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>
41 #include <lnet/lib-lnet.h>
43 #define D_LNI D_CONSOLE
46 * initialize ln_api_mutex statically, since it needs to be used in
47 * discovery_set callback. That module parameter callback can be called
48 * before module init completes. The mutex needs to be ready for use then.
50 struct lnet the_lnet = {
51 .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
52 }; /* THE state of the network */
53 EXPORT_SYMBOL(the_lnet);
55 static char *ip2nets = "";
56 module_param(ip2nets, charp, 0444);
57 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
59 static char *networks = "";
60 module_param(networks, charp, 0444);
61 MODULE_PARM_DESC(networks, "local networks");
63 static char *routes = "";
64 module_param(routes, charp, 0444);
65 MODULE_PARM_DESC(routes, "routes to non-local networks");
67 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
68 module_param(rnet_htable_size, int, 0444);
69 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
71 static int use_tcp_bonding = false;
72 module_param(use_tcp_bonding, int, 0444);
73 MODULE_PARM_DESC(use_tcp_bonding,
74 "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
76 unsigned int lnet_numa_range = 0;
77 module_param(lnet_numa_range, uint, 0444);
78 MODULE_PARM_DESC(lnet_numa_range,
79 "NUMA range to consider during Multi-Rail selection");
82 * lnet_health_sensitivity determines by how much we decrement the health
83 * value on sending error. The value defaults to 100, which means health
84 * interface health is decremented by 100 points every failure.
86 unsigned int lnet_health_sensitivity = 100;
87 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
88 #ifdef HAVE_KERNEL_PARAM_OPS
89 static struct kernel_param_ops param_ops_health_sensitivity = {
90 .set = sensitivity_set,
93 #define param_check_health_sensitivity(name, p) \
94 __param_check(name, p, int)
95 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
97 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
98 &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
100 MODULE_PARM_DESC(lnet_health_sensitivity,
101 "Value to decrement the health value by on error");
104 * lnet_recovery_interval determines how often we should perform recovery
105 * on unhealthy interfaces.
107 unsigned int lnet_recovery_interval = 1;
108 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
109 #ifdef HAVE_KERNEL_PARAM_OPS
110 static struct kernel_param_ops param_ops_recovery_interval = {
111 .set = recovery_interval_set,
112 .get = param_get_int,
114 #define param_check_recovery_interval(name, p) \
115 __param_check(name, p, int)
116 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
118 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
119 &lnet_recovery_interval, S_IRUGO|S_IWUSR);
121 MODULE_PARM_DESC(lnet_recovery_interval,
122 "Interval to recover unhealthy interfaces in seconds");
124 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
125 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
127 static struct kernel_param_ops param_ops_interfaces_max = {
129 .get = param_get_int,
132 #define param_check_interfaces_max(name, p) \
133 __param_check(name, p, int)
135 #ifdef HAVE_KERNEL_PARAM_OPS
136 module_param(lnet_interfaces_max, interfaces_max, 0644);
138 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
139 ¶m_ops_interfaces_max, 0644);
141 MODULE_PARM_DESC(lnet_interfaces_max,
142 "Maximum number of interfaces in a node.");
144 unsigned lnet_peer_discovery_disabled = 0;
145 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
147 static struct kernel_param_ops param_ops_discovery_disabled = {
148 .set = discovery_set,
149 .get = param_get_int,
152 #define param_check_discovery_disabled(name, p) \
153 __param_check(name, p, int)
154 #ifdef HAVE_KERNEL_PARAM_OPS
155 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
157 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
158 ¶m_ops_discovery_disabled, 0644);
160 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
161 "Set to 1 to disable peer discovery on this node.");
163 unsigned int lnet_drop_asym_route;
164 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
166 static struct kernel_param_ops param_ops_drop_asym_route = {
167 .set = drop_asym_route_set,
168 .get = param_get_int,
171 #define param_check_drop_asym_route(name, p) \
172 __param_check(name, p, int)
173 #ifdef HAVE_KERNEL_PARAM_OPS
174 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
176 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
177 ¶m_ops_drop_asym_route, 0644);
179 MODULE_PARM_DESC(lnet_drop_asym_route,
180 "Set to 1 to drop asymmetrical route messages.");
182 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
183 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 50
185 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
186 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
187 #ifdef HAVE_KERNEL_PARAM_OPS
188 static struct kernel_param_ops param_ops_transaction_timeout = {
189 .set = transaction_to_set,
190 .get = param_get_int,
193 #define param_check_transaction_timeout(name, p) \
194 __param_check(name, p, int)
195 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
197 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
198 &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
200 MODULE_PARM_DESC(lnet_transaction_timeout,
201 "Maximum number of seconds to wait for a peer response.");
203 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 2
204 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
205 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
206 #ifdef HAVE_KERNEL_PARAM_OPS
207 static struct kernel_param_ops param_ops_retry_count = {
208 .set = retry_count_set,
209 .get = param_get_int,
212 #define param_check_retry_count(name, p) \
213 __param_check(name, p, int)
214 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
216 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
217 &lnet_retry_count, S_IRUGO|S_IWUSR);
219 MODULE_PARM_DESC(lnet_retry_count,
220 "Maximum number of times to retry transmitting a message");
223 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
224 unsigned int lnet_current_net_count;
227 * This sequence number keeps track of how many times DLC was used to
228 * update the local NIs. It is incremented when a NI is added or
229 * removed and checked when sending a message to determine if there is
230 * a need to re-run the selection algorithm. See lnet_select_pathway()
231 * for more details on its usage.
233 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
235 static int lnet_ping(struct lnet_process_id id, signed long timeout,
236 struct lnet_process_id __user *ids, int n_ids);
238 static int lnet_discover(struct lnet_process_id id, __u32 force,
239 struct lnet_process_id __user *ids, int n_ids);
242 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
245 unsigned *sensitivity = (unsigned *)kp->arg;
248 rc = kstrtoul(val, 0, &value);
250 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
255 * The purpose of locking the api_mutex here is to ensure that
256 * the correct value ends up stored properly.
258 mutex_lock(&the_lnet.ln_api_mutex);
260 if (value > LNET_MAX_HEALTH_VALUE) {
261 mutex_unlock(&the_lnet.ln_api_mutex);
262 CERROR("Invalid health value. Maximum: %d value = %lu\n",
263 LNET_MAX_HEALTH_VALUE, value);
268 * if we're turning on health then use the health timeout
271 if (*sensitivity == 0 && value != 0) {
272 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
273 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
275 * if we're turning off health then use the no health timeout
278 } else if (*sensitivity != 0 && value == 0) {
279 lnet_transaction_timeout =
280 LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
281 lnet_retry_count = 0;
284 *sensitivity = value;
286 mutex_unlock(&the_lnet.ln_api_mutex);
292 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
295 unsigned *interval = (unsigned *)kp->arg;
298 rc = kstrtoul(val, 0, &value);
300 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
305 CERROR("lnet_recovery_interval must be at least 1 second\n");
310 * The purpose of locking the api_mutex here is to ensure that
311 * the correct value ends up stored properly.
313 mutex_lock(&the_lnet.ln_api_mutex);
317 mutex_unlock(&the_lnet.ln_api_mutex);
323 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
326 unsigned *discovery = (unsigned *)kp->arg;
328 struct lnet_ping_buffer *pbuf;
330 rc = kstrtoul(val, 0, &value);
332 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
336 value = (value) ? 1 : 0;
339 * The purpose of locking the api_mutex here is to ensure that
340 * the correct value ends up stored properly.
342 mutex_lock(&the_lnet.ln_api_mutex);
344 if (value == *discovery) {
345 mutex_unlock(&the_lnet.ln_api_mutex);
351 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
352 mutex_unlock(&the_lnet.ln_api_mutex);
356 /* tell peers that discovery setting has changed */
357 lnet_net_lock(LNET_LOCK_EX);
358 pbuf = the_lnet.ln_ping_target;
360 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
362 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
363 lnet_net_unlock(LNET_LOCK_EX);
365 lnet_push_update_to_peers(1);
367 mutex_unlock(&the_lnet.ln_api_mutex);
373 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
376 unsigned int *drop_asym_route = (unsigned int *)kp->arg;
379 rc = kstrtoul(val, 0, &value);
381 CERROR("Invalid module parameter value for "
382 "'lnet_drop_asym_route'\n");
387 * The purpose of locking the api_mutex here is to ensure that
388 * the correct value ends up stored properly.
390 mutex_lock(&the_lnet.ln_api_mutex);
392 if (value == *drop_asym_route) {
393 mutex_unlock(&the_lnet.ln_api_mutex);
397 *drop_asym_route = value;
399 mutex_unlock(&the_lnet.ln_api_mutex);
405 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
408 unsigned *transaction_to = (unsigned *)kp->arg;
411 rc = kstrtoul(val, 0, &value);
413 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
418 * The purpose of locking the api_mutex here is to ensure that
419 * the correct value ends up stored properly.
421 mutex_lock(&the_lnet.ln_api_mutex);
423 if (value < lnet_retry_count || value == 0) {
424 mutex_unlock(&the_lnet.ln_api_mutex);
425 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
426 "Has to be greater than lnet_retry_count (%u)\n",
427 value, lnet_retry_count);
431 if (value == *transaction_to) {
432 mutex_unlock(&the_lnet.ln_api_mutex);
436 *transaction_to = value;
437 if (lnet_retry_count == 0)
438 lnet_lnd_timeout = value;
440 lnet_lnd_timeout = value / lnet_retry_count;
442 mutex_unlock(&the_lnet.ln_api_mutex);
448 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
451 unsigned *retry_count = (unsigned *)kp->arg;
454 rc = kstrtoul(val, 0, &value);
456 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
461 * The purpose of locking the api_mutex here is to ensure that
462 * the correct value ends up stored properly.
464 mutex_lock(&the_lnet.ln_api_mutex);
466 if (lnet_health_sensitivity == 0) {
467 mutex_unlock(&the_lnet.ln_api_mutex);
468 CERROR("Can not set retry_count when health feature is turned off\n");
472 if (value > lnet_transaction_timeout) {
473 mutex_unlock(&the_lnet.ln_api_mutex);
474 CERROR("Invalid value for lnet_retry_count (%lu). "
475 "Has to be smaller than lnet_transaction_timeout (%u)\n",
476 value, lnet_transaction_timeout);
480 *retry_count = value;
483 lnet_lnd_timeout = lnet_transaction_timeout;
485 lnet_lnd_timeout = lnet_transaction_timeout / value;
487 mutex_unlock(&the_lnet.ln_api_mutex);
493 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
497 rc = kstrtoint(val, 0, &value);
499 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
503 if (value < LNET_INTERFACES_MIN) {
504 CWARN("max interfaces provided are too small, setting to %d\n",
505 LNET_INTERFACES_MAX_DEFAULT);
506 value = LNET_INTERFACES_MAX_DEFAULT;
509 *(int *)kp->arg = value;
515 lnet_get_routes(void)
521 lnet_get_networks(void)
526 if (*networks != 0 && *ip2nets != 0) {
527 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
528 "'ip2nets' but not both at once\n");
533 rc = lnet_parse_ip2nets(&nets, ip2nets);
534 return (rc == 0) ? nets : NULL;
544 lnet_init_locks(void)
546 spin_lock_init(&the_lnet.ln_eq_wait_lock);
547 spin_lock_init(&the_lnet.ln_msg_resend_lock);
548 init_waitqueue_head(&the_lnet.ln_eq_waitq);
549 init_completion(&the_lnet.ln_mt_wait_complete);
550 mutex_init(&the_lnet.ln_lnd_mutex);
553 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
554 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
556 struct kmem_cache *lnet_rspt_cachep; /* response tracker cache */
557 struct kmem_cache *lnet_msg_cachep;
560 lnet_slab_setup(void)
562 /* create specific kmem_cache for MEs and small MDs (i.e., originally
563 * allocated in <size-xxx> kmem_cache).
