4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_LNET
35 #include <linux/ctype.h>
36 #include <linux/log2.h>
37 #include <linux/ktime.h>
38 #include <linux/moduleparam.h>
39 #include <linux/uaccess.h>
40 #ifdef HAVE_SCHED_HEADERS
41 #include <linux/sched/signal.h>
44 #include <lnet/lib-lnet.h>
46 #define D_LNI D_CONSOLE
49 * initialize ln_api_mutex statically, since it needs to be used in
50 * discovery_set callback. That module parameter callback can be called
51 * before module init completes. The mutex needs to be ready for use then.
53 struct lnet the_lnet = {
54 .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
55 }; /* THE state of the network */
56 EXPORT_SYMBOL(the_lnet);
58 static char *ip2nets = "";
59 module_param(ip2nets, charp, 0444);
60 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
62 static char *networks = "";
63 module_param(networks, charp, 0444);
64 MODULE_PARM_DESC(networks, "local networks");
66 static char *routes = "";
67 module_param(routes, charp, 0444);
68 MODULE_PARM_DESC(routes, "routes to non-local networks");
70 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
71 module_param(rnet_htable_size, int, 0444);
72 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
74 static int use_tcp_bonding = false;
75 module_param(use_tcp_bonding, int, 0444);
76 MODULE_PARM_DESC(use_tcp_bonding,
77 "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
79 unsigned int lnet_numa_range = 0;
80 module_param(lnet_numa_range, uint, 0444);
81 MODULE_PARM_DESC(lnet_numa_range,
82 "NUMA range to consider during Multi-Rail selection");
85 * lnet_health_sensitivity determines by how much we decrement the health
86 * value on sending error. The value defaults to 100, which means health
87 * interface health is decremented by 100 points every failure.
89 unsigned int lnet_health_sensitivity = 100;
90 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
91 #ifdef HAVE_KERNEL_PARAM_OPS
92 static struct kernel_param_ops param_ops_health_sensitivity = {
93 .set = sensitivity_set,
96 #define param_check_health_sensitivity(name, p) \
97 __param_check(name, p, int)
98 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
100 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
101 &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
103 MODULE_PARM_DESC(lnet_health_sensitivity,
104 "Value to decrement the health value by on error");
107 * lnet_recovery_interval determines how often we should perform recovery
108 * on unhealthy interfaces.
110 unsigned int lnet_recovery_interval = 1;
111 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
112 #ifdef HAVE_KERNEL_PARAM_OPS
113 static struct kernel_param_ops param_ops_recovery_interval = {
114 .set = recovery_interval_set,
115 .get = param_get_int,
117 #define param_check_recovery_interval(name, p) \
118 __param_check(name, p, int)
119 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
121 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
122 &lnet_recovery_interval, S_IRUGO|S_IWUSR);
124 MODULE_PARM_DESC(lnet_recovery_interval,
125 "Interval to recover unhealthy interfaces in seconds");
127 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
128 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
130 static struct kernel_param_ops param_ops_interfaces_max = {
132 .get = param_get_int,
135 #define param_check_interfaces_max(name, p) \
136 __param_check(name, p, int)
138 #ifdef HAVE_KERNEL_PARAM_OPS
139 module_param(lnet_interfaces_max, interfaces_max, 0644);
141 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
142 ¶m_ops_interfaces_max, 0644);
144 MODULE_PARM_DESC(lnet_interfaces_max,
145 "Maximum number of interfaces in a node.");
147 unsigned lnet_peer_discovery_disabled = 0;
148 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
150 static struct kernel_param_ops param_ops_discovery_disabled = {
151 .set = discovery_set,
152 .get = param_get_int,
155 #define param_check_discovery_disabled(name, p) \
156 __param_check(name, p, int)
157 #ifdef HAVE_KERNEL_PARAM_OPS
158 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
160 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
161 ¶m_ops_discovery_disabled, 0644);
163 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
164 "Set to 1 to disable peer discovery on this node.");
166 unsigned int lnet_drop_asym_route;
167 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
169 static struct kernel_param_ops param_ops_drop_asym_route = {
170 .set = drop_asym_route_set,
171 .get = param_get_int,
174 #define param_check_drop_asym_route(name, p) \
175 __param_check(name, p, int)
176 #ifdef HAVE_KERNEL_PARAM_OPS
177 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
179 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
180 ¶m_ops_drop_asym_route, 0644);
182 MODULE_PARM_DESC(lnet_drop_asym_route,
183 "Set to 1 to drop asymmetrical route messages.");
185 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
186 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 50
188 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
189 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
190 #ifdef HAVE_KERNEL_PARAM_OPS
191 static struct kernel_param_ops param_ops_transaction_timeout = {
192 .set = transaction_to_set,
193 .get = param_get_int,
196 #define param_check_transaction_timeout(name, p) \
197 __param_check(name, p, int)
198 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
200 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
201 &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
203 MODULE_PARM_DESC(lnet_transaction_timeout,
204 "Maximum number of seconds to wait for a peer response.");
206 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 2
207 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
208 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
209 #ifdef HAVE_KERNEL_PARAM_OPS
210 static struct kernel_param_ops param_ops_retry_count = {
211 .set = retry_count_set,
212 .get = param_get_int,
215 #define param_check_retry_count(name, p) \
216 __param_check(name, p, int)
217 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
219 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
220 &lnet_retry_count, S_IRUGO|S_IWUSR);
222 MODULE_PARM_DESC(lnet_retry_count,
223 "Maximum number of times to retry transmitting a message");
226 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
227 unsigned int lnet_current_net_count;
230 * This sequence number keeps track of how many times DLC was used to
231 * update the local NIs. It is incremented when a NI is added or
232 * removed and checked when sending a message to determine if there is
233 * a need to re-run the selection algorithm. See lnet_select_pathway()
234 * for more details on its usage.
236 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
238 static int lnet_ping(struct lnet_process_id id, signed long timeout,
239 struct lnet_process_id __user *ids, int n_ids);
241 static int lnet_discover(struct lnet_process_id id, __u32 force,
242 struct lnet_process_id __user *ids, int n_ids);
245 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
248 unsigned *sensitivity = (unsigned *)kp->arg;
251 rc = kstrtoul(val, 0, &value);
253 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
258 * The purpose of locking the api_mutex here is to ensure that
259 * the correct value ends up stored properly.
261 mutex_lock(&the_lnet.ln_api_mutex);
263 if (value > LNET_MAX_HEALTH_VALUE) {
264 mutex_unlock(&the_lnet.ln_api_mutex);
265 CERROR("Invalid health value. Maximum: %d value = %lu\n",
266 LNET_MAX_HEALTH_VALUE, value);
271 * if we're turning on health then use the health timeout
274 if (*sensitivity == 0 && value != 0) {
275 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
276 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
278 * if we're turning off health then use the no health timeout
281 } else if (*sensitivity != 0 && value == 0) {
282 lnet_transaction_timeout =
283 LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
284 lnet_retry_count = 0;
287 *sensitivity = value;
289 mutex_unlock(&the_lnet.ln_api_mutex);
295 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
298 unsigned *interval = (unsigned *)kp->arg;
301 rc = kstrtoul(val, 0, &value);
303 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
308 CERROR("lnet_recovery_interval must be at least 1 second\n");
313 * The purpose of locking the api_mutex here is to ensure that
314 * the correct value ends up stored properly.
316 mutex_lock(&the_lnet.ln_api_mutex);
320 mutex_unlock(&the_lnet.ln_api_mutex);
326 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
329 unsigned *discovery = (unsigned *)kp->arg;
331 struct lnet_ping_buffer *pbuf;
333 rc = kstrtoul(val, 0, &value);
335 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
339 value = (value) ? 1 : 0;
342 * The purpose of locking the api_mutex here is to ensure that
343 * the correct value ends up stored properly.
345 mutex_lock(&the_lnet.ln_api_mutex);
347 if (value == *discovery) {
348 mutex_unlock(&the_lnet.ln_api_mutex);
353 * We still want to set the discovery value even when LNet is not
354 * running. This is the case when LNet is being loaded and we want
355 * the module parameters to take effect. Otherwise if we're
356 * changing the value dynamically, we want to set it after
359 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
361 mutex_unlock(&the_lnet.ln_api_mutex);
365 /* tell peers that discovery setting has changed */
366 lnet_net_lock(LNET_LOCK_EX);
367 pbuf = the_lnet.ln_ping_target;
369 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
371 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
372 lnet_net_unlock(LNET_LOCK_EX);
375 * Always update the peers. This will result in a push to the
376 * peers with the updated capabilities feature mask. The peer can
377 * then take appropriate action to update its representation of
380 * If discovery is already off, turn it on first before pushing
381 * the update. The discovery flag must be on before pushing.
382 * otherwise if the flag is on and we're turning it off then push
383 * first before turning the flag off. In the former case the flag
384 * is being set twice, but I find it's better to do that rather
385 * than have duplicate code in an if/else statement.
387 if (*discovery > 0 && value == 0)
389 lnet_push_update_to_peers(1);
392 mutex_unlock(&the_lnet.ln_api_mutex);
398 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
401 unsigned int *drop_asym_route = (unsigned int *)kp->arg;
404 rc = kstrtoul(val, 0, &value);
406 CERROR("Invalid module parameter value for "
407 "'lnet_drop_asym_route'\n");
412 * The purpose of locking the api_mutex here is to ensure that
413 * the correct value ends up stored properly.
415 mutex_lock(&the_lnet.ln_api_mutex);
417 if (value == *drop_asym_route) {
418 mutex_unlock(&the_lnet.ln_api_mutex);
422 *drop_asym_route = value;
424 mutex_unlock(&the_lnet.ln_api_mutex);
430 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
433 unsigned *transaction_to = (unsigned *)kp->arg;
436 rc = kstrtoul(val, 0, &value);
438 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
443 * The purpose of locking the api_mutex here is to ensure that
444 * the correct value ends up stored properly.
446 mutex_lock(&the_lnet.ln_api_mutex);
448 if (value < lnet_retry_count || value == 0) {
449 mutex_unlock(&the_lnet.ln_api_mutex);
450 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
451 "Has to be greater than lnet_retry_count (%u)\n",
452 value, lnet_retry_count);
456 if (value == *transaction_to) {
457 mutex_unlock(&the_lnet.ln_api_mutex);
461 *transaction_to = value;
462 if (lnet_retry_count == 0)
463 lnet_lnd_timeout = value;
465 lnet_lnd_timeout = value / lnet_retry_count;
467 mutex_unlock(&the_lnet.ln_api_mutex);
473 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
476 unsigned *retry_count = (unsigned *)kp->arg;
479 rc = kstrtoul(val, 0, &value);
481 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
486 * The purpose of locking the api_mutex here is to ensure that
487 * the correct value ends up stored properly.
