, we want to set it after
351 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
352 *discovery_off = value;
353 mutex_unlock(&the_lnet.ln_api_mutex);
357 /* tell peers that discovery setting has changed */
358 lnet_net_lock(LNET_LOCK_EX);
359 pbuf = the_lnet.ln_ping_target;
361 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
363 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
364 lnet_net_unlock(LNET_LOCK_EX);
366 /* only send a push when we're turning off discovery */
367 if (*discovery_off <= 0 && value > 0)
368 lnet_push_update_to_peers(1);
369 *discovery_off = value;
371 mutex_unlock(&the_lnet.ln_api_mutex);
377 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
380 unsigned int *drop_asym_route = (unsigned int *)kp->arg;
383 rc = kstrtoul(val, 0, &value);
385 CERROR("Invalid module parameter value for "
386 "'lnet_drop_asym_route'\n");
391 * The purpose of locking the api_mutex here is to ensure that
392 * the correct value ends up stored properly.
394 mutex_lock(&the_lnet.ln_api_mutex);
396 if (value == *drop_asym_route) {
397 mutex_unlock(&the_lnet.ln_api_mutex);
401 *drop_asym_route = value;
403 mutex_unlock(&the_lnet.ln_api_mutex);
409 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
412 unsigned *transaction_to = (unsigned *)kp->arg;
415 rc = kstrtoul(val, 0, &value);
417 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
422 * The purpose of locking the api_mutex here is to ensure that
423 * the correct value ends up stored properly.
425 mutex_lock(&the_lnet.ln_api_mutex);
427 if (value <= lnet_retry_count || value == 0) {
428 mutex_unlock(&the_lnet.ln_api_mutex);
429 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
430 "Has to be greater than lnet_retry_count (%u)\n",
431 value, lnet_retry_count);
435 if (value == *transaction_to) {
436 mutex_unlock(&the_lnet.ln_api_mutex);
440 *transaction_to = value;
441 /* Update the lnet_lnd_timeout now that we've modified the
442 * transaction timeout
444 lnet_set_lnd_timeout();
446 mutex_unlock(&the_lnet.ln_api_mutex);
452 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
455 unsigned *retry_count = (unsigned *)kp->arg;
458 rc = kstrtoul(val, 0, &value);
460 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
465 * The purpose of locking the api_mutex here is to ensure that
466 * the correct value ends up stored properly.
468 mutex_lock(&the_lnet.ln_api_mutex);
470 if (lnet_health_sensitivity == 0 && value > 0) {
471 mutex_unlock(&the_lnet.ln_api_mutex);
472 CERROR("Can not set lnet_retry_count when health feature is turned off\n");
476 if (value > lnet_transaction_timeout) {
477 mutex_unlock(&the_lnet.ln_api_mutex);
478 CERROR("Invalid value for lnet_retry_count (%lu). "
479 "Has to be smaller than lnet_transaction_timeout (%u)\n",
480 value, lnet_transaction_timeout);
484 *retry_count = value;
486 /* Update the lnet_lnd_timeout now that we've modified the
489 lnet_set_lnd_timeout();
491 mutex_unlock(&the_lnet.ln_api_mutex);
497 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
501 rc = kstrtoint(val, 0, &value);
503 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
507 if (value < LNET_INTERFACES_MIN) {
508 CWARN("max interfaces provided are too small, setting to %d\n",
509 LNET_INTERFACES_MAX_DEFAULT);
510 value = LNET_INTERFACES_MAX_DEFAULT;
513 *(int *)kp->arg = value;
519 response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp)
522 unsigned long new_value;
524 rc = kstrtoul(val, 0, &new_value);
526 CERROR("Invalid value for 'lnet_response_tracking'\n");
530 if (new_value < 0 || new_value > 3) {
531 CWARN("Invalid value (%lu) for 'lnet_response_tracking'\n",
536 lnet_response_tracking = new_value;
542 lnet_get_routes(void)
548 lnet_get_networks(void)
553 if (*networks != 0 && *ip2nets != 0) {
554 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
555 "'ip2nets' but not both at once\n");
560 rc = lnet_parse_ip2nets(&nets, ip2nets);
561 return (rc == 0) ? nets : NULL;
571 lnet_init_locks(void)
573 spin_lock_init(&the_lnet.ln_eq_wait_lock);
574 spin_lock_init(&the_lnet.ln_msg_resend_lock);
575 init_completion(&the_lnet.ln_mt_wait_complete);
576 mutex_init(&the_lnet.ln_lnd_mutex);
579 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
580 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
582 struct kmem_cache *lnet_udsp_cachep; /* udsp cache */
583 struct kmem_cache *lnet_rspt_cachep; /* response tracker cache */
584 struct kmem_cache *lnet_msg_cachep;
587 lnet_slab_setup(void)
589 /* create specific kmem_cache for MEs and small MDs (i.e., originally
590 * allocated in <size-xxx> kmem_cache).
592 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
594 if (!lnet_mes_cachep)
597 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
598 LNET_SMALL_MD_SIZE, 0, 0,
600 if (!lnet_small_mds_cachep)
603 lnet_udsp_cachep = kmem_cache_create("lnet_udsp",
604 sizeof(struct lnet_udsp),
606 if (!lnet_udsp_cachep)
609 lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
611 if (!lnet_rspt_cachep)
614 lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
616 if (!lnet_msg_cachep)
623 lnet_slab_cleanup(void)
625 if (lnet_msg_cachep) {
626 kmem_cache_destroy(lnet_msg_cachep);
627 lnet_msg_cachep = NULL;
630 if (lnet_rspt_cachep) {
631 kmem_cache_destroy(lnet_rspt_cachep);
632 lnet_rspt_cachep = NULL;
635 if (lnet_udsp_cachep) {
636 kmem_cache_destroy(lnet_udsp_cachep);
637 lnet_udsp_cachep = NULL;
640 if (lnet_small_mds_cachep) {
641 kmem_cache_destroy(lnet_small_mds_cachep);
642 lnet_small_mds_cachep = NULL;
645 if (lnet_mes_cachep) {
646 kmem_cache_destroy(lnet_mes_cachep);
647 lnet_mes_cachep = NULL;
652 lnet_create_remote_nets_table(void)
655 struct list_head *hash;
657 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
658 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
659 CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
661 CERROR("Failed to create remote nets hash table\n");
665 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
666 INIT_LIST_HEAD(&hash[i]);
667 the_lnet.ln_remote_nets_hash = hash;
672 lnet_destroy_remote_nets_table(void)
676 if (the_lnet.ln_remote_nets_hash == NULL)
679 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
680 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
682 CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
683 LNET_REMOTE_NETS_HASH_SIZE);
684 the_lnet.ln_remote_nets_hash = NULL;
688 lnet_destroy_locks(void)
690 if (the_lnet.ln_res_lock != NULL) {
691 cfs_percpt_lock_free(the_lnet.ln_res_lock);
692 the_lnet.ln_res_lock = NULL;
695 if (the_lnet.ln_net_lock != NULL) {
696 cfs_percpt_lock_free(the_lnet.ln_net_lock);
697 the_lnet.ln_net_lock = NULL;
702 lnet_create_locks(void)
706 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
707 if (the_lnet.ln_res_lock == NULL)
710 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
711 if (the_lnet.ln_net_lock == NULL)
717 lnet_destroy_locks();
721 static void lnet_assert_wire_constants(void)
723 /* Wire protocol assertions generated by 'wirecheck'
724 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
725 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
726 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
730 BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
731 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
732 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
733 BUILD_BUG_ON(LNET_MSG_ACK != 0);
734 BUILD_BUG_ON(LNET_MSG_PUT != 1);
735 BUILD_BUG_ON(LNET_MSG_GET != 2);
736 BUILD_BUG_ON(LNET_MSG_REPLY != 3);
737 BUILD_BUG_ON(LNET_MSG_HELLO != 4);
739 BUILD_BUG_ON((int)sizeof(lnet_nid_t) != 8);
740 BUILD_BUG_ON((int)sizeof(lnet_pid_t) != 4);
742 /* Checks for struct lnet_nid */
743 BUILD_BUG_ON((int)sizeof(struct lnet_nid) != 20);
744 BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_size) != 0);
745 BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_size) != 1);
746 BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_type) != 1);
747 BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_type) != 1);
748 BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_num) != 2);
749 BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_num) != 2);
750 BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_addr) != 4);
751 BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_addr) != 16);
753 /* Checks for struct lnet_process_id_packed */
754 BUILD_BUG_ON((int)sizeof(struct lnet_process_id_packed) != 12);
755 BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, nid) != 0);
756 BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->nid) != 8);
757 BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, pid) != 8);
758 BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->pid) != 4);
760 /* Checks for struct lnet_handle_wire */
761 BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
762 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
763 wh_interface_cookie) != 0);
764 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
765 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
766 wh_object_cookie) != 8);
767 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
769 /* Checks for struct struct lnet_magicversion */
770 BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
771 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
772 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
773 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
774 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
775 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
776 version_minor) != 6);
777 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
779 /* Checks for struct _lnet_hdr_nid4 */
780 BUILD_BUG_ON((int)sizeof(struct _lnet_hdr_nid4) != 72);
781 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, dest_nid) != 0);
782 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->dest_nid) != 8);
783 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, src_nid) != 8);
784 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->src_nid) != 8);
785 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, dest_pid) != 16);
786 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->dest_pid) != 4);
787 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, src_pid) != 20);
788 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->src_pid) != 4);
789 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, type) != 24);
790 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->type) != 4);
791 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, payload_length) != 28);
792 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->payload_length) != 4);
793 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg) != 32);
794 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg) != 40);
797 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.dst_wmd) != 32);
798 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.dst_wmd) != 16);
799 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.match_bits) != 48);
800 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.match_bits) != 8);
801 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.mlength) != 56);
802 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.mlength) != 4);
805 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.ack_wmd) != 32);
806 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.ack_wmd) != 16);
807 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.match_bits) != 48);
808 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.match_bits) != 8);
809 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.hdr_data) != 56);
810 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.hdr_data) != 8);
811 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.ptl_index) != 64);
812 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.ptl_index) != 4);
813 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.offset) != 68);
814 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.offset) != 4);
817 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.return_wmd) != 32);
818 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.return_wmd) != 16);
819 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.match_bits) != 48);
820 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.match_bits) != 8);
821 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.ptl_index) != 56);
822 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.ptl_index) != 4);
823 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.src_offset) != 60);
824 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.src_offset) != 4);
825 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.sink_length) != 64);
826 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.sink_length) != 4);
829 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.reply.dst_wmd) != 32);
830 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.reply.dst_wmd) != 16);
833 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.hello.incarnation) != 32);
834 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.hello.incarnation) != 8);
835 BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.hello.type) != 40);
836 BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.hello.type) != 4);
838 /* Checks for struct lnet_ni_status and related constants */
839 BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
840 BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
841 BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
843 /* Checks for struct lnet_ni_status */
844 BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
845 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
846 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
847 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
848 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
849 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
850 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
852 /* Checks for struct lnet_ping_info and related constants */
853 BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
854 BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
855 BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
856 BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
857 BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
858 BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
859 BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
860 BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
862 /* Checks for struct lnet_ping_info */
863 BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
864 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
865 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
866 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
867 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
868 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
869 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
870 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
871 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
872 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
873 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
875 /* Acceptor connection request */
876 BUILD_BUG_ON(LNET_PROTO_ACCEPTOR_VERSION != 1);
878 /* Checks for struct lnet_acceptor_connreq */
879 BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq) != 16);
880 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_magic) != 0);
881 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_magic) != 4);
882 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_version) != 4);
883 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_version) != 4);
884 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_nid) != 8);
885 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_nid) != 8);
887 /* Checks for struct lnet_acceptor_connreq_v2 */
888 BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq_v2) != 28);
889 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_magic) != 0);
890 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_magic) != 4);
891 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_version) != 4);
892 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_version) != 4);
893 BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_nid) != 8);
894 BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_nid) != 20);
896 /* Checks for struct lnet_counters_common */
897 BUILD_BUG_ON((int)sizeof(struct lnet_counters_common) != 60);
898 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_alloc) != 0);
899 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_alloc) != 4);
900 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_max) != 4);
901 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_max) != 4);
902 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_errors) != 8);
903 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_errors) != 4);
904 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_count) != 12);
905 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_count) != 4);
906 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_count) != 16);
907 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_count) != 4);
908 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_count) != 20);
909 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_count) != 4);
910 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_count) != 24);
911 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_count) != 4);
912 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_length) != 28);
913 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_length) != 8);
914 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_length) != 36);
915 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_length) != 8);
916 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_length) != 44);
917 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_length) != 8);
918 BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_length) != 52);
919 BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_length) != 8);
922 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
924 const struct lnet_lnd *lnd;
926 /* holding lnd mutex */
927 if (type >= NUM_LNDS)
929 lnd = the_lnet.ln_lnds[type];
930 LASSERT(!lnd || lnd->lnd_type == type);
936 lnet_get_lnd_timeout(void)
938 return lnet_lnd_timeout;
940 EXPORT_SYMBOL(lnet_get_lnd_timeout);
943 lnet_register_lnd(const struct lnet_lnd *lnd)
945 mutex_lock(&the_lnet.ln_lnd_mutex);
947 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
948 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
950 the_lnet.ln_lnds[lnd->lnd_type] = lnd;
952 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
954 mutex_unlock(&the_lnet.ln_lnd_mutex);
956 EXPORT_SYMBOL(lnet_register_lnd);
959 lnet_unregister_lnd(const struct lnet_lnd *lnd)
961 mutex_lock(&the_lnet.ln_lnd_mutex);
963 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
965 the_lnet.ln_lnds[lnd->lnd_type] = NULL;
966 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
968 mutex_unlock(&the_lnet.ln_lnd_mutex);
970 EXPORT_SYMBOL(lnet_unregister_lnd);
973 lnet_counters_get_common_locked(struct lnet_counters_common *common)
975 struct lnet_counters *ctr;
978 /* FIXME !!! Their is no assert_lnet_net_locked() to ensure this
979 * actually called under the protection of the lnet_net_lock.
