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 0, which means health
84 * checking is turned off by default.
86 unsigned int lnet_health_sensitivity = 0;
87 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
88 static struct kernel_param_ops param_ops_health_sensitivity = {
89 .set = sensitivity_set,
92 #define param_check_health_sensitivity(name, p) \
93 __param_check(name, p, int)
94 #ifdef HAVE_KERNEL_PARAM_OPS
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");
103 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
104 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
106 static struct kernel_param_ops param_ops_interfaces_max = {
108 .get = param_get_int,
111 #define param_check_interfaces_max(name, p) \
112 __param_check(name, p, int)
114 #ifdef HAVE_KERNEL_PARAM_OPS
115 module_param(lnet_interfaces_max, interfaces_max, 0644);
117 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
118 ¶m_ops_interfaces_max, 0644);
120 MODULE_PARM_DESC(lnet_interfaces_max,
121 "Maximum number of interfaces in a node.");
123 unsigned lnet_peer_discovery_disabled = 0;
124 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
126 static struct kernel_param_ops param_ops_discovery_disabled = {
127 .set = discovery_set,
128 .get = param_get_int,
131 #define param_check_discovery_disabled(name, p) \
132 __param_check(name, p, int)
133 #ifdef HAVE_KERNEL_PARAM_OPS
134 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
136 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
137 ¶m_ops_discovery_disabled, 0644);
139 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
140 "Set to 1 to disable peer discovery on this node.");
142 unsigned lnet_transaction_timeout = 5;
143 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
144 static struct kernel_param_ops param_ops_transaction_timeout = {
145 .set = transaction_to_set,
146 .get = param_get_int,
149 #define param_check_transaction_timeout(name, p) \
150 __param_check(name, p, int)
151 #ifdef HAVE_KERNEL_PARAM_OPS
152 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
154 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
155 &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
157 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
158 "Set to 1 to disable peer discovery on this node.");
160 unsigned lnet_retry_count = 0;
161 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
162 static struct kernel_param_ops param_ops_retry_count = {
163 .set = retry_count_set,
164 .get = param_get_int,
167 #define param_check_retry_count(name, p) \
168 __param_check(name, p, int)
169 #ifdef HAVE_KERNEL_PARAM_OPS
170 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
172 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
173 &lnet_retry_count, S_IRUGO|S_IWUSR);
175 MODULE_PARM_DESC(lnet_retry_count,
176 "Maximum number of times to retry transmitting a message");
179 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
182 * This sequence number keeps track of how many times DLC was used to
183 * update the local NIs. It is incremented when a NI is added or
184 * removed and checked when sending a message to determine if there is
185 * a need to re-run the selection algorithm. See lnet_select_pathway()
186 * for more details on its usage.
188 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
190 static int lnet_ping(struct lnet_process_id id, signed long timeout,
191 struct lnet_process_id __user *ids, int n_ids);
193 static int lnet_discover(struct lnet_process_id id, __u32 force,
194 struct lnet_process_id __user *ids, int n_ids);
197 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
200 unsigned *sensitivity = (unsigned *)kp->arg;
203 rc = kstrtoul(val, 0, &value);
205 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
210 * The purpose of locking the api_mutex here is to ensure that
211 * the correct value ends up stored properly.
213 mutex_lock(&the_lnet.ln_api_mutex);
215 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
216 mutex_unlock(&the_lnet.ln_api_mutex);
220 if (value == *sensitivity) {
221 mutex_unlock(&the_lnet.ln_api_mutex);
225 *sensitivity = value;
227 mutex_unlock(&the_lnet.ln_api_mutex);
233 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
236 unsigned *discovery = (unsigned *)kp->arg;
238 struct lnet_ping_buffer *pbuf;
240 rc = kstrtoul(val, 0, &value);
242 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
246 value = (value) ? 1 : 0;
249 * The purpose of locking the api_mutex here is to ensure that
250 * the correct value ends up stored properly.
252 mutex_lock(&the_lnet.ln_api_mutex);
254 if (value == *discovery) {
255 mutex_unlock(&the_lnet.ln_api_mutex);
261 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
262 mutex_unlock(&the_lnet.ln_api_mutex);
266 /* tell peers that discovery setting has changed */
267 lnet_net_lock(LNET_LOCK_EX);
268 pbuf = the_lnet.ln_ping_target;
270 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
272 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
273 lnet_net_unlock(LNET_LOCK_EX);
275 lnet_push_update_to_peers(1);
277 mutex_unlock(&the_lnet.ln_api_mutex);
283 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
286 unsigned *transaction_to = (unsigned *)kp->arg;
289 rc = kstrtoul(val, 0, &value);
291 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
296 * The purpose of locking the api_mutex here is to ensure that
297 * the correct value ends up stored properly.
299 mutex_lock(&the_lnet.ln_api_mutex);
301 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
302 mutex_unlock(&the_lnet.ln_api_mutex);
306 if (value < lnet_retry_count || value == 0) {
307 mutex_unlock(&the_lnet.ln_api_mutex);
308 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
309 "Has to be greater than lnet_retry_count (%u)\n",
310 value, lnet_retry_count);
314 if (value == *transaction_to) {
315 mutex_unlock(&the_lnet.ln_api_mutex);
319 *transaction_to = value;
320 if (lnet_retry_count == 0)
321 lnet_lnd_timeout = value;
323 lnet_lnd_timeout = value / lnet_retry_count;
325 mutex_unlock(&the_lnet.ln_api_mutex);
331 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
334 unsigned *retry_count = (unsigned *)kp->arg;
337 rc = kstrtoul(val, 0, &value);
339 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
344 * The purpose of locking the api_mutex here is to ensure that
345 * the correct value ends up stored properly.
347 mutex_lock(&the_lnet.ln_api_mutex);
349 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
350 mutex_unlock(&the_lnet.ln_api_mutex);
354 if (value > lnet_transaction_timeout) {
355 mutex_unlock(&the_lnet.ln_api_mutex);
356 CERROR("Invalid value for lnet_retry_count (%lu). "
357 "Has to be smaller than lnet_transaction_timeout (%u)\n",
358 value, lnet_transaction_timeout);
362 if (value == *retry_count) {
363 mutex_unlock(&the_lnet.ln_api_mutex);
367 *retry_count = value;
370 lnet_lnd_timeout = lnet_transaction_timeout;
372 lnet_lnd_timeout = lnet_transaction_timeout / value;
374 mutex_unlock(&the_lnet.ln_api_mutex);
380 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
384 rc = kstrtoint(val, 0, &value);
386 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
390 if (value < LNET_INTERFACES_MIN) {
391 CWARN("max interfaces provided are too small, setting to %d\n",
392 LNET_INTERFACES_MAX_DEFAULT);
393 value = LNET_INTERFACES_MAX_DEFAULT;
396 *(int *)kp->arg = value;
402 lnet_get_routes(void)
408 lnet_get_networks(void)
413 if (*networks != 0 && *ip2nets != 0) {
414 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
415 "'ip2nets' but not both at once\n");
420 rc = lnet_parse_ip2nets(&nets, ip2nets);
421 return (rc == 0) ? nets : NULL;
431 lnet_init_locks(void)
433 spin_lock_init(&the_lnet.ln_eq_wait_lock);
434 spin_lock_init(&the_lnet.ln_msg_resend_lock);
435 init_waitqueue_head(&the_lnet.ln_eq_waitq);
436 init_waitqueue_head(&the_lnet.ln_mt_waitq);
437 mutex_init(&the_lnet.ln_lnd_mutex);
441 lnet_fini_locks(void)
445 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
446 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
450 lnet_descriptor_setup(void)
452 /* create specific kmem_cache for MEs and small MDs (i.e., originally
453 * allocated in <size-xxx> kmem_cache).
455 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
457 if (!lnet_mes_cachep)
460 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
461 LNET_SMALL_MD_SIZE, 0, 0,
463 if (!lnet_small_mds_cachep)
470 lnet_descriptor_cleanup(void)
473 if (lnet_small_mds_cachep) {
474 kmem_cache_destroy(lnet_small_mds_cachep);
475 lnet_small_mds_cachep = NULL;
478 if (lnet_mes_cachep) {
479 kmem_cache_destroy(lnet_mes_cachep);
480 lnet_mes_cachep = NULL;
485 lnet_create_remote_nets_table(void)
488 struct list_head *hash;
490 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
491 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
492 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
494 CERROR("Failed to create remote nets hash table\n");
498 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
499 INIT_LIST_HEAD(&hash[i]);
500 the_lnet.ln_remote_nets_hash = hash;
505 lnet_destroy_remote_nets_table(void)
509 if (the_lnet.ln_remote_nets_hash == NULL)
512 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
513 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
515 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
516 LNET_REMOTE_NETS_HASH_SIZE *
517 sizeof(the_lnet.ln_remote_nets_hash[0]));
518 the_lnet.ln_remote_nets_hash = NULL;
522 lnet_destroy_locks(void)
524 if (the_lnet.ln_res_lock != NULL) {
525 cfs_percpt_lock_free(the_lnet.ln_res_lock);
526 the_lnet.ln_res_lock = NULL;
529 if (the_lnet.ln_net_lock != NULL) {
530 cfs_percpt_lock_free(the_lnet.ln_net_lock);
531 the_lnet.ln_net_lock = NULL;
538 lnet_create_locks(void)
542 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
543 if (the_lnet.ln_res_lock == NULL)
546 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
547 if (the_lnet.ln_net_lock == NULL)
553 lnet_destroy_locks();
557 static void lnet_assert_wire_constants(void)
559 /* Wire protocol assertions generated by 'wirecheck'
560 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
561 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
562 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
565 CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
566 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
567 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
568 CLASSERT(LNET_MSG_ACK == 0);
569 CLASSERT(LNET_MSG_PUT == 1);
570 CLASSERT(LNET_MSG_GET == 2);
571 CLASSERT(LNET_MSG_REPLY == 3);
572 CLASSERT(LNET_MSG_HELLO == 4);
574 /* Checks for struct lnet_handle_wire */
575 CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
576 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
577 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
578 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
579 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
581 /* Checks for struct struct lnet_magicversion */
582 CLASSERT((int)sizeof(struct lnet_magicversion) == 8);
583 CLASSERT((int)offsetof(struct lnet_magicversion, magic) == 0);
584 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->magic) == 4);
585 CLASSERT((int)offsetof(struct lnet_magicversion, version_major) == 4);
586 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_major) == 2);
587 CLASSERT((int)offsetof(struct lnet_magicversion, version_minor) == 6);
588 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_minor) == 2);
590 /* Checks for struct struct lnet_hdr */
591 CLASSERT((int)sizeof(struct lnet_hdr) == 72);
592 CLASSERT((int)offsetof(struct lnet_hdr, dest_nid) == 0);
593 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_nid) == 8);