489 mutex_lock(&the_lnet.ln_api_mutex);
491 if (lnet_health_sensitivity == 0) {
492 mutex_unlock(&the_lnet.ln_api_mutex);
493 CERROR("Can not set retry_count when health feature is turned off\n");
497 if (value > lnet_transaction_timeout) {
498 mutex_unlock(&the_lnet.ln_api_mutex);
499 CERROR("Invalid value for lnet_retry_count (%lu). "
500 "Has to be smaller than lnet_transaction_timeout (%u)\n",
501 value, lnet_transaction_timeout);
505 *retry_count = value;
508 lnet_lnd_timeout = lnet_transaction_timeout;
510 lnet_lnd_timeout = lnet_transaction_timeout / value;
512 mutex_unlock(&the_lnet.ln_api_mutex);
518 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
522 rc = kstrtoint(val, 0, &value);
524 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
528 if (value < LNET_INTERFACES_MIN) {
529 CWARN("max interfaces provided are too small, setting to %d\n",
530 LNET_INTERFACES_MAX_DEFAULT);
531 value = LNET_INTERFACES_MAX_DEFAULT;
534 *(int *)kp->arg = value;
540 lnet_get_routes(void)
546 lnet_get_networks(void)
551 if (*networks != 0 && *ip2nets != 0) {
552 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
553 "'ip2nets' but not both at once\n");
558 rc = lnet_parse_ip2nets(&nets, ip2nets);
559 return (rc == 0) ? nets : NULL;
569 lnet_init_locks(void)
571 spin_lock_init(&the_lnet.ln_eq_wait_lock);
572 spin_lock_init(&the_lnet.ln_msg_resend_lock);
573 init_completion(&the_lnet.ln_mt_wait_complete);
574 mutex_init(&the_lnet.ln_lnd_mutex);
577 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
578 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
580 struct kmem_cache *lnet_rspt_cachep; /* response tracker cache */
581 struct kmem_cache *lnet_msg_cachep;
584 lnet_slab_setup(void)
586 /* create specific kmem_cache for MEs and small MDs (i.e., originally
587 * allocated in <size-xxx> kmem_cache).
589 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
591 if (!lnet_mes_cachep)
594 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
595 LNET_SMALL_MD_SIZE, 0, 0,
597 if (!lnet_small_mds_cachep)
600 lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
602 if (!lnet_rspt_cachep)
605 lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
607 if (!lnet_msg_cachep)
614 lnet_slab_cleanup(void)
616 if (lnet_msg_cachep) {
617 kmem_cache_destroy(lnet_msg_cachep);
618 lnet_msg_cachep = NULL;
622 if (lnet_rspt_cachep) {
623 kmem_cache_destroy(lnet_rspt_cachep);
624 lnet_rspt_cachep = NULL;
627 if (lnet_small_mds_cachep) {
628 kmem_cache_destroy(lnet_small_mds_cachep);
629 lnet_small_mds_cachep = NULL;
632 if (lnet_mes_cachep) {
633 kmem_cache_destroy(lnet_mes_cachep);
634 lnet_mes_cachep = NULL;
639 lnet_create_remote_nets_table(void)
642 struct list_head *hash;
644 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
645 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
646 CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
648 CERROR("Failed to create remote nets hash table\n");
652 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
653 INIT_LIST_HEAD(&hash[i]);
654 the_lnet.ln_remote_nets_hash = hash;
659 lnet_destroy_remote_nets_table(void)
663 if (the_lnet.ln_remote_nets_hash == NULL)
666 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
667 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
669 CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
670 LNET_REMOTE_NETS_HASH_SIZE);
671 the_lnet.ln_remote_nets_hash = NULL;
675 lnet_destroy_locks(void)
677 if (the_lnet.ln_res_lock != NULL) {
678 cfs_percpt_lock_free(the_lnet.ln_res_lock);
679 the_lnet.ln_res_lock = NULL;
682 if (the_lnet.ln_net_lock != NULL) {
683 cfs_percpt_lock_free(the_lnet.ln_net_lock);
684 the_lnet.ln_net_lock = NULL;
689 lnet_create_locks(void)
693 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
694 if (the_lnet.ln_res_lock == NULL)
697 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
698 if (the_lnet.ln_net_lock == NULL)
704 lnet_destroy_locks();
708 static void lnet_assert_wire_constants(void)
710 /* Wire protocol assertions generated by 'wirecheck'
711 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
712 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
713 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
716 BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
717 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
718 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
719 BUILD_BUG_ON(LNET_MSG_ACK != 0);
720 BUILD_BUG_ON(LNET_MSG_PUT != 1);
721 BUILD_BUG_ON(LNET_MSG_GET != 2);
722 BUILD_BUG_ON(LNET_MSG_REPLY != 3);
723 BUILD_BUG_ON(LNET_MSG_HELLO != 4);
725 /* Checks for struct lnet_handle_wire */
726 BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
727 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
728 wh_interface_cookie) != 0);
729 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
730 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
731 wh_object_cookie) != 8);
732 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
734 /* Checks for struct struct lnet_magicversion */
735 BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
736 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
737 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
738 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
739 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
740 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
741 version_minor) != 6);
742 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
744 /* Checks for struct struct lnet_hdr */
745 BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
746 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
747 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
748 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
749 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
750 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
751 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
752 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
753 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
754 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
755 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
756 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
757 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
758 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
759 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
762 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
763 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
764 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
765 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
766 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
767 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
770 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
771 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
772 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
773 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
774 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
775 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
776 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
777 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
778 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
779 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
782 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
783 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
784 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
785 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
786 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
787 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
788 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
789 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
790 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
791 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
794 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
795 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
798 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
799 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
800 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
801 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
803 /* Checks for struct lnet_ni_status and related constants */
804 BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
805 BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
806 BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
808 /* Checks for struct lnet_ni_status */
809 BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
810 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
811 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
812 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
813 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
814 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
815 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
817 /* Checks for struct lnet_ping_info and related constants */
818 BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
819 BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
820 BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
821 BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
822 BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
823 BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
824 BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
825 BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
827 /* Checks for struct lnet_ping_info */
828 BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
829 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
830 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
831 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
832 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
833 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
834 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
835 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
836 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
837 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
838 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
841 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
843 const struct lnet_lnd *lnd;
845 /* holding lnd mutex */
846 if (type >= NUM_LNDS)
848 lnd = the_lnet.ln_lnds[type];
849 LASSERT(!lnd || lnd->lnd_type == type);
855 lnet_get_lnd_timeout(void)
857 return lnet_lnd_timeout;
859 EXPORT_SYMBOL(lnet_get_lnd_timeout);
862 lnet_register_lnd(const struct lnet_lnd *lnd)
864 mutex_lock(&the_lnet.ln_lnd_mutex);
866 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
867 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
869 the_lnet.ln_lnds[lnd->lnd_type] = lnd;
871 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
873 mutex_unlock(&the_lnet.ln_lnd_mutex);
875 EXPORT_SYMBOL(lnet_register_lnd);
878 lnet_unregister_lnd(const struct lnet_lnd *lnd)
880 mutex_lock(&the_lnet.ln_lnd_mutex);
882 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
884 the_lnet.ln_lnds[lnd->lnd_type] = NULL;
885 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
887 mutex_unlock(&the_lnet.ln_lnd_mutex);
889 EXPORT_SYMBOL(lnet_unregister_lnd);
892 lnet_counters_get_common(struct lnet_counters_common *common)
894 struct lnet_counters *ctr;
897 memset(common, 0, sizeof(*common));
899 lnet_net_lock(LNET_LOCK_EX);
901 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
902 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
903 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
904 common->lcc_errors += ctr->lct_common.lcc_errors;
905 common->lcc_send_count += ctr->lct_common.lcc_send_count;
906 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
907 common->lcc_route_count += ctr->lct_common.lcc_route_count;
908 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
909 common->lcc_send_length += ctr->lct_common.lcc_send_length;
910 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
911 common->lcc_route_length += ctr->lct_common.lcc_route_length;
912 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
914 lnet_net_unlock(LNET_LOCK_EX);
916 EXPORT_SYMBOL(lnet_counters_get_common);
919 lnet_counters_get(struct lnet_counters *counters)
921 struct lnet_counters *ctr;
922 struct lnet_counters_health *health = &counters->lct_health;
925 memset(counters, 0, sizeof(*counters));
927 lnet_counters_get_common(&counters->lct_common);
929 lnet_net_lock(LNET_LOCK_EX);
931 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
932 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
933 health->lch_resend_count += ctr->lct_health.lch_resend_count;
934 health->lch_response_timeout_count +=
935 ctr->lct_health.lch_response_timeout_count;
936 health->lch_local_interrupt_count +=
937 ctr->lct_health.lch_local_interrupt_count;
938 health->lch_local_dropped_count +=
939 ctr->lct_health.lch_local_dropped_count;
940 health->lch_local_aborted_count +=
941 ctr->lct_health.lch_local_aborted_count;
942 health->lch_local_no_route_count +=
943 ctr->lct_health.lch_local_no_route_count;
944 health->lch_local_timeout_count +=
945 ctr->lct_health.lch_local_timeout_count;
946 health->lch_local_error_count +=
947 ctr->lct_health.lch_local_error_count;
948 health->lch_remote_dropped_count +=
949 ctr->lct_health.lch_remote_dropped_count;
950 health->lch_remote_error_count +=
951 ctr->lct_health.lch_remote_error_count;
952 health->lch_remote_timeout_count +=
953 ctr->lct_health.lch_remote_timeout_count;
954 health->lch_network_timeout_count +=
955 ctr->lct_health.lch_network_timeout_count;
957 lnet_net_unlock(LNET_LOCK_EX);
959 EXPORT_SYMBOL(lnet_counters_get);
962 lnet_counters_reset(void)
964 struct lnet_counters *counters;
967 lnet_net_lock(LNET_LOCK_EX);
969 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
970 memset(counters, 0, sizeof(struct lnet_counters));
972 lnet_net_unlock(LNET_LOCK_EX);
976 lnet_res_type2str(int type)
981 case LNET_COOKIE_TYPE_MD:
983 case LNET_COOKIE_TYPE_ME:
985 case LNET_COOKIE_TYPE_EQ:
991 lnet_res_container_cleanup(struct lnet_res_container *rec)
995 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
998 while (!list_empty(&rec->rec_active)) {
999 struct list_head *e = rec->rec_active.next;
1002 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1003 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1005 } else { /* NB: Active MEs should be attached on portals */
1012 /* Found alive MD/ME/EQ, user really should unlink/free
1013 * all of them before finalize LNet, but if someone didn't,
1014 * we have to recycle garbage for him */
1015 CERROR("%d active elements on exit of %s container\n",
1016 count, lnet_res_type2str(rec->rec_type));
1019 if (rec->rec_lh_hash != NULL) {
1020 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1021 rec->rec_lh_hash = NULL;
1024 rec->rec_type = 0; /* mark it as finalized */
1028 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1033 LASSERT(rec->rec_type == 0);
1035 rec->rec_type = type;
1036 INIT_LIST_HEAD(&rec->rec_active);
1038 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1040 /* Arbitrary choice of hash table size */
1041 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1042 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1043 if (rec->rec_lh_hash == NULL) {
1048 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1049 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1054 CERROR("Failed to setup %s resource container\n",
1055 lnet_res_type2str(type));
1056 lnet_res_container_cleanup(rec);
1061 lnet_res_containers_destroy(struct lnet_res_container **recs)
1063 struct lnet_res_container *rec;
1066 cfs_percpt_for_each(rec, i, recs)
1067 lnet_res_container_cleanup(rec);
1069 cfs_percpt_free(recs);
1072 static struct lnet_res_container **
1073 lnet_res_containers_create(int type)
1075 struct lnet_res_container **recs;
1076 struct lnet_res_container *rec;
1080 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1082 CERROR("Failed to allocate %s resource containers\n",
1083 lnet_res_type2str(type));
1087 cfs_percpt_for_each(rec, i, recs) {
1088 rc = lnet_res_container_setup(rec, i, type);
1090 lnet_res_containers_destroy(recs);
1098 struct lnet_libhandle *
1099 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1101 /* ALWAYS called with lnet_res_lock held */
1102 struct list_head *head;
1103 struct lnet_libhandle *lh;
1106 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1109 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1110 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1112 list_for_each_entry(lh, head, lh_hash_chain) {
1113 if (lh->lh_cookie == cookie)
1121 lnet_res_lh_initialize(struct lnet_res_container *rec,
1122 struct lnet_libhandle *lh)
1124 /* ALWAYS called with lnet_res_lock held */
1125 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1128 lh->lh_cookie = rec->rec_lh_cookie;
1129 rec->rec_lh_cookie += 1 << ibits;
1131 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1133 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1137 lnet_create_array_of_queues(void)
1139 struct list_head **qs;
1140 struct list_head *q;
1143 qs = cfs_percpt_alloc(lnet_cpt_table(),
1144 sizeof(struct list_head));
1146 CERROR("Failed to allocate queues\n");
1150 cfs_percpt_for_each(q, i, qs)
1156 static int lnet_unprepare(void);
1159 lnet_prepare(lnet_pid_t requested_pid)
1161 /* Prepare to bring up the network */
1162 struct lnet_res_container **recs;
1165 if (requested_pid == LNET_PID_ANY) {
1166 /* Don't instantiate LNET just for me */
1170 LASSERT(the_lnet.ln_refcount == 0);
1172 the_lnet.ln_routing = 0;
1174 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1175 the_lnet.ln_pid = requested_pid;
1177 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1178 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1179 INIT_LIST_HEAD(&the_lnet.ln_nets);
1180 INIT_LIST_HEAD(&the_lnet.ln_routers);
1181 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1182 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1183 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1184 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1185 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1186 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1187 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1188 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1189 the_lnet.ln_mt_eq = NULL;
1190 init_completion(&the_lnet.ln_started);
1192 rc = lnet_slab_setup();
1196 rc = lnet_create_remote_nets_table();
1201 * NB the interface cookie in wire handles guards against delayed
1202 * replies and ACKs appearing valid after reboot.