981 memset(common, 0, sizeof(*common));
983 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
984 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
985 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
986 common->lcc_errors += ctr->lct_common.lcc_errors;
987 common->lcc_send_count += ctr->lct_common.lcc_send_count;
988 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
989 common->lcc_route_count += ctr->lct_common.lcc_route_count;
990 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
991 common->lcc_send_length += ctr->lct_common.lcc_send_length;
992 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
993 common->lcc_route_length += ctr->lct_common.lcc_route_length;
994 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
999 lnet_counters_get_common(struct lnet_counters_common *common)
1001 lnet_net_lock(LNET_LOCK_EX);
1002 lnet_counters_get_common_locked(common);
1003 lnet_net_unlock(LNET_LOCK_EX);
1005 EXPORT_SYMBOL(lnet_counters_get_common);
1008 lnet_counters_get(struct lnet_counters *counters)
1010 struct lnet_counters *ctr;
1011 struct lnet_counters_health *health = &counters->lct_health;
1014 memset(counters, 0, sizeof(*counters));
1016 lnet_net_lock(LNET_LOCK_EX);
1018 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1019 GOTO(out_unlock, rc = -ENODEV);
1021 lnet_counters_get_common_locked(&counters->lct_common);
1023 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
1024 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
1025 health->lch_resend_count += ctr->lct_health.lch_resend_count;
1026 health->lch_response_timeout_count +=
1027 ctr->lct_health.lch_response_timeout_count;
1028 health->lch_local_interrupt_count +=
1029 ctr->lct_health.lch_local_interrupt_count;
1030 health->lch_local_dropped_count +=
1031 ctr->lct_health.lch_local_dropped_count;
1032 health->lch_local_aborted_count +=
1033 ctr->lct_health.lch_local_aborted_count;
1034 health->lch_local_no_route_count +=
1035 ctr->lct_health.lch_local_no_route_count;
1036 health->lch_local_timeout_count +=
1037 ctr->lct_health.lch_local_timeout_count;
1038 health->lch_local_error_count +=
1039 ctr->lct_health.lch_local_error_count;
1040 health->lch_remote_dropped_count +=
1041 ctr->lct_health.lch_remote_dropped_count;
1042 health->lch_remote_error_count +=
1043 ctr->lct_health.lch_remote_error_count;
1044 health->lch_remote_timeout_count +=
1045 ctr->lct_health.lch_remote_timeout_count;
1046 health->lch_network_timeout_count +=
1047 ctr->lct_health.lch_network_timeout_count;
1050 lnet_net_unlock(LNET_LOCK_EX);
1053 EXPORT_SYMBOL(lnet_counters_get);
1056 lnet_counters_reset(void)
1058 struct lnet_counters *counters;
1061 lnet_net_lock(LNET_LOCK_EX);
1063 if (the_lnet.ln_state != LNET_STATE_RUNNING)
1066 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
1067 memset(counters, 0, sizeof(struct lnet_counters));
1069 lnet_net_unlock(LNET_LOCK_EX);
1073 lnet_res_type2str(int type)
1078 case LNET_COOKIE_TYPE_MD:
1080 case LNET_COOKIE_TYPE_ME:
1082 case LNET_COOKIE_TYPE_EQ:
1088 lnet_res_container_cleanup(struct lnet_res_container *rec)
1092 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
1095 while (!list_empty(&rec->rec_active)) {
1096 struct list_head *e = rec->rec_active.next;
1099 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1100 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1102 } else { /* NB: Active MEs should be attached on portals */
1109 /* Found alive MD/ME/EQ, user really should unlink/free
1110 * all of them before finalize LNet, but if someone didn't,
1111 * we have to recycle garbage for him */
1112 CERROR("%d active elements on exit of %s container\n",
1113 count, lnet_res_type2str(rec->rec_type));
1116 if (rec->rec_lh_hash != NULL) {
1117 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1118 rec->rec_lh_hash = NULL;
1121 rec->rec_type = 0; /* mark it as finalized */
1125 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1130 LASSERT(rec->rec_type == 0);
1132 rec->rec_type = type;
1133 INIT_LIST_HEAD(&rec->rec_active);
1135 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1137 /* Arbitrary choice of hash table size */
1138 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1139 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1140 if (rec->rec_lh_hash == NULL) {
1145 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1146 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1151 CERROR("Failed to setup %s resource container\n",
1152 lnet_res_type2str(type));
1153 lnet_res_container_cleanup(rec);
1158 lnet_res_containers_destroy(struct lnet_res_container **recs)
1160 struct lnet_res_container *rec;
1163 cfs_percpt_for_each(rec, i, recs)
1164 lnet_res_container_cleanup(rec);
1166 cfs_percpt_free(recs);
1169 static struct lnet_res_container **
1170 lnet_res_containers_create(int type)
1172 struct lnet_res_container **recs;
1173 struct lnet_res_container *rec;
1177 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1179 CERROR("Failed to allocate %s resource containers\n",
1180 lnet_res_type2str(type));
1184 cfs_percpt_for_each(rec, i, recs) {
1185 rc = lnet_res_container_setup(rec, i, type);
1187 lnet_res_containers_destroy(recs);
1195 struct lnet_libhandle *
1196 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1198 /* ALWAYS called with lnet_res_lock held */
1199 struct list_head *head;
1200 struct lnet_libhandle *lh;
1203 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1206 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1207 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1209 list_for_each_entry(lh, head, lh_hash_chain) {
1210 if (lh->lh_cookie == cookie)
1218 lnet_res_lh_initialize(struct lnet_res_container *rec,
1219 struct lnet_libhandle *lh)
1221 /* ALWAYS called with lnet_res_lock held */
1222 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1225 lh->lh_cookie = rec->rec_lh_cookie;
1226 rec->rec_lh_cookie += 1 << ibits;
1228 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1230 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1234 lnet_create_array_of_queues(void)
1236 struct list_head **qs;
1237 struct list_head *q;
1240 qs = cfs_percpt_alloc(lnet_cpt_table(),
1241 sizeof(struct list_head));
1243 CERROR("Failed to allocate queues\n");
1247 cfs_percpt_for_each(q, i, qs)
1253 static int lnet_unprepare(void);
1256 lnet_prepare(lnet_pid_t requested_pid)
1258 /* Prepare to bring up the network */
1259 struct lnet_res_container **recs;
1262 if (requested_pid == LNET_PID_ANY) {
1263 /* Don't instantiate LNET just for me */
1267 LASSERT(the_lnet.ln_refcount == 0);
1269 the_lnet.ln_routing = 0;
1271 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1272 the_lnet.ln_pid = requested_pid;
1274 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1275 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1276 INIT_LIST_HEAD(&the_lnet.ln_nets);
1277 INIT_LIST_HEAD(&the_lnet.ln_routers);
1278 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1279 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1280 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1281 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1282 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1283 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1284 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1285 INIT_LIST_HEAD(&the_lnet.ln_udsp_list);
1286 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1287 the_lnet.ln_mt_handler = NULL;
1288 init_completion(&the_lnet.ln_started);
1290 rc = lnet_slab_setup();
1294 rc = lnet_create_remote_nets_table();
1299 * NB the interface cookie in wire handles guards against delayed
1300 * replies and ACKs appearing valid after reboot.
1302 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1304 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1305 sizeof(struct lnet_counters));
1306 if (the_lnet.ln_counters == NULL) {
1307 CERROR("Failed to allocate counters for LNet\n");
1312 rc = lnet_peer_tables_create();
1316 rc = lnet_msg_containers_create();
1320 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1321 LNET_COOKIE_TYPE_EQ);
1325 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1331 the_lnet.ln_md_containers = recs;
1333 rc = lnet_portals_create();
1335 CERROR("Failed to create portals for LNet: %d\n", rc);
1339 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1340 if (!the_lnet.ln_mt_zombie_rstqs) {
1353 lnet_unprepare (void)
1355 /* NB no LNET_LOCK since this is the last reference. All LND instances
1356 * have shut down already, so it is safe to unlink and free all
1357 * descriptors, even those that appear committed to a network op (eg MD
1358 * with non-zero pending count) */
1360 lnet_fail_nid(LNET_NID_ANY, 0);
1362 LASSERT(the_lnet.ln_refcount == 0);
1363 LASSERT(list_empty(&the_lnet.ln_test_peers));
1364 LASSERT(list_empty(&the_lnet.ln_nets));
1366 if (the_lnet.ln_mt_zombie_rstqs) {
1367 lnet_clean_zombie_rstqs();
1368 the_lnet.ln_mt_zombie_rstqs = NULL;
1371 lnet_assert_handler_unused(the_lnet.ln_mt_handler);
1372 the_lnet.ln_mt_handler = NULL;
1374 lnet_portals_destroy();
1376 if (the_lnet.ln_md_containers != NULL) {
1377 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1378 the_lnet.ln_md_containers = NULL;
1381 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1383 lnet_msg_containers_destroy();
1385 lnet_rtrpools_free(0);
1387 if (the_lnet.ln_counters != NULL) {
1388 cfs_percpt_free(the_lnet.ln_counters);
1389 the_lnet.ln_counters = NULL;
1391 lnet_destroy_remote_nets_table();
1392 lnet_udsp_destroy(true);
1393 lnet_slab_cleanup();
1399 lnet_net2ni_locked(__u32 net_id, int cpt)
1402 struct lnet_net *net;
1404 LASSERT(cpt != LNET_LOCK_EX);
1406 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1407 if (net->net_id == net_id) {
1408 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1418 lnet_net2ni_addref(__u32 net)
1423 ni = lnet_net2ni_locked(net, 0);
1425 lnet_ni_addref_locked(ni, 0);
1430 EXPORT_SYMBOL(lnet_net2ni_addref);
1433 lnet_get_net_locked(__u32 net_id)
1435 struct lnet_net *net;
1437 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1438 if (net->net_id == net_id)
1446 lnet_net_clr_pref_rtrs(struct lnet_net *net)
1448 struct list_head zombies;
1449 struct lnet_nid_list *ne;
1450 struct lnet_nid_list *tmp;
1452 INIT_LIST_HEAD(&zombies);
1454 lnet_net_lock(LNET_LOCK_EX);
1455 list_splice_init(&net->net_rtr_pref_nids, &zombies);
1456 lnet_net_unlock(LNET_LOCK_EX);
1458 list_for_each_entry_safe(ne, tmp, &zombies, nl_list) {
1459 list_del_init(&ne->nl_list);
1460 LIBCFS_FREE(ne, sizeof(*ne));
1465 lnet_net_add_pref_rtr(struct lnet_net *net,
1466 struct lnet_nid *gw_nid)
1467 __must_hold(&the_lnet.ln_api_mutex)
1469 struct lnet_nid_list *ne;
1471 /* This function is called with api_mutex held. When the api_mutex
1472 * is held the list can not be modified, as it is only modified as
1473 * a result of applying a UDSP and that happens under api_mutex
1476 list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1477 if (nid_same(&ne->nl_nid, gw_nid))
1481 LIBCFS_ALLOC(ne, sizeof(*ne));
1485 ne->nl_nid = *gw_nid;
1487 /* Lock the cpt to protect against addition and checks in the
1488 * selection algorithm
1490 lnet_net_lock(LNET_LOCK_EX);
1491 list_add(&ne->nl_list, &net->net_rtr_pref_nids);
1492 lnet_net_unlock(LNET_LOCK_EX);
1498 lnet_net_is_pref_rtr_locked(struct lnet_net *net, struct lnet_nid *rtr_nid)
1500 struct lnet_nid_list *ne;
1502 CDEBUG(D_NET, "%s: rtr pref empty: %d\n",
1503 libcfs_net2str(net->net_id),
1504 list_empty(&net->net_rtr_pref_nids));
1506 if (list_empty(&net->net_rtr_pref_nids))
1509 list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1510 CDEBUG(D_NET, "Comparing pref %s with gw %s\n",
1511 libcfs_nidstr(&ne->nl_nid),
1512 libcfs_nidstr(rtr_nid));
1513 if (nid_same(rtr_nid, &ne->nl_nid))
1521 lnet_nid4_cpt_hash(lnet_nid_t nid, unsigned int number)
1524 __u64 pair_bits = 0x0001000100010001LLU;
1525 __u64 mask = pair_bits * 0xFF;
1528 /* Use (sum-by-multiplication of nid bytes) mod (number of CPTs)
1529 * to match nid to a CPT.
