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_waitqueue_head(&the_lnet.ln_eq_waitq);
552 init_completion(&the_lnet.ln_mt_wait_complete);
553 mutex_init(&the_lnet.ln_lnd_mutex);
556 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
557 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
559 struct kmem_cache *lnet_rspt_cachep; /* response tracker cache */
560 struct kmem_cache *lnet_msg_cachep;
563 lnet_slab_setup(void)
565 /* create specific kmem_cache for MEs and small MDs (i.e., originally
566 * allocated in <size-xxx> kmem_cache).
568 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
570 if (!lnet_mes_cachep)
573 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
574 LNET_SMALL_MD_SIZE, 0, 0,
576 if (!lnet_small_mds_cachep)
579 lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
581 if (!lnet_rspt_cachep)
584 lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
586 if (!lnet_msg_cachep)
593 lnet_slab_cleanup(void)
595 if (lnet_msg_cachep) {
596 kmem_cache_destroy(lnet_msg_cachep);
597 lnet_msg_cachep = NULL;
601 if (lnet_rspt_cachep) {
602 kmem_cache_destroy(lnet_rspt_cachep);
603 lnet_rspt_cachep = NULL;
606 if (lnet_small_mds_cachep) {
607 kmem_cache_destroy(lnet_small_mds_cachep);
608 lnet_small_mds_cachep = NULL;
611 if (lnet_mes_cachep) {
612 kmem_cache_destroy(lnet_mes_cachep);
613 lnet_mes_cachep = NULL;
618 lnet_create_remote_nets_table(void)
621 struct list_head *hash;
623 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
624 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
625 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
627 CERROR("Failed to create remote nets hash table\n");
631 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
632 INIT_LIST_HEAD(&hash[i]);
633 the_lnet.ln_remote_nets_hash = hash;
638 lnet_destroy_remote_nets_table(void)
642 if (the_lnet.ln_remote_nets_hash == NULL)
645 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
646 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
648 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
649 LNET_REMOTE_NETS_HASH_SIZE *
650 sizeof(the_lnet.ln_remote_nets_hash[0]));
651 the_lnet.ln_remote_nets_hash = NULL;
655 lnet_destroy_locks(void)
657 if (the_lnet.ln_res_lock != NULL) {
658 cfs_percpt_lock_free(the_lnet.ln_res_lock);
659 the_lnet.ln_res_lock = NULL;
662 if (the_lnet.ln_net_lock != NULL) {
663 cfs_percpt_lock_free(the_lnet.ln_net_lock);
664 the_lnet.ln_net_lock = NULL;
669 lnet_create_locks(void)
673 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
674 if (the_lnet.ln_res_lock == NULL)
677 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
678 if (the_lnet.ln_net_lock == NULL)
684 lnet_destroy_locks();
688 static void lnet_assert_wire_constants(void)
690 /* Wire protocol assertions generated by 'wirecheck'
691 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
692 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
693 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
696 BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
697 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
698 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
699 BUILD_BUG_ON(LNET_MSG_ACK != 0);
700 BUILD_BUG_ON(LNET_MSG_PUT != 1);
701 BUILD_BUG_ON(LNET_MSG_GET != 2);
702 BUILD_BUG_ON(LNET_MSG_REPLY != 3);
703 BUILD_BUG_ON(LNET_MSG_HELLO != 4);
705 /* Checks for struct lnet_handle_wire */
706 BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
707 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
708 wh_interface_cookie) != 0);
709 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
710 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
711 wh_object_cookie) != 8);
712 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
714 /* Checks for struct struct lnet_magicversion */
715 BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
716 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
717 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
718 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
719 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
720 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
721 version_minor) != 6);
722 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
724 /* Checks for struct struct lnet_hdr */
725 BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
726 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
727 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
728 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
729 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
730 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
731 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
732 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
733 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
734 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
735 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
736 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
737 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
738 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
739 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
742 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
743 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
744 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
745 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
746 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
747 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
750 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
751 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
752 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
753 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
754 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
755 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
756 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
757 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
758 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
759 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
762 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
763 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
764 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
765 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
766 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
767 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
768 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
769 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
770 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
771 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
774 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
775 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
778 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
779 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
780 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
781 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
783 /* Checks for struct lnet_ni_status and related constants */
784 BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
785 BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
786 BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
788 /* Checks for struct lnet_ni_status */
789 BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
790 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
791 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
792 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
793 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
794 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
795 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
797 /* Checks for struct lnet_ping_info and related constants */
798 BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
799 BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
800 BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
801 BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
802 BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
803 BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
804 BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
805 BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
807 /* Checks for struct lnet_ping_info */
808 BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
809 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
810 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
811 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
812 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
813 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
814 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
815 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
816 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
817 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
818 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
821 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
823 const struct lnet_lnd *lnd;
825 /* holding lnd mutex */
826 if (type >= NUM_LNDS)
828 lnd = the_lnet.ln_lnds[type];
829 LASSERT(!lnd || lnd->lnd_type == type);
835 lnet_get_lnd_timeout(void)
837 return lnet_lnd_timeout;
839 EXPORT_SYMBOL(lnet_get_lnd_timeout);
842 lnet_register_lnd(const struct lnet_lnd *lnd)
844 mutex_lock(&the_lnet.ln_lnd_mutex);
846 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
847 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
849 the_lnet.ln_lnds[lnd->lnd_type] = lnd;
851 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
853 mutex_unlock(&the_lnet.ln_lnd_mutex);
855 EXPORT_SYMBOL(lnet_register_lnd);
858 lnet_unregister_lnd(const struct lnet_lnd *lnd)
860 mutex_lock(&the_lnet.ln_lnd_mutex);
862 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
864 the_lnet.ln_lnds[lnd->lnd_type] = NULL;
865 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
867 mutex_unlock(&the_lnet.ln_lnd_mutex);
869 EXPORT_SYMBOL(lnet_unregister_lnd);
872 lnet_counters_get_common(struct lnet_counters_common *common)
874 struct lnet_counters *ctr;
877 memset(common, 0, sizeof(*common));
879 lnet_net_lock(LNET_LOCK_EX);
881 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
882 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
883 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
884 common->lcc_errors += ctr->lct_common.lcc_errors;
885 common->lcc_send_count += ctr->lct_common.lcc_send_count;
886 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
887 common->lcc_route_count += ctr->lct_common.lcc_route_count;
888 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
889 common->lcc_send_length += ctr->lct_common.lcc_send_length;
890 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
891 common->lcc_route_length += ctr->lct_common.lcc_route_length;
892 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
894 lnet_net_unlock(LNET_LOCK_EX);
896 EXPORT_SYMBOL(lnet_counters_get_common);
899 lnet_counters_get(struct lnet_counters *counters)
901 struct lnet_counters *ctr;
902 struct lnet_counters_health *health = &counters->lct_health;
905 memset(counters, 0, sizeof(*counters));
907 lnet_counters_get_common(&counters->lct_common);
909 lnet_net_lock(LNET_LOCK_EX);
911 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
912 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
913 health->lch_resend_count += ctr->lct_health.lch_resend_count;
914 health->lch_response_timeout_count +=
915 ctr->lct_health.lch_response_timeout_count;
916 health->lch_local_interrupt_count +=
917 ctr->lct_health.lch_local_interrupt_count;
918 health->lch_local_dropped_count +=
919 ctr->lct_health.lch_local_dropped_count;
920 health->lch_local_aborted_count +=
921 ctr->lct_health.lch_local_aborted_count;
922 health->lch_local_no_route_count +=
923 ctr->lct_health.lch_local_no_route_count;
924 health->lch_local_timeout_count +=
925 ctr->lct_health.lch_local_timeout_count;
926 health->lch_local_error_count +=
927 ctr->lct_health.lch_local_error_count;
928 health->lch_remote_dropped_count +=
929 ctr->lct_health.lch_remote_dropped_count;
930 health->lch_remote_error_count +=
931 ctr->lct_health.lch_remote_error_count;
932 health->lch_remote_timeout_count +=
933 ctr->lct_health.lch_remote_timeout_count;
934 health->lch_network_timeout_count +=
935 ctr->lct_health.lch_network_timeout_count;
937 lnet_net_unlock(LNET_LOCK_EX);
939 EXPORT_SYMBOL(lnet_counters_get);
942 lnet_counters_reset(void)
944 struct lnet_counters *counters;
947 lnet_net_lock(LNET_LOCK_EX);
949 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
950 memset(counters, 0, sizeof(struct lnet_counters));
952 lnet_net_unlock(LNET_LOCK_EX);
956 lnet_res_type2str(int type)
961 case LNET_COOKIE_TYPE_MD:
963 case LNET_COOKIE_TYPE_ME:
965 case LNET_COOKIE_TYPE_EQ:
971 lnet_res_container_cleanup(struct lnet_res_container *rec)
975 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
978 while (!list_empty(&rec->rec_active)) {
979 struct list_head *e = rec->rec_active.next;
982 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
983 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
985 } else { /* NB: Active MEs should be attached on portals */
992 /* Found alive MD/ME/EQ, user really should unlink/free
993 * all of them before finalize LNet, but if someone didn't,
994 * we have to recycle garbage for him */
995 CERROR("%d active elements on exit of %s container\n",
996 count, lnet_res_type2str(rec->rec_type));
999 if (rec->rec_lh_hash != NULL) {
1000 LIBCFS_FREE(rec->rec_lh_hash,
1001 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1002 rec->rec_lh_hash = NULL;
1005 rec->rec_type = 0; /* mark it as finalized */
1009 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1014 LASSERT(rec->rec_type == 0);
1016 rec->rec_type = type;
1017 INIT_LIST_HEAD(&rec->rec_active);
1019 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1021 /* Arbitrary choice of hash table size */
1022 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1023 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1024 if (rec->rec_lh_hash == NULL) {
1029 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1030 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1035 CERROR("Failed to setup %s resource container\n",
1036 lnet_res_type2str(type));
1037 lnet_res_container_cleanup(rec);
1042 lnet_res_containers_destroy(struct lnet_res_container **recs)
1044 struct lnet_res_container *rec;
1047 cfs_percpt_for_each(rec, i, recs)
1048 lnet_res_container_cleanup(rec);
1050 cfs_percpt_free(recs);
1053 static struct lnet_res_container **
1054 lnet_res_containers_create(int type)
1056 struct lnet_res_container **recs;
1057 struct lnet_res_container *rec;
1061 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1063 CERROR("Failed to allocate %s resource containers\n",
1064 lnet_res_type2str(type));
1068 cfs_percpt_for_each(rec, i, recs) {
1069 rc = lnet_res_container_setup(rec, i, type);
1071 lnet_res_containers_destroy(recs);
1079 struct lnet_libhandle *
1080 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1082 /* ALWAYS called with lnet_res_lock held */
1083 struct list_head *head;
1084 struct lnet_libhandle *lh;
1087 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1090 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1091 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1093 list_for_each_entry(lh, head, lh_hash_chain) {
1094 if (lh->lh_cookie == cookie)
1102 lnet_res_lh_initialize(struct lnet_res_container *rec,
1103 struct lnet_libhandle *lh)
1105 /* ALWAYS called with lnet_res_lock held */
1106 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1109 lh->lh_cookie = rec->rec_lh_cookie;
1110 rec->rec_lh_cookie += 1 << ibits;
1112 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1114 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1118 lnet_create_array_of_queues(void)
1120 struct list_head **qs;
1121 struct list_head *q;
1124 qs = cfs_percpt_alloc(lnet_cpt_table(),
1125 sizeof(struct list_head));
1127 CERROR("Failed to allocate queues\n");
1131 cfs_percpt_for_each(q, i, qs)
1137 static int lnet_unprepare(void);
1140 lnet_prepare(lnet_pid_t requested_pid)
1142 /* Prepare to bring up the network */
1143 struct lnet_res_container **recs;
1146 if (requested_pid == LNET_PID_ANY) {
1147 /* Don't instantiate LNET just for me */
1151 LASSERT(the_lnet.ln_refcount == 0);
1153 the_lnet.ln_routing = 0;
1155 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1156 the_lnet.ln_pid = requested_pid;
1158 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1159 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1160 INIT_LIST_HEAD(&the_lnet.ln_nets);
1161 INIT_LIST_HEAD(&the_lnet.ln_routers);
1162 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1163 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1164 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1165 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1166 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1167 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1168 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1169 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1170 the_lnet.ln_mt_eq = NULL;
1171 init_completion(&the_lnet.ln_started);
1173 rc = lnet_slab_setup();
1177 rc = lnet_create_remote_nets_table();
1182 * NB the interface cookie in wire handles guards against delayed
1183 * replies and ACKs appearing valid after reboot.
