4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_LNET
35 #include <linux/ctype.h>
36 #include <linux/log2.h>
37 #include <linux/ktime.h>
38 #include <linux/moduleparam.h>
39 #include <linux/uaccess.h>
40 #ifdef HAVE_SCHED_HEADERS
41 #include <linux/sched/signal.h>
44 #include <lnet/lib-lnet.h>
46 #define D_LNI D_CONSOLE
49 * initialize ln_api_mutex statically, since it needs to be used in
50 * discovery_set callback. That module parameter callback can be called
51 * before module init completes. The mutex needs to be ready for use then.
53 struct lnet the_lnet = {
54 .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
55 }; /* THE state of the network */
56 EXPORT_SYMBOL(the_lnet);
58 static char *ip2nets = "";
59 module_param(ip2nets, charp, 0444);
60 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
62 static char *networks = "";
63 module_param(networks, charp, 0444);
64 MODULE_PARM_DESC(networks, "local networks");
66 static char *routes = "";
67 module_param(routes, charp, 0444);
68 MODULE_PARM_DESC(routes, "routes to non-local networks");
70 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
71 module_param(rnet_htable_size, int, 0444);
72 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
74 static int use_tcp_bonding = false;
75 module_param(use_tcp_bonding, int, 0444);
76 MODULE_PARM_DESC(use_tcp_bonding,
77 "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
79 unsigned int lnet_numa_range = 0;
80 module_param(lnet_numa_range, uint, 0444);
81 MODULE_PARM_DESC(lnet_numa_range,
82 "NUMA range to consider during Multi-Rail selection");
85 * lnet_health_sensitivity determines by how much we decrement the health
86 * value on sending error. The value defaults to 100, which means health
87 * interface health is decremented by 100 points every failure.
89 unsigned int lnet_health_sensitivity = 100;
90 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
91 #ifdef HAVE_KERNEL_PARAM_OPS
92 static struct kernel_param_ops param_ops_health_sensitivity = {
93 .set = sensitivity_set,
96 #define param_check_health_sensitivity(name, p) \
97 __param_check(name, p, int)
98 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
100 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
101 &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
103 MODULE_PARM_DESC(lnet_health_sensitivity,
104 "Value to decrement the health value by on error");
107 * lnet_recovery_interval determines how often we should perform recovery
108 * on unhealthy interfaces.
110 unsigned int lnet_recovery_interval = 1;
111 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
112 #ifdef HAVE_KERNEL_PARAM_OPS
113 static struct kernel_param_ops param_ops_recovery_interval = {
114 .set = recovery_interval_set,
115 .get = param_get_int,
117 #define param_check_recovery_interval(name, p) \
118 __param_check(name, p, int)
119 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
121 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
122 &lnet_recovery_interval, S_IRUGO|S_IWUSR);
124 MODULE_PARM_DESC(lnet_recovery_interval,
125 "Interval to recover unhealthy interfaces in seconds");
127 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
128 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
130 static struct kernel_param_ops param_ops_interfaces_max = {
132 .get = param_get_int,
135 #define param_check_interfaces_max(name, p) \
136 __param_check(name, p, int)
138 #ifdef HAVE_KERNEL_PARAM_OPS
139 module_param(lnet_interfaces_max, interfaces_max, 0644);
141 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
142 ¶m_ops_interfaces_max, 0644);
144 MODULE_PARM_DESC(lnet_interfaces_max,
145 "Maximum number of interfaces in a node.");
147 unsigned lnet_peer_discovery_disabled = 0;
148 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
150 static struct kernel_param_ops param_ops_discovery_disabled = {
151 .set = discovery_set,
152 .get = param_get_int,
155 #define param_check_discovery_disabled(name, p) \
156 __param_check(name, p, int)
157 #ifdef HAVE_KERNEL_PARAM_OPS
158 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
160 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
161 ¶m_ops_discovery_disabled, 0644);
163 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
164 "Set to 1 to disable peer discovery on this node.");
166 unsigned int lnet_drop_asym_route;
167 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
169 static struct kernel_param_ops param_ops_drop_asym_route = {
170 .set = drop_asym_route_set,
171 .get = param_get_int,
174 #define param_check_drop_asym_route(name, p) \
175 __param_check(name, p, int)
176 #ifdef HAVE_KERNEL_PARAM_OPS
177 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
179 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
180 ¶m_ops_drop_asym_route, 0644);
182 MODULE_PARM_DESC(lnet_drop_asym_route,
183 "Set to 1 to drop asymmetrical route messages.");
185 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
186 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 50
188 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
189 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
190 #ifdef HAVE_KERNEL_PARAM_OPS
191 static struct kernel_param_ops param_ops_transaction_timeout = {
192 .set = transaction_to_set,
193 .get = param_get_int,
196 #define param_check_transaction_timeout(name, p) \
197 __param_check(name, p, int)
198 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
200 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
201 &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
203 MODULE_PARM_DESC(lnet_transaction_timeout,
204 "Maximum number of seconds to wait for a peer response.");
206 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 2
207 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
208 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
209 #ifdef HAVE_KERNEL_PARAM_OPS
210 static struct kernel_param_ops param_ops_retry_count = {
211 .set = retry_count_set,
212 .get = param_get_int,
215 #define param_check_retry_count(name, p) \
216 __param_check(name, p, int)
217 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
219 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
220 &lnet_retry_count, S_IRUGO|S_IWUSR);
222 MODULE_PARM_DESC(lnet_retry_count,
223 "Maximum number of times to retry transmitting a message");
226 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
227 unsigned int lnet_current_net_count;
230 * This sequence number keeps track of how many times DLC was used to
231 * update the local NIs. It is incremented when a NI is added or
232 * removed and checked when sending a message to determine if there is
233 * a need to re-run the selection algorithm. See lnet_select_pathway()
234 * for more details on its usage.
236 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
238 static int lnet_ping(struct lnet_process_id id, signed long timeout,
239 struct lnet_process_id __user *ids, int n_ids);
241 static int lnet_discover(struct lnet_process_id id, __u32 force,
242 struct lnet_process_id __user *ids, int n_ids);
245 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
248 unsigned *sensitivity = (unsigned *)kp->arg;
251 rc = kstrtoul(val, 0, &value);
253 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
258 * The purpose of locking the api_mutex here is to ensure that
259 * the correct value ends up stored properly.
261 mutex_lock(&the_lnet.ln_api_mutex);
263 if (value > LNET_MAX_HEALTH_VALUE) {
264 mutex_unlock(&the_lnet.ln_api_mutex);
265 CERROR("Invalid health value. Maximum: %d value = %lu\n",
266 LNET_MAX_HEALTH_VALUE, value);
271 * if we're turning on health then use the health timeout
274 if (*sensitivity == 0 && value != 0) {
275 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
276 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
278 * if we're turning off health then use the no health timeout
281 } else if (*sensitivity != 0 && value == 0) {
282 lnet_transaction_timeout =
283 LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
284 lnet_retry_count = 0;
287 *sensitivity = value;
289 mutex_unlock(&the_lnet.ln_api_mutex);
295 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
298 unsigned *interval = (unsigned *)kp->arg;
301 rc = kstrtoul(val, 0, &value);
303 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
308 CERROR("lnet_recovery_interval must be at least 1 second\n");
313 * The purpose of locking the api_mutex here is to ensure that
314 * the correct value ends up stored properly.
316 mutex_lock(&the_lnet.ln_api_mutex);
320 mutex_unlock(&the_lnet.ln_api_mutex);
326 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
329 unsigned *discovery = (unsigned *)kp->arg;
331 struct lnet_ping_buffer *pbuf;
333 rc = kstrtoul(val, 0, &value);
335 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
339 value = (value) ? 1 : 0;
342 * The purpose of locking the api_mutex here is to ensure that
343 * the correct value ends up stored properly.
345 mutex_lock(&the_lnet.ln_api_mutex);
347 if (value == *discovery) {
348 mutex_unlock(&the_lnet.ln_api_mutex);
353 * We still want to set the discovery value even when LNet is not
354 * running. This is the case when LNet is being loaded and we want
355 * the module parameters to take effect. Otherwise if we're
356 * changing the value dynamically, we want to set it after
359 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
361 mutex_unlock(&the_lnet.ln_api_mutex);
365 /* tell peers that discovery setting has changed */
366 lnet_net_lock(LNET_LOCK_EX);
367 pbuf = the_lnet.ln_ping_target;
369 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
371 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
372 lnet_net_unlock(LNET_LOCK_EX);
375 * Always update the peers. This will result in a push to the
376 * peers with the updated capabilities feature mask. The peer can
377 * then take appropriate action to update its representation of
380 * If discovery is already off, turn it on first before pushing
381 * the update. The discovery flag must be on before pushing.
382 * otherwise if the flag is on and we're turning it off then push
383 * first before turning the flag off. In the former case the flag
384 * is being set twice, but I find it's better to do that rather
385 * than have duplicate code in an if/else statement.
387 if (*discovery > 0 && value == 0)
389 lnet_push_update_to_peers(1);
392 mutex_unlock(&the_lnet.ln_api_mutex);
398 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
401 unsigned int *drop_asym_route = (unsigned int *)kp->arg;
404 rc = kstrtoul(val, 0, &value);
406 CERROR("Invalid module parameter value for "
407 "'lnet_drop_asym_route'\n");
412 * The purpose of locking the api_mutex here is to ensure that
413 * the correct value ends up stored properly.
415 mutex_lock(&the_lnet.ln_api_mutex);
417 if (value == *drop_asym_route) {
418 mutex_unlock(&the_lnet.ln_api_mutex);
422 *drop_asym_route = value;
424 mutex_unlock(&the_lnet.ln_api_mutex);
430 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
433 unsigned *transaction_to = (unsigned *)kp->arg;
436 rc = kstrtoul(val, 0, &value);
438 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
443 * The purpose of locking the api_mutex here is to ensure that
444 * the correct value ends up stored properly.
446 mutex_lock(&the_lnet.ln_api_mutex);
448 if (value < lnet_retry_count || value == 0) {
449 mutex_unlock(&the_lnet.ln_api_mutex);
450 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
451 "Has to be greater than lnet_retry_count (%u)\n",
452 value, lnet_retry_count);
456 if (value == *transaction_to) {
457 mutex_unlock(&the_lnet.ln_api_mutex);
461 *transaction_to = value;
462 if (lnet_retry_count == 0)
463 lnet_lnd_timeout = value;
465 lnet_lnd_timeout = value / lnet_retry_count;
467 mutex_unlock(&the_lnet.ln_api_mutex);
473 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
476 unsigned *retry_count = (unsigned *)kp->arg;
479 rc = kstrtoul(val, 0, &value);
481 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
486 * The purpose of locking the api_mutex here is to ensure that
487 * the correct value ends up stored properly.