1204 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1206 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1207 sizeof(struct lnet_counters));
1208 if (the_lnet.ln_counters == NULL) {
1209 CERROR("Failed to allocate counters for LNet\n");
1214 rc = lnet_peer_tables_create();
1218 rc = lnet_msg_containers_create();
1222 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1223 LNET_COOKIE_TYPE_EQ);
1227 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1233 the_lnet.ln_md_containers = recs;
1235 rc = lnet_portals_create();
1237 CERROR("Failed to create portals for LNet: %d\n", rc);
1241 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1242 if (!the_lnet.ln_mt_zombie_rstqs) {
1255 lnet_unprepare (void)
1259 /* NB no LNET_LOCK since this is the last reference. All LND instances
1260 * have shut down already, so it is safe to unlink and free all
1261 * descriptors, even those that appear committed to a network op (eg MD
1262 * with non-zero pending count) */
1264 lnet_fail_nid(LNET_NID_ANY, 0);
1266 LASSERT(the_lnet.ln_refcount == 0);
1267 LASSERT(list_empty(&the_lnet.ln_test_peers));
1268 LASSERT(list_empty(&the_lnet.ln_nets));
1270 if (the_lnet.ln_mt_zombie_rstqs) {
1271 lnet_clean_zombie_rstqs();
1272 the_lnet.ln_mt_zombie_rstqs = NULL;
1275 if (the_lnet.ln_mt_eq) {
1276 rc = LNetEQFree(the_lnet.ln_mt_eq);
1277 the_lnet.ln_mt_eq = NULL;
1281 lnet_portals_destroy();
1283 if (the_lnet.ln_md_containers != NULL) {
1284 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1285 the_lnet.ln_md_containers = NULL;
1288 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1290 lnet_msg_containers_destroy();
1292 lnet_rtrpools_free(0);
1294 if (the_lnet.ln_counters != NULL) {
1295 cfs_percpt_free(the_lnet.ln_counters);
1296 the_lnet.ln_counters = NULL;
1298 lnet_destroy_remote_nets_table();
1299 lnet_slab_cleanup();
1305 lnet_net2ni_locked(__u32 net_id, int cpt)
1308 struct lnet_net *net;
1310 LASSERT(cpt != LNET_LOCK_EX);
1312 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1313 if (net->net_id == net_id) {
1314 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1324 lnet_net2ni_addref(__u32 net)
1329 ni = lnet_net2ni_locked(net, 0);
1331 lnet_ni_addref_locked(ni, 0);
1336 EXPORT_SYMBOL(lnet_net2ni_addref);
1339 lnet_get_net_locked(__u32 net_id)
1341 struct lnet_net *net;
1343 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1344 if (net->net_id == net_id)
1352 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1357 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1362 val = hash_long(key, LNET_CPT_BITS);
1363 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1367 return (unsigned int)(key + val + (val >> 1)) % number;
1371 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1373 struct lnet_net *net;
1375 /* must called with hold of lnet_net_lock */
1376 if (LNET_CPT_NUMBER == 1)
1377 return 0; /* the only one */
1380 * If NI is provided then use the CPT identified in the NI cpt
1381 * list if one exists. If one doesn't exist, then that NI is
1382 * associated with all CPTs and it follows that the net it belongs
1383 * to is implicitly associated with all CPTs, so just hash the nid
1387 if (ni->ni_cpts != NULL)
1388 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1391 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1394 /* no NI provided so look at the net */
1395 net = lnet_get_net_locked(LNET_NIDNET(nid));
1397 if (net != NULL && net->net_cpts != NULL) {
1398 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1401 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1405 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1410 if (LNET_CPT_NUMBER == 1)
1411 return 0; /* the only one */
1413 cpt = lnet_net_lock_current();
1415 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1417 lnet_net_unlock(cpt);
1421 EXPORT_SYMBOL(lnet_cpt_of_nid);
1424 lnet_islocalnet_locked(__u32 net_id)
1426 struct lnet_net *net;
1429 net = lnet_get_net_locked(net_id);
1431 local = net != NULL;
1437 lnet_islocalnet(__u32 net_id)
1442 cpt = lnet_net_lock_current();
1444 local = lnet_islocalnet_locked(net_id);
1446 lnet_net_unlock(cpt);
1452 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1454 struct lnet_net *net;
1457 LASSERT(cpt != LNET_LOCK_EX);
1459 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1460 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1461 if (ni->ni_nid == nid)
1470 lnet_nid2ni_addref(lnet_nid_t nid)
1475 ni = lnet_nid2ni_locked(nid, 0);
1477 lnet_ni_addref_locked(ni, 0);
1482 EXPORT_SYMBOL(lnet_nid2ni_addref);
1485 lnet_islocalnid(lnet_nid_t nid)
1490 cpt = lnet_net_lock_current();
1491 ni = lnet_nid2ni_locked(nid, cpt);
1492 lnet_net_unlock(cpt);
1498 lnet_count_acceptor_nets(void)
1500 /* Return the # of NIs that need the acceptor. */
1502 struct lnet_net *net;
1505 cpt = lnet_net_lock_current();
1506 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1507 /* all socklnd type networks should have the acceptor
1509 if (net->net_lnd->lnd_accept != NULL)
1513 lnet_net_unlock(cpt);
1518 struct lnet_ping_buffer *
1519 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1521 struct lnet_ping_buffer *pbuf;
1523 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1525 pbuf->pb_nnis = nnis;
1526 pbuf->pb_needs_post = false;
1527 atomic_set(&pbuf->pb_refcnt, 1);
1534 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1536 LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1537 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1540 static struct lnet_ping_buffer *
1541 lnet_ping_target_create(int nnis)
1543 struct lnet_ping_buffer *pbuf;
1545 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1547 CERROR("Can't allocate ping source [%d]\n", nnis);
1551 pbuf->pb_info.pi_nnis = nnis;
1552 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1553 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1554 pbuf->pb_info.pi_features =
1555 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1561 lnet_get_net_ni_count_locked(struct lnet_net *net)
1566 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1573 lnet_get_net_ni_count_pre(struct lnet_net *net)
1578 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1585 lnet_get_ni_count(void)
1588 struct lnet_net *net;
1593 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1594 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1604 lnet_get_net_count(void)
1606 struct lnet_net *net;
1611 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1621 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1623 struct lnet_ni_status *stat;
1627 __swab32s(&pbuf->pb_info.pi_magic);
1628 __swab32s(&pbuf->pb_info.pi_features);
1629 __swab32s(&pbuf->pb_info.pi_pid);
1630 __swab32s(&pbuf->pb_info.pi_nnis);
1631 nnis = pbuf->pb_info.pi_nnis;
1632 if (nnis > pbuf->pb_nnis)
1633 nnis = pbuf->pb_nnis;
1634 for (i = 0; i < nnis; i++) {
1635 stat = &pbuf->pb_info.pi_ni[i];
1636 __swab64s(&stat->ns_nid);
1637 __swab32s(&stat->ns_status);
1642 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1646 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1648 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1650 /* Loopback is guaranteed to be present */
1651 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1653 if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1659 lnet_ping_target_destroy(void)
1661 struct lnet_net *net;
1664 lnet_net_lock(LNET_LOCK_EX);
1666 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1667 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1669 ni->ni_status = NULL;
1674 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1675 the_lnet.ln_ping_target = NULL;
1677 lnet_net_unlock(LNET_LOCK_EX);
1681 lnet_ping_target_event_handler(struct lnet_event *event)
1683 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1685 if (event->unlinked)
1686 lnet_ping_buffer_decref(pbuf);
1690 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1691 struct lnet_handle_md *ping_mdh,
1692 int ni_count, bool set_eq)
1694 struct lnet_process_id id = {
1695 .nid = LNET_NID_ANY,
1699 struct lnet_md md = { NULL };
1703 the_lnet.ln_ping_target_eq =
1704 LNetEQAlloc(lnet_ping_target_event_handler);
1705 if (IS_ERR(the_lnet.ln_ping_target_eq)) {
1706 rc = PTR_ERR(the_lnet.ln_ping_target_eq);
1707 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1712 *ppbuf = lnet_ping_target_create(ni_count);
1713 if (*ppbuf == NULL) {
1718 /* Ping target ME/MD */
1719 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1720 LNET_PROTO_PING_MATCHBITS, 0,
1721 LNET_UNLINK, LNET_INS_AFTER);
1724 CERROR("Can't create ping target ME: %d\n", rc);
1725 goto fail_decref_ping_buffer;
1728 /* initialize md content */
1729 md.start = &(*ppbuf)->pb_info;
1730 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1731 md.threshold = LNET_MD_THRESH_INF;
1733 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1734 LNET_MD_MANAGE_REMOTE;
1735 md.eq_handle = the_lnet.ln_ping_target_eq;
1736 md.user_ptr = *ppbuf;
1738 rc = LNetMDAttach(me, md, LNET_RETAIN, ping_mdh);
1740 CERROR("Can't attach ping target MD: %d\n", rc);
1741 goto fail_unlink_ping_me;
1743 lnet_ping_buffer_addref(*ppbuf);
1747 fail_unlink_ping_me:
1749 fail_decref_ping_buffer:
1750 LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1751 lnet_ping_buffer_decref(*ppbuf);
1755 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1762 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1763 struct lnet_handle_md *ping_mdh)
1765 LNetMDUnlink(*ping_mdh);
1766 LNetInvalidateMDHandle(ping_mdh);
1768 /* NB the MD could be busy; this just starts the unlink */
1769 wait_var_event_warning(&pbuf->pb_refcnt,
1770 atomic_read(&pbuf->pb_refcnt) <= 1,
1771 "Still waiting for ping data MD to unlink\n");
1775 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1778 struct lnet_net *net;
1779 struct lnet_ni_status *ns;
1784 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1785 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1786 LASSERT(i < pbuf->pb_nnis);
1788 ns = &pbuf->pb_info.pi_ni[i];
1790 ns->ns_nid = ni->ni_nid;
1793 ns->ns_status = (ni->ni_status != NULL) ?