1185 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1187 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1188 sizeof(struct lnet_counters));
1189 if (the_lnet.ln_counters == NULL) {
1190 CERROR("Failed to allocate counters for LNet\n");
1195 rc = lnet_peer_tables_create();
1199 rc = lnet_msg_containers_create();
1203 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1204 LNET_COOKIE_TYPE_EQ);
1208 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1214 the_lnet.ln_md_containers = recs;
1216 rc = lnet_portals_create();
1218 CERROR("Failed to create portals for LNet: %d\n", rc);
1222 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1223 if (!the_lnet.ln_mt_zombie_rstqs) {
1236 lnet_unprepare (void)
1240 /* NB no LNET_LOCK since this is the last reference. All LND instances
1241 * have shut down already, so it is safe to unlink and free all
1242 * descriptors, even those that appear committed to a network op (eg MD
1243 * with non-zero pending count) */
1245 lnet_fail_nid(LNET_NID_ANY, 0);
1247 LASSERT(the_lnet.ln_refcount == 0);
1248 LASSERT(list_empty(&the_lnet.ln_test_peers));
1249 LASSERT(list_empty(&the_lnet.ln_nets));
1251 if (the_lnet.ln_mt_zombie_rstqs) {
1252 lnet_clean_zombie_rstqs();
1253 the_lnet.ln_mt_zombie_rstqs = NULL;
1256 if (the_lnet.ln_mt_eq) {
1257 rc = LNetEQFree(the_lnet.ln_mt_eq);
1258 the_lnet.ln_mt_eq = NULL;
1262 lnet_portals_destroy();
1264 if (the_lnet.ln_md_containers != NULL) {
1265 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1266 the_lnet.ln_md_containers = NULL;
1269 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1271 lnet_msg_containers_destroy();
1273 lnet_rtrpools_free(0);
1275 if (the_lnet.ln_counters != NULL) {
1276 cfs_percpt_free(the_lnet.ln_counters);
1277 the_lnet.ln_counters = NULL;
1279 lnet_destroy_remote_nets_table();
1280 lnet_slab_cleanup();
1286 lnet_net2ni_locked(__u32 net_id, int cpt)
1289 struct lnet_net *net;
1291 LASSERT(cpt != LNET_LOCK_EX);
1293 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1294 if (net->net_id == net_id) {
1295 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1305 lnet_net2ni_addref(__u32 net)
1310 ni = lnet_net2ni_locked(net, 0);
1312 lnet_ni_addref_locked(ni, 0);
1317 EXPORT_SYMBOL(lnet_net2ni_addref);
1320 lnet_get_net_locked(__u32 net_id)
1322 struct lnet_net *net;
1324 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1325 if (net->net_id == net_id)
1333 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1338 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1343 val = hash_long(key, LNET_CPT_BITS);
1344 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1348 return (unsigned int)(key + val + (val >> 1)) % number;
1352 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1354 struct lnet_net *net;
1356 /* must called with hold of lnet_net_lock */
1357 if (LNET_CPT_NUMBER == 1)
1358 return 0; /* the only one */
1361 * If NI is provided then use the CPT identified in the NI cpt
1362 * list if one exists. If one doesn't exist, then that NI is
1363 * associated with all CPTs and it follows that the net it belongs
1364 * to is implicitly associated with all CPTs, so just hash the nid
1368 if (ni->ni_cpts != NULL)
1369 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1372 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1375 /* no NI provided so look at the net */
1376 net = lnet_get_net_locked(LNET_NIDNET(nid));
1378 if (net != NULL && net->net_cpts != NULL) {
1379 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1382 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1386 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1391 if (LNET_CPT_NUMBER == 1)
1392 return 0; /* the only one */
1394 cpt = lnet_net_lock_current();
1396 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1398 lnet_net_unlock(cpt);
1402 EXPORT_SYMBOL(lnet_cpt_of_nid);
1405 lnet_islocalnet_locked(__u32 net_id)
1407 struct lnet_net *net;
1410 net = lnet_get_net_locked(net_id);
1412 local = net != NULL;
1418 lnet_islocalnet(__u32 net_id)
1423 cpt = lnet_net_lock_current();
1425 local = lnet_islocalnet_locked(net_id);
1427 lnet_net_unlock(cpt);
1433 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1435 struct lnet_net *net;
1438 LASSERT(cpt != LNET_LOCK_EX);
1440 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1441 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1442 if (ni->ni_nid == nid)
1451 lnet_nid2ni_addref(lnet_nid_t nid)
1456 ni = lnet_nid2ni_locked(nid, 0);
1458 lnet_ni_addref_locked(ni, 0);
1463 EXPORT_SYMBOL(lnet_nid2ni_addref);
1466 lnet_islocalnid(lnet_nid_t nid)
1471 cpt = lnet_net_lock_current();
1472 ni = lnet_nid2ni_locked(nid, cpt);
1473 lnet_net_unlock(cpt);
1479 lnet_count_acceptor_nets(void)
1481 /* Return the # of NIs that need the acceptor. */
1483 struct lnet_net *net;
1486 cpt = lnet_net_lock_current();
1487 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1488 /* all socklnd type networks should have the acceptor
1490 if (net->net_lnd->lnd_accept != NULL)
1494 lnet_net_unlock(cpt);
1499 struct lnet_ping_buffer *
1500 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1502 struct lnet_ping_buffer *pbuf;
1504 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1506 pbuf->pb_nnis = nnis;
1507 atomic_set(&pbuf->pb_refcnt, 1);
1514 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1516 LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1517 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1520 static struct lnet_ping_buffer *
1521 lnet_ping_target_create(int nnis)
1523 struct lnet_ping_buffer *pbuf;
1525 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1527 CERROR("Can't allocate ping source [%d]\n", nnis);
1531 pbuf->pb_info.pi_nnis = nnis;
1532 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1533 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1534 pbuf->pb_info.pi_features =
1535 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1541 lnet_get_net_ni_count_locked(struct lnet_net *net)
1546 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1553 lnet_get_net_ni_count_pre(struct lnet_net *net)
1558 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1565 lnet_get_ni_count(void)
1568 struct lnet_net *net;
1573 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1574 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1584 lnet_get_net_count(void)
1586 struct lnet_net *net;
1591 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1601 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1603 struct lnet_ni_status *stat;
1607 __swab32s(&pbuf->pb_info.pi_magic);
1608 __swab32s(&pbuf->pb_info.pi_features);
1609 __swab32s(&pbuf->pb_info.pi_pid);
1610 __swab32s(&pbuf->pb_info.pi_nnis);
1611 nnis = pbuf->pb_info.pi_nnis;
1612 if (nnis > pbuf->pb_nnis)
1613 nnis = pbuf->pb_nnis;
1614 for (i = 0; i < nnis; i++) {
1615 stat = &pbuf->pb_info.pi_ni[i];
1616 __swab64s(&stat->ns_nid);
1617 __swab32s(&stat->ns_status);
1622 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1626 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1628 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1630 /* Loopback is guaranteed to be present */
1631 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1633 if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1639 lnet_ping_target_destroy(void)
1641 struct lnet_net *net;
1644 lnet_net_lock(LNET_LOCK_EX);
1646 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1647 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1649 ni->ni_status = NULL;
1654 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1655 the_lnet.ln_ping_target = NULL;
1657 lnet_net_unlock(LNET_LOCK_EX);
1661 lnet_ping_target_event_handler(struct lnet_event *event)
1663 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1665 if (event->unlinked)
1666 lnet_ping_buffer_decref(pbuf);
1670 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1671 struct lnet_handle_md *ping_mdh,
1672 int ni_count, bool set_eq)
1674 struct lnet_process_id id = {
1675 .nid = LNET_NID_ANY,
1679 struct lnet_md md = { NULL };
1683 the_lnet.ln_ping_target_eq =
1684 LNetEQAlloc(0, lnet_ping_target_event_handler);
1685 if (IS_ERR(the_lnet.ln_ping_target_eq)) {
1686 rc = PTR_ERR(the_lnet.ln_ping_target_eq);
1687 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1692 *ppbuf = lnet_ping_target_create(ni_count);
1693 if (*ppbuf == NULL) {
1698 /* Ping target ME/MD */
1699 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1700 LNET_PROTO_PING_MATCHBITS, 0,
1701 LNET_UNLINK, LNET_INS_AFTER);
1704 CERROR("Can't create ping target ME: %d\n", rc);
1705 goto fail_decref_ping_buffer;
1708 /* initialize md content */
1709 md.start = &(*ppbuf)->pb_info;
1710 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1711 md.threshold = LNET_MD_THRESH_INF;
1713 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1714 LNET_MD_MANAGE_REMOTE;
1715 md.eq_handle = the_lnet.ln_ping_target_eq;
1716 md.user_ptr = *ppbuf;
1718 rc = LNetMDAttach(me, md, LNET_RETAIN, ping_mdh);
1720 CERROR("Can't attach ping target MD: %d\n", rc);
1721 goto fail_unlink_ping_me;
1723 lnet_ping_buffer_addref(*ppbuf);
1727 fail_unlink_ping_me:
1729 fail_decref_ping_buffer:
1730 LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1731 lnet_ping_buffer_decref(*ppbuf);
1735 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1742 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1743 struct lnet_handle_md *ping_mdh)
1745 LNetMDUnlink(*ping_mdh);
1746 LNetInvalidateMDHandle(ping_mdh);
1748 /* NB the MD could be busy; this just starts the unlink */
1749 while (atomic_read(&pbuf->pb_refcnt) > 1) {
1750 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1751 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1756 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1759 struct lnet_net *net;
1760 struct lnet_ni_status *ns;
1765 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1766 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1767 LASSERT(i < pbuf->pb_nnis);
1769 ns = &pbuf->pb_info.pi_ni[i];
1771 ns->ns_nid = ni->ni_nid;
1774 ns->ns_status = (ni->ni_status != NULL) ?