489 mutex_lock(&the_lnet.ln_api_mutex);
491 if (lnet_health_sensitivity == 0) {
492 mutex_unlock(&the_lnet.ln_api_mutex);
493 CERROR("Can not set retry_count when health feature is turned off\n");
497 if (value > lnet_transaction_timeout) {
498 mutex_unlock(&the_lnet.ln_api_mutex);
499 CERROR("Invalid value for lnet_retry_count (%lu). "
500 "Has to be smaller than lnet_transaction_timeout (%u)\n",
501 value, lnet_transaction_timeout);
505 *retry_count = value;
508 lnet_lnd_timeout = lnet_transaction_timeout;
510 lnet_lnd_timeout = lnet_transaction_timeout / value;
512 mutex_unlock(&the_lnet.ln_api_mutex);
518 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
522 rc = kstrtoint(val, 0, &value);
524 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
528 if (value < LNET_INTERFACES_MIN) {
529 CWARN("max interfaces provided are too small, setting to %d\n",
530 LNET_INTERFACES_MAX_DEFAULT);
531 value = LNET_INTERFACES_MAX_DEFAULT;
534 *(int *)kp->arg = value;
540 lnet_get_routes(void)
546 lnet_get_networks(void)
551 if (*networks != 0 && *ip2nets != 0) {
552 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
553 "'ip2nets' but not both at once\n");
558 rc = lnet_parse_ip2nets(&nets, ip2nets);
559 return (rc == 0) ? nets : NULL;
569 lnet_init_locks(void)
571 spin_lock_init(&the_lnet.ln_eq_wait_lock);
572 spin_lock_init(&the_lnet.ln_msg_resend_lock);
573 init_completion(&the_lnet.ln_mt_wait_complete);
574 mutex_init(&the_lnet.ln_lnd_mutex);
577 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
578 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
580 struct kmem_cache *lnet_rspt_cachep; /* response tracker cache */
581 struct kmem_cache *lnet_msg_cachep;
584 lnet_slab_setup(void)
586 /* create specific kmem_cache for MEs and small MDs (i.e., originally
587 * allocated in <size-xxx> kmem_cache).
589 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
591 if (!lnet_mes_cachep)
594 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
595 LNET_SMALL_MD_SIZE, 0, 0,
597 if (!lnet_small_mds_cachep)
600 lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
602 if (!lnet_rspt_cachep)
605 lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
607 if (!lnet_msg_cachep)
614 lnet_slab_cleanup(void)
616 if (lnet_msg_cachep) {
617 kmem_cache_destroy(lnet_msg_cachep);
618 lnet_msg_cachep = NULL;
622 if (lnet_rspt_cachep) {
623 kmem_cache_destroy(lnet_rspt_cachep);
624 lnet_rspt_cachep = NULL;
627 if (lnet_small_mds_cachep) {
628 kmem_cache_destroy(lnet_small_mds_cachep);
629 lnet_small_mds_cachep = NULL;
632 if (lnet_mes_cachep) {
633 kmem_cache_destroy(lnet_mes_cachep);
634 lnet_mes_cachep = NULL;
639 lnet_create_remote_nets_table(void)
642 struct list_head *hash;
644 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
645 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
646 CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
648 CERROR("Failed to create remote nets hash table\n");
652 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
653 INIT_LIST_HEAD(&hash[i]);
654 the_lnet.ln_remote_nets_hash = hash;
659 lnet_destroy_remote_nets_table(void)
663 if (the_lnet.ln_remote_nets_hash == NULL)
666 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
667 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
669 CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
670 LNET_REMOTE_NETS_HASH_SIZE);
671 the_lnet.ln_remote_nets_hash = NULL;
675 lnet_destroy_locks(void)
677 if (the_lnet.ln_res_lock != NULL) {
678 cfs_percpt_lock_free(the_lnet.ln_res_lock);
679 the_lnet.ln_res_lock = NULL;
682 if (the_lnet.ln_net_lock != NULL) {
683 cfs_percpt_lock_free(the_lnet.ln_net_lock);
684 the_lnet.ln_net_lock = NULL;
689 lnet_create_locks(void)
693 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
694 if (the_lnet.ln_res_lock == NULL)
697 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
698 if (the_lnet.ln_net_lock == NULL)
704 lnet_destroy_locks();
708 static void lnet_assert_wire_constants(void)
710 /* Wire protocol assertions generated by 'wirecheck'
711 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
712 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
713 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
716 BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
717 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
718 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
719 BUILD_BUG_ON(LNET_MSG_ACK != 0);
720 BUILD_BUG_ON(LNET_MSG_PUT != 1);
721 BUILD_BUG_ON(LNET_MSG_GET != 2);
722 BUILD_BUG_ON(LNET_MSG_REPLY != 3);
723 BUILD_BUG_ON(LNET_MSG_HELLO != 4);
725 /* Checks for struct lnet_handle_wire */
726 BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
727 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
728 wh_interface_cookie) != 0);
729 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
730 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
731 wh_object_cookie) != 8);
732 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
734 /* Checks for struct struct lnet_magicversion */
735 BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
736 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
737 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
738 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
739 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
740 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
741 version_minor) != 6);
742 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
744 /* Checks for struct struct lnet_hdr */
745 BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
746 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
747 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
748 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
749 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
750 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
751 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
752 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
753 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
754 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
755 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
756 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
757 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
758 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
759 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
762 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
763 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
764 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
765 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
766 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
767 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
770 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
771 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
772 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
773 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
774 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
775 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
776 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
777 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
778 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
779 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
782 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
783 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
784 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
785 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
786 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
787 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
788 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
789 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
790 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
791 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
794 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
795 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
798 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
799 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
800 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
801 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
803 /* Checks for struct lnet_ni_status and related constants */
804 BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
805 BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
806 BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
808 /* Checks for struct lnet_ni_status */
809 BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
810 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
811 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
812 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
813 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
814 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
815 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
817 /* Checks for struct lnet_ping_info and related constants */
818 BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
819 BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
820 BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
821 BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
822 BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
823 BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
824 BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
825 BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
827 /* Checks for struct lnet_ping_info */
828 BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
829 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
830 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
831 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
832 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
833 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
834 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
835 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
836 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
837 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
838 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
841 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
843 const struct lnet_lnd *lnd;
845 /* holding lnd mutex */
846 if (type >= NUM_LNDS)
848 lnd = the_lnet.ln_lnds[type];
849 LASSERT(!lnd || lnd->lnd_type == type);
855 lnet_get_lnd_timeout(void)
857 return lnet_lnd_timeout;
859 EXPORT_SYMBOL(lnet_get_lnd_timeout);
862 lnet_register_lnd(const struct lnet_lnd *lnd)
864 mutex_lock(&the_lnet.ln_lnd_mutex);
866 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
867 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
869 the_lnet.ln_lnds[lnd->lnd_type] = lnd;
871 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
873 mutex_unlock(&the_lnet.ln_lnd_mutex);
875 EXPORT_SYMBOL(lnet_register_lnd);
878 lnet_unregister_lnd(const struct lnet_lnd *lnd)
880 mutex_lock(&the_lnet.ln_lnd_mutex);
882 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
884 the_lnet.ln_lnds[lnd->lnd_type] = NULL;
885 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
887 mutex_unlock(&the_lnet.ln_lnd_mutex);
889 EXPORT_SYMBOL(lnet_unregister_lnd);
892 lnet_counters_get_common(struct lnet_counters_common *common)
894 struct lnet_counters *ctr;
897 memset(common, 0, sizeof(*common));
899 lnet_net_lock(LNET_LOCK_EX);
901 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
902 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
903 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
904 common->lcc_errors += ctr->lct_common.lcc_errors;
905 common->lcc_send_count += ctr->lct_common.lcc_send_count;
906 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
907 common->lcc_route_count += ctr->lct_common.lcc_route_count;
908 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
909 common->lcc_send_length += ctr->lct_common.lcc_send_length;
910 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
911 common->lcc_route_length += ctr->lct_common.lcc_route_length;
912 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
914 lnet_net_unlock(LNET_LOCK_EX);
916 EXPORT_SYMBOL(lnet_counters_get_common);
919 lnet_counters_get(struct lnet_counters *counters)
921 struct lnet_counters *ctr;
922 struct lnet_counters_health *health = &counters->lct_health;
925 memset(counters, 0, sizeof(*counters));
927 lnet_counters_get_common(&counters->lct_common);
929 lnet_net_lock(LNET_LOCK_EX);
931 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
932 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
933 health->lch_resend_count += ctr->lct_health.lch_resend_count;
934 health->lch_response_timeout_count +=
935 ctr->lct_health.lch_response_timeout_count;
936 health->lch_local_interrupt_count +=
937 ctr->lct_health.lch_local_interrupt_count;
938 health->lch_local_dropped_count +=
939 ctr->lct_health.lch_local_dropped_count;
940 health->lch_local_aborted_count +=
941 ctr->lct_health.lch_local_aborted_count;
942 health->lch_local_no_route_count +=
943 ctr->lct_health.lch_local_no_route_count;
944 health->lch_local_timeout_count +=
945 ctr->lct_health.lch_local_timeout_count;
946 health->lch_local_error_count +=
947 ctr->lct_health.lch_local_error_count;
948 health->lch_remote_dropped_count +=
949 ctr->lct_health.lch_remote_dropped_count;
950 health->lch_remote_error_count +=
951 ctr->lct_health.lch_remote_error_count;
952 health->lch_remote_timeout_count +=
953 ctr->lct_health.lch_remote_timeout_count;
954 health->lch_network_timeout_count +=
955 ctr->lct_health.lch_network_timeout_count;
957 lnet_net_unlock(LNET_LOCK_EX);
959 EXPORT_SYMBOL(lnet_counters_get);
962 lnet_counters_reset(void)
964 struct lnet_counters *counters;
967 lnet_net_lock(LNET_LOCK_EX);
969 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
970 memset(counters, 0, sizeof(struct lnet_counters));
972 lnet_net_unlock(LNET_LOCK_EX);
976 lnet_res_type2str(int type)
981 case LNET_COOKIE_TYPE_MD:
983 case LNET_COOKIE_TYPE_ME:
985 case LNET_COOKIE_TYPE_EQ:
991 lnet_res_container_cleanup(struct lnet_res_container *rec)
995 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
998 while (!list_empty(&rec->rec_active)) {
999 struct list_head *e = rec->rec_active.next;
1002 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1003 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1005 } else { /* NB: Active MEs should be attached on portals */
1012 /* Found alive MD/ME/EQ, user really should unlink/free
1013 * all of them before finalize LNet, but if someone didn't,
1014 * we have to recycle garbage for him */
1015 CERROR("%d active elements on exit of %s container\n",
1016 count, lnet_res_type2str(rec->rec_type));
1019 if (rec->rec_lh_hash != NULL) {
1020 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1021 rec->rec_lh_hash = NULL;
1024 rec->rec_type = 0; /* mark it as finalized */
1028 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1033 LASSERT(rec->rec_type == 0);
1035 rec->rec_type = type;
1036 INIT_LIST_HEAD(&rec->rec_active);
1038 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1040 /* Arbitrary choice of hash table size */
1041 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1042 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1043 if (rec->rec_lh_hash == NULL) {
1048 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1049 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1054 CERROR("Failed to setup %s resource container\n",
1055 lnet_res_type2str(type));
1056 lnet_res_container_cleanup(rec);
1061 lnet_res_containers_destroy(struct lnet_res_container **recs)
1063 struct lnet_res_container *rec;
1066 cfs_percpt_for_each(rec, i, recs)
1067 lnet_res_container_cleanup(rec);
1069 cfs_percpt_free(recs);
1072 static struct lnet_res_container **
1073 lnet_res_containers_create(int type)
1075 struct lnet_res_container **recs;
1076 struct lnet_res_container *rec;
1080 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1082 CERROR("Failed to allocate %s resource containers\n",
1083 lnet_res_type2str(type));
1087 cfs_percpt_for_each(rec, i, recs) {
1088 rc = lnet_res_container_setup(rec, i, type);
1090 lnet_res_containers_destroy(recs);
1098 struct lnet_libhandle *
1099 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1101 /* ALWAYS called with lnet_res_lock held */
1102 struct list_head *head;
1103 struct lnet_libhandle *lh;
1106 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1109 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1110 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1112 list_for_each_entry(lh, head, lh_hash_chain) {
1113 if (lh->lh_cookie == cookie)
1121 lnet_res_lh_initialize(struct lnet_res_container *rec,
1122 struct lnet_libhandle *lh)
1124 /* ALWAYS called with lnet_res_lock held */
1125 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1128 lh->lh_cookie = rec->rec_lh_cookie;
1129 rec->rec_lh_cookie += 1 << ibits;
1131 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1133 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1137 lnet_create_array_of_queues(void)
1139 struct list_head **qs;
1140 struct list_head *q;
1143 qs = cfs_percpt_alloc(lnet_cpt_table(),
1144 sizeof(struct list_head));
1146 CERROR("Failed to allocate queues\n");
1150 cfs_percpt_for_each(q, i, qs)
1156 static int lnet_unprepare(void);
1159 lnet_prepare(lnet_pid_t requested_pid)
1161 /* Prepare to bring up the network */
1162 struct lnet_res_container **recs;
1165 if (requested_pid == LNET_PID_ANY) {
1166 /* Don't instantiate LNET just for me */
1170 LASSERT(the_lnet.ln_refcount == 0);
1172 the_lnet.ln_routing = 0;
1174 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1175 the_lnet.ln_pid = requested_pid;
1177 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1178 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1179 INIT_LIST_HEAD(&the_lnet.ln_nets);
1180 INIT_LIST_HEAD(&the_lnet.ln_routers);
1181 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1182 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1183 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1184 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1185 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1186 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1187 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1188 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1189 the_lnet.ln_mt_eq = NULL;
1190 init_completion(&the_lnet.ln_started);
1192 rc = lnet_slab_setup();
1196 rc = lnet_create_remote_nets_table();
1201 * NB the interface cookie in wire handles guards against delayed
1202 * replies and ACKs appearing valid after reboot.
