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);
354 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
355 mutex_unlock(&the_lnet.ln_api_mutex);
359 /* tell peers that discovery setting has changed */
360 lnet_net_lock(LNET_LOCK_EX);
361 pbuf = the_lnet.ln_ping_target;
363 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
365 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
366 lnet_net_unlock(LNET_LOCK_EX);
368 lnet_push_update_to_peers(1);
370 mutex_unlock(&the_lnet.ln_api_mutex);
376 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
379 unsigned int *drop_asym_route = (unsigned int *)kp->arg;
382 rc = kstrtoul(val, 0, &value);
384 CERROR("Invalid module parameter value for "
385 "'lnet_drop_asym_route'\n");
390 * The purpose of locking the api_mutex here is to ensure that
391 * the correct value ends up stored properly.
393 mutex_lock(&the_lnet.ln_api_mutex);
395 if (value == *drop_asym_route) {
396 mutex_unlock(&the_lnet.ln_api_mutex);
400 *drop_asym_route = value;
402 mutex_unlock(&the_lnet.ln_api_mutex);
408 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
411 unsigned *transaction_to = (unsigned *)kp->arg;
414 rc = kstrtoul(val, 0, &value);
416 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
421 * The purpose of locking the api_mutex here is to ensure that
422 * the correct value ends up stored properly.
424 mutex_lock(&the_lnet.ln_api_mutex);
426 if (value < lnet_retry_count || value == 0) {
427 mutex_unlock(&the_lnet.ln_api_mutex);
428 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
429 "Has to be greater than lnet_retry_count (%u)\n",
430 value, lnet_retry_count);
434 if (value == *transaction_to) {
435 mutex_unlock(&the_lnet.ln_api_mutex);
439 *transaction_to = value;
440 if (lnet_retry_count == 0)
441 lnet_lnd_timeout = value;
443 lnet_lnd_timeout = value / lnet_retry_count;
445 mutex_unlock(&the_lnet.ln_api_mutex);
451 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
454 unsigned *retry_count = (unsigned *)kp->arg;
457 rc = kstrtoul(val, 0, &value);
459 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
464 * The purpose of locking the api_mutex here is to ensure that
465 * the correct value ends up stored properly.
467 mutex_lock(&the_lnet.ln_api_mutex);
469 if (lnet_health_sensitivity == 0) {
470 mutex_unlock(&the_lnet.ln_api_mutex);
471 CERROR("Can not set retry_count when health feature is turned off\n");
475 if (value > lnet_transaction_timeout) {
476 mutex_unlock(&the_lnet.ln_api_mutex);
477 CERROR("Invalid value for lnet_retry_count (%lu). "
478 "Has to be smaller than lnet_transaction_timeout (%u)\n",
479 value, lnet_transaction_timeout);
483 *retry_count = value;
486 lnet_lnd_timeout = lnet_transaction_timeout;
488 lnet_lnd_timeout = lnet_transaction_timeout / value;
490 mutex_unlock(&the_lnet.ln_api_mutex);
496 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
500 rc = kstrtoint(val, 0, &value);
502 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
506 if (value < LNET_INTERFACES_MIN) {
507 CWARN("max interfaces provided are too small, setting to %d\n",
508 LNET_INTERFACES_MAX_DEFAULT);
509 value = LNET_INTERFACES_MAX_DEFAULT;
512 *(int *)kp->arg = value;
518 lnet_get_routes(void)
524 lnet_get_networks(void)
529 if (*networks != 0 && *ip2nets != 0) {
530 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
531 "'ip2nets' but not both at once\n");
536 rc = lnet_parse_ip2nets(&nets, ip2nets);
537 return (rc == 0) ? nets : NULL;
547 lnet_init_locks(void)
549 spin_lock_init(&the_lnet.ln_eq_wait_lock);
550 spin_lock_init(&the_lnet.ln_msg_resend_lock);
551 init_completion(&the_lnet.ln_mt_wait_complete);
552 mutex_init(&the_lnet.ln_lnd_mutex);
555 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
556 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
558 struct kmem_cache *lnet_rspt_cachep; /* response tracker cache */
559 struct kmem_cache *lnet_msg_cachep;
562 lnet_slab_setup(void)
564 /* create specific kmem_cache for MEs and small MDs (i.e., originally
565 * allocated in <size-xxx> kmem_cache).
567 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
569 if (!lnet_mes_cachep)
572 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
573 LNET_SMALL_MD_SIZE, 0, 0,
575 if (!lnet_small_mds_cachep)
578 lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
580 if (!lnet_rspt_cachep)
583 lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
585 if (!lnet_msg_cachep)
592 lnet_slab_cleanup(void)
594 if (lnet_msg_cachep) {
595 kmem_cache_destroy(lnet_msg_cachep);
596 lnet_msg_cachep = NULL;
600 if (lnet_rspt_cachep) {
601 kmem_cache_destroy(lnet_rspt_cachep);
602 lnet_rspt_cachep = NULL;
605 if (lnet_small_mds_cachep) {
606 kmem_cache_destroy(lnet_small_mds_cachep);
607 lnet_small_mds_cachep = NULL;
610 if (lnet_mes_cachep) {
611 kmem_cache_destroy(lnet_mes_cachep);
612 lnet_mes_cachep = NULL;
617 lnet_create_remote_nets_table(void)
620 struct list_head *hash;
622 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
623 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
624 CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
626 CERROR("Failed to create remote nets hash table\n");
630 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
631 INIT_LIST_HEAD(&hash[i]);
632 the_lnet.ln_remote_nets_hash = hash;
637 lnet_destroy_remote_nets_table(void)
641 if (the_lnet.ln_remote_nets_hash == NULL)
644 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
645 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
647 CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
648 LNET_REMOTE_NETS_HASH_SIZE);
649 the_lnet.ln_remote_nets_hash = NULL;
653 lnet_destroy_locks(void)
655 if (the_lnet.ln_res_lock != NULL) {
656 cfs_percpt_lock_free(the_lnet.ln_res_lock);
657 the_lnet.ln_res_lock = NULL;
660 if (the_lnet.ln_net_lock != NULL) {
661 cfs_percpt_lock_free(the_lnet.ln_net_lock);
662 the_lnet.ln_net_lock = NULL;
667 lnet_create_locks(void)
671 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
672 if (the_lnet.ln_res_lock == NULL)
675 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
676 if (the_lnet.ln_net_lock == NULL)
682 lnet_destroy_locks();
686 static void lnet_assert_wire_constants(void)
688 /* Wire protocol assertions generated by 'wirecheck'
689 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
690 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
691 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
694 BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
695 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
696 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
697 BUILD_BUG_ON(LNET_MSG_ACK != 0);
698 BUILD_BUG_ON(LNET_MSG_PUT != 1);
699 BUILD_BUG_ON(LNET_MSG_GET != 2);
700 BUILD_BUG_ON(LNET_MSG_REPLY != 3);
701 BUILD_BUG_ON(LNET_MSG_HELLO != 4);
703 /* Checks for struct lnet_handle_wire */
704 BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
705 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
706 wh_interface_cookie) != 0);
707 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
708 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
709 wh_object_cookie) != 8);
710 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
712 /* Checks for struct struct lnet_magicversion */
713 BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
714 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
715 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
716 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
717 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
718 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
719 version_minor) != 6);
720 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
722 /* Checks for struct struct lnet_hdr */
723 BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
724 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
725 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
726 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
727 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
728 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
729 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
730 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
731 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
732 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
733 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
734 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
735 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
736 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
737 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
740 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
741 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
742 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
743 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
744 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
745 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
748 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
749 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
750 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
751 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
752 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
753 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
754 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
755 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
756 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
757 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
760 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
761 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
762 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
763 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
764 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
765 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
766 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
767 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
768 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
769 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
772 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
773 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
776 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
777 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
778 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
779 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
781 /* Checks for struct lnet_ni_status and related constants */
782 BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
783 BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
784 BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
786 /* Checks for struct lnet_ni_status */
787 BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
788 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
789 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
790 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
791 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
792 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
793 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
795 /* Checks for struct lnet_ping_info and related constants */
796 BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
797 BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
798 BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
799 BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
800 BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
801 BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
802 BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
803 BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
805 /* Checks for struct lnet_ping_info */
806 BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
807 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
808 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
809 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
810 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
811 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
812 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
813 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
814 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
815 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
816 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
819 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
821 const struct lnet_lnd *lnd;
823 /* holding lnd mutex */
824 if (type >= NUM_LNDS)
826 lnd = the_lnet.ln_lnds[type];
827 LASSERT(!lnd || lnd->lnd_type == type);
833 lnet_get_lnd_timeout(void)
835 return lnet_lnd_timeout;
837 EXPORT_SYMBOL(lnet_get_lnd_timeout);
840 lnet_register_lnd(const struct lnet_lnd *lnd)
842 mutex_lock(&the_lnet.ln_lnd_mutex);
844 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
845 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
847 the_lnet.ln_lnds[lnd->lnd_type] = lnd;
