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 10
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 3
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);
554 lnet_fini_locks(void)
558 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
559 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
563 lnet_descriptor_setup(void)
565 /* create specific kmem_cache for MEs and small MDs (i.e., originally
566 * allocated in <size-xxx> kmem_cache).
568 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
570 if (!lnet_mes_cachep)
573 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
574 LNET_SMALL_MD_SIZE, 0, 0,
576 if (!lnet_small_mds_cachep)
583 lnet_descriptor_cleanup(void)
586 if (lnet_small_mds_cachep) {
587 kmem_cache_destroy(lnet_small_mds_cachep);
588 lnet_small_mds_cachep = NULL;
591 if (lnet_mes_cachep) {
592 kmem_cache_destroy(lnet_mes_cachep);
593 lnet_mes_cachep = NULL;
598 lnet_create_remote_nets_table(void)
601 struct list_head *hash;
603 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
604 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
605 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
607 CERROR("Failed to create remote nets hash table\n");
611 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
612 INIT_LIST_HEAD(&hash[i]);
613 the_lnet.ln_remote_nets_hash = hash;
618 lnet_destroy_remote_nets_table(void)
622 if (the_lnet.ln_remote_nets_hash == NULL)
625 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
626 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
628 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
629 LNET_REMOTE_NETS_HASH_SIZE *
630 sizeof(the_lnet.ln_remote_nets_hash[0]));
631 the_lnet.ln_remote_nets_hash = NULL;
635 lnet_destroy_locks(void)
637 if (the_lnet.ln_res_lock != NULL) {
638 cfs_percpt_lock_free(the_lnet.ln_res_lock);
639 the_lnet.ln_res_lock = NULL;
642 if (the_lnet.ln_net_lock != NULL) {
643 cfs_percpt_lock_free(the_lnet.ln_net_lock);
644 the_lnet.ln_net_lock = NULL;
651 lnet_create_locks(void)
655 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
656 if (the_lnet.ln_res_lock == NULL)
659 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
660 if (the_lnet.ln_net_lock == NULL)
666 lnet_destroy_locks();
670 static void lnet_assert_wire_constants(void)
672 /* Wire protocol assertions generated by 'wirecheck'
673 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
674 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
675 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
678 CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
679 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
680 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
681 CLASSERT(LNET_MSG_ACK == 0);
682 CLASSERT(LNET_MSG_PUT == 1);
683 CLASSERT(LNET_MSG_GET == 2);
684 CLASSERT(LNET_MSG_REPLY == 3);
685 CLASSERT(LNET_MSG_HELLO == 4);
687 /* Checks for struct lnet_handle_wire */
688 CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
689 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
690 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
691 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
692 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
694 /* Checks for struct struct lnet_magicversion */
695 CLASSERT((int)sizeof(struct lnet_magicversion) == 8);
696 CLASSERT((int)offsetof(struct lnet_magicversion, magic) == 0);
697 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->magic) == 4);
698 CLASSERT((int)offsetof(struct lnet_magicversion, version_major) == 4);
699 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_major) == 2);
700 CLASSERT((int)offsetof(struct lnet_magicversion, version_minor) == 6);
701 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_minor) == 2);
703 /* Checks for struct struct lnet_hdr */
704 CLASSERT((int)sizeof(struct lnet_hdr) == 72);
705 CLASSERT((int)offsetof(struct lnet_hdr, dest_nid) == 0);
706 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_nid) == 8);
707 CLASSERT((int)offsetof(struct lnet_hdr, src_nid) == 8);
708 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_nid) == 8);
709 CLASSERT((int)offsetof(struct lnet_hdr, dest_pid) == 16);
710 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_pid) == 4);
711 CLASSERT((int)offsetof(struct lnet_hdr, src_pid) == 20);
712 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_pid) == 4);
713 CLASSERT((int)offsetof(struct lnet_hdr, type) == 24);
714 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->type) == 4);
715 CLASSERT((int)offsetof(struct lnet_hdr, payload_length) == 28);
716 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->payload_length) == 4);
717 CLASSERT((int)offsetof(struct lnet_hdr, msg) == 32);
718 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg) == 40);
721 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) == 32);
722 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) == 16);
723 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.match_bits) == 48);
724 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) == 8);
725 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.mlength) == 56);
726 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) == 4);
729 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) == 32);
730 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) == 16);
731 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.match_bits) == 48);
732 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) == 8);
733 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.hdr_data) == 56);
734 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) == 8);
735 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ptl_index) == 64);
736 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) == 4);
737 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.offset) == 68);
738 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) == 4);
741 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.return_wmd) == 32);
742 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) == 16);
743 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.match_bits) == 48);
744 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) == 8);
745 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.ptl_index) == 56);
746 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) == 4);
747 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.src_offset) == 60);
748 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) == 4);
749 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.sink_length) == 64);
750 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) == 4);
753 CLASSERT((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) == 32);
754 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) == 16);
757 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.incarnation) == 32);
758 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) == 8);
759 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.type) == 40);
760 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) == 4);
762 /* Checks for struct lnet_ni_status and related constants */
763 CLASSERT(LNET_NI_STATUS_INVALID == 0x00000000);
764 CLASSERT(LNET_NI_STATUS_UP == 0x15aac0de);
765 CLASSERT(LNET_NI_STATUS_DOWN == 0xdeadface);
767 /* Checks for struct lnet_ni_status */
768 CLASSERT((int)sizeof(struct lnet_ni_status) == 16);
769 CLASSERT((int)offsetof(struct lnet_ni_status, ns_nid) == 0);
770 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) == 8);
771 CLASSERT((int)offsetof(struct lnet_ni_status, ns_status) == 8);
772 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_status) == 4);
773 CLASSERT((int)offsetof(struct lnet_ni_status, ns_unused) == 12);
774 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) == 4);
776 /* Checks for struct lnet_ping_info and related constants */
777 CLASSERT(LNET_PROTO_PING_MAGIC == 0x70696E67);
778 CLASSERT(LNET_PING_FEAT_INVAL == 0);
779 CLASSERT(LNET_PING_FEAT_BASE == 1);
780 CLASSERT(LNET_PING_FEAT_NI_STATUS == 2);
781 CLASSERT(LNET_PING_FEAT_RTE_DISABLED == 4);
782 CLASSERT(LNET_PING_FEAT_MULTI_RAIL == 8);
783 CLASSERT(LNET_PING_FEAT_DISCOVERY == 16);
784 CLASSERT(LNET_PING_FEAT_BITS == 31);
786 /* Checks for struct lnet_ping_info */
787 CLASSERT((int)sizeof(struct lnet_ping_info) == 16);
788 CLASSERT((int)offsetof(struct lnet_ping_info, pi_magic) == 0);
789 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) == 4);
790 CLASSERT((int)offsetof(struct lnet_ping_info, pi_features) == 4);
791 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_features) == 4);
792 CLASSERT((int)offsetof(struct lnet_ping_info, pi_pid) == 8);
793 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) == 4);
794 CLASSERT((int)offsetof(struct lnet_ping_info, pi_nnis) == 12);
795 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) == 4);
796 CLASSERT((int)offsetof(struct lnet_ping_info, pi_ni) == 16);
797 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) == 0);
800 static struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
802 struct lnet_lnd *lnd;
803 struct list_head *tmp;
805 /* holding lnd mutex */
806 list_for_each(tmp, &the_lnet.ln_lnds) {
807 lnd = list_entry(tmp, struct lnet_lnd, lnd_list);
809 if (lnd->lnd_type == type)
816 lnet_get_lnd_timeout(void)
818 return lnet_lnd_timeout;
820 EXPORT_SYMBOL(lnet_get_lnd_timeout);
823 lnet_register_lnd(struct lnet_lnd *lnd)
825 mutex_lock(&the_lnet.ln_lnd_mutex);
827 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
828 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
830 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
832 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
834 mutex_unlock(&the_lnet.ln_lnd_mutex);
836 EXPORT_SYMBOL(lnet_register_lnd);
839 lnet_unregister_lnd(struct lnet_lnd *lnd)
841 mutex_lock(&the_lnet.ln_lnd_mutex);
843 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
845 list_del(&lnd->lnd_list);
846 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
848 mutex_unlock(&the_lnet.ln_lnd_mutex);
850 EXPORT_SYMBOL(lnet_unregister_lnd);
853 lnet_counters_get_common(struct lnet_counters_common *common)
855 struct lnet_counters *ctr;
858 memset(common, 0, sizeof(*common));
860 lnet_net_lock(LNET_LOCK_EX);
862 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
863 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
864 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
865 common->lcc_errors += ctr->lct_common.lcc_errors;
866 common->lcc_send_count += ctr->lct_common.lcc_send_count;
867 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
868 common->lcc_route_count += ctr->lct_common.lcc_route_count;
869 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
870 common->lcc_send_length += ctr->lct_common.lcc_send_length;
871 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