1204 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1206 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1207 sizeof(struct lnet_counters));
1208 if (the_lnet.ln_counters == NULL) {
1209 CERROR("Failed to allocate counters for LNet\n");
1214 rc = lnet_peer_tables_create();
1218 rc = lnet_msg_containers_create();
1222 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1223 LNET_COOKIE_TYPE_EQ);
1227 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1233 the_lnet.ln_md_containers = recs;
1235 rc = lnet_portals_create();
1237 CERROR("Failed to create portals for LNet: %d\n", rc);
1241 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1242 if (!the_lnet.ln_mt_zombie_rstqs) {
1255 lnet_unprepare (void)
1257 /* NB no LNET_LOCK since this is the last reference. All LND instances
1258 * have shut down already, so it is safe to unlink and free all
1259 * descriptors, even those that appear committed to a network op (eg MD
1260 * with non-zero pending count) */
1262 lnet_fail_nid(LNET_NID_ANY, 0);
1264 LASSERT(the_lnet.ln_refcount == 0);
1265 LASSERT(list_empty(&the_lnet.ln_test_peers));
1266 LASSERT(list_empty(&the_lnet.ln_nets));
1268 if (the_lnet.ln_mt_zombie_rstqs) {
1269 lnet_clean_zombie_rstqs();
1270 the_lnet.ln_mt_zombie_rstqs = NULL;
1273 if (the_lnet.ln_mt_eq) {
1274 LNetEQFree(the_lnet.ln_mt_eq);
1275 the_lnet.ln_mt_eq = NULL;
1278 lnet_portals_destroy();
1280 if (the_lnet.ln_md_containers != NULL) {
1281 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1282 the_lnet.ln_md_containers = NULL;
1285 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1287 lnet_msg_containers_destroy();
1289 lnet_rtrpools_free(0);
1291 if (the_lnet.ln_counters != NULL) {
1292 cfs_percpt_free(the_lnet.ln_counters);
1293 the_lnet.ln_counters = NULL;
1295 lnet_destroy_remote_nets_table();
1296 lnet_slab_cleanup();
1302 lnet_net2ni_locked(__u32 net_id, int cpt)
1305 struct lnet_net *net;
1307 LASSERT(cpt != LNET_LOCK_EX);
1309 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1310 if (net->net_id == net_id) {
1311 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1321 lnet_net2ni_addref(__u32 net)
1326 ni = lnet_net2ni_locked(net, 0);
1328 lnet_ni_addref_locked(ni, 0);
1333 EXPORT_SYMBOL(lnet_net2ni_addref);
1336 lnet_get_net_locked(__u32 net_id)
1338 struct lnet_net *net;
1340 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1341 if (net->net_id == net_id)
1349 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1354 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1359 val = hash_long(key, LNET_CPT_BITS);
1360 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1364 return (unsigned int)(key + val + (val >> 1)) % number;
1368 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1370 struct lnet_net *net;
1372 /* must called with hold of lnet_net_lock */
1373 if (LNET_CPT_NUMBER == 1)
1374 return 0; /* the only one */
1377 * If NI is provided then use the CPT identified in the NI cpt
1378 * list if one exists. If one doesn't exist, then that NI is
1379 * associated with all CPTs and it follows that the net it belongs
1380 * to is implicitly associated with all CPTs, so just hash the nid
1384 if (ni->ni_cpts != NULL)
1385 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1388 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1391 /* no NI provided so look at the net */
1392 net = lnet_get_net_locked(LNET_NIDNET(nid));
1394 if (net != NULL && net->net_cpts != NULL) {
1395 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1398 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1402 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1407 if (LNET_CPT_NUMBER == 1)
1408 return 0; /* the only one */
1410 cpt = lnet_net_lock_current();
1412 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1414 lnet_net_unlock(cpt);
1418 EXPORT_SYMBOL(lnet_cpt_of_nid);
1421 lnet_islocalnet_locked(__u32 net_id)
1423 struct lnet_net *net;
1426 net = lnet_get_net_locked(net_id);
1428 local = net != NULL;
1434 lnet_islocalnet(__u32 net_id)
1439 cpt = lnet_net_lock_current();
1441 local = lnet_islocalnet_locked(net_id);
1443 lnet_net_unlock(cpt);
1449 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1451 struct lnet_net *net;
1454 LASSERT(cpt != LNET_LOCK_EX);
1456 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1457 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1458 if (ni->ni_nid == nid)
1467 lnet_nid2ni_addref(lnet_nid_t nid)
1472 ni = lnet_nid2ni_locked(nid, 0);
1474 lnet_ni_addref_locked(ni, 0);
1479 EXPORT_SYMBOL(lnet_nid2ni_addref);
1482 lnet_islocalnid(lnet_nid_t nid)
1487 cpt = lnet_net_lock_current();
1488 ni = lnet_nid2ni_locked(nid, cpt);
1489 lnet_net_unlock(cpt);
1495 lnet_count_acceptor_nets(void)
1497 /* Return the # of NIs that need the acceptor. */
1499 struct lnet_net *net;
1502 cpt = lnet_net_lock_current();
1503 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1504 /* all socklnd type networks should have the acceptor
1506 if (net->net_lnd->lnd_accept != NULL)
1510 lnet_net_unlock(cpt);
1515 struct lnet_ping_buffer *
1516 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1518 struct lnet_ping_buffer *pbuf;
1520 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1522 pbuf->pb_nnis = nnis;
1523 pbuf->pb_needs_post = false;
1524 atomic_set(&pbuf->pb_refcnt, 1);
1531 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1533 LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1534 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1537 static struct lnet_ping_buffer *
1538 lnet_ping_target_create(int nnis)
1540 struct lnet_ping_buffer *pbuf;
1542 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1544 CERROR("Can't allocate ping source [%d]\n", nnis);
1548 pbuf->pb_info.pi_nnis = nnis;
1549 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1550 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1551 pbuf->pb_info.pi_features =
1552 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1558 lnet_get_net_ni_count_locked(struct lnet_net *net)
1563 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1570 lnet_get_net_ni_count_pre(struct lnet_net *net)
1575 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1582 lnet_get_ni_count(void)
1585 struct lnet_net *net;
1590 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1591 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1601 lnet_get_net_count(void)
1603 struct lnet_net *net;
1608 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1618 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1620 struct lnet_ni_status *stat;
1624 __swab32s(&pbuf->pb_info.pi_magic);
1625 __swab32s(&pbuf->pb_info.pi_features);
1626 __swab32s(&pbuf->pb_info.pi_pid);
1627 __swab32s(&pbuf->pb_info.pi_nnis);
1628 nnis = pbuf->pb_info.pi_nnis;
1629 if (nnis > pbuf->pb_nnis)
1630 nnis = pbuf->pb_nnis;
1631 for (i = 0; i < nnis; i++) {
1632 stat = &pbuf->pb_info.pi_ni[i];
1633 __swab64s(&stat->ns_nid);
1634 __swab32s(&stat->ns_status);
1639 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1643 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1645 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1647 /* Loopback is guaranteed to be present */
1648 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1650 if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1656 lnet_ping_target_destroy(void)
1658 struct lnet_net *net;
1661 lnet_net_lock(LNET_LOCK_EX);
1663 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1664 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1666 ni->ni_status = NULL;
1671 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1672 the_lnet.ln_ping_target = NULL;
1674 lnet_net_unlock(LNET_LOCK_EX);
1678 lnet_ping_target_event_handler(struct lnet_event *event)
1680 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1682 if (event->unlinked)
1683 lnet_ping_buffer_decref(pbuf);
1687 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1688 struct lnet_handle_md *ping_mdh,
1689 int ni_count, bool set_eq)
1691 struct lnet_process_id id = {
1692 .nid = LNET_NID_ANY,
1696 struct lnet_md md = { NULL };
1700 the_lnet.ln_ping_target_eq =
1701 LNetEQAlloc(lnet_ping_target_event_handler);
1702 if (IS_ERR(the_lnet.ln_ping_target_eq)) {
1703 rc = PTR_ERR(the_lnet.ln_ping_target_eq);
1704 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1709 *ppbuf = lnet_ping_target_create(ni_count);
1710 if (*ppbuf == NULL) {
1715 /* Ping target ME/MD */
1716 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1717 LNET_PROTO_PING_MATCHBITS, 0,
1718 LNET_UNLINK, LNET_INS_AFTER);
1721 CERROR("Can't create ping target ME: %d\n", rc);
1722 goto fail_decref_ping_buffer;
1725 /* initialize md content */
1726 md.start = &(*ppbuf)->pb_info;
1727 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1728 md.threshold = LNET_MD_THRESH_INF;
1730 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1731 LNET_MD_MANAGE_REMOTE;
1732 md.eq_handle = the_lnet.ln_ping_target_eq;
1733 md.user_ptr = *ppbuf;
1735 rc = LNetMDAttach(me, md, LNET_RETAIN, ping_mdh);
1737 CERROR("Can't attach ping target MD: %d\n", rc);
1738 goto fail_unlink_ping_me;
1740 lnet_ping_buffer_addref(*ppbuf);
1744 fail_unlink_ping_me:
1746 fail_decref_ping_buffer:
1747 LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1748 lnet_ping_buffer_decref(*ppbuf);
1752 LNetEQFree(the_lnet.ln_ping_target_eq);
1758 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1759 struct lnet_handle_md *ping_mdh)
1761 LNetMDUnlink(*ping_mdh);
1762 LNetInvalidateMDHandle(ping_mdh);
1764 /* NB the MD could be busy; this just starts the unlink */
1765 wait_var_event_warning(&pbuf->pb_refcnt,
1766 atomic_read(&pbuf->pb_refcnt) <= 1,
1767 "Still waiting for ping data MD to unlink\n");
1771 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1774 struct lnet_net *net;
1775 struct lnet_ni_status *ns;
1780 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1781 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1782 LASSERT(i < pbuf->pb_nnis);
1784 ns = &pbuf->pb_info.pi_ni[i];
1786 ns->ns_nid = ni->ni_nid;
1789 ns->ns_status = (ni->ni_status != NULL) ?