849 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
851 mutex_unlock(&the_lnet.ln_lnd_mutex);
853 EXPORT_SYMBOL(lnet_register_lnd);
856 lnet_unregister_lnd(const struct lnet_lnd *lnd)
858 mutex_lock(&the_lnet.ln_lnd_mutex);
860 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
862 the_lnet.ln_lnds[lnd->lnd_type] = NULL;
863 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
865 mutex_unlock(&the_lnet.ln_lnd_mutex);
867 EXPORT_SYMBOL(lnet_unregister_lnd);
870 lnet_counters_get_common(struct lnet_counters_common *common)
872 struct lnet_counters *ctr;
875 memset(common, 0, sizeof(*common));
877 lnet_net_lock(LNET_LOCK_EX);
879 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
880 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
881 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
882 common->lcc_errors += ctr->lct_common.lcc_errors;
883 common->lcc_send_count += ctr->lct_common.lcc_send_count;
884 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
885 common->lcc_route_count += ctr->lct_common.lcc_route_count;
886 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
887 common->lcc_send_length += ctr->lct_common.lcc_send_length;
888 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
889 common->lcc_route_length += ctr->lct_common.lcc_route_length;
890 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
892 lnet_net_unlock(LNET_LOCK_EX);
894 EXPORT_SYMBOL(lnet_counters_get_common);
897 lnet_counters_get(struct lnet_counters *counters)
899 struct lnet_counters *ctr;
900 struct lnet_counters_health *health = &counters->lct_health;
903 memset(counters, 0, sizeof(*counters));
905 lnet_counters_get_common(&counters->lct_common);
907 lnet_net_lock(LNET_LOCK_EX);
909 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
910 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
911 health->lch_resend_count += ctr->lct_health.lch_resend_count;
912 health->lch_response_timeout_count +=
913 ctr->lct_health.lch_response_timeout_count;
914 health->lch_local_interrupt_count +=
915 ctr->lct_health.lch_local_interrupt_count;
916 health->lch_local_dropped_count +=
917 ctr->lct_health.lch_local_dropped_count;
918 health->lch_local_aborted_count +=
919 ctr->lct_health.lch_local_aborted_count;
920 health->lch_local_no_route_count +=
921 ctr->lct_health.lch_local_no_route_count;
922 health->lch_local_timeout_count +=
923 ctr->lct_health.lch_local_timeout_count;
924 health->lch_local_error_count +=
925 ctr->lct_health.lch_local_error_count;
926 health->lch_remote_dropped_count +=
927 ctr->lct_health.lch_remote_dropped_count;
928 health->lch_remote_error_count +=
929 ctr->lct_health.lch_remote_error_count;
930 health->lch_remote_timeout_count +=
931 ctr->lct_health.lch_remote_timeout_count;
932 health->lch_network_timeout_count +=
933 ctr->lct_health.lch_network_timeout_count;
935 lnet_net_unlock(LNET_LOCK_EX);
937 EXPORT_SYMBOL(lnet_counters_get);
940 lnet_counters_reset(void)
942 struct lnet_counters *counters;
945 lnet_net_lock(LNET_LOCK_EX);
947 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
948 memset(counters, 0, sizeof(struct lnet_counters));
950 lnet_net_unlock(LNET_LOCK_EX);
954 lnet_res_type2str(int type)
959 case LNET_COOKIE_TYPE_MD:
961 case LNET_COOKIE_TYPE_ME:
963 case LNET_COOKIE_TYPE_EQ:
969 lnet_res_container_cleanup(struct lnet_res_container *rec)
973 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
976 while (!list_empty(&rec->rec_active)) {
977 struct list_head *e = rec->rec_active.next;
980 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
981 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
983 } else { /* NB: Active MEs should be attached on portals */
990 /* Found alive MD/ME/EQ, user really should unlink/free
991 * all of them before finalize LNet, but if someone didn't,
992 * we have to recycle garbage for him */
993 CERROR("%d active elements on exit of %s container\n",
994 count, lnet_res_type2str(rec->rec_type));
997 if (rec->rec_lh_hash != NULL) {
998 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
999 rec->rec_lh_hash = NULL;
1002 rec->rec_type = 0; /* mark it as finalized */
1006 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1011 LASSERT(rec->rec_type == 0);
1013 rec->rec_type = type;
1014 INIT_LIST_HEAD(&rec->rec_active);
1016 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1018 /* Arbitrary choice of hash table size */
1019 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1020 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1021 if (rec->rec_lh_hash == NULL) {
1026 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1027 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1032 CERROR("Failed to setup %s resource container\n",
1033 lnet_res_type2str(type));
1034 lnet_res_container_cleanup(rec);
1039 lnet_res_containers_destroy(struct lnet_res_container **recs)
1041 struct lnet_res_container *rec;
1044 cfs_percpt_for_each(rec, i, recs)
1045 lnet_res_container_cleanup(rec);
1047 cfs_percpt_free(recs);
1050 static struct lnet_res_container **
1051 lnet_res_containers_create(int type)
1053 struct lnet_res_container **recs;
1054 struct lnet_res_container *rec;
1058 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1060 CERROR("Failed to allocate %s resource containers\n",
1061 lnet_res_type2str(type));
1065 cfs_percpt_for_each(rec, i, recs) {
1066 rc = lnet_res_container_setup(rec, i, type);
1068 lnet_res_containers_destroy(recs);
1076 struct lnet_libhandle *
1077 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1079 /* ALWAYS called with lnet_res_lock held */
1080 struct list_head *head;
1081 struct lnet_libhandle *lh;
1084 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1087 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1088 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1090 list_for_each_entry(lh, head, lh_hash_chain) {
1091 if (lh->lh_cookie == cookie)
1099 lnet_res_lh_initialize(struct lnet_res_container *rec,
1100 struct lnet_libhandle *lh)
1102 /* ALWAYS called with lnet_res_lock held */
1103 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1106 lh->lh_cookie = rec->rec_lh_cookie;
1107 rec->rec_lh_cookie += 1 << ibits;
1109 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1111 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1115 lnet_create_array_of_queues(void)
1117 struct list_head **qs;
1118 struct list_head *q;
1121 qs = cfs_percpt_alloc(lnet_cpt_table(),
1122 sizeof(struct list_head));
1124 CERROR("Failed to allocate queues\n");
1128 cfs_percpt_for_each(q, i, qs)
1134 static int lnet_unprepare(void);
1137 lnet_prepare(lnet_pid_t requested_pid)
1139 /* Prepare to bring up the network */
1140 struct lnet_res_container **recs;
1143 if (requested_pid == LNET_PID_ANY) {
1144 /* Don't instantiate LNET just for me */
1148 LASSERT(the_lnet.ln_refcount == 0);
1150 the_lnet.ln_routing = 0;
1152 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1153 the_lnet.ln_pid = requested_pid;
1155 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1156 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1157 INIT_LIST_HEAD(&the_lnet.ln_nets);
1158 INIT_LIST_HEAD(&the_lnet.ln_routers);
1159 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1160 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1161 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1162 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1163 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1164 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1165 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1166 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1167 the_lnet.ln_mt_eq = NULL;
1168 init_completion(&the_lnet.ln_started);
1170 rc = lnet_slab_setup();
1174 rc = lnet_create_remote_nets_table();
1179 * NB the interface cookie in wire handles guards against delayed
1180 * replies and ACKs appearing valid after reboot.
1182 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1184 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1185 sizeof(struct lnet_counters));
1186 if (the_lnet.ln_counters == NULL) {
1187 CERROR("Failed to allocate counters for LNet\n");
1192 rc = lnet_peer_tables_create();
1196 rc = lnet_msg_containers_create();
1200 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1201 LNET_COOKIE_TYPE_EQ);
1205 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1211 the_lnet.ln_md_containers = recs;
1213 rc = lnet_portals_create();
1215 CERROR("Failed to create portals for LNet: %d\n", rc);
1219 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1220 if (!the_lnet.ln_mt_zombie_rstqs) {
1233 lnet_unprepare (void)
1237 /* NB no LNET_LOCK since this is the last reference. All LND instances
1238 * have shut down already, so it is safe to unlink and free all
1239 * descriptors, even those that appear committed to a network op (eg MD
1240 * with non-zero pending count) */
1242 lnet_fail_nid(LNET_NID_ANY, 0);
1244 LASSERT(the_lnet.ln_refcount == 0);
1245 LASSERT(list_empty(&the_lnet.ln_test_peers));
1246 LASSERT(list_empty(&the_lnet.ln_nets));
1248 if (the_lnet.ln_mt_zombie_rstqs) {
1249 lnet_clean_zombie_rstqs();
1250 the_lnet.ln_mt_zombie_rstqs = NULL;
1253 if (the_lnet.ln_mt_eq) {
1254 rc = LNetEQFree(the_lnet.ln_mt_eq);
1255 the_lnet.ln_mt_eq = NULL;
1259 lnet_portals_destroy();
1261 if (the_lnet.ln_md_containers != NULL) {
1262 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1263 the_lnet.ln_md_containers = NULL;
1266 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1268 lnet_msg_containers_destroy();
1270 lnet_rtrpools_free(0);
1272 if (the_lnet.ln_counters != NULL) {
1273 cfs_percpt_free(the_lnet.ln_counters);
1274 the_lnet.ln_counters = NULL;
1276 lnet_destroy_remote_nets_table();
1277 lnet_slab_cleanup();
1283 lnet_net2ni_locked(__u32 net_id, int cpt)
1286 struct lnet_net *net;
1288 LASSERT(cpt != LNET_LOCK_EX);
1290 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1291 if (net->net_id == net_id) {
1292 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1302 lnet_net2ni_addref(__u32 net)
1307 ni = lnet_net2ni_locked(net, 0);
1309 lnet_ni_addref_locked(ni, 0);
1314 EXPORT_SYMBOL(lnet_net2ni_addref);
1317 lnet_get_net_locked(__u32 net_id)
1319 struct lnet_net *net;
1321 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1322 if (net->net_id == net_id)
1330 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1335 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1340 val = hash_long(key, LNET_CPT_BITS);
1341 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1345 return (unsigned int)(key + val + (val >> 1)) % number;
1349 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1351 struct lnet_net *net;
1353 /* must called with hold of lnet_net_lock */
1354 if (LNET_CPT_NUMBER == 1)
1355 return 0; /* the only one */
1358 * If NI is provided then use the CPT identified in the NI cpt
1359 * list if one exists. If one doesn't exist, then that NI is
1360 * associated with all CPTs and it follows that the net it belongs
1361 * to is implicitly associated with all CPTs, so just hash the nid
1365 if (ni->ni_cpts != NULL)
1366 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1369 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1372 /* no NI provided so look at the net */
1373 net = lnet_get_net_locked(LNET_NIDNET(nid));
1375 if (net != NULL && net->net_cpts != NULL) {
1376 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1379 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1383 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1388 if (LNET_CPT_NUMBER == 1)
1389 return 0; /* the only one */
1391 cpt = lnet_net_lock_current();
1393 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1395 lnet_net_unlock(cpt);
1399 EXPORT_SYMBOL(lnet_cpt_of_nid);
1402 lnet_islocalnet_locked(__u32 net_id)
1404 struct lnet_net *net;
1407 net = lnet_get_net_locked(net_id);
1409 local = net != NULL;
1415 lnet_islocalnet(__u32 net_id)
1420 cpt = lnet_net_lock_current();
1422 local = lnet_islocalnet_locked(net_id);
1424 lnet_net_unlock(cpt);
1430 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1432 struct lnet_net *net;
1435 LASSERT(cpt != LNET_LOCK_EX);
1437 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1438 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1439 if (ni->ni_nid == nid)
1448 lnet_nid2ni_addref(lnet_nid_t nid)
1453 ni = lnet_nid2ni_locked(nid, 0);
1455 lnet_ni_addref_locked(ni, 0);
1460 EXPORT_SYMBOL(lnet_nid2ni_addref);
1463 lnet_islocalnid(lnet_nid_t nid)
1468 cpt = lnet_net_lock_current();
1469 ni = lnet_nid2ni_locked(nid, cpt);
1470 lnet_net_unlock(cpt);
1476 lnet_count_acceptor_nets(void)
1478 /* Return the # of NIs that need the acceptor. */
1480 struct lnet_net *net;
1483 cpt = lnet_net_lock_current();
1484 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1485 /* all socklnd type networks should have the acceptor
1487 if (net->net_lnd->lnd_accept != NULL)
1491 lnet_net_unlock(cpt);
1496 struct lnet_ping_buffer *