1775 ni->ni_status->ns_status :
1784 * We (ab)use the ns_status of the loopback interface to
1785 * transmit the sequence number. The first interface listed
1786 * must be the loopback interface.
1788 rc = lnet_ping_info_validate(&pbuf->pb_info);
1790 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1793 LNET_PING_BUFFER_SEQNO(pbuf) =
1794 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1798 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1799 struct lnet_handle_md ping_mdh)
1801 struct lnet_ping_buffer *old_pbuf = NULL;
1802 struct lnet_handle_md old_ping_md;
1804 /* switch the NIs to point to the new ping info created */
1805 lnet_net_lock(LNET_LOCK_EX);
1807 if (!the_lnet.ln_routing)
1808 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1809 if (!lnet_peer_discovery_disabled)
1810 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1812 /* Ensure only known feature bits have been set. */
1813 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1814 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1816 lnet_ping_target_install_locked(pbuf);
1818 if (the_lnet.ln_ping_target) {
1819 old_pbuf = the_lnet.ln_ping_target;
1820 old_ping_md = the_lnet.ln_ping_target_md;
1822 the_lnet.ln_ping_target_md = ping_mdh;
1823 the_lnet.ln_ping_target = pbuf;
1825 lnet_net_unlock(LNET_LOCK_EX);
1828 /* unlink and free the old ping info */
1829 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1830 lnet_ping_buffer_decref(old_pbuf);
1833 lnet_push_update_to_peers(0);
1837 lnet_ping_target_fini(void)
1841 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1842 &the_lnet.ln_ping_target_md);
1844 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1847 lnet_ping_target_destroy();
1850 /* Resize the push target. */
1851 int lnet_push_target_resize(void)
1853 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1854 struct lnet_md md = { NULL };
1856 struct lnet_handle_md mdh;
1857 struct lnet_handle_md old_mdh;
1858 struct lnet_ping_buffer *pbuf;
1859 struct lnet_ping_buffer *old_pbuf;
1860 int nnis = the_lnet.ln_push_target_nnis;
1868 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1874 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1875 LNET_PROTO_PING_MATCHBITS, 0,
1876 LNET_UNLINK, LNET_INS_AFTER);
1880 CERROR("Can't create push target ME: %d\n", rc);
1881 goto fail_decref_pbuf;
1884 /* initialize md content */
1885 md.start = &pbuf->pb_info;
1886 md.length = LNET_PING_INFO_SIZE(nnis);
1887 md.threshold = LNET_MD_THRESH_INF;
1889 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1890 LNET_MD_MANAGE_REMOTE;
1892 md.eq_handle = the_lnet.ln_push_target_eq;
1894 rc = LNetMDAttach(me, md, LNET_RETAIN, &mdh);
1896 CERROR("Can't attach push MD: %d\n", rc);
1897 goto fail_unlink_me;
1899 lnet_ping_buffer_addref(pbuf);
1901 lnet_net_lock(LNET_LOCK_EX);
1902 old_pbuf = the_lnet.ln_push_target;
1903 old_mdh = the_lnet.ln_push_target_md;
1904 the_lnet.ln_push_target = pbuf;
1905 the_lnet.ln_push_target_md = mdh;
1906 lnet_net_unlock(LNET_LOCK_EX);
1909 LNetMDUnlink(old_mdh);
1910 lnet_ping_buffer_decref(old_pbuf);
1913 if (nnis < the_lnet.ln_push_target_nnis)
1916 CDEBUG(D_NET, "nnis %d success\n", nnis);
1923 lnet_ping_buffer_decref(pbuf);
1925 CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1929 static void lnet_push_target_event_handler(struct lnet_event *ev)
1931 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1933 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1934 lnet_swap_pinginfo(pbuf);
1936 lnet_peer_push_event(ev);
1938 lnet_ping_buffer_decref(pbuf);
1941 /* Initialize the push target. */
1942 static int lnet_push_target_init(void)
1946 if (the_lnet.ln_push_target)
1949 the_lnet.ln_push_target_eq =
1950 LNetEQAlloc(0, lnet_push_target_event_handler);
1951 if (IS_ERR(the_lnet.ln_push_target_eq)) {
1952 rc = PTR_ERR(the_lnet.ln_push_target_eq);
1953 CERROR("Can't allocated push target EQ: %d\n", rc);
1957 /* Start at the required minimum, we'll enlarge if required. */
1958 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1960 rc = lnet_push_target_resize();
1963 LNetEQFree(the_lnet.ln_push_target_eq);
1964 the_lnet.ln_push_target_eq = NULL;
1970 /* Clean up the push target. */
1971 static void lnet_push_target_fini(void)
1973 if (!the_lnet.ln_push_target)
1976 /* Unlink and invalidate to prevent new references. */
1977 LNetMDUnlink(the_lnet.ln_push_target_md);
1978 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1980 /* Wait for the unlink to complete. */
1981 while (atomic_read(&the_lnet.ln_push_target->pb_refcnt) > 1) {
1982 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1983 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1986 lnet_ping_buffer_decref(the_lnet.ln_push_target);
1987 the_lnet.ln_push_target = NULL;
1988 the_lnet.ln_push_target_nnis = 0;
1990 LNetEQFree(the_lnet.ln_push_target_eq);
1991 the_lnet.ln_push_target_eq = NULL;
1995 lnet_ni_tq_credits(struct lnet_ni *ni)
1999 LASSERT(ni->ni_ncpts >= 1);
2001 if (ni->ni_ncpts == 1)
2002 return ni->ni_net->net_tunables.lct_max_tx_credits;
2004 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2005 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2006 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2012 lnet_ni_unlink_locked(struct lnet_ni *ni)
2014 /* move it to zombie list and nobody can find it anymore */
2015 LASSERT(!list_empty(&ni->ni_netlist));
2016 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2017 lnet_ni_decref_locked(ni, 0);
2021 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2026 struct list_head *zombie_list = &net->net_ni_zombie;
2029 * Now wait for the NIs I just nuked to show up on the zombie
2030 * list and shut them down in guaranteed thread context
2033 while (!list_empty(zombie_list)) {
2037 ni = list_entry(zombie_list->next,
2038 struct lnet_ni, ni_netlist);
2039 list_del_init(&ni->ni_netlist);
2040 /* the ni should be in deleting state. If it's not it's
2042 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2043 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2046 /* still busy, add it back to zombie list */
2047 list_add(&ni->ni_netlist, zombie_list);
2051 if (!list_empty(&ni->ni_netlist)) {
2052 lnet_net_unlock(LNET_LOCK_EX);
2054 if ((i & (-i)) == i) {
2056 "Waiting for zombie LNI %s\n",
2057 libcfs_nid2str(ni->ni_nid));
2059 schedule_timeout_uninterruptible(cfs_time_seconds(1));
2060 lnet_net_lock(LNET_LOCK_EX);
2064 lnet_net_unlock(LNET_LOCK_EX);
2066 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2068 LASSERT(!in_interrupt());
2069 /* Holding the mutex makes it safe for lnd_shutdown
2070 * to call module_put(). Module unload cannot finish
2071 * until lnet_unregister_lnd() completes, and that
2072 * requires the mutex.