1790 ni->ni_status->ns_status :
1799 * We (ab)use the ns_status of the loopback interface to
1800 * transmit the sequence number. The first interface listed
1801 * must be the loopback interface.
1803 rc = lnet_ping_info_validate(&pbuf->pb_info);
1805 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1808 LNET_PING_BUFFER_SEQNO(pbuf) =
1809 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1813 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1814 struct lnet_handle_md ping_mdh)
1816 struct lnet_ping_buffer *old_pbuf = NULL;
1817 struct lnet_handle_md old_ping_md;
1819 /* switch the NIs to point to the new ping info created */
1820 lnet_net_lock(LNET_LOCK_EX);
1822 if (!the_lnet.ln_routing)
1823 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1824 if (!lnet_peer_discovery_disabled)
1825 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1827 /* Ensure only known feature bits have been set. */
1828 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1829 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1831 lnet_ping_target_install_locked(pbuf);
1833 if (the_lnet.ln_ping_target) {
1834 old_pbuf = the_lnet.ln_ping_target;
1835 old_ping_md = the_lnet.ln_ping_target_md;
1837 the_lnet.ln_ping_target_md = ping_mdh;
1838 the_lnet.ln_ping_target = pbuf;
1840 lnet_net_unlock(LNET_LOCK_EX);
1843 /* unlink and free the old ping info */
1844 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1845 lnet_ping_buffer_decref(old_pbuf);
1848 lnet_push_update_to_peers(0);
1852 lnet_ping_target_fini(void)
1854 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1855 &the_lnet.ln_ping_target_md);
1857 LNetEQFree(the_lnet.ln_ping_target_eq);
1859 lnet_ping_target_destroy();
1862 /* Resize the push target. */
1863 int lnet_push_target_resize(void)
1865 struct lnet_handle_md mdh;
1866 struct lnet_handle_md old_mdh;
1867 struct lnet_ping_buffer *pbuf;
1868 struct lnet_ping_buffer *old_pbuf;
1873 nnis = the_lnet.ln_push_target_nnis;
1875 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
1879 /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
1880 * dropped when we need to resize again (see "old_pbuf" below) or when
1881 * LNet is shutdown (see lnet_push_target_fini())
1883 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1885 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
1889 rc = lnet_push_target_post(pbuf, &mdh);
1891 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
1892 lnet_ping_buffer_decref(pbuf);
1896 lnet_net_lock(LNET_LOCK_EX);
1897 old_pbuf = the_lnet.ln_push_target;
1898 old_mdh = the_lnet.ln_push_target_md;
1899 the_lnet.ln_push_target = pbuf;
1900 the_lnet.ln_push_target_md = mdh;
1901 lnet_net_unlock(LNET_LOCK_EX);
1904 LNetMDUnlink(old_mdh);
1905 /* Drop ref set by lnet_ping_buffer_alloc() */
1906 lnet_ping_buffer_decref(old_pbuf);
1909 /* Received another push or reply that requires a larger buffer */
1910 if (nnis < the_lnet.ln_push_target_nnis)
1913 CDEBUG(D_NET, "nnis %d success\n", nnis);
1917 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
1918 struct lnet_handle_md *mdhp)
1920 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1921 struct lnet_md md = { NULL };
1925 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1926 LNET_PROTO_PING_MATCHBITS, 0,
1927 LNET_UNLINK, LNET_INS_AFTER);
1930 CERROR("Can't create push target ME: %d\n", rc);
1934 pbuf->pb_needs_post = false;
1936 /* This reference is dropped by lnet_push_target_event_handler() */
1937 lnet_ping_buffer_addref(pbuf);
1939 /* initialize md content */
1940 md.start = &pbuf->pb_info;
1941 md.length = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
1944 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
1946 md.eq_handle = the_lnet.ln_push_target_eq;
1948 rc = LNetMDAttach(me, md, LNET_UNLINK, mdhp);
1950 CERROR("Can't attach push MD: %d\n", rc);
1952 lnet_ping_buffer_decref(pbuf);
1953 pbuf->pb_needs_post = true;
1957 CDEBUG(D_NET, "posted push target %p\n", pbuf);
1962 static void lnet_push_target_event_handler(struct lnet_event *ev)
1964 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1966 CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
1969 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1970 lnet_swap_pinginfo(pbuf);
1972 if (ev->type == LNET_EVENT_UNLINK) {
1973 /* Drop ref added by lnet_push_target_post() */
1974 lnet_ping_buffer_decref(pbuf);
1978 lnet_peer_push_event(ev);
1980 /* Drop ref added by lnet_push_target_post */
1981 lnet_ping_buffer_decref(pbuf);
1984 /* Initialize the push target. */
1985 static int lnet_push_target_init(void)
1989 if (the_lnet.ln_push_target)
1992 the_lnet.ln_push_target_eq =
1993 LNetEQAlloc(lnet_push_target_event_handler);
1994 if (IS_ERR(the_lnet.ln_push_target_eq)) {
1995 rc = PTR_ERR(the_lnet.ln_push_target_eq);
1996 CERROR("Can't allocated push target EQ: %d\n", rc);
2000 rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2003 /* Start at the required minimum, we'll enlarge if required. */
2004 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2006 rc = lnet_push_target_resize();
2009 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2010 LNetEQFree(the_lnet.ln_push_target_eq);
2011 the_lnet.ln_push_target_eq = NULL;
2017 /* Clean up the push target. */
2018 static void lnet_push_target_fini(void)
2020 if (!the_lnet.ln_push_target)
2023 /* Unlink and invalidate to prevent new references. */
2024 LNetMDUnlink(the_lnet.ln_push_target_md);
2025 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2027 /* Wait for the unlink to complete. */
2028 wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2029 atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2030 "Still waiting for ping data MD to unlink\n");
2032 /* Drop ref set by lnet_ping_buffer_alloc() */
2033 lnet_ping_buffer_decref(the_lnet.ln_push_target);
2034 the_lnet.ln_push_target = NULL;
2035 the_lnet.ln_push_target_nnis = 0;
2037 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2038 LNetEQFree(the_lnet.ln_push_target_eq);
2039 the_lnet.ln_push_target_eq = NULL;
2043 lnet_ni_tq_credits(struct lnet_ni *ni)
2047 LASSERT(ni->ni_ncpts >= 1);
2049 if (ni->ni_ncpts == 1)
2050 return ni->ni_net->net_tunables.lct_max_tx_credits;
2052 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2053 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2054 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2060 lnet_ni_unlink_locked(struct lnet_ni *ni)
2062 /* move it to zombie list and nobody can find it anymore */
2063 LASSERT(!list_empty(&ni->ni_netlist));
2064 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2065 lnet_ni_decref_locked(ni, 0);
2069 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2074 struct list_head *zombie_list = &net->net_ni_zombie;
2077 * Now wait for the NIs I just nuked to show up on the zombie
2078 * list and shut them down in guaranteed thread context
2081 while (!list_empty(zombie_list)) {
2085 ni = list_entry(zombie_list->next,
2086 struct lnet_ni, ni_netlist);
2087 list_del_init(&ni->ni_netlist);
2088 /* the ni should be in deleting state. If it's not it's
2090 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2091 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2094 /* still busy, add it back to zombie list */
2095 list_add(&ni->ni_netlist, zombie_list);
2099 if (!list_empty(&ni->ni_netlist)) {
2100 lnet_net_unlock(LNET_LOCK_EX);
2102 if ((i & (-i)) == i) {
2104 "Waiting for zombie LNI %s\n",
2105 libcfs_nid2str(ni->ni_nid));
2107 schedule_timeout_uninterruptible(cfs_time_seconds(1));
2108 lnet_net_lock(LNET_LOCK_EX);
2112 lnet_net_unlock(LNET_LOCK_EX);
2114 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2116 LASSERT(!in_interrupt());
2117 /* Holding the mutex makes it safe for lnd_shutdown
2118 * to call module_put(). Module unload cannot finish
2119 * until lnet_unregister_lnd() completes, and that
2120 * requires the mutex.
2122 mutex_lock(&the_lnet.ln_lnd_mutex);
2123 (net->net_lnd->lnd_shutdown)(ni);
2124 mutex_unlock(&the_lnet.ln_lnd_mutex);
2127 CDEBUG(D_LNI, "Removed LNI %s\n",
2128 libcfs_nid2str(ni->ni_nid));
2132 lnet_net_lock(LNET_LOCK_EX);
2136 /* shutdown down the NI and release refcount */
2138 lnet_shutdown_lndni(struct lnet_ni *ni)
2141 struct lnet_net *net = ni->ni_net;
2143 lnet_net_lock(LNET_LOCK_EX);
2145 ni->ni_state = LNET_NI_STATE_DELETING;
2147 lnet_ni_unlink_locked(ni);
2148 lnet_incr_dlc_seq();
2149 lnet_net_unlock(LNET_LOCK_EX);
2151 /* clear messages for this NI on the lazy portal */
2152 for (i = 0; i < the_lnet.ln_nportals; i++)
2153 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2155 lnet_net_lock(LNET_LOCK_EX);
2156 lnet_clear_zombies_nis_locked(net);
2157 lnet_net_unlock(LNET_LOCK_EX);
2161 lnet_shutdown_lndnet(struct lnet_net *net)
2165 lnet_net_lock(LNET_LOCK_EX);
2167 list_del_init(&net->net_list);
2169 while (!list_empty(&net->net_ni_list)) {
2170 ni = list_entry(net->net_ni_list.next,
2171 struct lnet_ni, ni_netlist);
2172 lnet_net_unlock(LNET_LOCK_EX);
2173 lnet_shutdown_lndni(ni);
2174 lnet_net_lock(LNET_LOCK_EX);
2177 lnet_net_unlock(LNET_LOCK_EX);
2179 /* Do peer table cleanup for this net */
2180 lnet_peer_tables_cleanup(net);
2186 lnet_shutdown_lndnets(void)
2188 struct lnet_net *net;
2190 struct lnet_msg *msg, *tmp;
2192 /* NB called holding the global mutex */
2194 /* All quiet on the API front */
2195 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2196 LASSERT(the_lnet.ln_refcount == 0);
2198 lnet_net_lock(LNET_LOCK_EX);
2199 the_lnet.ln_state = LNET_STATE_STOPPING;
2202 * move the nets to the zombie list to avoid them being
2203 * picked up for new work. LONET is also included in the
2204 * Nets that will be moved to the zombie list
2206 list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2208 /* Drop the cached loopback Net. */
2209 if (the_lnet.ln_loni != NULL) {
2210 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2211 the_lnet.ln_loni = NULL;
2213 lnet_net_unlock(LNET_LOCK_EX);
2215 /* iterate through the net zombie list and delete each net */
2216 while (!list_empty(&the_lnet.ln_net_zombie)) {
2217 net = list_entry(the_lnet.ln_net_zombie.next,
2218 struct lnet_net, net_list);
2219 lnet_shutdown_lndnet(net);
2222 spin_lock(&the_lnet.ln_msg_resend_lock);
2223 list_splice(&the_lnet.ln_msg_resend, &resend);
2224 spin_unlock(&the_lnet.ln_msg_resend_lock);
2226 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2227 list_del_init(&msg->msg_list);
2228 msg->msg_no_resend = true;
2229 lnet_finalize(msg, -ECANCELED);
2232 lnet_net_lock(LNET_LOCK_EX);
2233 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2234 lnet_net_unlock(LNET_LOCK_EX);
2238 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2241 struct lnet_tx_queue *tq;
2243 struct lnet_net *net = ni->ni_net;
2245 mutex_lock(&the_lnet.ln_lnd_mutex);
2248 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2249 ni->ni_lnd_tunables_set = true;
2252 rc = (net->net_lnd->lnd_startup)(ni);
2254 mutex_unlock(&the_lnet.ln_lnd_mutex);
2257 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2258 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2263 ni->ni_state = LNET_NI_STATE_ACTIVE;
2266 /* We keep a reference on the loopback net through the loopback NI */
2267 if (net->net_lnd->lnd_type == LOLND) {
2269 LASSERT(the_lnet.ln_loni == NULL);
2270 the_lnet.ln_loni = ni;
2271 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2272 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2273 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2274 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2278 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2279 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2280 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2281 libcfs_lnd2str(net->net_lnd->lnd_type),
2282 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2284 /* shutdown the NI since if we get here then it must've already
2287 lnet_shutdown_lndni(ni);
2291 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2292 tq->tq_credits_min =
2293 tq->tq_credits_max =
2294 tq->tq_credits = lnet_ni_tq_credits(ni);
2297 atomic_set(&ni->ni_tx_credits,
2298 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2299 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2301 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2302 libcfs_nid2str(ni->ni_nid),
2303 ni->ni_net->net_tunables.lct_peer_tx_credits,
2304 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2305 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2306 ni->ni_net->net_tunables.lct_peer_timeout);
2315 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2318 struct lnet_net *net_l = NULL;
2319 LIST_HEAD(local_ni_list);
2323 const struct lnet_lnd *lnd;
2325 net->net_tunables.lct_peer_timeout;
2327 net->net_tunables.lct_max_tx_credits;
2328 int peerrtrcredits =
2329 net->net_tunables.lct_peer_rtr_credits;
2332 * make sure that this net is unique. If it isn't then
2333 * we are adding interfaces to an already existing network, and
2334 * 'net' is just a convenient way to pass in the list.
