Whamcloud - gitweb
d94f9f385cea6f77884e823d2b455519884f0c71
[fs/lustre-release.git] / lnet / lnet / api-ni.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  */
32
33 #define DEBUG_SUBSYSTEM S_LNET
34
35 #include <linux/ctype.h>
36 #include <linux/log2.h>
37 #include <linux/ktime.h>
38 #include <linux/moduleparam.h>
39 #include <linux/uaccess.h>
40
41 #include <lnet/lib-lnet.h>
42
43 #define D_LNI D_CONSOLE
44
45 /*
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.
49  */
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);
54
55 static char *ip2nets = "";
56 module_param(ip2nets, charp, 0444);
57 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
58
59 static char *networks = "";
60 module_param(networks, charp, 0444);
61 MODULE_PARM_DESC(networks, "local networks");
62
63 static char *routes = "";
64 module_param(routes, charp, 0444);
65 MODULE_PARM_DESC(routes, "routes to non-local networks");
66
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");
70
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");
75
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");
80
81 /*
82  * lnet_health_sensitivity determines by how much we decrement the health
83  * value on sending error. The value defaults to 100, which means health
84  * interface health is decremented by 100 points every failure.
85  */
86 unsigned int lnet_health_sensitivity = 100;
87 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
88 #ifdef HAVE_KERNEL_PARAM_OPS
89 static struct kernel_param_ops param_ops_health_sensitivity = {
90         .set = sensitivity_set,
91         .get = param_get_int,
92 };
93 #define param_check_health_sensitivity(name, p) \
94                 __param_check(name, p, int)
95 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
96 #else
97 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
98                   &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
99 #endif
100 MODULE_PARM_DESC(lnet_health_sensitivity,
101                 "Value to decrement the health value by on error");
102
103 /*
104  * lnet_recovery_interval determines how often we should perform recovery
105  * on unhealthy interfaces.
106  */
107 unsigned int lnet_recovery_interval = 1;
108 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
109 #ifdef HAVE_KERNEL_PARAM_OPS
110 static struct kernel_param_ops param_ops_recovery_interval = {
111         .set = recovery_interval_set,
112         .get = param_get_int,
113 };
114 #define param_check_recovery_interval(name, p) \
115                 __param_check(name, p, int)
116 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
117 #else
118 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
119                   &lnet_recovery_interval, S_IRUGO|S_IWUSR);
120 #endif
121 MODULE_PARM_DESC(lnet_recovery_interval,
122                 "Interval to recover unhealthy interfaces in seconds");
123
124 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
125 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
126
127 static struct kernel_param_ops param_ops_interfaces_max = {
128         .set = intf_max_set,
129         .get = param_get_int,
130 };
131
132 #define param_check_interfaces_max(name, p) \
133                 __param_check(name, p, int)
134
135 #ifdef HAVE_KERNEL_PARAM_OPS
136 module_param(lnet_interfaces_max, interfaces_max, 0644);
137 #else
138 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
139                   &param_ops_interfaces_max, 0644);
140 #endif
141 MODULE_PARM_DESC(lnet_interfaces_max,
142                 "Maximum number of interfaces in a node.");
143
144 unsigned lnet_peer_discovery_disabled = 0;
145 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
146
147 static struct kernel_param_ops param_ops_discovery_disabled = {
148         .set = discovery_set,
149         .get = param_get_int,
150 };
151
152 #define param_check_discovery_disabled(name, p) \
153                 __param_check(name, p, int)
154 #ifdef HAVE_KERNEL_PARAM_OPS
155 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
156 #else
157 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
158                   &param_ops_discovery_disabled, 0644);
159 #endif
160 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
161                 "Set to 1 to disable peer discovery on this node.");
162
163 unsigned int lnet_drop_asym_route;
164 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
165
166 static struct kernel_param_ops param_ops_drop_asym_route = {
167         .set = drop_asym_route_set,
168         .get = param_get_int,
169 };
170
171 #define param_check_drop_asym_route(name, p)    \
172         __param_check(name, p, int)
173 #ifdef HAVE_KERNEL_PARAM_OPS
174 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
175 #else
176 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
177                   &param_ops_drop_asym_route, 0644);
178 #endif
179 MODULE_PARM_DESC(lnet_drop_asym_route,
180                  "Set to 1 to drop asymmetrical route messages.");
181
182 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
183 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 10
184
185 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
186 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
187 #ifdef HAVE_KERNEL_PARAM_OPS
188 static struct kernel_param_ops param_ops_transaction_timeout = {
189         .set = transaction_to_set,
190         .get = param_get_int,
191 };
192
193 #define param_check_transaction_timeout(name, p) \
194                 __param_check(name, p, int)
195 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
196 #else
197 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
198                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
199 #endif
200 MODULE_PARM_DESC(lnet_transaction_timeout,
201                 "Maximum number of seconds to wait for a peer response.");
202
203 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 3
204 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
205 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
206 #ifdef HAVE_KERNEL_PARAM_OPS
207 static struct kernel_param_ops param_ops_retry_count = {
208         .set = retry_count_set,
209         .get = param_get_int,
210 };
211
212 #define param_check_retry_count(name, p) \
213                 __param_check(name, p, int)
214 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
215 #else
216 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
217                   &lnet_retry_count, S_IRUGO|S_IWUSR);
218 #endif
219 MODULE_PARM_DESC(lnet_retry_count,
220                  "Maximum number of times to retry transmitting a message");
221
222
223 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
224 unsigned int lnet_current_net_count;
225
226 /*
227  * This sequence number keeps track of how many times DLC was used to
228  * update the local NIs. It is incremented when a NI is added or
229  * removed and checked when sending a message to determine if there is
230  * a need to re-run the selection algorithm. See lnet_select_pathway()
231  * for more details on its usage.