872 common->lcc_route_length += ctr->lct_common.lcc_route_length;
873 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
875 lnet_net_unlock(LNET_LOCK_EX);
877 EXPORT_SYMBOL(lnet_counters_get_common);
880 lnet_counters_get(struct lnet_counters *counters)
882 struct lnet_counters *ctr;
883 struct lnet_counters_health *health = &counters->lct_health;
886 memset(counters, 0, sizeof(*counters));
888 lnet_counters_get_common(&counters->lct_common);
890 lnet_net_lock(LNET_LOCK_EX);
892 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
893 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
894 health->lch_resend_count += ctr->lct_health.lch_resend_count;
895 health->lch_response_timeout_count +=
896 ctr->lct_health.lch_response_timeout_count;
897 health->lch_local_interrupt_count +=
898 ctr->lct_health.lch_local_interrupt_count;
899 health->lch_local_dropped_count +=
900 ctr->lct_health.lch_local_dropped_count;
901 health->lch_local_aborted_count +=
902 ctr->lct_health.lch_local_aborted_count;
903 health->lch_local_no_route_count +=
904 ctr->lct_health.lch_local_no_route_count;
905 health->lch_local_timeout_count +=
906 ctr->lct_health.lch_local_timeout_count;
907 health->lch_local_error_count +=
908 ctr->lct_health.lch_local_error_count;
909 health->lch_remote_dropped_count +=
910 ctr->lct_health.lch_remote_dropped_count;
911 health->lch_remote_error_count +=
912 ctr->lct_health.lch_remote_error_count;
913 health->lch_remote_timeout_count +=
914 ctr->lct_health.lch_remote_timeout_count;
915 health->lch_network_timeout_count +=
916 ctr->lct_health.lch_network_timeout_count;
918 lnet_net_unlock(LNET_LOCK_EX);
920 EXPORT_SYMBOL(lnet_counters_get);
923 lnet_counters_reset(void)
925 struct lnet_counters *counters;
928 lnet_net_lock(LNET_LOCK_EX);
930 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
931 memset(counters, 0, sizeof(struct lnet_counters));
933 lnet_net_unlock(LNET_LOCK_EX);
937 lnet_res_type2str(int type)
942 case LNET_COOKIE_TYPE_MD:
944 case LNET_COOKIE_TYPE_ME:
946 case LNET_COOKIE_TYPE_EQ:
952 lnet_res_container_cleanup(struct lnet_res_container *rec)
956 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
959 while (!list_empty(&rec->rec_active)) {
960 struct list_head *e = rec->rec_active.next;
963 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
964 lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
966 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
967 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
969 } else { /* NB: Active MEs should be attached on portals */
976 /* Found alive MD/ME/EQ, user really should unlink/free
977 * all of them before finalize LNet, but if someone didn't,
978 * we have to recycle garbage for him */
979 CERROR("%d active elements on exit of %s container\n",
980 count, lnet_res_type2str(rec->rec_type));
983 if (rec->rec_lh_hash != NULL) {
984 LIBCFS_FREE(rec->rec_lh_hash,
985 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
986 rec->rec_lh_hash = NULL;
989 rec->rec_type = 0; /* mark it as finalized */
993 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
998 LASSERT(rec->rec_type == 0);
1000 rec->rec_type = type;
1001 INIT_LIST_HEAD(&rec->rec_active);
1003 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1005 /* Arbitrary choice of hash table size */
1006 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1007 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1008 if (rec->rec_lh_hash == NULL) {
1013 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1014 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1019 CERROR("Failed to setup %s resource container\n",
1020 lnet_res_type2str(type));
1021 lnet_res_container_cleanup(rec);
1026 lnet_res_containers_destroy(struct lnet_res_container **recs)
1028 struct lnet_res_container *rec;
1031 cfs_percpt_for_each(rec, i, recs)
1032 lnet_res_container_cleanup(rec);
1034 cfs_percpt_free(recs);
1037 static struct lnet_res_container **
1038 lnet_res_containers_create(int type)
1040 struct lnet_res_container **recs;
1041 struct lnet_res_container *rec;
1045 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1047 CERROR("Failed to allocate %s resource containers\n",
1048 lnet_res_type2str(type));
1052 cfs_percpt_for_each(rec, i, recs) {
1053 rc = lnet_res_container_setup(rec, i, type);
1055 lnet_res_containers_destroy(recs);
1063 struct lnet_libhandle *
1064 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1066 /* ALWAYS called with lnet_res_lock held */
1067 struct list_head *head;
1068 struct lnet_libhandle *lh;
1071 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1074 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1075 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1077 list_for_each_entry(lh, head, lh_hash_chain) {
1078 if (lh->lh_cookie == cookie)
1086 lnet_res_lh_initialize(struct lnet_res_container *rec,
1087 struct lnet_libhandle *lh)
1089 /* ALWAYS called with lnet_res_lock held */
1090 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1093 lh->lh_cookie = rec->rec_lh_cookie;
1094 rec->rec_lh_cookie += 1 << ibits;
1096 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1098 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1102 lnet_create_array_of_queues(void)
1104 struct list_head **qs;
1105 struct list_head *q;
1108 qs = cfs_percpt_alloc(lnet_cpt_table(),
1109 sizeof(struct list_head));
1111 CERROR("Failed to allocate queues\n");
1115 cfs_percpt_for_each(q, i, qs)
1121 static int lnet_unprepare(void);
1124 lnet_prepare(lnet_pid_t requested_pid)
1126 /* Prepare to bring up the network */
1127 struct lnet_res_container **recs;
1130 if (requested_pid == LNET_PID_ANY) {
1131 /* Don't instantiate LNET just for me */
1135 LASSERT(the_lnet.ln_refcount == 0);
1137 the_lnet.ln_routing = 0;
1139 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1140 the_lnet.ln_pid = requested_pid;
1142 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1143 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1144 INIT_LIST_HEAD(&the_lnet.ln_nets);
1145 INIT_LIST_HEAD(&the_lnet.ln_routers);
1146 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1147 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1148 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1149 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1150 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1151 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1152 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1153 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1154 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1155 init_completion(&the_lnet.ln_started);
1157 rc = lnet_descriptor_setup();
1161 rc = lnet_create_remote_nets_table();
1166 * NB the interface cookie in wire handles guards against delayed
1167 * replies and ACKs appearing valid after reboot.
1169 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1171 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1172 sizeof(struct lnet_counters));
1173 if (the_lnet.ln_counters == NULL) {
1174 CERROR("Failed to allocate counters for LNet\n");
1179 rc = lnet_peer_tables_create();
1183 rc = lnet_msg_containers_create();
1187 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1188 LNET_COOKIE_TYPE_EQ);
1192 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
1198 the_lnet.ln_me_containers = recs;
1200 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1206 the_lnet.ln_md_containers = recs;
1208 rc = lnet_portals_create();
1210 CERROR("Failed to create portals for LNet: %d\n", rc);
1214 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1215 if (!the_lnet.ln_mt_zombie_rstqs) {
1228 lnet_unprepare (void)
1232 /* NB no LNET_LOCK since this is the last reference. All LND instances
1233 * have shut down already, so it is safe to unlink and free all
1234 * descriptors, even those that appear committed to a network op (eg MD
1235 * with non-zero pending count) */
1237 lnet_fail_nid(LNET_NID_ANY, 0);
1239 LASSERT(the_lnet.ln_refcount == 0);
1240 LASSERT(list_empty(&the_lnet.ln_test_peers));
1241 LASSERT(list_empty(&the_lnet.ln_nets));
1243 if (the_lnet.ln_mt_zombie_rstqs) {
1244 lnet_clean_zombie_rstqs();
1245 the_lnet.ln_mt_zombie_rstqs = NULL;
1248 if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1249 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1250 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1254 lnet_portals_destroy();
1256 if (the_lnet.ln_md_containers != NULL) {
1257 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1258 the_lnet.ln_md_containers = NULL;
1261 if (the_lnet.ln_me_containers != NULL) {
1262 lnet_res_containers_destroy(the_lnet.ln_me_containers);
1263 the_lnet.ln_me_containers = NULL;
1266 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1268 lnet_msg_containers_destroy();
1270 lnet_rtrpools_free(0);
1272 if (the_lnet.ln_counters != NULL) {
1273 cfs_percpt_free(the_lnet.ln_counters);
1274 the_lnet.ln_counters = NULL;
1276 lnet_destroy_remote_nets_table();
1277 lnet_descriptor_cleanup();
1283 lnet_net2ni_locked(__u32 net_id, int cpt)
1286 struct lnet_net *net;
1288 LASSERT(cpt != LNET_LOCK_EX);
1290 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1291 if (net->net_id == net_id) {
1292 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1302 lnet_net2ni_addref(__u32 net)
1307 ni = lnet_net2ni_locked(net, 0);
1309 lnet_ni_addref_locked(ni, 0);
1314 EXPORT_SYMBOL(lnet_net2ni_addref);
1317 lnet_get_net_locked(__u32 net_id)
1319 struct lnet_net *net;
1321 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1322 if (net->net_id == net_id)
1330 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1335 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1340 val = hash_long(key, LNET_CPT_BITS);
1341 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1345 return (unsigned int)(key + val + (val >> 1)) % number;
1349 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1351 struct lnet_net *net;
1353 /* must called with hold of lnet_net_lock */
1354 if (LNET_CPT_NUMBER == 1)
1355 return 0; /* the only one */
1358 * If NI is provided then use the CPT identified in the NI cpt
1359 * list if one exists. If one doesn't exist, then that NI is
1360 * associated with all CPTs and it follows that the net it belongs
1361 * to is implicitly associated with all CPTs, so just hash the nid
1365 if (ni->ni_cpts != NULL)
1366 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1369 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1372 /* no NI provided so look at the net */
1373 net = lnet_get_net_locked(LNET_NIDNET(nid));
1375 if (net != NULL && net->net_cpts != NULL) {
1376 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1379 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1383 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1388 if (LNET_CPT_NUMBER == 1)
1389 return 0; /* the only one */