2335 * if it is unique we need to find the LND and load it if
2338 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2339 lnd_type = LNET_NETTYP(net->net_id);
2341 mutex_lock(&the_lnet.ln_lnd_mutex);
2342 lnd = lnet_find_lnd_by_type(lnd_type);
2345 mutex_unlock(&the_lnet.ln_lnd_mutex);
2346 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2347 mutex_lock(&the_lnet.ln_lnd_mutex);
2349 lnd = lnet_find_lnd_by_type(lnd_type);
2351 mutex_unlock(&the_lnet.ln_lnd_mutex);
2352 CERROR("Can't load LND %s, module %s, rc=%d\n",
2353 libcfs_lnd2str(lnd_type),
2354 libcfs_lnd2modname(lnd_type), rc);
2355 #ifndef HAVE_MODULE_LOADING_SUPPORT
2356 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2357 "compiled with kernel module "
2358 "loading support.");
2367 mutex_unlock(&the_lnet.ln_lnd_mutex);
2373 * net_l: if the network being added is unique then net_l
2374 * will point to that network
2375 * if the network being added is not unique then
2376 * net_l points to the existing network.
2378 * When we enter the loop below, we'll pick NIs off he
2379 * network beign added and start them up, then add them to
2380 * a local ni list. Once we've successfully started all
2381 * the NIs then we join the local NI list (of started up
2382 * networks) with the net_l->net_ni_list, which should
2383 * point to the correct network to add the new ni list to
2385 * If any of the new NIs fail to start up, then we want to
2386 * iterate through the local ni list, which should include
2387 * any NIs which were successfully started up, and shut
2390 * After than we want to delete the network being added,
2391 * to avoid a memory leak.
2395 * When a network uses TCP bonding then all its interfaces
2396 * must be specified when the network is first defined: the
2397 * TCP bonding code doesn't allow for interfaces to be added
2400 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2401 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2406 while (!list_empty(&net->net_ni_added)) {
2407 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2409 list_del_init(&ni->ni_netlist);
2411 /* make sure that the the NI we're about to start
2412 * up is actually unique. if it's not fail. */
2413 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2414 ni->ni_interfaces[0])) {
2419 /* adjust the pointer the parent network, just in case it
2420 * the net is a duplicate */
2423 rc = lnet_startup_lndni(ni, tun);
2429 list_add_tail(&ni->ni_netlist, &local_ni_list);
2434 lnet_net_lock(LNET_LOCK_EX);
2435 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2436 lnet_incr_dlc_seq();
2437 lnet_net_unlock(LNET_LOCK_EX);
2439 /* if the network is not unique then we don't want to keep
2440 * it around after we're done. Free it. Otherwise add that
2441 * net to the global the_lnet.ln_nets */
2442 if (net_l != net && net_l != NULL) {
2444 * TODO - note. currently the tunables can not be updated
2450 * restore tunables after it has been overwitten by the
2453 if (peer_timeout != -1)
2454 net->net_tunables.lct_peer_timeout = peer_timeout;
2455 if (maxtxcredits != -1)
2456 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2457 if (peerrtrcredits != -1)
2458 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2460 lnet_net_lock(LNET_LOCK_EX);
2461 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2462 lnet_net_unlock(LNET_LOCK_EX);
2465 /* update net count */
2466 lnet_current_net_count = lnet_get_net_count();
2472 * shutdown the new NIs that are being started up
2473 * free the NET being started
2475 while (!list_empty(&local_ni_list)) {
2476 ni = list_entry(local_ni_list.next, struct lnet_ni,
2479 lnet_shutdown_lndni(ni);
2489 lnet_startup_lndnets(struct list_head *netlist)
2491 struct lnet_net *net;
2496 * Change to running state before bringing up the LNDs. This
2497 * allows lnet_shutdown_lndnets() to assert that we've passed
2500 lnet_net_lock(LNET_LOCK_EX);
2501 the_lnet.ln_state = LNET_STATE_RUNNING;
2502 lnet_net_unlock(LNET_LOCK_EX);
2504 while (!list_empty(netlist)) {
2505 net = list_entry(netlist->next, struct lnet_net, net_list);
2506 list_del_init(&net->net_list);
2508 rc = lnet_startup_lndnet(net, NULL);
2518 lnet_shutdown_lndnets();
2524 * Initialize LNet library.
2526 * Automatically called at module loading time. Caller has to call
2527 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2528 * latter returned 0. It must be called exactly once.
2530 * \retval 0 on success
2531 * \retval -ve on failures.
2533 int lnet_lib_init(void)
2537 lnet_assert_wire_constants();
2539 /* refer to global cfs_cpt_table for now */
2540 the_lnet.ln_cpt_table = cfs_cpt_tab;
2541 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2543 LASSERT(the_lnet.ln_cpt_number > 0);
2544 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2545 /* we are under risk of consuming all lh_cookie */
2546 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2547 "please change setting of CPT-table and retry\n",
2548 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2552 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2553 the_lnet.ln_cpt_bits++;
2555 rc = lnet_create_locks();
2557 CERROR("Can't create LNet global locks: %d\n", rc);
2561 the_lnet.ln_refcount = 0;
2562 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2563 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2565 /* The hash table size is the number of bits it takes to express the set
2566 * ln_num_routes, minus 1 (better to under estimate than over so we
2567 * don't waste memory). */
2568 if (rnet_htable_size <= 0)
2569 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2570 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2571 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2572 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2573 order_base_2(rnet_htable_size) - 1);
2575 /* All LNDs apart from the LOLND are in separate modules. They
2576 * register themselves when their module loads, and unregister
2577 * themselves when their module is unloaded. */
2578 lnet_register_lnd(&the_lolnd);
2583 * Finalize LNet library.
2585 * \pre lnet_lib_init() called with success.
2586 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2588 * As this happens at module-unload, all lnds must already be unloaded,
2589 * so they must already be unregistered.
2591 void lnet_lib_exit(void)
2595 LASSERT(the_lnet.ln_refcount == 0);
2596 lnet_unregister_lnd(&the_lolnd);
2597 for (i = 0; i < NUM_LNDS; i++)
2598 LASSERT(!the_lnet.ln_lnds[i]);
2599 lnet_destroy_locks();
2603 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2605 * Users must call this function at least once before any other functions.
2606 * For each successful call there must be a corresponding call to
2607 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2610 * The PID used by LNet may be different from the one requested.
2613 * \param requested_pid PID requested by the caller.
2615 * \return >= 0 on success, and < 0 error code on failures.
2618 LNetNIInit(lnet_pid_t requested_pid)
2620 int im_a_router = 0;
2623 struct lnet_ping_buffer *pbuf;
2624 struct lnet_handle_md ping_mdh;
2625 LIST_HEAD(net_head);
2626 struct lnet_net *net;
2628 mutex_lock(&the_lnet.ln_api_mutex);
2630 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2632 if (the_lnet.ln_refcount > 0) {
2633 rc = the_lnet.ln_refcount++;
2634 mutex_unlock(&the_lnet.ln_api_mutex);
2638 rc = lnet_prepare(requested_pid);
2640 mutex_unlock(&the_lnet.ln_api_mutex);
2644 /* create a network for Loopback network */
2645 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2648 goto err_empty_list;
2651 /* Add in the loopback NI */
2652 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2654 goto err_empty_list;
2657 /* If LNet is being initialized via DLC it is possible
2658 * that the user requests not to load module parameters (ones which
2659 * are supported by DLC) on initialization. Therefore, make sure not
2660 * to load networks, routes and forwarding from module parameters
2661 * in this case. On cleanup in case of failure only clean up
2662 * routes if it has been loaded */
2663 if (!the_lnet.ln_nis_from_mod_params) {
2664 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2667 goto err_empty_list;
2670 ni_count = lnet_startup_lndnets(&net_head);
2673 goto err_empty_list;
2676 if (!the_lnet.ln_nis_from_mod_params) {
2677 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2679 goto err_shutdown_lndnis;
2681 rc = lnet_rtrpools_alloc(im_a_router);
2683 goto err_destroy_routes;
2686 rc = lnet_acceptor_start();
2688 goto err_destroy_routes;
2690 the_lnet.ln_refcount = 1;
2691 /* Now I may use my own API functions... */
2693 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2695 goto err_acceptor_stop;
2697 lnet_ping_target_update(pbuf, ping_mdh);
2699 the_lnet.ln_mt_eq = LNetEQAlloc(lnet_mt_event_handler);
2700 if (IS_ERR(the_lnet.ln_mt_eq)) {
2701 rc = PTR_ERR(the_lnet.ln_mt_eq);
2702 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2706 rc = lnet_push_target_init();
2710 rc = lnet_peer_discovery_start();
2712 goto err_destroy_push_target;
2714 rc = lnet_monitor_thr_start();
2716 goto err_stop_discovery_thr;
2719 lnet_router_debugfs_init();
2721 mutex_unlock(&the_lnet.ln_api_mutex);
2723 complete_all(&the_lnet.ln_started);
2725 /* wait for all routers to start */
2726 lnet_wait_router_start();
2730 err_stop_discovery_thr:
2731 lnet_peer_discovery_stop();
2732 err_destroy_push_target:
2733 lnet_push_target_fini();
2735 lnet_ping_target_fini();
2737 the_lnet.ln_refcount = 0;
2738 lnet_acceptor_stop();
2740 if (!the_lnet.ln_nis_from_mod_params)
2741 lnet_destroy_routes();
2742 err_shutdown_lndnis:
2743 lnet_shutdown_lndnets();
2747 mutex_unlock(&the_lnet.ln_api_mutex);
2748 while (!list_empty(&net_head)) {
2749 struct lnet_net *net;
2751 net = list_entry(net_head.next, struct lnet_net, net_list);
2752 list_del_init(&net->net_list);
2757 EXPORT_SYMBOL(LNetNIInit);
2760 * Stop LNet interfaces, routing, and forwarding.
2762 * Users must call this function once for each successful call to LNetNIInit().