565 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
567 if (!lnet_mes_cachep)
570 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
571 LNET_SMALL_MD_SIZE, 0, 0,
573 if (!lnet_small_mds_cachep)
576 lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
578 if (!lnet_rspt_cachep)
581 lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
583 if (!lnet_msg_cachep)
590 lnet_slab_cleanup(void)
592 if (lnet_msg_cachep) {
593 kmem_cache_destroy(lnet_msg_cachep);
594 lnet_msg_cachep = NULL;
598 if (lnet_rspt_cachep) {
599 kmem_cache_destroy(lnet_rspt_cachep);
600 lnet_rspt_cachep = NULL;
603 if (lnet_small_mds_cachep) {
604 kmem_cache_destroy(lnet_small_mds_cachep);
605 lnet_small_mds_cachep = NULL;
608 if (lnet_mes_cachep) {
609 kmem_cache_destroy(lnet_mes_cachep);
610 lnet_mes_cachep = NULL;
615 lnet_create_remote_nets_table(void)
618 struct list_head *hash;
620 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
621 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
622 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
624 CERROR("Failed to create remote nets hash table\n");
628 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
629 INIT_LIST_HEAD(&hash[i]);
630 the_lnet.ln_remote_nets_hash = hash;
635 lnet_destroy_remote_nets_table(void)
639 if (the_lnet.ln_remote_nets_hash == NULL)
642 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
643 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
645 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
646 LNET_REMOTE_NETS_HASH_SIZE *
647 sizeof(the_lnet.ln_remote_nets_hash[0]));
648 the_lnet.ln_remote_nets_hash = NULL;
652 lnet_destroy_locks(void)
654 if (the_lnet.ln_res_lock != NULL) {
655 cfs_percpt_lock_free(the_lnet.ln_res_lock);
656 the_lnet.ln_res_lock = NULL;
659 if (the_lnet.ln_net_lock != NULL) {
660 cfs_percpt_lock_free(the_lnet.ln_net_lock);
661 the_lnet.ln_net_lock = NULL;
666 lnet_create_locks(void)
670 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
671 if (the_lnet.ln_res_lock == NULL)
674 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
675 if (the_lnet.ln_net_lock == NULL)
681 lnet_destroy_locks();
685 static void lnet_assert_wire_constants(void)
687 /* Wire protocol assertions generated by 'wirecheck'
688 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
689 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
690 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
693 BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
694 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
695 BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
696 BUILD_BUG_ON(LNET_MSG_ACK != 0);
697 BUILD_BUG_ON(LNET_MSG_PUT != 1);
698 BUILD_BUG_ON(LNET_MSG_GET != 2);
699 BUILD_BUG_ON(LNET_MSG_REPLY != 3);
700 BUILD_BUG_ON(LNET_MSG_HELLO != 4);
702 /* Checks for struct lnet_handle_wire */
703 BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
704 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
705 wh_interface_cookie) != 0);
706 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
707 BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
708 wh_object_cookie) != 8);
709 BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
711 /* Checks for struct struct lnet_magicversion */
712 BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
713 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
714 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
715 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
716 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
717 BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
718 version_minor) != 6);
719 BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
721 /* Checks for struct struct lnet_hdr */
722 BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
723 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
724 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
725 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
726 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
727 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
728 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
729 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
730 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
731 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
732 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
733 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
734 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
735 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
736 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
739 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
740 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
741 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
742 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
743 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
744 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
747 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
748 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
749 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
750 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
751 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
752 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
753 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
754 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
755 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
756 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
759 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
760 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
761 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
762 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
763 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
764 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
765 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
766 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
767 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
768 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
771 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
772 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
775 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
776 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
777 BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
778 BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
780 /* Checks for struct lnet_ni_status and related constants */
781 BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
782 BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
783 BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
785 /* Checks for struct lnet_ni_status */
786 BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
787 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
788 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
789 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
790 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
791 BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
792 BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
794 /* Checks for struct lnet_ping_info and related constants */
795 BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
796 BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
797 BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
798 BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
799 BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
800 BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
801 BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
802 BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
804 /* Checks for struct lnet_ping_info */
805 BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
806 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
807 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
808 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
809 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
810 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
811 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
812 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
813 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
814 BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
815 BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
818 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
820 const struct lnet_lnd *lnd;
822 /* holding lnd mutex */
823 if (type >= NUM_LNDS)
825 lnd = the_lnet.ln_lnds[type];
826 LASSERT(!lnd || lnd->lnd_type == type);
832 lnet_get_lnd_timeout(void)
834 return lnet_lnd_timeout;
836 EXPORT_SYMBOL(lnet_get_lnd_timeout);
839 lnet_register_lnd(const struct lnet_lnd *lnd)
841 mutex_lock(&the_lnet.ln_lnd_mutex);
843 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
844 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
846 the_lnet.ln_lnds[lnd->lnd_type] = lnd;
848 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
850 mutex_unlock(&the_lnet.ln_lnd_mutex);
852 EXPORT_SYMBOL(lnet_register_lnd);
855 lnet_unregister_lnd(const struct lnet_lnd *lnd)
857 mutex_lock(&the_lnet.ln_lnd_mutex);
859 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
861 the_lnet.ln_lnds[lnd->lnd_type] = NULL;
862 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
864 mutex_unlock(&the_lnet.ln_lnd_mutex);
866 EXPORT_SYMBOL(lnet_unregister_lnd);
869 lnet_counters_get_common(struct lnet_counters_common *common)
871 struct lnet_counters *ctr;
874 memset(common, 0, sizeof(*common));
876 lnet_net_lock(LNET_LOCK_EX);
878 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
879 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
880 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
881 common->lcc_errors += ctr->lct_common.lcc_errors;
882 common->lcc_send_count += ctr->lct_common.lcc_send_count;
883 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
884 common->lcc_route_count += ctr->lct_common.lcc_route_count;
885 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
886 common->lcc_send_length += ctr->lct_common.lcc_send_length;
887 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
888 common->lcc_route_length += ctr->lct_common.lcc_route_length;
889 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
891 lnet_net_unlock(LNET_LOCK_EX);
893 EXPORT_SYMBOL(lnet_counters_get_common);
896 lnet_counters_get(struct lnet_counters *counters)
898 struct lnet_counters *ctr;
899 struct lnet_counters_health *health = &counters->lct_health;
902 memset(counters, 0, sizeof(*counters));
904 lnet_counters_get_common(&counters->lct_common);
906 lnet_net_lock(LNET_LOCK_EX);
908 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
909 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
910 health->lch_resend_count += ctr->lct_health.lch_resend_count;
911 health->lch_response_timeout_count +=
912 ctr->lct_health.lch_response_timeout_count;
913 health->lch_local_interrupt_count +=
914 ctr->lct_health.lch_local_interrupt_count;
915 health->lch_local_dropped_count +=
916 ctr->lct_health.lch_local_dropped_count;
917 health->lch_local_aborted_count +=
918 ctr->lct_health.lch_local_aborted_count;
919 health->lch_local_no_route_count +=
920 ctr->lct_health.lch_local_no_route_count;
921 health->lch_local_timeout_count +=
922 ctr->lct_health.lch_local_timeout_count;
923 health->lch_local_error_count +=
924 ctr->lct_health.lch_local_error_count;
925 health->lch_remote_dropped_count +=
926 ctr->lct_health.lch_remote_dropped_count;
927 health->lch_remote_error_count +=
928 ctr->lct_health.lch_remote_error_count;
929 health->lch_remote_timeout_count +=
930 ctr->lct_health.lch_remote_timeout_count;
931 health->lch_network_timeout_count +=
932 ctr->lct_health.lch_network_timeout_count;
934 lnet_net_unlock(LNET_LOCK_EX);
936 EXPORT_SYMBOL(lnet_counters_get);
939 lnet_counters_reset(void)
941 struct lnet_counters *counters;
944 lnet_net_lock(LNET_LOCK_EX);
946 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
947 memset(counters, 0, sizeof(struct lnet_counters));
949 lnet_net_unlock(LNET_LOCK_EX);
953 lnet_res_type2str(int type)
958 case LNET_COOKIE_TYPE_MD:
960 case LNET_COOKIE_TYPE_ME:
962 case LNET_COOKIE_TYPE_EQ:
968 lnet_res_container_cleanup(struct lnet_res_container *rec)
972 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
975 while (!list_empty(&rec->rec_active)) {
976 struct list_head *e = rec->rec_active.next;
979 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
980 lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
982 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
983 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
985 } else { /* NB: Active MEs should be attached on portals */
992 /* Found alive MD/ME/EQ, user really should unlink/free
993 * all of them before finalize LNet, but if someone didn't,
994 * we have to recycle garbage for him */
995 CERROR("%d active elements on exit of %s container\n",
996 count, lnet_res_type2str(rec->rec_type));
999 if (rec->rec_lh_hash != NULL) {
1000 LIBCFS_FREE(rec->rec_lh_hash,
1001 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1002 rec->rec_lh_hash = NULL;
1005 rec->rec_type = 0; /* mark it as finalized */
1009 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1014 LASSERT(rec->rec_type == 0);
1016 rec->rec_type = type;
1017 INIT_LIST_HEAD(&rec->rec_active);
1019 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1021 /* Arbitrary choice of hash table size */
1022 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1023 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1024 if (rec->rec_lh_hash == NULL) {
1029 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1030 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1035 CERROR("Failed to setup %s resource container\n",
1036 lnet_res_type2str(type));
1037 lnet_res_container_cleanup(rec);
1042 lnet_res_containers_destroy(struct lnet_res_container **recs)
1044 struct lnet_res_container *rec;
1047 cfs_percpt_for_each(rec, i, recs)
1048 lnet_res_container_cleanup(rec);
1050 cfs_percpt_free(recs);
1053 static struct lnet_res_container **
1054 lnet_res_containers_create(int type)
1056 struct lnet_res_container **recs;
1057 struct lnet_res_container *rec;
1061 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1063 CERROR("Failed to allocate %s resource containers\n",
1064 lnet_res_type2str(type));
1068 cfs_percpt_for_each(rec, i, recs) {
1069 rc = lnet_res_container_setup(rec, i, type);
1071 lnet_res_containers_destroy(recs);
1079 struct lnet_libhandle *
1080 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1082 /* ALWAYS called with lnet_res_lock held */
1083 struct list_head *head;
1084 struct lnet_libhandle *lh;
1087 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1090 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1091 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1093 list_for_each_entry(lh, head, lh_hash_chain) {
1094 if (lh->lh_cookie == cookie)
1102 lnet_res_lh_initialize(struct lnet_res_container *rec,
1103 struct lnet_libhandle *lh)
1105 /* ALWAYS called with lnet_res_lock held */
1106 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1109 lh->lh_cookie = rec->rec_lh_cookie;
1110 rec->rec_lh_cookie += 1 << ibits;
1112 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1114 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1118 lnet_create_array_of_queues(void)
1120 struct list_head **qs;
1121 struct list_head *q;
1124 qs = cfs_percpt_alloc(lnet_cpt_table(),
1125 sizeof(struct list_head));
1127 CERROR("Failed to allocate queues\n");
1131 cfs_percpt_for_each(q, i, qs)
1137 static int lnet_unprepare(void);
1140 lnet_prepare(lnet_pid_t requested_pid)
1142 /* Prepare to bring up the network */
1143 struct lnet_res_container **recs;
1146 if (requested_pid == LNET_PID_ANY) {
1147 /* Don't instantiate LNET just for me */
1151 LASSERT(the_lnet.ln_refcount == 0);
1153 the_lnet.ln_routing = 0;
1155 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1156 the_lnet.ln_pid = requested_pid;
1158 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1159 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1160 INIT_LIST_HEAD(&the_lnet.ln_nets);
1161 INIT_LIST_HEAD(&the_lnet.ln_routers);
1162 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1163 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1164 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1165 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1166 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1167 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1168 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1169 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1170 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1171 init_completion(&the_lnet.ln_started);
1173 rc = lnet_slab_setup();
1177 rc = lnet_create_remote_nets_table();
1182 * NB the interface cookie in wire handles guards against delayed
1183 * replies and ACKs appearing valid after reboot.