594 CLASSERT((int)offsetof(struct lnet_hdr, src_nid) == 8);
595 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_nid) == 8);
596 CLASSERT((int)offsetof(struct lnet_hdr, dest_pid) == 16);
597 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_pid) == 4);
598 CLASSERT((int)offsetof(struct lnet_hdr, src_pid) == 20);
599 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_pid) == 4);
600 CLASSERT((int)offsetof(struct lnet_hdr, type) == 24);
601 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->type) == 4);
602 CLASSERT((int)offsetof(struct lnet_hdr, payload_length) == 28);
603 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->payload_length) == 4);
604 CLASSERT((int)offsetof(struct lnet_hdr, msg) == 32);
605 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg) == 40);
608 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) == 32);
609 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) == 16);
610 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.match_bits) == 48);
611 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) == 8);
612 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.mlength) == 56);
613 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) == 4);
616 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) == 32);
617 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) == 16);
618 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.match_bits) == 48);
619 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) == 8);
620 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.hdr_data) == 56);
621 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) == 8);
622 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ptl_index) == 64);
623 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) == 4);
624 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.offset) == 68);
625 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) == 4);
628 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.return_wmd) == 32);
629 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) == 16);
630 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.match_bits) == 48);
631 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) == 8);
632 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.ptl_index) == 56);
633 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) == 4);
634 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.src_offset) == 60);
635 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) == 4);
636 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.sink_length) == 64);
637 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) == 4);
640 CLASSERT((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) == 32);
641 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) == 16);
644 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.incarnation) == 32);
645 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) == 8);
646 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.type) == 40);
647 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) == 4);
649 /* Checks for struct lnet_ni_status and related constants */
650 CLASSERT(LNET_NI_STATUS_INVALID == 0x00000000);
651 CLASSERT(LNET_NI_STATUS_UP == 0x15aac0de);
652 CLASSERT(LNET_NI_STATUS_DOWN == 0xdeadface);
654 /* Checks for struct lnet_ni_status */
655 CLASSERT((int)sizeof(struct lnet_ni_status) == 16);
656 CLASSERT((int)offsetof(struct lnet_ni_status, ns_nid) == 0);
657 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) == 8);
658 CLASSERT((int)offsetof(struct lnet_ni_status, ns_status) == 8);
659 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_status) == 4);
660 CLASSERT((int)offsetof(struct lnet_ni_status, ns_unused) == 12);
661 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) == 4);
663 /* Checks for struct lnet_ping_info and related constants */
664 CLASSERT(LNET_PROTO_PING_MAGIC == 0x70696E67);
665 CLASSERT(LNET_PING_FEAT_INVAL == 0);
666 CLASSERT(LNET_PING_FEAT_BASE == 1);
667 CLASSERT(LNET_PING_FEAT_NI_STATUS == 2);
668 CLASSERT(LNET_PING_FEAT_RTE_DISABLED == 4);
669 CLASSERT(LNET_PING_FEAT_MULTI_RAIL == 8);
670 CLASSERT(LNET_PING_FEAT_DISCOVERY == 16);
671 CLASSERT(LNET_PING_FEAT_BITS == 31);
673 /* Checks for struct lnet_ping_info */
674 CLASSERT((int)sizeof(struct lnet_ping_info) == 16);
675 CLASSERT((int)offsetof(struct lnet_ping_info, pi_magic) == 0);
676 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) == 4);
677 CLASSERT((int)offsetof(struct lnet_ping_info, pi_features) == 4);
678 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_features) == 4);
679 CLASSERT((int)offsetof(struct lnet_ping_info, pi_pid) == 8);
680 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) == 4);
681 CLASSERT((int)offsetof(struct lnet_ping_info, pi_nnis) == 12);
682 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) == 4);
683 CLASSERT((int)offsetof(struct lnet_ping_info, pi_ni) == 16);
684 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) == 0);
687 static struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
689 struct lnet_lnd *lnd;
690 struct list_head *tmp;
692 /* holding lnd mutex */
693 list_for_each(tmp, &the_lnet.ln_lnds) {
694 lnd = list_entry(tmp, struct lnet_lnd, lnd_list);
696 if (lnd->lnd_type == type)
703 lnet_get_lnd_timeout(void)
705 return lnet_lnd_timeout;
707 EXPORT_SYMBOL(lnet_get_lnd_timeout);
710 lnet_register_lnd(struct lnet_lnd *lnd)
712 mutex_lock(&the_lnet.ln_lnd_mutex);
714 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
715 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
717 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
718 lnd->lnd_refcount = 0;
720 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
722 mutex_unlock(&the_lnet.ln_lnd_mutex);
724 EXPORT_SYMBOL(lnet_register_lnd);
727 lnet_unregister_lnd(struct lnet_lnd *lnd)
729 mutex_lock(&the_lnet.ln_lnd_mutex);
731 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
732 LASSERT(lnd->lnd_refcount == 0);
734 list_del(&lnd->lnd_list);
735 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
737 mutex_unlock(&the_lnet.ln_lnd_mutex);
739 EXPORT_SYMBOL(lnet_unregister_lnd);
742 lnet_counters_get(struct lnet_counters *counters)
744 struct lnet_counters *ctr;
747 memset(counters, 0, sizeof(*counters));
749 lnet_net_lock(LNET_LOCK_EX);
751 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
752 counters->msgs_max += ctr->msgs_max;
753 counters->msgs_alloc += ctr->msgs_alloc;
754 counters->errors += ctr->errors;
755 counters->send_count += ctr->send_count;
756 counters->recv_count += ctr->recv_count;
757 counters->route_count += ctr->route_count;
758 counters->drop_count += ctr->drop_count;
759 counters->send_length += ctr->send_length;
760 counters->recv_length += ctr->recv_length;
761 counters->route_length += ctr->route_length;
762 counters->drop_length += ctr->drop_length;
765 lnet_net_unlock(LNET_LOCK_EX);
767 EXPORT_SYMBOL(lnet_counters_get);
770 lnet_counters_reset(void)
772 struct lnet_counters *counters;
775 lnet_net_lock(LNET_LOCK_EX);
777 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
778 memset(counters, 0, sizeof(struct lnet_counters));
780 lnet_net_unlock(LNET_LOCK_EX);
784 lnet_res_type2str(int type)
789 case LNET_COOKIE_TYPE_MD:
791 case LNET_COOKIE_TYPE_ME:
793 case LNET_COOKIE_TYPE_EQ:
799 lnet_res_container_cleanup(struct lnet_res_container *rec)
803 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
806 while (!list_empty(&rec->rec_active)) {
807 struct list_head *e = rec->rec_active.next;
810 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
811 lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
813 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
814 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
816 } else { /* NB: Active MEs should be attached on portals */
823 /* Found alive MD/ME/EQ, user really should unlink/free
824 * all of them before finalize LNet, but if someone didn't,
825 * we have to recycle garbage for him */
826 CERROR("%d active elements on exit of %s container\n",
827 count, lnet_res_type2str(rec->rec_type));
830 if (rec->rec_lh_hash != NULL) {
831 LIBCFS_FREE(rec->rec_lh_hash,
832 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
833 rec->rec_lh_hash = NULL;
836 rec->rec_type = 0; /* mark it as finalized */
840 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
845 LASSERT(rec->rec_type == 0);
847 rec->rec_type = type;
848 INIT_LIST_HEAD(&rec->rec_active);
850 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
852 /* Arbitrary choice of hash table size */
853 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
854 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
855 if (rec->rec_lh_hash == NULL) {
860 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
861 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
866 CERROR("Failed to setup %s resource container\n",
867 lnet_res_type2str(type));
868 lnet_res_container_cleanup(rec);
873 lnet_res_containers_destroy(struct lnet_res_container **recs)
875 struct lnet_res_container *rec;
878 cfs_percpt_for_each(rec, i, recs)
879 lnet_res_container_cleanup(rec);
881 cfs_percpt_free(recs);
884 static struct lnet_res_container **
885 lnet_res_containers_create(int type)
887 struct lnet_res_container **recs;
888 struct lnet_res_container *rec;
892 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
894 CERROR("Failed to allocate %s resource containers\n",
895 lnet_res_type2str(type));
899 cfs_percpt_for_each(rec, i, recs) {
900 rc = lnet_res_container_setup(rec, i, type);
902 lnet_res_containers_destroy(recs);
910 struct lnet_libhandle *
911 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
913 /* ALWAYS called with lnet_res_lock held */
914 struct list_head *head;
915 struct lnet_libhandle *lh;
918 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
921 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
922 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
924 list_for_each_entry(lh, head, lh_hash_chain) {
925 if (lh->lh_cookie == cookie)
933 lnet_res_lh_initialize(struct lnet_res_container *rec,
934 struct lnet_libhandle *lh)
936 /* ALWAYS called with lnet_res_lock held */
937 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
940 lh->lh_cookie = rec->rec_lh_cookie;
941 rec->rec_lh_cookie += 1 << ibits;
943 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
945 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
948 static int lnet_unprepare(void);
951 lnet_prepare(lnet_pid_t requested_pid)
953 /* Prepare to bring up the network */
954 struct lnet_res_container **recs;
957 if (requested_pid == LNET_PID_ANY) {
958 /* Don't instantiate LNET just for me */
962 LASSERT(the_lnet.ln_refcount == 0);
964 the_lnet.ln_routing = 0;
966 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
967 the_lnet.ln_pid = requested_pid;
969 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
970 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
971 INIT_LIST_HEAD(&the_lnet.ln_nets);
972 INIT_LIST_HEAD(&the_lnet.ln_routers);
973 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
974 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
975 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
976 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
977 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
978 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
979 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
980 init_waitqueue_head(&the_lnet.ln_dc_waitq);
982 rc = lnet_descriptor_setup();
986 rc = lnet_create_remote_nets_table();
991 * NB the interface cookie in wire handles guards against delayed
992 * replies and ACKs appearing valid after reboot.