1497 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1499 struct lnet_ping_buffer *pbuf;
1501 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1503 pbuf->pb_nnis = nnis;
1504 pbuf->pb_needs_post = false;
1505 atomic_set(&pbuf->pb_refcnt, 1);
1512 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1514 LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1515 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1518 static struct lnet_ping_buffer *
1519 lnet_ping_target_create(int nnis)
1521 struct lnet_ping_buffer *pbuf;
1523 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1525 CERROR("Can't allocate ping source [%d]\n", nnis);
1529 pbuf->pb_info.pi_nnis = nnis;
1530 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1531 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1532 pbuf->pb_info.pi_features =
1533 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1539 lnet_get_net_ni_count_locked(struct lnet_net *net)
1544 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1551 lnet_get_net_ni_count_pre(struct lnet_net *net)
1556 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1563 lnet_get_ni_count(void)
1566 struct lnet_net *net;
1571 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1572 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1582 lnet_get_net_count(void)
1584 struct lnet_net *net;
1589 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1599 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1601 struct lnet_ni_status *stat;
1605 __swab32s(&pbuf->pb_info.pi_magic);
1606 __swab32s(&pbuf->pb_info.pi_features);
1607 __swab32s(&pbuf->pb_info.pi_pid);
1608 __swab32s(&pbuf->pb_info.pi_nnis);
1609 nnis = pbuf->pb_info.pi_nnis;
1610 if (nnis > pbuf->pb_nnis)
1611 nnis = pbuf->pb_nnis;
1612 for (i = 0; i < nnis; i++) {
1613 stat = &pbuf->pb_info.pi_ni[i];
1614 __swab64s(&stat->ns_nid);
1615 __swab32s(&stat->ns_status);
1620 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1624 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1626 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1628 /* Loopback is guaranteed to be present */
1629 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1631 if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1637 lnet_ping_target_destroy(void)
1639 struct lnet_net *net;
1642 lnet_net_lock(LNET_LOCK_EX);
1644 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1645 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1647 ni->ni_status = NULL;
1652 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1653 the_lnet.ln_ping_target = NULL;
1655 lnet_net_unlock(LNET_LOCK_EX);
1659 lnet_ping_target_event_handler(struct lnet_event *event)
1661 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1663 if (event->unlinked)
1664 lnet_ping_buffer_decref(pbuf);
1668 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1669 struct lnet_handle_md *ping_mdh,
1670 int ni_count, bool set_eq)
1672 struct lnet_process_id id = {
1673 .nid = LNET_NID_ANY,
1677 struct lnet_md md = { NULL };
1681 the_lnet.ln_ping_target_eq =
1682 LNetEQAlloc(lnet_ping_target_event_handler);
1683 if (IS_ERR(the_lnet.ln_ping_target_eq)) {
1684 rc = PTR_ERR(the_lnet.ln_ping_target_eq);
1685 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1690 *ppbuf = lnet_ping_target_create(ni_count);
1691 if (*ppbuf == NULL) {
1696 /* Ping target ME/MD */
1697 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1698 LNET_PROTO_PING_MATCHBITS, 0,
1699 LNET_UNLINK, LNET_INS_AFTER);
1702 CERROR("Can't create ping target ME: %d\n", rc);
1703 goto fail_decref_ping_buffer;
1706 /* initialize md content */
1707 md.start = &(*ppbuf)->pb_info;
1708 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1709 md.threshold = LNET_MD_THRESH_INF;
1711 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1712 LNET_MD_MANAGE_REMOTE;
1713 md.eq_handle = the_lnet.ln_ping_target_eq;
1714 md.user_ptr = *ppbuf;
1716 rc = LNetMDAttach(me, md, LNET_RETAIN, ping_mdh);
1718 CERROR("Can't attach ping target MD: %d\n", rc);
1719 goto fail_unlink_ping_me;
1721 lnet_ping_buffer_addref(*ppbuf);
1725 fail_unlink_ping_me:
1727 fail_decref_ping_buffer:
1728 LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1729 lnet_ping_buffer_decref(*ppbuf);
1733 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1740 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1741 struct lnet_handle_md *ping_mdh)
1743 LNetMDUnlink(*ping_mdh);
1744 LNetInvalidateMDHandle(ping_mdh);
1746 /* NB the MD could be busy; this just starts the unlink */
1747 while (atomic_read(&pbuf->pb_refcnt) > 1) {
1748 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1749 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1754 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1757 struct lnet_net *net;
1758 struct lnet_ni_status *ns;
1763 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1764 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1765 LASSERT(i < pbuf->pb_nnis);
1767 ns = &pbuf->pb_info.pi_ni[i];
1769 ns->ns_nid = ni->ni_nid;
1772 ns->ns_status = (ni->ni_status != NULL) ?
1773 ni->ni_status->ns_status :
1782 * We (ab)use the ns_status of the loopback interface to
1783 * transmit the sequence number. The first interface listed
1784 * must be the loopback interface.
1786 rc = lnet_ping_info_validate(&pbuf->pb_info);
1788 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1791 LNET_PING_BUFFER_SEQNO(pbuf) =
1792 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1796 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1797 struct lnet_handle_md ping_mdh)
1799 struct lnet_ping_buffer *old_pbuf = NULL;
1800 struct lnet_handle_md old_ping_md;
1802 /* switch the NIs to point to the new ping info created */
1803 lnet_net_lock(LNET_LOCK_EX);
1805 if (!the_lnet.ln_routing)
1806 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1807 if (!lnet_peer_discovery_disabled)
1808 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1810 /* Ensure only known feature bits have been set. */
1811 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1812 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1814 lnet_ping_target_install_locked(pbuf);
1816 if (the_lnet.ln_ping_target) {
1817 old_pbuf = the_lnet.ln_ping_target;
1818 old_ping_md = the_lnet.ln_ping_target_md;
1820 the_lnet.ln_ping_target_md = ping_mdh;
1821 the_lnet.ln_ping_target = pbuf;
1823 lnet_net_unlock(LNET_LOCK_EX);
1826 /* unlink and free the old ping info */
1827 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1828 lnet_ping_buffer_decref(old_pbuf);
1831 lnet_push_update_to_peers(0);
1835 lnet_ping_target_fini(void)
1839 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1840 &the_lnet.ln_ping_target_md);
1842 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1845 lnet_ping_target_destroy();
1848 /* Resize the push target. */
1849 int lnet_push_target_resize(void)
1851 struct lnet_handle_md mdh;
1852 struct lnet_handle_md old_mdh;
1853 struct lnet_ping_buffer *pbuf;
1854 struct lnet_ping_buffer *old_pbuf;
1859 nnis = the_lnet.ln_push_target_nnis;
1861 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
1865 /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
1866 * dropped when we need to resize again (see "old_pbuf" below) or when
1867 * LNet is shutdown (see lnet_push_target_fini())
1869 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1871 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
1875 rc = lnet_push_target_post(pbuf, &mdh);
1877 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
1878 lnet_ping_buffer_decref(pbuf);
1882 lnet_net_lock(LNET_LOCK_EX);
1883 old_pbuf = the_lnet.ln_push_target;
1884 old_mdh = the_lnet.ln_push_target_md;
1885 the_lnet.ln_push_target = pbuf;
1886 the_lnet.ln_push_target_md = mdh;
1887 lnet_net_unlock(LNET_LOCK_EX);
1890 LNetMDUnlink(old_mdh);
1891 /* Drop ref set by lnet_ping_buffer_alloc() */
1892 lnet_ping_buffer_decref(old_pbuf);
1895 /* Received another push or reply that requires a larger buffer */
1896 if (nnis < the_lnet.ln_push_target_nnis)
1899 CDEBUG(D_NET, "nnis %d success\n", nnis);
1903 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
1904 struct lnet_handle_md *mdhp)
1906 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1907 struct lnet_md md = { NULL };
1911 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1912 LNET_PROTO_PING_MATCHBITS, 0,
1913 LNET_UNLINK, LNET_INS_AFTER);
1916 CERROR("Can't create push target ME: %d\n", rc);
1920 pbuf->pb_needs_post = false;
1922 /* This reference is dropped by lnet_push_target_event_handler() */
1923 lnet_ping_buffer_addref(pbuf);
1925 /* initialize md content */
1926 md.start = &pbuf->pb_info;
1927 md.length = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
1930 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
1932 md.eq_handle = the_lnet.ln_push_target_eq;
1934 rc = LNetMDAttach(me, md, LNET_UNLINK, mdhp);
1936 CERROR("Can't attach push MD: %d\n", rc);
1938 lnet_ping_buffer_decref(pbuf);
1939 pbuf->pb_needs_post = true;
1943 CDEBUG(D_NET, "posted push target %p\n", pbuf);
1948 static void lnet_push_target_event_handler(struct lnet_event *ev)
1950 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1952 CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
1955 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1956 lnet_swap_pinginfo(pbuf);
1958 if (ev->type == LNET_EVENT_UNLINK) {
1959 /* Drop ref added by lnet_push_target_post() */
1960 lnet_ping_buffer_decref(pbuf);
1964 lnet_peer_push_event(ev);
1966 /* Drop ref added by lnet_push_target_post */
1967 lnet_ping_buffer_decref(pbuf);
1970 /* Initialize the push target. */
1971 static int lnet_push_target_init(void)
1975 if (the_lnet.ln_push_target)
1978 the_lnet.ln_push_target_eq =
1979 LNetEQAlloc(lnet_push_target_event_handler);
1980 if (IS_ERR(the_lnet.ln_push_target_eq)) {
1981 rc = PTR_ERR(the_lnet.ln_push_target_eq);
1982 CERROR("Can't allocated push target EQ: %d\n", rc);
1986 rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
1989 /* Start at the required minimum, we'll enlarge if required. */
1990 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1992 rc = lnet_push_target_resize();
1995 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
1996 LNetEQFree(the_lnet.ln_push_target_eq);
1997 the_lnet.ln_push_target_eq = NULL;
2003 /* Clean up the push target. */
2004 static void lnet_push_target_fini(void)
2006 if (!the_lnet.ln_push_target)
2009 /* Unlink and invalidate to prevent new references. */
2010 LNetMDUnlink(the_lnet.ln_push_target_md);
2011 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2013 /* Wait for the unlink to complete. */
2014 while (atomic_read(&the_lnet.ln_push_target->pb_refcnt) > 1) {
2015 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
2016 schedule_timeout_uninterruptible(cfs_time_seconds(1));
2019 /* Drop ref set by lnet_ping_buffer_alloc() */
2020 lnet_ping_buffer_decref(the_lnet.ln_push_target);
2021 the_lnet.ln_push_target = NULL;
2022 the_lnet.ln_push_target_nnis = 0;
2024 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2025 LNetEQFree(the_lnet.ln_push_target_eq);
2026 the_lnet.ln_push_target_eq = NULL;
2030 lnet_ni_tq_credits(struct lnet_ni *ni)
2034 LASSERT(ni->ni_ncpts >= 1);
2036 if (ni->ni_ncpts == 1)
2037 return ni->ni_net->net_tunables.lct_max_tx_credits;
2039 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2040 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2041 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2047 lnet_ni_unlink_locked(struct lnet_ni *ni)
2049 /* move it to zombie list and nobody can find it anymore */
2050 LASSERT(!list_empty(&ni->ni_netlist));
2051 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2052 lnet_ni_decref_locked(ni, 0);
2056 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2061 struct list_head *zombie_list = &net->net_ni_zombie;
2064 * Now wait for the NIs I just nuked to show up on the zombie
2065 * list and shut them down in guaranteed thread context
2068 while (!list_empty(zombie_list)) {
2072 ni = list_entry(zombie_list->next,
2073 struct lnet_ni, ni_netlist);
2074 list_del_init(&ni->ni_netlist);
2075 /* the ni should be in deleting state. If it's not it's
2077 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2078 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2081 /* still busy, add it back to zombie list */
2082 list_add(&ni->ni_netlist, zombie_list);
2086 if (!list_empty(&ni->ni_netlist)) {
2087 lnet_net_unlock(LNET_LOCK_EX);
2089 if ((i & (-i)) == i) {
2091 "Waiting for zombie LNI %s\n",
2092 libcfs_nid2str(ni->ni_nid));
2094 schedule_timeout_uninterruptible(cfs_time_seconds(1));
2095 lnet_net_lock(LNET_LOCK_EX);
2099 lnet_net_unlock(LNET_LOCK_EX);
2101 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2103 LASSERT(!in_interrupt());
2104 /* Holding the mutex makes it safe for lnd_shutdown
2105 * to call module_put(). Module unload cannot finish
2106 * until lnet_unregister_lnd() completes, and that
2107 * requires the mutex.