1794 ni->ni_status->ns_status :
1803 * We (ab)use the ns_status of the loopback interface to
1804 * transmit the sequence number. The first interface listed
1805 * must be the loopback interface.
1807 rc = lnet_ping_info_validate(&pbuf->pb_info);
1809 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1812 LNET_PING_BUFFER_SEQNO(pbuf) =
1813 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1817 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1818 struct lnet_handle_md ping_mdh)
1820 struct lnet_ping_buffer *old_pbuf = NULL;
1821 struct lnet_handle_md old_ping_md;
1823 /* switch the NIs to point to the new ping info created */
1824 lnet_net_lock(LNET_LOCK_EX);
1826 if (!the_lnet.ln_routing)
1827 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1828 if (!lnet_peer_discovery_disabled)
1829 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1831 /* Ensure only known feature bits have been set. */
1832 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1833 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1835 lnet_ping_target_install_locked(pbuf);
1837 if (the_lnet.ln_ping_target) {
1838 old_pbuf = the_lnet.ln_ping_target;
1839 old_ping_md = the_lnet.ln_ping_target_md;
1841 the_lnet.ln_ping_target_md = ping_mdh;
1842 the_lnet.ln_ping_target = pbuf;
1844 lnet_net_unlock(LNET_LOCK_EX);
1847 /* unlink and free the old ping info */
1848 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1849 lnet_ping_buffer_decref(old_pbuf);
1852 lnet_push_update_to_peers(0);
1856 lnet_ping_target_fini(void)
1860 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1861 &the_lnet.ln_ping_target_md);
1863 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1866 lnet_ping_target_destroy();
1869 /* Resize the push target. */
1870 int lnet_push_target_resize(void)
1872 struct lnet_handle_md mdh;
1873 struct lnet_handle_md old_mdh;
1874 struct lnet_ping_buffer *pbuf;
1875 struct lnet_ping_buffer *old_pbuf;
1880 nnis = the_lnet.ln_push_target_nnis;
1882 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
1886 /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
1887 * dropped when we need to resize again (see "old_pbuf" below) or when
1888 * LNet is shutdown (see lnet_push_target_fini())
1890 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1892 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
1896 rc = lnet_push_target_post(pbuf, &mdh);
1898 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
1899 lnet_ping_buffer_decref(pbuf);
1903 lnet_net_lock(LNET_LOCK_EX);
1904 old_pbuf = the_lnet.ln_push_target;
1905 old_mdh = the_lnet.ln_push_target_md;
1906 the_lnet.ln_push_target = pbuf;
1907 the_lnet.ln_push_target_md = mdh;
1908 lnet_net_unlock(LNET_LOCK_EX);
1911 LNetMDUnlink(old_mdh);
1912 /* Drop ref set by lnet_ping_buffer_alloc() */
1913 lnet_ping_buffer_decref(old_pbuf);
1916 /* Received another push or reply that requires a larger buffer */
1917 if (nnis < the_lnet.ln_push_target_nnis)
1920 CDEBUG(D_NET, "nnis %d success\n", nnis);
1924 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
1925 struct lnet_handle_md *mdhp)
1927 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1928 struct lnet_md md = { NULL };
1932 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1933 LNET_PROTO_PING_MATCHBITS, 0,
1934 LNET_UNLINK, LNET_INS_AFTER);
1937 CERROR("Can't create push target ME: %d\n", rc);
1941 pbuf->pb_needs_post = false;
1943 /* This reference is dropped by lnet_push_target_event_handler() */
1944 lnet_ping_buffer_addref(pbuf);
1946 /* initialize md content */
1947 md.start = &pbuf->pb_info;
1948 md.length = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
1951 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
1953 md.eq_handle = the_lnet.ln_push_target_eq;
1955 rc = LNetMDAttach(me, md, LNET_UNLINK, mdhp);
1957 CERROR("Can't attach push MD: %d\n", rc);
1959 lnet_ping_buffer_decref(pbuf);
1960 pbuf->pb_needs_post = true;
1964 CDEBUG(D_NET, "posted push target %p\n", pbuf);
1969 static void lnet_push_target_event_handler(struct lnet_event *ev)
1971 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1973 CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
1976 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1977 lnet_swap_pinginfo(pbuf);
1979 if (ev->type == LNET_EVENT_UNLINK) {
1980 /* Drop ref added by lnet_push_target_post() */
1981 lnet_ping_buffer_decref(pbuf);
1985 lnet_peer_push_event(ev);
1987 /* Drop ref added by lnet_push_target_post */
1988 lnet_ping_buffer_decref(pbuf);
1991 /* Initialize the push target. */
1992 static int lnet_push_target_init(void)
1996 if (the_lnet.ln_push_target)
1999 the_lnet.ln_push_target_eq =
2000 LNetEQAlloc(lnet_push_target_event_handler);
2001 if (IS_ERR(the_lnet.ln_push_target_eq)) {
2002 rc = PTR_ERR(the_lnet.ln_push_target_eq);
2003 CERROR("Can't allocated push target EQ: %d\n", rc);
2007 rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2010 /* Start at the required minimum, we'll enlarge if required. */
2011 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2013 rc = lnet_push_target_resize();
2016 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2017 LNetEQFree(the_lnet.ln_push_target_eq);
2018 the_lnet.ln_push_target_eq = NULL;
2024 /* Clean up the push target. */
2025 static void lnet_push_target_fini(void)
2027 if (!the_lnet.ln_push_target)
2030 /* Unlink and invalidate to prevent new references. */
2031 LNetMDUnlink(the_lnet.ln_push_target_md);
2032 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2034 /* Wait for the unlink to complete. */
2035 wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2036 atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2037 "Still waiting for ping data MD to unlink\n");
2039 /* Drop ref set by lnet_ping_buffer_alloc() */
2040 lnet_ping_buffer_decref(the_lnet.ln_push_target);
2041 the_lnet.ln_push_target = NULL;
2042 the_lnet.ln_push_target_nnis = 0;
2044 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2045 LNetEQFree(the_lnet.ln_push_target_eq);
2046 the_lnet.ln_push_target_eq = NULL;
2050 lnet_ni_tq_credits(struct lnet_ni *ni)
2054 LASSERT(ni->ni_ncpts >= 1);
2056 if (ni->ni_ncpts == 1)
2057 return ni->ni_net->net_tunables.lct_max_tx_credits;
2059 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2060 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2061 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2067 lnet_ni_unlink_locked(struct lnet_ni *ni)
2069 /* move it to zombie list and nobody can find it anymore */
2070 LASSERT(!list_empty(&ni->ni_netlist));
2071 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2072 lnet_ni_decref_locked(ni, 0);
2076 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2081 struct list_head *zombie_list = &net->net_ni_zombie;
2084 * Now wait for the NIs I just nuked to show up on the zombie
2085 * list and shut them down in guaranteed thread context
2088 while (!list_empty(zombie_list)) {
2092 ni = list_entry(zombie_list->next,
2093 struct lnet_ni, ni_netlist);
2094 list_del_init(&ni->ni_netlist);
2095 /* the ni should be in deleting state. If it's not it's
2097 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2098 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2101 /* still busy, add it back to zombie list */
2102 list_add(&ni->ni_netlist, zombie_list);
2106 if (!list_empty(&ni->ni_netlist)) {
2107 lnet_net_unlock(LNET_LOCK_EX);
2109 if ((i & (-i)) == i) {
2111 "Waiting for zombie LNI %s\n",
2112 libcfs_nid2str(ni->ni_nid));
2114 schedule_timeout_uninterruptible(cfs_time_seconds(1));
2115 lnet_net_lock(LNET_LOCK_EX);
2119 lnet_net_unlock(LNET_LOCK_EX);
2121 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2123 LASSERT(!in_interrupt());
2124 /* Holding the mutex makes it safe for lnd_shutdown
2125 * to call module_put(). Module unload cannot finish
2126 * until lnet_unregister_lnd() completes, and that
2127 * requires the mutex.
2129 mutex_lock(&the_lnet.ln_lnd_mutex);
2130 (net->net_lnd->lnd_shutdown)(ni);
2131 mutex_unlock(&the_lnet.ln_lnd_mutex);
2134 CDEBUG(D_LNI, "Removed LNI %s\n",
2135 libcfs_nid2str(ni->ni_nid));
2139 lnet_net_lock(LNET_LOCK_EX);
2143 /* shutdown down the NI and release refcount */
2145 lnet_shutdown_lndni(struct lnet_ni *ni)
2148 struct lnet_net *net = ni->ni_net;
2150 lnet_net_lock(LNET_LOCK_EX);
2152 ni->ni_state = LNET_NI_STATE_DELETING;
2154 lnet_ni_unlink_locked(ni);
2155 lnet_incr_dlc_seq();
2156 lnet_net_unlock(LNET_LOCK_EX);
2158 /* clear messages for this NI on the lazy portal */
2159 for (i = 0; i < the_lnet.ln_nportals; i++)
2160 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2162 lnet_net_lock(LNET_LOCK_EX);
2163 lnet_clear_zombies_nis_locked(net);
2164 lnet_net_unlock(LNET_LOCK_EX);
2168 lnet_shutdown_lndnet(struct lnet_net *net)
2172 lnet_net_lock(LNET_LOCK_EX);
2174 list_del_init(&net->net_list);
2176 while (!list_empty(&net->net_ni_list)) {
2177 ni = list_entry(net->net_ni_list.next,
2178 struct lnet_ni, ni_netlist);
2179 lnet_net_unlock(LNET_LOCK_EX);
2180 lnet_shutdown_lndni(ni);
2181 lnet_net_lock(LNET_LOCK_EX);
2184 lnet_net_unlock(LNET_LOCK_EX);
2186 /* Do peer table cleanup for this net */
2187 lnet_peer_tables_cleanup(net);
2193 lnet_shutdown_lndnets(void)
2195 struct lnet_net *net;
2197 struct lnet_msg *msg, *tmp;
2199 /* NB called holding the global mutex */
2201 /* All quiet on the API front */
2202 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2203 LASSERT(the_lnet.ln_refcount == 0);
2205 lnet_net_lock(LNET_LOCK_EX);
2206 the_lnet.ln_state = LNET_STATE_STOPPING;
2209 * move the nets to the zombie list to avoid them being
2210 * picked up for new work. LONET is also included in the
2211 * Nets that will be moved to the zombie list
2213 list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2215 /* Drop the cached loopback Net. */
2216 if (the_lnet.ln_loni != NULL) {
2217 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2218 the_lnet.ln_loni = NULL;
2220 lnet_net_unlock(LNET_LOCK_EX);
2222 /* iterate through the net zombie list and delete each net */
2223 while (!list_empty(&the_lnet.ln_net_zombie)) {
2224 net = list_entry(the_lnet.ln_net_zombie.next,
2225 struct lnet_net, net_list);
2226 lnet_shutdown_lndnet(net);
2229 spin_lock(&the_lnet.ln_msg_resend_lock);
2230 list_splice(&the_lnet.ln_msg_resend, &resend);
2231 spin_unlock(&the_lnet.ln_msg_resend_lock);
2233 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2234 list_del_init(&msg->msg_list);
2235 msg->msg_no_resend = true;
2236 lnet_finalize(msg, -ECANCELED);
2239 lnet_net_lock(LNET_LOCK_EX);
2240 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2241 lnet_net_unlock(LNET_LOCK_EX);
2245 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2248 struct lnet_tx_queue *tq;
2250 struct lnet_net *net = ni->ni_net;
2252 mutex_lock(&the_lnet.ln_lnd_mutex);
2255 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2256 ni->ni_lnd_tunables_set = true;
2259 rc = (net->net_lnd->lnd_startup)(ni);
2261 mutex_unlock(&the_lnet.ln_lnd_mutex);
2264 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2265 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2270 ni->ni_state = LNET_NI_STATE_ACTIVE;
2273 /* We keep a reference on the loopback net through the loopback NI */
2274 if (net->net_lnd->lnd_type == LOLND) {
2276 LASSERT(the_lnet.ln_loni == NULL);
2277 the_lnet.ln_loni = ni;
2278 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2279 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2280 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2281 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2285 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2286 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2287 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2288 libcfs_lnd2str(net->net_lnd->lnd_type),
2289 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2291 /* shutdown the NI since if we get here then it must've already
2294 lnet_shutdown_lndni(ni);
2298 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2299 tq->tq_credits_min =
2300 tq->tq_credits_max =
2301 tq->tq_credits = lnet_ni_tq_credits(ni);
2304 atomic_set(&ni->ni_tx_credits,
2305 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2306 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2308 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2309 libcfs_nid2str(ni->ni_nid),
2310 ni->ni_net->net_tunables.lct_peer_tx_credits,
2311 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2312 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2313 ni->ni_net->net_tunables.lct_peer_timeout);
2322 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2325 struct lnet_net *net_l = NULL;
2326 LIST_HEAD(local_ni_list);
2330 const struct lnet_lnd *lnd;
2332 net->net_tunables.lct_peer_timeout;
2334 net->net_tunables.lct_max_tx_credits;
2335 int peerrtrcredits =
2336 net->net_tunables.lct_peer_rtr_credits;
2339 * make sure that this net is unique. If it isn't then
2340 * we are adding interfaces to an already existing network, and
2341 * 'net' is just a convenient way to pass in the list.