1391 cpt = lnet_net_lock_current();
1393 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1395 lnet_net_unlock(cpt);
1399 EXPORT_SYMBOL(lnet_cpt_of_nid);
1402 lnet_islocalnet_locked(__u32 net_id)
1404 struct lnet_net *net;
1407 net = lnet_get_net_locked(net_id);
1409 local = net != NULL;
1415 lnet_islocalnet(__u32 net_id)
1420 cpt = lnet_net_lock_current();
1422 local = lnet_islocalnet_locked(net_id);
1424 lnet_net_unlock(cpt);
1430 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1432 struct lnet_net *net;
1435 LASSERT(cpt != LNET_LOCK_EX);
1437 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1438 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1439 if (ni->ni_nid == nid)
1448 lnet_nid2ni_addref(lnet_nid_t nid)
1453 ni = lnet_nid2ni_locked(nid, 0);
1455 lnet_ni_addref_locked(ni, 0);
1460 EXPORT_SYMBOL(lnet_nid2ni_addref);
1463 lnet_islocalnid(lnet_nid_t nid)
1468 cpt = lnet_net_lock_current();
1469 ni = lnet_nid2ni_locked(nid, cpt);
1470 lnet_net_unlock(cpt);
1476 lnet_count_acceptor_nets(void)
1478 /* Return the # of NIs that need the acceptor. */
1480 struct lnet_net *net;
1483 cpt = lnet_net_lock_current();
1484 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1485 /* all socklnd type networks should have the acceptor
1487 if (net->net_lnd->lnd_accept != NULL)
1491 lnet_net_unlock(cpt);
1496 struct lnet_ping_buffer *
1497 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1499 struct lnet_ping_buffer *pbuf;
1501 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1503 pbuf->pb_nnis = nnis;
1504 atomic_set(&pbuf->pb_refcnt, 1);
1511 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1513 LASSERT(lnet_ping_buffer_numref(pbuf) == 0);
1514 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1517 static struct lnet_ping_buffer *
1518 lnet_ping_target_create(int nnis)
1520 struct lnet_ping_buffer *pbuf;
1522 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1524 CERROR("Can't allocate ping source [%d]\n", nnis);
1528 pbuf->pb_info.pi_nnis = nnis;
1529 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1530 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1531 pbuf->pb_info.pi_features =
1532 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1538 lnet_get_net_ni_count_locked(struct lnet_net *net)
1543 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1550 lnet_get_net_ni_count_pre(struct lnet_net *net)
1555 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1562 lnet_get_ni_count(void)
1565 struct lnet_net *net;
1570 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1571 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1581 lnet_get_net_count(void)
1583 struct lnet_net *net;
1588 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1598 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1600 struct lnet_ni_status *stat;
1604 __swab32s(&pbuf->pb_info.pi_magic);
1605 __swab32s(&pbuf->pb_info.pi_features);
1606 __swab32s(&pbuf->pb_info.pi_pid);
1607 __swab32s(&pbuf->pb_info.pi_nnis);
1608 nnis = pbuf->pb_info.pi_nnis;
1609 if (nnis > pbuf->pb_nnis)
1610 nnis = pbuf->pb_nnis;
1611 for (i = 0; i < nnis; i++) {
1612 stat = &pbuf->pb_info.pi_ni[i];
1613 __swab64s(&stat->ns_nid);
1614 __swab32s(&stat->ns_status);
1619 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1623 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1625 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1627 /* Loopback is guaranteed to be present */
1628 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1630 if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1636 lnet_ping_target_destroy(void)
1638 struct lnet_net *net;
1641 lnet_net_lock(LNET_LOCK_EX);
1643 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1644 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1646 ni->ni_status = NULL;
1651 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1652 the_lnet.ln_ping_target = NULL;
1654 lnet_net_unlock(LNET_LOCK_EX);
1658 lnet_ping_target_event_handler(struct lnet_event *event)
1660 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1662 if (event->unlinked)
1663 lnet_ping_buffer_decref(pbuf);
1667 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1668 struct lnet_handle_md *ping_mdh,
1669 int ni_count, bool set_eq)
1671 struct lnet_process_id id = {
1672 .nid = LNET_NID_ANY,
1675 struct lnet_handle_me me_handle;
1676 struct lnet_md md = { NULL };
1680 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1681 &the_lnet.ln_ping_target_eq);
1683 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1688 *ppbuf = lnet_ping_target_create(ni_count);
1689 if (*ppbuf == NULL) {
1694 /* Ping target ME/MD */
1695 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1696 LNET_PROTO_PING_MATCHBITS, 0,
1697 LNET_UNLINK, LNET_INS_AFTER,
1700 CERROR("Can't create ping target ME: %d\n", rc);
1701 goto fail_decref_ping_buffer;
1704 /* initialize md content */
1705 md.start = &(*ppbuf)->pb_info;
1706 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1707 md.threshold = LNET_MD_THRESH_INF;
1709 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1710 LNET_MD_MANAGE_REMOTE;
1711 md.eq_handle = the_lnet.ln_ping_target_eq;
1712 md.user_ptr = *ppbuf;
1714 rc = LNetMDAttach(me_handle, md, LNET_RETAIN, ping_mdh);
1716 CERROR("Can't attach ping target MD: %d\n", rc);
1717 goto fail_unlink_ping_me;
1719 lnet_ping_buffer_addref(*ppbuf);
1723 fail_unlink_ping_me:
1724 rc2 = LNetMEUnlink(me_handle);
1726 fail_decref_ping_buffer:
1727 LASSERT(lnet_ping_buffer_numref(*ppbuf) == 1);
1728 lnet_ping_buffer_decref(*ppbuf);
1732 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1739 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1740 struct lnet_handle_md *ping_mdh)
1742 sigset_t blocked = cfs_block_allsigs();
1744 LNetMDUnlink(*ping_mdh);
1745 LNetInvalidateMDHandle(ping_mdh);
1747 /* NB the MD could be busy; this just starts the unlink */
1748 while (lnet_ping_buffer_numref(pbuf) > 1) {
1749 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1750 set_current_state(TASK_UNINTERRUPTIBLE);
1751 schedule_timeout(cfs_time_seconds(1));
1754 cfs_restore_sigs(blocked);
1758 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1761 struct lnet_net *net;
1762 struct lnet_ni_status *ns;
1767 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1768 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1769 LASSERT(i < pbuf->pb_nnis);
1771 ns = &pbuf->pb_info.pi_ni[i];
1773 ns->ns_nid = ni->ni_nid;
1776 ns->ns_status = (ni->ni_status != NULL) ?
1777 ni->ni_status->ns_status :
1786 * We (ab)use the ns_status of the loopback interface to
1787 * transmit the sequence number. The first interface listed
1788 * must be the loopback interface.
1790 rc = lnet_ping_info_validate(&pbuf->pb_info);
1792 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1795 LNET_PING_BUFFER_SEQNO(pbuf) =
1796 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1800 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1801 struct lnet_handle_md ping_mdh)
1803 struct lnet_ping_buffer *old_pbuf = NULL;
1804 struct lnet_handle_md old_ping_md;
1806 /* switch the NIs to point to the new ping info created */
1807 lnet_net_lock(LNET_LOCK_EX);
1809 if (!the_lnet.ln_routing)
1810 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1811 if (!lnet_peer_discovery_disabled)
1812 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1814 /* Ensure only known feature bits have been set. */
1815 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1816 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1818 lnet_ping_target_install_locked(pbuf);
1820 if (the_lnet.ln_ping_target) {
1821 old_pbuf = the_lnet.ln_ping_target;
1822 old_ping_md = the_lnet.ln_ping_target_md;
1824 the_lnet.ln_ping_target_md = ping_mdh;
1825 the_lnet.ln_ping_target = pbuf;
1827 lnet_net_unlock(LNET_LOCK_EX);
1830 /* unlink and free the old ping info */
1831 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1832 lnet_ping_buffer_decref(old_pbuf);
1835 lnet_push_update_to_peers(0);
1839 lnet_ping_target_fini(void)
1843 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1844 &the_lnet.ln_ping_target_md);
1846 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1849 lnet_ping_target_destroy();
1852 /* Resize the push target. */
1853 int lnet_push_target_resize(void)
1855 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1856 struct lnet_md md = { NULL };
1857 struct lnet_handle_me meh;
1858 struct lnet_handle_md mdh;
1859 struct lnet_handle_md old_mdh;
1860 struct lnet_ping_buffer *pbuf;
1861 struct lnet_ping_buffer *old_pbuf;
1862 int nnis = the_lnet.ln_push_target_nnis;
1870 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1876 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1877 LNET_PROTO_PING_MATCHBITS, 0,
1878 LNET_UNLINK, LNET_INS_AFTER,
1881 CERROR("Can't create push target ME: %d\n", rc);
1882 goto fail_decref_pbuf;
1885 /* initialize md content */
1886 md.start = &pbuf->pb_info;
1887 md.length = LNET_PING_INFO_SIZE(nnis);
1888 md.threshold = LNET_MD_THRESH_INF;
1890 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1891 LNET_MD_MANAGE_REMOTE;
1893 md.eq_handle = the_lnet.ln_push_target_eq;
1895 rc = LNetMDAttach(meh, md, LNET_RETAIN, &mdh);
1897 CERROR("Can't attach push MD: %d\n", rc);
1898 goto fail_unlink_meh;
1900 lnet_ping_buffer_addref(pbuf);
1902 lnet_net_lock(LNET_LOCK_EX);
1903 old_pbuf = the_lnet.ln_push_target;
1904 old_mdh = the_lnet.ln_push_target_md;
1905 the_lnet.ln_push_target = pbuf;
1906 the_lnet.ln_push_target_md = mdh;
1907 lnet_net_unlock(LNET_LOCK_EX);
1910 LNetMDUnlink(old_mdh);
1911 lnet_ping_buffer_decref(old_pbuf);
1914 if (nnis < the_lnet.ln_push_target_nnis)
1917 CDEBUG(D_NET, "nnis %d success\n", nnis);
1924 lnet_ping_buffer_decref(pbuf);
1926 CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1930 static void lnet_push_target_event_handler(struct lnet_event *ev)
1932 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1934 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1935 lnet_swap_pinginfo(pbuf);
1937 lnet_peer_push_event(ev);
1939 lnet_ping_buffer_decref(pbuf);
1942 /* Initialize the push target. */
1943 static int lnet_push_target_init(void)
1947 if (the_lnet.ln_push_target)
1950 rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1951 &the_lnet.ln_push_target_eq);
1953 CERROR("Can't allocated push target EQ: %d\n", rc);
1957 /* Start at the required minimum, we'll enlarge if required. */
1958 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1960 rc = lnet_push_target_resize();
1963 LNetEQFree(the_lnet.ln_push_target_eq);
1964 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1970 /* Clean up the push target. */
1971 static void lnet_push_target_fini(void)
1973 if (!the_lnet.ln_push_target)
1976 /* Unlink and invalidate to prevent new references. */
1977 LNetMDUnlink(the_lnet.ln_push_target_md);