2109 mutex_lock(&the_lnet.ln_lnd_mutex);
2110 (net->net_lnd->lnd_shutdown)(ni);
2111 mutex_unlock(&the_lnet.ln_lnd_mutex);
2114 CDEBUG(D_LNI, "Removed LNI %s\n",
2115 libcfs_nid2str(ni->ni_nid));
2119 lnet_net_lock(LNET_LOCK_EX);
2123 /* shutdown down the NI and release refcount */
2125 lnet_shutdown_lndni(struct lnet_ni *ni)
2128 struct lnet_net *net = ni->ni_net;
2130 lnet_net_lock(LNET_LOCK_EX);
2132 ni->ni_state = LNET_NI_STATE_DELETING;
2134 lnet_ni_unlink_locked(ni);
2135 lnet_incr_dlc_seq();
2136 lnet_net_unlock(LNET_LOCK_EX);
2138 /* clear messages for this NI on the lazy portal */
2139 for (i = 0; i < the_lnet.ln_nportals; i++)
2140 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2142 lnet_net_lock(LNET_LOCK_EX);
2143 lnet_clear_zombies_nis_locked(net);
2144 lnet_net_unlock(LNET_LOCK_EX);
2148 lnet_shutdown_lndnet(struct lnet_net *net)
2152 lnet_net_lock(LNET_LOCK_EX);
2154 list_del_init(&net->net_list);
2156 while (!list_empty(&net->net_ni_list)) {
2157 ni = list_entry(net->net_ni_list.next,
2158 struct lnet_ni, ni_netlist);
2159 lnet_net_unlock(LNET_LOCK_EX);
2160 lnet_shutdown_lndni(ni);
2161 lnet_net_lock(LNET_LOCK_EX);
2164 lnet_net_unlock(LNET_LOCK_EX);
2166 /* Do peer table cleanup for this net */
2167 lnet_peer_tables_cleanup(net);
2173 lnet_shutdown_lndnets(void)
2175 struct lnet_net *net;
2177 struct lnet_msg *msg, *tmp;
2179 /* NB called holding the global mutex */
2181 /* All quiet on the API front */
2182 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2183 LASSERT(the_lnet.ln_refcount == 0);
2185 lnet_net_lock(LNET_LOCK_EX);
2186 the_lnet.ln_state = LNET_STATE_STOPPING;
2189 * move the nets to the zombie list to avoid them being
2190 * picked up for new work. LONET is also included in the
2191 * Nets that will be moved to the zombie list
2193 list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2195 /* Drop the cached loopback Net. */
2196 if (the_lnet.ln_loni != NULL) {
2197 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2198 the_lnet.ln_loni = NULL;
2200 lnet_net_unlock(LNET_LOCK_EX);
2202 /* iterate through the net zombie list and delete each net */
2203 while (!list_empty(&the_lnet.ln_net_zombie)) {
2204 net = list_entry(the_lnet.ln_net_zombie.next,
2205 struct lnet_net, net_list);
2206 lnet_shutdown_lndnet(net);
2209 spin_lock(&the_lnet.ln_msg_resend_lock);
2210 list_splice(&the_lnet.ln_msg_resend, &resend);
2211 spin_unlock(&the_lnet.ln_msg_resend_lock);
2213 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2214 list_del_init(&msg->msg_list);
2215 msg->msg_no_resend = true;
2216 lnet_finalize(msg, -ECANCELED);
2219 lnet_net_lock(LNET_LOCK_EX);
2220 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2221 lnet_net_unlock(LNET_LOCK_EX);
2225 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2228 struct lnet_tx_queue *tq;
2230 struct lnet_net *net = ni->ni_net;
2232 mutex_lock(&the_lnet.ln_lnd_mutex);
2235 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2236 ni->ni_lnd_tunables_set = true;
2239 rc = (net->net_lnd->lnd_startup)(ni);
2241 mutex_unlock(&the_lnet.ln_lnd_mutex);
2244 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2245 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2250 ni->ni_state = LNET_NI_STATE_ACTIVE;
2253 /* We keep a reference on the loopback net through the loopback NI */
2254 if (net->net_lnd->lnd_type == LOLND) {
2256 LASSERT(the_lnet.ln_loni == NULL);
2257 the_lnet.ln_loni = ni;
2258 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2259 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2260 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2261 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2265 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2266 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2267 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2268 libcfs_lnd2str(net->net_lnd->lnd_type),
2269 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2271 /* shutdown the NI since if we get here then it must've already
2274 lnet_shutdown_lndni(ni);
2278 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2279 tq->tq_credits_min =
2280 tq->tq_credits_max =
2281 tq->tq_credits = lnet_ni_tq_credits(ni);
2284 atomic_set(&ni->ni_tx_credits,
2285 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2286 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2288 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2289 libcfs_nid2str(ni->ni_nid),
2290 ni->ni_net->net_tunables.lct_peer_tx_credits,
2291 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2292 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2293 ni->ni_net->net_tunables.lct_peer_timeout);
2302 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2305 struct lnet_net *net_l = NULL;
2306 LIST_HEAD(local_ni_list);
2310 const struct lnet_lnd *lnd;
2312 net->net_tunables.lct_peer_timeout;
2314 net->net_tunables.lct_max_tx_credits;
2315 int peerrtrcredits =
2316 net->net_tunables.lct_peer_rtr_credits;
2319 * make sure that this net is unique. If it isn't then
2320 * we are adding interfaces to an already existing network, and
2321 * 'net' is just a convenient way to pass in the list.
2322 * if it is unique we need to find the LND and load it if
2325 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2326 lnd_type = LNET_NETTYP(net->net_id);
2328 mutex_lock(&the_lnet.ln_lnd_mutex);
2329 lnd = lnet_find_lnd_by_type(lnd_type);
2332 mutex_unlock(&the_lnet.ln_lnd_mutex);
2333 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2334 mutex_lock(&the_lnet.ln_lnd_mutex);
2336 lnd = lnet_find_lnd_by_type(lnd_type);
2338 mutex_unlock(&the_lnet.ln_lnd_mutex);
2339 CERROR("Can't load LND %s, module %s, rc=%d\n",
2340 libcfs_lnd2str(lnd_type),
2341 libcfs_lnd2modname(lnd_type), rc);
2342 #ifndef HAVE_MODULE_LOADING_SUPPORT
2343 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2344 "compiled with kernel module "
2345 "loading support.");
2354 mutex_unlock(&the_lnet.ln_lnd_mutex);
2360 * net_l: if the network being added is unique then net_l
2361 * will point to that network
2362 * if the network being added is not unique then
2363 * net_l points to the existing network.
2365 * When we enter the loop below, we'll pick NIs off he
2366 * network beign added and start them up, then add them to
2367 * a local ni list. Once we've successfully started all
2368 * the NIs then we join the local NI list (of started up
2369 * networks) with the net_l->net_ni_list, which should
2370 * point to the correct network to add the new ni list to
2372 * If any of the new NIs fail to start up, then we want to
2373 * iterate through the local ni list, which should include
2374 * any NIs which were successfully started up, and shut
2377 * After than we want to delete the network being added,
2378 * to avoid a memory leak.