2342 * if it is unique we need to find the LND and load it if
2345 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2346 lnd_type = LNET_NETTYP(net->net_id);
2348 mutex_lock(&the_lnet.ln_lnd_mutex);
2349 lnd = lnet_find_lnd_by_type(lnd_type);
2352 mutex_unlock(&the_lnet.ln_lnd_mutex);
2353 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2354 mutex_lock(&the_lnet.ln_lnd_mutex);
2356 lnd = lnet_find_lnd_by_type(lnd_type);
2358 mutex_unlock(&the_lnet.ln_lnd_mutex);
2359 CERROR("Can't load LND %s, module %s, rc=%d\n",
2360 libcfs_lnd2str(lnd_type),
2361 libcfs_lnd2modname(lnd_type), rc);
2362 #ifndef HAVE_MODULE_LOADING_SUPPORT
2363 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2364 "compiled with kernel module "
2365 "loading support.");
2374 mutex_unlock(&the_lnet.ln_lnd_mutex);
2380 * net_l: if the network being added is unique then net_l
2381 * will point to that network
2382 * if the network being added is not unique then
2383 * net_l points to the existing network.
2385 * When we enter the loop below, we'll pick NIs off he
2386 * network beign added and start them up, then add them to
2387 * a local ni list. Once we've successfully started all
2388 * the NIs then we join the local NI list (of started up
2389 * networks) with the net_l->net_ni_list, which should
2390 * point to the correct network to add the new ni list to
2392 * If any of the new NIs fail to start up, then we want to
2393 * iterate through the local ni list, which should include
2394 * any NIs which were successfully started up, and shut
2397 * After than we want to delete the network being added,
2398 * to avoid a memory leak.
2402 * When a network uses TCP bonding then all its interfaces
2403 * must be specified when the network is first defined: the
2404 * TCP bonding code doesn't allow for interfaces to be added
2407 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2408 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2413 while (!list_empty(&net->net_ni_added)) {
2414 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2416 list_del_init(&ni->ni_netlist);
2418 /* make sure that the the NI we're about to start
2419 * up is actually unique. if it's not fail. */
2420 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2421 ni->ni_interfaces[0])) {
2426 /* adjust the pointer the parent network, just in case it
2427 * the net is a duplicate */
2430 rc = lnet_startup_lndni(ni, tun);
2436 list_add_tail(&ni->ni_netlist, &local_ni_list);
2441 lnet_net_lock(LNET_LOCK_EX);
2442 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2443 lnet_incr_dlc_seq();
2444 lnet_net_unlock(LNET_LOCK_EX);
2446 /* if the network is not unique then we don't want to keep
2447 * it around after we're done. Free it. Otherwise add that
2448 * net to the global the_lnet.ln_nets */
2449 if (net_l != net && net_l != NULL) {
2451 * TODO - note. currently the tunables can not be updated
2457 * restore tunables after it has been overwitten by the
2460 if (peer_timeout != -1)
2461 net->net_tunables.lct_peer_timeout = peer_timeout;
2462 if (maxtxcredits != -1)
2463 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2464 if (peerrtrcredits != -1)
2465 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2467 lnet_net_lock(LNET_LOCK_EX);
2468 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2469 lnet_net_unlock(LNET_LOCK_EX);
2472 /* update net count */
2473 lnet_current_net_count = lnet_get_net_count();
2479 * shutdown the new NIs that are being started up
2480 * free the NET being started
2482 while (!list_empty(&local_ni_list)) {
2483 ni = list_entry(local_ni_list.next, struct lnet_ni,
2486 lnet_shutdown_lndni(ni);
2496 lnet_startup_lndnets(struct list_head *netlist)
2498 struct lnet_net *net;
2503 * Change to running state before bringing up the LNDs. This
2504 * allows lnet_shutdown_lndnets() to assert that we've passed
2507 lnet_net_lock(LNET_LOCK_EX);
2508 the_lnet.ln_state = LNET_STATE_RUNNING;
2509 lnet_net_unlock(LNET_LOCK_EX);
2511 while (!list_empty(netlist)) {
2512 net = list_entry(netlist->next, struct lnet_net, net_list);
2513 list_del_init(&net->net_list);
2515 rc = lnet_startup_lndnet(net, NULL);
2525 lnet_shutdown_lndnets();
2531 * Initialize LNet library.
2533 * Automatically called at module loading time. Caller has to call
2534 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2535 * latter returned 0. It must be called exactly once.
2537 * \retval 0 on success
2538 * \retval -ve on failures.
2540 int lnet_lib_init(void)
2544 lnet_assert_wire_constants();
2546 /* refer to global cfs_cpt_table for now */
2547 the_lnet.ln_cpt_table = cfs_cpt_tab;
2548 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2550 LASSERT(the_lnet.ln_cpt_number > 0);
2551 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2552 /* we are under risk of consuming all lh_cookie */
2553 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2554 "please change setting of CPT-table and retry\n",
2555 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2559 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2560 the_lnet.ln_cpt_bits++;
2562 rc = lnet_create_locks();
2564 CERROR("Can't create LNet global locks: %d\n", rc);
2568 the_lnet.ln_refcount = 0;
2569 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2570 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2572 /* The hash table size is the number of bits it takes to express the set
2573 * ln_num_routes, minus 1 (better to under estimate than over so we
2574 * don't waste memory). */
2575 if (rnet_htable_size <= 0)
2576 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2577 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2578 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2579 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2580 order_base_2(rnet_htable_size) - 1);
2582 /* All LNDs apart from the LOLND are in separate modules. They
2583 * register themselves when their module loads, and unregister
2584 * themselves when their module is unloaded. */
2585 lnet_register_lnd(&the_lolnd);
2590 * Finalize LNet library.
2592 * \pre lnet_lib_init() called with success.
2593 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2595 * As this happens at module-unload, all lnds must already be unloaded,
2596 * so they must already be unregistered.
2598 void lnet_lib_exit(void)
2602 LASSERT(the_lnet.ln_refcount == 0);
2603 lnet_unregister_lnd(&the_lolnd);
2604 for (i = 0; i < NUM_LNDS; i++)
2605 LASSERT(!the_lnet.ln_lnds[i]);
2606 lnet_destroy_locks();
2610 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2612 * Users must call this function at least once before any other functions.
2613 * For each successful call there must be a corresponding call to
2614 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2617 * The PID used by LNet may be different from the one requested.
2620 * \param requested_pid PID requested by the caller.
2622 * \return >= 0 on success, and < 0 error code on failures.
2625 LNetNIInit(lnet_pid_t requested_pid)
2627 int im_a_router = 0;
2630 struct lnet_ping_buffer *pbuf;
2631 struct lnet_handle_md ping_mdh;
2632 LIST_HEAD(net_head);
2633 struct lnet_net *net;
2635 mutex_lock(&the_lnet.ln_api_mutex);
2637 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2639 if (the_lnet.ln_refcount > 0) {
2640 rc = the_lnet.ln_refcount++;
2641 mutex_unlock(&the_lnet.ln_api_mutex);
2645 rc = lnet_prepare(requested_pid);
2647 mutex_unlock(&the_lnet.ln_api_mutex);
2651 /* create a network for Loopback network */
2652 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2655 goto err_empty_list;
2658 /* Add in the loopback NI */
2659 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2661 goto err_empty_list;
2664 /* If LNet is being initialized via DLC it is possible
2665 * that the user requests not to load module parameters (ones which
2666 * are supported by DLC) on initialization. Therefore, make sure not
2667 * to load networks, routes and forwarding from module parameters
2668 * in this case. On cleanup in case of failure only clean up
2669 * routes if it has been loaded */
2670 if (!the_lnet.ln_nis_from_mod_params) {
2671 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2674 goto err_empty_list;
2677 ni_count = lnet_startup_lndnets(&net_head);
2680 goto err_empty_list;
2683 if (!the_lnet.ln_nis_from_mod_params) {
2684 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2686 goto err_shutdown_lndnis;
2688 rc = lnet_rtrpools_alloc(im_a_router);
2690 goto err_destroy_routes;
2693 rc = lnet_acceptor_start();
2695 goto err_destroy_routes;
2697 the_lnet.ln_refcount = 1;
2698 /* Now I may use my own API functions... */
2700 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2702 goto err_acceptor_stop;
2704 lnet_ping_target_update(pbuf, ping_mdh);
2706 the_lnet.ln_mt_eq = LNetEQAlloc(lnet_mt_event_handler);
2707 if (IS_ERR(the_lnet.ln_mt_eq)) {
2708 rc = PTR_ERR(the_lnet.ln_mt_eq);
2709 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2713 rc = lnet_push_target_init();
2717 rc = lnet_peer_discovery_start();
2719 goto err_destroy_push_target;
2721 rc = lnet_monitor_thr_start();
2723 goto err_stop_discovery_thr;
2726 lnet_router_debugfs_init();
2728 mutex_unlock(&the_lnet.ln_api_mutex);
2730 complete_all(&the_lnet.ln_started);
2732 /* wait for all routers to start */
2733 lnet_wait_router_start();
2737 err_stop_discovery_thr:
2738 lnet_peer_discovery_stop();
2739 err_destroy_push_target:
2740 lnet_push_target_fini();
2742 lnet_ping_target_fini();
2744 the_lnet.ln_refcount = 0;
2745 lnet_acceptor_stop();
2747 if (!the_lnet.ln_nis_from_mod_params)
2748 lnet_destroy_routes();
2749 err_shutdown_lndnis:
2750 lnet_shutdown_lndnets();
2754 mutex_unlock(&the_lnet.ln_api_mutex);
2755 while (!list_empty(&net_head)) {
2756 struct lnet_net *net;
2758 net = list_entry(net_head.next, struct lnet_net, net_list);
2759 list_del_init(&net->net_list);
2764 EXPORT_SYMBOL(LNetNIInit);
2767 * Stop LNet interfaces, routing, and forwarding.
2769 * Users must call this function once for each successful call to LNetNIInit().
2770 * Once the LNetNIFini() operation has been started, the results of pending
2771 * API operations are undefined.