2074 mutex_lock(&the_lnet.ln_lnd_mutex);
2075 (net->net_lnd->lnd_shutdown)(ni);
2076 mutex_unlock(&the_lnet.ln_lnd_mutex);
2079 CDEBUG(D_LNI, "Removed LNI %s\n",
2080 libcfs_nid2str(ni->ni_nid));
2084 lnet_net_lock(LNET_LOCK_EX);
2088 /* shutdown down the NI and release refcount */
2090 lnet_shutdown_lndni(struct lnet_ni *ni)
2093 struct lnet_net *net = ni->ni_net;
2095 lnet_net_lock(LNET_LOCK_EX);
2097 ni->ni_state = LNET_NI_STATE_DELETING;
2099 lnet_ni_unlink_locked(ni);
2100 lnet_incr_dlc_seq();
2101 lnet_net_unlock(LNET_LOCK_EX);
2103 /* clear messages for this NI on the lazy portal */
2104 for (i = 0; i < the_lnet.ln_nportals; i++)
2105 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2107 lnet_net_lock(LNET_LOCK_EX);
2108 lnet_clear_zombies_nis_locked(net);
2109 lnet_net_unlock(LNET_LOCK_EX);
2113 lnet_shutdown_lndnet(struct lnet_net *net)
2117 lnet_net_lock(LNET_LOCK_EX);
2119 list_del_init(&net->net_list);
2121 while (!list_empty(&net->net_ni_list)) {
2122 ni = list_entry(net->net_ni_list.next,
2123 struct lnet_ni, ni_netlist);
2124 lnet_net_unlock(LNET_LOCK_EX);
2125 lnet_shutdown_lndni(ni);
2126 lnet_net_lock(LNET_LOCK_EX);
2129 lnet_net_unlock(LNET_LOCK_EX);
2131 /* Do peer table cleanup for this net */
2132 lnet_peer_tables_cleanup(net);
2138 lnet_shutdown_lndnets(void)
2140 struct lnet_net *net;
2142 struct lnet_msg *msg, *tmp;
2144 /* NB called holding the global mutex */
2146 /* All quiet on the API front */
2147 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2148 LASSERT(the_lnet.ln_refcount == 0);
2150 lnet_net_lock(LNET_LOCK_EX);
2151 the_lnet.ln_state = LNET_STATE_STOPPING;
2154 * move the nets to the zombie list to avoid them being
2155 * picked up for new work. LONET is also included in the
2156 * Nets that will be moved to the zombie list
2158 list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2160 /* Drop the cached loopback Net. */
2161 if (the_lnet.ln_loni != NULL) {
2162 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2163 the_lnet.ln_loni = NULL;
2165 lnet_net_unlock(LNET_LOCK_EX);
2167 /* iterate through the net zombie list and delete each net */
2168 while (!list_empty(&the_lnet.ln_net_zombie)) {
2169 net = list_entry(the_lnet.ln_net_zombie.next,
2170 struct lnet_net, net_list);
2171 lnet_shutdown_lndnet(net);
2174 spin_lock(&the_lnet.ln_msg_resend_lock);
2175 list_splice(&the_lnet.ln_msg_resend, &resend);
2176 spin_unlock(&the_lnet.ln_msg_resend_lock);
2178 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2179 list_del_init(&msg->msg_list);
2180 msg->msg_no_resend = true;
2181 lnet_finalize(msg, -ECANCELED);
2184 lnet_net_lock(LNET_LOCK_EX);
2185 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2186 lnet_net_unlock(LNET_LOCK_EX);
2190 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2193 struct lnet_tx_queue *tq;
2195 struct lnet_net *net = ni->ni_net;
2197 mutex_lock(&the_lnet.ln_lnd_mutex);
2200 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2201 ni->ni_lnd_tunables_set = true;
2204 rc = (net->net_lnd->lnd_startup)(ni);
2206 mutex_unlock(&the_lnet.ln_lnd_mutex);
2209 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2210 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2215 ni->ni_state = LNET_NI_STATE_ACTIVE;
2218 /* We keep a reference on the loopback net through the loopback NI */
2219 if (net->net_lnd->lnd_type == LOLND) {
2221 LASSERT(the_lnet.ln_loni == NULL);
2222 the_lnet.ln_loni = ni;
2223 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2224 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2225 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2226 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2230 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2231 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2232 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2233 libcfs_lnd2str(net->net_lnd->lnd_type),
2234 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2236 /* shutdown the NI since if we get here then it must've already
2239 lnet_shutdown_lndni(ni);
2243 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2244 tq->tq_credits_min =
2245 tq->tq_credits_max =
2246 tq->tq_credits = lnet_ni_tq_credits(ni);
2249 atomic_set(&ni->ni_tx_credits,
2250 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2251 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2253 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2254 libcfs_nid2str(ni->ni_nid),
2255 ni->ni_net->net_tunables.lct_peer_tx_credits,
2256 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2257 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2258 ni->ni_net->net_tunables.lct_peer_timeout);
2267 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2270 struct lnet_net *net_l = NULL;
2271 LIST_HEAD(local_ni_list);
2275 const struct lnet_lnd *lnd;
2277 net->net_tunables.lct_peer_timeout;
2279 net->net_tunables.lct_max_tx_credits;
2280 int peerrtrcredits =
2281 net->net_tunables.lct_peer_rtr_credits;
2284 * make sure that this net is unique. If it isn't then
2285 * we are adding interfaces to an already existing network, and
2286 * 'net' is just a convenient way to pass in the list.
2287 * if it is unique we need to find the LND and load it if
2290 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2291 lnd_type = LNET_NETTYP(net->net_id);
2293 mutex_lock(&the_lnet.ln_lnd_mutex);
2294 lnd = lnet_find_lnd_by_type(lnd_type);
2297 mutex_unlock(&the_lnet.ln_lnd_mutex);
2298 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2299 mutex_lock(&the_lnet.ln_lnd_mutex);
2301 lnd = lnet_find_lnd_by_type(lnd_type);
2303 mutex_unlock(&the_lnet.ln_lnd_mutex);
2304 CERROR("Can't load LND %s, module %s, rc=%d\n",
2305 libcfs_lnd2str(lnd_type),
2306 libcfs_lnd2modname(lnd_type), rc);
2307 #ifndef HAVE_MODULE_LOADING_SUPPORT
2308 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2309 "compiled with kernel module "
2310 "loading support.");
2319 mutex_unlock(&the_lnet.ln_lnd_mutex);
2325 * net_l: if the network being added is unique then net_l
2326 * will point to that network
2327 * if the network being added is not unique then
2328 * net_l points to the existing network.
2330 * When we enter the loop below, we'll pick NIs off he
2331 * network beign added and start them up, then add them to
2332 * a local ni list. Once we've successfully started all
2333 * the NIs then we join the local NI list (of started up
2334 * networks) with the net_l->net_ni_list, which should
2335 * point to the correct network to add the new ni list to
2337 * If any of the new NIs fail to start up, then we want to
2338 * iterate through the local ni list, which should include
2339 * any NIs which were successfully started up, and shut
2342 * After than we want to delete the network being added,
2343 * to avoid a memory leak.
2347 * When a network uses TCP bonding then all its interfaces
2348 * must be specified when the network is first defined: the
2349 * TCP bonding code doesn't allow for interfaces to be added
2352 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2353 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2358 while (!list_empty(&net->net_ni_added)) {
2359 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2361 list_del_init(&ni->ni_netlist);
2363 /* make sure that the the NI we're about to start
2364 * up is actually unique. if it's not fail. */
2365 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2366 ni->ni_interfaces[0])) {
2371 /* adjust the pointer the parent network, just in case it
2372 * the net is a duplicate */
2375 rc = lnet_startup_lndni(ni, tun);
2381 list_add_tail(&ni->ni_netlist, &local_ni_list);
2386 lnet_net_lock(LNET_LOCK_EX);
2387 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2388 lnet_incr_dlc_seq();
2389 lnet_net_unlock(LNET_LOCK_EX);
2391 /* if the network is not unique then we don't want to keep
2392 * it around after we're done. Free it. Otherwise add that
2393 * net to the global the_lnet.ln_nets */
2394 if (net_l != net && net_l != NULL) {
2396 * TODO - note. currently the tunables can not be updated
2402 * restore tunables after it has been overwitten by the
2405 if (peer_timeout != -1)
2406 net->net_tunables.lct_peer_timeout = peer_timeout;
2407 if (maxtxcredits != -1)
2408 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2409 if (peerrtrcredits != -1)
2410 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2412 lnet_net_lock(LNET_LOCK_EX);
2413 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2414 lnet_net_unlock(LNET_LOCK_EX);
2417 /* update net count */
2418 lnet_current_net_count = lnet_get_net_count();
2424 * shutdown the new NIs that are being started up
2425 * free the NET being started
2427 while (!list_empty(&local_ni_list)) {
2428 ni = list_entry(local_ni_list.next, struct lnet_ni,
2431 lnet_shutdown_lndni(ni);
2441 lnet_startup_lndnets(struct list_head *netlist)
2443 struct lnet_net *net;
2448 * Change to running state before bringing up the LNDs. This
2449 * allows lnet_shutdown_lndnets() to assert that we've passed
2452 lnet_net_lock(LNET_LOCK_EX);
2453 the_lnet.ln_state = LNET_STATE_RUNNING;
2454 lnet_net_unlock(LNET_LOCK_EX);
2456 while (!list_empty(netlist)) {
2457 net = list_entry(netlist->next, struct lnet_net, net_list);
2458 list_del_init(&net->net_list);
2460 rc = lnet_startup_lndnet(net, NULL);
2470 lnet_shutdown_lndnets();
2476 * Initialize LNet library.
2478 * Automatically called at module loading time. Caller has to call
2479 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2480 * latter returned 0. It must be called exactly once.
2482 * \retval 0 on success
2483 * \retval -ve on failures.
2485 int lnet_lib_init(void)
2489 lnet_assert_wire_constants();
2491 /* refer to global cfs_cpt_table for now */
2492 the_lnet.ln_cpt_table = cfs_cpt_tab;
2493 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2495 LASSERT(the_lnet.ln_cpt_number > 0);
2496 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2497 /* we are under risk of consuming all lh_cookie */
2498 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2499 "please change setting of CPT-table and retry\n",
2500 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2504 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2505 the_lnet.ln_cpt_bits++;
2507 rc = lnet_create_locks();
2509 CERROR("Can't create LNet global locks: %d\n", rc);
2513 the_lnet.ln_refcount = 0;
2514 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2515 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2517 /* The hash table size is the number of bits it takes to express the set
2518 * ln_num_routes, minus 1 (better to under estimate than over so we
2519 * don't waste memory). */
2520 if (rnet_htable_size <= 0)
2521 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2522 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2523 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2524 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2525 order_base_2(rnet_htable_size) - 1);
2527 /* All LNDs apart from the LOLND are in separate modules. They
2528 * register themselves when their module loads, and unregister
2529 * themselves when their module is unloaded. */
2530 lnet_register_lnd(&the_lolnd);
2535 * Finalize LNet library.