1978 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1980 /* Wait for the unlink to complete. */
1981 while (lnet_ping_buffer_numref(the_lnet.ln_push_target) > 1) {
1982 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1983 set_current_state(TASK_UNINTERRUPTIBLE);
1984 schedule_timeout(cfs_time_seconds(1));
1987 lnet_ping_buffer_decref(the_lnet.ln_push_target);
1988 the_lnet.ln_push_target = NULL;
1989 the_lnet.ln_push_target_nnis = 0;
1991 LNetEQFree(the_lnet.ln_push_target_eq);
1992 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1996 lnet_ni_tq_credits(struct lnet_ni *ni)
2000 LASSERT(ni->ni_ncpts >= 1);
2002 if (ni->ni_ncpts == 1)
2003 return ni->ni_net->net_tunables.lct_max_tx_credits;
2005 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2006 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2007 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2013 lnet_ni_unlink_locked(struct lnet_ni *ni)
2015 /* move it to zombie list and nobody can find it anymore */
2016 LASSERT(!list_empty(&ni->ni_netlist));
2017 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2018 lnet_ni_decref_locked(ni, 0);
2022 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2027 struct list_head *zombie_list = &net->net_ni_zombie;
2030 * Now wait for the NIs I just nuked to show up on the zombie
2031 * list and shut them down in guaranteed thread context
2034 while (!list_empty(zombie_list)) {
2038 ni = list_entry(zombie_list->next,
2039 struct lnet_ni, ni_netlist);
2040 list_del_init(&ni->ni_netlist);
2041 /* the ni should be in deleting state. If it's not it's
2043 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2044 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2047 /* still busy, add it back to zombie list */
2048 list_add(&ni->ni_netlist, zombie_list);
2052 if (!list_empty(&ni->ni_netlist)) {
2053 lnet_net_unlock(LNET_LOCK_EX);
2055 if ((i & (-i)) == i) {
2057 "Waiting for zombie LNI %s\n",
2058 libcfs_nid2str(ni->ni_nid));
2060 set_current_state(TASK_UNINTERRUPTIBLE);
2061 schedule_timeout(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 (net->net_lnd->lnd_shutdown)(ni);
2074 CDEBUG(D_LNI, "Removed LNI %s\n",
2075 libcfs_nid2str(ni->ni_nid));
2079 lnet_net_lock(LNET_LOCK_EX);
2083 /* shutdown down the NI and release refcount */
2085 lnet_shutdown_lndni(struct lnet_ni *ni)
2088 struct lnet_net *net = ni->ni_net;
2090 lnet_net_lock(LNET_LOCK_EX);
2092 ni->ni_state = LNET_NI_STATE_DELETING;
2094 lnet_ni_unlink_locked(ni);
2095 lnet_incr_dlc_seq();
2096 lnet_net_unlock(LNET_LOCK_EX);
2098 /* clear messages for this NI on the lazy portal */
2099 for (i = 0; i < the_lnet.ln_nportals; i++)
2100 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2102 lnet_net_lock(LNET_LOCK_EX);
2103 lnet_clear_zombies_nis_locked(net);
2104 lnet_net_unlock(LNET_LOCK_EX);
2108 lnet_shutdown_lndnet(struct lnet_net *net)
2112 lnet_net_lock(LNET_LOCK_EX);
2114 list_del_init(&net->net_list);
2116 while (!list_empty(&net->net_ni_list)) {
2117 ni = list_entry(net->net_ni_list.next,
2118 struct lnet_ni, ni_netlist);
2119 lnet_net_unlock(LNET_LOCK_EX);
2120 lnet_shutdown_lndni(ni);
2121 lnet_net_lock(LNET_LOCK_EX);
2124 lnet_net_unlock(LNET_LOCK_EX);
2126 /* Do peer table cleanup for this net */
2127 lnet_peer_tables_cleanup(net);
2133 lnet_shutdown_lndnets(void)
2135 struct lnet_net *net;
2136 struct list_head resend;
2137 struct lnet_msg *msg, *tmp;
2139 INIT_LIST_HEAD(&resend);
2141 /* NB called holding the global mutex */
2143 /* All quiet on the API front */
2144 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2145 LASSERT(the_lnet.ln_refcount == 0);
2147 lnet_net_lock(LNET_LOCK_EX);
2148 the_lnet.ln_state = LNET_STATE_STOPPING;
2150 while (!list_empty(&the_lnet.ln_nets)) {
2152 * move the nets to the zombie list to avoid them being
2153 * picked up for new work. LONET is also included in the
2154 * Nets that will be moved to the zombie list
2156 net = list_entry(the_lnet.ln_nets.next,
2157 struct lnet_net, net_list);
2158 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2161 /* Drop the cached loopback Net. */
2162 if (the_lnet.ln_loni != NULL) {
2163 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2164 the_lnet.ln_loni = NULL;
2166 lnet_net_unlock(LNET_LOCK_EX);
2168 /* iterate through the net zombie list and delete each net */
2169 while (!list_empty(&the_lnet.ln_net_zombie)) {
2170 net = list_entry(the_lnet.ln_net_zombie.next,
2171 struct lnet_net, net_list);
2172 lnet_shutdown_lndnet(net);
2175 spin_lock(&the_lnet.ln_msg_resend_lock);
2176 list_splice(&the_lnet.ln_msg_resend, &resend);
2177 spin_unlock(&the_lnet.ln_msg_resend_lock);
2179 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2180 list_del_init(&msg->msg_list);
2181 msg->msg_no_resend = true;
2182 lnet_finalize(msg, -ECANCELED);
2185 lnet_net_lock(LNET_LOCK_EX);
2186 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2187 lnet_net_unlock(LNET_LOCK_EX);
2191 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2194 struct lnet_tx_queue *tq;
2196 struct lnet_net *net = ni->ni_net;
2198 mutex_lock(&the_lnet.ln_lnd_mutex);
2201 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2202 ni->ni_lnd_tunables_set = true;
2205 rc = (net->net_lnd->lnd_startup)(ni);
2207 mutex_unlock(&the_lnet.ln_lnd_mutex);
2210 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2211 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2216 ni->ni_state = LNET_NI_STATE_ACTIVE;
2219 /* We keep a reference on the loopback net through the loopback NI */
2220 if (net->net_lnd->lnd_type == LOLND) {
2222 LASSERT(the_lnet.ln_loni == NULL);
2223 the_lnet.ln_loni = ni;
2224 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2225 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2226 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2227 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2231 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2232 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2233 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2234 libcfs_lnd2str(net->net_lnd->lnd_type),
2235 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2237 /* shutdown the NI since if we get here then it must've already
2240 lnet_shutdown_lndni(ni);
2244 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2245 tq->tq_credits_min =
2246 tq->tq_credits_max =
2247 tq->tq_credits = lnet_ni_tq_credits(ni);
2250 atomic_set(&ni->ni_tx_credits,
2251 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2252 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2254 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2255 libcfs_nid2str(ni->ni_nid),
2256 ni->ni_net->net_tunables.lct_peer_tx_credits,
2257 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2258 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2259 ni->ni_net->net_tunables.lct_peer_timeout);
2268 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2271 struct lnet_net *net_l = NULL;
2272 struct list_head local_ni_list;
2276 struct lnet_lnd *lnd;
2278 net->net_tunables.lct_peer_timeout;
2280 net->net_tunables.lct_max_tx_credits;
2281 int peerrtrcredits =
2282 net->net_tunables.lct_peer_rtr_credits;
2284 INIT_LIST_HEAD(&local_ni_list);
2287 * make sure that this net is unique. If it isn't then
2288 * we are adding interfaces to an already existing network, and
2289 * 'net' is just a convenient way to pass in the list.
2290 * if it is unique we need to find the LND and load it if
2293 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2294 lnd_type = LNET_NETTYP(net->net_id);
2296 mutex_lock(&the_lnet.ln_lnd_mutex);
2297 lnd = lnet_find_lnd_by_type(lnd_type);
2300 mutex_unlock(&the_lnet.ln_lnd_mutex);
2301 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2302 mutex_lock(&the_lnet.ln_lnd_mutex);
2304 lnd = lnet_find_lnd_by_type(lnd_type);
2306 mutex_unlock(&the_lnet.ln_lnd_mutex);
2307 CERROR("Can't load LND %s, module %s, rc=%d\n",
2308 libcfs_lnd2str(lnd_type),
2309 libcfs_lnd2modname(lnd_type), rc);
2310 #ifndef HAVE_MODULE_LOADING_SUPPORT
2311 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2312 "compiled with kernel module "
2313 "loading support.");
2322 mutex_unlock(&the_lnet.ln_lnd_mutex);
2328 * net_l: if the network being added is unique then net_l
2329 * will point to that network
2330 * if the network being added is not unique then
2331 * net_l points to the existing network.
2333 * When we enter the loop below, we'll pick NIs off he
2334 * network beign added and start them up, then add them to
2335 * a local ni list. Once we've successfully started all
2336 * the NIs then we join the local NI list (of started up
2337 * networks) with the net_l->net_ni_list, which should
2338 * point to the correct network to add the new ni list to
2340 * If any of the new NIs fail to start up, then we want to
2341 * iterate through the local ni list, which should include
2342 * any NIs which were successfully started up, and shut
2345 * After than we want to delete the network being added,
2346 * to avoid a memory leak.
2350 * When a network uses TCP bonding then all its interfaces
2351 * must be specified when the network is first defined: the
2352 * TCP bonding code doesn't allow for interfaces to be added
2355 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2356 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2361 while (!list_empty(&net->net_ni_added)) {
2362 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2364 list_del_init(&ni->ni_netlist);
2366 /* make sure that the the NI we're about to start
2367 * up is actually unique. if it's not fail. */
2368 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2369 ni->ni_interfaces[0])) {
2374 /* adjust the pointer the parent network, just in case it
2375 * the net is a duplicate */
2378 rc = lnet_startup_lndni(ni, tun);
2380 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
2381 ni->ni_net->net_lnd->lnd_query != NULL);
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_lnds);
2521 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2522 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2524 /* The hash table size is the number of bits it takes to express the set
2525 * ln_num_routes, minus 1 (better to under estimate than over so we
2526 * don't waste memory). */
2527 if (rnet_htable_size <= 0)
2528 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2529 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2530 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2531 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2532 order_base_2(rnet_htable_size) - 1);
2534 /* All LNDs apart from the LOLND are in separate modules. They
2535 * register themselves when their module loads, and unregister
2536 * themselves when their module is unloaded. */
2537 lnet_register_lnd(&the_lolnd);
2542 * Finalize LNet library.