1185 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1187 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1188 sizeof(struct lnet_counters));
1189 if (the_lnet.ln_counters == NULL) {
1190 CERROR("Failed to allocate counters for LNet\n");
1195 rc = lnet_peer_tables_create();
1199 rc = lnet_msg_containers_create();
1203 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1204 LNET_COOKIE_TYPE_EQ);
1208 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1214 the_lnet.ln_md_containers = recs;
1216 rc = lnet_portals_create();
1218 CERROR("Failed to create portals for LNet: %d\n", rc);
1222 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1223 if (!the_lnet.ln_mt_zombie_rstqs) {
1236 lnet_unprepare (void)
1240 /* NB no LNET_LOCK since this is the last reference. All LND instances
1241 * have shut down already, so it is safe to unlink and free all
1242 * descriptors, even those that appear committed to a network op (eg MD
1243 * with non-zero pending count) */
1245 lnet_fail_nid(LNET_NID_ANY, 0);
1247 LASSERT(the_lnet.ln_refcount == 0);
1248 LASSERT(list_empty(&the_lnet.ln_test_peers));
1249 LASSERT(list_empty(&the_lnet.ln_nets));
1251 if (the_lnet.ln_mt_zombie_rstqs) {
1252 lnet_clean_zombie_rstqs();
1253 the_lnet.ln_mt_zombie_rstqs = NULL;
1256 if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1257 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1258 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1262 lnet_portals_destroy();
1264 if (the_lnet.ln_md_containers != NULL) {
1265 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1266 the_lnet.ln_md_containers = NULL;
1269 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1271 lnet_msg_containers_destroy();
1273 lnet_rtrpools_free(0);
1275 if (the_lnet.ln_counters != NULL) {
1276 cfs_percpt_free(the_lnet.ln_counters);
1277 the_lnet.ln_counters = NULL;
1279 lnet_destroy_remote_nets_table();
1280 lnet_slab_cleanup();
1286 lnet_net2ni_locked(__u32 net_id, int cpt)
1289 struct lnet_net *net;
1291 LASSERT(cpt != LNET_LOCK_EX);
1293 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1294 if (net->net_id == net_id) {
1295 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1305 lnet_net2ni_addref(__u32 net)
1310 ni = lnet_net2ni_locked(net, 0);
1312 lnet_ni_addref_locked(ni, 0);
1317 EXPORT_SYMBOL(lnet_net2ni_addref);
1320 lnet_get_net_locked(__u32 net_id)
1322 struct lnet_net *net;
1324 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1325 if (net->net_id == net_id)
1333 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1338 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1343 val = hash_long(key, LNET_CPT_BITS);
1344 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1348 return (unsigned int)(key + val + (val >> 1)) % number;
1352 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1354 struct lnet_net *net;
1356 /* must called with hold of lnet_net_lock */
1357 if (LNET_CPT_NUMBER == 1)
1358 return 0; /* the only one */
1361 * If NI is provided then use the CPT identified in the NI cpt
1362 * list if one exists. If one doesn't exist, then that NI is
1363 * associated with all CPTs and it follows that the net it belongs
1364 * to is implicitly associated with all CPTs, so just hash the nid
1368 if (ni->ni_cpts != NULL)
1369 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1372 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1375 /* no NI provided so look at the net */
1376 net = lnet_get_net_locked(LNET_NIDNET(nid));
1378 if (net != NULL && net->net_cpts != NULL) {
1379 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1382 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1386 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1391 if (LNET_CPT_NUMBER == 1)
1392 return 0; /* the only one */
1394 cpt = lnet_net_lock_current();
1396 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1398 lnet_net_unlock(cpt);
1402 EXPORT_SYMBOL(lnet_cpt_of_nid);
1405 lnet_islocalnet_locked(__u32 net_id)
1407 struct lnet_net *net;
1410 net = lnet_get_net_locked(net_id);
1412 local = net != NULL;
1418 lnet_islocalnet(__u32 net_id)
1423 cpt = lnet_net_lock_current();
1425 local = lnet_islocalnet_locked(net_id);
1427 lnet_net_unlock(cpt);
1433 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1435 struct lnet_net *net;
1438 LASSERT(cpt != LNET_LOCK_EX);
1440 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1441 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1442 if (ni->ni_nid == nid)
1451 lnet_nid2ni_addref(lnet_nid_t nid)
1456 ni = lnet_nid2ni_locked(nid, 0);
1458 lnet_ni_addref_locked(ni, 0);
1463 EXPORT_SYMBOL(lnet_nid2ni_addref);
1466 lnet_islocalnid(lnet_nid_t nid)
1471 cpt = lnet_net_lock_current();
1472 ni = lnet_nid2ni_locked(nid, cpt);
1473 lnet_net_unlock(cpt);
1479 lnet_count_acceptor_nets(void)
1481 /* Return the # of NIs that need the acceptor. */
1483 struct lnet_net *net;
1486 cpt = lnet_net_lock_current();
1487 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1488 /* all socklnd type networks should have the acceptor
1490 if (net->net_lnd->lnd_accept != NULL)
1494 lnet_net_unlock(cpt);
1499 struct lnet_ping_buffer *
1500 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1502 struct lnet_ping_buffer *pbuf;
1504 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1506 pbuf->pb_nnis = nnis;
1507 atomic_set(&pbuf->pb_refcnt, 1);
1514 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1516 LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1517 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1520 static struct lnet_ping_buffer *
1521 lnet_ping_target_create(int nnis)
1523 struct lnet_ping_buffer *pbuf;
1525 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1527 CERROR("Can't allocate ping source [%d]\n", nnis);
1531 pbuf->pb_info.pi_nnis = nnis;
1532 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1533 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1534 pbuf->pb_info.pi_features =
1535 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1541 lnet_get_net_ni_count_locked(struct lnet_net *net)
1546 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1553 lnet_get_net_ni_count_pre(struct lnet_net *net)
1558 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1565 lnet_get_ni_count(void)
1568 struct lnet_net *net;
1573 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1574 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1584 lnet_get_net_count(void)
1586 struct lnet_net *net;
1591 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1601 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1603 struct lnet_ni_status *stat;
1607 __swab32s(&pbuf->pb_info.pi_magic);
1608 __swab32s(&pbuf->pb_info.pi_features);
1609 __swab32s(&pbuf->pb_info.pi_pid);
1610 __swab32s(&pbuf->pb_info.pi_nnis);
1611 nnis = pbuf->pb_info.pi_nnis;
1612 if (nnis > pbuf->pb_nnis)
1613 nnis = pbuf->pb_nnis;
1614 for (i = 0; i < nnis; i++) {
1615 stat = &pbuf->pb_info.pi_ni[i];
1616 __swab64s(&stat->ns_nid);
1617 __swab32s(&stat->ns_status);
1622 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1626 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1628 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1630 /* Loopback is guaranteed to be present */
1631 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1633 if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1639 lnet_ping_target_destroy(void)
1641 struct lnet_net *net;
1644 lnet_net_lock(LNET_LOCK_EX);
1646 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1647 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1649 ni->ni_status = NULL;
1654 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1655 the_lnet.ln_ping_target = NULL;
1657 lnet_net_unlock(LNET_LOCK_EX);
1661 lnet_ping_target_event_handler(struct lnet_event *event)
1663 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1665 if (event->unlinked)
1666 lnet_ping_buffer_decref(pbuf);
1670 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1671 struct lnet_handle_md *ping_mdh,
1672 int ni_count, bool set_eq)
1674 struct lnet_process_id id = {
1675 .nid = LNET_NID_ANY,
1679 struct lnet_md md = { NULL };
1683 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1684 &the_lnet.ln_ping_target_eq);
1686 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1691 *ppbuf = lnet_ping_target_create(ni_count);
1692 if (*ppbuf == NULL) {
1697 /* Ping target ME/MD */
1698 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1699 LNET_PROTO_PING_MATCHBITS, 0,
1700 LNET_UNLINK, LNET_INS_AFTER);
1703 CERROR("Can't create ping target ME: %d\n", rc);
1704 goto fail_decref_ping_buffer;
1707 /* initialize md content */
1708 md.start = &(*ppbuf)->pb_info;
1709 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1710 md.threshold = LNET_MD_THRESH_INF;
1712 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1713 LNET_MD_MANAGE_REMOTE;
1714 md.eq_handle = the_lnet.ln_ping_target_eq;
1715 md.user_ptr = *ppbuf;
1717 rc = LNetMDAttach(me, md, LNET_RETAIN, ping_mdh);
1719 CERROR("Can't attach ping target MD: %d\n", rc);
1720 goto fail_unlink_ping_me;
1722 lnet_ping_buffer_addref(*ppbuf);
1726 fail_unlink_ping_me:
1728 fail_decref_ping_buffer:
1729 LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1730 lnet_ping_buffer_decref(*ppbuf);
1734 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1741 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1742 struct lnet_handle_md *ping_mdh)
1744 sigset_t blocked = cfs_block_allsigs();
1746 LNetMDUnlink(*ping_mdh);
1747 LNetInvalidateMDHandle(ping_mdh);
1749 /* NB the MD could be busy; this just starts the unlink */
1750 while (atomic_read(&pbuf->pb_refcnt) > 1) {
1751 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1752 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1755 cfs_restore_sigs(blocked);
1759 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1762 struct lnet_net *net;
1763 struct lnet_ni_status *ns;
1768 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1769 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1770 LASSERT(i < pbuf->pb_nnis);
1772 ns = &pbuf->pb_info.pi_ni[i];
1774 ns->ns_nid = ni->ni_nid;
1777 ns->ns_status = (ni->ni_status != NULL) ?
1778 ni->ni_status->ns_status :
1787 * We (ab)use the ns_status of the loopback interface to
1788 * transmit the sequence number. The first interface listed
1789 * must be the loopback interface.