994 the_lnet.ln_interface_cookie = ktime_get_real_ns();
996 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
997 sizeof(struct lnet_counters));
998 if (the_lnet.ln_counters == NULL) {
999 CERROR("Failed to allocate counters for LNet\n");
1004 rc = lnet_peer_tables_create();
1008 rc = lnet_msg_containers_create();
1012 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1013 LNET_COOKIE_TYPE_EQ);
1017 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
1023 the_lnet.ln_me_containers = recs;
1025 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1031 the_lnet.ln_md_containers = recs;
1033 rc = lnet_portals_create();
1035 CERROR("Failed to create portals for LNet: %d\n", rc);
1047 lnet_unprepare (void)
1049 /* NB no LNET_LOCK since this is the last reference. All LND instances
1050 * have shut down already, so it is safe to unlink and free all
1051 * descriptors, even those that appear committed to a network op (eg MD
1052 * with non-zero pending count) */
1054 lnet_fail_nid(LNET_NID_ANY, 0);
1056 LASSERT(the_lnet.ln_refcount == 0);
1057 LASSERT(list_empty(&the_lnet.ln_test_peers));
1058 LASSERT(list_empty(&the_lnet.ln_nets));
1060 lnet_portals_destroy();
1062 if (the_lnet.ln_md_containers != NULL) {
1063 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1064 the_lnet.ln_md_containers = NULL;
1067 if (the_lnet.ln_me_containers != NULL) {
1068 lnet_res_containers_destroy(the_lnet.ln_me_containers);
1069 the_lnet.ln_me_containers = NULL;
1072 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1074 lnet_msg_containers_destroy();
1076 lnet_rtrpools_free(0);
1078 if (the_lnet.ln_counters != NULL) {
1079 cfs_percpt_free(the_lnet.ln_counters);
1080 the_lnet.ln_counters = NULL;
1082 lnet_destroy_remote_nets_table();
1083 lnet_descriptor_cleanup();
1089 lnet_net2ni_locked(__u32 net_id, int cpt)
1092 struct lnet_net *net;
1094 LASSERT(cpt != LNET_LOCK_EX);
1096 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1097 if (net->net_id == net_id) {
1098 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1108 lnet_net2ni_addref(__u32 net)
1113 ni = lnet_net2ni_locked(net, 0);
1115 lnet_ni_addref_locked(ni, 0);
1120 EXPORT_SYMBOL(lnet_net2ni_addref);
1123 lnet_get_net_locked(__u32 net_id)
1125 struct lnet_net *net;
1127 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1128 if (net->net_id == net_id)
1136 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1141 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1146 val = hash_long(key, LNET_CPT_BITS);
1147 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1151 return (unsigned int)(key + val + (val >> 1)) % number;
1155 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1157 struct lnet_net *net;
1159 /* must called with hold of lnet_net_lock */
1160 if (LNET_CPT_NUMBER == 1)
1161 return 0; /* the only one */
1164 * If NI is provided then use the CPT identified in the NI cpt
1165 * list if one exists. If one doesn't exist, then that NI is
1166 * associated with all CPTs and it follows that the net it belongs
1167 * to is implicitly associated with all CPTs, so just hash the nid
1171 if (ni->ni_cpts != NULL)
1172 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1175 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1178 /* no NI provided so look at the net */
1179 net = lnet_get_net_locked(LNET_NIDNET(nid));
1181 if (net != NULL && net->net_cpts != NULL) {
1182 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1185 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1189 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1194 if (LNET_CPT_NUMBER == 1)
1195 return 0; /* the only one */
1197 cpt = lnet_net_lock_current();
1199 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1201 lnet_net_unlock(cpt);
1205 EXPORT_SYMBOL(lnet_cpt_of_nid);
1208 lnet_islocalnet(__u32 net_id)
1210 struct lnet_net *net;
1214 cpt = lnet_net_lock_current();
1216 net = lnet_get_net_locked(net_id);
1218 local = net != NULL;
1220 lnet_net_unlock(cpt);
1226 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1228 struct lnet_net *net;
1231 LASSERT(cpt != LNET_LOCK_EX);
1233 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1234 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1235 if (ni->ni_nid == nid)
1244 lnet_nid2ni_addref(lnet_nid_t nid)
1249 ni = lnet_nid2ni_locked(nid, 0);
1251 lnet_ni_addref_locked(ni, 0);
1256 EXPORT_SYMBOL(lnet_nid2ni_addref);
1259 lnet_islocalnid(lnet_nid_t nid)
1264 cpt = lnet_net_lock_current();
1265 ni = lnet_nid2ni_locked(nid, cpt);
1266 lnet_net_unlock(cpt);
1272 lnet_count_acceptor_nets(void)
1274 /* Return the # of NIs that need the acceptor. */
1276 struct lnet_net *net;
1279 cpt = lnet_net_lock_current();
1280 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1281 /* all socklnd type networks should have the acceptor
1283 if (net->net_lnd->lnd_accept != NULL)
1287 lnet_net_unlock(cpt);
1292 struct lnet_ping_buffer *
1293 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1295 struct lnet_ping_buffer *pbuf;
1297 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1299 pbuf->pb_nnis = nnis;
1300 atomic_set(&pbuf->pb_refcnt, 1);
1307 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1309 LASSERT(lnet_ping_buffer_numref(pbuf) == 0);
1310 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1313 static struct lnet_ping_buffer *
1314 lnet_ping_target_create(int nnis)
1316 struct lnet_ping_buffer *pbuf;
1318 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1320 CERROR("Can't allocate ping source [%d]\n", nnis);
1324 pbuf->pb_info.pi_nnis = nnis;
1325 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1326 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1327 pbuf->pb_info.pi_features =
1328 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1334 lnet_get_net_ni_count_locked(struct lnet_net *net)
1339 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1346 lnet_get_net_ni_count_pre(struct lnet_net *net)
1351 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1358 lnet_get_ni_count(void)
1361 struct lnet_net *net;
1366 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1367 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1377 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1381 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1383 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1385 /* Loopback is guaranteed to be present */
1386 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1388 if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1394 lnet_ping_target_destroy(void)
1396 struct lnet_net *net;
1399 lnet_net_lock(LNET_LOCK_EX);
1401 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1402 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1404 ni->ni_status = NULL;
1409 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1410 the_lnet.ln_ping_target = NULL;
1412 lnet_net_unlock(LNET_LOCK_EX);
1416 lnet_ping_target_event_handler(struct lnet_event *event)
1418 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1420 if (event->unlinked)
1421 lnet_ping_buffer_decref(pbuf);
1425 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1426 struct lnet_handle_md *ping_mdh,
1427 int ni_count, bool set_eq)
1429 struct lnet_process_id id = {
1430 .nid = LNET_NID_ANY,
1433 struct lnet_handle_me me_handle;
1434 struct lnet_md md = { NULL };
1438 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1439 &the_lnet.ln_ping_target_eq);
1441 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1446 *ppbuf = lnet_ping_target_create(ni_count);
1447 if (*ppbuf == NULL) {
1452 /* Ping target ME/MD */
1453 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1454 LNET_PROTO_PING_MATCHBITS, 0,
1455 LNET_UNLINK, LNET_INS_AFTER,
1458 CERROR("Can't create ping target ME: %d\n", rc);
1459 goto fail_decref_ping_buffer;
1462 /* initialize md content */
1463 md.start = &(*ppbuf)->pb_info;
1464 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1465 md.threshold = LNET_MD_THRESH_INF;
1467 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1468 LNET_MD_MANAGE_REMOTE;
1469 md.eq_handle = the_lnet.ln_ping_target_eq;
1470 md.user_ptr = *ppbuf;
1472 rc = LNetMDAttach(me_handle, md, LNET_RETAIN, ping_mdh);
1474 CERROR("Can't attach ping target MD: %d\n", rc);
1475 goto fail_unlink_ping_me;
1477 lnet_ping_buffer_addref(*ppbuf);
1481 fail_unlink_ping_me:
1482 rc2 = LNetMEUnlink(me_handle);
1484 fail_decref_ping_buffer:
1485 LASSERT(lnet_ping_buffer_numref(*ppbuf) == 1);
1486 lnet_ping_buffer_decref(*ppbuf);
1490 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1497 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1498 struct lnet_handle_md *ping_mdh)
1500 sigset_t blocked = cfs_block_allsigs();
1502 LNetMDUnlink(*ping_mdh);
1503 LNetInvalidateMDHandle(ping_mdh);
1505 /* NB the MD could be busy; this just starts the unlink */
1506 while (lnet_ping_buffer_numref(pbuf) > 1) {
1507 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1508 set_current_state(TASK_UNINTERRUPTIBLE);
1509 schedule_timeout(cfs_time_seconds(1));
1512 cfs_restore_sigs(blocked);
1516 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1519 struct lnet_net *net;
1520 struct lnet_ni_status *ns;
1525 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1526 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1527 LASSERT(i < pbuf->pb_nnis);
1529 ns = &pbuf->pb_info.pi_ni[i];
1531 ns->ns_nid = ni->ni_nid;
1534 ns->ns_status = (ni->ni_status != NULL) ?
1535 ni->ni_status->ns_status :
1544 * We (ab)use the ns_status of the loopback interface to
1545 * transmit the sequence number. The first interface listed
1546 * must be the loopback interface.