2382 * When a network uses TCP bonding then all its interfaces
2383 * must be specified when the network is first defined: the
2384 * TCP bonding code doesn't allow for interfaces to be added
2387 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2388 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2393 while (!list_empty(&net->net_ni_added)) {
2394 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2396 list_del_init(&ni->ni_netlist);
2398 /* make sure that the the NI we're about to start
2399 * up is actually unique. if it's not fail. */
2400 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2401 ni->ni_interfaces[0])) {
2406 /* adjust the pointer the parent network, just in case it
2407 * the net is a duplicate */
2410 rc = lnet_startup_lndni(ni, tun);
2416 list_add_tail(&ni->ni_netlist, &local_ni_list);
2421 lnet_net_lock(LNET_LOCK_EX);
2422 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2423 lnet_incr_dlc_seq();
2424 lnet_net_unlock(LNET_LOCK_EX);
2426 /* if the network is not unique then we don't want to keep
2427 * it around after we're done. Free it. Otherwise add that
2428 * net to the global the_lnet.ln_nets */
2429 if (net_l != net && net_l != NULL) {
2431 * TODO - note. currently the tunables can not be updated
2437 * restore tunables after it has been overwitten by the
2440 if (peer_timeout != -1)
2441 net->net_tunables.lct_peer_timeout = peer_timeout;
2442 if (maxtxcredits != -1)
2443 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2444 if (peerrtrcredits != -1)
2445 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2447 lnet_net_lock(LNET_LOCK_EX);
2448 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2449 lnet_net_unlock(LNET_LOCK_EX);
2452 /* update net count */
2453 lnet_current_net_count = lnet_get_net_count();
2459 * shutdown the new NIs that are being started up
2460 * free the NET being started
2462 while (!list_empty(&local_ni_list)) {
2463 ni = list_entry(local_ni_list.next, struct lnet_ni,
2466 lnet_shutdown_lndni(ni);
2476 lnet_startup_lndnets(struct list_head *netlist)
2478 struct lnet_net *net;
2483 * Change to running state before bringing up the LNDs. This
2484 * allows lnet_shutdown_lndnets() to assert that we've passed
2487 lnet_net_lock(LNET_LOCK_EX);
2488 the_lnet.ln_state = LNET_STATE_RUNNING;
2489 lnet_net_unlock(LNET_LOCK_EX);
2491 while (!list_empty(netlist)) {
2492 net = list_entry(netlist->next, struct lnet_net, net_list);
2493 list_del_init(&net->net_list);
2495 rc = lnet_startup_lndnet(net, NULL);
2505 lnet_shutdown_lndnets();
2511 * Initialize LNet library.
2513 * Automatically called at module loading time. Caller has to call
2514 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2515 * latter returned 0. It must be called exactly once.
2517 * \retval 0 on success
2518 * \retval -ve on failures.
2520 int lnet_lib_init(void)
2524 lnet_assert_wire_constants();
2526 /* refer to global cfs_cpt_table for now */
2527 the_lnet.ln_cpt_table = cfs_cpt_tab;
2528 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2530 LASSERT(the_lnet.ln_cpt_number > 0);
2531 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2532 /* we are under risk of consuming all lh_cookie */
2533 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2534 "please change setting of CPT-table and retry\n",
2535 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2539 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2540 the_lnet.ln_cpt_bits++;
2542 rc = lnet_create_locks();
2544 CERROR("Can't create LNet global locks: %d\n", rc);
2548 the_lnet.ln_refcount = 0;
2549 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2550 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2552 /* The hash table size is the number of bits it takes to express the set
2553 * ln_num_routes, minus 1 (better to under estimate than over so we
2554 * don't waste memory). */
2555 if (rnet_htable_size <= 0)
2556 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2557 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2558 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2559 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2560 order_base_2(rnet_htable_size) - 1);
2562 /* All LNDs apart from the LOLND are in separate modules. They
2563 * register themselves when their module loads, and unregister
2564 * themselves when their module is unloaded. */
2565 lnet_register_lnd(&the_lolnd);
2570 * Finalize LNet library.
2572 * \pre lnet_lib_init() called with success.
2573 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2575 * As this happens at module-unload, all lnds must already be unloaded,
2576 * so they must already be unregistered.
2578 void lnet_lib_exit(void)
2582 LASSERT(the_lnet.ln_refcount == 0);
2583 lnet_unregister_lnd(&the_lolnd);
2584 for (i = 0; i < NUM_LNDS; i++)
2585 LASSERT(!the_lnet.ln_lnds[i]);
2586 lnet_destroy_locks();
2590 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2592 * Users must call this function at least once before any other functions.
2593 * For each successful call there must be a corresponding call to
2594 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2597 * The PID used by LNet may be different from the one requested.
2600 * \param requested_pid PID requested by the caller.
2602 * \return >= 0 on success, and < 0 error code on failures.
2605 LNetNIInit(lnet_pid_t requested_pid)
2607 int im_a_router = 0;
2610 struct lnet_ping_buffer *pbuf;
2611 struct lnet_handle_md ping_mdh;
2612 LIST_HEAD(net_head);
2613 struct lnet_net *net;
2615 mutex_lock(&the_lnet.ln_api_mutex);
2617 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2619 if (the_lnet.ln_refcount > 0) {
2620 rc = the_lnet.ln_refcount++;
2621 mutex_unlock(&the_lnet.ln_api_mutex);
2625 rc = lnet_prepare(requested_pid);
2627 mutex_unlock(&the_lnet.ln_api_mutex);
2631 /* create a network for Loopback network */
2632 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2635 goto err_empty_list;
2638 /* Add in the loopback NI */
2639 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2641 goto err_empty_list;
2644 /* If LNet is being initialized via DLC it is possible
2645 * that the user requests not to load module parameters (ones which
2646 * are supported by DLC) on initialization. Therefore, make sure not
2647 * to load networks, routes and forwarding from module parameters
2648 * in this case. On cleanup in case of failure only clean up
2649 * routes if it has been loaded */
2650 if (!the_lnet.ln_nis_from_mod_params) {
2651 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2654 goto err_empty_list;
2657 ni_count = lnet_startup_lndnets(&net_head);
2660 goto err_empty_list;
2663 if (!the_lnet.ln_nis_from_mod_params) {
2664 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2666 goto err_shutdown_lndnis;
2668 rc = lnet_rtrpools_alloc(im_a_router);
2670 goto err_destroy_routes;
2673 rc = lnet_acceptor_start();
2675 goto err_destroy_routes;
2677 the_lnet.ln_refcount = 1;
2678 /* Now I may use my own API functions... */
2680 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2682 goto err_acceptor_stop;
2684 lnet_ping_target_update(pbuf, ping_mdh);
2686 the_lnet.ln_mt_eq = LNetEQAlloc(lnet_mt_event_handler);
2687 if (IS_ERR(the_lnet.ln_mt_eq)) {
2688 rc = PTR_ERR(the_lnet.ln_mt_eq);
2689 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2693 rc = lnet_push_target_init();
2697 rc = lnet_peer_discovery_start();
2699 goto err_destroy_push_target;
2701 rc = lnet_monitor_thr_start();
2703 goto err_stop_discovery_thr;
2706 lnet_router_debugfs_init();
2708 mutex_unlock(&the_lnet.ln_api_mutex);
2710 complete_all(&the_lnet.ln_started);
2712 /* wait for all routers to start */
2713 lnet_wait_router_start();
2717 err_stop_discovery_thr:
2718 lnet_peer_discovery_stop();
2719 err_destroy_push_target:
2720 lnet_push_target_fini();
2722 lnet_ping_target_fini();
2724 the_lnet.ln_refcount = 0;
2725 lnet_acceptor_stop();
2727 if (!the_lnet.ln_nis_from_mod_params)
2728 lnet_destroy_routes();
2729 err_shutdown_lndnis:
2730 lnet_shutdown_lndnets();
2734 mutex_unlock(&the_lnet.ln_api_mutex);
2735 while (!list_empty(&net_head)) {
2736 struct lnet_net *net;
2738 net = list_entry(net_head.next, struct lnet_net, net_list);
2739 list_del_init(&net->net_list);
2744 EXPORT_SYMBOL(LNetNIInit);
2747 * Stop LNet interfaces, routing, and forwarding.
2749 * Users must call this function once for each successful call to LNetNIInit().
2750 * Once the LNetNIFini() operation has been started, the results of pending
2751 * API operations are undefined.
2753 * \return always 0 for current implementation.
2758 mutex_lock(&the_lnet.ln_api_mutex);
2760 LASSERT(the_lnet.ln_refcount > 0);
2762 if (the_lnet.ln_refcount != 1) {
2763 the_lnet.ln_refcount--;
2765 LASSERT(!the_lnet.ln_niinit_self);
2769 lnet_router_debugfs_fini();
2770 lnet_monitor_thr_stop();
2771 lnet_peer_discovery_stop();
2772 lnet_push_target_fini();
2773 lnet_ping_target_fini();
2775 /* Teardown fns that use my own API functions BEFORE here */
2776 the_lnet.ln_refcount = 0;
2778 lnet_acceptor_stop();
2779 lnet_destroy_routes();
2780 lnet_shutdown_lndnets();
2784 mutex_unlock(&the_lnet.ln_api_mutex);
2787 EXPORT_SYMBOL(LNetNIFini);
2790 * Grabs the ni data from the ni structure and fills the out
2793 * \param[in] ni network interface structure
2794 * \param[out] cfg_ni NI config information
2795 * \param[out] tun network and LND tunables
2798 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2799 struct lnet_ioctl_config_lnd_tunables *tun,
2800 struct lnet_ioctl_element_stats *stats,
2803 size_t min_size = 0;
2806 if (!ni || !cfg_ni || !tun)
2809 if (ni->ni_interfaces[0] != NULL) {
2810 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2811 if (ni->ni_interfaces[i] != NULL) {
2812 strncpy(cfg_ni->lic_ni_intf[i],
2813 ni->ni_interfaces[i],
2814 sizeof(cfg_ni->lic_ni_intf[i]));
2819 cfg_ni->lic_nid = ni->ni_nid;
2820 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2821 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2823 cfg_ni->lic_status = ni->ni_status->ns_status;
2824 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2825 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2827 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2830 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2831 LNET_STATS_TYPE_SEND);
2832 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2833 LNET_STATS_TYPE_RECV);
2834 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2835 LNET_STATS_TYPE_DROP);
2839 * tun->lt_tun will always be present, but in order to be
2840 * backwards compatible, we need to deal with the cases when
2841 * tun->lt_tun is smaller than what the kernel has, because it
2842 * comes from an older version of a userspace program, then we'll
2843 * need to copy as much information as we have available space.
2845 min_size = tun_size - sizeof(tun->lt_cmn);
2846 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2848 /* copy over the cpts */
2849 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2850 ni->ni_cpts == NULL) {
2851 for (i = 0; i < ni->ni_ncpts; i++)
2852 cfg_ni->lic_cpts[i] = i;
2855 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2856 i < LNET_MAX_SHOW_NUM_CPT;
2858 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2860 cfg_ni->lic_ncpts = ni->ni_ncpts;
2864 * NOTE: This is a legacy function left in the code to be backwards
2865 * compatible with older userspace programs. It should eventually be
2868 * Grabs the ni data from the ni structure and fills the out
2871 * \param[in] ni network interface structure
2872 * \param[out] config config information
2875 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2876 struct lnet_ioctl_config_data *config)
2878 struct lnet_ioctl_net_config *net_config;
2879 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2880 size_t min_size, tunable_size = 0;
2886 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2890 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2891 ARRAY_SIZE(net_config->ni_interfaces));
2893 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2894 if (!ni->ni_interfaces[i])
2897 strncpy(net_config->ni_interfaces[i],
2898 ni->ni_interfaces[i],
2899 sizeof(net_config->ni_interfaces[i]));
2902 config->cfg_nid = ni->ni_nid;
2903 config->cfg_config_u.cfg_net.net_peer_timeout =
2904 ni->ni_net->net_tunables.lct_peer_timeout;
2905 config->cfg_config_u.cfg_net.net_max_tx_credits =
2906 ni->ni_net->net_tunables.lct_max_tx_credits;
2907 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2908 ni->ni_net->net_tunables.lct_peer_tx_credits;
2909 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2910 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2912 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2913 net_config->ni_status = LNET_NI_STATUS_UP;
2915 net_config->ni_status = ni->ni_status->ns_status;
2918 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2920 for (i = 0; i < num_cpts; i++)
2921 net_config->ni_cpts[i] = ni->ni_cpts[i];
2923 config->cfg_ncpts = num_cpts;
2927 * See if user land tools sent in a newer and larger version
2928 * of struct lnet_tunables than what the kernel uses.