2537 * \pre lnet_lib_init() called with success.
2538 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2540 * As this happens at module-unload, all lnds must already be unloaded,
2541 * so they must already be unregistered.
2543 void lnet_lib_exit(void)
2547 LASSERT(the_lnet.ln_refcount == 0);
2548 lnet_unregister_lnd(&the_lolnd);
2549 for (i = 0; i < NUM_LNDS; i++)
2550 LASSERT(!the_lnet.ln_lnds[i]);
2551 lnet_destroy_locks();
2555 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2557 * Users must call this function at least once before any other functions.
2558 * For each successful call there must be a corresponding call to
2559 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2562 * The PID used by LNet may be different from the one requested.
2565 * \param requested_pid PID requested by the caller.
2567 * \return >= 0 on success, and < 0 error code on failures.
2570 LNetNIInit(lnet_pid_t requested_pid)
2572 int im_a_router = 0;
2575 struct lnet_ping_buffer *pbuf;
2576 struct lnet_handle_md ping_mdh;
2577 LIST_HEAD(net_head);
2578 struct lnet_net *net;
2580 mutex_lock(&the_lnet.ln_api_mutex);
2582 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2584 if (the_lnet.ln_refcount > 0) {
2585 rc = the_lnet.ln_refcount++;
2586 mutex_unlock(&the_lnet.ln_api_mutex);
2590 rc = lnet_prepare(requested_pid);
2592 mutex_unlock(&the_lnet.ln_api_mutex);
2596 /* create a network for Loopback network */
2597 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2600 goto err_empty_list;
2603 /* Add in the loopback NI */
2604 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2606 goto err_empty_list;
2609 /* If LNet is being initialized via DLC it is possible
2610 * that the user requests not to load module parameters (ones which
2611 * are supported by DLC) on initialization. Therefore, make sure not
2612 * to load networks, routes and forwarding from module parameters
2613 * in this case. On cleanup in case of failure only clean up
2614 * routes if it has been loaded */
2615 if (!the_lnet.ln_nis_from_mod_params) {
2616 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2619 goto err_empty_list;
2622 ni_count = lnet_startup_lndnets(&net_head);
2625 goto err_empty_list;
2628 if (!the_lnet.ln_nis_from_mod_params) {
2629 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2631 goto err_shutdown_lndnis;
2633 rc = lnet_rtrpools_alloc(im_a_router);
2635 goto err_destroy_routes;
2638 rc = lnet_acceptor_start();
2640 goto err_destroy_routes;
2642 the_lnet.ln_refcount = 1;
2643 /* Now I may use my own API functions... */
2645 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2647 goto err_acceptor_stop;
2649 lnet_ping_target_update(pbuf, ping_mdh);
2651 the_lnet.ln_mt_eq = LNetEQAlloc(0, lnet_mt_event_handler);
2652 if (IS_ERR(the_lnet.ln_mt_eq)) {
2653 rc = PTR_ERR(the_lnet.ln_mt_eq);
2654 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2658 rc = lnet_push_target_init();
2662 rc = lnet_peer_discovery_start();
2664 goto err_destroy_push_target;
2666 rc = lnet_monitor_thr_start();
2668 goto err_stop_discovery_thr;
2671 lnet_router_debugfs_init();
2673 mutex_unlock(&the_lnet.ln_api_mutex);
2675 complete_all(&the_lnet.ln_started);
2677 /* wait for all routers to start */
2678 lnet_wait_router_start();
2682 err_stop_discovery_thr:
2683 lnet_peer_discovery_stop();
2684 err_destroy_push_target:
2685 lnet_push_target_fini();
2687 lnet_ping_target_fini();
2689 the_lnet.ln_refcount = 0;
2690 lnet_acceptor_stop();
2692 if (!the_lnet.ln_nis_from_mod_params)
2693 lnet_destroy_routes();
2694 err_shutdown_lndnis:
2695 lnet_shutdown_lndnets();
2699 mutex_unlock(&the_lnet.ln_api_mutex);
2700 while (!list_empty(&net_head)) {
2701 struct lnet_net *net;
2703 net = list_entry(net_head.next, struct lnet_net, net_list);
2704 list_del_init(&net->net_list);
2709 EXPORT_SYMBOL(LNetNIInit);
2712 * Stop LNet interfaces, routing, and forwarding.
2714 * Users must call this function once for each successful call to LNetNIInit().
2715 * Once the LNetNIFini() operation has been started, the results of pending
2716 * API operations are undefined.
2718 * \return always 0 for current implementation.
2723 mutex_lock(&the_lnet.ln_api_mutex);
2725 LASSERT(the_lnet.ln_refcount > 0);
2727 if (the_lnet.ln_refcount != 1) {
2728 the_lnet.ln_refcount--;
2730 LASSERT(!the_lnet.ln_niinit_self);
2734 lnet_router_debugfs_fini();
2735 lnet_monitor_thr_stop();
2736 lnet_peer_discovery_stop();
2737 lnet_push_target_fini();
2738 lnet_ping_target_fini();
2740 /* Teardown fns that use my own API functions BEFORE here */
2741 the_lnet.ln_refcount = 0;
2743 lnet_acceptor_stop();
2744 lnet_destroy_routes();
2745 lnet_shutdown_lndnets();
2749 mutex_unlock(&the_lnet.ln_api_mutex);
2752 EXPORT_SYMBOL(LNetNIFini);
2755 * Grabs the ni data from the ni structure and fills the out
2758 * \param[in] ni network interface structure
2759 * \param[out] cfg_ni NI config information
2760 * \param[out] tun network and LND tunables
2763 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2764 struct lnet_ioctl_config_lnd_tunables *tun,
2765 struct lnet_ioctl_element_stats *stats,
2768 size_t min_size = 0;
2771 if (!ni || !cfg_ni || !tun)
2774 if (ni->ni_interfaces[0] != NULL) {
2775 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2776 if (ni->ni_interfaces[i] != NULL) {
2777 strncpy(cfg_ni->lic_ni_intf[i],
2778 ni->ni_interfaces[i],
2779 sizeof(cfg_ni->lic_ni_intf[i]));
2784 cfg_ni->lic_nid = ni->ni_nid;
2785 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2786 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2788 cfg_ni->lic_status = ni->ni_status->ns_status;
2789 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2790 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2792 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2795 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2796 LNET_STATS_TYPE_SEND);
2797 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2798 LNET_STATS_TYPE_RECV);
2799 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2800 LNET_STATS_TYPE_DROP);
2804 * tun->lt_tun will always be present, but in order to be
2805 * backwards compatible, we need to deal with the cases when
2806 * tun->lt_tun is smaller than what the kernel has, because it
2807 * comes from an older version of a userspace program, then we'll
2808 * need to copy as much information as we have available space.
2810 min_size = tun_size - sizeof(tun->lt_cmn);
2811 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2813 /* copy over the cpts */
2814 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2815 ni->ni_cpts == NULL) {
2816 for (i = 0; i < ni->ni_ncpts; i++)
2817 cfg_ni->lic_cpts[i] = i;
2820 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2821 i < LNET_MAX_SHOW_NUM_CPT;
2823 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2825 cfg_ni->lic_ncpts = ni->ni_ncpts;
2829 * NOTE: This is a legacy function left in the code to be backwards
2830 * compatible with older userspace programs. It should eventually be
2833 * Grabs the ni data from the ni structure and fills the out
2836 * \param[in] ni network interface structure
2837 * \param[out] config config information
2840 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2841 struct lnet_ioctl_config_data *config)
2843 struct lnet_ioctl_net_config *net_config;
2844 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2845 size_t min_size, tunable_size = 0;
2851 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2855 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2856 ARRAY_SIZE(net_config->ni_interfaces));
2858 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2859 if (!ni->ni_interfaces[i])
2862 strncpy(net_config->ni_interfaces[i],
2863 ni->ni_interfaces[i],
2864 sizeof(net_config->ni_interfaces[i]));
2867 config->cfg_nid = ni->ni_nid;
2868 config->cfg_config_u.cfg_net.net_peer_timeout =
2869 ni->ni_net->net_tunables.lct_peer_timeout;
2870 config->cfg_config_u.cfg_net.net_max_tx_credits =
2871 ni->ni_net->net_tunables.lct_max_tx_credits;
2872 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2873 ni->ni_net->net_tunables.lct_peer_tx_credits;
2874 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2875 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2877 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2878 net_config->ni_status = LNET_NI_STATUS_UP;
2880 net_config->ni_status = ni->ni_status->ns_status;
2883 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2885 for (i = 0; i < num_cpts; i++)
2886 net_config->ni_cpts[i] = ni->ni_cpts[i];
2888 config->cfg_ncpts = num_cpts;
2892 * See if user land tools sent in a newer and larger version
2893 * of struct lnet_tunables than what the kernel uses.