2544 * \pre lnet_lib_init() called with success.
2545 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2547 void lnet_lib_exit(void)
2549 LASSERT(the_lnet.ln_refcount == 0);
2551 while (!list_empty(&the_lnet.ln_lnds))
2552 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
2553 struct lnet_lnd, lnd_list));
2554 lnet_destroy_locks();
2558 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2560 * Users must call this function at least once before any other functions.
2561 * For each successful call there must be a corresponding call to
2562 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2565 * The PID used by LNet may be different from the one requested.
2568 * \param requested_pid PID requested by the caller.
2570 * \return >= 0 on success, and < 0 error code on failures.
2573 LNetNIInit(lnet_pid_t requested_pid)
2575 int im_a_router = 0;
2578 struct lnet_ping_buffer *pbuf;
2579 struct lnet_handle_md ping_mdh;
2580 struct list_head net_head;
2581 struct lnet_net *net;
2583 INIT_LIST_HEAD(&net_head);
2585 mutex_lock(&the_lnet.ln_api_mutex);
2587 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2589 if (the_lnet.ln_refcount > 0) {
2590 rc = the_lnet.ln_refcount++;
2591 mutex_unlock(&the_lnet.ln_api_mutex);
2595 rc = lnet_prepare(requested_pid);
2597 mutex_unlock(&the_lnet.ln_api_mutex);
2601 /* create a network for Loopback network */
2602 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2605 goto err_empty_list;
2608 /* Add in the loopback NI */
2609 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2611 goto err_empty_list;
2614 /* If LNet is being initialized via DLC it is possible
2615 * that the user requests not to load module parameters (ones which
2616 * are supported by DLC) on initialization. Therefore, make sure not
2617 * to load networks, routes and forwarding from module parameters
2618 * in this case. On cleanup in case of failure only clean up
2619 * routes if it has been loaded */
2620 if (!the_lnet.ln_nis_from_mod_params) {
2621 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2624 goto err_empty_list;
2627 ni_count = lnet_startup_lndnets(&net_head);
2630 goto err_empty_list;
2633 if (!the_lnet.ln_nis_from_mod_params) {
2634 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2636 goto err_shutdown_lndnis;
2638 rc = lnet_rtrpools_alloc(im_a_router);
2640 goto err_destroy_routes;
2643 rc = lnet_acceptor_start();
2645 goto err_destroy_routes;
2647 the_lnet.ln_refcount = 1;
2648 /* Now I may use my own API functions... */
2650 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2652 goto err_acceptor_stop;
2654 lnet_ping_target_update(pbuf, ping_mdh);
2656 rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2658 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2662 rc = lnet_push_target_init();
2666 rc = lnet_peer_discovery_start();
2668 goto err_destroy_push_target;
2670 rc = lnet_monitor_thr_start();
2672 goto err_stop_discovery_thr;
2675 lnet_router_debugfs_init();
2677 mutex_unlock(&the_lnet.ln_api_mutex);
2679 complete_all(&the_lnet.ln_started);
2681 /* wait for all routers to start */
2682 lnet_wait_router_start();
2686 err_stop_discovery_thr:
2687 lnet_peer_discovery_stop();
2688 err_destroy_push_target:
2689 lnet_push_target_fini();
2691 lnet_ping_target_fini();
2693 the_lnet.ln_refcount = 0;
2694 lnet_acceptor_stop();
2696 if (!the_lnet.ln_nis_from_mod_params)
2697 lnet_destroy_routes();
2698 err_shutdown_lndnis:
2699 lnet_shutdown_lndnets();
2703 mutex_unlock(&the_lnet.ln_api_mutex);
2704 while (!list_empty(&net_head)) {
2705 struct lnet_net *net;
2707 net = list_entry(net_head.next, struct lnet_net, net_list);
2708 list_del_init(&net->net_list);
2713 EXPORT_SYMBOL(LNetNIInit);
2716 * Stop LNet interfaces, routing, and forwarding.
2718 * Users must call this function once for each successful call to LNetNIInit().
2719 * Once the LNetNIFini() operation has been started, the results of pending
2720 * API operations are undefined.
2722 * \return always 0 for current implementation.
2727 mutex_lock(&the_lnet.ln_api_mutex);
2729 LASSERT(the_lnet.ln_refcount > 0);
2731 if (the_lnet.ln_refcount != 1) {
2732 the_lnet.ln_refcount--;
2734 LASSERT(!the_lnet.ln_niinit_self);
2738 lnet_router_debugfs_fini();
2739 lnet_monitor_thr_stop();
2740 lnet_peer_discovery_stop();
2741 lnet_push_target_fini();
2742 lnet_ping_target_fini();
2744 /* Teardown fns that use my own API functions BEFORE here */
2745 the_lnet.ln_refcount = 0;
2747 lnet_acceptor_stop();
2748 lnet_destroy_routes();
2749 lnet_shutdown_lndnets();
2753 mutex_unlock(&the_lnet.ln_api_mutex);
2756 EXPORT_SYMBOL(LNetNIFini);
2759 * Grabs the ni data from the ni structure and fills the out
2762 * \param[in] ni network interface structure
2763 * \param[out] cfg_ni NI config information
2764 * \param[out] tun network and LND tunables
2767 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2768 struct lnet_ioctl_config_lnd_tunables *tun,
2769 struct lnet_ioctl_element_stats *stats,
2772 size_t min_size = 0;
2775 if (!ni || !cfg_ni || !tun)
2778 if (ni->ni_interfaces[0] != NULL) {
2779 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2780 if (ni->ni_interfaces[i] != NULL) {
2781 strncpy(cfg_ni->lic_ni_intf[i],
2782 ni->ni_interfaces[i],
2783 sizeof(cfg_ni->lic_ni_intf[i]));
2788 cfg_ni->lic_nid = ni->ni_nid;
2789 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2790 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2792 cfg_ni->lic_status = ni->ni_status->ns_status;
2793 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2794 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2796 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2799 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2800 LNET_STATS_TYPE_SEND);
2801 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2802 LNET_STATS_TYPE_RECV);
2803 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2804 LNET_STATS_TYPE_DROP);
2808 * tun->lt_tun will always be present, but in order to be
2809 * backwards compatible, we need to deal with the cases when
2810 * tun->lt_tun is smaller than what the kernel has, because it
2811 * comes from an older version of a userspace program, then we'll
2812 * need to copy as much information as we have available space.
2814 min_size = tun_size - sizeof(tun->lt_cmn);
2815 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2817 /* copy over the cpts */
2818 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2819 ni->ni_cpts == NULL) {
2820 for (i = 0; i < ni->ni_ncpts; i++)
2821 cfg_ni->lic_cpts[i] = i;
2824 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2825 i < LNET_MAX_SHOW_NUM_CPT;
2827 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2829 cfg_ni->lic_ncpts = ni->ni_ncpts;
2833 * NOTE: This is a legacy function left in the code to be backwards
2834 * compatible with older userspace programs. It should eventually be
2837 * Grabs the ni data from the ni structure and fills the out
2840 * \param[in] ni network interface structure
2841 * \param[out] config config information
2844 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2845 struct lnet_ioctl_config_data *config)
2847 struct lnet_ioctl_net_config *net_config;
2848 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2849 size_t min_size, tunable_size = 0;
2855 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2859 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2860 ARRAY_SIZE(net_config->ni_interfaces));
2862 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2863 if (!ni->ni_interfaces[i])
2866 strncpy(net_config->ni_interfaces[i],
2867 ni->ni_interfaces[i],
2868 sizeof(net_config->ni_interfaces[i]));
2871 config->cfg_nid = ni->ni_nid;
2872 config->cfg_config_u.cfg_net.net_peer_timeout =
2873 ni->ni_net->net_tunables.lct_peer_timeout;
2874 config->cfg_config_u.cfg_net.net_max_tx_credits =
2875 ni->ni_net->net_tunables.lct_max_tx_credits;
2876 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2877 ni->ni_net->net_tunables.lct_peer_tx_credits;
2878 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2879 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2881 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2882 net_config->ni_status = LNET_NI_STATUS_UP;
2884 net_config->ni_status = ni->ni_status->ns_status;
2887 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2889 for (i = 0; i < num_cpts; i++)
2890 net_config->ni_cpts[i] = ni->ni_cpts[i];
2892 config->cfg_ncpts = num_cpts;
2896 * See if user land tools sent in a newer and larger version
2897 * of struct lnet_tunables than what the kernel uses.