2930 min_size = sizeof(*config) + sizeof(*net_config);
2932 if (config->cfg_hdr.ioc_len > min_size)
2933 tunable_size = config->cfg_hdr.ioc_len - min_size;
2935 /* Don't copy too much data to user space */
2936 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2937 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2939 if (lnd_cfg && min_size) {
2940 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2941 config->cfg_config_u.cfg_net.net_interface_count = 1;
2943 /* Tell user land that kernel side has less data */
2944 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2945 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2946 config->cfg_hdr.ioc_len -= min_size;
2952 lnet_get_ni_idx_locked(int idx)
2955 struct lnet_net *net;
2957 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2958 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2968 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2971 struct lnet_net *net = mynet;
2974 * It is possible that the net has been cleaned out while there is
2975 * a message being sent. This function accessed the net without
2976 * checking if the list is empty
2980 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2982 if (list_empty(&net->net_ni_list))
2984 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2990 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2991 /* if you reached the end of the ni list and the net is
2992 * specified, then there are no more nis in that net */
2996 /* we reached the end of this net ni list. move to the
2998 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2999 /* no more nets and no more NIs. */
3002 /* get the next net */
3003 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3005 if (list_empty(&net->net_ni_list))
3007 /* get the ni on it */
3008 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3014 if (list_empty(&prev->ni_netlist))
3017 /* there are more nis left */
3018 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3024 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3029 int idx = config->cfg_count;
3031 cpt = lnet_net_lock_current();
3033 ni = lnet_get_ni_idx_locked(idx);
3038 lnet_fill_ni_info_legacy(ni, config);
3042 lnet_net_unlock(cpt);
3047 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3048 struct lnet_ioctl_config_lnd_tunables *tun,
3049 struct lnet_ioctl_element_stats *stats,
3056 if (!cfg_ni || !tun || !stats)
3059 cpt = lnet_net_lock_current();
3061 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3066 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3070 lnet_net_unlock(cpt);
3074 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3083 cpt = lnet_net_lock_current();
3085 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3088 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3092 lnet_net_unlock(cpt);
3097 static int lnet_add_net_common(struct lnet_net *net,
3098 struct lnet_ioctl_config_lnd_tunables *tun)
3101 struct lnet_ping_buffer *pbuf;
3102 struct lnet_handle_md ping_mdh;
3104 struct lnet_remotenet *rnet;
3106 int num_acceptor_nets;
3108 lnet_net_lock(LNET_LOCK_EX);
3109 rnet = lnet_find_rnet_locked(net->net_id);
3110 lnet_net_unlock(LNET_LOCK_EX);
3112 * make sure that the net added doesn't invalidate the current
3113 * configuration LNet is keeping
3116 CERROR("Adding net %s will invalidate routing configuration\n",
3117 libcfs_net2str(net->net_id));
3123 * make sure you calculate the correct number of slots in the ping
3124 * buffer. Since the ping info is a flattened list of all the NIs,
3125 * we should allocate enough slots to accomodate the number of NIs
3126 * which will be added.
3128 * since ni hasn't been configured yet, use
3129 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3131 net_ni_count = lnet_get_net_ni_count_pre(net);
3133 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3134 net_ni_count + lnet_get_ni_count(),
3142 memcpy(&net->net_tunables,
3143 &tun->lt_cmn, sizeof(net->net_tunables));
3145 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3148 * before starting this network get a count of the current TCP
3149 * networks which require the acceptor thread running. If that
3150 * count is == 0 before we start up this network, then we'd want to
3151 * start up the acceptor thread after starting up this network
3153 num_acceptor_nets = lnet_count_acceptor_nets();
3155 net_id = net->net_id;
3157 rc = lnet_startup_lndnet(net,
3158 (tun) ? &tun->lt_tun : NULL);
3162 lnet_net_lock(LNET_LOCK_EX);
3163 net = lnet_get_net_locked(net_id);
3164 lnet_net_unlock(LNET_LOCK_EX);
3169 * Start the acceptor thread if this is the first network
3170 * being added that requires the thread.
3172 if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3173 rc = lnet_acceptor_start();
3175 /* shutdown the net that we just started */
3176 CERROR("Failed to start up acceptor thread\n");
3177 lnet_shutdown_lndnet(net);
3182 lnet_net_lock(LNET_LOCK_EX);
3183 lnet_peer_net_added(net);
3184 lnet_net_unlock(LNET_LOCK_EX);
3186 lnet_ping_target_update(pbuf, ping_mdh);
3191 lnet_ping_md_unlink(pbuf, &ping_mdh);
3192 lnet_ping_buffer_decref(pbuf);
3196 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3197 struct lnet_ioctl_config_lnd_tunables *tun)
3199 struct lnet_net *net;
3202 LIST_HEAD(net_head);
3204 rc = lnet_parse_ip2nets(&nets, ip2nets);
3208 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3212 mutex_lock(&the_lnet.ln_api_mutex);
3213 while (!list_empty(&net_head)) {
3214 net = list_entry(net_head.next, struct lnet_net, net_list);
3215 list_del_init(&net->net_list);
3216 rc = lnet_add_net_common(net, tun);
3222 mutex_unlock(&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);
3232 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3234 struct lnet_net *net;
3236 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3238 __u32 net_id, lnd_type;
3240 /* get the tunables if they are available */
3241 if (conf->lic_cfg_hdr.ioc_len >=
3242 sizeof(*conf) + sizeof(*tun))
3243 tun = (struct lnet_ioctl_config_lnd_tunables *)
3246 /* handle legacy ip2nets from DLC */
3247 if (conf->lic_legacy_ip2nets[0] != '\0')
3248 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3251 net_id = LNET_NIDNET(conf->lic_nid);
3252 lnd_type = LNET_NETTYP(net_id);
3254 if (!libcfs_isknown_lnd(lnd_type)) {
3255 CERROR("No valid net and lnd information provided\n");
3259 net = lnet_net_alloc(net_id, NULL);
3263 for (i = 0; i < conf->lic_ncpts; i++) {
3264 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3268 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3269 conf->lic_ni_intf[0]);
3273 mutex_lock(&the_lnet.ln_api_mutex);
3275 rc = lnet_add_net_common(net, tun);
3277 mutex_unlock(&the_lnet.ln_api_mutex);
3282 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3284 struct lnet_net *net;
3286 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3287 struct lnet_ping_buffer *pbuf;
3288 struct lnet_handle_md ping_mdh;
3293 /* don't allow userspace to shutdown the LOLND */
3294 if (LNET_NETTYP(net_id) == LOLND)
3297 mutex_lock(&the_lnet.ln_api_mutex);
3301 net = lnet_get_net_locked(net_id);
3303 CERROR("net %s not found\n",
3304 libcfs_net2str(net_id));
3309 addr = LNET_NIDADDR(conf->lic_nid);
3311 /* remove the entire net */
3312 net_count = lnet_get_net_ni_count_locked(net);
3316 /* create and link a new ping info, before removing the old one */
3317 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3318 lnet_get_ni_count() - net_count,
3321 goto unlock_api_mutex;
3323 lnet_shutdown_lndnet(net);
3325 if (lnet_count_acceptor_nets() == 0)
3326 lnet_acceptor_stop();
3328 lnet_ping_target_update(pbuf, ping_mdh);
3330 goto unlock_api_mutex;
3333 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3335 CERROR("nid %s not found\n",
3336 libcfs_nid2str(conf->lic_nid));
3341 net_count = lnet_get_net_ni_count_locked(net);
3345 /* create and link a new ping info, before removing the old one */
3346 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3347 lnet_get_ni_count() - 1, false);
3349 goto unlock_api_mutex;
3351 lnet_shutdown_lndni(ni);
3353 if (lnet_count_acceptor_nets() == 0)
3354 lnet_acceptor_stop();
3356 lnet_ping_target_update(pbuf, ping_mdh);
3358 /* check if the net is empty and remove it if it is */
3360 lnet_shutdown_lndnet(net);
3362 goto unlock_api_mutex;
3367 mutex_unlock(&the_lnet.ln_api_mutex);
3373 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3374 * They are only expected to be called for unique networks.