2895 min_size = sizeof(*config) + sizeof(*net_config);
2897 if (config->cfg_hdr.ioc_len > min_size)
2898 tunable_size = config->cfg_hdr.ioc_len - min_size;
2900 /* Don't copy too much data to user space */
2901 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2902 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2904 if (lnd_cfg && min_size) {
2905 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2906 config->cfg_config_u.cfg_net.net_interface_count = 1;
2908 /* Tell user land that kernel side has less data */
2909 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2910 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2911 config->cfg_hdr.ioc_len -= min_size;
2917 lnet_get_ni_idx_locked(int idx)
2920 struct lnet_net *net;
2922 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2923 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2933 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2936 struct lnet_net *net = mynet;
2939 * It is possible that the net has been cleaned out while there is
2940 * a message being sent. This function accessed the net without
2941 * checking if the list is empty
2945 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2947 if (list_empty(&net->net_ni_list))
2949 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2955 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2956 /* if you reached the end of the ni list and the net is
2957 * specified, then there are no more nis in that net */
2961 /* we reached the end of this net ni list. move to the
2963 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2964 /* no more nets and no more NIs. */
2967 /* get the next net */
2968 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2970 if (list_empty(&net->net_ni_list))
2972 /* get the ni on it */
2973 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2979 if (list_empty(&prev->ni_netlist))
2982 /* there are more nis left */
2983 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2989 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2994 int idx = config->cfg_count;
2996 cpt = lnet_net_lock_current();
2998 ni = lnet_get_ni_idx_locked(idx);
3003 lnet_fill_ni_info_legacy(ni, config);
3007 lnet_net_unlock(cpt);
3012 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3013 struct lnet_ioctl_config_lnd_tunables *tun,
3014 struct lnet_ioctl_element_stats *stats,
3021 if (!cfg_ni || !tun || !stats)
3024 cpt = lnet_net_lock_current();
3026 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3031 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3035 lnet_net_unlock(cpt);
3039 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3048 cpt = lnet_net_lock_current();
3050 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3053 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3057 lnet_net_unlock(cpt);
3062 static int lnet_add_net_common(struct lnet_net *net,
3063 struct lnet_ioctl_config_lnd_tunables *tun)
3066 struct lnet_ping_buffer *pbuf;
3067 struct lnet_handle_md ping_mdh;
3069 struct lnet_remotenet *rnet;
3071 int num_acceptor_nets;
3073 lnet_net_lock(LNET_LOCK_EX);
3074 rnet = lnet_find_rnet_locked(net->net_id);
3075 lnet_net_unlock(LNET_LOCK_EX);
3077 * make sure that the net added doesn't invalidate the current
3078 * configuration LNet is keeping
3081 CERROR("Adding net %s will invalidate routing configuration\n",
3082 libcfs_net2str(net->net_id));
3088 * make sure you calculate the correct number of slots in the ping
3089 * buffer. Since the ping info is a flattened list of all the NIs,
3090 * we should allocate enough slots to accomodate the number of NIs
3091 * which will be added.
3093 * since ni hasn't been configured yet, use
3094 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3096 net_ni_count = lnet_get_net_ni_count_pre(net);
3098 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3099 net_ni_count + lnet_get_ni_count(),
3107 memcpy(&net->net_tunables,
3108 &tun->lt_cmn, sizeof(net->net_tunables));
3110 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3113 * before starting this network get a count of the current TCP
3114 * networks which require the acceptor thread running. If that
3115 * count is == 0 before we start up this network, then we'd want to
3116 * start up the acceptor thread after starting up this network
3118 num_acceptor_nets = lnet_count_acceptor_nets();
3120 net_id = net->net_id;
3122 rc = lnet_startup_lndnet(net,
3123 (tun) ? &tun->lt_tun : NULL);
3127 lnet_net_lock(LNET_LOCK_EX);
3128 net = lnet_get_net_locked(net_id);
3129 lnet_net_unlock(LNET_LOCK_EX);
3134 * Start the acceptor thread if this is the first network
3135 * being added that requires the thread.
3137 if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3138 rc = lnet_acceptor_start();
3140 /* shutdown the net that we just started */
3141 CERROR("Failed to start up acceptor thread\n");
3142 lnet_shutdown_lndnet(net);
3147 lnet_net_lock(LNET_LOCK_EX);
3148 lnet_peer_net_added(net);
3149 lnet_net_unlock(LNET_LOCK_EX);
3151 lnet_ping_target_update(pbuf, ping_mdh);
3156 lnet_ping_md_unlink(pbuf, &ping_mdh);
3157 lnet_ping_buffer_decref(pbuf);
3161 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3162 struct lnet_ioctl_config_lnd_tunables *tun)
3164 struct lnet_net *net;
3167 LIST_HEAD(net_head);
3169 rc = lnet_parse_ip2nets(&nets, ip2nets);
3173 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3177 mutex_lock(&the_lnet.ln_api_mutex);
3178 while (!list_empty(&net_head)) {
3179 net = list_entry(net_head.next, struct lnet_net, net_list);
3180 list_del_init(&net->net_list);
3181 rc = lnet_add_net_common(net, tun);
3187 mutex_unlock(&the_lnet.ln_api_mutex);
3189 while (!list_empty(&net_head)) {
3190 net = list_entry(net_head.next, struct lnet_net, net_list);
3191 list_del_init(&net->net_list);
3197 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3199 struct lnet_net *net;
3201 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3203 __u32 net_id, lnd_type;
3205 /* get the tunables if they are available */
3206 if (conf->lic_cfg_hdr.ioc_len >=
3207 sizeof(*conf) + sizeof(*tun))
3208 tun = (struct lnet_ioctl_config_lnd_tunables *)
3211 /* handle legacy ip2nets from DLC */
3212 if (conf->lic_legacy_ip2nets[0] != '\0')
3213 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3216 net_id = LNET_NIDNET(conf->lic_nid);
3217 lnd_type = LNET_NETTYP(net_id);
3219 if (!libcfs_isknown_lnd(lnd_type)) {
3220 CERROR("No valid net and lnd information provided\n");
3224 net = lnet_net_alloc(net_id, NULL);
3228 for (i = 0; i < conf->lic_ncpts; i++) {
3229 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3233 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3234 conf->lic_ni_intf[0]);
3238 mutex_lock(&the_lnet.ln_api_mutex);
3240 rc = lnet_add_net_common(net, tun);
3242 mutex_unlock(&the_lnet.ln_api_mutex);
3247 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3249 struct lnet_net *net;
3251 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3252 struct lnet_ping_buffer *pbuf;
3253 struct lnet_handle_md ping_mdh;
3258 /* don't allow userspace to shutdown the LOLND */
3259 if (LNET_NETTYP(net_id) == LOLND)
3262 mutex_lock(&the_lnet.ln_api_mutex);
3266 net = lnet_get_net_locked(net_id);
3268 CERROR("net %s not found\n",
3269 libcfs_net2str(net_id));
3274 addr = LNET_NIDADDR(conf->lic_nid);
3276 /* remove the entire net */
3277 net_count = lnet_get_net_ni_count_locked(net);
3281 /* create and link a new ping info, before removing the old one */
3282 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3283 lnet_get_ni_count() - net_count,
3286 goto unlock_api_mutex;
3288 lnet_shutdown_lndnet(net);
3290 if (lnet_count_acceptor_nets() == 0)
3291 lnet_acceptor_stop();
3293 lnet_ping_target_update(pbuf, ping_mdh);
3295 goto unlock_api_mutex;
3298 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3300 CERROR("nid %s not found\n",
3301 libcfs_nid2str(conf->lic_nid));
3306 net_count = lnet_get_net_ni_count_locked(net);
3310 /* create and link a new ping info, before removing the old one */
3311 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3312 lnet_get_ni_count() - 1, false);
3314 goto unlock_api_mutex;
3316 lnet_shutdown_lndni(ni);
3318 if (lnet_count_acceptor_nets() == 0)
3319 lnet_acceptor_stop();
3321 lnet_ping_target_update(pbuf, ping_mdh);
3323 /* check if the net is empty and remove it if it is */
3325 lnet_shutdown_lndnet(net);
3327 goto unlock_api_mutex;
3332 mutex_unlock(&the_lnet.ln_api_mutex);
3338 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3339 * They are only expected to be called for unique networks.
3340 * That can be as a result of older DLC library
3341 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3344 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3346 struct lnet_net *net;
3347 LIST_HEAD(net_head);
3349 struct lnet_ioctl_config_lnd_tunables tun;
3350 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3352 /* Create a net/ni structures for the network string */
3353 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3355 return rc == 0 ? -EINVAL : rc;
3357 mutex_lock(&the_lnet.ln_api_mutex);
3360 rc = -EINVAL; /* only add one network per call */
3361 goto out_unlock_clean;
3364 net = list_entry(net_head.next, struct lnet_net, net_list);
3365 list_del_init(&net->net_list);
3367 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3369 memset(&tun, 0, sizeof(tun));
3371 tun.lt_cmn.lct_peer_timeout =
3372 conf->cfg_config_u.cfg_net.net_peer_timeout;
3373 tun.lt_cmn.lct_peer_tx_credits =
3374 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3375 tun.lt_cmn.lct_peer_rtr_credits =
3376 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3377 tun.lt_cmn.lct_max_tx_credits =
3378 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3380 rc = lnet_add_net_common(net, &tun);
3383 mutex_unlock(&the_lnet.ln_api_mutex);
3384 while (!list_empty(&net_head)) {
3385 /* net_head list is empty in success case */
3386 net = list_entry(net_head.next, struct lnet_net, net_list);
3387 list_del_init(&net->net_list);
3394 lnet_dyn_del_net(__u32 net_id)
3396 struct lnet_net *net;
3397 struct lnet_ping_buffer *pbuf;
3398 struct lnet_handle_md ping_mdh;
3402 /* don't allow userspace to shutdown the LOLND */
3403 if (LNET_NETTYP(net_id) == LOLND)
3406 mutex_lock(&the_lnet.ln_api_mutex);
3410 net = lnet_get_net_locked(net_id);
3417 net_ni_count = lnet_get_net_ni_count_locked(net);
3421 /* create and link a new ping info, before removing the old one */
3422 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3423 lnet_get_ni_count() - net_ni_count, false);
3427 lnet_shutdown_lndnet(net);
3429 if (lnet_count_acceptor_nets() == 0)
3430 lnet_acceptor_stop();
3432 lnet_ping_target_update(pbuf, ping_mdh);
3435 mutex_unlock(&the_lnet.ln_api_mutex);
3440 void lnet_incr_dlc_seq(void)
3442 atomic_inc(&lnet_dlc_seq_no);
3445 __u32 lnet_get_dlc_seq_locked(void)
3447 return atomic_read(&lnet_dlc_seq_no);
3451 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3453 struct lnet_net *net;
3456 lnet_net_lock(LNET_LOCK_EX);
3457 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3458 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3459 if (ni->ni_nid == nid || all) {
3460 atomic_set(&ni->ni_healthv, value);
3461 if (list_empty(&ni->ni_recovery) &&
3462 value < LNET_MAX_HEALTH_VALUE) {
3463 CERROR("manually adding local NI %s to recovery\n",
3464 libcfs_nid2str(ni->ni_nid));
3465 list_add_tail(&ni->ni_recovery,
3466 &the_lnet.ln_mt_localNIRecovq);
3467 lnet_ni_addref_locked(ni, 0);
3470 lnet_net_unlock(LNET_LOCK_EX);
3476 lnet_net_unlock(LNET_LOCK_EX);
3480 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3484 lnet_nid_t nid = stats->hlni_nid;
3486 cpt = lnet_net_lock_current();
3487 ni = lnet_nid2ni_locked(nid, cpt);
3494 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3495 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3496 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3497 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3498 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3499 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3500 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3503 lnet_net_unlock(cpt);
3509 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3514 lnet_net_lock(LNET_LOCK_EX);
3515 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3516 list->rlst_nid_array[i] = ni->ni_nid;
3518 if (i >= LNET_MAX_SHOW_NUM_NID)
3521 lnet_net_unlock(LNET_LOCK_EX);
3522 list->rlst_num_nids = i;
3528 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3530 struct lnet_peer_ni *lpni;
3533 lnet_net_lock(LNET_LOCK_EX);
3534 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3535 list->rlst_nid_array[i] = lpni->lpni_nid;
3537 if (i >= LNET_MAX_SHOW_NUM_NID)
3540 lnet_net_unlock(LNET_LOCK_EX);
3541 list->rlst_num_nids = i;
3547 * LNet ioctl handler.