2773 * \return always 0 for current implementation.
2778 mutex_lock(&the_lnet.ln_api_mutex);
2780 LASSERT(the_lnet.ln_refcount > 0);
2782 if (the_lnet.ln_refcount != 1) {
2783 the_lnet.ln_refcount--;
2785 LASSERT(!the_lnet.ln_niinit_self);
2789 lnet_router_debugfs_fini();
2790 lnet_monitor_thr_stop();
2791 lnet_peer_discovery_stop();
2792 lnet_push_target_fini();
2793 lnet_ping_target_fini();
2795 /* Teardown fns that use my own API functions BEFORE here */
2796 the_lnet.ln_refcount = 0;
2798 lnet_acceptor_stop();
2799 lnet_destroy_routes();
2800 lnet_shutdown_lndnets();
2804 mutex_unlock(&the_lnet.ln_api_mutex);
2807 EXPORT_SYMBOL(LNetNIFini);
2810 * Grabs the ni data from the ni structure and fills the out
2813 * \param[in] ni network interface structure
2814 * \param[out] cfg_ni NI config information
2815 * \param[out] tun network and LND tunables
2818 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2819 struct lnet_ioctl_config_lnd_tunables *tun,
2820 struct lnet_ioctl_element_stats *stats,
2823 size_t min_size = 0;
2826 if (!ni || !cfg_ni || !tun)
2829 if (ni->ni_interfaces[0] != NULL) {
2830 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2831 if (ni->ni_interfaces[i] != NULL) {
2832 strncpy(cfg_ni->lic_ni_intf[i],
2833 ni->ni_interfaces[i],
2834 sizeof(cfg_ni->lic_ni_intf[i]));
2839 cfg_ni->lic_nid = ni->ni_nid;
2840 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2841 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2843 cfg_ni->lic_status = ni->ni_status->ns_status;
2844 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2845 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2847 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2850 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2851 LNET_STATS_TYPE_SEND);
2852 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2853 LNET_STATS_TYPE_RECV);
2854 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2855 LNET_STATS_TYPE_DROP);
2859 * tun->lt_tun will always be present, but in order to be
2860 * backwards compatible, we need to deal with the cases when
2861 * tun->lt_tun is smaller than what the kernel has, because it
2862 * comes from an older version of a userspace program, then we'll
2863 * need to copy as much information as we have available space.
2865 min_size = tun_size - sizeof(tun->lt_cmn);
2866 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2868 /* copy over the cpts */
2869 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2870 ni->ni_cpts == NULL) {
2871 for (i = 0; i < ni->ni_ncpts; i++)
2872 cfg_ni->lic_cpts[i] = i;
2875 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2876 i < LNET_MAX_SHOW_NUM_CPT;
2878 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2880 cfg_ni->lic_ncpts = ni->ni_ncpts;
2884 * NOTE: This is a legacy function left in the code to be backwards
2885 * compatible with older userspace programs. It should eventually be
2888 * Grabs the ni data from the ni structure and fills the out
2891 * \param[in] ni network interface structure
2892 * \param[out] config config information
2895 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2896 struct lnet_ioctl_config_data *config)
2898 struct lnet_ioctl_net_config *net_config;
2899 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2900 size_t min_size, tunable_size = 0;
2906 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2910 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2911 ARRAY_SIZE(net_config->ni_interfaces));
2913 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2914 if (!ni->ni_interfaces[i])
2917 strncpy(net_config->ni_interfaces[i],
2918 ni->ni_interfaces[i],
2919 sizeof(net_config->ni_interfaces[i]));
2922 config->cfg_nid = ni->ni_nid;
2923 config->cfg_config_u.cfg_net.net_peer_timeout =
2924 ni->ni_net->net_tunables.lct_peer_timeout;
2925 config->cfg_config_u.cfg_net.net_max_tx_credits =
2926 ni->ni_net->net_tunables.lct_max_tx_credits;
2927 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2928 ni->ni_net->net_tunables.lct_peer_tx_credits;
2929 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2930 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2932 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2933 net_config->ni_status = LNET_NI_STATUS_UP;
2935 net_config->ni_status = ni->ni_status->ns_status;
2938 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2940 for (i = 0; i < num_cpts; i++)
2941 net_config->ni_cpts[i] = ni->ni_cpts[i];
2943 config->cfg_ncpts = num_cpts;
2947 * See if user land tools sent in a newer and larger version
2948 * of struct lnet_tunables than what the kernel uses.
2950 min_size = sizeof(*config) + sizeof(*net_config);
2952 if (config->cfg_hdr.ioc_len > min_size)
2953 tunable_size = config->cfg_hdr.ioc_len - min_size;
2955 /* Don't copy too much data to user space */
2956 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2957 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2959 if (lnd_cfg && min_size) {
2960 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2961 config->cfg_config_u.cfg_net.net_interface_count = 1;
2963 /* Tell user land that kernel side has less data */
2964 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2965 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2966 config->cfg_hdr.ioc_len -= min_size;
2972 lnet_get_ni_idx_locked(int idx)
2975 struct lnet_net *net;
2977 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2978 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2988 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2991 struct lnet_net *net = mynet;
2994 * It is possible that the net has been cleaned out while there is
2995 * a message being sent. This function accessed the net without
2996 * checking if the list is empty
3000 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3002 if (list_empty(&net->net_ni_list))
3004 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3010 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3011 /* if you reached the end of the ni list and the net is
3012 * specified, then there are no more nis in that net */
3016 /* we reached the end of this net ni list. move to the
3018 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3019 /* no more nets and no more NIs. */
3022 /* get the next net */
3023 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3025 if (list_empty(&net->net_ni_list))
3027 /* get the ni on it */
3028 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3034 if (list_empty(&prev->ni_netlist))
3037 /* there are more nis left */
3038 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3044 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3049 int idx = config->cfg_count;
3051 cpt = lnet_net_lock_current();
3053 ni = lnet_get_ni_idx_locked(idx);
3058 lnet_fill_ni_info_legacy(ni, config);
3062 lnet_net_unlock(cpt);
3067 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3068 struct lnet_ioctl_config_lnd_tunables *tun,
3069 struct lnet_ioctl_element_stats *stats,
3076 if (!cfg_ni || !tun || !stats)
3079 cpt = lnet_net_lock_current();
3081 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3086 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3090 lnet_net_unlock(cpt);
3094 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3103 cpt = lnet_net_lock_current();
3105 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3108 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3112 lnet_net_unlock(cpt);
3117 static int lnet_add_net_common(struct lnet_net *net,
3118 struct lnet_ioctl_config_lnd_tunables *tun)
3121 struct lnet_ping_buffer *pbuf;
3122 struct lnet_handle_md ping_mdh;
3124 struct lnet_remotenet *rnet;
3127 lnet_net_lock(LNET_LOCK_EX);
3128 rnet = lnet_find_rnet_locked(net->net_id);
3129 lnet_net_unlock(LNET_LOCK_EX);
3131 * make sure that the net added doesn't invalidate the current
3132 * configuration LNet is keeping
3135 CERROR("Adding net %s will invalidate routing configuration\n",
3136 libcfs_net2str(net->net_id));
3142 * make sure you calculate the correct number of slots in the ping
3143 * buffer. Since the ping info is a flattened list of all the NIs,
3144 * we should allocate enough slots to accomodate the number of NIs
3145 * which will be added.
3147 * since ni hasn't been configured yet, use
3148 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3150 net_ni_count = lnet_get_net_ni_count_pre(net);
3152 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3153 net_ni_count + lnet_get_ni_count(),
3161 memcpy(&net->net_tunables,
3162 &tun->lt_cmn, sizeof(net->net_tunables));
3164 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3166 net_id = net->net_id;
3168 rc = lnet_startup_lndnet(net,
3169 (tun) ? &tun->lt_tun : NULL);
3173 lnet_net_lock(LNET_LOCK_EX);
3174 net = lnet_get_net_locked(net_id);
3175 lnet_net_unlock(LNET_LOCK_EX);
3180 * Start the acceptor thread if this is the first network
3181 * being added that requires the thread.
3183 if (net->net_lnd->lnd_accept) {
3184 rc = lnet_acceptor_start();
3186 /* shutdown the net that we just started */
3187 CERROR("Failed to start up acceptor thread\n");
3188 lnet_shutdown_lndnet(net);
3193 lnet_net_lock(LNET_LOCK_EX);
3194 lnet_peer_net_added(net);
3195 lnet_net_unlock(LNET_LOCK_EX);
3197 lnet_ping_target_update(pbuf, ping_mdh);
3202 lnet_ping_md_unlink(pbuf, &ping_mdh);
3203 lnet_ping_buffer_decref(pbuf);
3207 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3208 struct lnet_ioctl_config_lnd_tunables *tun)
3210 struct lnet_net *net;
3213 LIST_HEAD(net_head);
3215 rc = lnet_parse_ip2nets(&nets, ip2nets);
3219 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3223 mutex_lock(&the_lnet.ln_api_mutex);
3224 while (!list_empty(&net_head)) {
3225 net = list_entry(net_head.next, struct lnet_net, net_list);
3226 list_del_init(&net->net_list);
3227 rc = lnet_add_net_common(net, tun);
3233 mutex_unlock(&the_lnet.ln_api_mutex);
3235 while (!list_empty(&net_head)) {
3236 net = list_entry(net_head.next, struct lnet_net, net_list);
3237 list_del_init(&net->net_list);
3243 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3245 struct lnet_net *net;
3247 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3249 __u32 net_id, lnd_type;
3251 /* get the tunables if they are available */
3252 if (conf->lic_cfg_hdr.ioc_len >=
3253 sizeof(*conf) + sizeof(*tun))
3254 tun = (struct lnet_ioctl_config_lnd_tunables *)
3257 /* handle legacy ip2nets from DLC */
3258 if (conf->lic_legacy_ip2nets[0] != '\0')
3259 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3262 net_id = LNET_NIDNET(conf->lic_nid);
3263 lnd_type = LNET_NETTYP(net_id);
3265 if (!libcfs_isknown_lnd(lnd_type)) {
3266 CERROR("No valid net and lnd information provided\n");
3270 net = lnet_net_alloc(net_id, NULL);
3274 for (i = 0; i < conf->lic_ncpts; i++) {
3275 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3279 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3280 conf->lic_ni_intf[0]);
3284 mutex_lock(&the_lnet.ln_api_mutex);
3286 rc = lnet_add_net_common(net, tun);
3288 mutex_unlock(&the_lnet.ln_api_mutex);
3293 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3295 struct lnet_net *net;
3297 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3298 struct lnet_ping_buffer *pbuf;
3299 struct lnet_handle_md ping_mdh;
3304 /* don't allow userspace to shutdown the LOLND */
3305 if (LNET_NETTYP(net_id) == LOLND)
3308 mutex_lock(&the_lnet.ln_api_mutex);
3312 net = lnet_get_net_locked(net_id);
3314 CERROR("net %s not found\n",
3315 libcfs_net2str(net_id));
3320 addr = LNET_NIDADDR(conf->lic_nid);
3322 /* remove the entire net */
3323 net_count = lnet_get_net_ni_count_locked(net);
3327 /* create and link a new ping info, before removing the old one */
3328 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3329 lnet_get_ni_count() - net_count,
3332 goto unlock_api_mutex;
3334 lnet_shutdown_lndnet(net);
3336 lnet_acceptor_stop();
3338 lnet_ping_target_update(pbuf, ping_mdh);
3340 goto unlock_api_mutex;
3343 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3345 CERROR("nid %s not found\n",
3346 libcfs_nid2str(conf->lic_nid));
3351 net_count = lnet_get_net_ni_count_locked(net);
3355 /* create and link a new ping info, before removing the old one */
3356 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3357 lnet_get_ni_count() - 1, false);
3359 goto unlock_api_mutex;
3361 lnet_shutdown_lndni(ni);
3363 lnet_acceptor_stop();
3365 lnet_ping_target_update(pbuf, ping_mdh);
3367 /* check if the net is empty and remove it if it is */
3369 lnet_shutdown_lndnet(net);
3371 goto unlock_api_mutex;
3376 mutex_unlock(&the_lnet.ln_api_mutex);
3382 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3383 * They are only expected to be called for unique networks.