2763 * Once the LNetNIFini() operation has been started, the results of pending
2764 * API operations are undefined.
2766 * \return always 0 for current implementation.
2771 mutex_lock(&the_lnet.ln_api_mutex);
2773 LASSERT(the_lnet.ln_refcount > 0);
2775 if (the_lnet.ln_refcount != 1) {
2776 the_lnet.ln_refcount--;
2778 LASSERT(!the_lnet.ln_niinit_self);
2782 lnet_router_debugfs_fini();
2783 lnet_monitor_thr_stop();
2784 lnet_peer_discovery_stop();
2785 lnet_push_target_fini();
2786 lnet_ping_target_fini();
2788 /* Teardown fns that use my own API functions BEFORE here */
2789 the_lnet.ln_refcount = 0;
2791 lnet_acceptor_stop();
2792 lnet_destroy_routes();
2793 lnet_shutdown_lndnets();
2797 mutex_unlock(&the_lnet.ln_api_mutex);
2800 EXPORT_SYMBOL(LNetNIFini);
2803 * Grabs the ni data from the ni structure and fills the out
2806 * \param[in] ni network interface structure
2807 * \param[out] cfg_ni NI config information
2808 * \param[out] tun network and LND tunables
2811 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2812 struct lnet_ioctl_config_lnd_tunables *tun,
2813 struct lnet_ioctl_element_stats *stats,
2816 size_t min_size = 0;
2819 if (!ni || !cfg_ni || !tun)
2822 if (ni->ni_interfaces[0] != NULL) {
2823 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2824 if (ni->ni_interfaces[i] != NULL) {
2825 strncpy(cfg_ni->lic_ni_intf[i],
2826 ni->ni_interfaces[i],
2827 sizeof(cfg_ni->lic_ni_intf[i]));
2832 cfg_ni->lic_nid = ni->ni_nid;
2833 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2834 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2836 cfg_ni->lic_status = ni->ni_status->ns_status;
2837 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2838 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2840 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2843 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2844 LNET_STATS_TYPE_SEND);
2845 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2846 LNET_STATS_TYPE_RECV);
2847 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2848 LNET_STATS_TYPE_DROP);
2852 * tun->lt_tun will always be present, but in order to be
2853 * backwards compatible, we need to deal with the cases when
2854 * tun->lt_tun is smaller than what the kernel has, because it
2855 * comes from an older version of a userspace program, then we'll
2856 * need to copy as much information as we have available space.
2858 min_size = tun_size - sizeof(tun->lt_cmn);
2859 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2861 /* copy over the cpts */
2862 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2863 ni->ni_cpts == NULL) {
2864 for (i = 0; i < ni->ni_ncpts; i++)
2865 cfg_ni->lic_cpts[i] = i;
2868 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2869 i < LNET_MAX_SHOW_NUM_CPT;
2871 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2873 cfg_ni->lic_ncpts = ni->ni_ncpts;
2877 * NOTE: This is a legacy function left in the code to be backwards
2878 * compatible with older userspace programs. It should eventually be
2881 * Grabs the ni data from the ni structure and fills the out
2884 * \param[in] ni network interface structure
2885 * \param[out] config config information
2888 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2889 struct lnet_ioctl_config_data *config)
2891 struct lnet_ioctl_net_config *net_config;
2892 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2893 size_t min_size, tunable_size = 0;
2899 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2903 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2904 ARRAY_SIZE(net_config->ni_interfaces));
2906 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2907 if (!ni->ni_interfaces[i])
2910 strncpy(net_config->ni_interfaces[i],
2911 ni->ni_interfaces[i],
2912 sizeof(net_config->ni_interfaces[i]));
2915 config->cfg_nid = ni->ni_nid;
2916 config->cfg_config_u.cfg_net.net_peer_timeout =
2917 ni->ni_net->net_tunables.lct_peer_timeout;
2918 config->cfg_config_u.cfg_net.net_max_tx_credits =
2919 ni->ni_net->net_tunables.lct_max_tx_credits;
2920 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2921 ni->ni_net->net_tunables.lct_peer_tx_credits;
2922 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2923 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2925 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2926 net_config->ni_status = LNET_NI_STATUS_UP;
2928 net_config->ni_status = ni->ni_status->ns_status;
2931 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2933 for (i = 0; i < num_cpts; i++)
2934 net_config->ni_cpts[i] = ni->ni_cpts[i];
2936 config->cfg_ncpts = num_cpts;
2940 * See if user land tools sent in a newer and larger version
2941 * of struct lnet_tunables than what the kernel uses.
2943 min_size = sizeof(*config) + sizeof(*net_config);
2945 if (config->cfg_hdr.ioc_len > min_size)
2946 tunable_size = config->cfg_hdr.ioc_len - min_size;
2948 /* Don't copy too much data to user space */
2949 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2950 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2952 if (lnd_cfg && min_size) {
2953 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2954 config->cfg_config_u.cfg_net.net_interface_count = 1;
2956 /* Tell user land that kernel side has less data */
2957 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2958 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2959 config->cfg_hdr.ioc_len -= min_size;
2965 lnet_get_ni_idx_locked(int idx)
2968 struct lnet_net *net;
2970 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2971 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2981 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2984 struct lnet_net *net = mynet;
2987 * It is possible that the net has been cleaned out while there is
2988 * a message being sent. This function accessed the net without
2989 * checking if the list is empty
2993 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2995 if (list_empty(&net->net_ni_list))
2997 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3003 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3004 /* if you reached the end of the ni list and the net is
3005 * specified, then there are no more nis in that net */
3009 /* we reached the end of this net ni list. move to the
3011 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3012 /* no more nets and no more NIs. */
3015 /* get the next net */
3016 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3018 if (list_empty(&net->net_ni_list))
3020 /* get the ni on it */
3021 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3027 if (list_empty(&prev->ni_netlist))
3030 /* there are more nis left */
3031 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3037 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3042 int idx = config->cfg_count;
3044 cpt = lnet_net_lock_current();
3046 ni = lnet_get_ni_idx_locked(idx);
3051 lnet_fill_ni_info_legacy(ni, config);
3055 lnet_net_unlock(cpt);
3060 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3061 struct lnet_ioctl_config_lnd_tunables *tun,
3062 struct lnet_ioctl_element_stats *stats,
3069 if (!cfg_ni || !tun || !stats)
3072 cpt = lnet_net_lock_current();
3074 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3079 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3083 lnet_net_unlock(cpt);
3087 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3096 cpt = lnet_net_lock_current();
3098 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3101 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3105 lnet_net_unlock(cpt);
3110 static int lnet_add_net_common(struct lnet_net *net,
3111 struct lnet_ioctl_config_lnd_tunables *tun)
3114 struct lnet_ping_buffer *pbuf;
3115 struct lnet_handle_md ping_mdh;
3117 struct lnet_remotenet *rnet;
3120 lnet_net_lock(LNET_LOCK_EX);
3121 rnet = lnet_find_rnet_locked(net->net_id);
3122 lnet_net_unlock(LNET_LOCK_EX);
3124 * make sure that the net added doesn't invalidate the current
3125 * configuration LNet is keeping
3128 CERROR("Adding net %s will invalidate routing configuration\n",
3129 libcfs_net2str(net->net_id));
3135 * make sure you calculate the correct number of slots in the ping
3136 * buffer. Since the ping info is a flattened list of all the NIs,
3137 * we should allocate enough slots to accomodate the number of NIs
3138 * which will be added.
3140 * since ni hasn't been configured yet, use
3141 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3143 net_ni_count = lnet_get_net_ni_count_pre(net);
3145 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3146 net_ni_count + lnet_get_ni_count(),
3154 memcpy(&net->net_tunables,
3155 &tun->lt_cmn, sizeof(net->net_tunables));
3157 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3159 net_id = net->net_id;
3161 rc = lnet_startup_lndnet(net,
3162 (tun) ? &tun->lt_tun : NULL);
3166 lnet_net_lock(LNET_LOCK_EX);
3167 net = lnet_get_net_locked(net_id);
3168 lnet_net_unlock(LNET_LOCK_EX);
3173 * Start the acceptor thread if this is the first network
3174 * being added that requires the thread.
3176 if (net->net_lnd->lnd_accept) {
3177 rc = lnet_acceptor_start();
3179 /* shutdown the net that we just started */
3180 CERROR("Failed to start up acceptor thread\n");
3181 lnet_shutdown_lndnet(net);
3186 lnet_net_lock(LNET_LOCK_EX);
3187 lnet_peer_net_added(net);
3188 lnet_net_unlock(LNET_LOCK_EX);
3190 lnet_ping_target_update(pbuf, ping_mdh);
3195 lnet_ping_md_unlink(pbuf, &ping_mdh);
3196 lnet_ping_buffer_decref(pbuf);
3200 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3201 struct lnet_ioctl_config_lnd_tunables *tun)
3203 struct lnet_net *net;
3206 LIST_HEAD(net_head);
3208 rc = lnet_parse_ip2nets(&nets, ip2nets);
3212 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3216 mutex_lock(&the_lnet.ln_api_mutex);
3217 while (!list_empty(&net_head)) {
3218 net = list_entry(net_head.next, struct lnet_net, net_list);
3219 list_del_init(&net->net_list);
3220 rc = lnet_add_net_common(net, tun);
3226 mutex_unlock(&the_lnet.ln_api_mutex);
3228 while (!list_empty(&net_head)) {
3229 net = list_entry(net_head.next, struct lnet_net, net_list);
3230 list_del_init(&net->net_list);
3236 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3238 struct lnet_net *net;
3240 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3242 __u32 net_id, lnd_type;
3244 /* get the tunables if they are available */
3245 if (conf->lic_cfg_hdr.ioc_len >=
3246 sizeof(*conf) + sizeof(*tun))
3247 tun = (struct lnet_ioctl_config_lnd_tunables *)
3250 /* handle legacy ip2nets from DLC */
3251 if (conf->lic_legacy_ip2nets[0] != '\0')
3252 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3255 net_id = LNET_NIDNET(conf->lic_nid);
3256 lnd_type = LNET_NETTYP(net_id);
3258 if (!libcfs_isknown_lnd(lnd_type)) {
3259 CERROR("No valid net and lnd information provided\n");
3263 net = lnet_net_alloc(net_id, NULL);
3267 for (i = 0; i < conf->lic_ncpts; i++) {
3268 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3272 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3273 conf->lic_ni_intf[0]);
3277 mutex_lock(&the_lnet.ln_api_mutex);
3279 rc = lnet_add_net_common(net, tun);
3281 mutex_unlock(&the_lnet.ln_api_mutex);
3286 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3288 struct lnet_net *net;
3290 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3291 struct lnet_ping_buffer *pbuf;
3292 struct lnet_handle_md ping_mdh;
3297 /* don't allow userspace to shutdown the LOLND */
3298 if (LNET_NETTYP(net_id) == LOLND)
3301 mutex_lock(&the_lnet.ln_api_mutex);
3305 net = lnet_get_net_locked(net_id);
3307 CERROR("net %s not found\n",
3308 libcfs_net2str(net_id));
3313 addr = LNET_NIDADDR(conf->lic_nid);
3315 /* remove the entire net */
3316 net_count = lnet_get_net_ni_count_locked(net);
3320 /* create and link a new ping info, before removing the old one */
3321 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3322 lnet_get_ni_count() - net_count,
3325 goto unlock_api_mutex;
3327 lnet_shutdown_lndnet(net);
3329 lnet_acceptor_stop();
3331 lnet_ping_target_update(pbuf, ping_mdh);
3333 goto unlock_api_mutex;
3336 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3338 CERROR("nid %s not found\n",
3339 libcfs_nid2str(conf->lic_nid));
3344 net_count = lnet_get_net_ni_count_locked(net);
3348 /* create and link a new ping info, before removing the old one */
3349 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3350 lnet_get_ni_count() - 1, false);
3352 goto unlock_api_mutex;
3354 lnet_shutdown_lndni(ni);
3356 lnet_acceptor_stop();
3358 lnet_ping_target_update(pbuf, ping_mdh);
3360 /* check if the net is empty and remove it if it is */
3362 lnet_shutdown_lndnet(net);
3364 goto unlock_api_mutex;
3369 mutex_unlock(&the_lnet.ln_api_mutex);
3375 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3376 * They are only expected to be called for unique networks.