3551 LNetCtl(unsigned int cmd, void *arg)
3553 struct libcfs_ioctl_data *data = arg;
3554 struct lnet_ioctl_config_data *config;
3555 struct lnet_process_id id = {0};
3559 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3560 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3563 case IOC_LIBCFS_GET_NI:
3564 rc = LNetGetId(data->ioc_count, &id);
3565 data->ioc_nid = id.nid;
3568 case IOC_LIBCFS_FAIL_NID:
3569 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3571 case IOC_LIBCFS_ADD_ROUTE: {
3572 /* default router sensitivity to 1 */
3573 unsigned int sensitivity = 1;
3576 if (config->cfg_hdr.ioc_len < sizeof(*config))
3579 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3581 config->cfg_config_u.cfg_route.rtr_sensitivity;
3584 mutex_lock(&the_lnet.ln_api_mutex);
3585 rc = lnet_add_route(config->cfg_net,
3586 config->cfg_config_u.cfg_route.rtr_hop,
3588 config->cfg_config_u.cfg_route.
3589 rtr_priority, sensitivity);
3590 mutex_unlock(&the_lnet.ln_api_mutex);
3594 case IOC_LIBCFS_DEL_ROUTE:
3597 if (config->cfg_hdr.ioc_len < sizeof(*config))
3600 mutex_lock(&the_lnet.ln_api_mutex);
3601 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3602 mutex_unlock(&the_lnet.ln_api_mutex);
3605 case IOC_LIBCFS_GET_ROUTE:
3608 if (config->cfg_hdr.ioc_len < sizeof(*config))
3611 mutex_lock(&the_lnet.ln_api_mutex);
3612 rc = lnet_get_route(config->cfg_count,
3614 &config->cfg_config_u.cfg_route.rtr_hop,
3616 &config->cfg_config_u.cfg_route.rtr_flags,
3617 &config->cfg_config_u.cfg_route.
3619 &config->cfg_config_u.cfg_route.
3621 mutex_unlock(&the_lnet.ln_api_mutex);
3624 case IOC_LIBCFS_GET_LOCAL_NI: {
3625 struct lnet_ioctl_config_ni *cfg_ni;
3626 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3627 struct lnet_ioctl_element_stats *stats;
3632 /* get the tunables if they are available */
3633 if (cfg_ni->lic_cfg_hdr.ioc_len <
3634 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3637 stats = (struct lnet_ioctl_element_stats *)
3639 tun = (struct lnet_ioctl_config_lnd_tunables *)
3640 (cfg_ni->lic_bulk + sizeof(*stats));
3642 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3645 mutex_lock(&the_lnet.ln_api_mutex);
3646 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3647 mutex_unlock(&the_lnet.ln_api_mutex);
3651 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3652 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3654 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3657 mutex_lock(&the_lnet.ln_api_mutex);
3658 rc = lnet_get_ni_stats(msg_stats);
3659 mutex_unlock(&the_lnet.ln_api_mutex);
3664 case IOC_LIBCFS_GET_NET: {
3665 size_t total = sizeof(*config) +
3666 sizeof(struct lnet_ioctl_net_config);
3669 if (config->cfg_hdr.ioc_len < total)
3672 mutex_lock(&the_lnet.ln_api_mutex);
3673 rc = lnet_get_net_config(config);
3674 mutex_unlock(&the_lnet.ln_api_mutex);
3678 case IOC_LIBCFS_GET_LNET_STATS:
3680 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3682 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3685 mutex_lock(&the_lnet.ln_api_mutex);
3686 lnet_counters_get(&lnet_stats->st_cntrs);
3687 mutex_unlock(&the_lnet.ln_api_mutex);
3691 case IOC_LIBCFS_CONFIG_RTR:
3694 if (config->cfg_hdr.ioc_len < sizeof(*config))
3697 mutex_lock(&the_lnet.ln_api_mutex);
3698 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3699 rc = lnet_rtrpools_enable();
3700 mutex_unlock(&the_lnet.ln_api_mutex);
3703 lnet_rtrpools_disable();
3704 mutex_unlock(&the_lnet.ln_api_mutex);
3707 case IOC_LIBCFS_ADD_BUF:
3710 if (config->cfg_hdr.ioc_len < sizeof(*config))
3713 mutex_lock(&the_lnet.ln_api_mutex);
3714 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3716 config->cfg_config_u.cfg_buffers.
3718 config->cfg_config_u.cfg_buffers.
3720 mutex_unlock(&the_lnet.ln_api_mutex);
3723 case IOC_LIBCFS_SET_NUMA_RANGE: {
3724 struct lnet_ioctl_set_value *numa;
3726 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3728 lnet_net_lock(LNET_LOCK_EX);
3729 lnet_numa_range = numa->sv_value;
3730 lnet_net_unlock(LNET_LOCK_EX);
3734 case IOC_LIBCFS_GET_NUMA_RANGE: {
3735 struct lnet_ioctl_set_value *numa;
3737 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3739 numa->sv_value = lnet_numa_range;
3743 case IOC_LIBCFS_GET_BUF: {
3744 struct lnet_ioctl_pool_cfg *pool_cfg;
3745 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3749 if (config->cfg_hdr.ioc_len < total)
3752 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3754 mutex_lock(&the_lnet.ln_api_mutex);
3755 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3756 mutex_unlock(&the_lnet.ln_api_mutex);
3760 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3761 struct lnet_ioctl_local_ni_hstats *stats = arg;
3763 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3766 mutex_lock(&the_lnet.ln_api_mutex);
3767 rc = lnet_get_local_ni_hstats(stats);
3768 mutex_unlock(&the_lnet.ln_api_mutex);
3773 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3774 struct lnet_ioctl_recovery_list *list = arg;
3775 if (list->rlst_hdr.ioc_len < sizeof(*list))
3778 mutex_lock(&the_lnet.ln_api_mutex);
3779 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3780 rc = lnet_get_local_ni_recovery_list(list);
3782 rc = lnet_get_peer_ni_recovery_list(list);
3783 mutex_unlock(&the_lnet.ln_api_mutex);
3787 case IOC_LIBCFS_ADD_PEER_NI: {
3788 struct lnet_ioctl_peer_cfg *cfg = arg;
3790 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3793 mutex_lock(&the_lnet.ln_api_mutex);
3794 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3797 mutex_unlock(&the_lnet.ln_api_mutex);
3801 case IOC_LIBCFS_DEL_PEER_NI: {
3802 struct lnet_ioctl_peer_cfg *cfg = arg;
3804 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3807 mutex_lock(&the_lnet.ln_api_mutex);
3808 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3809 cfg->prcfg_cfg_nid);
3810 mutex_unlock(&the_lnet.ln_api_mutex);
3814 case IOC_LIBCFS_GET_PEER_INFO: {
3815 struct lnet_ioctl_peer *peer_info = arg;
3817 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3820 mutex_lock(&the_lnet.ln_api_mutex);
3821 rc = lnet_get_peer_ni_info(
3822 peer_info->pr_count,
3824 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3825 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3826 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3827 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3828 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3829 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3830 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3831 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3832 mutex_unlock(&the_lnet.ln_api_mutex);
3836 case IOC_LIBCFS_GET_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_get_peer_info(cfg,
3844 (void __user *)cfg->prcfg_bulk);
3845 mutex_unlock(&the_lnet.ln_api_mutex);
3849 case IOC_LIBCFS_GET_PEER_LIST: {
3850 struct lnet_ioctl_peer_cfg *cfg = arg;
3852 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3855 mutex_lock(&the_lnet.ln_api_mutex);
3856 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3857 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3858 mutex_unlock(&the_lnet.ln_api_mutex);
3862 case IOC_LIBCFS_SET_HEALHV: {
3863 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3865 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3867 if (cfg->rh_value < 0 ||
3868 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3869 value = LNET_MAX_HEALTH_VALUE;
3871 value = cfg->rh_value;
3872 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3873 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3874 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3875 mutex_lock(&the_lnet.ln_api_mutex);
3876 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3877 lnet_ni_set_healthv(cfg->rh_nid, value,
3880 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3882 mutex_unlock(&the_lnet.ln_api_mutex);
3886 case IOC_LIBCFS_NOTIFY_ROUTER: {
3887 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3889 /* The deadline passed in by the user should be some time in
3890 * seconds in the future since the UNIX epoch. We have to map
3891 * that deadline to the wall clock.