1531 pair_sum = (key & mask) + ((key >> 8) & mask);
1532 pair_sum = (pair_sum * pair_bits) >> 48;
1534 CDEBUG(D_NET, "Match nid %s to cpt %u\n",
1535 libcfs_nid2str(nid), (unsigned int)(pair_sum) % number);
1537 return (unsigned int)(pair_sum) % number;
1541 lnet_nid_cpt_hash(struct lnet_nid *nid, unsigned int number)
1547 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1552 if (nid_is_nid4(nid))
1553 return lnet_nid4_cpt_hash(lnet_nid_to_nid4(nid), number);
1555 for (i = 0; i < 4; i++)
1556 h = hash_32(nid->nid_addr[i]^h, 32);
1557 val = hash_32(LNET_NID_NET(nid) ^ h, LNET_CPT_BITS);
1560 return (unsigned int)(h + val + (val >> 1)) % number;
1564 lnet_cpt_of_nid_locked(struct lnet_nid *nid, struct lnet_ni *ni)
1566 struct lnet_net *net;
1568 /* must called with hold of lnet_net_lock */
1569 if (LNET_CPT_NUMBER == 1)
1570 return 0; /* the only one */
1573 * If NI is provided then use the CPT identified in the NI cpt
1574 * list if one exists. If one doesn't exist, then that NI is
1575 * associated with all CPTs and it follows that the net it belongs
1576 * to is implicitly associated with all CPTs, so just hash the nid
1580 if (ni->ni_cpts != NULL)
1581 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1584 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1587 /* no NI provided so look at the net */
1588 net = lnet_get_net_locked(LNET_NID_NET(nid));
1590 if (net != NULL && net->net_cpts != NULL) {
1591 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1594 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1598 lnet_nid2cpt(struct lnet_nid *nid, struct lnet_ni *ni)
1603 if (LNET_CPT_NUMBER == 1)
1604 return 0; /* the only one */
1606 cpt = lnet_net_lock_current();
1608 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1610 lnet_net_unlock(cpt);
1614 EXPORT_SYMBOL(lnet_nid2cpt);
1617 lnet_cpt_of_nid(lnet_nid_t nid4, struct lnet_ni *ni)
1619 struct lnet_nid nid;
1621 if (LNET_CPT_NUMBER == 1)
1622 return 0; /* the only one */
1624 lnet_nid4_to_nid(nid4, &nid);
1625 return lnet_nid2cpt(&nid, ni);
1627 EXPORT_SYMBOL(lnet_cpt_of_nid);
1630 lnet_islocalnet_locked(__u32 net_id)
1632 struct lnet_net *net;
1635 net = lnet_get_net_locked(net_id);
1637 local = net != NULL;
1643 lnet_islocalnet(__u32 net_id)
1648 cpt = lnet_net_lock_current();
1650 local = lnet_islocalnet_locked(net_id);
1652 lnet_net_unlock(cpt);
1658 lnet_nid_to_ni_locked(struct lnet_nid *nid, int cpt)
1660 struct lnet_net *net;
1663 LASSERT(cpt != LNET_LOCK_EX);
1665 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1666 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1667 if (nid_same(&ni->ni_nid, nid))
1676 lnet_nid2ni_locked(lnet_nid_t nid4, int cpt)
1678 struct lnet_nid nid;
1680 lnet_nid4_to_nid(nid4, &nid);
1681 return lnet_nid_to_ni_locked(&nid, cpt);
1685 lnet_nid2ni_addref(lnet_nid_t nid4)
1688 struct lnet_nid nid;
1690 lnet_nid4_to_nid(nid4, &nid);
1693 ni = lnet_nid_to_ni_locked(&nid, 0);
1695 lnet_ni_addref_locked(ni, 0);
1700 EXPORT_SYMBOL(lnet_nid2ni_addref);
1703 lnet_nid_to_ni_addref(struct lnet_nid *nid)
1708 ni = lnet_nid_to_ni_locked(nid, 0);
1710 lnet_ni_addref_locked(ni, 0);
1715 EXPORT_SYMBOL(lnet_nid_to_ni_addref);
1718 lnet_islocalnid(struct lnet_nid *nid)
1723 cpt = lnet_net_lock_current();
1724 ni = lnet_nid_to_ni_locked(nid, cpt);
1725 lnet_net_unlock(cpt);
1731 lnet_count_acceptor_nets(void)
1733 /* Return the # of NIs that need the acceptor. */
1735 struct lnet_net *net;
1738 cpt = lnet_net_lock_current();
1739 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1740 /* all socklnd type networks should have the acceptor
1742 if (net->net_lnd->lnd_accept != NULL)
1746 lnet_net_unlock(cpt);
1751 struct lnet_ping_buffer *
1752 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1754 struct lnet_ping_buffer *pbuf;
1756 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1758 pbuf->pb_nnis = nnis;
1759 pbuf->pb_needs_post = false;
1760 atomic_set(&pbuf->pb_refcnt, 1);
1767 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1769 LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1770 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1773 static struct lnet_ping_buffer *
1774 lnet_ping_target_create(int nnis)
1776 struct lnet_ping_buffer *pbuf;
1778 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1780 CERROR("Can't allocate ping source [%d]\n", nnis);
1784 pbuf->pb_info.pi_nnis = nnis;
1785 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1786 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1787 pbuf->pb_info.pi_features =
1788 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1794 lnet_get_net_ni_count_locked(struct lnet_net *net)
1799 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1806 lnet_get_net_ni_count_pre(struct lnet_net *net)
1811 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1818 lnet_get_ni_count(void)
1821 struct lnet_net *net;
1826 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1827 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1837 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1839 struct lnet_ni_status *stat;
1843 __swab32s(&pbuf->pb_info.pi_magic);
1844 __swab32s(&pbuf->pb_info.pi_features);
1845 __swab32s(&pbuf->pb_info.pi_pid);
1846 __swab32s(&pbuf->pb_info.pi_nnis);
1847 nnis = pbuf->pb_info.pi_nnis;
1848 if (nnis > pbuf->pb_nnis)
1849 nnis = pbuf->pb_nnis;
1850 for (i = 0; i < nnis; i++) {
1851 stat = &pbuf->pb_info.pi_ni[i];
1852 __swab64s(&stat->ns_nid);
1853 __swab32s(&stat->ns_status);
1858 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1862 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1864 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1866 /* Loopback is guaranteed to be present */
1867 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1869 if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1875 lnet_ping_target_destroy(void)
1877 struct lnet_net *net;
1880 lnet_net_lock(LNET_LOCK_EX);
1882 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1883 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1885 ni->ni_status = NULL;
1890 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1891 the_lnet.ln_ping_target = NULL;
1893 lnet_net_unlock(LNET_LOCK_EX);
1897 lnet_ping_target_event_handler(struct lnet_event *event)
1899 struct lnet_ping_buffer *pbuf = event->md_user_ptr;
1901 if (event->unlinked)
1902 lnet_ping_buffer_decref(pbuf);
1906 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1907 struct lnet_handle_md *ping_mdh,
1908 int ni_count, bool set_eq)
1910 struct lnet_processid id = {
1911 .nid = LNET_ANY_NID,
1915 struct lnet_md md = { NULL };
1919 the_lnet.ln_ping_target_handler =
1920 lnet_ping_target_event_handler;
1922 *ppbuf = lnet_ping_target_create(ni_count);
1923 if (*ppbuf == NULL) {
1928 /* Ping target ME/MD */
1929 me = LNetMEAttach(LNET_RESERVED_PORTAL, &id,
1930 LNET_PROTO_PING_MATCHBITS, 0,
1931 LNET_UNLINK, LNET_INS_AFTER);
1934 CERROR("Can't create ping target ME: %d\n", rc);
1935 goto fail_decref_ping_buffer;
1938 /* initialize md content */
1939 md.start = &(*ppbuf)->pb_info;
1940 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1941 md.threshold = LNET_MD_THRESH_INF;
1943 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1944 LNET_MD_MANAGE_REMOTE;
1945 md.handler = the_lnet.ln_ping_target_handler;
1946 md.user_ptr = *ppbuf;
1948 rc = LNetMDAttach(me, &md, LNET_RETAIN, ping_mdh);
1950 CERROR("Can't attach ping target MD: %d\n", rc);
1951 goto fail_decref_ping_buffer;
1953 lnet_ping_buffer_addref(*ppbuf);
1957 fail_decref_ping_buffer:
1958 LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1959 lnet_ping_buffer_decref(*ppbuf);
1966 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1967 struct lnet_handle_md *ping_mdh)
1969 LNetMDUnlink(*ping_mdh);
1970 LNetInvalidateMDHandle(ping_mdh);
1972 /* NB the MD could be busy; this just starts the unlink */
1973 wait_var_event_warning(&pbuf->pb_refcnt,
1974 atomic_read(&pbuf->pb_refcnt) <= 1,
1975 "Still waiting for ping data MD to unlink\n");
1979 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1982 struct lnet_net *net;
1983 struct lnet_ni_status *ns;
1988 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1989 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1990 LASSERT(i < pbuf->pb_nnis);
1992 ns = &pbuf->pb_info.pi_ni[i];
1994 if (!nid_is_nid4(&ni->ni_nid))
1996 ns->ns_nid = lnet_nid_to_nid4(&ni->ni_nid);
1999 ns->ns_status = lnet_ni_get_status_locked(ni);
2007 * We (ab)use the ns_status of the loopback interface to
2008 * transmit the sequence number. The first interface listed
2009 * must be the loopback interface.