3384 * That can be as a result of older DLC library
3385 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3388 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3390 struct lnet_net *net;
3391 LIST_HEAD(net_head);
3393 struct lnet_ioctl_config_lnd_tunables tun;
3394 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3396 /* Create a net/ni structures for the network string */
3397 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3399 return rc == 0 ? -EINVAL : rc;
3401 mutex_lock(&the_lnet.ln_api_mutex);
3404 rc = -EINVAL; /* only add one network per call */
3405 goto out_unlock_clean;
3408 net = list_entry(net_head.next, struct lnet_net, net_list);
3409 list_del_init(&net->net_list);
3411 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3413 memset(&tun, 0, sizeof(tun));
3415 tun.lt_cmn.lct_peer_timeout =
3416 conf->cfg_config_u.cfg_net.net_peer_timeout;
3417 tun.lt_cmn.lct_peer_tx_credits =
3418 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3419 tun.lt_cmn.lct_peer_rtr_credits =
3420 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3421 tun.lt_cmn.lct_max_tx_credits =
3422 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3424 rc = lnet_add_net_common(net, &tun);
3427 mutex_unlock(&the_lnet.ln_api_mutex);
3428 while (!list_empty(&net_head)) {
3429 /* net_head list is empty in success case */
3430 net = list_entry(net_head.next, struct lnet_net, net_list);
3431 list_del_init(&net->net_list);
3438 lnet_dyn_del_net(__u32 net_id)
3440 struct lnet_net *net;
3441 struct lnet_ping_buffer *pbuf;
3442 struct lnet_handle_md ping_mdh;
3446 /* don't allow userspace to shutdown the LOLND */
3447 if (LNET_NETTYP(net_id) == LOLND)
3450 mutex_lock(&the_lnet.ln_api_mutex);
3454 net = lnet_get_net_locked(net_id);
3461 net_ni_count = lnet_get_net_ni_count_locked(net);
3465 /* create and link a new ping info, before removing the old one */
3466 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3467 lnet_get_ni_count() - net_ni_count, false);
3471 lnet_shutdown_lndnet(net);
3473 lnet_acceptor_stop();
3475 lnet_ping_target_update(pbuf, ping_mdh);
3478 mutex_unlock(&the_lnet.ln_api_mutex);
3483 void lnet_incr_dlc_seq(void)
3485 atomic_inc(&lnet_dlc_seq_no);
3488 __u32 lnet_get_dlc_seq_locked(void)
3490 return atomic_read(&lnet_dlc_seq_no);
3494 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3496 struct lnet_net *net;
3499 lnet_net_lock(LNET_LOCK_EX);
3500 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3501 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3502 if (ni->ni_nid == nid || all) {
3503 atomic_set(&ni->ni_healthv, value);
3504 if (list_empty(&ni->ni_recovery) &&
3505 value < LNET_MAX_HEALTH_VALUE) {
3506 CERROR("manually adding local NI %s to recovery\n",
3507 libcfs_nid2str(ni->ni_nid));
3508 list_add_tail(&ni->ni_recovery,
3509 &the_lnet.ln_mt_localNIRecovq);
3510 lnet_ni_addref_locked(ni, 0);
3513 lnet_net_unlock(LNET_LOCK_EX);
3519 lnet_net_unlock(LNET_LOCK_EX);
3523 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3527 lnet_nid_t nid = stats->hlni_nid;
3529 cpt = lnet_net_lock_current();
3530 ni = lnet_nid2ni_locked(nid, cpt);
3537 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3538 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3539 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3540 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3541 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3542 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3543 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3546 lnet_net_unlock(cpt);
3552 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3557 lnet_net_lock(LNET_LOCK_EX);
3558 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3559 list->rlst_nid_array[i] = ni->ni_nid;
3561 if (i >= LNET_MAX_SHOW_NUM_NID)
3564 lnet_net_unlock(LNET_LOCK_EX);
3565 list->rlst_num_nids = i;
3571 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3573 struct lnet_peer_ni *lpni;
3576 lnet_net_lock(LNET_LOCK_EX);
3577 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3578 list->rlst_nid_array[i] = lpni->lpni_nid;
3580 if (i >= LNET_MAX_SHOW_NUM_NID)
3583 lnet_net_unlock(LNET_LOCK_EX);
3584 list->rlst_num_nids = i;
3590 * LNet ioctl handler.
3594 LNetCtl(unsigned int cmd, void *arg)
3596 struct libcfs_ioctl_data *data = arg;
3597 struct lnet_ioctl_config_data *config;
3598 struct lnet_process_id id = {0};
3602 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3603 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3606 case IOC_LIBCFS_GET_NI:
3607 rc = LNetGetId(data->ioc_count, &id);
3608 data->ioc_nid = id.nid;
3611 case IOC_LIBCFS_FAIL_NID:
3612 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3614 case IOC_LIBCFS_ADD_ROUTE: {
3615 /* default router sensitivity to 1 */
3616 unsigned int sensitivity = 1;
3619 if (config->cfg_hdr.ioc_len < sizeof(*config))
3622 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3624 config->cfg_config_u.cfg_route.rtr_sensitivity;
3627 mutex_lock(&the_lnet.ln_api_mutex);
3628 rc = lnet_add_route(config->cfg_net,
3629 config->cfg_config_u.cfg_route.rtr_hop,
3631 config->cfg_config_u.cfg_route.
3632 rtr_priority, sensitivity);
3633 mutex_unlock(&the_lnet.ln_api_mutex);
3637 case IOC_LIBCFS_DEL_ROUTE:
3640 if (config->cfg_hdr.ioc_len < sizeof(*config))
3643 mutex_lock(&the_lnet.ln_api_mutex);
3644 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3645 mutex_unlock(&the_lnet.ln_api_mutex);
3648 case IOC_LIBCFS_GET_ROUTE:
3651 if (config->cfg_hdr.ioc_len < sizeof(*config))
3654 mutex_lock(&the_lnet.ln_api_mutex);
3655 rc = lnet_get_route(config->cfg_count,
3657 &config->cfg_config_u.cfg_route.rtr_hop,
3659 &config->cfg_config_u.cfg_route.rtr_flags,
3660 &config->cfg_config_u.cfg_route.
3662 &config->cfg_config_u.cfg_route.
3664 mutex_unlock(&the_lnet.ln_api_mutex);
3667 case IOC_LIBCFS_GET_LOCAL_NI: {
3668 struct lnet_ioctl_config_ni *cfg_ni;
3669 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3670 struct lnet_ioctl_element_stats *stats;
3675 /* get the tunables if they are available */
3676 if (cfg_ni->lic_cfg_hdr.ioc_len <
3677 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3680 stats = (struct lnet_ioctl_element_stats *)
3682 tun = (struct lnet_ioctl_config_lnd_tunables *)
3683 (cfg_ni->lic_bulk + sizeof(*stats));
3685 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3688 mutex_lock(&the_lnet.ln_api_mutex);
3689 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3690 mutex_unlock(&the_lnet.ln_api_mutex);
3694 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3695 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3697 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3700 mutex_lock(&the_lnet.ln_api_mutex);
3701 rc = lnet_get_ni_stats(msg_stats);
3702 mutex_unlock(&the_lnet.ln_api_mutex);
3707 case IOC_LIBCFS_GET_NET: {
3708 size_t total = sizeof(*config) +
3709 sizeof(struct lnet_ioctl_net_config);
3712 if (config->cfg_hdr.ioc_len < total)
3715 mutex_lock(&the_lnet.ln_api_mutex);
3716 rc = lnet_get_net_config(config);
3717 mutex_unlock(&the_lnet.ln_api_mutex);
3721 case IOC_LIBCFS_GET_LNET_STATS:
3723 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3725 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3728 mutex_lock(&the_lnet.ln_api_mutex);
3729 lnet_counters_get(&lnet_stats->st_cntrs);
3730 mutex_unlock(&the_lnet.ln_api_mutex);
3734 case IOC_LIBCFS_CONFIG_RTR:
3737 if (config->cfg_hdr.ioc_len < sizeof(*config))
3740 mutex_lock(&the_lnet.ln_api_mutex);
3741 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3742 rc = lnet_rtrpools_enable();
3743 mutex_unlock(&the_lnet.ln_api_mutex);
3746 lnet_rtrpools_disable();
3747 mutex_unlock(&the_lnet.ln_api_mutex);
3750 case IOC_LIBCFS_ADD_BUF:
3753 if (config->cfg_hdr.ioc_len < sizeof(*config))
3756 mutex_lock(&the_lnet.ln_api_mutex);
3757 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3759 config->cfg_config_u.cfg_buffers.
3761 config->cfg_config_u.cfg_buffers.
3763 mutex_unlock(&the_lnet.ln_api_mutex);
3766 case IOC_LIBCFS_SET_NUMA_RANGE: {
3767 struct lnet_ioctl_set_value *numa;
3769 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3771 lnet_net_lock(LNET_LOCK_EX);
3772 lnet_numa_range = numa->sv_value;
3773 lnet_net_unlock(LNET_LOCK_EX);
3777 case IOC_LIBCFS_GET_NUMA_RANGE: {
3778 struct lnet_ioctl_set_value *numa;
3780 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3782 numa->sv_value = lnet_numa_range;
3786 case IOC_LIBCFS_GET_BUF: {
3787 struct lnet_ioctl_pool_cfg *pool_cfg;
3788 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3792 if (config->cfg_hdr.ioc_len < total)
3795 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3797 mutex_lock(&the_lnet.ln_api_mutex);
3798 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3799 mutex_unlock(&the_lnet.ln_api_mutex);
3803 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3804 struct lnet_ioctl_local_ni_hstats *stats = arg;
3806 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3809 mutex_lock(&the_lnet.ln_api_mutex);
3810 rc = lnet_get_local_ni_hstats(stats);
3811 mutex_unlock(&the_lnet.ln_api_mutex);
3816 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3817 struct lnet_ioctl_recovery_list *list = arg;
3818 if (list->rlst_hdr.ioc_len < sizeof(*list))
3821 mutex_lock(&the_lnet.ln_api_mutex);
3822 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3823 rc = lnet_get_local_ni_recovery_list(list);
3825 rc = lnet_get_peer_ni_recovery_list(list);
3826 mutex_unlock(&the_lnet.ln_api_mutex);
3830 case IOC_LIBCFS_ADD_PEER_NI: {
3831 struct lnet_ioctl_peer_cfg *cfg = arg;
3833 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3836 mutex_lock(&the_lnet.ln_api_mutex);
3837 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3840 mutex_unlock(&the_lnet.ln_api_mutex);
3844 case IOC_LIBCFS_DEL_PEER_NI: {
3845 struct lnet_ioctl_peer_cfg *cfg = arg;
3847 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3850 mutex_lock(&the_lnet.ln_api_mutex);
3851 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3852 cfg->prcfg_cfg_nid);
3853 mutex_unlock(&the_lnet.ln_api_mutex);
3857 case IOC_LIBCFS_GET_PEER_INFO: {
3858 struct lnet_ioctl_peer *peer_info = arg;
3860 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3863 mutex_lock(&the_lnet.ln_api_mutex);
3864 rc = lnet_get_peer_ni_info(
3865 peer_info->pr_count,
3867 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3868 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3869 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3870 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3871 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3872 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3873 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3874 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3875 mutex_unlock(&the_lnet.ln_api_mutex);
3879 case IOC_LIBCFS_GET_PEER_NI: {
3880 struct lnet_ioctl_peer_cfg *cfg = arg;
3882 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3885 mutex_lock(&the_lnet.ln_api_mutex);
3886 rc = lnet_get_peer_info(cfg,
3887 (void __user *)cfg->prcfg_bulk);
3888 mutex_unlock(&the_lnet.ln_api_mutex);
3892 case IOC_LIBCFS_GET_PEER_LIST: {
3893 struct lnet_ioctl_peer_cfg *cfg = arg;
3895 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3898 mutex_lock(&the_lnet.ln_api_mutex);
3899 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3900 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3901 mutex_unlock(&the_lnet.ln_api_mutex);
3905 case IOC_LIBCFS_SET_HEALHV: {
3906 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3908 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3910 if (cfg->rh_value < 0 ||
3911 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3912 value = LNET_MAX_HEALTH_VALUE;
3914 value = cfg->rh_value;
3915 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3916 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3917 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3918 mutex_lock(&the_lnet.ln_api_mutex);
3919 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3920 lnet_ni_set_healthv(cfg->rh_nid, value,
3923 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3925 mutex_unlock(&the_lnet.ln_api_mutex);
3929 case IOC_LIBCFS_NOTIFY_ROUTER: {
3930 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3932 /* The deadline passed in by the user should be some time in
3933 * seconds in the future since the UNIX epoch. We have to map
3934 * that deadline to the wall clock.