1548 rc = lnet_ping_info_validate(&pbuf->pb_info);
1550 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1553 LNET_PING_BUFFER_SEQNO(pbuf) =
1554 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1558 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1559 struct lnet_handle_md ping_mdh)
1561 struct lnet_ping_buffer *old_pbuf = NULL;
1562 struct lnet_handle_md old_ping_md;
1564 /* switch the NIs to point to the new ping info created */
1565 lnet_net_lock(LNET_LOCK_EX);
1567 if (!the_lnet.ln_routing)
1568 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1569 if (!lnet_peer_discovery_disabled)
1570 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1572 /* Ensure only known feature bits have been set. */
1573 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1574 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1576 lnet_ping_target_install_locked(pbuf);
1578 if (the_lnet.ln_ping_target) {
1579 old_pbuf = the_lnet.ln_ping_target;
1580 old_ping_md = the_lnet.ln_ping_target_md;
1582 the_lnet.ln_ping_target_md = ping_mdh;
1583 the_lnet.ln_ping_target = pbuf;
1585 lnet_net_unlock(LNET_LOCK_EX);
1588 /* unlink and free the old ping info */
1589 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1590 lnet_ping_buffer_decref(old_pbuf);
1593 lnet_push_update_to_peers(0);
1597 lnet_ping_target_fini(void)
1601 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1602 &the_lnet.ln_ping_target_md);
1604 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1607 lnet_ping_target_destroy();
1610 /* Resize the push target. */
1611 int lnet_push_target_resize(void)
1613 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1614 struct lnet_md md = { NULL };
1615 struct lnet_handle_me meh;
1616 struct lnet_handle_md mdh;
1617 struct lnet_handle_md old_mdh;
1618 struct lnet_ping_buffer *pbuf;
1619 struct lnet_ping_buffer *old_pbuf;
1620 int nnis = the_lnet.ln_push_target_nnis;
1628 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1634 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1635 LNET_PROTO_PING_MATCHBITS, 0,
1636 LNET_UNLINK, LNET_INS_AFTER,
1639 CERROR("Can't create push target ME: %d\n", rc);
1640 goto fail_decref_pbuf;
1643 /* initialize md content */
1644 md.start = &pbuf->pb_info;
1645 md.length = LNET_PING_INFO_SIZE(nnis);
1646 md.threshold = LNET_MD_THRESH_INF;
1648 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1649 LNET_MD_MANAGE_REMOTE;
1651 md.eq_handle = the_lnet.ln_push_target_eq;
1653 rc = LNetMDAttach(meh, md, LNET_RETAIN, &mdh);
1655 CERROR("Can't attach push MD: %d\n", rc);
1656 goto fail_unlink_meh;
1658 lnet_ping_buffer_addref(pbuf);
1660 lnet_net_lock(LNET_LOCK_EX);
1661 old_pbuf = the_lnet.ln_push_target;
1662 old_mdh = the_lnet.ln_push_target_md;
1663 the_lnet.ln_push_target = pbuf;
1664 the_lnet.ln_push_target_md = mdh;
1665 lnet_net_unlock(LNET_LOCK_EX);
1668 LNetMDUnlink(old_mdh);
1669 lnet_ping_buffer_decref(old_pbuf);
1672 if (nnis < the_lnet.ln_push_target_nnis)
1675 CDEBUG(D_NET, "nnis %d success\n", nnis);
1682 lnet_ping_buffer_decref(pbuf);
1684 CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1688 static void lnet_push_target_event_handler(struct lnet_event *ev)
1690 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1692 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1693 lnet_swap_pinginfo(pbuf);
1695 lnet_peer_push_event(ev);
1697 lnet_ping_buffer_decref(pbuf);
1700 /* Initialize the push target. */
1701 static int lnet_push_target_init(void)
1705 if (the_lnet.ln_push_target)
1708 rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1709 &the_lnet.ln_push_target_eq);
1711 CERROR("Can't allocated push target EQ: %d\n", rc);
1715 /* Start at the required minimum, we'll enlarge if required. */
1716 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1718 rc = lnet_push_target_resize();
1721 LNetEQFree(the_lnet.ln_push_target_eq);
1722 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1728 /* Clean up the push target. */
1729 static void lnet_push_target_fini(void)
1731 if (!the_lnet.ln_push_target)
1734 /* Unlink and invalidate to prevent new references. */
1735 LNetMDUnlink(the_lnet.ln_push_target_md);
1736 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1738 /* Wait for the unlink to complete. */
1739 while (lnet_ping_buffer_numref(the_lnet.ln_push_target) > 1) {
1740 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1741 set_current_state(TASK_UNINTERRUPTIBLE);
1742 schedule_timeout(cfs_time_seconds(1));
1745 lnet_ping_buffer_decref(the_lnet.ln_push_target);
1746 the_lnet.ln_push_target = NULL;
1747 the_lnet.ln_push_target_nnis = 0;
1749 LNetEQFree(the_lnet.ln_push_target_eq);
1750 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1754 lnet_ni_tq_credits(struct lnet_ni *ni)
1758 LASSERT(ni->ni_ncpts >= 1);
1760 if (ni->ni_ncpts == 1)
1761 return ni->ni_net->net_tunables.lct_max_tx_credits;
1763 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
1764 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
1765 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
1771 lnet_ni_unlink_locked(struct lnet_ni *ni)
1773 if (!list_empty(&ni->ni_cptlist)) {
1774 list_del_init(&ni->ni_cptlist);
1775 lnet_ni_decref_locked(ni, 0);
1778 /* move it to zombie list and nobody can find it anymore */
1779 LASSERT(!list_empty(&ni->ni_netlist));
1780 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
1781 lnet_ni_decref_locked(ni, 0);
1785 lnet_clear_zombies_nis_locked(struct lnet_net *net)
1790 struct list_head *zombie_list = &net->net_ni_zombie;
1793 * Now wait for the NIs I just nuked to show up on the zombie
1794 * list and shut them down in guaranteed thread context
1797 while (!list_empty(zombie_list)) {
1801 ni = list_entry(zombie_list->next,
1802 struct lnet_ni, ni_netlist);
1803 list_del_init(&ni->ni_netlist);
1804 /* the ni should be in deleting state. If it's not it's
1806 LASSERT(ni->ni_state & LNET_NI_STATE_DELETING);
1807 cfs_percpt_for_each(ref, j, ni->ni_refs) {
1810 /* still busy, add it back to zombie list */
1811 list_add(&ni->ni_netlist, zombie_list);
1815 if (!list_empty(&ni->ni_netlist)) {
1816 lnet_net_unlock(LNET_LOCK_EX);
1818 if ((i & (-i)) == i) {
1820 "Waiting for zombie LNI %s\n",
1821 libcfs_nid2str(ni->ni_nid));
1823 set_current_state(TASK_UNINTERRUPTIBLE);
1824 schedule_timeout(cfs_time_seconds(1));
1825 lnet_net_lock(LNET_LOCK_EX);
1829 lnet_net_unlock(LNET_LOCK_EX);
1831 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
1833 LASSERT(!in_interrupt());
1834 (net->net_lnd->lnd_shutdown)(ni);
1837 CDEBUG(D_LNI, "Removed LNI %s\n",
1838 libcfs_nid2str(ni->ni_nid));
1842 lnet_net_lock(LNET_LOCK_EX);
1846 /* shutdown down the NI and release refcount */
1848 lnet_shutdown_lndni(struct lnet_ni *ni)
1851 struct lnet_net *net = ni->ni_net;
1853 lnet_net_lock(LNET_LOCK_EX);
1855 ni->ni_state |= LNET_NI_STATE_DELETING;
1856 ni->ni_state &= ~LNET_NI_STATE_ACTIVE;
1858 lnet_ni_unlink_locked(ni);
1859 lnet_incr_dlc_seq();
1860 lnet_net_unlock(LNET_LOCK_EX);
1862 /* clear messages for this NI on the lazy portal */
1863 for (i = 0; i < the_lnet.ln_nportals; i++)
1864 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
1866 lnet_net_lock(LNET_LOCK_EX);
1867 lnet_clear_zombies_nis_locked(net);
1868 lnet_net_unlock(LNET_LOCK_EX);
1872 lnet_shutdown_lndnet(struct lnet_net *net)
1876 lnet_net_lock(LNET_LOCK_EX);
1878 net->net_state = LNET_NET_STATE_DELETING;
1880 list_del_init(&net->net_list);
1882 while (!list_empty(&net->net_ni_list)) {
1883 ni = list_entry(net->net_ni_list.next,
1884 struct lnet_ni, ni_netlist);
1885 lnet_net_unlock(LNET_LOCK_EX);
1886 lnet_shutdown_lndni(ni);
1887 lnet_net_lock(LNET_LOCK_EX);
1890 lnet_net_unlock(LNET_LOCK_EX);
1892 /* Do peer table cleanup for this net */
1893 lnet_peer_tables_cleanup(net);
1895 lnet_net_lock(LNET_LOCK_EX);
1897 * decrement ref count on lnd only when the entire network goes
1900 net->net_lnd->lnd_refcount--;
1902 lnet_net_unlock(LNET_LOCK_EX);
1908 lnet_shutdown_lndnets(void)
1910 struct lnet_net *net;
1911 struct list_head resend;
1912 struct lnet_msg *msg, *tmp;
1914 INIT_LIST_HEAD(&resend);
1916 /* NB called holding the global mutex */
1918 /* All quiet on the API front */
1919 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
1920 LASSERT(the_lnet.ln_refcount == 0);
1922 lnet_net_lock(LNET_LOCK_EX);
1923 the_lnet.ln_state = LNET_STATE_STOPPING;
1925 while (!list_empty(&the_lnet.ln_nets)) {
1927 * move the nets to the zombie list to avoid them being
1928 * picked up for new work. LONET is also included in the
1929 * Nets that will be moved to the zombie list
1931 net = list_entry(the_lnet.ln_nets.next,
1932 struct lnet_net, net_list);
1933 list_move(&net->net_list, &the_lnet.ln_net_zombie);
1936 /* Drop the cached loopback Net. */
1937 if (the_lnet.ln_loni != NULL) {
1938 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
1939 the_lnet.ln_loni = NULL;
1941 lnet_net_unlock(LNET_LOCK_EX);
1943 /* iterate through the net zombie list and delete each net */
1944 while (!list_empty(&the_lnet.ln_net_zombie)) {
1945 net = list_entry(the_lnet.ln_net_zombie.next,
1946 struct lnet_net, net_list);
1947 lnet_shutdown_lndnet(net);
1950 spin_lock(&the_lnet.ln_msg_resend_lock);
1951 list_splice(&the_lnet.ln_msg_resend, &resend);
1952 spin_unlock(&the_lnet.ln_msg_resend_lock);
1954 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
1955 list_del_init(&msg->msg_list);
1956 msg->msg_no_resend = true;
1957 lnet_finalize(msg, -ECANCELED);
1960 lnet_net_lock(LNET_LOCK_EX);
1961 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
1962 lnet_net_unlock(LNET_LOCK_EX);
1966 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
1969 struct lnet_tx_queue *tq;
1971 struct lnet_net *net = ni->ni_net;
1973 mutex_lock(&the_lnet.ln_lnd_mutex);
1976 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
1977 ni->ni_lnd_tunables_set = true;
1980 rc = (net->net_lnd->lnd_startup)(ni);
1982 mutex_unlock(&the_lnet.ln_lnd_mutex);
1985 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1986 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
1987 lnet_net_lock(LNET_LOCK_EX);
1988 net->net_lnd->lnd_refcount--;
1989 lnet_net_unlock(LNET_LOCK_EX);
1994 ni->ni_state |= LNET_NI_STATE_ACTIVE;
1995 ni->ni_state &= ~LNET_NI_STATE_INIT;
1998 /* We keep a reference on the loopback net through the loopback NI */
1999 if (net->net_lnd->lnd_type == LOLND) {
2001 LASSERT(the_lnet.ln_loni == NULL);
2002 the_lnet.ln_loni = ni;
2003 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2004 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2005 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2006 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2010 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2011 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2012 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2013 libcfs_lnd2str(net->net_lnd->lnd_type),
2014 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2016 /* shutdown the NI since if we get here then it must've already
2019 lnet_shutdown_lndni(ni);
2023 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2024 tq->tq_credits_min =
2025 tq->tq_credits_max =
2026 tq->tq_credits = lnet_ni_tq_credits(ni);
2029 atomic_set(&ni->ni_tx_credits,
2030 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2031 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2033 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2034 libcfs_nid2str(ni->ni_nid),
2035 ni->ni_net->net_tunables.lct_peer_tx_credits,
2036 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2037 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2038 ni->ni_net->net_tunables.lct_peer_timeout);
2047 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2050 struct lnet_net *net_l = NULL;
2051 struct list_head local_ni_list;
2055 struct lnet_lnd *lnd;
2057 net->net_tunables.lct_peer_timeout;
2059 net->net_tunables.lct_max_tx_credits;
2060 int peerrtrcredits =
2061 net->net_tunables.lct_peer_rtr_credits;
2063 INIT_LIST_HEAD(&local_ni_list);
2066 * make sure that this net is unique. If it isn't then
2067 * we are adding interfaces to an already existing network, and
2068 * 'net' is just a convenient way to pass in the list.
2069 * if it is unique we need to find the LND and load it if
2072 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2073 lnd_type = LNET_NETTYP(net->net_id);
2075 mutex_lock(&the_lnet.ln_lnd_mutex);
2076 lnd = lnet_find_lnd_by_type(lnd_type);
2079 mutex_unlock(&the_lnet.ln_lnd_mutex);
2080 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2081 mutex_lock(&the_lnet.ln_lnd_mutex);
2083 lnd = lnet_find_lnd_by_type(lnd_type);
2085 mutex_unlock(&the_lnet.ln_lnd_mutex);
2086 CERROR("Can't load LND %s, module %s, rc=%d\n",
2087 libcfs_lnd2str(lnd_type),
2088 libcfs_lnd2modname(lnd_type), rc);
2089 #ifndef HAVE_MODULE_LOADING_SUPPORT
2090 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2091 "compiled with kernel module "
2092 "loading support.");
2099 lnet_net_lock(LNET_LOCK_EX);
2100 lnd->lnd_refcount++;
2101 lnet_net_unlock(LNET_LOCK_EX);
2105 mutex_unlock(&the_lnet.ln_lnd_mutex);
2111 * net_l: if the network being added is unique then net_l
2112 * will point to that network
2113 * if the network being added is not unique then
2114 * net_l points to the existing network.