232  */
233 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
234
235 static int lnet_ping(struct lnet_process_id id, signed long timeout,
236                      struct lnet_process_id __user *ids, int n_ids);
237
238 static int lnet_discover(struct lnet_process_id id, __u32 force,
239                          struct lnet_process_id __user *ids, int n_ids);
240
241 static int
242 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
243 {
244         int rc;
245         unsigned *sensitivity = (unsigned *)kp->arg;
246         unsigned long value;
247
248         rc = kstrtoul(val, 0, &value);
249         if (rc) {
250                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
251                 return rc;
252         }
253
254         /*
255          * The purpose of locking the api_mutex here is to ensure that
256          * the correct value ends up stored properly.
257          */
258         mutex_lock(&the_lnet.ln_api_mutex);
259
260         if (value > LNET_MAX_HEALTH_VALUE) {
261                 mutex_unlock(&the_lnet.ln_api_mutex);
262                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
263                        LNET_MAX_HEALTH_VALUE, value);
264                 return -EINVAL;
265         }
266
267         /*
268          * if we're turning on health then use the health timeout
269          * defaults.
270          */
271         if (*sensitivity == 0 && value != 0) {
272                 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
273                 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
274         /*
275          * if we're turning off health then use the no health timeout
276          * default.
277          */
278         } else if (*sensitivity != 0 && value == 0) {
279                 lnet_transaction_timeout =
280                         LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
281                 lnet_retry_count = 0;
282         }
283
284         *sensitivity = value;
285
286         mutex_unlock(&the_lnet.ln_api_mutex);
287
288         return 0;
289 }
290
291 static int
292 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
293 {
294         int rc;
295         unsigned *interval = (unsigned *)kp->arg;
296         unsigned long value;
297
298         rc = kstrtoul(val, 0, &value);
299         if (rc) {
300                 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
301                 return rc;
302         }
303
304         if (value < 1) {
305                 CERROR("lnet_recovery_interval must be at least 1 second\n");
306                 return -EINVAL;
307         }
308
309         /*
310          * The purpose of locking the api_mutex here is to ensure that
311          * the correct value ends up stored properly.
312          */
313         mutex_lock(&the_lnet.ln_api_mutex);
314
315         *interval = value;
316
317         mutex_unlock(&the_lnet.ln_api_mutex);
318
319         return 0;
320 }
321
322 static int
323 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
324 {
325         int rc;
326         unsigned *discovery = (unsigned *)kp->arg;
327         unsigned long value;
328         struct lnet_ping_buffer *pbuf;
329
330         rc = kstrtoul(val, 0, &value);
331         if (rc) {
332                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
333                 return rc;
334         }
335
336         value = (value) ? 1 : 0;
337
338         /*
339          * The purpose of locking the api_mutex here is to ensure that
340          * the correct value ends up stored properly.
341          */
342         mutex_lock(&the_lnet.ln_api_mutex);
343
344         if (value == *discovery) {
345                 mutex_unlock(&the_lnet.ln_api_mutex);
346                 return 0;
347         }
348
349         *discovery = value;
350
351         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
352                 mutex_unlock(&the_lnet.ln_api_mutex);
353                 return 0;
354         }
355
356         /* tell peers that discovery setting has changed */
357         lnet_net_lock(LNET_LOCK_EX);
358         pbuf = the_lnet.ln_ping_target;
359         if (value)
360                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
361         else
362                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
363         lnet_net_unlock(LNET_LOCK_EX);
364
365         lnet_push_update_to_peers(1);
366
367         mutex_unlock(&the_lnet.ln_api_mutex);
368
369         return 0;
370 }
371
372 static int
373 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
374 {
375         int rc;
376         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
377         unsigned long value;
378
379         rc = kstrtoul(val, 0, &value);
380         if (rc) {
381                 CERROR("Invalid module parameter value for "
382                        "'lnet_drop_asym_route'\n");
383                 return rc;
384         }
385
386         /*
387          * The purpose of locking the api_mutex here is to ensure that
388          * the correct value ends up stored properly.
389          */
390         mutex_lock(&the_lnet.ln_api_mutex);
391
392         if (value == *drop_asym_route) {
393                 mutex_unlock(&the_lnet.ln_api_mutex);
394                 return 0;
395         }
396
397         *drop_asym_route = value;
398
399         mutex_unlock(&the_lnet.ln_api_mutex);
400
401         return 0;
402 }
403
404 static int
405 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
406 {
407         int rc;
408         unsigned *transaction_to = (unsigned *)kp->arg;
409         unsigned long value;
410
411         rc = kstrtoul(val, 0, &value);
412         if (rc) {
413                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
414                 return rc;
415         }
416
417         /*
418          * The purpose of locking the api_mutex here is to ensure that
419          * the correct value ends up stored properly.
420          */
421         mutex_lock(&the_lnet.ln_api_mutex);
422
423         if (value < lnet_retry_count || value == 0) {
424                 mutex_unlock(&the_lnet.ln_api_mutex);
425                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
426                        "Has to be greater than lnet_retry_count (%u)\n",
427                        value, lnet_retry_count);
428                 return -EINVAL;
429         }
430
431         if (value == *transaction_to) {
432                 mutex_unlock(&the_lnet.ln_api_mutex);
433                 return 0;
434         }
435
436         *transaction_to = value;
437         if (lnet_retry_count == 0)
438                 lnet_lnd_timeout = value;
439         else
440                 lnet_lnd_timeout = value / lnet_retry_count;
441
442         mutex_unlock(&the_lnet.ln_api_mutex);
443
444         return 0;
445 }
446
447 static int
448 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
449 {
450         int rc;
451         unsigned *retry_count = (unsigned *)kp->arg;
452         unsigned long value;
453
454         rc = kstrtoul(val, 0, &value);
455         if (rc) {
456                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
457                 return rc;
458         }
459
460         /*
461          * The purpose of locking the api_mutex here is to ensure that
462          * the correct value ends up stored properly.