2899 min_size = sizeof(*config) + sizeof(*net_config);
2901 if (config->cfg_hdr.ioc_len > min_size)
2902 tunable_size = config->cfg_hdr.ioc_len - min_size;
2904 /* Don't copy too much data to user space */
2905 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2906 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2908 if (lnd_cfg && min_size) {
2909 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2910 config->cfg_config_u.cfg_net.net_interface_count = 1;
2912 /* Tell user land that kernel side has less data */
2913 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2914 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2915 config->cfg_hdr.ioc_len -= min_size;
2921 lnet_get_ni_idx_locked(int idx)
2924 struct lnet_net *net;
2926 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2927 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2937 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2940 struct lnet_net *net = mynet;
2943 * It is possible that the net has been cleaned out while there is
2944 * a message being sent. This function accessed the net without
2945 * checking if the list is empty
2949 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2951 if (list_empty(&net->net_ni_list))
2953 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2959 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2960 /* if you reached the end of the ni list and the net is
2961 * specified, then there are no more nis in that net */
2965 /* we reached the end of this net ni list. move to the
2967 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2968 /* no more nets and no more NIs. */
2971 /* get the next net */
2972 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2974 if (list_empty(&net->net_ni_list))
2976 /* get the ni on it */
2977 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2983 if (list_empty(&prev->ni_netlist))
2986 /* there are more nis left */
2987 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2993 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2998 int idx = config->cfg_count;
3000 cpt = lnet_net_lock_current();
3002 ni = lnet_get_ni_idx_locked(idx);
3007 lnet_fill_ni_info_legacy(ni, config);
3011 lnet_net_unlock(cpt);
3016 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3017 struct lnet_ioctl_config_lnd_tunables *tun,
3018 struct lnet_ioctl_element_stats *stats,
3025 if (!cfg_ni || !tun || !stats)
3028 cpt = lnet_net_lock_current();
3030 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3035 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3039 lnet_net_unlock(cpt);
3043 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3052 cpt = lnet_net_lock_current();
3054 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3057 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3061 lnet_net_unlock(cpt);
3066 static int lnet_add_net_common(struct lnet_net *net,
3067 struct lnet_ioctl_config_lnd_tunables *tun)
3070 struct lnet_ping_buffer *pbuf;
3071 struct lnet_handle_md ping_mdh;
3073 struct lnet_remotenet *rnet;
3075 int num_acceptor_nets;
3077 lnet_net_lock(LNET_LOCK_EX);
3078 rnet = lnet_find_rnet_locked(net->net_id);
3079 lnet_net_unlock(LNET_LOCK_EX);
3081 * make sure that the net added doesn't invalidate the current
3082 * configuration LNet is keeping
3085 CERROR("Adding net %s will invalidate routing configuration\n",
3086 libcfs_net2str(net->net_id));
3092 * make sure you calculate the correct number of slots in the ping
3093 * buffer. Since the ping info is a flattened list of all the NIs,
3094 * we should allocate enough slots to accomodate the number of NIs
3095 * which will be added.
3097 * since ni hasn't been configured yet, use
3098 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3100 net_ni_count = lnet_get_net_ni_count_pre(net);
3102 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3103 net_ni_count + lnet_get_ni_count(),
3111 memcpy(&net->net_tunables,
3112 &tun->lt_cmn, sizeof(net->net_tunables));
3114 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3117 * before starting this network get a count of the current TCP
3118 * networks which require the acceptor thread running. If that
3119 * count is == 0 before we start up this network, then we'd want to
3120 * start up the acceptor thread after starting up this network
3122 num_acceptor_nets = lnet_count_acceptor_nets();
3124 net_id = net->net_id;
3126 rc = lnet_startup_lndnet(net,
3127 (tun) ? &tun->lt_tun : NULL);
3131 lnet_net_lock(LNET_LOCK_EX);
3132 net = lnet_get_net_locked(net_id);
3133 lnet_net_unlock(LNET_LOCK_EX);
3138 * Start the acceptor thread if this is the first network
3139 * being added that requires the thread.
3141 if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3142 rc = lnet_acceptor_start();
3144 /* shutdown the net that we just started */
3145 CERROR("Failed to start up acceptor thread\n");
3146 lnet_shutdown_lndnet(net);
3151 lnet_net_lock(LNET_LOCK_EX);
3152 lnet_peer_net_added(net);
3153 lnet_net_unlock(LNET_LOCK_EX);
3155 lnet_ping_target_update(pbuf, ping_mdh);
3160 lnet_ping_md_unlink(pbuf, &ping_mdh);
3161 lnet_ping_buffer_decref(pbuf);
3165 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3166 struct lnet_ioctl_config_lnd_tunables *tun)
3168 struct lnet_net *net;
3171 struct list_head net_head;
3173 INIT_LIST_HEAD(&net_head);
3175 rc = lnet_parse_ip2nets(&nets, ip2nets);
3179 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3183 mutex_lock(&the_lnet.ln_api_mutex);
3184 while (!list_empty(&net_head)) {
3185 net = list_entry(net_head.next, struct lnet_net, net_list);
3186 list_del_init(&net->net_list);
3187 rc = lnet_add_net_common(net, tun);
3193 mutex_unlock(&the_lnet.ln_api_mutex);
3195 while (!list_empty(&net_head)) {
3196 net = list_entry(net_head.next, struct lnet_net, net_list);
3197 list_del_init(&net->net_list);
3203 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3205 struct lnet_net *net;
3207 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3209 __u32 net_id, lnd_type;
3211 /* get the tunables if they are available */
3212 if (conf->lic_cfg_hdr.ioc_len >=
3213 sizeof(*conf) + sizeof(*tun))
3214 tun = (struct lnet_ioctl_config_lnd_tunables *)
3217 /* handle legacy ip2nets from DLC */
3218 if (conf->lic_legacy_ip2nets[0] != '\0')
3219 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3222 net_id = LNET_NIDNET(conf->lic_nid);
3223 lnd_type = LNET_NETTYP(net_id);
3225 if (!libcfs_isknown_lnd(lnd_type)) {
3226 CERROR("No valid net and lnd information provided\n");
3230 net = lnet_net_alloc(net_id, NULL);
3234 for (i = 0; i < conf->lic_ncpts; i++) {
3235 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3239 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3240 conf->lic_ni_intf[0]);
3244 mutex_lock(&the_lnet.ln_api_mutex);
3246 rc = lnet_add_net_common(net, tun);
3248 mutex_unlock(&the_lnet.ln_api_mutex);
3253 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3255 struct lnet_net *net;
3257 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3258 struct lnet_ping_buffer *pbuf;
3259 struct lnet_handle_md ping_mdh;
3264 /* don't allow userspace to shutdown the LOLND */
3265 if (LNET_NETTYP(net_id) == LOLND)
3268 mutex_lock(&the_lnet.ln_api_mutex);
3272 net = lnet_get_net_locked(net_id);
3274 CERROR("net %s not found\n",
3275 libcfs_net2str(net_id));
3280 addr = LNET_NIDADDR(conf->lic_nid);
3282 /* remove the entire net */
3283 net_count = lnet_get_net_ni_count_locked(net);
3287 /* create and link a new ping info, before removing the old one */
3288 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3289 lnet_get_ni_count() - net_count,
3292 goto unlock_api_mutex;
3294 lnet_shutdown_lndnet(net);
3296 if (lnet_count_acceptor_nets() == 0)
3297 lnet_acceptor_stop();
3299 lnet_ping_target_update(pbuf, ping_mdh);
3301 goto unlock_api_mutex;
3304 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3306 CERROR("nid %s not found\n",
3307 libcfs_nid2str(conf->lic_nid));
3312 net_count = lnet_get_net_ni_count_locked(net);
3316 /* create and link a new ping info, before removing the old one */
3317 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3318 lnet_get_ni_count() - 1, false);
3320 goto unlock_api_mutex;
3322 lnet_shutdown_lndni(ni);
3324 if (lnet_count_acceptor_nets() == 0)
3325 lnet_acceptor_stop();
3327 lnet_ping_target_update(pbuf, ping_mdh);
3329 /* check if the net is empty and remove it if it is */
3331 lnet_shutdown_lndnet(net);
3333 goto unlock_api_mutex;
3338 mutex_unlock(&the_lnet.ln_api_mutex);
3344 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3345 * They are only expected to be called for unique networks.
3346 * That can be as a result of older DLC library
3347 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3350 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3352 struct lnet_net *net;
3353 struct list_head net_head;
3355 struct lnet_ioctl_config_lnd_tunables tun;
3356 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3358 INIT_LIST_HEAD(&net_head);
3360 /* Create a net/ni structures for the network string */
3361 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3363 return rc == 0 ? -EINVAL : rc;
3365 mutex_lock(&the_lnet.ln_api_mutex);
3368 rc = -EINVAL; /* only add one network per call */
3369 goto out_unlock_clean;
3372 net = list_entry(net_head.next, struct lnet_net, net_list);
3373 list_del_init(&net->net_list);
3375 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3377 memset(&tun, 0, sizeof(tun));
3379 tun.lt_cmn.lct_peer_timeout =
3380 conf->cfg_config_u.cfg_net.net_peer_timeout;
3381 tun.lt_cmn.lct_peer_tx_credits =
3382 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3383 tun.lt_cmn.lct_peer_rtr_credits =
3384 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3385 tun.lt_cmn.lct_max_tx_credits =
3386 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3388 rc = lnet_add_net_common(net, &tun);
3391 mutex_unlock(&the_lnet.ln_api_mutex);
3392 while (!list_empty(&net_head)) {
3393 /* net_head list is empty in success case */
3394 net = list_entry(net_head.next, struct lnet_net, net_list);
3395 list_del_init(&net->net_list);
3402 lnet_dyn_del_net(__u32 net_id)
3404 struct lnet_net *net;
3405 struct lnet_ping_buffer *pbuf;
3406 struct lnet_handle_md ping_mdh;
3410 /* don't allow userspace to shutdown the LOLND */
3411 if (LNET_NETTYP(net_id) == LOLND)
3414 mutex_lock(&the_lnet.ln_api_mutex);
3418 net = lnet_get_net_locked(net_id);
3425 net_ni_count = lnet_get_net_ni_count_locked(net);
3429 /* create and link a new ping info, before removing the old one */
3430 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3431 lnet_get_ni_count() - net_ni_count, false);
3435 lnet_shutdown_lndnet(net);
3437 if (lnet_count_acceptor_nets() == 0)
3438 lnet_acceptor_stop();
3440 lnet_ping_target_update(pbuf, ping_mdh);
3443 mutex_unlock(&the_lnet.ln_api_mutex);
3448 void lnet_incr_dlc_seq(void)
3450 atomic_inc(&lnet_dlc_seq_no);
3453 __u32 lnet_get_dlc_seq_locked(void)
3455 return atomic_read(&lnet_dlc_seq_no);
3459 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3461 struct lnet_net *net;
3464 lnet_net_lock(LNET_LOCK_EX);
3465 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3466 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3467 if (ni->ni_nid == nid || all) {
3468 atomic_set(&ni->ni_healthv, value);
3469 if (list_empty(&ni->ni_recovery) &&
3470 value < LNET_MAX_HEALTH_VALUE) {
3471 CERROR("manually adding local NI %s to recovery\n",
3472 libcfs_nid2str(ni->ni_nid));
3473 list_add_tail(&ni->ni_recovery,
3474 &the_lnet.ln_mt_localNIRecovq);
3475 lnet_ni_addref_locked(ni, 0);
3478 lnet_net_unlock(LNET_LOCK_EX);
3484 lnet_net_unlock(LNET_LOCK_EX);
3488 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3492 lnet_nid_t nid = stats->hlni_nid;
3494 cpt = lnet_net_lock_current();
3495 ni = lnet_nid2ni_locked(nid, cpt);
3502 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3503 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3504 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3505 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3506 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3507 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3508 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3511 lnet_net_unlock(cpt);
3517 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3522 lnet_net_lock(LNET_LOCK_EX);
3523 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3524 list->rlst_nid_array[i] = ni->ni_nid;
3526 if (i >= LNET_MAX_SHOW_NUM_NID)
3529 lnet_net_unlock(LNET_LOCK_EX);
3530 list->rlst_num_nids = i;
3536 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3538 struct lnet_peer_ni *lpni;
3541 lnet_net_lock(LNET_LOCK_EX);
3542 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3543 list->rlst_nid_array[i] = lpni->lpni_nid;
3545 if (i >= LNET_MAX_SHOW_NUM_NID)
3548 lnet_net_unlock(LNET_LOCK_EX);
3549 list->rlst_num_nids = i;
3555 * LNet ioctl handler.