2116 * When we enter the loop below, we'll pick NIs off he
2117 * network beign added and start them up, then add them to
2118 * a local ni list. Once we've successfully started all
2119 * the NIs then we join the local NI list (of started up
2120 * networks) with the net_l->net_ni_list, which should
2121 * point to the correct network to add the new ni list to
2123 * If any of the new NIs fail to start up, then we want to
2124 * iterate through the local ni list, which should include
2125 * any NIs which were successfully started up, and shut
2128 * After than we want to delete the network being added,
2129 * to avoid a memory leak.
2133 * When a network uses TCP bonding then all its interfaces
2134 * must be specified when the network is first defined: the
2135 * TCP bonding code doesn't allow for interfaces to be added
2138 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2139 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2144 while (!list_empty(&net->net_ni_added)) {
2145 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2147 list_del_init(&ni->ni_netlist);
2149 /* make sure that the the NI we're about to start
2150 * up is actually unique. if it's not fail. */
2151 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2152 ni->ni_interfaces[0])) {
2157 /* adjust the pointer the parent network, just in case it
2158 * the net is a duplicate */
2161 rc = lnet_startup_lndni(ni, tun);
2163 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
2164 ni->ni_net->net_lnd->lnd_query != NULL);
2170 list_add_tail(&ni->ni_netlist, &local_ni_list);
2175 lnet_net_lock(LNET_LOCK_EX);
2176 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2177 lnet_incr_dlc_seq();
2178 lnet_net_unlock(LNET_LOCK_EX);
2180 /* if the network is not unique then we don't want to keep
2181 * it around after we're done. Free it. Otherwise add that
2182 * net to the global the_lnet.ln_nets */
2183 if (net_l != net && net_l != NULL) {
2185 * TODO - note. currently the tunables can not be updated
2190 net->net_state = LNET_NET_STATE_ACTIVE;
2192 * restore tunables after it has been overwitten by the
2195 if (peer_timeout != -1)
2196 net->net_tunables.lct_peer_timeout = peer_timeout;
2197 if (maxtxcredits != -1)
2198 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2199 if (peerrtrcredits != -1)
2200 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2202 lnet_net_lock(LNET_LOCK_EX);
2203 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2204 lnet_net_unlock(LNET_LOCK_EX);
2211 * shutdown the new NIs that are being started up
2212 * free the NET being started
2214 while (!list_empty(&local_ni_list)) {
2215 ni = list_entry(local_ni_list.next, struct lnet_ni,
2218 lnet_shutdown_lndni(ni);
2228 lnet_startup_lndnets(struct list_head *netlist)
2230 struct lnet_net *net;
2235 * Change to running state before bringing up the LNDs. This
2236 * allows lnet_shutdown_lndnets() to assert that we've passed
2239 lnet_net_lock(LNET_LOCK_EX);
2240 the_lnet.ln_state = LNET_STATE_RUNNING;
2241 lnet_net_unlock(LNET_LOCK_EX);
2243 while (!list_empty(netlist)) {
2244 net = list_entry(netlist->next, struct lnet_net, net_list);
2245 list_del_init(&net->net_list);
2247 rc = lnet_startup_lndnet(net, NULL);
2257 lnet_shutdown_lndnets();
2263 * Initialize LNet library.
2265 * Automatically called at module loading time. Caller has to call
2266 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2267 * latter returned 0. It must be called exactly once.
2269 * \retval 0 on success
2270 * \retval -ve on failures.
2272 int lnet_lib_init(void)
2276 lnet_assert_wire_constants();
2278 /* refer to global cfs_cpt_table for now */
2279 the_lnet.ln_cpt_table = cfs_cpt_table;
2280 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
2282 LASSERT(the_lnet.ln_cpt_number > 0);
2283 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2284 /* we are under risk of consuming all lh_cookie */
2285 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2286 "please change setting of CPT-table and retry\n",
2287 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2291 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2292 the_lnet.ln_cpt_bits++;
2294 rc = lnet_create_locks();
2296 CERROR("Can't create LNet global locks: %d\n", rc);
2300 the_lnet.ln_refcount = 0;
2301 LNetInvalidateEQHandle(&the_lnet.ln_rc_eqh);
2302 INIT_LIST_HEAD(&the_lnet.ln_lnds);
2303 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2304 INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
2305 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2306 INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
2308 /* The hash table size is the number of bits it takes to express the set
2309 * ln_num_routes, minus 1 (better to under estimate than over so we
2310 * don't waste memory). */
2311 if (rnet_htable_size <= 0)
2312 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2313 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2314 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2315 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2316 order_base_2(rnet_htable_size) - 1);
2318 /* All LNDs apart from the LOLND are in separate modules. They
2319 * register themselves when their module loads, and unregister
2320 * themselves when their module is unloaded. */
2321 lnet_register_lnd(&the_lolnd);
2326 * Finalize LNet library.
2328 * \pre lnet_lib_init() called with success.
2329 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2331 void lnet_lib_exit(void)
2333 LASSERT(the_lnet.ln_refcount == 0);
2335 while (!list_empty(&the_lnet.ln_lnds))
2336 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
2337 struct lnet_lnd, lnd_list));
2338 lnet_destroy_locks();
2342 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2344 * Users must call this function at least once before any other functions.
2345 * For each successful call there must be a corresponding call to
2346 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2349 * The PID used by LNet may be different from the one requested.
2352 * \param requested_pid PID requested by the caller.
2354 * \return >= 0 on success, and < 0 error code on failures.
2357 LNetNIInit(lnet_pid_t requested_pid)
2359 int im_a_router = 0;
2362 struct lnet_ping_buffer *pbuf;
2363 struct lnet_handle_md ping_mdh;
2364 struct list_head net_head;
2365 struct lnet_net *net;
2367 INIT_LIST_HEAD(&net_head);
2369 mutex_lock(&the_lnet.ln_api_mutex);
2371 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2373 if (the_lnet.ln_refcount > 0) {
2374 rc = the_lnet.ln_refcount++;
2375 mutex_unlock(&the_lnet.ln_api_mutex);
2379 rc = lnet_prepare(requested_pid);
2381 mutex_unlock(&the_lnet.ln_api_mutex);
2385 /* create a network for Loopback network */
2386 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2389 goto err_empty_list;
2392 /* Add in the loopback NI */
2393 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2395 goto err_empty_list;
2398 /* If LNet is being initialized via DLC it is possible
2399 * that the user requests not to load module parameters (ones which
2400 * are supported by DLC) on initialization. Therefore, make sure not
2401 * to load networks, routes and forwarding from module parameters
2402 * in this case. On cleanup in case of failure only clean up
2403 * routes if it has been loaded */
2404 if (!the_lnet.ln_nis_from_mod_params) {
2405 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2408 goto err_empty_list;
2411 ni_count = lnet_startup_lndnets(&net_head);
2414 goto err_empty_list;
2417 if (!the_lnet.ln_nis_from_mod_params) {
2418 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2420 goto err_shutdown_lndnis;
2422 rc = lnet_check_routes();
2424 goto err_destroy_routes;
2426 rc = lnet_rtrpools_alloc(im_a_router);
2428 goto err_destroy_routes;
2431 rc = lnet_acceptor_start();
2433 goto err_destroy_routes;
2435 the_lnet.ln_refcount = 1;
2436 /* Now I may use my own API functions... */
2438 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2440 goto err_acceptor_stop;
2442 lnet_ping_target_update(pbuf, ping_mdh);
2444 rc = lnet_monitor_thr_start();
2448 rc = lnet_push_target_init();
2450 goto err_stop_monitor_thr;
2452 rc = lnet_peer_discovery_start();
2454 goto err_destroy_push_target;
2457 lnet_router_debugfs_init();
2459 mutex_unlock(&the_lnet.ln_api_mutex);
2463 err_destroy_push_target:
2464 lnet_push_target_fini();
2465 err_stop_monitor_thr:
2466 lnet_monitor_thr_stop();
2468 lnet_ping_target_fini();
2470 the_lnet.ln_refcount = 0;
2471 lnet_acceptor_stop();
2473 if (!the_lnet.ln_nis_from_mod_params)
2474 lnet_destroy_routes();
2475 err_shutdown_lndnis:
2476 lnet_shutdown_lndnets();
2480 mutex_unlock(&the_lnet.ln_api_mutex);
2481 while (!list_empty(&net_head)) {
2482 struct lnet_net *net;
2484 net = list_entry(net_head.next, struct lnet_net, net_list);
2485 list_del_init(&net->net_list);
2490 EXPORT_SYMBOL(LNetNIInit);
2493 * Stop LNet interfaces, routing, and forwarding.
2495 * Users must call this function once for each successful call to LNetNIInit().
2496 * Once the LNetNIFini() operation has been started, the results of pending
2497 * API operations are undefined.
2499 * \return always 0 for current implementation.
2504 mutex_lock(&the_lnet.ln_api_mutex);
2506 LASSERT(the_lnet.ln_refcount > 0);
2508 if (the_lnet.ln_refcount != 1) {
2509 the_lnet.ln_refcount--;
2511 LASSERT(!the_lnet.ln_niinit_self);
2515 lnet_router_debugfs_init();
2516 lnet_peer_discovery_stop();
2517 lnet_push_target_fini();
2518 lnet_monitor_thr_stop();
2519 lnet_ping_target_fini();
2521 /* Teardown fns that use my own API functions BEFORE here */
2522 the_lnet.ln_refcount = 0;
2524 lnet_acceptor_stop();
2525 lnet_destroy_routes();
2526 lnet_shutdown_lndnets();
2530 mutex_unlock(&the_lnet.ln_api_mutex);
2533 EXPORT_SYMBOL(LNetNIFini);
2536 * Grabs the ni data from the ni structure and fills the out
2539 * \param[in] ni network interface structure
2540 * \param[out] cfg_ni NI config information
2541 * \param[out] tun network and LND tunables
2544 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2545 struct lnet_ioctl_config_lnd_tunables *tun,
2546 struct lnet_ioctl_element_stats *stats,
2549 size_t min_size = 0;
2552 if (!ni || !cfg_ni || !tun)
2555 if (ni->ni_interfaces[0] != NULL) {
2556 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2557 if (ni->ni_interfaces[i] != NULL) {
2558 strncpy(cfg_ni->lic_ni_intf[i],
2559 ni->ni_interfaces[i],
2560 sizeof(cfg_ni->lic_ni_intf[i]));
2565 cfg_ni->lic_nid = ni->ni_nid;
2566 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2567 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2569 cfg_ni->lic_status = ni->ni_status->ns_status;
2570 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2571 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2573 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2576 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2577 LNET_STATS_TYPE_SEND);
2578 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2579 LNET_STATS_TYPE_RECV);
2580 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2581 LNET_STATS_TYPE_DROP);
2585 * tun->lt_tun will always be present, but in order to be
2586 * backwards compatible, we need to deal with the cases when
2587 * tun->lt_tun is smaller than what the kernel has, because it
2588 * comes from an older version of a userspace program, then we'll
2589 * need to copy as much information as we have available space.