2011 rc = lnet_ping_info_validate(&pbuf->pb_info);
2013 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
2016 LNET_PING_BUFFER_SEQNO(pbuf) =
2017 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
2021 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
2022 struct lnet_handle_md ping_mdh)
2024 struct lnet_ping_buffer *old_pbuf = NULL;
2025 struct lnet_handle_md old_ping_md;
2027 /* switch the NIs to point to the new ping info created */
2028 lnet_net_lock(LNET_LOCK_EX);
2030 if (!the_lnet.ln_routing)
2031 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
2032 if (!lnet_peer_discovery_disabled)
2033 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
2035 /* Ensure only known feature bits have been set. */
2036 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
2037 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
2039 lnet_ping_target_install_locked(pbuf);
2041 if (the_lnet.ln_ping_target) {
2042 old_pbuf = the_lnet.ln_ping_target;
2043 old_ping_md = the_lnet.ln_ping_target_md;
2045 the_lnet.ln_ping_target_md = ping_mdh;
2046 the_lnet.ln_ping_target = pbuf;
2048 lnet_net_unlock(LNET_LOCK_EX);
2051 /* unlink and free the old ping info */
2052 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
2053 lnet_ping_buffer_decref(old_pbuf);
2056 lnet_push_update_to_peers(0);
2060 lnet_ping_target_fini(void)
2062 lnet_ping_md_unlink(the_lnet.ln_ping_target,
2063 &the_lnet.ln_ping_target_md);
2065 lnet_assert_handler_unused(the_lnet.ln_ping_target_handler);
2066 lnet_ping_target_destroy();
2069 /* Resize the push target. */
2070 int lnet_push_target_resize(void)
2072 struct lnet_handle_md mdh;
2073 struct lnet_handle_md old_mdh;
2074 struct lnet_ping_buffer *pbuf;
2075 struct lnet_ping_buffer *old_pbuf;
2080 nnis = the_lnet.ln_push_target_nnis;
2082 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
2086 /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
2087 * dropped when we need to resize again (see "old_pbuf" below) or when
2088 * LNet is shutdown (see lnet_push_target_fini())
2090 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
2092 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
2096 rc = lnet_push_target_post(pbuf, &mdh);
2098 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
2099 lnet_ping_buffer_decref(pbuf);
2103 lnet_net_lock(LNET_LOCK_EX);
2104 old_pbuf = the_lnet.ln_push_target;
2105 old_mdh = the_lnet.ln_push_target_md;
2106 the_lnet.ln_push_target = pbuf;
2107 the_lnet.ln_push_target_md = mdh;
2108 lnet_net_unlock(LNET_LOCK_EX);
2111 LNetMDUnlink(old_mdh);
2112 /* Drop ref set by lnet_ping_buffer_alloc() */
2113 lnet_ping_buffer_decref(old_pbuf);
2116 /* Received another push or reply that requires a larger buffer */
2117 if (nnis < the_lnet.ln_push_target_nnis)
2120 CDEBUG(D_NET, "nnis %d success\n", nnis);
2124 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
2125 struct lnet_handle_md *mdhp)
2127 struct lnet_processid id = { LNET_ANY_NID, LNET_PID_ANY };
2128 struct lnet_md md = { NULL };
2132 me = LNetMEAttach(LNET_RESERVED_PORTAL, &id,
2133 LNET_PROTO_PING_MATCHBITS, 0,
2134 LNET_UNLINK, LNET_INS_AFTER);
2137 CERROR("Can't create push target ME: %d\n", rc);
2141 pbuf->pb_needs_post = false;
2143 /* This reference is dropped by lnet_push_target_event_handler() */
2144 lnet_ping_buffer_addref(pbuf);
2146 /* initialize md content */
2147 md.start = &pbuf->pb_info;
2148 md.length = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2151 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
2153 md.handler = the_lnet.ln_push_target_handler;
2155 rc = LNetMDAttach(me, &md, LNET_UNLINK, mdhp);
2157 CERROR("Can't attach push MD: %d\n", rc);
2158 lnet_ping_buffer_decref(pbuf);
2159 pbuf->pb_needs_post = true;
2163 CDEBUG(D_NET, "posted push target %p\n", pbuf);
2168 static void lnet_push_target_event_handler(struct lnet_event *ev)
2170 struct lnet_ping_buffer *pbuf = ev->md_user_ptr;
2172 CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
2175 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2176 lnet_swap_pinginfo(pbuf);
2178 if (ev->type == LNET_EVENT_UNLINK) {
2179 /* Drop ref added by lnet_push_target_post() */
2180 lnet_ping_buffer_decref(pbuf);
2184 lnet_peer_push_event(ev);
2186 /* Drop ref added by lnet_push_target_post */
2187 lnet_ping_buffer_decref(pbuf);
2190 /* Initialize the push target. */
2191 static int lnet_push_target_init(void)
2195 if (the_lnet.ln_push_target)
2198 the_lnet.ln_push_target_handler =
2199 lnet_push_target_event_handler;
2201 rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2204 /* Start at the required minimum, we'll enlarge if required. */
2205 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2207 rc = lnet_push_target_resize();
2210 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2211 the_lnet.ln_push_target_handler = NULL;
2217 /* Clean up the push target. */
2218 static void lnet_push_target_fini(void)
2220 if (!the_lnet.ln_push_target)
2223 /* Unlink and invalidate to prevent new references. */
2224 LNetMDUnlink(the_lnet.ln_push_target_md);
2225 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2227 /* Wait for the unlink to complete. */
2228 wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2229 atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2230 "Still waiting for ping data MD to unlink\n");
2232 /* Drop ref set by lnet_ping_buffer_alloc() */
2233 lnet_ping_buffer_decref(the_lnet.ln_push_target);
2234 the_lnet.ln_push_target = NULL;
2235 the_lnet.ln_push_target_nnis = 0;
2237 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2238 lnet_assert_handler_unused(the_lnet.ln_push_target_handler);
2239 the_lnet.ln_push_target_handler = NULL;
2243 lnet_ni_tq_credits(struct lnet_ni *ni)
2247 LASSERT(ni->ni_ncpts >= 1);
2249 if (ni->ni_ncpts == 1)
2250 return ni->ni_net->net_tunables.lct_max_tx_credits;
2252 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2253 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2254 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2260 lnet_ni_unlink_locked(struct lnet_ni *ni)
2262 /* move it to zombie list and nobody can find it anymore */
2263 LASSERT(!list_empty(&ni->ni_netlist));
2264 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2265 lnet_ni_decref_locked(ni, 0);
2269 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2274 struct list_head *zombie_list = &net->net_ni_zombie;
2277 * Now wait for the NIs I just nuked to show up on the zombie
2278 * list and shut them down in guaranteed thread context
2281 while (!list_empty(zombie_list)) {
2285 ni = list_entry(zombie_list->next,
2286 struct lnet_ni, ni_netlist);
2287 list_del_init(&ni->ni_netlist);
2288 /* the ni should be in deleting state. If it's not it's
2290 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2291 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2294 /* still busy, add it back to zombie list */
2295 list_add(&ni->ni_netlist, zombie_list);
2299 if (!list_empty(&ni->ni_netlist)) {
2300 /* Unlock mutex while waiting to allow other
2301 * threads to read the LNet state and fall through
2304 lnet_net_unlock(LNET_LOCK_EX);
2305 mutex_unlock(&the_lnet.ln_api_mutex);
2308 if ((i & (-i)) == i) {
2310 "Waiting for zombie LNI %s\n",
2311 libcfs_nidstr(&ni->ni_nid));
2313 schedule_timeout_uninterruptible(cfs_time_seconds(1));
2315 mutex_lock(&the_lnet.ln_api_mutex);
2316 lnet_net_lock(LNET_LOCK_EX);
2320 lnet_net_unlock(LNET_LOCK_EX);
2322 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2324 LASSERT(!in_interrupt());
2325 /* Holding the mutex makes it safe for lnd_shutdown
2326 * to call module_put(). Module unload cannot finish
2327 * until lnet_unregister_lnd() completes, and that
2328 * requires the mutex.
2330 mutex_lock(&the_lnet.ln_lnd_mutex);
2331 (net->net_lnd->lnd_shutdown)(ni);
2332 mutex_unlock(&the_lnet.ln_lnd_mutex);
2335 CDEBUG(D_LNI, "Removed LNI %s\n",
2336 libcfs_nidstr(&ni->ni_nid));
2340 lnet_net_lock(LNET_LOCK_EX);
2344 /* shutdown down the NI and release refcount */
2346 lnet_shutdown_lndni(struct lnet_ni *ni)
2349 struct lnet_net *net = ni->ni_net;
2351 lnet_net_lock(LNET_LOCK_EX);
2353 ni->ni_state = LNET_NI_STATE_DELETING;
2355 lnet_ni_unlink_locked(ni);
2356 lnet_incr_dlc_seq();
2357 lnet_net_unlock(LNET_LOCK_EX);
2359 /* clear messages for this NI on the lazy portal */
2360 for (i = 0; i < the_lnet.ln_nportals; i++)
2361 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2363 lnet_net_lock(LNET_LOCK_EX);
2364 lnet_clear_zombies_nis_locked(net);
2365 lnet_net_unlock(LNET_LOCK_EX);
2369 lnet_shutdown_lndnet(struct lnet_net *net)
2373 lnet_net_lock(LNET_LOCK_EX);
2375 list_del_init(&net->net_list);
2377 while (!list_empty(&net->net_ni_list)) {
2378 ni = list_entry(net->net_ni_list.next,
2379 struct lnet_ni, ni_netlist);
2380 lnet_net_unlock(LNET_LOCK_EX);
2381 lnet_shutdown_lndni(ni);
2382 lnet_net_lock(LNET_LOCK_EX);
2385 lnet_net_unlock(LNET_LOCK_EX);
2387 /* Do peer table cleanup for this net */
2388 lnet_peer_tables_cleanup(net);
2394 lnet_shutdown_lndnets(void)
2396 struct lnet_net *net;
2398 struct lnet_msg *msg, *tmp;
2400 /* NB called holding the global mutex */
2402 /* All quiet on the API front */
2403 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2404 LASSERT(the_lnet.ln_refcount == 0);
2406 lnet_net_lock(LNET_LOCK_EX);
2407 the_lnet.ln_state = LNET_STATE_STOPPING;
2410 * move the nets to the zombie list to avoid them being
2411 * picked up for new work. LONET is also included in the
2412 * Nets that will be moved to the zombie list
2414 list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2416 /* Drop the cached loopback Net. */
2417 if (the_lnet.ln_loni != NULL) {
2418 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2419 the_lnet.ln_loni = NULL;
2421 lnet_net_unlock(LNET_LOCK_EX);
2423 /* iterate through the net zombie list and delete each net */
2424 while (!list_empty(&the_lnet.ln_net_zombie)) {
2425 net = list_entry(the_lnet.ln_net_zombie.next,
2426 struct lnet_net, net_list);
2427 lnet_shutdown_lndnet(net);
2430 spin_lock(&the_lnet.ln_msg_resend_lock);
2431 list_splice(&the_lnet.ln_msg_resend, &resend);
2432 spin_unlock(&the_lnet.ln_msg_resend_lock);
2434 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2435 list_del_init(&msg->msg_list);
2436 msg->msg_no_resend = true;
2437 lnet_finalize(msg, -ECANCELED);
2440 lnet_net_lock(LNET_LOCK_EX);
2441 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2442 lnet_net_unlock(LNET_LOCK_EX);
2446 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2449 struct lnet_tx_queue *tq;
2451 struct lnet_net *net = ni->ni_net;
2453 mutex_lock(&the_lnet.ln_lnd_mutex);
2456 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2457 ni->ni_lnd_tunables_set = true;
2460 rc = (net->net_lnd->lnd_startup)(ni);
2462 mutex_unlock(&the_lnet.ln_lnd_mutex);
2465 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2466 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2471 ni->ni_state = LNET_NI_STATE_ACTIVE;
2474 /* We keep a reference on the loopback net through the loopback NI */
2475 if (net->net_lnd->lnd_type == LOLND) {
2477 LASSERT(the_lnet.ln_loni == NULL);
2478 the_lnet.ln_loni = ni;
2479 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2480 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2481 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2482 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2486 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2487 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2488 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2489 libcfs_lnd2str(net->net_lnd->lnd_type),
2490 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2492 /* shutdown the NI since if we get here then it must've already
2495 lnet_shutdown_lndni(ni);
2499 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2500 tq->tq_credits_min =
2501 tq->tq_credits_max =
2502 tq->tq_credits = lnet_ni_tq_credits(ni);
2505 atomic_set(&ni->ni_tx_credits,
2506 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2507 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2509 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2510 libcfs_nidstr(&ni->ni_nid),
2511 ni->ni_net->net_tunables.lct_peer_tx_credits,
2512 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2513 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2514 ni->ni_net->net_tunables.lct_peer_timeout);
2523 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2526 struct lnet_net *net_l = NULL;
2527 LIST_HEAD(local_ni_list);
2531 const struct lnet_lnd *lnd;
2533 net->net_tunables.lct_peer_timeout;
2535 net->net_tunables.lct_max_tx_credits;
2536 int peerrtrcredits =
2537 net->net_tunables.lct_peer_rtr_credits;
2540 * make sure that this net is unique. If it isn't then
2541 * we are adding interfaces to an already existing network, and
2542 * 'net' is just a convenient way to pass in the list.
2543 * if it is unique we need to find the LND and load it if
2546 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2547 lnd_type = LNET_NETTYP(net->net_id);
2549 mutex_lock(&the_lnet.ln_lnd_mutex);
2550 lnd = lnet_find_lnd_by_type(lnd_type);
2553 mutex_unlock(&the_lnet.ln_lnd_mutex);
2554 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2555 mutex_lock(&the_lnet.ln_lnd_mutex);
2557 lnd = lnet_find_lnd_by_type(lnd_type);
2559 mutex_unlock(&the_lnet.ln_lnd_mutex);
2560 CERROR("Can't load LND %s, module %s, rc=%d\n",
2561 libcfs_lnd2str(lnd_type),
2562 libcfs_lnd2modname(lnd_type), rc);
2563 #ifndef HAVE_MODULE_LOADING_SUPPORT
2564 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2565 "compiled with kernel module "
2566 "loading support.");
2575 mutex_unlock(&the_lnet.ln_lnd_mutex);
2581 * net_l: if the network being added is unique then net_l
2582 * will point to that network
2583 * if the network being added is not unique then
2584 * net_l points to the existing network.
2586 * When we enter the loop below, we'll pick NIs off he
2587 * network beign added and start them up, then add them to
2588 * a local ni list. Once we've successfully started all
2589 * the NIs then we join the local NI list (of started up
2590 * networks) with the net_l->net_ni_list, which should
2591 * point to the correct network to add the new ni list to
2593 * If any of the new NIs fail to start up, then we want to
2594 * iterate through the local ni list, which should include
2595 * any NIs which were successfully started up, and shut
2598 * After than we want to delete the network being added,
2599 * to avoid a memory leak.