1791 rc = lnet_ping_info_validate(&pbuf->pb_info);
1793 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1796 LNET_PING_BUFFER_SEQNO(pbuf) =
1797 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1801 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1802 struct lnet_handle_md ping_mdh)
1804 struct lnet_ping_buffer *old_pbuf = NULL;
1805 struct lnet_handle_md old_ping_md;
1807 /* switch the NIs to point to the new ping info created */
1808 lnet_net_lock(LNET_LOCK_EX);
1810 if (!the_lnet.ln_routing)
1811 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1812 if (!lnet_peer_discovery_disabled)
1813 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1815 /* Ensure only known feature bits have been set. */
1816 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1817 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1819 lnet_ping_target_install_locked(pbuf);
1821 if (the_lnet.ln_ping_target) {
1822 old_pbuf = the_lnet.ln_ping_target;
1823 old_ping_md = the_lnet.ln_ping_target_md;
1825 the_lnet.ln_ping_target_md = ping_mdh;
1826 the_lnet.ln_ping_target = pbuf;
1828 lnet_net_unlock(LNET_LOCK_EX);
1831 /* unlink and free the old ping info */
1832 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1833 lnet_ping_buffer_decref(old_pbuf);
1836 lnet_push_update_to_peers(0);
1840 lnet_ping_target_fini(void)
1844 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1845 &the_lnet.ln_ping_target_md);
1847 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1850 lnet_ping_target_destroy();
1853 /* Resize the push target. */
1854 int lnet_push_target_resize(void)
1856 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1857 struct lnet_md md = { NULL };
1859 struct lnet_handle_md mdh;
1860 struct lnet_handle_md old_mdh;
1861 struct lnet_ping_buffer *pbuf;
1862 struct lnet_ping_buffer *old_pbuf;
1863 int nnis = the_lnet.ln_push_target_nnis;
1871 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1877 me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1878 LNET_PROTO_PING_MATCHBITS, 0,
1879 LNET_UNLINK, LNET_INS_AFTER);
1883 CERROR("Can't create push target ME: %d\n", rc);
1884 goto fail_decref_pbuf;
1887 /* initialize md content */
1888 md.start = &pbuf->pb_info;
1889 md.length = LNET_PING_INFO_SIZE(nnis);
1890 md.threshold = LNET_MD_THRESH_INF;
1892 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1893 LNET_MD_MANAGE_REMOTE;
1895 md.eq_handle = the_lnet.ln_push_target_eq;
1897 rc = LNetMDAttach(me, md, LNET_RETAIN, &mdh);
1899 CERROR("Can't attach push MD: %d\n", rc);
1900 goto fail_unlink_me;
1902 lnet_ping_buffer_addref(pbuf);
1904 lnet_net_lock(LNET_LOCK_EX);
1905 old_pbuf = the_lnet.ln_push_target;
1906 old_mdh = the_lnet.ln_push_target_md;
1907 the_lnet.ln_push_target = pbuf;
1908 the_lnet.ln_push_target_md = mdh;
1909 lnet_net_unlock(LNET_LOCK_EX);
1912 LNetMDUnlink(old_mdh);
1913 lnet_ping_buffer_decref(old_pbuf);
1916 if (nnis < the_lnet.ln_push_target_nnis)
1919 CDEBUG(D_NET, "nnis %d success\n", nnis);
1926 lnet_ping_buffer_decref(pbuf);
1928 CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1932 static void lnet_push_target_event_handler(struct lnet_event *ev)
1934 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1936 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1937 lnet_swap_pinginfo(pbuf);
1939 lnet_peer_push_event(ev);
1941 lnet_ping_buffer_decref(pbuf);
1944 /* Initialize the push target. */
1945 static int lnet_push_target_init(void)
1949 if (the_lnet.ln_push_target)
1952 rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1953 &the_lnet.ln_push_target_eq);
1955 CERROR("Can't allocated push target EQ: %d\n", rc);
1959 /* Start at the required minimum, we'll enlarge if required. */
1960 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1962 rc = lnet_push_target_resize();
1965 LNetEQFree(the_lnet.ln_push_target_eq);
1966 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1972 /* Clean up the push target. */
1973 static void lnet_push_target_fini(void)
1975 if (!the_lnet.ln_push_target)
1978 /* Unlink and invalidate to prevent new references. */
1979 LNetMDUnlink(the_lnet.ln_push_target_md);
1980 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1982 /* Wait for the unlink to complete. */
1983 while (atomic_read(&the_lnet.ln_push_target->pb_refcnt) > 1) {
1984 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1985 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1988 lnet_ping_buffer_decref(the_lnet.ln_push_target);
1989 the_lnet.ln_push_target = NULL;
1990 the_lnet.ln_push_target_nnis = 0;
1992 LNetEQFree(the_lnet.ln_push_target_eq);
1993 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1997 lnet_ni_tq_credits(struct lnet_ni *ni)
2001 LASSERT(ni->ni_ncpts >= 1);
2003 if (ni->ni_ncpts == 1)
2004 return ni->ni_net->net_tunables.lct_max_tx_credits;
2006 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2007 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2008 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2014 lnet_ni_unlink_locked(struct lnet_ni *ni)
2016 /* move it to zombie list and nobody can find it anymore */
2017 LASSERT(!list_empty(&ni->ni_netlist));
2018 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2019 lnet_ni_decref_locked(ni, 0);
2023 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2028 struct list_head *zombie_list = &net->net_ni_zombie;
2031 * Now wait for the NIs I just nuked to show up on the zombie
2032 * list and shut them down in guaranteed thread context
2035 while (!list_empty(zombie_list)) {
2039 ni = list_entry(zombie_list->next,
2040 struct lnet_ni, ni_netlist);
2041 list_del_init(&ni->ni_netlist);
2042 /* the ni should be in deleting state. If it's not it's
2044 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2045 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2048 /* still busy, add it back to zombie list */
2049 list_add(&ni->ni_netlist, zombie_list);
2053 if (!list_empty(&ni->ni_netlist)) {
2054 lnet_net_unlock(LNET_LOCK_EX);
2056 if ((i & (-i)) == i) {
2058 "Waiting for zombie LNI %s\n",
2059 libcfs_nid2str(ni->ni_nid));
2061 schedule_timeout_uninterruptible(cfs_time_seconds(1));
2062 lnet_net_lock(LNET_LOCK_EX);
2066 lnet_net_unlock(LNET_LOCK_EX);
2068 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2070 LASSERT(!in_interrupt());
2071 /* Holding the mutex makes it safe for lnd_shutdown
2072 * to call module_put(). Module unload cannot finish
2073 * until lnet_unregister_lnd() completes, and that
2074 * requires the mutex.
2076 mutex_lock(&the_lnet.ln_lnd_mutex);
2077 (net->net_lnd->lnd_shutdown)(ni);
2078 mutex_unlock(&the_lnet.ln_lnd_mutex);
2081 CDEBUG(D_LNI, "Removed LNI %s\n",
2082 libcfs_nid2str(ni->ni_nid));
2086 lnet_net_lock(LNET_LOCK_EX);
2090 /* shutdown down the NI and release refcount */
2092 lnet_shutdown_lndni(struct lnet_ni *ni)
2095 struct lnet_net *net = ni->ni_net;
2097 lnet_net_lock(LNET_LOCK_EX);
2099 ni->ni_state = LNET_NI_STATE_DELETING;
2101 lnet_ni_unlink_locked(ni);
2102 lnet_incr_dlc_seq();
2103 lnet_net_unlock(LNET_LOCK_EX);
2105 /* clear messages for this NI on the lazy portal */
2106 for (i = 0; i < the_lnet.ln_nportals; i++)
2107 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2109 lnet_net_lock(LNET_LOCK_EX);
2110 lnet_clear_zombies_nis_locked(net);
2111 lnet_net_unlock(LNET_LOCK_EX);
2115 lnet_shutdown_lndnet(struct lnet_net *net)
2119 lnet_net_lock(LNET_LOCK_EX);
2121 list_del_init(&net->net_list);
2123 while (!list_empty(&net->net_ni_list)) {
2124 ni = list_entry(net->net_ni_list.next,
2125 struct lnet_ni, ni_netlist);
2126 lnet_net_unlock(LNET_LOCK_EX);
2127 lnet_shutdown_lndni(ni);
2128 lnet_net_lock(LNET_LOCK_EX);
2131 lnet_net_unlock(LNET_LOCK_EX);
2133 /* Do peer table cleanup for this net */
2134 lnet_peer_tables_cleanup(net);
2140 lnet_shutdown_lndnets(void)
2142 struct lnet_net *net;
2144 struct lnet_msg *msg, *tmp;
2146 /* NB called holding the global mutex */
2148 /* All quiet on the API front */
2149 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2150 LASSERT(the_lnet.ln_refcount == 0);
2152 lnet_net_lock(LNET_LOCK_EX);
2153 the_lnet.ln_state = LNET_STATE_STOPPING;
2155 while (!list_empty(&the_lnet.ln_nets)) {
2157 * move the nets to the zombie list to avoid them being
2158 * picked up for new work. LONET is also included in the
2159 * Nets that will be moved to the zombie list
2161 net = list_entry(the_lnet.ln_nets.next,
2162 struct lnet_net, net_list);
2163 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2166 /* Drop the cached loopback Net. */
2167 if (the_lnet.ln_loni != NULL) {
2168 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2169 the_lnet.ln_loni = NULL;
2171 lnet_net_unlock(LNET_LOCK_EX);
2173 /* iterate through the net zombie list and delete each net */
2174 while (!list_empty(&the_lnet.ln_net_zombie)) {
2175 net = list_entry(the_lnet.ln_net_zombie.next,
2176 struct lnet_net, net_list);
2177 lnet_shutdown_lndnet(net);
2180 spin_lock(&the_lnet.ln_msg_resend_lock);
2181 list_splice(&the_lnet.ln_msg_resend, &resend);
2182 spin_unlock(&the_lnet.ln_msg_resend_lock);
2184 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2185 list_del_init(&msg->msg_list);
2186 msg->msg_no_resend = true;
2187 lnet_finalize(msg, -ECANCELED);
2190 lnet_net_lock(LNET_LOCK_EX);
2191 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2192 lnet_net_unlock(LNET_LOCK_EX);
2196 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2199 struct lnet_tx_queue *tq;
2201 struct lnet_net *net = ni->ni_net;
2203 mutex_lock(&the_lnet.ln_lnd_mutex);
2206 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2207 ni->ni_lnd_tunables_set = true;
2210 rc = (net->net_lnd->lnd_startup)(ni);
2212 mutex_unlock(&the_lnet.ln_lnd_mutex);
2215 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2216 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2221 ni->ni_state = LNET_NI_STATE_ACTIVE;
2224 /* We keep a reference on the loopback net through the loopback NI */
2225 if (net->net_lnd->lnd_type == LOLND) {
2227 LASSERT(the_lnet.ln_loni == NULL);
2228 the_lnet.ln_loni = ni;
2229 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2230 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2231 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2232 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2236 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2237 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2238 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2239 libcfs_lnd2str(net->net_lnd->lnd_type),
2240 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2242 /* shutdown the NI since if we get here then it must've already
2245 lnet_shutdown_lndni(ni);
2249 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2250 tq->tq_credits_min =
2251 tq->tq_credits_max =
2252 tq->tq_credits = lnet_ni_tq_credits(ni);
2255 atomic_set(&ni->ni_tx_credits,
2256 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2257 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2259 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2260 libcfs_nid2str(ni->ni_nid),
2261 ni->ni_net->net_tunables.lct_peer_tx_credits,
2262 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2263 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2264 ni->ni_net->net_tunables.lct_peer_timeout);
2273 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2276 struct lnet_net *net_l = NULL;
2277 LIST_HEAD(local_ni_list);
2281 const struct lnet_lnd *lnd;
2283 net->net_tunables.lct_peer_timeout;
2285 net->net_tunables.lct_max_tx_credits;
2286 int peerrtrcredits =
2287 net->net_tunables.lct_peer_rtr_credits;
2290 * make sure that this net is unique. If it isn't then
2291 * we are adding interfaces to an already existing network, and
2292 * 'net' is just a convenient way to pass in the list.