2591 min_size = tun_size - sizeof(tun->lt_cmn);
2592 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2594 /* copy over the cpts */
2595 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2596 ni->ni_cpts == NULL) {
2597 for (i = 0; i < ni->ni_ncpts; i++)
2598 cfg_ni->lic_cpts[i] = i;
2601 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2602 i < LNET_MAX_SHOW_NUM_CPT;
2604 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2606 cfg_ni->lic_ncpts = ni->ni_ncpts;
2610 * NOTE: This is a legacy function left in the code to be backwards
2611 * compatible with older userspace programs. It should eventually be
2614 * Grabs the ni data from the ni structure and fills the out
2617 * \param[in] ni network interface structure
2618 * \param[out] config config information
2621 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2622 struct lnet_ioctl_config_data *config)
2624 struct lnet_ioctl_net_config *net_config;
2625 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2626 size_t min_size, tunable_size = 0;
2632 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2636 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2637 ARRAY_SIZE(net_config->ni_interfaces));
2639 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2640 if (!ni->ni_interfaces[i])
2643 strncpy(net_config->ni_interfaces[i],
2644 ni->ni_interfaces[i],
2645 sizeof(net_config->ni_interfaces[i]));
2648 config->cfg_nid = ni->ni_nid;
2649 config->cfg_config_u.cfg_net.net_peer_timeout =
2650 ni->ni_net->net_tunables.lct_peer_timeout;
2651 config->cfg_config_u.cfg_net.net_max_tx_credits =
2652 ni->ni_net->net_tunables.lct_max_tx_credits;
2653 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2654 ni->ni_net->net_tunables.lct_peer_tx_credits;
2655 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2656 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2658 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2659 net_config->ni_status = LNET_NI_STATUS_UP;
2661 net_config->ni_status = ni->ni_status->ns_status;
2664 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2666 for (i = 0; i < num_cpts; i++)
2667 net_config->ni_cpts[i] = ni->ni_cpts[i];
2669 config->cfg_ncpts = num_cpts;
2673 * See if user land tools sent in a newer and larger version
2674 * of struct lnet_tunables than what the kernel uses.
2676 min_size = sizeof(*config) + sizeof(*net_config);
2678 if (config->cfg_hdr.ioc_len > min_size)
2679 tunable_size = config->cfg_hdr.ioc_len - min_size;
2681 /* Don't copy too much data to user space */
2682 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2683 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2685 if (lnd_cfg && min_size) {
2686 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2687 config->cfg_config_u.cfg_net.net_interface_count = 1;
2689 /* Tell user land that kernel side has less data */
2690 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2691 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2692 config->cfg_hdr.ioc_len -= min_size;
2698 lnet_get_ni_idx_locked(int idx)
2701 struct lnet_net *net;
2703 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2704 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2714 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2717 struct lnet_net *net = mynet;
2720 * It is possible that the net has been cleaned out while there is
2721 * a message being sent. This function accessed the net without
2722 * checking if the list is empty
2726 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2728 if (list_empty(&net->net_ni_list))
2730 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2736 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2737 /* if you reached the end of the ni list and the net is
2738 * specified, then there are no more nis in that net */
2742 /* we reached the end of this net ni list. move to the
2744 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2745 /* no more nets and no more NIs. */
2748 /* get the next net */
2749 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2751 if (list_empty(&net->net_ni_list))
2753 /* get the ni on it */
2754 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2760 if (list_empty(&prev->ni_netlist))
2763 /* there are more nis left */
2764 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2770 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2775 int idx = config->cfg_count;
2777 cpt = lnet_net_lock_current();
2779 ni = lnet_get_ni_idx_locked(idx);
2784 lnet_fill_ni_info_legacy(ni, config);
2788 lnet_net_unlock(cpt);
2793 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
2794 struct lnet_ioctl_config_lnd_tunables *tun,
2795 struct lnet_ioctl_element_stats *stats,
2802 if (!cfg_ni || !tun || !stats)
2805 cpt = lnet_net_lock_current();
2807 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
2812 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
2816 lnet_net_unlock(cpt);
2820 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
2829 cpt = lnet_net_lock_current();
2831 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
2834 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
2838 lnet_net_unlock(cpt);
2843 static int lnet_add_net_common(struct lnet_net *net,
2844 struct lnet_ioctl_config_lnd_tunables *tun)
2847 struct lnet_ping_buffer *pbuf;
2848 struct lnet_handle_md ping_mdh;
2850 struct lnet_remotenet *rnet;
2852 int num_acceptor_nets;
2854 lnet_net_lock(LNET_LOCK_EX);
2855 rnet = lnet_find_rnet_locked(net->net_id);
2856 lnet_net_unlock(LNET_LOCK_EX);
2858 * make sure that the net added doesn't invalidate the current
2859 * configuration LNet is keeping
2862 CERROR("Adding net %s will invalidate routing configuration\n",
2863 libcfs_net2str(net->net_id));
2869 * make sure you calculate the correct number of slots in the ping
2870 * buffer. Since the ping info is a flattened list of all the NIs,
2871 * we should allocate enough slots to accomodate the number of NIs
2872 * which will be added.
2874 * since ni hasn't been configured yet, use
2875 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
2877 net_ni_count = lnet_get_net_ni_count_pre(net);
2879 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
2880 net_ni_count + lnet_get_ni_count(),
2888 memcpy(&net->net_tunables,
2889 &tun->lt_cmn, sizeof(net->net_tunables));
2891 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
2894 * before starting this network get a count of the current TCP
2895 * networks which require the acceptor thread running. If that
2896 * count is == 0 before we start up this network, then we'd want to
2897 * start up the acceptor thread after starting up this network
2899 num_acceptor_nets = lnet_count_acceptor_nets();
2901 net_id = net->net_id;
2903 rc = lnet_startup_lndnet(net,
2904 (tun) ? &tun->lt_tun : NULL);
2908 lnet_net_lock(LNET_LOCK_EX);
2909 net = lnet_get_net_locked(net_id);
2910 lnet_net_unlock(LNET_LOCK_EX);
2915 * Start the acceptor thread if this is the first network
2916 * being added that requires the thread.
2918 if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
2919 rc = lnet_acceptor_start();
2921 /* shutdown the net that we just started */
2922 CERROR("Failed to start up acceptor thread\n");
2923 lnet_shutdown_lndnet(net);
2928 lnet_net_lock(LNET_LOCK_EX);
2929 lnet_peer_net_added(net);
2930 lnet_net_unlock(LNET_LOCK_EX);
2932 lnet_ping_target_update(pbuf, ping_mdh);
2937 lnet_ping_md_unlink(pbuf, &ping_mdh);
2938 lnet_ping_buffer_decref(pbuf);
2942 static int lnet_handle_legacy_ip2nets(char *ip2nets,
2943 struct lnet_ioctl_config_lnd_tunables *tun)
2945 struct lnet_net *net;
2948 struct list_head net_head;
2950 INIT_LIST_HEAD(&net_head);
2952 rc = lnet_parse_ip2nets(&nets, ip2nets);
2956 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
2960 mutex_lock(&the_lnet.ln_api_mutex);
2961 while (!list_empty(&net_head)) {
2962 net = list_entry(net_head.next, struct lnet_net, net_list);
2963 list_del_init(&net->net_list);
2964 rc = lnet_add_net_common(net, tun);
2970 mutex_unlock(&the_lnet.ln_api_mutex);
2972 while (!list_empty(&net_head)) {
2973 net = list_entry(net_head.next, struct lnet_net, net_list);
2974 list_del_init(&net->net_list);
2980 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
2982 struct lnet_net *net;
2984 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
2986 __u32 net_id, lnd_type;
2988 /* get the tunables if they are available */
2989 if (conf->lic_cfg_hdr.ioc_len >=
2990 sizeof(*conf) + sizeof(*tun))
2991 tun = (struct lnet_ioctl_config_lnd_tunables *)
2994 /* handle legacy ip2nets from DLC */
2995 if (conf->lic_legacy_ip2nets[0] != '\0')
2996 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
2999 net_id = LNET_NIDNET(conf->lic_nid);
3000 lnd_type = LNET_NETTYP(net_id);
3002 if (!libcfs_isknown_lnd(lnd_type)) {
3003 CERROR("No valid net and lnd information provided\n");
3007 net = lnet_net_alloc(net_id, NULL);
3011 for (i = 0; i < conf->lic_ncpts; i++) {
3012 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3016 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3017 conf->lic_ni_intf[0]);
3021 mutex_lock(&the_lnet.ln_api_mutex);
3023 rc = lnet_add_net_common(net, tun);
3025 mutex_unlock(&the_lnet.ln_api_mutex);
3030 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3032 struct lnet_net *net;
3034 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3035 struct lnet_ping_buffer *pbuf;
3036 struct lnet_handle_md ping_mdh;
3041 /* don't allow userspace to shutdown the LOLND */
3042 if (LNET_NETTYP(net_id) == LOLND)
3045 mutex_lock(&the_lnet.ln_api_mutex);
3049 net = lnet_get_net_locked(net_id);
3051 CERROR("net %s not found\n",
3052 libcfs_net2str(net_id));
3057 addr = LNET_NIDADDR(conf->lic_nid);
3059 /* remove the entire net */
3060 net_count = lnet_get_net_ni_count_locked(net);
3064 /* create and link a new ping info, before removing the old one */
3065 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3066 lnet_get_ni_count() - net_count,
3069 goto unlock_api_mutex;
3071 lnet_shutdown_lndnet(net);
3073 if (lnet_count_acceptor_nets() == 0)
3074 lnet_acceptor_stop();
3076 lnet_ping_target_update(pbuf, ping_mdh);
3078 goto unlock_api_mutex;
3081 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3083 CERROR("nid %s not found\n",
3084 libcfs_nid2str(conf->lic_nid));
3089 net_count = lnet_get_net_ni_count_locked(net);
3093 /* create and link a new ping info, before removing the old one */
3094 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3095 lnet_get_ni_count() - 1, false);
3097 goto unlock_api_mutex;
3099 lnet_shutdown_lndni(ni);
3101 if (lnet_count_acceptor_nets() == 0)
3102 lnet_acceptor_stop();
3104 lnet_ping_target_update(pbuf, ping_mdh);
3106 /* check if the net is empty and remove it if it is */
3108 lnet_shutdown_lndnet(net);
3110 goto unlock_api_mutex;
3115 mutex_unlock(&the_lnet.ln_api_mutex);
3121 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3122 * They are only expected to be called for unique networks.