3377 * That can be as a result of older DLC library
3378 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3381 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3383 struct lnet_net *net;
3384 LIST_HEAD(net_head);
3386 struct lnet_ioctl_config_lnd_tunables tun;
3387 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3389 /* Create a net/ni structures for the network string */
3390 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3392 return rc == 0 ? -EINVAL : rc;
3394 mutex_lock(&the_lnet.ln_api_mutex);
3397 rc = -EINVAL; /* only add one network per call */
3398 goto out_unlock_clean;
3401 net = list_entry(net_head.next, struct lnet_net, net_list);
3402 list_del_init(&net->net_list);
3404 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3406 memset(&tun, 0, sizeof(tun));
3408 tun.lt_cmn.lct_peer_timeout =
3409 conf->cfg_config_u.cfg_net.net_peer_timeout;
3410 tun.lt_cmn.lct_peer_tx_credits =
3411 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3412 tun.lt_cmn.lct_peer_rtr_credits =
3413 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3414 tun.lt_cmn.lct_max_tx_credits =
3415 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3417 rc = lnet_add_net_common(net, &tun);
3420 mutex_unlock(&the_lnet.ln_api_mutex);
3421 while (!list_empty(&net_head)) {
3422 /* net_head list is empty in success case */
3423 net = list_entry(net_head.next, struct lnet_net, net_list);
3424 list_del_init(&net->net_list);
3431 lnet_dyn_del_net(__u32 net_id)
3433 struct lnet_net *net;
3434 struct lnet_ping_buffer *pbuf;
3435 struct lnet_handle_md ping_mdh;
3439 /* don't allow userspace to shutdown the LOLND */
3440 if (LNET_NETTYP(net_id) == LOLND)
3443 mutex_lock(&the_lnet.ln_api_mutex);
3447 net = lnet_get_net_locked(net_id);
3454 net_ni_count = lnet_get_net_ni_count_locked(net);
3458 /* create and link a new ping info, before removing the old one */
3459 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3460 lnet_get_ni_count() - net_ni_count, false);
3464 lnet_shutdown_lndnet(net);
3466 lnet_acceptor_stop();
3468 lnet_ping_target_update(pbuf, ping_mdh);
3471 mutex_unlock(&the_lnet.ln_api_mutex);
3476 void lnet_incr_dlc_seq(void)
3478 atomic_inc(&lnet_dlc_seq_no);
3481 __u32 lnet_get_dlc_seq_locked(void)
3483 return atomic_read(&lnet_dlc_seq_no);
3487 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3489 struct lnet_net *net;
3492 lnet_net_lock(LNET_LOCK_EX);
3493 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3494 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3495 if (ni->ni_nid == nid || all) {
3496 atomic_set(&ni->ni_healthv, value);
3497 if (list_empty(&ni->ni_recovery) &&
3498 value < LNET_MAX_HEALTH_VALUE) {
3499 CERROR("manually adding local NI %s to recovery\n",
3500 libcfs_nid2str(ni->ni_nid));
3501 list_add_tail(&ni->ni_recovery,
3502 &the_lnet.ln_mt_localNIRecovq);
3503 lnet_ni_addref_locked(ni, 0);
3506 lnet_net_unlock(LNET_LOCK_EX);
3512 lnet_net_unlock(LNET_LOCK_EX);
3516 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3520 lnet_nid_t nid = stats->hlni_nid;
3522 cpt = lnet_net_lock_current();
3523 ni = lnet_nid2ni_locked(nid, cpt);
3530 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3531 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3532 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3533 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3534 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3535 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3536 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3539 lnet_net_unlock(cpt);
3545 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3550 lnet_net_lock(LNET_LOCK_EX);
3551 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3552 list->rlst_nid_array[i] = ni->ni_nid;
3554 if (i >= LNET_MAX_SHOW_NUM_NID)
3557 lnet_net_unlock(LNET_LOCK_EX);
3558 list->rlst_num_nids = i;
3564 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3566 struct lnet_peer_ni *lpni;
3569 lnet_net_lock(LNET_LOCK_EX);
3570 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3571 list->rlst_nid_array[i] = lpni->lpni_nid;
3573 if (i >= LNET_MAX_SHOW_NUM_NID)
3576 lnet_net_unlock(LNET_LOCK_EX);
3577 list->rlst_num_nids = i;
3583 * LNet ioctl handler.
3587 LNetCtl(unsigned int cmd, void *arg)
3589 struct libcfs_ioctl_data *data = arg;
3590 struct lnet_ioctl_config_data *config;
3591 struct lnet_process_id id = {0};
3595 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3596 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3599 case IOC_LIBCFS_GET_NI:
3600 rc = LNetGetId(data->ioc_count, &id);
3601 data->ioc_nid = id.nid;
3604 case IOC_LIBCFS_FAIL_NID:
3605 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3607 case IOC_LIBCFS_ADD_ROUTE: {
3608 /* default router sensitivity to 1 */
3609 unsigned int sensitivity = 1;
3612 if (config->cfg_hdr.ioc_len < sizeof(*config))
3615 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3617 config->cfg_config_u.cfg_route.rtr_sensitivity;
3620 mutex_lock(&the_lnet.ln_api_mutex);
3621 rc = lnet_add_route(config->cfg_net,
3622 config->cfg_config_u.cfg_route.rtr_hop,
3624 config->cfg_config_u.cfg_route.
3625 rtr_priority, sensitivity);
3626 mutex_unlock(&the_lnet.ln_api_mutex);
3630 case IOC_LIBCFS_DEL_ROUTE:
3633 if (config->cfg_hdr.ioc_len < sizeof(*config))
3636 mutex_lock(&the_lnet.ln_api_mutex);
3637 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3638 mutex_unlock(&the_lnet.ln_api_mutex);
3641 case IOC_LIBCFS_GET_ROUTE:
3644 if (config->cfg_hdr.ioc_len < sizeof(*config))
3647 mutex_lock(&the_lnet.ln_api_mutex);
3648 rc = lnet_get_route(config->cfg_count,
3650 &config->cfg_config_u.cfg_route.rtr_hop,
3652 &config->cfg_config_u.cfg_route.rtr_flags,
3653 &config->cfg_config_u.cfg_route.
3655 &config->cfg_config_u.cfg_route.
3657 mutex_unlock(&the_lnet.ln_api_mutex);
3660 case IOC_LIBCFS_GET_LOCAL_NI: {
3661 struct lnet_ioctl_config_ni *cfg_ni;
3662 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3663 struct lnet_ioctl_element_stats *stats;
3668 /* get the tunables if they are available */
3669 if (cfg_ni->lic_cfg_hdr.ioc_len <
3670 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3673 stats = (struct lnet_ioctl_element_stats *)
3675 tun = (struct lnet_ioctl_config_lnd_tunables *)
3676 (cfg_ni->lic_bulk + sizeof(*stats));
3678 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3681 mutex_lock(&the_lnet.ln_api_mutex);
3682 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3683 mutex_unlock(&the_lnet.ln_api_mutex);
3687 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3688 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3690 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3693 mutex_lock(&the_lnet.ln_api_mutex);
3694 rc = lnet_get_ni_stats(msg_stats);
3695 mutex_unlock(&the_lnet.ln_api_mutex);
3700 case IOC_LIBCFS_GET_NET: {
3701 size_t total = sizeof(*config) +
3702 sizeof(struct lnet_ioctl_net_config);
3705 if (config->cfg_hdr.ioc_len < total)
3708 mutex_lock(&the_lnet.ln_api_mutex);
3709 rc = lnet_get_net_config(config);
3710 mutex_unlock(&the_lnet.ln_api_mutex);
3714 case IOC_LIBCFS_GET_LNET_STATS:
3716 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3718 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3721 mutex_lock(&the_lnet.ln_api_mutex);
3722 lnet_counters_get(&lnet_stats->st_cntrs);
3723 mutex_unlock(&the_lnet.ln_api_mutex);
3727 case IOC_LIBCFS_CONFIG_RTR:
3730 if (config->cfg_hdr.ioc_len < sizeof(*config))
3733 mutex_lock(&the_lnet.ln_api_mutex);
3734 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3735 rc = lnet_rtrpools_enable();
3736 mutex_unlock(&the_lnet.ln_api_mutex);
3739 lnet_rtrpools_disable();
3740 mutex_unlock(&the_lnet.ln_api_mutex);
3743 case IOC_LIBCFS_ADD_BUF:
3746 if (config->cfg_hdr.ioc_len < sizeof(*config))
3749 mutex_lock(&the_lnet.ln_api_mutex);
3750 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3752 config->cfg_config_u.cfg_buffers.
3754 config->cfg_config_u.cfg_buffers.
3756 mutex_unlock(&the_lnet.ln_api_mutex);
3759 case IOC_LIBCFS_SET_NUMA_RANGE: {
3760 struct lnet_ioctl_set_value *numa;
3762 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3764 lnet_net_lock(LNET_LOCK_EX);
3765 lnet_numa_range = numa->sv_value;
3766 lnet_net_unlock(LNET_LOCK_EX);
3770 case IOC_LIBCFS_GET_NUMA_RANGE: {
3771 struct lnet_ioctl_set_value *numa;
3773 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3775 numa->sv_value = lnet_numa_range;
3779 case IOC_LIBCFS_GET_BUF: {
3780 struct lnet_ioctl_pool_cfg *pool_cfg;
3781 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3785 if (config->cfg_hdr.ioc_len < total)
3788 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3790 mutex_lock(&the_lnet.ln_api_mutex);
3791 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3792 mutex_unlock(&the_lnet.ln_api_mutex);
3796 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3797 struct lnet_ioctl_local_ni_hstats *stats = arg;
3799 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3802 mutex_lock(&the_lnet.ln_api_mutex);
3803 rc = lnet_get_local_ni_hstats(stats);
3804 mutex_unlock(&the_lnet.ln_api_mutex);
3809 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3810 struct lnet_ioctl_recovery_list *list = arg;
3811 if (list->rlst_hdr.ioc_len < sizeof(*list))
3814 mutex_lock(&the_lnet.ln_api_mutex);
3815 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3816 rc = lnet_get_local_ni_recovery_list(list);
3818 rc = lnet_get_peer_ni_recovery_list(list);
3819 mutex_unlock(&the_lnet.ln_api_mutex);
3823 case IOC_LIBCFS_ADD_PEER_NI: {
3824 struct lnet_ioctl_peer_cfg *cfg = arg;
3826 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3829 mutex_lock(&the_lnet.ln_api_mutex);
3830 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3833 mutex_unlock(&the_lnet.ln_api_mutex);
3837 case IOC_LIBCFS_DEL_PEER_NI: {
3838 struct lnet_ioctl_peer_cfg *cfg = arg;
3840 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3843 mutex_lock(&the_lnet.ln_api_mutex);
3844 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3845 cfg->prcfg_cfg_nid);
3846 mutex_unlock(&the_lnet.ln_api_mutex);
3850 case IOC_LIBCFS_GET_PEER_INFO: {
3851 struct lnet_ioctl_peer *peer_info = arg;
3853 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3856 mutex_lock(&the_lnet.ln_api_mutex);
3857 rc = lnet_get_peer_ni_info(
3858 peer_info->pr_count,
3860 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3861 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3862 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3863 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3864 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3865 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3866 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3867 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3868 mutex_unlock(&the_lnet.ln_api_mutex);
3872 case IOC_LIBCFS_GET_PEER_NI: {
3873 struct lnet_ioctl_peer_cfg *cfg = arg;
3875 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3878 mutex_lock(&the_lnet.ln_api_mutex);
3879 rc = lnet_get_peer_info(cfg,
3880 (void __user *)cfg->prcfg_bulk);
3881 mutex_unlock(&the_lnet.ln_api_mutex);
3885 case IOC_LIBCFS_GET_PEER_LIST: {
3886 struct lnet_ioctl_peer_cfg *cfg = arg;
3888 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3891 mutex_lock(&the_lnet.ln_api_mutex);
3892 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3893 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3894 mutex_unlock(&the_lnet.ln_api_mutex);
3898 case IOC_LIBCFS_SET_HEALHV: {
3899 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3901 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3903 if (cfg->rh_value < 0 ||
3904 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3905 value = LNET_MAX_HEALTH_VALUE;
3907 value = cfg->rh_value;
3908 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3909 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3910 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3911 mutex_lock(&the_lnet.ln_api_mutex);
3912 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3913 lnet_ni_set_healthv(cfg->rh_nid, value,
3916 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3918 mutex_unlock(&the_lnet.ln_api_mutex);
3922 case IOC_LIBCFS_NOTIFY_ROUTER: {
3923 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3925 /* The deadline passed in by the user should be some time in
3926 * seconds in the future since the UNIX epoch. We have to map
3927 * that deadline to the wall clock.