2293 * if it is unique we need to find the LND and load it if
2296 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2297 lnd_type = LNET_NETTYP(net->net_id);
2299 mutex_lock(&the_lnet.ln_lnd_mutex);
2300 lnd = lnet_find_lnd_by_type(lnd_type);
2303 mutex_unlock(&the_lnet.ln_lnd_mutex);
2304 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2305 mutex_lock(&the_lnet.ln_lnd_mutex);
2307 lnd = lnet_find_lnd_by_type(lnd_type);
2309 mutex_unlock(&the_lnet.ln_lnd_mutex);
2310 CERROR("Can't load LND %s, module %s, rc=%d\n",
2311 libcfs_lnd2str(lnd_type),
2312 libcfs_lnd2modname(lnd_type), rc);
2313 #ifndef HAVE_MODULE_LOADING_SUPPORT
2314 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2315 "compiled with kernel module "
2316 "loading support.");
2325 mutex_unlock(&the_lnet.ln_lnd_mutex);
2331 * net_l: if the network being added is unique then net_l
2332 * will point to that network
2333 * if the network being added is not unique then
2334 * net_l points to the existing network.
2336 * When we enter the loop below, we'll pick NIs off he
2337 * network beign added and start them up, then add them to
2338 * a local ni list. Once we've successfully started all
2339 * the NIs then we join the local NI list (of started up
2340 * networks) with the net_l->net_ni_list, which should
2341 * point to the correct network to add the new ni list to
2343 * If any of the new NIs fail to start up, then we want to
2344 * iterate through the local ni list, which should include
2345 * any NIs which were successfully started up, and shut
2348 * After than we want to delete the network being added,
2349 * to avoid a memory leak.
2353 * When a network uses TCP bonding then all its interfaces
2354 * must be specified when the network is first defined: the
2355 * TCP bonding code doesn't allow for interfaces to be added
2358 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2359 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2364 while (!list_empty(&net->net_ni_added)) {
2365 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2367 list_del_init(&ni->ni_netlist);
2369 /* make sure that the the NI we're about to start
2370 * up is actually unique. if it's not fail. */
2371 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2372 ni->ni_interfaces[0])) {
2377 /* adjust the pointer the parent network, just in case it
2378 * the net is a duplicate */
2381 rc = lnet_startup_lndni(ni, tun);
2387 list_add_tail(&ni->ni_netlist, &local_ni_list);
2392 lnet_net_lock(LNET_LOCK_EX);
2393 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2394 lnet_incr_dlc_seq();
2395 lnet_net_unlock(LNET_LOCK_EX);
2397 /* if the network is not unique then we don't want to keep
2398 * it around after we're done. Free it. Otherwise add that
2399 * net to the global the_lnet.ln_nets */
2400 if (net_l != net && net_l != NULL) {
2402 * TODO - note. currently the tunables can not be updated
2408 * restore tunables after it has been overwitten by the
2411 if (peer_timeout != -1)
2412 net->net_tunables.lct_peer_timeout = peer_timeout;
2413 if (maxtxcredits != -1)
2414 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2415 if (peerrtrcredits != -1)
2416 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2418 lnet_net_lock(LNET_LOCK_EX);
2419 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2420 lnet_net_unlock(LNET_LOCK_EX);
2423 /* update net count */
2424 lnet_current_net_count = lnet_get_net_count();
2430 * shutdown the new NIs that are being started up
2431 * free the NET being started
2433 while (!list_empty(&local_ni_list)) {
2434 ni = list_entry(local_ni_list.next, struct lnet_ni,
2437 lnet_shutdown_lndni(ni);
2447 lnet_startup_lndnets(struct list_head *netlist)
2449 struct lnet_net *net;
2454 * Change to running state before bringing up the LNDs. This
2455 * allows lnet_shutdown_lndnets() to assert that we've passed
2458 lnet_net_lock(LNET_LOCK_EX);
2459 the_lnet.ln_state = LNET_STATE_RUNNING;
2460 lnet_net_unlock(LNET_LOCK_EX);
2462 while (!list_empty(netlist)) {
2463 net = list_entry(netlist->next, struct lnet_net, net_list);
2464 list_del_init(&net->net_list);
2466 rc = lnet_startup_lndnet(net, NULL);
2476 lnet_shutdown_lndnets();
2482 * Initialize LNet library.
2484 * Automatically called at module loading time. Caller has to call
2485 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2486 * latter returned 0. It must be called exactly once.
2488 * \retval 0 on success
2489 * \retval -ve on failures.
2491 int lnet_lib_init(void)
2495 lnet_assert_wire_constants();
2497 /* refer to global cfs_cpt_table for now */
2498 the_lnet.ln_cpt_table = cfs_cpt_table;
2499 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
2501 LASSERT(the_lnet.ln_cpt_number > 0);
2502 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2503 /* we are under risk of consuming all lh_cookie */
2504 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2505 "please change setting of CPT-table and retry\n",
2506 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2510 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2511 the_lnet.ln_cpt_bits++;
2513 rc = lnet_create_locks();
2515 CERROR("Can't create LNet global locks: %d\n", rc);
2519 the_lnet.ln_refcount = 0;
2520 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2521 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2523 /* The hash table size is the number of bits it takes to express the set
2524 * ln_num_routes, minus 1 (better to under estimate than over so we
2525 * don't waste memory). */
2526 if (rnet_htable_size <= 0)
2527 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2528 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2529 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2530 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2531 order_base_2(rnet_htable_size) - 1);
2533 /* All LNDs apart from the LOLND are in separate modules. They
2534 * register themselves when their module loads, and unregister
2535 * themselves when their module is unloaded. */
2536 lnet_register_lnd(&the_lolnd);
2541 * Finalize LNet library.
2543 * \pre lnet_lib_init() called with success.
2544 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2546 * As this happens at module-unload, all lnds must already be unloaded,
2547 * so they must already be unregistered.
2549 void lnet_lib_exit(void)
2553 LASSERT(the_lnet.ln_refcount == 0);
2554 lnet_unregister_lnd(&the_lolnd);
2555 for (i = 0; i < NUM_LNDS; i++)
2556 LASSERT(!the_lnet.ln_lnds[i]);
2557 lnet_destroy_locks();
2561 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2563 * Users must call this function at least once before any other functions.
2564 * For each successful call there must be a corresponding call to
2565 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2568 * The PID used by LNet may be different from the one requested.
2571 * \param requested_pid PID requested by the caller.
2573 * \return >= 0 on success, and < 0 error code on failures.
2576 LNetNIInit(lnet_pid_t requested_pid)
2578 int im_a_router = 0;
2581 struct lnet_ping_buffer *pbuf;
2582 struct lnet_handle_md ping_mdh;
2583 LIST_HEAD(net_head);
2584 struct lnet_net *net;
2586 mutex_lock(&the_lnet.ln_api_mutex);
2588 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2590 if (the_lnet.ln_refcount > 0) {
2591 rc = the_lnet.ln_refcount++;
2592 mutex_unlock(&the_lnet.ln_api_mutex);
2596 rc = lnet_prepare(requested_pid);
2598 mutex_unlock(&the_lnet.ln_api_mutex);
2602 /* create a network for Loopback network */
2603 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2606 goto err_empty_list;
2609 /* Add in the loopback NI */
2610 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2612 goto err_empty_list;
2615 /* If LNet is being initialized via DLC it is possible
2616 * that the user requests not to load module parameters (ones which
2617 * are supported by DLC) on initialization. Therefore, make sure not
2618 * to load networks, routes and forwarding from module parameters
2619 * in this case. On cleanup in case of failure only clean up
2620 * routes if it has been loaded */
2621 if (!the_lnet.ln_nis_from_mod_params) {
2622 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2625 goto err_empty_list;
2628 ni_count = lnet_startup_lndnets(&net_head);
2631 goto err_empty_list;
2634 if (!the_lnet.ln_nis_from_mod_params) {
2635 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2637 goto err_shutdown_lndnis;
2639 rc = lnet_rtrpools_alloc(im_a_router);
2641 goto err_destroy_routes;
2644 rc = lnet_acceptor_start();
2646 goto err_destroy_routes;
2648 the_lnet.ln_refcount = 1;
2649 /* Now I may use my own API functions... */
2651 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2653 goto err_acceptor_stop;
2655 lnet_ping_target_update(pbuf, ping_mdh);
2657 rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2659 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2663 rc = lnet_push_target_init();
2667 rc = lnet_peer_discovery_start();
2669 goto err_destroy_push_target;
2671 rc = lnet_monitor_thr_start();
2673 goto err_stop_discovery_thr;
2676 lnet_router_debugfs_init();
2678 mutex_unlock(&the_lnet.ln_api_mutex);
2680 complete_all(&the_lnet.ln_started);
2682 /* wait for all routers to start */
2683 lnet_wait_router_start();
2687 err_stop_discovery_thr:
2688 lnet_peer_discovery_stop();
2689 err_destroy_push_target:
2690 lnet_push_target_fini();
2692 lnet_ping_target_fini();
2694 the_lnet.ln_refcount = 0;
2695 lnet_acceptor_stop();
2697 if (!the_lnet.ln_nis_from_mod_params)
2698 lnet_destroy_routes();
2699 err_shutdown_lndnis:
2700 lnet_shutdown_lndnets();
2704 mutex_unlock(&the_lnet.ln_api_mutex);
2705 while (!list_empty(&net_head)) {
2706 struct lnet_net *net;
2708 net = list_entry(net_head.next, struct lnet_net, net_list);
2709 list_del_init(&net->net_list);
2714 EXPORT_SYMBOL(LNetNIInit);
2717 * Stop LNet interfaces, routing, and forwarding.
2719 * Users must call this function once for each successful call to LNetNIInit().
2720 * Once the LNetNIFini() operation has been started, the results of pending
2721 * API operations are undefined.
2723 * \return always 0 for current implementation.
2728 mutex_lock(&the_lnet.ln_api_mutex);
2730 LASSERT(the_lnet.ln_refcount > 0);
2732 if (the_lnet.ln_refcount != 1) {
2733 the_lnet.ln_refcount--;
2735 LASSERT(!the_lnet.ln_niinit_self);
2739 lnet_router_debugfs_fini();
2740 lnet_monitor_thr_stop();
2741 lnet_peer_discovery_stop();
2742 lnet_push_target_fini();
2743 lnet_ping_target_fini();
2745 /* Teardown fns that use my own API functions BEFORE here */
2746 the_lnet.ln_refcount = 0;
2748 lnet_acceptor_stop();
2749 lnet_destroy_routes();
2750 lnet_shutdown_lndnets();
2754 mutex_unlock(&the_lnet.ln_api_mutex);
2757 EXPORT_SYMBOL(LNetNIFini);
2760 * Grabs the ni data from the ni structure and fills the out
2763 * \param[in] ni network interface structure
2764 * \param[out] cfg_ni NI config information
2765 * \param[out] tun network and LND tunables
2768 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2769 struct lnet_ioctl_config_lnd_tunables *tun,
2770 struct lnet_ioctl_element_stats *stats,
2773 size_t min_size = 0;
2776 if (!ni || !cfg_ni || !tun)
2779 if (ni->ni_interfaces[0] != NULL) {
2780 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2781 if (ni->ni_interfaces[i] != NULL) {
2782 strncpy(cfg_ni->lic_ni_intf[i],
2783 ni->ni_interfaces[i],
2784 sizeof(cfg_ni->lic_ni_intf[i]));
2789 cfg_ni->lic_nid = ni->ni_nid;
2790 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2791 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2793 cfg_ni->lic_status = ni->ni_status->ns_status;
2794 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2795 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2797 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2800 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2801 LNET_STATS_TYPE_SEND);
2802 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2803 LNET_STATS_TYPE_RECV);
2804 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2805 LNET_STATS_TYPE_DROP);
2809 * tun->lt_tun will always be present, but in order to be
2810 * backwards compatible, we need to deal with the cases when
2811 * tun->lt_tun is smaller than what the kernel has, because it
2812 * comes from an older version of a userspace program, then we'll
2813 * need to copy as much information as we have available space.