2601 while (!list_empty(&net->net_ni_added)) {
2602 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2604 list_del_init(&ni->ni_netlist);
2606 /* make sure that the the NI we're about to start
2607 * up is actually unique. if it's not fail. */
2608 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2609 ni->ni_interface)) {
2614 /* adjust the pointer the parent network, just in case it
2615 * the net is a duplicate */
2618 rc = lnet_startup_lndni(ni, tun);
2624 list_add_tail(&ni->ni_netlist, &local_ni_list);
2629 lnet_net_lock(LNET_LOCK_EX);
2630 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2631 lnet_incr_dlc_seq();
2632 lnet_net_unlock(LNET_LOCK_EX);
2634 /* if the network is not unique then we don't want to keep
2635 * it around after we're done. Free it. Otherwise add that
2636 * net to the global the_lnet.ln_nets */
2637 if (net_l != net && net_l != NULL) {
2639 * TODO - note. currently the tunables can not be updated
2645 * restore tunables after it has been overwitten by the
2648 if (peer_timeout != -1)
2649 net->net_tunables.lct_peer_timeout = peer_timeout;
2650 if (maxtxcredits != -1)
2651 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2652 if (peerrtrcredits != -1)
2653 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2655 lnet_net_lock(LNET_LOCK_EX);
2656 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2657 lnet_net_unlock(LNET_LOCK_EX);
2664 * shutdown the new NIs that are being started up
2665 * free the NET being started
2667 while (!list_empty(&local_ni_list)) {
2668 ni = list_entry(local_ni_list.next, struct lnet_ni,
2671 lnet_shutdown_lndni(ni);
2681 lnet_startup_lndnets(struct list_head *netlist)
2683 struct lnet_net *net;
2688 * Change to running state before bringing up the LNDs. This
2689 * allows lnet_shutdown_lndnets() to assert that we've passed
2692 lnet_net_lock(LNET_LOCK_EX);
2693 the_lnet.ln_state = LNET_STATE_RUNNING;
2694 lnet_net_unlock(LNET_LOCK_EX);
2696 while (!list_empty(netlist)) {
2697 net = list_entry(netlist->next, struct lnet_net, net_list);
2698 list_del_init(&net->net_list);
2700 rc = lnet_startup_lndnet(net, NULL);
2710 lnet_shutdown_lndnets();
2715 static int lnet_genl_parse_list(struct sk_buff *msg,
2716 const struct ln_key_list *data[], u16 idx)
2718 const struct ln_key_list *list = data[idx];
2719 const struct ln_key_props *props;
2720 struct nlattr *node;
2726 if (!list->lkl_maxattr)
2729 props = list->lkl_list;
2733 node = nla_nest_start(msg, LN_SCALAR_ATTR_LIST);
2737 for (count = 1; count <= list->lkl_maxattr; count++) {
2738 struct nlattr *key = nla_nest_start(msg, count);
2741 nla_put_u16(msg, LN_SCALAR_ATTR_LIST_SIZE,
2744 nla_put_u16(msg, LN_SCALAR_ATTR_INDEX, count);
2745 if (props[count].lkp_value)
2746 nla_put_string(msg, LN_SCALAR_ATTR_VALUE,
2747 props[count].lkp_value);
2748 if (props[count].lkp_key_format)
2749 nla_put_u16(msg, LN_SCALAR_ATTR_KEY_FORMAT,
2750 props[count].lkp_key_format);
2751 nla_put_u16(msg, LN_SCALAR_ATTR_NLA_TYPE,
2752 props[count].lkp_data_type);
2753 if (props[count].lkp_data_type == NLA_NESTED) {
2756 rc = lnet_genl_parse_list(msg, data, ++idx);
2762 nla_nest_end(msg, key);
2765 nla_nest_end(msg, node);
2769 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
2770 const struct genl_family *family, int flags,
2771 u8 cmd, const struct ln_key_list *data[])
2779 hdr = genlmsg_put(msg, portid, seq, family, flags, cmd);
2781 GOTO(canceled, rc = -EMSGSIZE);
2783 rc = lnet_genl_parse_list(msg, data, 0);
2787 genlmsg_end(msg, hdr);
2790 genlmsg_cancel(msg, hdr);
2791 return rc > 0 ? 0 : rc;
2793 EXPORT_SYMBOL(lnet_genl_send_scalar_list);
2796 * Initialize LNet library.
2798 * Automatically called at module loading time. Caller has to call
2799 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2800 * latter returned 0. It must be called exactly once.
2802 * \retval 0 on success
2803 * \retval -ve on failures.
2805 int lnet_lib_init(void)
2809 lnet_assert_wire_constants();
2811 /* refer to global cfs_cpt_table for now */
2812 the_lnet.ln_cpt_table = cfs_cpt_tab;
2813 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2815 LASSERT(the_lnet.ln_cpt_number > 0);
2816 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2817 /* we are under risk of consuming all lh_cookie */
2818 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2819 "please change setting of CPT-table and retry\n",
2820 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2824 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2825 the_lnet.ln_cpt_bits++;
2827 rc = lnet_create_locks();
2829 CERROR("Can't create LNet global locks: %d\n", rc);
2833 the_lnet.ln_refcount = 0;
2834 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2835 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2837 /* The hash table size is the number of bits it takes to express the set
2838 * ln_num_routes, minus 1 (better to under estimate than over so we
2839 * don't waste memory). */
2840 if (rnet_htable_size <= 0)
2841 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2842 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2843 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2844 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2845 order_base_2(rnet_htable_size) - 1);
2847 /* All LNDs apart from the LOLND are in separate modules. They
2848 * register themselves when their module loads, and unregister
2849 * themselves when their module is unloaded. */
2850 lnet_register_lnd(&the_lolnd);
2855 * Finalize LNet library.
2857 * \pre lnet_lib_init() called with success.
2858 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2860 * As this happens at module-unload, all lnds must already be unloaded,
2861 * so they must already be unregistered.
2863 void lnet_lib_exit(void)
2867 LASSERT(the_lnet.ln_refcount == 0);
2868 lnet_unregister_lnd(&the_lolnd);
2869 for (i = 0; i < NUM_LNDS; i++)
2870 LASSERT(!the_lnet.ln_lnds[i]);
2871 lnet_destroy_locks();
2875 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2877 * Users must call this function at least once before any other functions.
2878 * For each successful call there must be a corresponding call to
2879 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2882 * The PID used by LNet may be different from the one requested.
2885 * \param requested_pid PID requested by the caller.
2887 * \return >= 0 on success, and < 0 error code on failures.
2890 LNetNIInit(lnet_pid_t requested_pid)
2892 int im_a_router = 0;
2895 struct lnet_ping_buffer *pbuf;
2896 struct lnet_handle_md ping_mdh;
2897 LIST_HEAD(net_head);
2898 struct lnet_net *net;
2900 mutex_lock(&the_lnet.ln_api_mutex);
2902 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2904 if (the_lnet.ln_refcount > 0) {
2905 rc = the_lnet.ln_refcount++;
2906 mutex_unlock(&the_lnet.ln_api_mutex);
2910 rc = lnet_prepare(requested_pid);
2912 mutex_unlock(&the_lnet.ln_api_mutex);
2916 /* create a network for Loopback network */
2917 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2920 goto err_empty_list;
2923 /* Add in the loopback NI */
2924 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2926 goto err_empty_list;
2929 if (use_tcp_bonding)
2930 CWARN("use_tcp_bonding has been removed. Use Multi-Rail and Dynamic Discovery instead, see LU-13641\n");
2932 /* If LNet is being initialized via DLC it is possible
2933 * that the user requests not to load module parameters (ones which
2934 * are supported by DLC) on initialization. Therefore, make sure not
2935 * to load networks, routes and forwarding from module parameters
2936 * in this case. On cleanup in case of failure only clean up
2937 * routes if it has been loaded */
2938 if (!the_lnet.ln_nis_from_mod_params) {
2939 rc = lnet_parse_networks(&net_head, lnet_get_networks());
2941 goto err_empty_list;
2944 ni_count = lnet_startup_lndnets(&net_head);
2947 goto err_empty_list;
2950 if (!the_lnet.ln_nis_from_mod_params) {
2951 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2953 goto err_shutdown_lndnis;
2955 rc = lnet_rtrpools_alloc(im_a_router);
2957 goto err_destroy_routes;
2960 rc = lnet_acceptor_start();
2962 goto err_destroy_routes;
2964 the_lnet.ln_refcount = 1;
2965 /* Now I may use my own API functions... */
2967 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2969 goto err_acceptor_stop;
2971 lnet_ping_target_update(pbuf, ping_mdh);
2973 the_lnet.ln_mt_handler = lnet_mt_event_handler;
2975 rc = lnet_push_target_init();
2979 rc = lnet_peer_discovery_start();
2981 goto err_destroy_push_target;
2983 rc = lnet_monitor_thr_start();
2985 goto err_stop_discovery_thr;
2988 lnet_router_debugfs_init();
2990 mutex_unlock(&the_lnet.ln_api_mutex);
2992 complete_all(&the_lnet.ln_started);
2994 /* wait for all routers to start */
2995 lnet_wait_router_start();
2999 err_stop_discovery_thr:
3000 lnet_peer_discovery_stop();
3001 err_destroy_push_target:
3002 lnet_push_target_fini();
3004 lnet_ping_target_fini();
3006 the_lnet.ln_refcount = 0;
3007 lnet_acceptor_stop();
3009 if (!the_lnet.ln_nis_from_mod_params)
3010 lnet_destroy_routes();
3011 err_shutdown_lndnis:
3012 lnet_shutdown_lndnets();
3016 mutex_unlock(&the_lnet.ln_api_mutex);
3017 while (!list_empty(&net_head)) {
3018 struct lnet_net *net;
3020 net = list_entry(net_head.next, struct lnet_net, net_list);
3021 list_del_init(&net->net_list);
3026 EXPORT_SYMBOL(LNetNIInit);
3029 * Stop LNet interfaces, routing, and forwarding.
3031 * Users must call this function once for each successful call to LNetNIInit().
3032 * Once the LNetNIFini() operation has been started, the results of pending
3033 * API operations are undefined.
3035 * \return always 0 for current implementation.
3040 mutex_lock(&the_lnet.ln_api_mutex);
3042 LASSERT(the_lnet.ln_refcount > 0);
3044 if (the_lnet.ln_refcount != 1) {
3045 the_lnet.ln_refcount--;
3047 LASSERT(!the_lnet.ln_niinit_self);
3051 lnet_router_debugfs_fini();
3052 lnet_monitor_thr_stop();
3053 lnet_peer_discovery_stop();
3054 lnet_push_target_fini();
3055 lnet_ping_target_fini();
3057 /* Teardown fns that use my own API functions BEFORE here */
3058 the_lnet.ln_refcount = 0;
3060 lnet_acceptor_stop();
3061 lnet_destroy_routes();
3062 lnet_shutdown_lndnets();
3066 mutex_unlock(&the_lnet.ln_api_mutex);
3069 EXPORT_SYMBOL(LNetNIFini);
3072 * Grabs the ni data from the ni structure and fills the out
3075 * \param[in] ni network interface structure
3076 * \param[out] cfg_ni NI config information
3077 * \param[out] tun network and LND tunables
3080 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
3081 struct lnet_ioctl_config_lnd_tunables *tun,
3082 struct lnet_ioctl_element_stats *stats,
3085 size_t min_size = 0;
3088 if (!ni || !cfg_ni || !tun || !nid_is_nid4(&ni->ni_nid))
3091 if (ni->ni_interface != NULL) {
3092 strncpy(cfg_ni->lic_ni_intf,
3094 sizeof(cfg_ni->lic_ni_intf));
3097 cfg_ni->lic_nid = lnet_nid_to_nid4(&ni->ni_nid);
3098 cfg_ni->lic_status = lnet_ni_get_status_locked(ni);
3099 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
3101 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
3104 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
3105 LNET_STATS_TYPE_SEND);
3106 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
3107 LNET_STATS_TYPE_RECV);
3108 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
3109 LNET_STATS_TYPE_DROP);
3113 * tun->lt_tun will always be present, but in order to be
3114 * backwards compatible, we need to deal with the cases when
3115 * tun->lt_tun is smaller than what the kernel has, because it
3116 * comes from an older version of a userspace program, then we'll
3117 * need to copy as much information as we have available space.
3119 min_size = tun_size - sizeof(tun->lt_cmn);
3120 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
3122 /* copy over the cpts */
3123 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
3124 ni->ni_cpts == NULL) {
3125 for (i = 0; i < ni->ni_ncpts; i++)
3126 cfg_ni->lic_cpts[i] = i;
3129 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
3130 i < LNET_MAX_SHOW_NUM_CPT;
3132 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
3134 cfg_ni->lic_ncpts = ni->ni_ncpts;
3138 * NOTE: This is a legacy function left in the code to be backwards
3139 * compatible with older userspace programs. It should eventually be
3142 * Grabs the ni data from the ni structure and fills the out
3145 * \param[in] ni network interface structure
3146 * \param[out] config config information
3149 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
3150 struct lnet_ioctl_config_data *config)
3152 struct lnet_ioctl_net_config *net_config;
3153 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
3154 size_t min_size, tunable_size = 0;
3157 if (!ni || !config || !nid_is_nid4(&ni->ni_nid))
3160 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
3164 if (!ni->ni_interface)
3167 strncpy(net_config->ni_interface,
3169 sizeof(net_config->ni_interface));
3171 config->cfg_nid = lnet_nid_to_nid4(&ni->ni_nid);
3172 config->cfg_config_u.cfg_net.net_peer_timeout =
3173 ni->ni_net->net_tunables.lct_peer_timeout;
3174 config->cfg_config_u.cfg_net.net_max_tx_credits =
3175 ni->ni_net->net_tunables.lct_max_tx_credits;
3176 config->cfg_config_u.cfg_net.net_peer_tx_credits =
3177 ni->ni_net->net_tunables.lct_peer_tx_credits;
3178 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
3179 ni->ni_net->net_tunables.lct_peer_rtr_credits;
3181 net_config->ni_status = lnet_ni_get_status_locked(ni);
3184 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
3186 for (i = 0; i < num_cpts; i++)
3187 net_config->ni_cpts[i] = ni->ni_cpts[i];
3189 config->cfg_ncpts = num_cpts;
3193 * See if user land tools sent in a newer and larger version
3194 * of struct lnet_tunables than what the kernel uses.