3123 * That can be as a result of older DLC library
3124 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3127 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3129 struct lnet_net *net;
3130 struct list_head net_head;
3132 struct lnet_ioctl_config_lnd_tunables tun;
3133 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3135 INIT_LIST_HEAD(&net_head);
3137 /* Create a net/ni structures for the network string */
3138 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3140 return rc == 0 ? -EINVAL : rc;
3142 mutex_lock(&the_lnet.ln_api_mutex);
3145 rc = -EINVAL; /* only add one network per call */
3146 goto out_unlock_clean;
3149 net = list_entry(net_head.next, struct lnet_net, net_list);
3150 list_del_init(&net->net_list);
3152 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3154 memset(&tun, 0, sizeof(tun));
3156 tun.lt_cmn.lct_peer_timeout =
3157 conf->cfg_config_u.cfg_net.net_peer_timeout;
3158 tun.lt_cmn.lct_peer_tx_credits =
3159 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3160 tun.lt_cmn.lct_peer_rtr_credits =
3161 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3162 tun.lt_cmn.lct_max_tx_credits =
3163 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3165 rc = lnet_add_net_common(net, &tun);
3168 mutex_unlock(&the_lnet.ln_api_mutex);
3169 while (!list_empty(&net_head)) {
3170 /* net_head list is empty in success case */
3171 net = list_entry(net_head.next, struct lnet_net, net_list);
3172 list_del_init(&net->net_list);
3179 lnet_dyn_del_net(__u32 net_id)
3181 struct lnet_net *net;
3182 struct lnet_ping_buffer *pbuf;
3183 struct lnet_handle_md ping_mdh;
3187 /* don't allow userspace to shutdown the LOLND */
3188 if (LNET_NETTYP(net_id) == LOLND)
3191 mutex_lock(&the_lnet.ln_api_mutex);
3195 net = lnet_get_net_locked(net_id);
3202 net_ni_count = lnet_get_net_ni_count_locked(net);
3206 /* create and link a new ping info, before removing the old one */
3207 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3208 lnet_get_ni_count() - net_ni_count, false);
3212 lnet_shutdown_lndnet(net);
3214 if (lnet_count_acceptor_nets() == 0)
3215 lnet_acceptor_stop();
3217 lnet_ping_target_update(pbuf, ping_mdh);
3220 mutex_unlock(&the_lnet.ln_api_mutex);
3225 void lnet_incr_dlc_seq(void)
3227 atomic_inc(&lnet_dlc_seq_no);
3230 __u32 lnet_get_dlc_seq_locked(void)
3232 return atomic_read(&lnet_dlc_seq_no);
3236 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3238 struct lnet_net *net;
3241 lnet_net_lock(LNET_LOCK_EX);
3242 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3243 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3244 if (ni->ni_nid == nid || all) {
3245 atomic_set(&ni->ni_healthv, value);
3246 if (list_empty(&ni->ni_recovery) &&
3247 value < LNET_MAX_HEALTH_VALUE) {
3248 CERROR("manually adding local NI %s to recovery\n",
3249 libcfs_nid2str(ni->ni_nid));
3250 list_add_tail(&ni->ni_recovery,
3251 &the_lnet.ln_mt_localNIRecovq);
3252 lnet_ni_addref_locked(ni, 0);
3255 lnet_net_unlock(LNET_LOCK_EX);
3261 lnet_net_unlock(LNET_LOCK_EX);
3265 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3269 lnet_nid_t nid = stats->hlni_nid;
3271 cpt = lnet_net_lock_current();
3272 ni = lnet_nid2ni_locked(nid, cpt);
3279 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3280 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3281 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3282 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3283 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3284 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3285 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3288 lnet_net_unlock(cpt);
3294 * LNet ioctl handler.
3298 LNetCtl(unsigned int cmd, void *arg)
3300 struct libcfs_ioctl_data *data = arg;
3301 struct lnet_ioctl_config_data *config;
3302 struct lnet_process_id id = {0};
3306 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3307 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3310 case IOC_LIBCFS_GET_NI:
3311 rc = LNetGetId(data->ioc_count, &id);
3312 data->ioc_nid = id.nid;
3315 case IOC_LIBCFS_FAIL_NID:
3316 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3318 case IOC_LIBCFS_ADD_ROUTE:
3321 if (config->cfg_hdr.ioc_len < sizeof(*config))
3324 mutex_lock(&the_lnet.ln_api_mutex);
3325 rc = lnet_add_route(config->cfg_net,
3326 config->cfg_config_u.cfg_route.rtr_hop,
3328 config->cfg_config_u.cfg_route.
3331 rc = lnet_check_routes();
3333 lnet_del_route(config->cfg_net,
3336 mutex_unlock(&the_lnet.ln_api_mutex);
3339 case IOC_LIBCFS_DEL_ROUTE:
3342 if (config->cfg_hdr.ioc_len < sizeof(*config))
3345 mutex_lock(&the_lnet.ln_api_mutex);
3346 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3347 mutex_unlock(&the_lnet.ln_api_mutex);
3350 case IOC_LIBCFS_GET_ROUTE:
3353 if (config->cfg_hdr.ioc_len < sizeof(*config))
3356 mutex_lock(&the_lnet.ln_api_mutex);
3357 rc = lnet_get_route(config->cfg_count,
3359 &config->cfg_config_u.cfg_route.rtr_hop,
3361 &config->cfg_config_u.cfg_route.rtr_flags,
3362 &config->cfg_config_u.cfg_route.
3364 mutex_unlock(&the_lnet.ln_api_mutex);
3367 case IOC_LIBCFS_GET_LOCAL_NI: {
3368 struct lnet_ioctl_config_ni *cfg_ni;
3369 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3370 struct lnet_ioctl_element_stats *stats;
3375 /* get the tunables if they are available */
3376 if (cfg_ni->lic_cfg_hdr.ioc_len <
3377 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3380 stats = (struct lnet_ioctl_element_stats *)
3382 tun = (struct lnet_ioctl_config_lnd_tunables *)
3383 (cfg_ni->lic_bulk + sizeof(*stats));
3385 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3388 mutex_lock(&the_lnet.ln_api_mutex);
3389 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3390 mutex_unlock(&the_lnet.ln_api_mutex);
3394 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3395 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3397 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3400 mutex_lock(&the_lnet.ln_api_mutex);
3401 rc = lnet_get_ni_stats(msg_stats);
3402 mutex_unlock(&the_lnet.ln_api_mutex);
3407 case IOC_LIBCFS_GET_NET: {
3408 size_t total = sizeof(*config) +
3409 sizeof(struct lnet_ioctl_net_config);
3412 if (config->cfg_hdr.ioc_len < total)
3415 mutex_lock(&the_lnet.ln_api_mutex);
3416 rc = lnet_get_net_config(config);
3417 mutex_unlock(&the_lnet.ln_api_mutex);
3421 case IOC_LIBCFS_GET_LNET_STATS:
3423 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3425 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3428 mutex_lock(&the_lnet.ln_api_mutex);
3429 lnet_counters_get(&lnet_stats->st_cntrs);
3430 mutex_unlock(&the_lnet.ln_api_mutex);
3434 case IOC_LIBCFS_CONFIG_RTR:
3437 if (config->cfg_hdr.ioc_len < sizeof(*config))
3440 mutex_lock(&the_lnet.ln_api_mutex);
3441 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3442 rc = lnet_rtrpools_enable();
3443 mutex_unlock(&the_lnet.ln_api_mutex);
3446 lnet_rtrpools_disable();
3447 mutex_unlock(&the_lnet.ln_api_mutex);
3450 case IOC_LIBCFS_ADD_BUF:
3453 if (config->cfg_hdr.ioc_len < sizeof(*config))
3456 mutex_lock(&the_lnet.ln_api_mutex);
3457 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3459 config->cfg_config_u.cfg_buffers.
3461 config->cfg_config_u.cfg_buffers.
3463 mutex_unlock(&the_lnet.ln_api_mutex);
3466 case IOC_LIBCFS_SET_NUMA_RANGE: {
3467 struct lnet_ioctl_set_value *numa;
3469 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3471 lnet_net_lock(LNET_LOCK_EX);
3472 lnet_numa_range = numa->sv_value;
3473 lnet_net_unlock(LNET_LOCK_EX);
3477 case IOC_LIBCFS_GET_NUMA_RANGE: {
3478 struct lnet_ioctl_set_value *numa;
3480 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3482 numa->sv_value = lnet_numa_range;
3486 case IOC_LIBCFS_GET_BUF: {
3487 struct lnet_ioctl_pool_cfg *pool_cfg;
3488 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3492 if (config->cfg_hdr.ioc_len < total)
3495 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3497 mutex_lock(&the_lnet.ln_api_mutex);
3498 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3499 mutex_unlock(&the_lnet.ln_api_mutex);
3503 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3504 struct lnet_ioctl_local_ni_hstats *stats = arg;
3506 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3509 mutex_lock(&the_lnet.ln_api_mutex);
3510 rc = lnet_get_local_ni_hstats(stats);
3511 mutex_unlock(&the_lnet.ln_api_mutex);
3516 case IOC_LIBCFS_ADD_PEER_NI: {
3517 struct lnet_ioctl_peer_cfg *cfg = arg;
3519 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3522 mutex_lock(&the_lnet.ln_api_mutex);
3523 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3526 mutex_unlock(&the_lnet.ln_api_mutex);
3530 case IOC_LIBCFS_DEL_PEER_NI: {
3531 struct lnet_ioctl_peer_cfg *cfg = arg;
3533 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3536 mutex_lock(&the_lnet.ln_api_mutex);
3537 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3538 cfg->prcfg_cfg_nid);
3539 mutex_unlock(&the_lnet.ln_api_mutex);
3543 case IOC_LIBCFS_GET_PEER_INFO: {
3544 struct lnet_ioctl_peer *peer_info = arg;
3546 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3549 mutex_lock(&the_lnet.ln_api_mutex);
3550 rc = lnet_get_peer_ni_info(
3551 peer_info->pr_count,
3553 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3554 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3555 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3556 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3557 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3558 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3559 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3560 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3561 mutex_unlock(&the_lnet.ln_api_mutex);
3565 case IOC_LIBCFS_GET_PEER_NI: {
3566 struct lnet_ioctl_peer_cfg *cfg = arg;
3568 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3571 mutex_lock(&the_lnet.ln_api_mutex);
3572 rc = lnet_get_peer_info(cfg,
3573 (void __user *)cfg->prcfg_bulk);
3574 mutex_unlock(&the_lnet.ln_api_mutex);
3578 case IOC_LIBCFS_GET_PEER_LIST: {
3579 struct lnet_ioctl_peer_cfg *cfg = arg;
3581 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3584 mutex_lock(&the_lnet.ln_api_mutex);
3585 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3586 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3587 mutex_unlock(&the_lnet.ln_api_mutex);
3591 case IOC_LIBCFS_SET_HEALHV: {
3592 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3594 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3596 if (cfg->rh_value < 0 ||
3597 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3598 value = LNET_MAX_HEALTH_VALUE;
3600 value = cfg->rh_value;
3601 mutex_lock(&the_lnet.ln_api_mutex);
3602 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3603 lnet_ni_set_healthv(cfg->rh_nid, value,
3606 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3608 mutex_unlock(&the_lnet.ln_api_mutex);
3611 case IOC_LIBCFS_NOTIFY_ROUTER: {
3612 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3614 /* The deadline passed in by the user should be some time in
3615 * seconds in the future since the UNIX epoch. We have to map
3616 * that deadline to the wall clock.