463          */
464         mutex_lock(&the_lnet.ln_api_mutex);
465
466         if (lnet_health_sensitivity == 0) {
467                 mutex_unlock(&the_lnet.ln_api_mutex);
468                 CERROR("Can not set retry_count when health feature is turned off\n");
469                 return -EINVAL;
470         }
471
472         if (value > lnet_transaction_timeout) {
473                 mutex_unlock(&the_lnet.ln_api_mutex);
474                 CERROR("Invalid value for lnet_retry_count (%lu). "
475                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
476                        value, lnet_transaction_timeout);
477                 return -EINVAL;
478         }
479
480         *retry_count = value;
481
482         if (value == 0)
483                 lnet_lnd_timeout = lnet_transaction_timeout;
484         else
485                 lnet_lnd_timeout = lnet_transaction_timeout / value;
486
487         mutex_unlock(&the_lnet.ln_api_mutex);
488
489         return 0;
490 }
491
492 static int
493 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
494 {
495         int value, rc;
496
497         rc = kstrtoint(val, 0, &value);
498         if (rc) {
499                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
500                 return rc;
501         }
502
503         if (value < LNET_INTERFACES_MIN) {
504                 CWARN("max interfaces provided are too small, setting to %d\n",
505                       LNET_INTERFACES_MAX_DEFAULT);
506                 value = LNET_INTERFACES_MAX_DEFAULT;
507         }
508
509         *(int *)kp->arg = value;
510
511         return 0;
512 }
513
514 static char *
515 lnet_get_routes(void)
516 {
517         return routes;
518 }
519
520 static char *
521 lnet_get_networks(void)
522 {
523         char   *nets;
524         int     rc;
525
526         if (*networks != 0 && *ip2nets != 0) {
527                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
528                                    "'ip2nets' but not both at once\n");
529                 return NULL;
530         }
531
532         if (*ip2nets != 0) {
533                 rc = lnet_parse_ip2nets(&nets, ip2nets);
534                 return (rc == 0) ? nets : NULL;
535         }
536
537         if (*networks != 0)
538                 return networks;
539
540         return "tcp";
541 }
542
543 static void
544 lnet_init_locks(void)
545 {
546         spin_lock_init(&the_lnet.ln_eq_wait_lock);
547         spin_lock_init(&the_lnet.ln_msg_resend_lock);
548         init_waitqueue_head(&the_lnet.ln_eq_waitq);
549         init_waitqueue_head(&the_lnet.ln_mt_waitq);
550         mutex_init(&the_lnet.ln_lnd_mutex);
551 }
552
553 static void
554 lnet_fini_locks(void)
555 {
556 }
557
558 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
559 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
560                                             *  MDs kmem_cache */
561
562 static int
563 lnet_descriptor_setup(void)
564 {
565         /* create specific kmem_cache for MEs and small MDs (i.e., originally
566          * allocated in <size-xxx> kmem_cache).
567          */
568         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
569                                             0, 0, NULL);
570         if (!lnet_mes_cachep)
571                 return -ENOMEM;
572
573         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
574                                                   LNET_SMALL_MD_SIZE, 0, 0,
575                                                   NULL);
576         if (!lnet_small_mds_cachep)
577                 return -ENOMEM;
578
579         return 0;
580 }
581
582 static void
583 lnet_descriptor_cleanup(void)
584 {
585
586         if (lnet_small_mds_cachep) {
587                 kmem_cache_destroy(lnet_small_mds_cachep);
588                 lnet_small_mds_cachep = NULL;
589         }
590
591         if (lnet_mes_cachep) {
592                 kmem_cache_destroy(lnet_mes_cachep);
593                 lnet_mes_cachep = NULL;
594         }
595 }
596
597 static int
598 lnet_create_remote_nets_table(void)
599 {
600         int               i;
601         struct list_head *hash;
602
603         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
604         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
605         LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
606         if (hash == NULL) {
607                 CERROR("Failed to create remote nets hash table\n");
608                 return -ENOMEM;
609         }
610
611         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
612                 INIT_LIST_HEAD(&hash[i]);
613         the_lnet.ln_remote_nets_hash = hash;
614         return 0;
615 }
616
617 static void
618 lnet_destroy_remote_nets_table(void)
619 {
620         int i;
621
622         if (the_lnet.ln_remote_nets_hash == NULL)
623                 return;
624
625         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
626                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
627
628         LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
629                     LNET_REMOTE_NETS_HASH_SIZE *
630                     sizeof(the_lnet.ln_remote_nets_hash[0]));
631         the_lnet.ln_remote_nets_hash = NULL;
632 }
633
634 static void
635 lnet_destroy_locks(void)
636 {
637         if (the_lnet.ln_res_lock != NULL) {
638                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
639                 the_lnet.ln_res_lock = NULL;
640         }
641
642         if (the_lnet.ln_net_lock != NULL) {
643                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
644                 the_lnet.ln_net_lock = NULL;
645         }
646
647         lnet_fini_locks();
648 }
649
650 static int
651 lnet_create_locks(void)
652 {
653         lnet_init_locks();
654
655         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
656         if (the_lnet.ln_res_lock == NULL)
657                 goto failed;
658
659         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
660         if (the_lnet.ln_net_lock == NULL)
661                 goto failed;
662
663         return 0;
664
665  failed:
666         lnet_destroy_locks();
667         return -ENOMEM;
668 }
669
670 static void lnet_assert_wire_constants(void)
671 {
672         /* Wire protocol assertions generated by 'wirecheck'
673          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
674          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
675          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
676
677         /* Constants... */
678         CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