2815 min_size = tun_size - sizeof(tun->lt_cmn);
2816 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2818 /* copy over the cpts */
2819 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2820 ni->ni_cpts == NULL) {
2821 for (i = 0; i < ni->ni_ncpts; i++)
2822 cfg_ni->lic_cpts[i] = i;
2825 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2826 i < LNET_MAX_SHOW_NUM_CPT;
2828 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2830 cfg_ni->lic_ncpts = ni->ni_ncpts;
2834 * NOTE: This is a legacy function left in the code to be backwards
2835 * compatible with older userspace programs. It should eventually be
2838 * Grabs the ni data from the ni structure and fills the out
2841 * \param[in] ni network interface structure
2842 * \param[out] config config information
2845 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2846 struct lnet_ioctl_config_data *config)
2848 struct lnet_ioctl_net_config *net_config;
2849 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2850 size_t min_size, tunable_size = 0;
2856 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2860 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2861 ARRAY_SIZE(net_config->ni_interfaces));
2863 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2864 if (!ni->ni_interfaces[i])
2867 strncpy(net_config->ni_interfaces[i],
2868 ni->ni_interfaces[i],
2869 sizeof(net_config->ni_interfaces[i]));
2872 config->cfg_nid = ni->ni_nid;
2873 config->cfg_config_u.cfg_net.net_peer_timeout =
2874 ni->ni_net->net_tunables.lct_peer_timeout;
2875 config->cfg_config_u.cfg_net.net_max_tx_credits =
2876 ni->ni_net->net_tunables.lct_max_tx_credits;
2877 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2878 ni->ni_net->net_tunables.lct_peer_tx_credits;
2879 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2880 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2882 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2883 net_config->ni_status = LNET_NI_STATUS_UP;
2885 net_config->ni_status = ni->ni_status->ns_status;
2888 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2890 for (i = 0; i < num_cpts; i++)
2891 net_config->ni_cpts[i] = ni->ni_cpts[i];
2893 config->cfg_ncpts = num_cpts;
2897 * See if user land tools sent in a newer and larger version
2898 * of struct lnet_tunables than what the kernel uses.
2900 min_size = sizeof(*config) + sizeof(*net_config);
2902 if (config->cfg_hdr.ioc_len > min_size)
2903 tunable_size = config->cfg_hdr.ioc_len - min_size;
2905 /* Don't copy too much data to user space */
2906 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2907 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2909 if (lnd_cfg && min_size) {
2910 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2911 config->cfg_config_u.cfg_net.net_interface_count = 1;
2913 /* Tell user land that kernel side has less data */
2914 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2915 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2916 config->cfg_hdr.ioc_len -= min_size;
2922 lnet_get_ni_idx_locked(int idx)
2925 struct lnet_net *net;
2927 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2928 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2938 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2941 struct lnet_net *net = mynet;
2944 * It is possible that the net has been cleaned out while there is
2945 * a message being sent. This function accessed the net without
2946 * checking if the list is empty
2950 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2952 if (list_empty(&net->net_ni_list))
2954 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2960 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2961 /* if you reached the end of the ni list and the net is
2962 * specified, then there are no more nis in that net */
2966 /* we reached the end of this net ni list. move to the
2968 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2969 /* no more nets and no more NIs. */
2972 /* get the next net */
2973 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2975 if (list_empty(&net->net_ni_list))
2977 /* get the ni on it */
2978 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2984 if (list_empty(&prev->ni_netlist))
2987 /* there are more nis left */
2988 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2994 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2999 int idx = config->cfg_count;
3001 cpt = lnet_net_lock_current();
3003 ni = lnet_get_ni_idx_locked(idx);
3008 lnet_fill_ni_info_legacy(ni, config);
3012 lnet_net_unlock(cpt);
3017 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3018 struct lnet_ioctl_config_lnd_tunables *tun,
3019 struct lnet_ioctl_element_stats *stats,
3026 if (!cfg_ni || !tun || !stats)
3029 cpt = lnet_net_lock_current();
3031 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3036 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3040 lnet_net_unlock(cpt);
3044 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3053 cpt = lnet_net_lock_current();
3055 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3058 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3062 lnet_net_unlock(cpt);
3067 static int lnet_add_net_common(struct lnet_net *net,
3068 struct lnet_ioctl_config_lnd_tunables *tun)
3071 struct lnet_ping_buffer *pbuf;
3072 struct lnet_handle_md ping_mdh;
3074 struct lnet_remotenet *rnet;
3076 int num_acceptor_nets;
3078 lnet_net_lock(LNET_LOCK_EX);
3079 rnet = lnet_find_rnet_locked(net->net_id);
3080 lnet_net_unlock(LNET_LOCK_EX);
3082 * make sure that the net added doesn't invalidate the current
3083 * configuration LNet is keeping
3086 CERROR("Adding net %s will invalidate routing configuration\n",
3087 libcfs_net2str(net->net_id));
3093 * make sure you calculate the correct number of slots in the ping
3094 * buffer. Since the ping info is a flattened list of all the NIs,
3095 * we should allocate enough slots to accomodate the number of NIs
3096 * which will be added.
3098 * since ni hasn't been configured yet, use
3099 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3101 net_ni_count = lnet_get_net_ni_count_pre(net);
3103 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3104 net_ni_count + lnet_get_ni_count(),
3112 memcpy(&net->net_tunables,
3113 &tun->lt_cmn, sizeof(net->net_tunables));
3115 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3118 * before starting this network get a count of the current TCP
3119 * networks which require the acceptor thread running. If that
3120 * count is == 0 before we start up this network, then we'd want to
3121 * start up the acceptor thread after starting up this network
3123 num_acceptor_nets = lnet_count_acceptor_nets();
3125 net_id = net->net_id;
3127 rc = lnet_startup_lndnet(net,
3128 (tun) ? &tun->lt_tun : NULL);
3132 lnet_net_lock(LNET_LOCK_EX);
3133 net = lnet_get_net_locked(net_id);
3134 lnet_net_unlock(LNET_LOCK_EX);
3139 * Start the acceptor thread if this is the first network
3140 * being added that requires the thread.
3142 if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3143 rc = lnet_acceptor_start();
3145 /* shutdown the net that we just started */
3146 CERROR("Failed to start up acceptor thread\n");
3147 lnet_shutdown_lndnet(net);
3152 lnet_net_lock(LNET_LOCK_EX);
3153 lnet_peer_net_added(net);
3154 lnet_net_unlock(LNET_LOCK_EX);
3156 lnet_ping_target_update(pbuf, ping_mdh);
3161 lnet_ping_md_unlink(pbuf, &ping_mdh);
3162 lnet_ping_buffer_decref(pbuf);
3166 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3167 struct lnet_ioctl_config_lnd_tunables *tun)
3169 struct lnet_net *net;
3172 LIST_HEAD(net_head);
3174 rc = lnet_parse_ip2nets(&nets, ip2nets);
3178 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3182 mutex_lock(&the_lnet.ln_api_mutex);
3183 while (!list_empty(&net_head)) {
3184 net = list_entry(net_head.next, struct lnet_net, net_list);
3185 list_del_init(&net->net_list);
3186 rc = lnet_add_net_common(net, tun);
3192 mutex_unlock(&the_lnet.ln_api_mutex);
3194 while (!list_empty(&net_head)) {
3195 net = list_entry(net_head.next, struct lnet_net, net_list);
3196 list_del_init(&net->net_list);
3202 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3204 struct lnet_net *net;
3206 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3208 __u32 net_id, lnd_type;
3210 /* get the tunables if they are available */
3211 if (conf->lic_cfg_hdr.ioc_len >=
3212 sizeof(*conf) + sizeof(*tun))
3213 tun = (struct lnet_ioctl_config_lnd_tunables *)
3216 /* handle legacy ip2nets from DLC */
3217 if (conf->lic_legacy_ip2nets[0] != '\0')
3218 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3221 net_id = LNET_NIDNET(conf->lic_nid);
3222 lnd_type = LNET_NETTYP(net_id);
3224 if (!libcfs_isknown_lnd(lnd_type)) {
3225 CERROR("No valid net and lnd information provided\n");
3229 net = lnet_net_alloc(net_id, NULL);
3233 for (i = 0; i < conf->lic_ncpts; i++) {
3234 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3238 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3239 conf->lic_ni_intf[0]);
3243 mutex_lock(&the_lnet.ln_api_mutex);
3245 rc = lnet_add_net_common(net, tun);
3247 mutex_unlock(&the_lnet.ln_api_mutex);
3252 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3254 struct lnet_net *net;
3256 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3257 struct lnet_ping_buffer *pbuf;
3258 struct lnet_handle_md ping_mdh;
3263 /* don't allow userspace to shutdown the LOLND */
3264 if (LNET_NETTYP(net_id) == LOLND)
3267 mutex_lock(&the_lnet.ln_api_mutex);
3271 net = lnet_get_net_locked(net_id);
3273 CERROR("net %s not found\n",
3274 libcfs_net2str(net_id));
3279 addr = LNET_NIDADDR(conf->lic_nid);
3281 /* remove the entire net */
3282 net_count = lnet_get_net_ni_count_locked(net);
3286 /* create and link a new ping info, before removing the old one */
3287 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3288 lnet_get_ni_count() - net_count,
3291 goto unlock_api_mutex;
3293 lnet_shutdown_lndnet(net);
3295 if (lnet_count_acceptor_nets() == 0)
3296 lnet_acceptor_stop();
3298 lnet_ping_target_update(pbuf, ping_mdh);
3300 goto unlock_api_mutex;
3303 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3305 CERROR("nid %s not found\n",
3306 libcfs_nid2str(conf->lic_nid));
3311 net_count = lnet_get_net_ni_count_locked(net);
3315 /* create and link a new ping info, before removing the old one */
3316 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3317 lnet_get_ni_count() - 1, false);
3319 goto unlock_api_mutex;
3321 lnet_shutdown_lndni(ni);
3323 if (lnet_count_acceptor_nets() == 0)
3324 lnet_acceptor_stop();
3326 lnet_ping_target_update(pbuf, ping_mdh);
3328 /* check if the net is empty and remove it if it is */
3330 lnet_shutdown_lndnet(net);
3332 goto unlock_api_mutex;
3337 mutex_unlock(&the_lnet.ln_api_mutex);
3343 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3344 * They are only expected to be called for unique networks.