3559 LNetCtl(unsigned int cmd, void *arg)
3561 struct libcfs_ioctl_data *data = arg;
3562 struct lnet_ioctl_config_data *config;
3563 struct lnet_process_id id = {0};
3567 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3568 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3571 case IOC_LIBCFS_GET_NI:
3572 rc = LNetGetId(data->ioc_count, &id);
3573 data->ioc_nid = id.nid;
3576 case IOC_LIBCFS_FAIL_NID:
3577 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3579 case IOC_LIBCFS_ADD_ROUTE: {
3580 /* default router sensitivity to 1 */
3581 unsigned int sensitivity = 1;
3584 if (config->cfg_hdr.ioc_len < sizeof(*config))
3587 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3589 config->cfg_config_u.cfg_route.rtr_sensitivity;
3592 mutex_lock(&the_lnet.ln_api_mutex);
3593 rc = lnet_add_route(config->cfg_net,
3594 config->cfg_config_u.cfg_route.rtr_hop,
3596 config->cfg_config_u.cfg_route.
3597 rtr_priority, sensitivity);
3598 mutex_unlock(&the_lnet.ln_api_mutex);
3602 case IOC_LIBCFS_DEL_ROUTE:
3605 if (config->cfg_hdr.ioc_len < sizeof(*config))
3608 mutex_lock(&the_lnet.ln_api_mutex);
3609 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3610 mutex_unlock(&the_lnet.ln_api_mutex);
3613 case IOC_LIBCFS_GET_ROUTE:
3616 if (config->cfg_hdr.ioc_len < sizeof(*config))
3619 mutex_lock(&the_lnet.ln_api_mutex);
3620 rc = lnet_get_route(config->cfg_count,
3622 &config->cfg_config_u.cfg_route.rtr_hop,
3624 &config->cfg_config_u.cfg_route.rtr_flags,
3625 &config->cfg_config_u.cfg_route.
3627 &config->cfg_config_u.cfg_route.
3629 mutex_unlock(&the_lnet.ln_api_mutex);
3632 case IOC_LIBCFS_GET_LOCAL_NI: {
3633 struct lnet_ioctl_config_ni *cfg_ni;
3634 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3635 struct lnet_ioctl_element_stats *stats;
3640 /* get the tunables if they are available */
3641 if (cfg_ni->lic_cfg_hdr.ioc_len <
3642 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3645 stats = (struct lnet_ioctl_element_stats *)
3647 tun = (struct lnet_ioctl_config_lnd_tunables *)
3648 (cfg_ni->lic_bulk + sizeof(*stats));
3650 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3653 mutex_lock(&the_lnet.ln_api_mutex);
3654 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3655 mutex_unlock(&the_lnet.ln_api_mutex);
3659 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3660 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3662 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3665 mutex_lock(&the_lnet.ln_api_mutex);
3666 rc = lnet_get_ni_stats(msg_stats);
3667 mutex_unlock(&the_lnet.ln_api_mutex);
3672 case IOC_LIBCFS_GET_NET: {
3673 size_t total = sizeof(*config) +
3674 sizeof(struct lnet_ioctl_net_config);
3677 if (config->cfg_hdr.ioc_len < total)
3680 mutex_lock(&the_lnet.ln_api_mutex);
3681 rc = lnet_get_net_config(config);
3682 mutex_unlock(&the_lnet.ln_api_mutex);
3686 case IOC_LIBCFS_GET_LNET_STATS:
3688 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3690 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3693 mutex_lock(&the_lnet.ln_api_mutex);
3694 lnet_counters_get(&lnet_stats->st_cntrs);
3695 mutex_unlock(&the_lnet.ln_api_mutex);
3699 case IOC_LIBCFS_CONFIG_RTR:
3702 if (config->cfg_hdr.ioc_len < sizeof(*config))
3705 mutex_lock(&the_lnet.ln_api_mutex);
3706 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3707 rc = lnet_rtrpools_enable();
3708 mutex_unlock(&the_lnet.ln_api_mutex);
3711 lnet_rtrpools_disable();
3712 mutex_unlock(&the_lnet.ln_api_mutex);
3715 case IOC_LIBCFS_ADD_BUF:
3718 if (config->cfg_hdr.ioc_len < sizeof(*config))
3721 mutex_lock(&the_lnet.ln_api_mutex);
3722 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3724 config->cfg_config_u.cfg_buffers.
3726 config->cfg_config_u.cfg_buffers.
3728 mutex_unlock(&the_lnet.ln_api_mutex);
3731 case IOC_LIBCFS_SET_NUMA_RANGE: {
3732 struct lnet_ioctl_set_value *numa;
3734 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3736 lnet_net_lock(LNET_LOCK_EX);
3737 lnet_numa_range = numa->sv_value;
3738 lnet_net_unlock(LNET_LOCK_EX);
3742 case IOC_LIBCFS_GET_NUMA_RANGE: {
3743 struct lnet_ioctl_set_value *numa;
3745 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3747 numa->sv_value = lnet_numa_range;
3751 case IOC_LIBCFS_GET_BUF: {
3752 struct lnet_ioctl_pool_cfg *pool_cfg;
3753 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3757 if (config->cfg_hdr.ioc_len < total)
3760 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3762 mutex_lock(&the_lnet.ln_api_mutex);
3763 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3764 mutex_unlock(&the_lnet.ln_api_mutex);
3768 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3769 struct lnet_ioctl_local_ni_hstats *stats = arg;
3771 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3774 mutex_lock(&the_lnet.ln_api_mutex);
3775 rc = lnet_get_local_ni_hstats(stats);
3776 mutex_unlock(&the_lnet.ln_api_mutex);
3781 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3782 struct lnet_ioctl_recovery_list *list = arg;
3783 if (list->rlst_hdr.ioc_len < sizeof(*list))
3786 mutex_lock(&the_lnet.ln_api_mutex);
3787 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3788 rc = lnet_get_local_ni_recovery_list(list);
3790 rc = lnet_get_peer_ni_recovery_list(list);
3791 mutex_unlock(&the_lnet.ln_api_mutex);
3795 case IOC_LIBCFS_ADD_PEER_NI: {
3796 struct lnet_ioctl_peer_cfg *cfg = arg;
3798 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3801 mutex_lock(&the_lnet.ln_api_mutex);
3802 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3805 mutex_unlock(&the_lnet.ln_api_mutex);
3809 case IOC_LIBCFS_DEL_PEER_NI: {
3810 struct lnet_ioctl_peer_cfg *cfg = arg;
3812 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3815 mutex_lock(&the_lnet.ln_api_mutex);
3816 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3817 cfg->prcfg_cfg_nid);
3818 mutex_unlock(&the_lnet.ln_api_mutex);
3822 case IOC_LIBCFS_GET_PEER_INFO: {
3823 struct lnet_ioctl_peer *peer_info = arg;
3825 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3828 mutex_lock(&the_lnet.ln_api_mutex);
3829 rc = lnet_get_peer_ni_info(
3830 peer_info->pr_count,
3832 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3833 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3834 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3835 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3836 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3837 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3838 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3839 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3840 mutex_unlock(&the_lnet.ln_api_mutex);
3844 case IOC_LIBCFS_GET_PEER_NI: {
3845 struct lnet_ioctl_peer_cfg *cfg = arg;
3847 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3850 mutex_lock(&the_lnet.ln_api_mutex);
3851 rc = lnet_get_peer_info(cfg,
3852 (void __user *)cfg->prcfg_bulk);
3853 mutex_unlock(&the_lnet.ln_api_mutex);
3857 case IOC_LIBCFS_GET_PEER_LIST: {
3858 struct lnet_ioctl_peer_cfg *cfg = arg;
3860 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3863 mutex_lock(&the_lnet.ln_api_mutex);
3864 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3865 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3866 mutex_unlock(&the_lnet.ln_api_mutex);
3870 case IOC_LIBCFS_SET_HEALHV: {
3871 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3873 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3875 if (cfg->rh_value < 0 ||
3876 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3877 value = LNET_MAX_HEALTH_VALUE;
3879 value = cfg->rh_value;
3880 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3881 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3882 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3883 mutex_lock(&the_lnet.ln_api_mutex);
3884 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3885 lnet_ni_set_healthv(cfg->rh_nid, value,
3888 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3890 mutex_unlock(&the_lnet.ln_api_mutex);
3894 case IOC_LIBCFS_NOTIFY_ROUTER: {
3895 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3897 /* The deadline passed in by the user should be some time in
3898 * seconds in the future since the UNIX epoch. We have to map
3899 * that deadline to the wall clock.