679         CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
680         CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
681         CLASSERT(LNET_MSG_ACK == 0);
682         CLASSERT(LNET_MSG_PUT == 1);
683         CLASSERT(LNET_MSG_GET == 2);
684         CLASSERT(LNET_MSG_REPLY == 3);
685         CLASSERT(LNET_MSG_HELLO == 4);
686
687         /* Checks for struct lnet_handle_wire */
688         CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
689         CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
690         CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
691         CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
692         CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
693
694         /* Checks for struct struct lnet_magicversion */
695         CLASSERT((int)sizeof(struct lnet_magicversion) == 8);
696         CLASSERT((int)offsetof(struct lnet_magicversion, magic) == 0);
697         CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->magic) == 4);
698         CLASSERT((int)offsetof(struct lnet_magicversion, version_major) == 4);
699         CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_major) == 2);
700         CLASSERT((int)offsetof(struct lnet_magicversion, version_minor) == 6);
701         CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_minor) == 2);
702
703         /* Checks for struct struct lnet_hdr */
704         CLASSERT((int)sizeof(struct lnet_hdr) == 72);
705         CLASSERT((int)offsetof(struct lnet_hdr, dest_nid) == 0);
706         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_nid) == 8);
707         CLASSERT((int)offsetof(struct lnet_hdr, src_nid) == 8);
708         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_nid) == 8);
709         CLASSERT((int)offsetof(struct lnet_hdr, dest_pid) == 16);
710         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_pid) == 4);
711         CLASSERT((int)offsetof(struct lnet_hdr, src_pid) == 20);
712         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_pid) == 4);
713         CLASSERT((int)offsetof(struct lnet_hdr, type) == 24);
714         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->type) == 4);
715         CLASSERT((int)offsetof(struct lnet_hdr, payload_length) == 28);
716         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->payload_length) == 4);
717         CLASSERT((int)offsetof(struct lnet_hdr, msg) == 32);
718         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg) == 40);
719
720         /* Ack */
721         CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) == 32);
722         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) == 16);
723         CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.match_bits) == 48);
724         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) == 8);
725         CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.mlength) == 56);
726         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) == 4);
727
728         /* Put */
729         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) == 32);
730         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) == 16);
731         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.match_bits) == 48);
732         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) == 8);
733         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.hdr_data) == 56);
734         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) == 8);
735         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ptl_index) == 64);
736         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) == 4);
737         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.offset) == 68);
738         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) == 4);
739
740         /* Get */
741         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.return_wmd) == 32);
742         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) == 16);
743         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.match_bits) == 48);
744         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) == 8);
745         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.ptl_index) == 56);
746         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) == 4);
747         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.src_offset) == 60);
748         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) == 4);
749         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.sink_length) == 64);
750         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) == 4);
751
752         /* Reply */
753         CLASSERT((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) == 32);
754         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) == 16);
755
756         /* Hello */
757         CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.incarnation) == 32);
758         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) == 8);
759         CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.type) == 40);
760         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) == 4);
761
762         /* Checks for struct lnet_ni_status and related constants */
763         CLASSERT(LNET_NI_STATUS_INVALID == 0x00000000);
764         CLASSERT(LNET_NI_STATUS_UP == 0x15aac0de);
765         CLASSERT(LNET_NI_STATUS_DOWN == 0xdeadface);
766
767         /* Checks for struct lnet_ni_status */
768         CLASSERT((int)sizeof(struct lnet_ni_status) == 16);
769         CLASSERT((int)offsetof(struct lnet_ni_status, ns_nid) == 0);
770         CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) == 8);
771         CLASSERT((int)offsetof(struct lnet_ni_status, ns_status) == 8);
772         CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_status) == 4);
773         CLASSERT((int)offsetof(struct lnet_ni_status, ns_unused) == 12);
774         CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) == 4);
775
776         /* Checks for struct lnet_ping_info and related constants */
777         CLASSERT(LNET_PROTO_PING_MAGIC == 0x70696E67);
778         CLASSERT(LNET_PING_FEAT_INVAL == 0);
779         CLASSERT(LNET_PING_FEAT_BASE == 1);
780         CLASSERT(LNET_PING_FEAT_NI_STATUS == 2);
781         CLASSERT(LNET_PING_FEAT_RTE_DISABLED == 4);
782         CLASSERT(LNET_PING_FEAT_MULTI_RAIL == 8);
783         CLASSERT(LNET_PING_FEAT_DISCOVERY == 16);
784         CLASSERT(LNET_PING_FEAT_BITS == 31);
785
786         /* Checks for struct lnet_ping_info */
787         CLASSERT((int)sizeof(struct lnet_ping_info) == 16);
788         CLASSERT((int)offsetof(struct lnet_ping_info, pi_magic) == 0);
789         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) == 4);