3345 * That can be as a result of older DLC library
3346 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3349 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3351 struct lnet_net *net;
3352 LIST_HEAD(net_head);
3354 struct lnet_ioctl_config_lnd_tunables tun;
3355 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3357 /* Create a net/ni structures for the network string */
3358 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3360 return rc == 0 ? -EINVAL : rc;
3362 mutex_lock(&the_lnet.ln_api_mutex);
3365 rc = -EINVAL; /* only add one network per call */
3366 goto out_unlock_clean;
3369 net = list_entry(net_head.next, struct lnet_net, net_list);
3370 list_del_init(&net->net_list);
3372 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3374 memset(&tun, 0, sizeof(tun));
3376 tun.lt_cmn.lct_peer_timeout =
3377 conf->cfg_config_u.cfg_net.net_peer_timeout;
3378 tun.lt_cmn.lct_peer_tx_credits =
3379 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3380 tun.lt_cmn.lct_peer_rtr_credits =
3381 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3382 tun.lt_cmn.lct_max_tx_credits =
3383 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3385 rc = lnet_add_net_common(net, &tun);
3388 mutex_unlock(&the_lnet.ln_api_mutex);
3389 while (!list_empty(&net_head)) {
3390 /* net_head list is empty in success case */
3391 net = list_entry(net_head.next, struct lnet_net, net_list);
3392 list_del_init(&net->net_list);
3399 lnet_dyn_del_net(__u32 net_id)
3401 struct lnet_net *net;
3402 struct lnet_ping_buffer *pbuf;
3403 struct lnet_handle_md ping_mdh;
3407 /* don't allow userspace to shutdown the LOLND */
3408 if (LNET_NETTYP(net_id) == LOLND)
3411 mutex_lock(&the_lnet.ln_api_mutex);
3415 net = lnet_get_net_locked(net_id);
3422 net_ni_count = lnet_get_net_ni_count_locked(net);
3426 /* create and link a new ping info, before removing the old one */
3427 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3428 lnet_get_ni_count() - net_ni_count, false);
3432 lnet_shutdown_lndnet(net);
3434 if (lnet_count_acceptor_nets() == 0)
3435 lnet_acceptor_stop();
3437 lnet_ping_target_update(pbuf, ping_mdh);
3440 mutex_unlock(&the_lnet.ln_api_mutex);
3445 void lnet_incr_dlc_seq(void)
3447 atomic_inc(&lnet_dlc_seq_no);
3450 __u32 lnet_get_dlc_seq_locked(void)
3452 return atomic_read(&lnet_dlc_seq_no);
3456 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3458 struct lnet_net *net;
3461 lnet_net_lock(LNET_LOCK_EX);
3462 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3463 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3464 if (ni->ni_nid == nid || all) {
3465 atomic_set(&ni->ni_healthv, value);
3466 if (list_empty(&ni->ni_recovery) &&
3467 value < LNET_MAX_HEALTH_VALUE) {
3468 CERROR("manually adding local NI %s to recovery\n",
3469 libcfs_nid2str(ni->ni_nid));
3470 list_add_tail(&ni->ni_recovery,
3471 &the_lnet.ln_mt_localNIRecovq);
3472 lnet_ni_addref_locked(ni, 0);
3475 lnet_net_unlock(LNET_LOCK_EX);
3481 lnet_net_unlock(LNET_LOCK_EX);
3485 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3489 lnet_nid_t nid = stats->hlni_nid;
3491 cpt = lnet_net_lock_current();
3492 ni = lnet_nid2ni_locked(nid, cpt);
3499 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3500 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3501 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3502 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3503 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3504 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3505 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3508 lnet_net_unlock(cpt);
3514 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3519 lnet_net_lock(LNET_LOCK_EX);
3520 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3521 list->rlst_nid_array[i] = ni->ni_nid;
3523 if (i >= LNET_MAX_SHOW_NUM_NID)
3526 lnet_net_unlock(LNET_LOCK_EX);
3527 list->rlst_num_nids = i;
3533 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3535 struct lnet_peer_ni *lpni;
3538 lnet_net_lock(LNET_LOCK_EX);
3539 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3540 list->rlst_nid_array[i] = lpni->lpni_nid;
3542 if (i >= LNET_MAX_SHOW_NUM_NID)
3545 lnet_net_unlock(LNET_LOCK_EX);
3546 list->rlst_num_nids = i;
3552 * LNet ioctl handler.
3556 LNetCtl(unsigned int cmd, void *arg)
3558 struct libcfs_ioctl_data *data = arg;
3559 struct lnet_ioctl_config_data *config;
3560 struct lnet_process_id id = {0};
3564 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3565 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3568 case IOC_LIBCFS_GET_NI:
3569 rc = LNetGetId(data->ioc_count, &id);
3570 data->ioc_nid = id.nid;
3573 case IOC_LIBCFS_FAIL_NID:
3574 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3576 case IOC_LIBCFS_ADD_ROUTE: {
3577 /* default router sensitivity to 1 */
3578 unsigned int sensitivity = 1;
3581 if (config->cfg_hdr.ioc_len < sizeof(*config))
3584 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3586 config->cfg_config_u.cfg_route.rtr_sensitivity;
3589 mutex_lock(&the_lnet.ln_api_mutex);
3590 rc = lnet_add_route(config->cfg_net,
3591 config->cfg_config_u.cfg_route.rtr_hop,
3593 config->cfg_config_u.cfg_route.
3594 rtr_priority, sensitivity);
3595 mutex_unlock(&the_lnet.ln_api_mutex);
3599 case IOC_LIBCFS_DEL_ROUTE:
3602 if (config->cfg_hdr.ioc_len < sizeof(*config))
3605 mutex_lock(&the_lnet.ln_api_mutex);
3606 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3607 mutex_unlock(&the_lnet.ln_api_mutex);
3610 case IOC_LIBCFS_GET_ROUTE:
3613 if (config->cfg_hdr.ioc_len < sizeof(*config))
3616 mutex_lock(&the_lnet.ln_api_mutex);
3617 rc = lnet_get_route(config->cfg_count,
3619 &config->cfg_config_u.cfg_route.rtr_hop,
3621 &config->cfg_config_u.cfg_route.rtr_flags,
3622 &config->cfg_config_u.cfg_route.
3624 &config->cfg_config_u.cfg_route.
3626 mutex_unlock(&the_lnet.ln_api_mutex);
3629 case IOC_LIBCFS_GET_LOCAL_NI: {
3630 struct lnet_ioctl_config_ni *cfg_ni;
3631 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3632 struct lnet_ioctl_element_stats *stats;
3637 /* get the tunables if they are available */
3638 if (cfg_ni->lic_cfg_hdr.ioc_len <
3639 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3642 stats = (struct lnet_ioctl_element_stats *)
3644 tun = (struct lnet_ioctl_config_lnd_tunables *)
3645 (cfg_ni->lic_bulk + sizeof(*stats));
3647 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3650 mutex_lock(&the_lnet.ln_api_mutex);
3651 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3652 mutex_unlock(&the_lnet.ln_api_mutex);
3656 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3657 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3659 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3662 mutex_lock(&the_lnet.ln_api_mutex);
3663 rc = lnet_get_ni_stats(msg_stats);
3664 mutex_unlock(&the_lnet.ln_api_mutex);
3669 case IOC_LIBCFS_GET_NET: {
3670 size_t total = sizeof(*config) +
3671 sizeof(struct lnet_ioctl_net_config);
3674 if (config->cfg_hdr.ioc_len < total)
3677 mutex_lock(&the_lnet.ln_api_mutex);
3678 rc = lnet_get_net_config(config);
3679 mutex_unlock(&the_lnet.ln_api_mutex);
3683 case IOC_LIBCFS_GET_LNET_STATS:
3685 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3687 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3690 mutex_lock(&the_lnet.ln_api_mutex);
3691 lnet_counters_get(&lnet_stats->st_cntrs);
3692 mutex_unlock(&the_lnet.ln_api_mutex);
3696 case IOC_LIBCFS_CONFIG_RTR:
3699 if (config->cfg_hdr.ioc_len < sizeof(*config))
3702 mutex_lock(&the_lnet.ln_api_mutex);
3703 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3704 rc = lnet_rtrpools_enable();
3705 mutex_unlock(&the_lnet.ln_api_mutex);
3708 lnet_rtrpools_disable();
3709 mutex_unlock(&the_lnet.ln_api_mutex);
3712 case IOC_LIBCFS_ADD_BUF:
3715 if (config->cfg_hdr.ioc_len < sizeof(*config))
3718 mutex_lock(&the_lnet.ln_api_mutex);
3719 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3721 config->cfg_config_u.cfg_buffers.
3723 config->cfg_config_u.cfg_buffers.
3725 mutex_unlock(&the_lnet.ln_api_mutex);
3728 case IOC_LIBCFS_SET_NUMA_RANGE: {
3729 struct lnet_ioctl_set_value *numa;
3731 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3733 lnet_net_lock(LNET_LOCK_EX);
3734 lnet_numa_range = numa->sv_value;
3735 lnet_net_unlock(LNET_LOCK_EX);
3739 case IOC_LIBCFS_GET_NUMA_RANGE: {
3740 struct lnet_ioctl_set_value *numa;
3742 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3744 numa->sv_value = lnet_numa_range;
3748 case IOC_LIBCFS_GET_BUF: {
3749 struct lnet_ioctl_pool_cfg *pool_cfg;
3750 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3754 if (config->cfg_hdr.ioc_len < total)
3757 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3759 mutex_lock(&the_lnet.ln_api_mutex);
3760 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3761 mutex_unlock(&the_lnet.ln_api_mutex);
3765 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3766 struct lnet_ioctl_local_ni_hstats *stats = arg;
3768 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3771 mutex_lock(&the_lnet.ln_api_mutex);
3772 rc = lnet_get_local_ni_hstats(stats);
3773 mutex_unlock(&the_lnet.ln_api_mutex);
3778 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3779 struct lnet_ioctl_recovery_list *list = arg;
3780 if (list->rlst_hdr.ioc_len < sizeof(*list))
3783 mutex_lock(&the_lnet.ln_api_mutex);
3784 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3785 rc = lnet_get_local_ni_recovery_list(list);
3787 rc = lnet_get_peer_ni_recovery_list(list);
3788 mutex_unlock(&the_lnet.ln_api_mutex);
3792 case IOC_LIBCFS_ADD_PEER_NI: {
3793 struct lnet_ioctl_peer_cfg *cfg = arg;
3795 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3798 mutex_lock(&the_lnet.ln_api_mutex);
3799 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3802 mutex_unlock(&the_lnet.ln_api_mutex);
3806 case IOC_LIBCFS_DEL_PEER_NI: {
3807 struct lnet_ioctl_peer_cfg *cfg = arg;
3809 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3812 mutex_lock(&the_lnet.ln_api_mutex);
3813 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3814 cfg->prcfg_cfg_nid);
3815 mutex_unlock(&the_lnet.ln_api_mutex);
3819 case IOC_LIBCFS_GET_PEER_INFO: {
3820 struct lnet_ioctl_peer *peer_info = arg;
3822 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3825 mutex_lock(&the_lnet.ln_api_mutex);
3826 rc = lnet_get_peer_ni_info(
3827 peer_info->pr_count,
3829 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3830 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3831 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3832 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3833 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3834 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3835 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3836 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3837 mutex_unlock(&the_lnet.ln_api_mutex);
3841 case IOC_LIBCFS_GET_PEER_NI: {
3842 struct lnet_ioctl_peer_cfg *cfg = arg;
3844 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3847 mutex_lock(&the_lnet.ln_api_mutex);
3848 rc = lnet_get_peer_info(cfg,
3849 (void __user *)cfg->prcfg_bulk);
3850 mutex_unlock(&the_lnet.ln_api_mutex);
3854 case IOC_LIBCFS_GET_PEER_LIST: {
3855 struct lnet_ioctl_peer_cfg *cfg = arg;
3857 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3860 mutex_lock(&the_lnet.ln_api_mutex);
3861 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3862 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3863 mutex_unlock(&the_lnet.ln_api_mutex);
3867 case IOC_LIBCFS_SET_HEALHV: {
3868 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3870 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3872 if (cfg->rh_value < 0 ||
3873 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3874 value = LNET_MAX_HEALTH_VALUE;
3876 value = cfg->rh_value;
3877 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3878 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3879 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3880 mutex_lock(&the_lnet.ln_api_mutex);
3881 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3882 lnet_ni_set_healthv(cfg->rh_nid, value,
3885 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3887 mutex_unlock(&the_lnet.ln_api_mutex);
3891 case IOC_LIBCFS_NOTIFY_ROUTER: {
3892 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3894 /* The deadline passed in by the user should be some time in
3895 * seconds in the future since the UNIX epoch. We have to map
3896 * that deadline to the wall clock.