3375 * That can be as a result of older DLC library
3376 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3379 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3381 struct lnet_net *net;
3382 LIST_HEAD(net_head);
3384 struct lnet_ioctl_config_lnd_tunables tun;
3385 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3387 /* Create a net/ni structures for the network string */
3388 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3390 return rc == 0 ? -EINVAL : rc;
3392 mutex_lock(&the_lnet.ln_api_mutex);
3395 rc = -EINVAL; /* only add one network per call */
3396 goto out_unlock_clean;
3399 net = list_entry(net_head.next, struct lnet_net, net_list);
3400 list_del_init(&net->net_list);
3402 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3404 memset(&tun, 0, sizeof(tun));
3406 tun.lt_cmn.lct_peer_timeout =
3407 conf->cfg_config_u.cfg_net.net_peer_timeout;
3408 tun.lt_cmn.lct_peer_tx_credits =
3409 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3410 tun.lt_cmn.lct_peer_rtr_credits =
3411 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3412 tun.lt_cmn.lct_max_tx_credits =
3413 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3415 rc = lnet_add_net_common(net, &tun);
3418 mutex_unlock(&the_lnet.ln_api_mutex);
3419 while (!list_empty(&net_head)) {
3420 /* net_head list is empty in success case */
3421 net = list_entry(net_head.next, struct lnet_net, net_list);
3422 list_del_init(&net->net_list);
3429 lnet_dyn_del_net(__u32 net_id)
3431 struct lnet_net *net;
3432 struct lnet_ping_buffer *pbuf;
3433 struct lnet_handle_md ping_mdh;
3437 /* don't allow userspace to shutdown the LOLND */
3438 if (LNET_NETTYP(net_id) == LOLND)
3441 mutex_lock(&the_lnet.ln_api_mutex);
3445 net = lnet_get_net_locked(net_id);
3452 net_ni_count = lnet_get_net_ni_count_locked(net);
3456 /* create and link a new ping info, before removing the old one */
3457 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3458 lnet_get_ni_count() - net_ni_count, false);
3462 lnet_shutdown_lndnet(net);
3464 if (lnet_count_acceptor_nets() == 0)
3465 lnet_acceptor_stop();
3467 lnet_ping_target_update(pbuf, ping_mdh);
3470 mutex_unlock(&the_lnet.ln_api_mutex);
3475 void lnet_incr_dlc_seq(void)
3477 atomic_inc(&lnet_dlc_seq_no);
3480 __u32 lnet_get_dlc_seq_locked(void)
3482 return atomic_read(&lnet_dlc_seq_no);
3486 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3488 struct lnet_net *net;
3491 lnet_net_lock(LNET_LOCK_EX);
3492 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3493 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3494 if (ni->ni_nid == nid || all) {
3495 atomic_set(&ni->ni_healthv, value);
3496 if (list_empty(&ni->ni_recovery) &&
3497 value < LNET_MAX_HEALTH_VALUE) {
3498 CERROR("manually adding local NI %s to recovery\n",
3499 libcfs_nid2str(ni->ni_nid));
3500 list_add_tail(&ni->ni_recovery,
3501 &the_lnet.ln_mt_localNIRecovq);
3502 lnet_ni_addref_locked(ni, 0);
3505 lnet_net_unlock(LNET_LOCK_EX);
3511 lnet_net_unlock(LNET_LOCK_EX);
3515 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3519 lnet_nid_t nid = stats->hlni_nid;
3521 cpt = lnet_net_lock_current();
3522 ni = lnet_nid2ni_locked(nid, cpt);
3529 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3530 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3531 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3532 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3533 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3534 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3535 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3538 lnet_net_unlock(cpt);
3544 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3549 lnet_net_lock(LNET_LOCK_EX);
3550 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3551 list->rlst_nid_array[i] = ni->ni_nid;
3553 if (i >= LNET_MAX_SHOW_NUM_NID)
3556 lnet_net_unlock(LNET_LOCK_EX);
3557 list->rlst_num_nids = i;
3563 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3565 struct lnet_peer_ni *lpni;
3568 lnet_net_lock(LNET_LOCK_EX);
3569 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3570 list->rlst_nid_array[i] = lpni->lpni_nid;
3572 if (i >= LNET_MAX_SHOW_NUM_NID)
3575 lnet_net_unlock(LNET_LOCK_EX);
3576 list->rlst_num_nids = i;
3582 * LNet ioctl handler.
3586 LNetCtl(unsigned int cmd, void *arg)
3588 struct libcfs_ioctl_data *data = arg;
3589 struct lnet_ioctl_config_data *config;
3590 struct lnet_process_id id = {0};
3594 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3595 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3598 case IOC_LIBCFS_GET_NI:
3599 rc = LNetGetId(data->ioc_count, &id);
3600 data->ioc_nid = id.nid;
3603 case IOC_LIBCFS_FAIL_NID:
3604 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3606 case IOC_LIBCFS_ADD_ROUTE: {
3607 /* default router sensitivity to 1 */
3608 unsigned int sensitivity = 1;
3611 if (config->cfg_hdr.ioc_len < sizeof(*config))
3614 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3616 config->cfg_config_u.cfg_route.rtr_sensitivity;
3619 mutex_lock(&the_lnet.ln_api_mutex);
3620 rc = lnet_add_route(config->cfg_net,
3621 config->cfg_config_u.cfg_route.rtr_hop,
3623 config->cfg_config_u.cfg_route.
3624 rtr_priority, sensitivity);
3625 mutex_unlock(&the_lnet.ln_api_mutex);
3629 case IOC_LIBCFS_DEL_ROUTE:
3632 if (config->cfg_hdr.ioc_len < sizeof(*config))
3635 mutex_lock(&the_lnet.ln_api_mutex);
3636 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3637 mutex_unlock(&the_lnet.ln_api_mutex);
3640 case IOC_LIBCFS_GET_ROUTE:
3643 if (config->cfg_hdr.ioc_len < sizeof(*config))
3646 mutex_lock(&the_lnet.ln_api_mutex);
3647 rc = lnet_get_route(config->cfg_count,
3649 &config->cfg_config_u.cfg_route.rtr_hop,
3651 &config->cfg_config_u.cfg_route.rtr_flags,
3652 &config->cfg_config_u.cfg_route.
3654 &config->cfg_config_u.cfg_route.
3656 mutex_unlock(&the_lnet.ln_api_mutex);
3659 case IOC_LIBCFS_GET_LOCAL_NI: {
3660 struct lnet_ioctl_config_ni *cfg_ni;
3661 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3662 struct lnet_ioctl_element_stats *stats;
3667 /* get the tunables if they are available */
3668 if (cfg_ni->lic_cfg_hdr.ioc_len <
3669 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3672 stats = (struct lnet_ioctl_element_stats *)
3674 tun = (struct lnet_ioctl_config_lnd_tunables *)
3675 (cfg_ni->lic_bulk + sizeof(*stats));
3677 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3680 mutex_lock(&the_lnet.ln_api_mutex);
3681 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3682 mutex_unlock(&the_lnet.ln_api_mutex);
3686 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3687 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3689 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3692 mutex_lock(&the_lnet.ln_api_mutex);
3693 rc = lnet_get_ni_stats(msg_stats);
3694 mutex_unlock(&the_lnet.ln_api_mutex);
3699 case IOC_LIBCFS_GET_NET: {
3700 size_t total = sizeof(*config) +
3701 sizeof(struct lnet_ioctl_net_config);
3704 if (config->cfg_hdr.ioc_len < total)
3707 mutex_lock(&the_lnet.ln_api_mutex);
3708 rc = lnet_get_net_config(config);
3709 mutex_unlock(&the_lnet.ln_api_mutex);
3713 case IOC_LIBCFS_GET_LNET_STATS:
3715 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3717 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3720 mutex_lock(&the_lnet.ln_api_mutex);
3721 lnet_counters_get(&lnet_stats->st_cntrs);
3722 mutex_unlock(&the_lnet.ln_api_mutex);
3726 case IOC_LIBCFS_CONFIG_RTR:
3729 if (config->cfg_hdr.ioc_len < sizeof(*config))
3732 mutex_lock(&the_lnet.ln_api_mutex);
3733 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3734 rc = lnet_rtrpools_enable();
3735 mutex_unlock(&the_lnet.ln_api_mutex);
3738 lnet_rtrpools_disable();
3739 mutex_unlock(&the_lnet.ln_api_mutex);
3742 case IOC_LIBCFS_ADD_BUF:
3745 if (config->cfg_hdr.ioc_len < sizeof(*config))
3748 mutex_lock(&the_lnet.ln_api_mutex);
3749 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3751 config->cfg_config_u.cfg_buffers.
3753 config->cfg_config_u.cfg_buffers.
3755 mutex_unlock(&the_lnet.ln_api_mutex);
3758 case IOC_LIBCFS_SET_NUMA_RANGE: {
3759 struct lnet_ioctl_set_value *numa;
3761 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3763 lnet_net_lock(LNET_LOCK_EX);
3764 lnet_numa_range = numa->sv_value;
3765 lnet_net_unlock(LNET_LOCK_EX);
3769 case IOC_LIBCFS_GET_NUMA_RANGE: {
3770 struct lnet_ioctl_set_value *numa;
3772 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3774 numa->sv_value = lnet_numa_range;
3778 case IOC_LIBCFS_GET_BUF: {
3779 struct lnet_ioctl_pool_cfg *pool_cfg;
3780 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3784 if (config->cfg_hdr.ioc_len < total)
3787 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3789 mutex_lock(&the_lnet.ln_api_mutex);
3790 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3791 mutex_unlock(&the_lnet.ln_api_mutex);
3795 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3796 struct lnet_ioctl_local_ni_hstats *stats = arg;
3798 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3801 mutex_lock(&the_lnet.ln_api_mutex);
3802 rc = lnet_get_local_ni_hstats(stats);
3803 mutex_unlock(&the_lnet.ln_api_mutex);
3808 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3809 struct lnet_ioctl_recovery_list *list = arg;
3810 if (list->rlst_hdr.ioc_len < sizeof(*list))
3813 mutex_lock(&the_lnet.ln_api_mutex);
3814 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3815 rc = lnet_get_local_ni_recovery_list(list);
3817 rc = lnet_get_peer_ni_recovery_list(list);
3818 mutex_unlock(&the_lnet.ln_api_mutex);
3822 case IOC_LIBCFS_ADD_PEER_NI: {
3823 struct lnet_ioctl_peer_cfg *cfg = arg;
3825 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3828 mutex_lock(&the_lnet.ln_api_mutex);
3829 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3832 mutex_unlock(&the_lnet.ln_api_mutex);
3836 case IOC_LIBCFS_DEL_PEER_NI: {
3837 struct lnet_ioctl_peer_cfg *cfg = arg;
3839 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3842 mutex_lock(&the_lnet.ln_api_mutex);
3843 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3844 cfg->prcfg_cfg_nid);
3845 mutex_unlock(&the_lnet.ln_api_mutex);
3849 case IOC_LIBCFS_GET_PEER_INFO: {
3850 struct lnet_ioctl_peer *peer_info = arg;
3852 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3855 mutex_lock(&the_lnet.ln_api_mutex);
3856 rc = lnet_get_peer_ni_info(
3857 peer_info->pr_count,
3859 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3860 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3861 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3862 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3863 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3864 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3865 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3866 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3867 mutex_unlock(&the_lnet.ln_api_mutex);
3871 case IOC_LIBCFS_GET_PEER_NI: {
3872 struct lnet_ioctl_peer_cfg *cfg = arg;
3874 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3877 mutex_lock(&the_lnet.ln_api_mutex);
3878 rc = lnet_get_peer_info(cfg,
3879 (void __user *)cfg->prcfg_bulk);
3880 mutex_unlock(&the_lnet.ln_api_mutex);
3884 case IOC_LIBCFS_GET_PEER_LIST: {
3885 struct lnet_ioctl_peer_cfg *cfg = arg;
3887 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3890 mutex_lock(&the_lnet.ln_api_mutex);
3891 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3892 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3893 mutex_unlock(&the_lnet.ln_api_mutex);
3897 case IOC_LIBCFS_SET_HEALHV: {
3898 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3900 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3902 if (cfg->rh_value < 0 ||
3903 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3904 value = LNET_MAX_HEALTH_VALUE;
3906 value = cfg->rh_value;
3907 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3908 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3909 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3910 mutex_lock(&the_lnet.ln_api_mutex);
3911 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3912 lnet_ni_set_healthv(cfg->rh_nid, value,
3915 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3917 mutex_unlock(&the_lnet.ln_api_mutex);
3921 case IOC_LIBCFS_NOTIFY_ROUTER: {
3922 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3924 /* The deadline passed in by the user should be some time in
3925 * seconds in the future since the UNIX epoch. We have to map
3926 * that deadline to the wall clock.