790         CLASSERT((int)offsetof(struct lnet_ping_info, pi_features) == 4);
791         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_features) == 4);
792         CLASSERT((int)offsetof(struct lnet_ping_info, pi_pid) == 8);
793         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) == 4);
794         CLASSERT((int)offsetof(struct lnet_ping_info, pi_nnis) == 12);
795         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) == 4);
796         CLASSERT((int)offsetof(struct lnet_ping_info, pi_ni) == 16);
797         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) == 0);
798 }
799
800 static struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
801 {
802         struct lnet_lnd *lnd;
803         struct list_head *tmp;
804
805         /* holding lnd mutex */
806         list_for_each(tmp, &the_lnet.ln_lnds) {
807                 lnd = list_entry(tmp, struct lnet_lnd, lnd_list);
808
809                 if (lnd->lnd_type == type)
810                         return lnd;
811         }
812         return NULL;
813 }
814
815 unsigned int
816 lnet_get_lnd_timeout(void)
817 {
818         return lnet_lnd_timeout;
819 }
820 EXPORT_SYMBOL(lnet_get_lnd_timeout);
821
822 void
823 lnet_register_lnd(struct lnet_lnd *lnd)
824 {
825         mutex_lock(&the_lnet.ln_lnd_mutex);
826
827         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
828         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
829
830         list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
831         lnd->lnd_refcount = 0;
832
833         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
834
835         mutex_unlock(&the_lnet.ln_lnd_mutex);
836 }
837 EXPORT_SYMBOL(lnet_register_lnd);
838
839 void
840 lnet_unregister_lnd(struct lnet_lnd *lnd)
841 {
842         mutex_lock(&the_lnet.ln_lnd_mutex);
843
844         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
845         LASSERT(lnd->lnd_refcount == 0);
846
847         list_del(&lnd->lnd_list);
848         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
849
850         mutex_unlock(&the_lnet.ln_lnd_mutex);
851 }
852 EXPORT_SYMBOL(lnet_unregister_lnd);
853
854 void
855 lnet_counters_get_common(struct lnet_counters_common *common)
856 {
857         struct lnet_counters *ctr;
858         int i;
859
860         memset(common, 0, sizeof(*common));
861
862         lnet_net_lock(LNET_LOCK_EX);
863
864         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
865                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
866                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
867                 common->lcc_errors       += ctr->lct_common.lcc_errors;
868                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
869                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
870                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
871                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
872                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
873                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
874                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
875                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
876         }
877         lnet_net_unlock(LNET_LOCK_EX);
878 }
879 EXPORT_SYMBOL(lnet_counters_get_common);
880
881 void
882 lnet_counters_get(struct lnet_counters *counters)
883 {
884         struct lnet_counters *ctr;
885         struct lnet_counters_health *health = &counters->lct_health;
886         int             i;
887
888         memset(counters, 0, sizeof(*counters));
889
890         lnet_counters_get_common(&counters->lct_common);
891
892         lnet_net_lock(LNET_LOCK_EX);
893
894         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
895                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
896                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
897                 health->lch_response_timeout_count +=
898                                 ctr->lct_health.lch_response_timeout_count;
899                 health->lch_local_interrupt_count +=
900                                 ctr->lct_health.lch_local_interrupt_count;
901                 health->lch_local_dropped_count +=
902                                 ctr->lct_health.lch_local_dropped_count;
903                 health->lch_local_aborted_count +=
904                                 ctr->lct_health.lch_local_aborted_count;
905                 health->lch_local_no_route_count +=
906                                 ctr->lct_health.lch_local_no_route_count;
907                 health->lch_local_timeout_count +=
908                                 ctr->lct_health.lch_local_timeout_count;
909                 health->lch_local_error_count +=
910                                 ctr->lct_health.lch_local_error_count;
911                 health->lch_remote_dropped_count +=
912                                 ctr->lct_health.lch_remote_dropped_count;
913                 health->lch_remote_error_count +=
914                                 ctr->lct_health.lch_remote_error_count;
915                 health->lch_remote_timeout_count +=
916                                 ctr->lct_health.lch_remote_timeout_count;
917                 health->lch_network_timeout_count +=
918                                 ctr->lct_health.lch_network_timeout_count;
919         }
920         lnet_net_unlock(LNET_LOCK_EX);
921 }
922 EXPORT_SYMBOL(lnet_counters_get);
923
924 void
925 lnet_counters_reset(void)
926 {
927         struct lnet_counters *counters;
928         int             i;
929
930         lnet_net_lock(LNET_LOCK_EX);
931
932         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
933                 memset(counters, 0, sizeof(struct lnet_counters));
934
935         lnet_net_unlock(LNET_LOCK_EX);
936 }
937
938 static char *
939 lnet_res_type2str(int type)
940 {
941         switch (type) {
942         default:
943                 LBUG();
944         case LNET_COOKIE_TYPE_MD:
945                 return "MD";
946         case LNET_COOKIE_TYPE_ME:
947                 return "ME";
948         case LNET_COOKIE_TYPE_EQ:
949                 return "EQ";
950         }
951 }
952
953 static void
954 lnet_res_container_cleanup(struct lnet_res_container *rec)
955 {
956         int     count = 0;
957
958         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
959                 return;
960
961         while (!list_empty(&rec->rec_active)) {
962                 struct list_head *e = rec->rec_active.next;
963
964                 list_del_init(e);