3618 deadline += ktime_get_seconds();
3619 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
3623 case IOC_LIBCFS_LNET_DIST:
3624 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3625 if (rc < 0 && rc != -EHOSTUNREACH)
3628 data->ioc_u32[0] = rc;
3631 case IOC_LIBCFS_TESTPROTOCOMPAT:
3632 lnet_net_lock(LNET_LOCK_EX);
3633 the_lnet.ln_testprotocompat = data->ioc_flags;
3634 lnet_net_unlock(LNET_LOCK_EX);
3637 case IOC_LIBCFS_LNET_FAULT:
3638 return lnet_fault_ctl(data->ioc_flags, data);
3640 case IOC_LIBCFS_PING: {
3641 signed long timeout;
3643 id.nid = data->ioc_nid;
3644 id.pid = data->ioc_u32[0];
3646 /* If timeout is negative then set default of 3 minutes */
3647 if (((s32)data->ioc_u32[1] <= 0) ||
3648 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3649 timeout = msecs_to_jiffies(DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC);
3651 timeout = msecs_to_jiffies(data->ioc_u32[1]);
3653 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3654 data->ioc_plen1 / sizeof(struct lnet_process_id));
3659 data->ioc_count = rc;
3663 case IOC_LIBCFS_PING_PEER: {
3664 struct lnet_ioctl_ping_data *ping = arg;
3665 struct lnet_peer *lp;
3666 signed long timeout;
3668 /* If timeout is negative then set default of 3 minutes */
3669 if (((s32)ping->op_param) <= 0 ||
3670 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3671 timeout = msecs_to_jiffies(DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC);
3673 timeout = msecs_to_jiffies(ping->op_param);
3675 rc = lnet_ping(ping->ping_id, timeout,
3681 mutex_lock(&the_lnet.ln_api_mutex);
3682 lp = lnet_find_peer(ping->ping_id.nid);
3684 ping->ping_id.nid = lp->lp_primary_nid;
3685 ping->mr_info = lnet_peer_is_multi_rail(lp);
3686 lnet_peer_decref_locked(lp);
3688 mutex_unlock(&the_lnet.ln_api_mutex);
3690 ping->ping_count = rc;
3694 case IOC_LIBCFS_DISCOVER: {
3695 struct lnet_ioctl_ping_data *discover = arg;
3696 struct lnet_peer *lp;
3698 rc = lnet_discover(discover->ping_id, discover->op_param,
3700 discover->ping_count);
3704 mutex_lock(&the_lnet.ln_api_mutex);
3705 lp = lnet_find_peer(discover->ping_id.nid);
3707 discover->ping_id.nid = lp->lp_primary_nid;
3708 discover->mr_info = lnet_peer_is_multi_rail(lp);
3709 lnet_peer_decref_locked(lp);
3711 mutex_unlock(&the_lnet.ln_api_mutex);
3713 discover->ping_count = rc;
3718 ni = lnet_net2ni_addref(data->ioc_net);
3722 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
3725 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
3732 EXPORT_SYMBOL(LNetCtl);
3734 void LNetDebugPeer(struct lnet_process_id id)
3736 lnet_debug_peer(id.nid);
3738 EXPORT_SYMBOL(LNetDebugPeer);
3741 * Determine if the specified peer \a nid is on the local node.
3743 * \param nid peer nid to check
3745 * \retval true If peer NID is on the local node.
3746 * \retval false If peer NID is not on the local node.
3748 bool LNetIsPeerLocal(lnet_nid_t nid)
3750 struct lnet_net *net;
3754 cpt = lnet_net_lock_current();
3755 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3756 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3757 if (ni->ni_nid == nid) {
3758 lnet_net_unlock(cpt);
3763 lnet_net_unlock(cpt);
3767 EXPORT_SYMBOL(LNetIsPeerLocal);
3770 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
3771 * Note that all interfaces share a same PID, as requested by LNetNIInit().
3773 * \param index Index of the interface to look up.
3774 * \param id On successful return, this location will hold the
3775 * struct lnet_process_id ID of the interface.
3777 * \retval 0 If an interface exists at \a index.
3778 * \retval -ENOENT If no interface has been found.
3781 LNetGetId(unsigned int index, struct lnet_process_id *id)
3784 struct lnet_net *net;
3788 LASSERT(the_lnet.ln_refcount > 0);
3790 cpt = lnet_net_lock_current();
3792 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3793 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3797 id->nid = ni->ni_nid;
3798 id->pid = the_lnet.ln_pid;
3804 lnet_net_unlock(cpt);
3807 EXPORT_SYMBOL(LNetGetId);
3809 static int lnet_ping(struct lnet_process_id id, signed long timeout,
3810 struct lnet_process_id __user *ids, int n_ids)
3812 struct lnet_handle_eq eqh;
3813 struct lnet_handle_md mdh;
3814 struct lnet_event event;
3815 struct lnet_md md = { NULL };
3819 const signed long a_long_time = msecs_to_jiffies(60 * MSEC_PER_SEC);
3820 struct lnet_ping_buffer *pbuf;
3821 struct lnet_process_id tmpid;
3828 /* n_ids limit is arbitrary */
3829 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
3833 * if the user buffer has more space than the lnet_interfaces_max
3834 * then only fill it up to lnet_interfaces_max
3836 if (n_ids > lnet_interfaces_max)
3837 n_ids = lnet_interfaces_max;
3839 if (id.pid == LNET_PID_ANY)
3840 id.pid = LNET_PID_LUSTRE;
3842 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
3846 /* NB 2 events max (including any unlink event) */
3847 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
3849 CERROR("Can't allocate EQ: %d\n", rc);
3850 goto fail_ping_buffer_decref;
3853 /* initialize md content */
3854 md.start = &pbuf->pb_info;
3855 md.length = LNET_PING_INFO_SIZE(n_ids);
3856 md.threshold = 2; /* GET/REPLY */
3858 md.options = LNET_MD_TRUNCATE;
3862 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
3864 CERROR("Can't bind MD: %d\n", rc);
3868 rc = LNetGet(LNET_NID_ANY, mdh, id,
3869 LNET_RESERVED_PORTAL,
3870 LNET_PROTO_PING_MATCHBITS, 0, false);
3873 /* Don't CERROR; this could be deliberate! */
3874 rc2 = LNetMDUnlink(mdh);
3877 /* NB must wait for the UNLINK event below... */
3879 timeout = a_long_time;
3883 /* MUST block for unlink to complete */
3885 blocked = cfs_block_allsigs();
3887 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
3890 cfs_restore_sigs(blocked);
3892 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
3893 (rc2 <= 0) ? -1 : event.type,
3894 (rc2 <= 0) ? -1 : event.status,
3895 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
3897 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
3899 if (rc2 <= 0 || event.status != 0) {
3900 /* timeout or error */
3901 if (!replied && rc == 0)
3902 rc = (rc2 < 0) ? rc2 :
3903 (rc2 == 0) ? -ETIMEDOUT :
3907 /* Ensure completion in finite time... */
3909 /* No assertion (racing with network) */
3911 timeout = a_long_time;
3912 } else if (rc2 == 0) {
3913 /* timed out waiting for unlink */
3914 CWARN("ping %s: late network completion\n",
3917 } else if (event.type == LNET_EVENT_REPLY) {
3921 } while (rc2 <= 0 || !event.unlinked);
3925 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
3932 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
3934 rc = -EPROTO; /* if I can't parse... */
3937 CERROR("%s: ping info too short %d\n",
3938 libcfs_id2str(id), nob);
3942 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
3943 lnet_swap_pinginfo(pbuf);
3944 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
3945 CERROR("%s: Unexpected magic %08x\n",
3946 libcfs_id2str(id), pbuf->pb_info.pi_magic);
3950 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
3951 CERROR("%s: ping w/o NI status: 0x%x\n",
3952 libcfs_id2str(id), pbuf->pb_info.pi_features);
3956 if (nob < LNET_PING_INFO_SIZE(0)) {
3957 CERROR("%s: Short reply %d(%d min)\n",
3959 nob, (int)LNET_PING_INFO_SIZE(0));
3963 if (pbuf->pb_info.pi_nnis < n_ids)
3964 n_ids = pbuf->pb_info.pi_nnis;
3966 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
3967 CERROR("%s: Short reply %d(%d expected)\n",
3969 nob, (int)LNET_PING_INFO_SIZE(n_ids));
3973 rc = -EFAULT; /* if I segv in copy_to_user()... */
3975 memset(&tmpid, 0, sizeof(tmpid));
3976 for (i = 0; i < n_ids; i++) {
3977 tmpid.pid = pbuf->pb_info.pi_pid;
3978 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
3979 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
3982 rc = pbuf->pb_info.pi_nnis;
3985 rc2 = LNetEQFree(eqh);
3987 CERROR("rc2 %d\n", rc2);
3990 fail_ping_buffer_decref:
3991 lnet_ping_buffer_decref(pbuf);
3996 lnet_discover(struct lnet_process_id id, __u32 force,
3997 struct lnet_process_id __user *ids, int n_ids)
3999 struct lnet_peer_ni *lpni;
4000 struct lnet_peer_ni *p;
4001 struct lnet_peer *lp;
4002 struct lnet_process_id *buf;
4006 int max_intf = lnet_interfaces_max;
4010 id.nid == LNET_NID_ANY)
4013 if (id.pid == LNET_PID_ANY)
4014 id.pid = LNET_PID_LUSTRE;
4017 * if the user buffer has more space than the max_intf
4018 * then only fill it up to max_intf
4020 if (n_ids > max_intf)
4023 buf_size = n_ids * sizeof(*buf);
4025 LIBCFS_ALLOC(buf, buf_size);
4029 cpt = lnet_net_lock_current();
4030 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4037 * Clearing the NIDS_UPTODATE flag ensures the peer will
4038 * be discovered, provided discovery has not been disabled.
4040 lp = lpni->lpni_peer_net->lpn_peer;
4041 spin_lock(&lp->lp_lock);
4042 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4043 /* If the force flag is set, force a PING and PUSH as well. */
4045 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4046 spin_unlock(&lp->lp_lock);
4047 rc = lnet_discover_peer_locked(lpni, cpt, true);
4051 /* Peer may have changed. */
4052 lp = lpni->lpni_peer_net->lpn_peer;
4053 if (lp->lp_nnis < n_ids)
4054 n_ids = lp->lp_nnis;
4058 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4059 buf[i].pid = id.pid;
4060 buf[i].nid = p->lpni_nid;
4065 lnet_net_unlock(cpt);
4068 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4074 lnet_peer_ni_decref_locked(lpni);
4076 lnet_net_unlock(cpt);
4078 LIBCFS_FREE(buf, buf_size);