3893 deadline += ktime_get_seconds();
3894 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3898 case IOC_LIBCFS_LNET_DIST:
3899 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3900 if (rc < 0 && rc != -EHOSTUNREACH)
3903 data->ioc_u32[0] = rc;
3906 case IOC_LIBCFS_TESTPROTOCOMPAT:
3907 the_lnet.ln_testprotocompat = data->ioc_flags;
3910 case IOC_LIBCFS_LNET_FAULT:
3911 return lnet_fault_ctl(data->ioc_flags, data);
3913 case IOC_LIBCFS_PING: {
3914 signed long timeout;
3916 id.nid = data->ioc_nid;
3917 id.pid = data->ioc_u32[0];
3919 /* If timeout is negative then set default of 3 minutes */
3920 if (((s32)data->ioc_u32[1] <= 0) ||
3921 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3922 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3924 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3926 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3927 data->ioc_plen1 / sizeof(struct lnet_process_id));
3932 data->ioc_count = rc;
3936 case IOC_LIBCFS_PING_PEER: {
3937 struct lnet_ioctl_ping_data *ping = arg;
3938 struct lnet_peer *lp;
3939 signed long timeout;
3941 /* If timeout is negative then set default of 3 minutes */
3942 if (((s32)ping->op_param) <= 0 ||
3943 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3944 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3946 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3948 rc = lnet_ping(ping->ping_id, timeout,
3954 mutex_lock(&the_lnet.ln_api_mutex);
3955 lp = lnet_find_peer(ping->ping_id.nid);
3957 ping->ping_id.nid = lp->lp_primary_nid;
3958 ping->mr_info = lnet_peer_is_multi_rail(lp);
3959 lnet_peer_decref_locked(lp);
3961 mutex_unlock(&the_lnet.ln_api_mutex);
3963 ping->ping_count = rc;
3967 case IOC_LIBCFS_DISCOVER: {
3968 struct lnet_ioctl_ping_data *discover = arg;
3969 struct lnet_peer *lp;
3971 rc = lnet_discover(discover->ping_id, discover->op_param,
3973 discover->ping_count);
3977 mutex_lock(&the_lnet.ln_api_mutex);
3978 lp = lnet_find_peer(discover->ping_id.nid);
3980 discover->ping_id.nid = lp->lp_primary_nid;
3981 discover->mr_info = lnet_peer_is_multi_rail(lp);
3982 lnet_peer_decref_locked(lp);
3984 mutex_unlock(&the_lnet.ln_api_mutex);
3986 discover->ping_count = rc;
3991 ni = lnet_net2ni_addref(data->ioc_net);
3995 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
3998 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4005 EXPORT_SYMBOL(LNetCtl);
4007 void LNetDebugPeer(struct lnet_process_id id)
4009 lnet_debug_peer(id.nid);
4011 EXPORT_SYMBOL(LNetDebugPeer);
4014 * Determine if the specified peer \a nid is on the local node.
4016 * \param nid peer nid to check
4018 * \retval true If peer NID is on the local node.
4019 * \retval false If peer NID is not on the local node.
4021 bool LNetIsPeerLocal(lnet_nid_t nid)
4023 struct lnet_net *net;
4027 cpt = lnet_net_lock_current();
4028 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4029 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4030 if (ni->ni_nid == nid) {
4031 lnet_net_unlock(cpt);
4036 lnet_net_unlock(cpt);
4040 EXPORT_SYMBOL(LNetIsPeerLocal);
4043 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4044 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4046 * \param index Index of the interface to look up.
4047 * \param id On successful return, this location will hold the
4048 * struct lnet_process_id ID of the interface.
4050 * \retval 0 If an interface exists at \a index.
4051 * \retval -ENOENT If no interface has been found.
4054 LNetGetId(unsigned int index, struct lnet_process_id *id)
4057 struct lnet_net *net;
4061 LASSERT(the_lnet.ln_refcount > 0);
4063 cpt = lnet_net_lock_current();
4065 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4066 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4070 id->nid = ni->ni_nid;
4071 id->pid = the_lnet.ln_pid;
4077 lnet_net_unlock(cpt);
4080 EXPORT_SYMBOL(LNetGetId);
4082 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4083 struct lnet_process_id __user *ids, int n_ids)
4086 struct lnet_handle_md mdh;
4087 struct lnet_event event;
4088 struct lnet_md md = { NULL };
4092 const signed long a_long_time = cfs_time_seconds(60);
4093 struct lnet_ping_buffer *pbuf;
4094 struct lnet_process_id tmpid;
4101 /* n_ids limit is arbitrary */
4102 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4106 * if the user buffer has more space than the lnet_interfaces_max
4107 * then only fill it up to lnet_interfaces_max
4109 if (n_ids > lnet_interfaces_max)
4110 n_ids = lnet_interfaces_max;
4112 if (id.pid == LNET_PID_ANY)
4113 id.pid = LNET_PID_LUSTRE;
4115 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4119 /* NB 2 events max (including any unlink event) */
4120 eq = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE);
4123 CERROR("Can't allocate EQ: %d\n", rc);
4124 goto fail_ping_buffer_decref;
4127 /* initialize md content */
4128 md.start = &pbuf->pb_info;
4129 md.length = LNET_PING_INFO_SIZE(n_ids);
4130 md.threshold = 2; /* GET/REPLY */
4132 md.options = LNET_MD_TRUNCATE;
4136 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4138 CERROR("Can't bind MD: %d\n", rc);
4142 rc = LNetGet(LNET_NID_ANY, mdh, id,
4143 LNET_RESERVED_PORTAL,
4144 LNET_PROTO_PING_MATCHBITS, 0, false);
4147 /* Don't CERROR; this could be deliberate! */
4148 rc2 = LNetMDUnlink(mdh);
4151 /* NB must wait for the UNLINK event below... */
4153 timeout = a_long_time;
4157 /* MUST block for unlink to complete */
4162 sigprocmask(SIG_SETMASK, &set, &blocked);
4165 rc2 = LNetEQPoll(&eq, 1, timeout, &event, &which);
4168 cfs_restore_sigs(blocked);
4170 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4171 (rc2 <= 0) ? -1 : event.type,
4172 (rc2 <= 0) ? -1 : event.status,
4173 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4175 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
4177 if (rc2 <= 0 || event.status != 0) {
4178 /* timeout or error */
4179 if (!replied && rc == 0)
4180 rc = (rc2 < 0) ? rc2 :
4181 (rc2 == 0) ? -ETIMEDOUT :
4185 /* Ensure completion in finite time... */
4187 /* No assertion (racing with network) */
4189 timeout = a_long_time;
4190 } else if (rc2 == 0) {
4191 /* timed out waiting for unlink */
4192 CWARN("ping %s: late network completion\n",
4195 } else if (event.type == LNET_EVENT_REPLY) {
4199 } while (rc2 <= 0 || !event.unlinked);
4203 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4210 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4212 rc = -EPROTO; /* if I can't parse... */
4215 CERROR("%s: ping info too short %d\n",
4216 libcfs_id2str(id), nob);
4220 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4221 lnet_swap_pinginfo(pbuf);
4222 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4223 CERROR("%s: Unexpected magic %08x\n",
4224 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4228 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4229 CERROR("%s: ping w/o NI status: 0x%x\n",
4230 libcfs_id2str(id), pbuf->pb_info.pi_features);
4234 if (nob < LNET_PING_INFO_SIZE(0)) {
4235 CERROR("%s: Short reply %d(%d min)\n",
4237 nob, (int)LNET_PING_INFO_SIZE(0));
4241 if (pbuf->pb_info.pi_nnis < n_ids)
4242 n_ids = pbuf->pb_info.pi_nnis;
4244 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4245 CERROR("%s: Short reply %d(%d expected)\n",
4247 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4251 rc = -EFAULT; /* if I segv in copy_to_user()... */
4253 memset(&tmpid, 0, sizeof(tmpid));
4254 for (i = 0; i < n_ids; i++) {
4255 tmpid.pid = pbuf->pb_info.pi_pid;
4256 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4257 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4260 rc = pbuf->pb_info.pi_nnis;
4263 rc2 = LNetEQFree(eq);
4265 CERROR("rc2 %d\n", rc2);
4268 fail_ping_buffer_decref:
4269 lnet_ping_buffer_decref(pbuf);
4274 lnet_discover(struct lnet_process_id id, __u32 force,
4275 struct lnet_process_id __user *ids, int n_ids)
4277 struct lnet_peer_ni *lpni;
4278 struct lnet_peer_ni *p;
4279 struct lnet_peer *lp;
4280 struct lnet_process_id *buf;
4284 int max_intf = lnet_interfaces_max;
4288 id.nid == LNET_NID_ANY)
4291 if (id.pid == LNET_PID_ANY)
4292 id.pid = LNET_PID_LUSTRE;
4295 * if the user buffer has more space than the max_intf
4296 * then only fill it up to max_intf
4298 if (n_ids > max_intf)
4301 buf_size = n_ids * sizeof(*buf);
4303 LIBCFS_ALLOC(buf, buf_size);
4307 cpt = lnet_net_lock_current();
4308 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4315 * Clearing the NIDS_UPTODATE flag ensures the peer will
4316 * be discovered, provided discovery has not been disabled.
4318 lp = lpni->lpni_peer_net->lpn_peer;
4319 spin_lock(&lp->lp_lock);
4320 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4321 /* If the force flag is set, force a PING and PUSH as well. */
4323 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4324 spin_unlock(&lp->lp_lock);
4325 rc = lnet_discover_peer_locked(lpni, cpt, true);
4329 /* Peer may have changed. */
4330 lp = lpni->lpni_peer_net->lpn_peer;
4331 if (lp->lp_nnis < n_ids)
4332 n_ids = lp->lp_nnis;
4336 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4337 buf[i].pid = id.pid;
4338 buf[i].nid = p->lpni_nid;
4343 lnet_net_unlock(cpt);
4346 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4352 lnet_peer_ni_decref_locked(lpni);
4354 lnet_net_unlock(cpt);
4356 LIBCFS_FREE(buf, buf_size);
4362 * Retrieve peer discovery status.
4364 * \retval 1 if lnet_peer_discovery_disabled is 0
4365 * \retval 0 if lnet_peer_discovery_disabled is 1
4368 LNetGetPeerDiscoveryStatus(void)
4370 return !lnet_peer_discovery_disabled;
4372 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);