965                 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
966                         lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
967
968                 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
969                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
970
971                 } else { /* NB: Active MEs should be attached on portals */
972                         LBUG();
973                 }
974                 count++;
975         }
976
977         if (count > 0) {
978                 /* Found alive MD/ME/EQ, user really should unlink/free
979                  * all of them before finalize LNet, but if someone didn't,
980                  * we have to recycle garbage for him */
981                 CERROR("%d active elements on exit of %s container\n",
982                        count, lnet_res_type2str(rec->rec_type));
983         }
984
985         if (rec->rec_lh_hash != NULL) {
986                 LIBCFS_FREE(rec->rec_lh_hash,
987                             LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
988                 rec->rec_lh_hash = NULL;
989         }
990
991         rec->rec_type = 0; /* mark it as finalized */
992 }
993
994 static int
995 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
996 {
997         int     rc = 0;
998         int     i;
999
1000         LASSERT(rec->rec_type == 0);
1001
1002         rec->rec_type = type;
1003         INIT_LIST_HEAD(&rec->rec_active);
1004
1005         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1006
1007         /* Arbitrary choice of hash table size */
1008         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1009                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1010         if (rec->rec_lh_hash == NULL) {
1011                 rc = -ENOMEM;
1012                 goto out;
1013         }
1014
1015         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1016                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1017
1018         return 0;
1019
1020 out:
1021         CERROR("Failed to setup %s resource container\n",
1022                lnet_res_type2str(type));
1023         lnet_res_container_cleanup(rec);
1024         return rc;
1025 }
1026
1027 static void
1028 lnet_res_containers_destroy(struct lnet_res_container **recs)
1029 {
1030         struct lnet_res_container       *rec;
1031         int                             i;
1032
1033         cfs_percpt_for_each(rec, i, recs)
1034                 lnet_res_container_cleanup(rec);
1035
1036         cfs_percpt_free(recs);
1037 }
1038
1039 static struct lnet_res_container **
1040 lnet_res_containers_create(int type)
1041 {
1042         struct lnet_res_container       **recs;
1043         struct lnet_res_container       *rec;
1044         int                             rc;
1045         int                             i;
1046
1047         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1048         if (recs == NULL) {
1049                 CERROR("Failed to allocate %s resource containers\n",
1050                        lnet_res_type2str(type));
1051                 return NULL;
1052         }
1053
1054         cfs_percpt_for_each(rec, i, recs) {
1055                 rc = lnet_res_container_setup(rec, i, type);
1056                 if (rc != 0) {
1057                         lnet_res_containers_destroy(recs);
1058                         return NULL;
1059                 }
1060         }
1061
1062         return recs;
1063 }
1064
1065 struct lnet_libhandle *
1066 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1067 {
1068         /* ALWAYS called with lnet_res_lock held */
1069         struct list_head        *head;
1070         struct lnet_libhandle   *lh;
1071         unsigned int            hash;
1072
1073         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1074                 return NULL;
1075
1076         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1077         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1078
1079         list_for_each_entry(lh, head, lh_hash_chain) {
1080                 if (lh->lh_cookie == cookie)
1081                         return lh;
1082         }
1083
1084         return NULL;
1085 }
1086
1087 void
1088 lnet_res_lh_initialize(struct lnet_res_container *rec,
1089                        struct lnet_libhandle *lh)
1090 {
1091         /* ALWAYS called with lnet_res_lock held */
1092         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1093         unsigned int    hash;
1094
1095         lh->lh_cookie = rec->rec_lh_cookie;
1096         rec->rec_lh_cookie += 1 << ibits;
1097
1098         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1099
1100         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1101 }
1102
1103 static int lnet_unprepare(void);
1104
1105 static int
1106 lnet_prepare(lnet_pid_t requested_pid)
1107 {
1108         /* Prepare to bring up the network */
1109         struct lnet_res_container **recs;
1110         int                       rc = 0;
1111
1112         if (requested_pid == LNET_PID_ANY) {
1113                 /* Don't instantiate LNET just for me */
1114                 return -ENETDOWN;
1115         }
1116
1117         LASSERT(the_lnet.ln_refcount == 0);
1118
1119         the_lnet.ln_routing = 0;
1120
1121         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1122         the_lnet.ln_pid = requested_pid;
1123
1124         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1125         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1126         INIT_LIST_HEAD(&the_lnet.ln_nets);
1127         INIT_LIST_HEAD(&the_lnet.ln_routers);
1128         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1129         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1130         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1131         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1132         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1133         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1134         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1135         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1136         LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1137         init_completion(&the_lnet.ln_started);
1138
1139         rc = lnet_descriptor_setup();
1140         if (rc != 0)
1141                 goto failed;
1142
1143         rc = lnet_create_remote_nets_table();
1144         if (rc != 0)
1145                 goto failed;
1146
1147         /*
1148          * NB the interface cookie in wire handles guards against delayed
1149          * replies and ACKs appearing valid after reboot.