3901 deadline += ktime_get_seconds();
3902 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3906 case IOC_LIBCFS_LNET_DIST:
3907 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3908 if (rc < 0 && rc != -EHOSTUNREACH)
3911 data->ioc_u32[0] = rc;
3914 case IOC_LIBCFS_TESTPROTOCOMPAT:
3915 lnet_net_lock(LNET_LOCK_EX);
3916 the_lnet.ln_testprotocompat = data->ioc_flags;
3917 lnet_net_unlock(LNET_LOCK_EX);
3920 case IOC_LIBCFS_LNET_FAULT:
3921 return lnet_fault_ctl(data->ioc_flags, data);
3923 case IOC_LIBCFS_PING: {
3924 signed long timeout;
3926 id.nid = data->ioc_nid;
3927 id.pid = data->ioc_u32[0];
3929 /* If timeout is negative then set default of 3 minutes */
3930 if (((s32)data->ioc_u32[1] <= 0) ||
3931 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3932 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3934 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3936 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3937 data->ioc_plen1 / sizeof(struct lnet_process_id));
3942 data->ioc_count = rc;
3946 case IOC_LIBCFS_PING_PEER: {
3947 struct lnet_ioctl_ping_data *ping = arg;
3948 struct lnet_peer *lp;
3949 signed long timeout;
3951 /* If timeout is negative then set default of 3 minutes */
3952 if (((s32)ping->op_param) <= 0 ||
3953 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3954 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3956 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3958 rc = lnet_ping(ping->ping_id, timeout,
3964 mutex_lock(&the_lnet.ln_api_mutex);
3965 lp = lnet_find_peer(ping->ping_id.nid);
3967 ping->ping_id.nid = lp->lp_primary_nid;
3968 ping->mr_info = lnet_peer_is_multi_rail(lp);
3969 lnet_peer_decref_locked(lp);
3971 mutex_unlock(&the_lnet.ln_api_mutex);
3973 ping->ping_count = rc;
3977 case IOC_LIBCFS_DISCOVER: {
3978 struct lnet_ioctl_ping_data *discover = arg;
3979 struct lnet_peer *lp;
3981 rc = lnet_discover(discover->ping_id, discover->op_param,
3983 discover->ping_count);
3987 mutex_lock(&the_lnet.ln_api_mutex);
3988 lp = lnet_find_peer(discover->ping_id.nid);
3990 discover->ping_id.nid = lp->lp_primary_nid;
3991 discover->mr_info = lnet_peer_is_multi_rail(lp);
3992 lnet_peer_decref_locked(lp);
3994 mutex_unlock(&the_lnet.ln_api_mutex);
3996 discover->ping_count = rc;
4001 ni = lnet_net2ni_addref(data->ioc_net);
4005 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4008 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4015 EXPORT_SYMBOL(LNetCtl);
4017 void LNetDebugPeer(struct lnet_process_id id)
4019 lnet_debug_peer(id.nid);
4021 EXPORT_SYMBOL(LNetDebugPeer);
4024 * Determine if the specified peer \a nid is on the local node.
4026 * \param nid peer nid to check
4028 * \retval true If peer NID is on the local node.
4029 * \retval false If peer NID is not on the local node.
4031 bool LNetIsPeerLocal(lnet_nid_t nid)
4033 struct lnet_net *net;
4037 cpt = lnet_net_lock_current();
4038 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4039 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4040 if (ni->ni_nid == nid) {
4041 lnet_net_unlock(cpt);
4046 lnet_net_unlock(cpt);
4050 EXPORT_SYMBOL(LNetIsPeerLocal);
4053 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4054 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4056 * \param index Index of the interface to look up.
4057 * \param id On successful return, this location will hold the
4058 * struct lnet_process_id ID of the interface.
4060 * \retval 0 If an interface exists at \a index.
4061 * \retval -ENOENT If no interface has been found.
4064 LNetGetId(unsigned int index, struct lnet_process_id *id)
4067 struct lnet_net *net;
4071 LASSERT(the_lnet.ln_refcount > 0);
4073 cpt = lnet_net_lock_current();
4075 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4076 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4080 id->nid = ni->ni_nid;
4081 id->pid = the_lnet.ln_pid;
4087 lnet_net_unlock(cpt);
4090 EXPORT_SYMBOL(LNetGetId);
4092 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4093 struct lnet_process_id __user *ids, int n_ids)
4095 struct lnet_handle_eq eqh;
4096 struct lnet_handle_md mdh;
4097 struct lnet_event event;
4098 struct lnet_md md = { NULL };
4102 const signed long a_long_time = cfs_time_seconds(60);
4103 struct lnet_ping_buffer *pbuf;
4104 struct lnet_process_id tmpid;
4111 /* n_ids limit is arbitrary */
4112 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4116 * if the user buffer has more space than the lnet_interfaces_max
4117 * then only fill it up to lnet_interfaces_max
4119 if (n_ids > lnet_interfaces_max)
4120 n_ids = lnet_interfaces_max;
4122 if (id.pid == LNET_PID_ANY)
4123 id.pid = LNET_PID_LUSTRE;
4125 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4129 /* NB 2 events max (including any unlink event) */
4130 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
4132 CERROR("Can't allocate EQ: %d\n", rc);
4133 goto fail_ping_buffer_decref;
4136 /* initialize md content */
4137 md.start = &pbuf->pb_info;
4138 md.length = LNET_PING_INFO_SIZE(n_ids);
4139 md.threshold = 2; /* GET/REPLY */
4141 md.options = LNET_MD_TRUNCATE;
4145 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4147 CERROR("Can't bind MD: %d\n", rc);
4151 rc = LNetGet(LNET_NID_ANY, mdh, id,
4152 LNET_RESERVED_PORTAL,
4153 LNET_PROTO_PING_MATCHBITS, 0, false);
4156 /* Don't CERROR; this could be deliberate! */
4157 rc2 = LNetMDUnlink(mdh);
4160 /* NB must wait for the UNLINK event below... */
4162 timeout = a_long_time;
4166 /* MUST block for unlink to complete */
4168 blocked = cfs_block_allsigs();
4170 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
4173 cfs_restore_sigs(blocked);
4175 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4176 (rc2 <= 0) ? -1 : event.type,
4177 (rc2 <= 0) ? -1 : event.status,
4178 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4180 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
4182 if (rc2 <= 0 || event.status != 0) {
4183 /* timeout or error */
4184 if (!replied && rc == 0)
4185 rc = (rc2 < 0) ? rc2 :
4186 (rc2 == 0) ? -ETIMEDOUT :
4190 /* Ensure completion in finite time... */
4192 /* No assertion (racing with network) */
4194 timeout = a_long_time;
4195 } else if (rc2 == 0) {
4196 /* timed out waiting for unlink */
4197 CWARN("ping %s: late network completion\n",
4200 } else if (event.type == LNET_EVENT_REPLY) {
4204 } while (rc2 <= 0 || !event.unlinked);
4208 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4215 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4217 rc = -EPROTO; /* if I can't parse... */
4220 CERROR("%s: ping info too short %d\n",
4221 libcfs_id2str(id), nob);
4225 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4226 lnet_swap_pinginfo(pbuf);
4227 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4228 CERROR("%s: Unexpected magic %08x\n",
4229 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4233 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4234 CERROR("%s: ping w/o NI status: 0x%x\n",
4235 libcfs_id2str(id), pbuf->pb_info.pi_features);
4239 if (nob < LNET_PING_INFO_SIZE(0)) {
4240 CERROR("%s: Short reply %d(%d min)\n",
4242 nob, (int)LNET_PING_INFO_SIZE(0));
4246 if (pbuf->pb_info.pi_nnis < n_ids)
4247 n_ids = pbuf->pb_info.pi_nnis;
4249 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4250 CERROR("%s: Short reply %d(%d expected)\n",
4252 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4256 rc = -EFAULT; /* if I segv in copy_to_user()... */
4258 memset(&tmpid, 0, sizeof(tmpid));
4259 for (i = 0; i < n_ids; i++) {
4260 tmpid.pid = pbuf->pb_info.pi_pid;
4261 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4262 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4265 rc = pbuf->pb_info.pi_nnis;
4268 rc2 = LNetEQFree(eqh);
4270 CERROR("rc2 %d\n", rc2);
4273 fail_ping_buffer_decref:
4274 lnet_ping_buffer_decref(pbuf);
4279 lnet_discover(struct lnet_process_id id, __u32 force,
4280 struct lnet_process_id __user *ids, int n_ids)
4282 struct lnet_peer_ni *lpni;
4283 struct lnet_peer_ni *p;
4284 struct lnet_peer *lp;
4285 struct lnet_process_id *buf;
4289 int max_intf = lnet_interfaces_max;
4293 id.nid == LNET_NID_ANY)
4296 if (id.pid == LNET_PID_ANY)
4297 id.pid = LNET_PID_LUSTRE;
4300 * if the user buffer has more space than the max_intf
4301 * then only fill it up to max_intf
4303 if (n_ids > max_intf)
4306 buf_size = n_ids * sizeof(*buf);
4308 LIBCFS_ALLOC(buf, buf_size);
4312 cpt = lnet_net_lock_current();
4313 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4320 * Clearing the NIDS_UPTODATE flag ensures the peer will
4321 * be discovered, provided discovery has not been disabled.
4323 lp = lpni->lpni_peer_net->lpn_peer;
4324 spin_lock(&lp->lp_lock);
4325 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4326 /* If the force flag is set, force a PING and PUSH as well. */
4328 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4329 spin_unlock(&lp->lp_lock);
4330 rc = lnet_discover_peer_locked(lpni, cpt, true);
4334 /* Peer may have changed. */
4335 lp = lpni->lpni_peer_net->lpn_peer;
4336 if (lp->lp_nnis < n_ids)
4337 n_ids = lp->lp_nnis;
4341 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4342 buf[i].pid = id.pid;
4343 buf[i].nid = p->lpni_nid;
4348 lnet_net_unlock(cpt);
4351 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4357 lnet_peer_ni_decref_locked(lpni);
4359 lnet_net_unlock(cpt);
4361 LIBCFS_FREE(buf, buf_size);
4367 * Retrieve peer discovery status.
4369 * \retval 1 if lnet_peer_discovery_disabled is 0
4370 * \retval 0 if lnet_peer_discovery_disabled is 1
4373 LNetGetPeerDiscoveryStatus(void)
4375 return !lnet_peer_discovery_disabled;
4377 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);