3929 deadline += ktime_get_seconds();
3930 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3934 case IOC_LIBCFS_LNET_DIST:
3935 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3936 if (rc < 0 && rc != -EHOSTUNREACH)
3939 data->ioc_u32[0] = rc;
3942 case IOC_LIBCFS_TESTPROTOCOMPAT:
3943 the_lnet.ln_testprotocompat = data->ioc_flags;
3946 case IOC_LIBCFS_LNET_FAULT:
3947 return lnet_fault_ctl(data->ioc_flags, data);
3949 case IOC_LIBCFS_PING: {
3950 signed long timeout;
3952 id.nid = data->ioc_nid;
3953 id.pid = data->ioc_u32[0];
3955 /* If timeout is negative then set default of 3 minutes */
3956 if (((s32)data->ioc_u32[1] <= 0) ||
3957 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3958 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3960 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3962 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3963 data->ioc_plen1 / sizeof(struct lnet_process_id));
3968 data->ioc_count = rc;
3972 case IOC_LIBCFS_PING_PEER: {
3973 struct lnet_ioctl_ping_data *ping = arg;
3974 struct lnet_peer *lp;
3975 signed long timeout;
3977 /* If timeout is negative then set default of 3 minutes */
3978 if (((s32)ping->op_param) <= 0 ||
3979 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3980 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3982 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3984 rc = lnet_ping(ping->ping_id, timeout,
3990 mutex_lock(&the_lnet.ln_api_mutex);
3991 lp = lnet_find_peer(ping->ping_id.nid);
3993 ping->ping_id.nid = lp->lp_primary_nid;
3994 ping->mr_info = lnet_peer_is_multi_rail(lp);
3995 lnet_peer_decref_locked(lp);
3997 mutex_unlock(&the_lnet.ln_api_mutex);
3999 ping->ping_count = rc;
4003 case IOC_LIBCFS_DISCOVER: {
4004 struct lnet_ioctl_ping_data *discover = arg;
4005 struct lnet_peer *lp;
4007 rc = lnet_discover(discover->ping_id, discover->op_param,
4009 discover->ping_count);
4013 mutex_lock(&the_lnet.ln_api_mutex);
4014 lp = lnet_find_peer(discover->ping_id.nid);
4016 discover->ping_id.nid = lp->lp_primary_nid;
4017 discover->mr_info = lnet_peer_is_multi_rail(lp);
4018 lnet_peer_decref_locked(lp);
4020 mutex_unlock(&the_lnet.ln_api_mutex);
4022 discover->ping_count = rc;
4027 ni = lnet_net2ni_addref(data->ioc_net);
4031 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4034 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4041 EXPORT_SYMBOL(LNetCtl);
4043 void LNetDebugPeer(struct lnet_process_id id)
4045 lnet_debug_peer(id.nid);
4047 EXPORT_SYMBOL(LNetDebugPeer);
4050 * Determine if the specified peer \a nid is on the local node.
4052 * \param nid peer nid to check
4054 * \retval true If peer NID is on the local node.
4055 * \retval false If peer NID is not on the local node.
4057 bool LNetIsPeerLocal(lnet_nid_t nid)
4059 struct lnet_net *net;
4063 cpt = lnet_net_lock_current();
4064 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4065 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4066 if (ni->ni_nid == nid) {
4067 lnet_net_unlock(cpt);
4072 lnet_net_unlock(cpt);
4076 EXPORT_SYMBOL(LNetIsPeerLocal);
4079 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4080 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4082 * \param index Index of the interface to look up.
4083 * \param id On successful return, this location will hold the
4084 * struct lnet_process_id ID of the interface.
4086 * \retval 0 If an interface exists at \a index.
4087 * \retval -ENOENT If no interface has been found.
4090 LNetGetId(unsigned int index, struct lnet_process_id *id)
4093 struct lnet_net *net;
4097 LASSERT(the_lnet.ln_refcount > 0);
4099 cpt = lnet_net_lock_current();
4101 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4102 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4106 id->nid = ni->ni_nid;
4107 id->pid = the_lnet.ln_pid;
4113 lnet_net_unlock(cpt);
4116 EXPORT_SYMBOL(LNetGetId);
4121 struct lnet_handle_md mdh;
4122 struct completion completion;
4126 lnet_ping_event_handler(struct lnet_event *event)
4128 struct ping_data *pd = event->md.user_ptr;
4130 CDEBUG(D_NET, "ping event (%d %d)%s\n",
4131 event->type, event->status,
4132 event->unlinked ? " unlinked" : "");
4134 if (event->status) {
4136 pd->rc = event->status;
4137 } else if (event->type == LNET_EVENT_REPLY) {
4139 pd->rc = event->mlength;
4141 if (event->unlinked)
4142 complete(&pd->completion);
4145 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4146 struct lnet_process_id __user *ids, int n_ids)
4149 struct lnet_md md = { NULL };
4150 struct ping_data pd = { 0 };
4151 struct lnet_ping_buffer *pbuf;
4152 struct lnet_process_id tmpid;
4158 /* n_ids limit is arbitrary */
4159 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4163 * if the user buffer has more space than the lnet_interfaces_max
4164 * then only fill it up to lnet_interfaces_max
4166 if (n_ids > lnet_interfaces_max)
4167 n_ids = lnet_interfaces_max;
4169 if (id.pid == LNET_PID_ANY)
4170 id.pid = LNET_PID_LUSTRE;
4172 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4176 eq = LNetEQAlloc(lnet_ping_event_handler);
4179 CERROR("Can't allocate EQ: %d\n", rc);
4180 goto fail_ping_buffer_decref;
4183 /* initialize md content */
4184 md.start = &pbuf->pb_info;
4185 md.length = LNET_PING_INFO_SIZE(n_ids);
4186 md.threshold = 2; /* GET/REPLY */
4188 md.options = LNET_MD_TRUNCATE;
4192 init_completion(&pd.completion);
4194 rc = LNetMDBind(md, LNET_UNLINK, &pd.mdh);
4196 CERROR("Can't bind MD: %d\n", rc);
4200 rc = LNetGet(LNET_NID_ANY, pd.mdh, id,
4201 LNET_RESERVED_PORTAL,
4202 LNET_PROTO_PING_MATCHBITS, 0, false);
4205 /* Don't CERROR; this could be deliberate! */
4206 rc2 = LNetMDUnlink(pd.mdh);
4209 /* NB must wait for the UNLINK event below... */
4212 if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4213 /* Ensure completion in finite time... */
4214 LNetMDUnlink(pd.mdh);
4215 wait_for_completion(&pd.completion);
4223 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4225 rc = -EPROTO; /* if I can't parse... */
4228 CERROR("%s: ping info too short %d\n",
4229 libcfs_id2str(id), nob);
4233 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4234 lnet_swap_pinginfo(pbuf);
4235 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4236 CERROR("%s: Unexpected magic %08x\n",
4237 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4241 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4242 CERROR("%s: ping w/o NI status: 0x%x\n",
4243 libcfs_id2str(id), pbuf->pb_info.pi_features);
4247 if (nob < LNET_PING_INFO_SIZE(0)) {
4248 CERROR("%s: Short reply %d(%d min)\n",
4250 nob, (int)LNET_PING_INFO_SIZE(0));
4254 if (pbuf->pb_info.pi_nnis < n_ids)
4255 n_ids = pbuf->pb_info.pi_nnis;
4257 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4258 CERROR("%s: Short reply %d(%d expected)\n",
4260 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4264 rc = -EFAULT; /* if I segv in copy_to_user()... */
4266 memset(&tmpid, 0, sizeof(tmpid));
4267 for (i = 0; i < n_ids; i++) {
4268 tmpid.pid = pbuf->pb_info.pi_pid;
4269 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4270 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4273 rc = pbuf->pb_info.pi_nnis;
4278 fail_ping_buffer_decref:
4279 lnet_ping_buffer_decref(pbuf);
4284 lnet_discover(struct lnet_process_id id, __u32 force,
4285 struct lnet_process_id __user *ids, int n_ids)
4287 struct lnet_peer_ni *lpni;
4288 struct lnet_peer_ni *p;
4289 struct lnet_peer *lp;
4290 struct lnet_process_id *buf;
4294 int max_intf = lnet_interfaces_max;
4297 id.nid == LNET_NID_ANY)
4300 if (id.pid == LNET_PID_ANY)
4301 id.pid = LNET_PID_LUSTRE;
4304 * if the user buffer has more space than the max_intf
4305 * then only fill it up to max_intf
4307 if (n_ids > max_intf)
4310 CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4314 cpt = lnet_net_lock_current();
4315 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4322 * Clearing the NIDS_UPTODATE flag ensures the peer will
4323 * be discovered, provided discovery has not been disabled.
4325 lp = lpni->lpni_peer_net->lpn_peer;
4326 spin_lock(&lp->lp_lock);
4327 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4328 /* If the force flag is set, force a PING and PUSH as well. */
4330 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4331 spin_unlock(&lp->lp_lock);
4332 rc = lnet_discover_peer_locked(lpni, cpt, true);
4336 /* Peer may have changed. */
4337 lp = lpni->lpni_peer_net->lpn_peer;
4338 if (lp->lp_nnis < n_ids)
4339 n_ids = lp->lp_nnis;
4343 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4344 buf[i].pid = id.pid;
4345 buf[i].nid = p->lpni_nid;
4350 lnet_net_unlock(cpt);
4353 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4359 lnet_peer_ni_decref_locked(lpni);
4361 lnet_net_unlock(cpt);
4363 CFS_FREE_PTR_ARRAY(buf, n_ids);
4369 * Retrieve peer discovery status.
4371 * \retval 1 if lnet_peer_discovery_disabled is 0
4372 * \retval 0 if lnet_peer_discovery_disabled is 1
4375 LNetGetPeerDiscoveryStatus(void)
4377 return !lnet_peer_discovery_disabled;
4379 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);