3196 min_size = sizeof(*config) + sizeof(*net_config);
3198 if (config->cfg_hdr.ioc_len > min_size)
3199 tunable_size = config->cfg_hdr.ioc_len - min_size;
3201 /* Don't copy too much data to user space */
3202 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
3203 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
3205 if (lnd_cfg && min_size) {
3206 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
3207 config->cfg_config_u.cfg_net.net_interface_count = 1;
3209 /* Tell user land that kernel side has less data */
3210 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
3211 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
3212 config->cfg_hdr.ioc_len -= min_size;
3218 lnet_get_ni_idx_locked(int idx)
3221 struct lnet_net *net;
3223 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3224 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3233 int lnet_get_net_healthv_locked(struct lnet_net *net)
3236 int best_healthv = 0;
3237 int healthv, ni_fatal;
3239 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3240 healthv = atomic_read(&ni->ni_healthv);
3241 ni_fatal = atomic_read(&ni->ni_fatal_error_on);
3242 if (!ni_fatal && healthv > best_healthv)
3243 best_healthv = healthv;
3246 return best_healthv;
3250 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
3253 struct lnet_net *net = mynet;
3256 * It is possible that the net has been cleaned out while there is
3257 * a message being sent. This function accessed the net without
3258 * checking if the list is empty
3262 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3264 if (list_empty(&net->net_ni_list))
3266 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3272 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3273 /* if you reached the end of the ni list and the net is
3274 * specified, then there are no more nis in that net */
3278 /* we reached the end of this net ni list. move to the
3280 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3281 /* no more nets and no more NIs. */
3284 /* get the next net */
3285 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3287 if (list_empty(&net->net_ni_list))
3289 /* get the ni on it */
3290 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3296 if (list_empty(&prev->ni_netlist))
3299 /* there are more nis left */
3300 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3306 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3311 int idx = config->cfg_count;
3313 cpt = lnet_net_lock_current();
3315 ni = lnet_get_ni_idx_locked(idx);
3320 lnet_fill_ni_info_legacy(ni, config);
3324 lnet_net_unlock(cpt);
3329 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3330 struct lnet_ioctl_config_lnd_tunables *tun,
3331 struct lnet_ioctl_element_stats *stats,
3338 if (!cfg_ni || !tun || !stats)
3341 cpt = lnet_net_lock_current();
3343 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3348 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3352 lnet_net_unlock(cpt);
3356 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3365 cpt = lnet_net_lock_current();
3367 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3370 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3374 lnet_net_unlock(cpt);
3379 static int lnet_add_net_common(struct lnet_net *net,
3380 struct lnet_ioctl_config_lnd_tunables *tun)
3382 struct lnet_handle_md ping_mdh;
3383 struct lnet_ping_buffer *pbuf;
3384 struct lnet_remotenet *rnet;
3390 lnet_net_lock(LNET_LOCK_EX);
3391 rnet = lnet_find_rnet_locked(net->net_id);
3392 lnet_net_unlock(LNET_LOCK_EX);
3394 * make sure that the net added doesn't invalidate the current
3395 * configuration LNet is keeping
3398 CERROR("Adding net %s will invalidate routing configuration\n",
3399 libcfs_net2str(net->net_id));
3405 * make sure you calculate the correct number of slots in the ping
3406 * buffer. Since the ping info is a flattened list of all the NIs,
3407 * we should allocate enough slots to accomodate the number of NIs
3408 * which will be added.
3410 * since ni hasn't been configured yet, use
3411 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3413 net_ni_count = lnet_get_net_ni_count_pre(net);
3415 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3416 net_ni_count + lnet_get_ni_count(),
3424 memcpy(&net->net_tunables,
3425 &tun->lt_cmn, sizeof(net->net_tunables));
3427 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3429 net_id = net->net_id;
3431 rc = lnet_startup_lndnet(net,
3432 (tun) ? &tun->lt_tun : NULL);
3436 lnet_net_lock(LNET_LOCK_EX);
3437 net = lnet_get_net_locked(net_id);
3440 /* apply the UDSPs */
3441 rc = lnet_udsp_apply_policies_on_net(net);
3443 CERROR("Failed to apply UDSPs on local net %s\n",
3444 libcfs_net2str(net->net_id));
3446 /* At this point we lost track of which NI was just added, so we
3447 * just re-apply the policies on all of the NIs on this net
3449 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3450 rc = lnet_udsp_apply_policies_on_ni(ni);
3452 CERROR("Failed to apply UDSPs on ni %s\n",
3453 libcfs_nidstr(&ni->ni_nid));
3455 lnet_net_unlock(LNET_LOCK_EX);
3458 * Start the acceptor thread if this is the first network
3459 * being added that requires the thread.
3461 if (net->net_lnd->lnd_accept) {
3462 rc = lnet_acceptor_start();
3464 /* shutdown the net that we just started */
3465 CERROR("Failed to start up acceptor thread\n");
3466 lnet_shutdown_lndnet(net);
3471 lnet_net_lock(LNET_LOCK_EX);
3472 lnet_peer_net_added(net);
3473 lnet_net_unlock(LNET_LOCK_EX);
3475 lnet_ping_target_update(pbuf, ping_mdh);
3480 lnet_ping_md_unlink(pbuf, &ping_mdh);
3481 lnet_ping_buffer_decref(pbuf);
3486 lnet_set_tune_defaults(struct lnet_ioctl_config_lnd_tunables *tun)
3489 if (!tun->lt_cmn.lct_peer_timeout)
3490 tun->lt_cmn.lct_peer_timeout = DEFAULT_PEER_TIMEOUT;
3491 if (!tun->lt_cmn.lct_peer_tx_credits)
3492 tun->lt_cmn.lct_peer_tx_credits = DEFAULT_PEER_CREDITS;
3493 if (!tun->lt_cmn.lct_max_tx_credits)
3494 tun->lt_cmn.lct_max_tx_credits = DEFAULT_CREDITS;
3498 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3499 struct lnet_ioctl_config_lnd_tunables *tun)
3501 struct lnet_net *net;
3504 LIST_HEAD(net_head);
3506 rc = lnet_parse_ip2nets(&nets, ip2nets);
3510 rc = lnet_parse_networks(&net_head, nets);
3514 lnet_set_tune_defaults(tun);
3516 mutex_lock(&the_lnet.ln_api_mutex);
3517 while (!list_empty(&net_head)) {
3518 net = list_entry(net_head.next, struct lnet_net, net_list);
3519 list_del_init(&net->net_list);
3520 rc = lnet_add_net_common(net, tun);
3526 mutex_unlock(&the_lnet.ln_api_mutex);
3528 while (!list_empty(&net_head)) {
3529 net = list_entry(net_head.next, struct lnet_net, net_list);
3530 list_del_init(&net->net_list);
3536 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3538 struct lnet_net *net;
3540 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3542 __u32 net_id, lnd_type;
3544 /* get the tunables if they are available */
3545 if (conf->lic_cfg_hdr.ioc_len >=
3546 sizeof(*conf) + sizeof(*tun))
3547 tun = (struct lnet_ioctl_config_lnd_tunables *)
3550 /* handle legacy ip2nets from DLC */
3551 if (conf->lic_legacy_ip2nets[0] != '\0')
3552 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3555 net_id = LNET_NIDNET(conf->lic_nid);
3556 lnd_type = LNET_NETTYP(net_id);
3558 if (!libcfs_isknown_lnd(lnd_type)) {
3559 CERROR("No valid net and lnd information provided\n");
3563 net = lnet_net_alloc(net_id, NULL);
3567 for (i = 0; i < conf->lic_ncpts; i++) {
3568 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3572 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3577 lnet_set_tune_defaults(tun);
3579 mutex_lock(&the_lnet.ln_api_mutex);
3581 rc = lnet_add_net_common(net, tun);
3583 mutex_unlock(&the_lnet.ln_api_mutex);
3588 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3590 struct lnet_net *net;
3592 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3593 struct lnet_ping_buffer *pbuf;
3594 struct lnet_handle_md ping_mdh;
3599 /* don't allow userspace to shutdown the LOLND */
3600 if (LNET_NETTYP(net_id) == LOLND)
3603 mutex_lock(&the_lnet.ln_api_mutex);
3607 net = lnet_get_net_locked(net_id);
3609 CERROR("net %s not found\n",
3610 libcfs_net2str(net_id));
3615 addr = LNET_NIDADDR(conf->lic_nid);
3617 /* remove the entire net */
3618 net_count = lnet_get_net_ni_count_locked(net);
3622 /* create and link a new ping info, before removing the old one */
3623 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3624 lnet_get_ni_count() - net_count,
3627 goto unlock_api_mutex;
3629 lnet_shutdown_lndnet(net);
3631 lnet_acceptor_stop();
3633 lnet_ping_target_update(pbuf, ping_mdh);
3635 goto unlock_api_mutex;
3638 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3640 CERROR("nid %s not found\n",
3641 libcfs_nid2str(conf->lic_nid));
3646 net_count = lnet_get_net_ni_count_locked(net);
3650 /* create and link a new ping info, before removing the old one */
3651 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3652 lnet_get_ni_count() - 1, false);
3654 goto unlock_api_mutex;
3656 lnet_shutdown_lndni(ni);
3658 lnet_acceptor_stop();
3660 lnet_ping_target_update(pbuf, ping_mdh);
3662 /* check if the net is empty and remove it if it is */
3664 lnet_shutdown_lndnet(net);
3666 goto unlock_api_mutex;
3671 mutex_unlock(&the_lnet.ln_api_mutex);
3677 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3678 * They are only expected to be called for unique networks.
3679 * That can be as a result of older DLC library
3680 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3683 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3685 struct lnet_net *net;
3686 LIST_HEAD(net_head);
3688 struct lnet_ioctl_config_lnd_tunables tun;
3689 const char *nets = conf->cfg_config_u.cfg_net.net_intf;
3691 /* Create a net/ni structures for the network string */
3692 rc = lnet_parse_networks(&net_head, nets);
3694 return rc == 0 ? -EINVAL : rc;
3696 mutex_lock(&the_lnet.ln_api_mutex);
3699 rc = -EINVAL; /* only add one network per call */
3700 goto out_unlock_clean;
3703 net = list_entry(net_head.next, struct lnet_net, net_list);
3704 list_del_init(&net->net_list);
3706 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3708 memset(&tun, 0, sizeof(tun));
3710 tun.lt_cmn.lct_peer_timeout =
3711 (!conf->cfg_config_u.cfg_net.net_peer_timeout) ? DEFAULT_PEER_TIMEOUT :
3712 conf->cfg_config_u.cfg_net.net_peer_timeout;
3713 tun.lt_cmn.lct_peer_tx_credits =
3714 (!conf->cfg_config_u.cfg_net.net_peer_tx_credits) ? DEFAULT_PEER_CREDITS :
3715 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3716 tun.lt_cmn.lct_peer_rtr_credits =
3717 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3718 tun.lt_cmn.lct_max_tx_credits =
3719 (!conf->cfg_config_u.cfg_net.net_max_tx_credits) ? DEFAULT_CREDITS :
3720 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3722 rc = lnet_add_net_common(net, &tun);
3725 mutex_unlock(&the_lnet.ln_api_mutex);
3726 while (!list_empty(&net_head)) {
3727 /* net_head list is empty in success case */
3728 net = list_entry(net_head.next, struct lnet_net, net_list);
3729 list_del_init(&net->net_list);
3736 lnet_dyn_del_net(__u32 net_id)
3738 struct lnet_net *net;
3739 struct lnet_ping_buffer *pbuf;
3740 struct lnet_handle_md ping_mdh;
3744 /* don't allow userspace to shutdown the LOLND */
3745 if (LNET_NETTYP(net_id) == LOLND)
3748 mutex_lock(&the_lnet.ln_api_mutex);
3752 net = lnet_get_net_locked(net_id);
3759 net_ni_count = lnet_get_net_ni_count_locked(net);
3763 /* create and link a new ping info, before removing the old one */
3764 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3765 lnet_get_ni_count() - net_ni_count, false);
3769 lnet_shutdown_lndnet(net);
3771 lnet_acceptor_stop();
3773 lnet_ping_target_update(pbuf, ping_mdh);
3776 mutex_unlock(&the_lnet.ln_api_mutex);
3781 void lnet_incr_dlc_seq(void)
3783 atomic_inc(&lnet_dlc_seq_no);
3786 __u32 lnet_get_dlc_seq_locked(void)
3788 return atomic_read(&lnet_dlc_seq_no);
3792 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3794 struct lnet_net *net;
3797 lnet_net_lock(LNET_LOCK_EX);
3798 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3799 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3800 if (all || (nid_is_nid4(&ni->ni_nid) &&
3801 lnet_nid_to_nid4(&ni->ni_nid) == nid)) {
3802 atomic_set(&ni->ni_healthv, value);
3803 if (list_empty(&ni->ni_recovery) &&
3804 value < LNET_MAX_HEALTH_VALUE) {
3805 CERROR("manually adding local NI %s to recovery\n",
3806 libcfs_nidstr(&ni->ni_nid));
3807 list_add_tail(&ni->ni_recovery,
3808 &the_lnet.ln_mt_localNIRecovq);
3809 lnet_ni_addref_locked(ni, 0);
3812 lnet_net_unlock(LNET_LOCK_EX);
3818 lnet_net_unlock(LNET_LOCK_EX);
3822 lnet_ni_set_conns_per_peer(lnet_nid_t nid, int value, bool all)
3824 struct lnet_net *net;
3827 lnet_net_lock(LNET_LOCK_EX);
3828 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3829 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3830 if (lnet_nid_to_nid4(&ni->ni_nid) != nid && !all)
3832 if (LNET_NETTYP(net->net_id) == SOCKLND)
3833 ni->ni_lnd_tunables.lnd_tun_u.lnd_sock.lnd_conns_per_peer = value;
3834 else if (LNET_NETTYP(net->net_id) == O2IBLND)
3835 ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib.lnd_conns_per_peer = value;
3837 lnet_net_unlock(LNET_LOCK_EX);
3842 lnet_net_unlock(LNET_LOCK_EX);
3846 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3850 lnet_nid_t nid = stats->hlni_nid;
3852 cpt = lnet_net_lock_current();
3853 ni = lnet_nid2ni_locked(nid, cpt);
3860 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3861 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3862 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3863 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3864 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3865 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3866 stats->hlni_fatal_error = atomic_read(&ni->ni_fatal_error_on);
3867 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3868 stats->hlni_ping_count = ni->ni_ping_count;
3869 stats->hlni_next_ping = ni->ni_next_ping;
3872 lnet_net_unlock(cpt);
3878 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3883 lnet_net_lock(LNET_LOCK_EX);
3884 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3885 if (!nid_is_nid4(&ni->ni_nid))
3887 list->rlst_nid_array[i] = lnet_nid_to_nid4(&ni->ni_nid);
3889 if (i >= LNET_MAX_SHOW_NUM_NID)
3892 lnet_net_unlock(LNET_LOCK_EX);
3893 list->rlst_num_nids = i;
3899 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3901 struct lnet_peer_ni *lpni;
3904 lnet_net_lock(LNET_LOCK_EX);
3905 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3906 list->rlst_nid_array[i] = lnet_nid_to_nid4(&lpni->lpni_nid);
3908 if (i >= LNET_MAX_SHOW_NUM_NID)
3911 lnet_net_unlock(LNET_LOCK_EX);
3912 list->rlst_num_nids = i;
3918 * LNet ioctl handler.