1150          */
1151         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1152
1153         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1154                                                 sizeof(struct lnet_counters));
1155         if (the_lnet.ln_counters == NULL) {
1156                 CERROR("Failed to allocate counters for LNet\n");
1157                 rc = -ENOMEM;
1158                 goto failed;
1159         }
1160
1161         rc = lnet_peer_tables_create();
1162         if (rc != 0)
1163                 goto failed;
1164
1165         rc = lnet_msg_containers_create();
1166         if (rc != 0)
1167                 goto failed;
1168
1169         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1170                                       LNET_COOKIE_TYPE_EQ);
1171         if (rc != 0)
1172                 goto failed;
1173
1174         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
1175         if (recs == NULL) {
1176                 rc = -ENOMEM;
1177                 goto failed;
1178         }
1179
1180         the_lnet.ln_me_containers = recs;
1181
1182         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1183         if (recs == NULL) {
1184                 rc = -ENOMEM;
1185                 goto failed;
1186         }
1187
1188         the_lnet.ln_md_containers = recs;
1189
1190         rc = lnet_portals_create();
1191         if (rc != 0) {
1192                 CERROR("Failed to create portals for LNet: %d\n", rc);
1193                 goto failed;
1194         }
1195
1196         return 0;
1197
1198  failed:
1199         lnet_unprepare();
1200         return rc;
1201 }
1202
1203 static int
1204 lnet_unprepare (void)
1205 {
1206         int rc;
1207
1208         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1209          * have shut down already, so it is safe to unlink and free all
1210          * descriptors, even those that appear committed to a network op (eg MD
1211          * with non-zero pending count) */
1212
1213         lnet_fail_nid(LNET_NID_ANY, 0);
1214
1215         LASSERT(the_lnet.ln_refcount == 0);
1216         LASSERT(list_empty(&the_lnet.ln_test_peers));
1217         LASSERT(list_empty(&the_lnet.ln_nets));
1218
1219         if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1220                 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1221                 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1222                 LASSERT(rc == 0);
1223         }
1224
1225         lnet_portals_destroy();
1226
1227         if (the_lnet.ln_md_containers != NULL) {
1228                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1229                 the_lnet.ln_md_containers = NULL;
1230         }
1231
1232         if (the_lnet.ln_me_containers != NULL) {
1233                 lnet_res_containers_destroy(the_lnet.ln_me_containers);
1234                 the_lnet.ln_me_containers = NULL;
1235         }
1236
1237         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1238
1239         lnet_msg_containers_destroy();
1240         lnet_peer_uninit();
1241         lnet_rtrpools_free(0);
1242
1243         if (the_lnet.ln_counters != NULL) {
1244                 cfs_percpt_free(the_lnet.ln_counters);
1245                 the_lnet.ln_counters = NULL;
1246         }
1247         lnet_destroy_remote_nets_table();
1248         lnet_descriptor_cleanup();
1249
1250         return 0;
1251 }
1252
1253 struct lnet_ni  *
1254 lnet_net2ni_locked(__u32 net_id, int cpt)
1255 {
1256         struct lnet_ni   *ni;
1257         struct lnet_net  *net;
1258
1259         LASSERT(cpt != LNET_LOCK_EX);
1260
1261         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1262                 if (net->net_id == net_id) {
1263                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1264                                         ni_netlist);
1265                         return ni;
1266                 }
1267         }
1268
1269         return NULL;
1270 }
1271
1272 struct lnet_ni *
1273 lnet_net2ni_addref(__u32 net)
1274 {
1275         struct lnet_ni *ni;
1276
1277         lnet_net_lock(0);
1278         ni = lnet_net2ni_locked(net, 0);
1279         if (ni)
1280                 lnet_ni_addref_locked(ni, 0);
1281         lnet_net_unlock(0);
1282
1283         return ni;
1284 }
1285 EXPORT_SYMBOL(lnet_net2ni_addref);
1286
1287 struct lnet_net *
1288 lnet_get_net_locked(__u32 net_id)
1289 {
1290         struct lnet_net  *net;
1291
1292         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1293                 if (net->net_id == net_id)
1294                         return net;
1295         }
1296
1297         return NULL;
1298 }
1299
1300 unsigned int
1301 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1302 {
1303         __u64           key = nid;
1304         unsigned int    val;
1305
1306         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1307
1308         if (number == 1)
1309                 return 0;
1310
1311         val = hash_long(key, LNET_CPT_BITS);
1312         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1313         if (val < number)
1314                 return val;
1315
1316         return (unsigned int)(key + val + (val >> 1)) % number;
1317 }
1318
1319 int
1320 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1321 {
1322         struct lnet_net *net;
1323
1324         /* must called with hold of lnet_net_lock */
1325         if (LNET_CPT_NUMBER == 1)
1326                 return 0; /* the only one */
1327
1328         /*
1329          * If NI is provided then use the CPT identified in the NI cpt
1330          * list if one exists. If one doesn't exist, then that NI is
1331          * associated with all CPTs and it follows that the net it belongs
1332          * to is implicitly associated with all CPTs, so just hash the nid
1333          * and return that.
1334          */
1335         if (ni != NULL) {
1336                 if (ni->ni_cpts != NULL)
1337                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1338                                                              ni->ni_ncpts)];
1339                 else
1340                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1341         }
1342
1343         /* no NI provided so look at the net */
1344         net = lnet_get_net_locked(LNET_NIDNET(nid));
1345
1346         if (net != NULL && net->net_cpts != NULL) {
1347                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1348         }
1349
1350         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1351 }
1352
1353 int
1354 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1355 {
1356         int     cpt;
1357         int     cpt2;
1358
1359         if (LNET_CPT_NUMBER == 1)
1360                 return 0; /* the only one */
1361
1362         cpt = lnet_net_lock_current();
1363
1364         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1365
1366         lnet_net_unlock(cpt);
1367
1368         return cpt2;
1369 }
1370 EXPORT_SYMBOL(lnet_cpt_of_nid);
1371
1372 int
1373 lnet_islocalnet_locked(__u32 net_id)
1374 {
1375         struct lnet_net *net;
1376         bool local;
1377
1378         net = lnet_get_net_locked(net_id);