3936 deadline += ktime_get_seconds();
3937 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3941 case IOC_LIBCFS_LNET_DIST:
3942 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3943 if (rc < 0 && rc != -EHOSTUNREACH)
3946 data->ioc_u32[0] = rc;
3949 case IOC_LIBCFS_TESTPROTOCOMPAT:
3950 the_lnet.ln_testprotocompat = data->ioc_flags;
3953 case IOC_LIBCFS_LNET_FAULT:
3954 return lnet_fault_ctl(data->ioc_flags, data);
3956 case IOC_LIBCFS_PING: {
3957 signed long timeout;
3959 id.nid = data->ioc_nid;
3960 id.pid = data->ioc_u32[0];
3962 /* If timeout is negative then set default of 3 minutes */
3963 if (((s32)data->ioc_u32[1] <= 0) ||
3964 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3965 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3967 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3969 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3970 data->ioc_plen1 / sizeof(struct lnet_process_id));
3975 data->ioc_count = rc;
3979 case IOC_LIBCFS_PING_PEER: {
3980 struct lnet_ioctl_ping_data *ping = arg;
3981 struct lnet_peer *lp;
3982 signed long timeout;
3984 /* If timeout is negative then set default of 3 minutes */
3985 if (((s32)ping->op_param) <= 0 ||
3986 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3987 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3989 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3991 rc = lnet_ping(ping->ping_id, timeout,
3997 mutex_lock(&the_lnet.ln_api_mutex);
3998 lp = lnet_find_peer(ping->ping_id.nid);
4000 ping->ping_id.nid = lp->lp_primary_nid;
4001 ping->mr_info = lnet_peer_is_multi_rail(lp);
4002 lnet_peer_decref_locked(lp);
4004 mutex_unlock(&the_lnet.ln_api_mutex);
4006 ping->ping_count = rc;
4010 case IOC_LIBCFS_DISCOVER: {
4011 struct lnet_ioctl_ping_data *discover = arg;
4012 struct lnet_peer *lp;
4014 rc = lnet_discover(discover->ping_id, discover->op_param,
4016 discover->ping_count);
4020 mutex_lock(&the_lnet.ln_api_mutex);
4021 lp = lnet_find_peer(discover->ping_id.nid);
4023 discover->ping_id.nid = lp->lp_primary_nid;
4024 discover->mr_info = lnet_peer_is_multi_rail(lp);
4025 lnet_peer_decref_locked(lp);
4027 mutex_unlock(&the_lnet.ln_api_mutex);
4029 discover->ping_count = rc;
4034 ni = lnet_net2ni_addref(data->ioc_net);
4038 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4041 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4048 EXPORT_SYMBOL(LNetCtl);
4050 void LNetDebugPeer(struct lnet_process_id id)
4052 lnet_debug_peer(id.nid);
4054 EXPORT_SYMBOL(LNetDebugPeer);
4057 * Determine if the specified peer \a nid is on the local node.
4059 * \param nid peer nid to check
4061 * \retval true If peer NID is on the local node.
4062 * \retval false If peer NID is not on the local node.
4064 bool LNetIsPeerLocal(lnet_nid_t nid)
4066 struct lnet_net *net;
4070 cpt = lnet_net_lock_current();
4071 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4072 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4073 if (ni->ni_nid == nid) {
4074 lnet_net_unlock(cpt);
4079 lnet_net_unlock(cpt);
4083 EXPORT_SYMBOL(LNetIsPeerLocal);
4086 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4087 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4089 * \param index Index of the interface to look up.
4090 * \param id On successful return, this location will hold the
4091 * struct lnet_process_id ID of the interface.
4093 * \retval 0 If an interface exists at \a index.
4094 * \retval -ENOENT If no interface has been found.
4097 LNetGetId(unsigned int index, struct lnet_process_id *id)
4100 struct lnet_net *net;
4104 LASSERT(the_lnet.ln_refcount > 0);
4106 cpt = lnet_net_lock_current();
4108 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4109 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4113 id->nid = ni->ni_nid;
4114 id->pid = the_lnet.ln_pid;
4120 lnet_net_unlock(cpt);
4123 EXPORT_SYMBOL(LNetGetId);
4128 struct lnet_handle_md mdh;
4129 struct completion completion;
4133 lnet_ping_event_handler(struct lnet_event *event)
4135 struct ping_data *pd = event->md.user_ptr;
4137 CDEBUG(D_NET, "ping event (%d %d)%s\n",
4138 event->type, event->status,
4139 event->unlinked ? " unlinked" : "");
4141 if (event->status) {
4143 pd->rc = event->status;
4144 } else if (event->type == LNET_EVENT_REPLY) {
4146 pd->rc = event->mlength;
4148 if (event->unlinked)
4149 complete(&pd->completion);
4152 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4153 struct lnet_process_id __user *ids, int n_ids)
4156 struct lnet_md md = { NULL };
4157 struct ping_data pd = { 0 };
4158 struct lnet_ping_buffer *pbuf;
4159 struct lnet_process_id tmpid;
4165 /* n_ids limit is arbitrary */
4166 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4170 * if the user buffer has more space than the lnet_interfaces_max
4171 * then only fill it up to lnet_interfaces_max
4173 if (n_ids > lnet_interfaces_max)
4174 n_ids = lnet_interfaces_max;
4176 if (id.pid == LNET_PID_ANY)
4177 id.pid = LNET_PID_LUSTRE;
4179 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4183 eq = LNetEQAlloc(lnet_ping_event_handler);
4186 CERROR("Can't allocate EQ: %d\n", rc);
4187 goto fail_ping_buffer_decref;
4190 /* initialize md content */
4191 md.start = &pbuf->pb_info;
4192 md.length = LNET_PING_INFO_SIZE(n_ids);
4193 md.threshold = 2; /* GET/REPLY */
4195 md.options = LNET_MD_TRUNCATE;
4199 init_completion(&pd.completion);
4201 rc = LNetMDBind(md, LNET_UNLINK, &pd.mdh);
4203 CERROR("Can't bind MD: %d\n", rc);
4207 rc = LNetGet(LNET_NID_ANY, pd.mdh, id,
4208 LNET_RESERVED_PORTAL,
4209 LNET_PROTO_PING_MATCHBITS, 0, false);
4212 /* Don't CERROR; this could be deliberate! */
4213 rc2 = LNetMDUnlink(pd.mdh);
4216 /* NB must wait for the UNLINK event below... */
4219 if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4220 /* Ensure completion in finite time... */
4221 LNetMDUnlink(pd.mdh);
4222 wait_for_completion(&pd.completion);
4230 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4232 rc = -EPROTO; /* if I can't parse... */
4235 CERROR("%s: ping info too short %d\n",
4236 libcfs_id2str(id), nob);
4240 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4241 lnet_swap_pinginfo(pbuf);
4242 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4243 CERROR("%s: Unexpected magic %08x\n",
4244 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4248 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4249 CERROR("%s: ping w/o NI status: 0x%x\n",
4250 libcfs_id2str(id), pbuf->pb_info.pi_features);
4254 if (nob < LNET_PING_INFO_SIZE(0)) {
4255 CERROR("%s: Short reply %d(%d min)\n",
4257 nob, (int)LNET_PING_INFO_SIZE(0));
4261 if (pbuf->pb_info.pi_nnis < n_ids)
4262 n_ids = pbuf->pb_info.pi_nnis;
4264 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4265 CERROR("%s: Short reply %d(%d expected)\n",
4267 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4271 rc = -EFAULT; /* if I segv in copy_to_user()... */
4273 memset(&tmpid, 0, sizeof(tmpid));
4274 for (i = 0; i < n_ids; i++) {
4275 tmpid.pid = pbuf->pb_info.pi_pid;
4276 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4277 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4280 rc = pbuf->pb_info.pi_nnis;
4283 rc2 = LNetEQFree(eq);
4285 CERROR("rc2 %d\n", rc2);
4288 fail_ping_buffer_decref:
4289 lnet_ping_buffer_decref(pbuf);
4294 lnet_discover(struct lnet_process_id id, __u32 force,
4295 struct lnet_process_id __user *ids, int n_ids)
4297 struct lnet_peer_ni *lpni;
4298 struct lnet_peer_ni *p;
4299 struct lnet_peer *lp;
4300 struct lnet_process_id *buf;
4304 int max_intf = lnet_interfaces_max;
4307 id.nid == LNET_NID_ANY)
4310 if (id.pid == LNET_PID_ANY)
4311 id.pid = LNET_PID_LUSTRE;
4314 * if the user buffer has more space than the max_intf
4315 * then only fill it up to max_intf
4317 if (n_ids > max_intf)
4320 CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4324 cpt = lnet_net_lock_current();
4325 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4332 * Clearing the NIDS_UPTODATE flag ensures the peer will
4333 * be discovered, provided discovery has not been disabled.
4335 lp = lpni->lpni_peer_net->lpn_peer;
4336 spin_lock(&lp->lp_lock);
4337 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4338 /* If the force flag is set, force a PING and PUSH as well. */
4340 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4341 spin_unlock(&lp->lp_lock);
4342 rc = lnet_discover_peer_locked(lpni, cpt, true);
4346 /* Peer may have changed. */
4347 lp = lpni->lpni_peer_net->lpn_peer;
4348 if (lp->lp_nnis < n_ids)
4349 n_ids = lp->lp_nnis;
4353 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4354 buf[i].pid = id.pid;
4355 buf[i].nid = p->lpni_nid;
4360 lnet_net_unlock(cpt);
4363 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4369 lnet_peer_ni_decref_locked(lpni);
4371 lnet_net_unlock(cpt);
4373 CFS_FREE_PTR_ARRAY(buf, n_ids);
4379 * Retrieve peer discovery status.
4381 * \retval 1 if lnet_peer_discovery_disabled is 0
4382 * \retval 0 if lnet_peer_discovery_disabled is 1
4385 LNetGetPeerDiscoveryStatus(void)
4387 return !lnet_peer_discovery_disabled;
4389 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);