3922 LNetCtl(unsigned int cmd, void *arg)
3924 struct libcfs_ioctl_data *data = arg;
3925 struct lnet_ioctl_config_data *config;
3926 struct lnet_process_id id4 = {};
3927 struct lnet_processid id = {};
3929 struct lnet_nid nid;
3932 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3933 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3936 case IOC_LIBCFS_GET_NI:
3937 rc = LNetGetId(data->ioc_count, &id);
3938 data->ioc_nid = lnet_nid_to_nid4(&id.nid);
3941 case IOC_LIBCFS_FAIL_NID:
3942 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3944 case IOC_LIBCFS_ADD_ROUTE: {
3945 /* default router sensitivity to 1 */
3946 unsigned int sensitivity = 1;
3949 if (config->cfg_hdr.ioc_len < sizeof(*config))
3952 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3954 config->cfg_config_u.cfg_route.rtr_sensitivity;
3957 lnet_nid4_to_nid(config->cfg_nid, &nid);
3958 mutex_lock(&the_lnet.ln_api_mutex);
3959 rc = lnet_add_route(config->cfg_net,
3960 config->cfg_config_u.cfg_route.rtr_hop,
3962 config->cfg_config_u.cfg_route.
3963 rtr_priority, sensitivity);
3964 mutex_unlock(&the_lnet.ln_api_mutex);
3968 case IOC_LIBCFS_DEL_ROUTE:
3971 if (config->cfg_hdr.ioc_len < sizeof(*config))
3974 lnet_nid4_to_nid(config->cfg_nid, &nid);
3975 mutex_lock(&the_lnet.ln_api_mutex);
3976 rc = lnet_del_route(config->cfg_net, &nid);
3977 mutex_unlock(&the_lnet.ln_api_mutex);
3980 case IOC_LIBCFS_GET_ROUTE:
3983 if (config->cfg_hdr.ioc_len < sizeof(*config))
3986 mutex_lock(&the_lnet.ln_api_mutex);
3987 rc = lnet_get_route(config->cfg_count,
3989 &config->cfg_config_u.cfg_route.rtr_hop,
3991 &config->cfg_config_u.cfg_route.rtr_flags,
3992 &config->cfg_config_u.cfg_route.
3994 &config->cfg_config_u.cfg_route.
3996 mutex_unlock(&the_lnet.ln_api_mutex);
3999 case IOC_LIBCFS_GET_LOCAL_NI: {
4000 struct lnet_ioctl_config_ni *cfg_ni;
4001 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
4002 struct lnet_ioctl_element_stats *stats;
4007 /* get the tunables if they are available */
4008 if (cfg_ni->lic_cfg_hdr.ioc_len <
4009 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
4012 stats = (struct lnet_ioctl_element_stats *)
4014 tun = (struct lnet_ioctl_config_lnd_tunables *)
4015 (cfg_ni->lic_bulk + sizeof(*stats));
4017 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
4020 mutex_lock(&the_lnet.ln_api_mutex);
4021 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
4022 mutex_unlock(&the_lnet.ln_api_mutex);
4026 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
4027 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
4029 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
4032 mutex_lock(&the_lnet.ln_api_mutex);
4033 rc = lnet_get_ni_stats(msg_stats);
4034 mutex_unlock(&the_lnet.ln_api_mutex);
4039 case IOC_LIBCFS_GET_NET: {
4040 size_t total = sizeof(*config) +
4041 sizeof(struct lnet_ioctl_net_config);
4044 if (config->cfg_hdr.ioc_len < total)
4047 mutex_lock(&the_lnet.ln_api_mutex);
4048 rc = lnet_get_net_config(config);
4049 mutex_unlock(&the_lnet.ln_api_mutex);
4053 case IOC_LIBCFS_GET_LNET_STATS:
4055 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
4057 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
4060 mutex_lock(&the_lnet.ln_api_mutex);
4061 rc = lnet_counters_get(&lnet_stats->st_cntrs);
4062 mutex_unlock(&the_lnet.ln_api_mutex);
4066 case IOC_LIBCFS_RESET_LNET_STATS:
4068 mutex_lock(&the_lnet.ln_api_mutex);
4069 lnet_counters_reset();
4070 mutex_unlock(&the_lnet.ln_api_mutex);
4074 case IOC_LIBCFS_CONFIG_RTR:
4077 if (config->cfg_hdr.ioc_len < sizeof(*config))
4080 mutex_lock(&the_lnet.ln_api_mutex);
4081 if (config->cfg_config_u.cfg_buffers.buf_enable) {
4082 rc = lnet_rtrpools_enable();
4083 mutex_unlock(&the_lnet.ln_api_mutex);
4086 lnet_rtrpools_disable();
4087 mutex_unlock(&the_lnet.ln_api_mutex);
4090 case IOC_LIBCFS_ADD_BUF:
4093 if (config->cfg_hdr.ioc_len < sizeof(*config))
4096 mutex_lock(&the_lnet.ln_api_mutex);
4097 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
4099 config->cfg_config_u.cfg_buffers.
4101 config->cfg_config_u.cfg_buffers.
4103 mutex_unlock(&the_lnet.ln_api_mutex);
4106 case IOC_LIBCFS_SET_NUMA_RANGE: {
4107 struct lnet_ioctl_set_value *numa;
4109 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4111 lnet_net_lock(LNET_LOCK_EX);
4112 lnet_numa_range = numa->sv_value;
4113 lnet_net_unlock(LNET_LOCK_EX);
4117 case IOC_LIBCFS_GET_NUMA_RANGE: {
4118 struct lnet_ioctl_set_value *numa;
4120 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4122 numa->sv_value = lnet_numa_range;
4126 case IOC_LIBCFS_GET_BUF: {
4127 struct lnet_ioctl_pool_cfg *pool_cfg;
4128 size_t total = sizeof(*config) + sizeof(*pool_cfg);
4132 if (config->cfg_hdr.ioc_len < total)
4135 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
4137 mutex_lock(&the_lnet.ln_api_mutex);
4138 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
4139 mutex_unlock(&the_lnet.ln_api_mutex);
4143 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
4144 struct lnet_ioctl_local_ni_hstats *stats = arg;
4146 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
4149 mutex_lock(&the_lnet.ln_api_mutex);
4150 rc = lnet_get_local_ni_hstats(stats);
4151 mutex_unlock(&the_lnet.ln_api_mutex);
4156 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
4157 struct lnet_ioctl_recovery_list *list = arg;
4158 if (list->rlst_hdr.ioc_len < sizeof(*list))
4161 mutex_lock(&the_lnet.ln_api_mutex);
4162 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
4163 rc = lnet_get_local_ni_recovery_list(list);
4165 rc = lnet_get_peer_ni_recovery_list(list);
4166 mutex_unlock(&the_lnet.ln_api_mutex);
4170 case IOC_LIBCFS_ADD_PEER_NI: {
4171 struct lnet_ioctl_peer_cfg *cfg = arg;
4173 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4176 mutex_lock(&the_lnet.ln_api_mutex);
4177 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
4179 cfg->prcfg_mr, false);
4180 mutex_unlock(&the_lnet.ln_api_mutex);
4184 case IOC_LIBCFS_DEL_PEER_NI: {
4185 struct lnet_ioctl_peer_cfg *cfg = arg;
4187 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4190 mutex_lock(&the_lnet.ln_api_mutex);
4191 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
4192 cfg->prcfg_cfg_nid);
4193 mutex_unlock(&the_lnet.ln_api_mutex);
4197 case IOC_LIBCFS_GET_PEER_INFO: {
4198 struct lnet_ioctl_peer *peer_info = arg;
4200 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
4203 mutex_lock(&the_lnet.ln_api_mutex);
4204 rc = lnet_get_peer_ni_info(
4205 peer_info->pr_count,
4207 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
4208 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
4209 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
4210 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
4211 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
4212 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
4213 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
4214 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
4215 mutex_unlock(&the_lnet.ln_api_mutex);
4219 case IOC_LIBCFS_GET_PEER_NI: {
4220 struct lnet_ioctl_peer_cfg *cfg = arg;
4222 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4225 mutex_lock(&the_lnet.ln_api_mutex);
4226 rc = lnet_get_peer_info(cfg,
4227 (void __user *)cfg->prcfg_bulk);
4228 mutex_unlock(&the_lnet.ln_api_mutex);
4232 case IOC_LIBCFS_GET_PEER_LIST: {
4233 struct lnet_ioctl_peer_cfg *cfg = arg;
4235 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4238 mutex_lock(&the_lnet.ln_api_mutex);
4239 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
4240 (struct lnet_process_id __user *)cfg->prcfg_bulk);
4241 mutex_unlock(&the_lnet.ln_api_mutex);
4245 case IOC_LIBCFS_SET_HEALHV: {
4246 struct lnet_ioctl_reset_health_cfg *cfg = arg;
4248 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
4250 if (cfg->rh_value < 0 ||
4251 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
4252 value = LNET_MAX_HEALTH_VALUE;
4254 value = cfg->rh_value;
4255 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
4256 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
4257 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
4258 mutex_lock(&the_lnet.ln_api_mutex);
4259 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
4260 lnet_ni_set_healthv(cfg->rh_nid, value,
4263 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
4265 mutex_unlock(&the_lnet.ln_api_mutex);
4269 case IOC_LIBCFS_SET_CONNS_PER_PEER: {
4270 struct lnet_ioctl_reset_conns_per_peer_cfg *cfg = arg;
4273 if (cfg->rcpp_hdr.ioc_len < sizeof(*cfg))
4275 if (cfg->rcpp_value < 0)
4278 value = cfg->rcpp_value;
4280 "Setting conns_per_peer to %d for %s. all = %d\n",
4281 value, libcfs_nid2str(cfg->rcpp_nid), cfg->rcpp_all);
4282 mutex_lock(&the_lnet.ln_api_mutex);
4283 lnet_ni_set_conns_per_peer(cfg->rcpp_nid, value, cfg->rcpp_all);
4284 mutex_unlock(&the_lnet.ln_api_mutex);
4288 case IOC_LIBCFS_NOTIFY_ROUTER: {
4289 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
4291 /* The deadline passed in by the user should be some time in
4292 * seconds in the future since the UNIX epoch. We have to map
4293 * that deadline to the wall clock.
4295 deadline += ktime_get_seconds();
4296 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
4300 case IOC_LIBCFS_LNET_DIST:
4301 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
4302 if (rc < 0 && rc != -EHOSTUNREACH)
4305 data->ioc_u32[0] = rc;
4308 case IOC_LIBCFS_TESTPROTOCOMPAT:
4309 the_lnet.ln_testprotocompat = data->ioc_flags;
4312 case IOC_LIBCFS_LNET_FAULT:
4313 return lnet_fault_ctl(data->ioc_flags, data);
4315 case IOC_LIBCFS_PING: {
4316 signed long timeout;
4318 id4.nid = data->ioc_nid;
4319 id4.pid = data->ioc_u32[0];
4321 /* If timeout is negative then set default of 3 minutes */
4322 if (((s32)data->ioc_u32[1] <= 0) ||
4323 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4324 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4326 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
4328 rc = lnet_ping(id4, &LNET_ANY_NID, timeout, data->ioc_pbuf1,
4329 data->ioc_plen1 / sizeof(struct lnet_process_id));
4334 data->ioc_count = rc;
4338 case IOC_LIBCFS_PING_PEER: {
4339 struct lnet_ioctl_ping_data *ping = arg;
4340 struct lnet_nid src_nid = LNET_ANY_NID;
4341 struct lnet_peer *lp;
4342 signed long timeout;
4344 /* Check if the supplied ping data supports source nid
4345 * NB: This check is sufficient if lnet_ioctl_ping_data has
4346 * additional fields added, but if they are re-ordered or
4347 * fields removed then this will break. It is expected that
4348 * these ioctls will be replaced with netlink implementation, so
4349 * it is probably not worth coming up with a more robust version
4350 * compatibility scheme.