3898 deadline += ktime_get_seconds();
3899 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3903 case IOC_LIBCFS_LNET_DIST:
3904 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3905 if (rc < 0 && rc != -EHOSTUNREACH)
3908 data->ioc_u32[0] = rc;
3911 case IOC_LIBCFS_TESTPROTOCOMPAT:
3912 the_lnet.ln_testprotocompat = data->ioc_flags;
3915 case IOC_LIBCFS_LNET_FAULT:
3916 return lnet_fault_ctl(data->ioc_flags, data);
3918 case IOC_LIBCFS_PING: {
3919 signed long timeout;
3921 id.nid = data->ioc_nid;
3922 id.pid = data->ioc_u32[0];
3924 /* If timeout is negative then set default of 3 minutes */
3925 if (((s32)data->ioc_u32[1] <= 0) ||
3926 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3927 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3929 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3931 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3932 data->ioc_plen1 / sizeof(struct lnet_process_id));
3937 data->ioc_count = rc;
3941 case IOC_LIBCFS_PING_PEER: {
3942 struct lnet_ioctl_ping_data *ping = arg;
3943 struct lnet_peer *lp;
3944 signed long timeout;
3946 /* If timeout is negative then set default of 3 minutes */
3947 if (((s32)ping->op_param) <= 0 ||
3948 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3949 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3951 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3953 rc = lnet_ping(ping->ping_id, timeout,
3959 mutex_lock(&the_lnet.ln_api_mutex);
3960 lp = lnet_find_peer(ping->ping_id.nid);
3962 ping->ping_id.nid = lp->lp_primary_nid;
3963 ping->mr_info = lnet_peer_is_multi_rail(lp);
3964 lnet_peer_decref_locked(lp);
3966 mutex_unlock(&the_lnet.ln_api_mutex);
3968 ping->ping_count = rc;
3972 case IOC_LIBCFS_DISCOVER: {
3973 struct lnet_ioctl_ping_data *discover = arg;
3974 struct lnet_peer *lp;
3976 rc = lnet_discover(discover->ping_id, discover->op_param,
3978 discover->ping_count);
3982 mutex_lock(&the_lnet.ln_api_mutex);
3983 lp = lnet_find_peer(discover->ping_id.nid);
3985 discover->ping_id.nid = lp->lp_primary_nid;
3986 discover->mr_info = lnet_peer_is_multi_rail(lp);
3987 lnet_peer_decref_locked(lp);
3989 mutex_unlock(&the_lnet.ln_api_mutex);
3991 discover->ping_count = rc;
3996 ni = lnet_net2ni_addref(data->ioc_net);
4000 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4003 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4010 EXPORT_SYMBOL(LNetCtl);
4012 void LNetDebugPeer(struct lnet_process_id id)
4014 lnet_debug_peer(id.nid);
4016 EXPORT_SYMBOL(LNetDebugPeer);
4019 * Determine if the specified peer \a nid is on the local node.
4021 * \param nid peer nid to check
4023 * \retval true If peer NID is on the local node.
4024 * \retval false If peer NID is not on the local node.
4026 bool LNetIsPeerLocal(lnet_nid_t nid)
4028 struct lnet_net *net;
4032 cpt = lnet_net_lock_current();
4033 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4034 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4035 if (ni->ni_nid == nid) {
4036 lnet_net_unlock(cpt);
4041 lnet_net_unlock(cpt);
4045 EXPORT_SYMBOL(LNetIsPeerLocal);
4048 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4049 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4051 * \param index Index of the interface to look up.
4052 * \param id On successful return, this location will hold the
4053 * struct lnet_process_id ID of the interface.
4055 * \retval 0 If an interface exists at \a index.
4056 * \retval -ENOENT If no interface has been found.
4059 LNetGetId(unsigned int index, struct lnet_process_id *id)
4062 struct lnet_net *net;
4066 LASSERT(the_lnet.ln_refcount > 0);
4068 cpt = lnet_net_lock_current();
4070 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4071 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4075 id->nid = ni->ni_nid;
4076 id->pid = the_lnet.ln_pid;
4082 lnet_net_unlock(cpt);
4085 EXPORT_SYMBOL(LNetGetId);
4087 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4088 struct lnet_process_id __user *ids, int n_ids)
4090 struct lnet_handle_eq eqh;
4091 struct lnet_handle_md mdh;
4092 struct lnet_event event;
4093 struct lnet_md md = { NULL };
4097 const signed long a_long_time = cfs_time_seconds(60);
4098 struct lnet_ping_buffer *pbuf;
4099 struct lnet_process_id tmpid;
4106 /* n_ids limit is arbitrary */
4107 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4111 * if the user buffer has more space than the lnet_interfaces_max
4112 * then only fill it up to lnet_interfaces_max
4114 if (n_ids > lnet_interfaces_max)
4115 n_ids = lnet_interfaces_max;
4117 if (id.pid == LNET_PID_ANY)
4118 id.pid = LNET_PID_LUSTRE;
4120 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4124 /* NB 2 events max (including any unlink event) */
4125 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
4127 CERROR("Can't allocate EQ: %d\n", rc);
4128 goto fail_ping_buffer_decref;
4131 /* initialize md content */
4132 md.start = &pbuf->pb_info;
4133 md.length = LNET_PING_INFO_SIZE(n_ids);
4134 md.threshold = 2; /* GET/REPLY */
4136 md.options = LNET_MD_TRUNCATE;
4140 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4142 CERROR("Can't bind MD: %d\n", rc);
4146 rc = LNetGet(LNET_NID_ANY, mdh, id,
4147 LNET_RESERVED_PORTAL,
4148 LNET_PROTO_PING_MATCHBITS, 0, false);
4151 /* Don't CERROR; this could be deliberate! */
4152 rc2 = LNetMDUnlink(mdh);
4155 /* NB must wait for the UNLINK event below... */
4157 timeout = a_long_time;
4161 /* MUST block for unlink to complete */
4163 blocked = cfs_block_allsigs();
4165 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
4168 cfs_restore_sigs(blocked);
4170 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4171 (rc2 <= 0) ? -1 : event.type,
4172 (rc2 <= 0) ? -1 : event.status,
4173 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4175 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
4177 if (rc2 <= 0 || event.status != 0) {
4178 /* timeout or error */
4179 if (!replied && rc == 0)
4180 rc = (rc2 < 0) ? rc2 :
4181 (rc2 == 0) ? -ETIMEDOUT :
4185 /* Ensure completion in finite time... */
4187 /* No assertion (racing with network) */
4189 timeout = a_long_time;
4190 } else if (rc2 == 0) {
4191 /* timed out waiting for unlink */
4192 CWARN("ping %s: late network completion\n",
4195 } else if (event.type == LNET_EVENT_REPLY) {
4199 } while (rc2 <= 0 || !event.unlinked);
4203 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4210 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4212 rc = -EPROTO; /* if I can't parse... */
4215 CERROR("%s: ping info too short %d\n",
4216 libcfs_id2str(id), nob);
4220 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4221 lnet_swap_pinginfo(pbuf);
4222 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4223 CERROR("%s: Unexpected magic %08x\n",
4224 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4228 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4229 CERROR("%s: ping w/o NI status: 0x%x\n",
4230 libcfs_id2str(id), pbuf->pb_info.pi_features);
4234 if (nob < LNET_PING_INFO_SIZE(0)) {
4235 CERROR("%s: Short reply %d(%d min)\n",
4237 nob, (int)LNET_PING_INFO_SIZE(0));
4241 if (pbuf->pb_info.pi_nnis < n_ids)
4242 n_ids = pbuf->pb_info.pi_nnis;
4244 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4245 CERROR("%s: Short reply %d(%d expected)\n",
4247 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4251 rc = -EFAULT; /* if I segv in copy_to_user()... */
4253 memset(&tmpid, 0, sizeof(tmpid));
4254 for (i = 0; i < n_ids; i++) {
4255 tmpid.pid = pbuf->pb_info.pi_pid;
4256 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4257 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4260 rc = pbuf->pb_info.pi_nnis;
4263 rc2 = LNetEQFree(eqh);
4265 CERROR("rc2 %d\n", rc2);
4268 fail_ping_buffer_decref:
4269 lnet_ping_buffer_decref(pbuf);
4274 lnet_discover(struct lnet_process_id id, __u32 force,
4275 struct lnet_process_id __user *ids, int n_ids)
4277 struct lnet_peer_ni *lpni;
4278 struct lnet_peer_ni *p;
4279 struct lnet_peer *lp;
4280 struct lnet_process_id *buf;
4284 int max_intf = lnet_interfaces_max;
4288 id.nid == LNET_NID_ANY)
4291 if (id.pid == LNET_PID_ANY)
4292 id.pid = LNET_PID_LUSTRE;
4295 * if the user buffer has more space than the max_intf
4296 * then only fill it up to max_intf
4298 if (n_ids > max_intf)
4301 buf_size = n_ids * sizeof(*buf);
4303 LIBCFS_ALLOC(buf, buf_size);
4307 cpt = lnet_net_lock_current();
4308 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4315 * Clearing the NIDS_UPTODATE flag ensures the peer will
4316 * be discovered, provided discovery has not been disabled.
4318 lp = lpni->lpni_peer_net->lpn_peer;
4319 spin_lock(&lp->lp_lock);
4320 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4321 /* If the force flag is set, force a PING and PUSH as well. */
4323 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4324 spin_unlock(&lp->lp_lock);
4325 rc = lnet_discover_peer_locked(lpni, cpt, true);
4329 /* Peer may have changed. */
4330 lp = lpni->lpni_peer_net->lpn_peer;
4331 if (lp->lp_nnis < n_ids)
4332 n_ids = lp->lp_nnis;
4336 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4337 buf[i].pid = id.pid;
4338 buf[i].nid = p->lpni_nid;
4343 lnet_net_unlock(cpt);
4346 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4352 lnet_peer_ni_decref_locked(lpni);
4354 lnet_net_unlock(cpt);
4356 LIBCFS_FREE(buf, buf_size);
4362 * Retrieve peer discovery status.
4364 * \retval 1 if lnet_peer_discovery_disabled is 0
4365 * \retval 0 if lnet_peer_discovery_disabled is 1
4368 LNetGetPeerDiscoveryStatus(void)
4370 return !lnet_peer_discovery_disabled;
4372 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);