3928 deadline += ktime_get_seconds();
3929 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3933 case IOC_LIBCFS_LNET_DIST:
3934 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3935 if (rc < 0 && rc != -EHOSTUNREACH)
3938 data->ioc_u32[0] = rc;
3941 case IOC_LIBCFS_TESTPROTOCOMPAT:
3942 the_lnet.ln_testprotocompat = data->ioc_flags;
3945 case IOC_LIBCFS_LNET_FAULT:
3946 return lnet_fault_ctl(data->ioc_flags, data);
3948 case IOC_LIBCFS_PING: {
3949 signed long timeout;
3951 id.nid = data->ioc_nid;
3952 id.pid = data->ioc_u32[0];
3954 /* If timeout is negative then set default of 3 minutes */
3955 if (((s32)data->ioc_u32[1] <= 0) ||
3956 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3957 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3959 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3961 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3962 data->ioc_plen1 / sizeof(struct lnet_process_id));
3967 data->ioc_count = rc;
3971 case IOC_LIBCFS_PING_PEER: {
3972 struct lnet_ioctl_ping_data *ping = arg;
3973 struct lnet_peer *lp;
3974 signed long timeout;
3976 /* If timeout is negative then set default of 3 minutes */
3977 if (((s32)ping->op_param) <= 0 ||
3978 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3979 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3981 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3983 rc = lnet_ping(ping->ping_id, timeout,
3989 mutex_lock(&the_lnet.ln_api_mutex);
3990 lp = lnet_find_peer(ping->ping_id.nid);
3992 ping->ping_id.nid = lp->lp_primary_nid;
3993 ping->mr_info = lnet_peer_is_multi_rail(lp);
3994 lnet_peer_decref_locked(lp);
3996 mutex_unlock(&the_lnet.ln_api_mutex);
3998 ping->ping_count = rc;
4002 case IOC_LIBCFS_DISCOVER: {
4003 struct lnet_ioctl_ping_data *discover = arg;
4004 struct lnet_peer *lp;
4006 rc = lnet_discover(discover->ping_id, discover->op_param,
4008 discover->ping_count);
4012 mutex_lock(&the_lnet.ln_api_mutex);
4013 lp = lnet_find_peer(discover->ping_id.nid);
4015 discover->ping_id.nid = lp->lp_primary_nid;
4016 discover->mr_info = lnet_peer_is_multi_rail(lp);
4017 lnet_peer_decref_locked(lp);
4019 mutex_unlock(&the_lnet.ln_api_mutex);
4021 discover->ping_count = rc;
4026 ni = lnet_net2ni_addref(data->ioc_net);
4030 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4033 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4040 EXPORT_SYMBOL(LNetCtl);
4042 void LNetDebugPeer(struct lnet_process_id id)
4044 lnet_debug_peer(id.nid);
4046 EXPORT_SYMBOL(LNetDebugPeer);
4049 * Determine if the specified peer \a nid is on the local node.
4051 * \param nid peer nid to check
4053 * \retval true If peer NID is on the local node.
4054 * \retval false If peer NID is not on the local node.
4056 bool LNetIsPeerLocal(lnet_nid_t nid)
4058 struct lnet_net *net;
4062 cpt = lnet_net_lock_current();
4063 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4064 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4065 if (ni->ni_nid == nid) {
4066 lnet_net_unlock(cpt);
4071 lnet_net_unlock(cpt);
4075 EXPORT_SYMBOL(LNetIsPeerLocal);
4078 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4079 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4081 * \param index Index of the interface to look up.
4082 * \param id On successful return, this location will hold the
4083 * struct lnet_process_id ID of the interface.
4085 * \retval 0 If an interface exists at \a index.
4086 * \retval -ENOENT If no interface has been found.
4089 LNetGetId(unsigned int index, struct lnet_process_id *id)
4092 struct lnet_net *net;
4096 LASSERT(the_lnet.ln_refcount > 0);
4098 cpt = lnet_net_lock_current();
4100 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4101 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4105 id->nid = ni->ni_nid;
4106 id->pid = the_lnet.ln_pid;
4112 lnet_net_unlock(cpt);
4115 EXPORT_SYMBOL(LNetGetId);
4120 struct lnet_handle_md mdh;
4121 struct completion completion;
4125 lnet_ping_event_handler(struct lnet_event *event)
4127 struct ping_data *pd = event->md.user_ptr;
4129 CDEBUG(D_NET, "ping event (%d %d)%s\n",
4130 event->type, event->status,
4131 event->unlinked ? " unlinked" : "");
4133 if (event->status) {
4135 pd->rc = event->status;
4136 } else if (event->type == LNET_EVENT_REPLY) {
4138 pd->rc = event->mlength;
4140 if (event->unlinked)
4141 complete(&pd->completion);
4144 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4145 struct lnet_process_id __user *ids, int n_ids)
4148 struct lnet_md md = { NULL };
4149 struct ping_data pd = { 0 };
4150 struct lnet_ping_buffer *pbuf;
4151 struct lnet_process_id tmpid;
4157 /* n_ids limit is arbitrary */
4158 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4162 * if the user buffer has more space than the lnet_interfaces_max
4163 * then only fill it up to lnet_interfaces_max
4165 if (n_ids > lnet_interfaces_max)
4166 n_ids = lnet_interfaces_max;
4168 if (id.pid == LNET_PID_ANY)
4169 id.pid = LNET_PID_LUSTRE;
4171 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4175 eq = LNetEQAlloc(lnet_ping_event_handler);
4178 CERROR("Can't allocate EQ: %d\n", rc);
4179 goto fail_ping_buffer_decref;
4182 /* initialize md content */
4183 md.start = &pbuf->pb_info;
4184 md.length = LNET_PING_INFO_SIZE(n_ids);
4185 md.threshold = 2; /* GET/REPLY */
4187 md.options = LNET_MD_TRUNCATE;
4191 init_completion(&pd.completion);
4193 rc = LNetMDBind(md, LNET_UNLINK, &pd.mdh);
4195 CERROR("Can't bind MD: %d\n", rc);
4199 rc = LNetGet(LNET_NID_ANY, pd.mdh, id,
4200 LNET_RESERVED_PORTAL,
4201 LNET_PROTO_PING_MATCHBITS, 0, false);
4204 /* Don't CERROR; this could be deliberate! */
4205 rc2 = LNetMDUnlink(pd.mdh);
4208 /* NB must wait for the UNLINK event below... */
4211 if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4212 /* Ensure completion in finite time... */
4213 LNetMDUnlink(pd.mdh);
4214 wait_for_completion(&pd.completion);
4222 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4224 rc = -EPROTO; /* if I can't parse... */
4227 CERROR("%s: ping info too short %d\n",
4228 libcfs_id2str(id), nob);
4232 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4233 lnet_swap_pinginfo(pbuf);
4234 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4235 CERROR("%s: Unexpected magic %08x\n",
4236 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4240 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4241 CERROR("%s: ping w/o NI status: 0x%x\n",
4242 libcfs_id2str(id), pbuf->pb_info.pi_features);
4246 if (nob < LNET_PING_INFO_SIZE(0)) {
4247 CERROR("%s: Short reply %d(%d min)\n",
4249 nob, (int)LNET_PING_INFO_SIZE(0));
4253 if (pbuf->pb_info.pi_nnis < n_ids)
4254 n_ids = pbuf->pb_info.pi_nnis;
4256 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4257 CERROR("%s: Short reply %d(%d expected)\n",
4259 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4263 rc = -EFAULT; /* if I segv in copy_to_user()... */
4265 memset(&tmpid, 0, sizeof(tmpid));
4266 for (i = 0; i < n_ids; i++) {
4267 tmpid.pid = pbuf->pb_info.pi_pid;
4268 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4269 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4272 rc = pbuf->pb_info.pi_nnis;
4275 rc2 = LNetEQFree(eq);
4277 CERROR("rc2 %d\n", rc2);
4280 fail_ping_buffer_decref:
4281 lnet_ping_buffer_decref(pbuf);
4286 lnet_discover(struct lnet_process_id id, __u32 force,
4287 struct lnet_process_id __user *ids, int n_ids)
4289 struct lnet_peer_ni *lpni;
4290 struct lnet_peer_ni *p;
4291 struct lnet_peer *lp;
4292 struct lnet_process_id *buf;
4296 int max_intf = lnet_interfaces_max;
4299 id.nid == LNET_NID_ANY)
4302 if (id.pid == LNET_PID_ANY)
4303 id.pid = LNET_PID_LUSTRE;
4306 * if the user buffer has more space than the max_intf
4307 * then only fill it up to max_intf
4309 if (n_ids > max_intf)
4312 CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4316 cpt = lnet_net_lock_current();
4317 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4324 * Clearing the NIDS_UPTODATE flag ensures the peer will
4325 * be discovered, provided discovery has not been disabled.
4327 lp = lpni->lpni_peer_net->lpn_peer;
4328 spin_lock(&lp->lp_lock);
4329 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4330 /* If the force flag is set, force a PING and PUSH as well. */
4332 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4333 spin_unlock(&lp->lp_lock);
4334 rc = lnet_discover_peer_locked(lpni, cpt, true);
4338 /* Peer may have changed. */
4339 lp = lpni->lpni_peer_net->lpn_peer;
4340 if (lp->lp_nnis < n_ids)
4341 n_ids = lp->lp_nnis;
4345 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4346 buf[i].pid = id.pid;
4347 buf[i].nid = p->lpni_nid;
4352 lnet_net_unlock(cpt);
4355 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4361 lnet_peer_ni_decref_locked(lpni);
4363 lnet_net_unlock(cpt);
4365 CFS_FREE_PTR_ARRAY(buf, n_ids);
4371 * Retrieve peer discovery status.
4373 * \retval 1 if lnet_peer_discovery_disabled is 0
4374 * \retval 0 if lnet_peer_discovery_disabled is 1
4377 LNetGetPeerDiscoveryStatus(void)
4379 return !lnet_peer_discovery_disabled;
4381 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);