4352 if (ping->ping_hdr.ioc_len >= sizeof(struct lnet_ioctl_ping_data))
4353 lnet_nid4_to_nid(ping->ping_src, &src_nid);
4355 /* If timeout is negative then set default of 3 minutes */
4356 if (((s32)ping->op_param) <= 0 ||
4357 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4358 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4360 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
4362 rc = lnet_ping(ping->ping_id, &src_nid, timeout,
4368 mutex_lock(&the_lnet.ln_api_mutex);
4369 lp = lnet_find_peer4(ping->ping_id.nid);
4372 lnet_nid_to_nid4(&lp->lp_primary_nid);
4373 ping->mr_info = lnet_peer_is_multi_rail(lp);
4374 lnet_peer_decref_locked(lp);
4376 mutex_unlock(&the_lnet.ln_api_mutex);
4378 ping->ping_count = rc;
4382 case IOC_LIBCFS_DISCOVER: {
4383 struct lnet_ioctl_ping_data *discover = arg;
4384 struct lnet_peer *lp;
4386 rc = lnet_discover(discover->ping_id, discover->op_param,
4388 discover->ping_count);
4392 mutex_lock(&the_lnet.ln_api_mutex);
4393 lp = lnet_find_peer4(discover->ping_id.nid);
4395 discover->ping_id.nid =
4396 lnet_nid_to_nid4(&lp->lp_primary_nid);
4397 discover->mr_info = lnet_peer_is_multi_rail(lp);
4398 lnet_peer_decref_locked(lp);
4400 mutex_unlock(&the_lnet.ln_api_mutex);
4402 discover->ping_count = rc;
4406 case IOC_LIBCFS_ADD_UDSP: {
4407 struct lnet_ioctl_udsp *ioc_udsp = arg;
4408 __u32 bulk_size = ioc_udsp->iou_hdr.ioc_len;
4410 mutex_lock(&the_lnet.ln_api_mutex);
4411 rc = lnet_udsp_demarshal_add(arg, bulk_size);
4413 rc = lnet_udsp_apply_policies(NULL, false);
4414 CDEBUG(D_NET, "policy application returned %d\n", rc);
4417 mutex_unlock(&the_lnet.ln_api_mutex);
4422 case IOC_LIBCFS_DEL_UDSP: {
4423 struct lnet_ioctl_udsp *ioc_udsp = arg;
4424 int idx = ioc_udsp->iou_idx;
4426 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4429 mutex_lock(&the_lnet.ln_api_mutex);
4430 rc = lnet_udsp_del_policy(idx);
4432 rc = lnet_udsp_apply_policies(NULL, false);
4433 CDEBUG(D_NET, "policy re-application returned %d\n",
4437 mutex_unlock(&the_lnet.ln_api_mutex);
4442 case IOC_LIBCFS_GET_UDSP_SIZE: {
4443 struct lnet_ioctl_udsp *ioc_udsp = arg;
4444 struct lnet_udsp *udsp;
4446 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4451 mutex_lock(&the_lnet.ln_api_mutex);
4452 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4456 /* coming in iou_idx will hold the idx of the udsp
4457 * to get the size of. going out the iou_idx will
4458 * hold the size of the UDSP found at the passed
4461 ioc_udsp->iou_idx = lnet_get_udsp_size(udsp);
4462 if (ioc_udsp->iou_idx < 0)
4465 mutex_unlock(&the_lnet.ln_api_mutex);
4470 case IOC_LIBCFS_GET_UDSP: {
4471 struct lnet_ioctl_udsp *ioc_udsp = arg;
4472 struct lnet_udsp *udsp;
4474 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4479 mutex_lock(&the_lnet.ln_api_mutex);
4480 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4484 rc = lnet_udsp_marshal(udsp, ioc_udsp);
4485 mutex_unlock(&the_lnet.ln_api_mutex);
4490 case IOC_LIBCFS_GET_CONST_UDSP_INFO: {
4491 struct lnet_ioctl_construct_udsp_info *info = arg;
4493 if (info->cud_hdr.ioc_len < sizeof(*info))
4496 CDEBUG(D_NET, "GET_UDSP_INFO for %s\n",
4497 libcfs_nid2str(info->cud_nid));
4499 mutex_lock(&the_lnet.ln_api_mutex);
4500 lnet_udsp_get_construct_info(info);
4501 mutex_unlock(&the_lnet.ln_api_mutex);
4507 ni = lnet_net2ni_addref(data->ioc_net);
4511 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4514 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4521 EXPORT_SYMBOL(LNetCtl);
4523 void LNetDebugPeer(struct lnet_processid *id)
4525 lnet_debug_peer(lnet_nid_to_nid4(&id->nid));
4527 EXPORT_SYMBOL(LNetDebugPeer);
4530 * Determine if the specified peer \a nid is on the local node.
4532 * \param nid peer nid to check
4534 * \retval true If peer NID is on the local node.
4535 * \retval false If peer NID is not on the local node.
4537 bool LNetIsPeerLocal(lnet_nid_t nid)
4539 struct lnet_net *net;
4543 cpt = lnet_net_lock_current();
4544 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4545 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4546 if (lnet_nid_to_nid4(&ni->ni_nid) == nid) {
4547 lnet_net_unlock(cpt);
4552 lnet_net_unlock(cpt);
4556 EXPORT_SYMBOL(LNetIsPeerLocal);
4559 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4560 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4562 * \param index Index of the interface to look up.
4563 * \param id On successful return, this location will hold the
4564 * struct lnet_process_id ID of the interface.
4566 * \retval 0 If an interface exists at \a index.
4567 * \retval -ENOENT If no interface has been found.
4570 LNetGetId(unsigned int index, struct lnet_processid *id)
4573 struct lnet_net *net;
4577 LASSERT(the_lnet.ln_refcount > 0);
4579 cpt = lnet_net_lock_current();
4581 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4582 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4583 if (!nid_is_nid4(&ni->ni_nid))
4584 /* FIXME this needs to be handled */
4589 id->nid = ni->ni_nid;
4590 id->pid = the_lnet.ln_pid;
4596 lnet_net_unlock(cpt);
4599 EXPORT_SYMBOL(LNetGetId);
4604 struct lnet_handle_md mdh;
4605 struct completion completion;
4609 lnet_ping_event_handler(struct lnet_event *event)
4611 struct ping_data *pd = event->md_user_ptr;
4613 CDEBUG(D_NET, "ping event (%d %d)%s\n",
4614 event->type, event->status,
4615 event->unlinked ? " unlinked" : "");
4617 if (event->status) {
4619 pd->rc = event->status;
4620 } else if (event->type == LNET_EVENT_REPLY) {
4622 pd->rc = event->mlength;
4624 if (event->unlinked)
4625 complete(&pd->completion);
4628 static int lnet_ping(struct lnet_process_id id, struct lnet_nid *src_nid,
4629 signed long timeout, struct lnet_process_id __user *ids,
4632 struct lnet_md md = { NULL };
4633 struct ping_data pd = { 0 };
4634 struct lnet_ping_buffer *pbuf;
4635 struct lnet_process_id tmpid;
4641 /* n_ids limit is arbitrary */
4642 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4646 * if the user buffer has more space than the lnet_interfaces_max
4647 * then only fill it up to lnet_interfaces_max
4649 if (n_ids > lnet_interfaces_max)
4650 n_ids = lnet_interfaces_max;
4652 if (id.pid == LNET_PID_ANY)
4653 id.pid = LNET_PID_LUSTRE;
4655 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4659 /* initialize md content */
4660 md.start = &pbuf->pb_info;
4661 md.length = LNET_PING_INFO_SIZE(n_ids);
4662 md.threshold = 2; /* GET/REPLY */
4664 md.options = LNET_MD_TRUNCATE;
4666 md.handler = lnet_ping_event_handler;
4668 init_completion(&pd.completion);
4670 rc = LNetMDBind(&md, LNET_UNLINK, &pd.mdh);
4672 CERROR("Can't bind MD: %d\n", rc);
4673 goto fail_ping_buffer_decref;
4676 rc = LNetGet(lnet_nid_to_nid4(src_nid), pd.mdh, id,
4677 LNET_RESERVED_PORTAL,
4678 LNET_PROTO_PING_MATCHBITS, 0, false);
4681 /* Don't CERROR; this could be deliberate! */
4682 rc2 = LNetMDUnlink(pd.mdh);
4685 /* NB must wait for the UNLINK event below... */
4688 if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4689 /* Ensure completion in finite time... */
4690 LNetMDUnlink(pd.mdh);
4691 wait_for_completion(&pd.completion);
4695 goto fail_ping_buffer_decref;
4699 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4701 rc = -EPROTO; /* if I can't parse... */
4704 CERROR("%s: ping info too short %d\n",
4705 libcfs_id2str(id), nob);
4706 goto fail_ping_buffer_decref;
4709 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4710 lnet_swap_pinginfo(pbuf);
4711 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4712 CERROR("%s: Unexpected magic %08x\n",
4713 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4714 goto fail_ping_buffer_decref;
4717 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4718 CERROR("%s: ping w/o NI status: 0x%x\n",
4719 libcfs_id2str(id), pbuf->pb_info.pi_features);
4720 goto fail_ping_buffer_decref;
4723 if (nob < LNET_PING_INFO_SIZE(0)) {
4724 CERROR("%s: Short reply %d(%d min)\n",
4726 nob, (int)LNET_PING_INFO_SIZE(0));
4727 goto fail_ping_buffer_decref;
4730 if (pbuf->pb_info.pi_nnis < n_ids)
4731 n_ids = pbuf->pb_info.pi_nnis;
4733 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4734 CERROR("%s: Short reply %d(%d expected)\n",
4736 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4737 goto fail_ping_buffer_decref;
4740 rc = -EFAULT; /* if I segv in copy_to_user()... */
4742 memset(&tmpid, 0, sizeof(tmpid));
4743 for (i = 0; i < n_ids; i++) {
4744 tmpid.pid = pbuf->pb_info.pi_pid;
4745 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4746 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4747 goto fail_ping_buffer_decref;
4749 rc = pbuf->pb_info.pi_nnis;
4751 fail_ping_buffer_decref:
4752 lnet_ping_buffer_decref(pbuf);
4757 lnet_discover(struct lnet_process_id id, __u32 force,
4758 struct lnet_process_id __user *ids, int n_ids)
4760 struct lnet_peer_ni *lpni;
4761 struct lnet_peer_ni *p;
4762 struct lnet_peer *lp;
4763 struct lnet_process_id *buf;
4769 id.nid == LNET_NID_ANY)
4772 if (id.pid == LNET_PID_ANY)
4773 id.pid = LNET_PID_LUSTRE;
4776 * If the user buffer has more space than the lnet_interfaces_max,
4777 * then only fill it up to lnet_interfaces_max.
4779 if (n_ids > lnet_interfaces_max)
4780 n_ids = lnet_interfaces_max;
4782 CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4786 cpt = lnet_net_lock_current();
4787 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4794 * Clearing the NIDS_UPTODATE flag ensures the peer will
4795 * be discovered, provided discovery has not been disabled.
4797 lp = lpni->lpni_peer_net->lpn_peer;
4798 spin_lock(&lp->lp_lock);
4799 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4800 /* If the force flag is set, force a PING and PUSH as well. */
4802 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4803 spin_unlock(&lp->lp_lock);
4804 rc = lnet_discover_peer_locked(lpni, cpt, true);
4808 /* The lpni (or lp) for this NID may have changed and our ref is
4809 * the only thing keeping the old one around. Release the ref
4810 * and lookup the lpni again
4812 lnet_peer_ni_decref_locked(lpni);
4813 lpni = lnet_find_peer_ni_locked(id.nid);
4818 lp = lpni->lpni_peer_net->lpn_peer;
4822 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4823 buf[i].pid = id.pid;
4824 buf[i].nid = lnet_nid_to_nid4(&p->lpni_nid);
4831 lnet_peer_ni_decref_locked(lpni);
4833 lnet_net_unlock(cpt);
4836 if (copy_to_user(ids, buf, rc * sizeof(*buf)))
4838 CFS_FREE_PTR_ARRAY(buf, n_ids);
4844 * Retrieve peer discovery status.
4846 * \retval 1 if lnet_peer_discovery_disabled is 0
4847 * \retval 0 if lnet_peer_discovery_disabled is 1
4850 LNetGetPeerDiscoveryStatus(void)
4852 return !lnet_peer_discovery_disabled;
4854 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);