1379
1380         local = net != NULL;
1381
1382         return local;
1383 }
1384
1385 int
1386 lnet_islocalnet(__u32 net_id)
1387 {
1388         int cpt;
1389         bool local;
1390
1391         cpt = lnet_net_lock_current();
1392
1393         local = lnet_islocalnet_locked(net_id);
1394
1395         lnet_net_unlock(cpt);
1396
1397         return local;
1398 }
1399
1400 struct lnet_ni  *
1401 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1402 {
1403         struct lnet_net  *net;
1404         struct lnet_ni   *ni;
1405
1406         LASSERT(cpt != LNET_LOCK_EX);
1407
1408         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1409                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1410                         if (ni->ni_nid == nid)
1411                                 return ni;
1412                 }
1413         }
1414
1415         return NULL;
1416 }
1417
1418 struct lnet_ni *
1419 lnet_nid2ni_addref(lnet_nid_t nid)
1420 {
1421         struct lnet_ni *ni;
1422
1423         lnet_net_lock(0);
1424         ni = lnet_nid2ni_locked(nid, 0);
1425         if (ni)
1426                 lnet_ni_addref_locked(ni, 0);
1427         lnet_net_unlock(0);
1428
1429         return ni;
1430 }
1431 EXPORT_SYMBOL(lnet_nid2ni_addref);
1432
1433 int
1434 lnet_islocalnid(lnet_nid_t nid)
1435 {
1436         struct lnet_ni  *ni;
1437         int             cpt;
1438
1439         cpt = lnet_net_lock_current();
1440         ni = lnet_nid2ni_locked(nid, cpt);
1441         lnet_net_unlock(cpt);
1442
1443         return ni != NULL;
1444 }
1445
1446 int
1447 lnet_count_acceptor_nets(void)
1448 {
1449         /* Return the # of NIs that need the acceptor. */
1450         int              count = 0;
1451         struct lnet_net  *net;
1452         int              cpt;
1453
1454         cpt = lnet_net_lock_current();
1455         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1456                 /* all socklnd type networks should have the acceptor
1457                  * thread started */
1458                 if (net->net_lnd->lnd_accept != NULL)
1459                         count++;
1460         }
1461
1462         lnet_net_unlock(cpt);
1463
1464         return count;
1465 }
1466
1467 struct lnet_ping_buffer *
1468 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1469 {
1470         struct lnet_ping_buffer *pbuf;
1471
1472         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1473         if (pbuf) {
1474                 pbuf->pb_nnis = nnis;
1475                 atomic_set(&pbuf->pb_refcnt, 1);
1476         }
1477
1478         return pbuf;
1479 }
1480
1481 void
1482 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1483 {
1484         LASSERT(lnet_ping_buffer_numref(pbuf) == 0);
1485         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1486 }
1487
1488 static struct lnet_ping_buffer *
1489 lnet_ping_target_create(int nnis)
1490 {
1491         struct lnet_ping_buffer *pbuf;
1492
1493         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1494         if (pbuf == NULL) {
1495                 CERROR("Can't allocate ping source [%d]\n", nnis);
1496                 return NULL;
1497         }
1498
1499         pbuf->pb_info.pi_nnis = nnis;
1500         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1501         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1502         pbuf->pb_info.pi_features =
1503                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1504
1505         return pbuf;
1506 }
1507
1508 static inline int
1509 lnet_get_net_ni_count_locked(struct lnet_net *net)
1510 {
1511         struct lnet_ni  *ni;
1512         int             count = 0;
1513
1514         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1515                 count++;
1516
1517         return count;
1518 }
1519
1520 static inline int
1521 lnet_get_net_ni_count_pre(struct lnet_net *net)
1522 {
1523         struct lnet_ni  *ni;
1524         int             count = 0;
1525
1526         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1527                 count++;
1528
1529         return count;
1530 }
1531
1532 static inline int
1533 lnet_get_ni_count(void)
1534 {
1535         struct lnet_ni  *ni;
1536         struct lnet_net *net;
1537         int             count = 0;
1538
1539         lnet_net_lock(0);
1540
1541         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1542                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1543                         count++;
1544         }
1545
1546         lnet_net_unlock(0);
1547
1548         return count;
1549 }
1550
1551 int
1552 lnet_get_net_count(void)
1553 {
1554         struct lnet_net *net;
1555         int count = 0;
1556
1557         lnet_net_lock(0);
1558
1559         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1560                 count++;
1561         }
1562
1563         lnet_net_unlock(0);
1564
1565         return count;
1566 }
1567
1568 void
1569 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1570 {
1571         struct lnet_ni_status *stat;
1572         int nnis;
1573         int i;
1574
1575         __swab32s(&pbuf->pb_info.pi_magic);
1576         __swab32s(&pbuf->pb_info.pi_features);
1577         __swab32s(&pbuf->pb_info.pi_pid);
1578         __swab32s(&pbuf->pb_info.pi_nnis);
1579         nnis = pbuf->pb_info.pi_nnis;
1580         if (nnis > pbuf->pb_nnis)
1581                 nnis = pbuf->pb_nnis;
1582         for (i = 0; i < nnis; i++) {
1583                 stat = &pbuf->pb_info.pi_ni[i];
1584                 __swab64s(&stat->ns_nid);
1585                 __swab32s(&stat->ns_status);
1586         }
1587         return;
1588 }
1589
1590 int
1591 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1592 {
1593         if (!pinfo)
1594                 return -EINVAL;
1595         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1596                 return -EPROTO;
1597         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1598                 return -EPROTO;
1599         /* Loopback is guaranteed to be present */
1600         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1601                 return -ERANGE;
1602         if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1603                 return -EPROTO;
1604         return 0;
1605 }
1606
1607 static void
1608 lnet_ping_target_destroy(void)
1609 {
1610         struct lnet_net *net;
1611         struct lnet_ni  *ni;
1612
1613         lnet_net_lock(LNET_LOCK_EX);
1614
1615         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1616                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1617                         lnet_ni_lock(ni);
1618                         ni->ni_status = NULL;
1619                         lnet_ni_unlock(ni);
1620                 }
1621         }
1622
1623         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1624         the_lnet.ln_ping_target = NULL;
1625
1626         lnet_net_unlock(LNET_LOCK_EX);
1627 }
1628
1629 static void
1630 lnet_ping_target_event_handler(struct lnet_event *event)
1631 {
1632         struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1633
1634         if (event->unlinked)
1635                 lnet_ping_buffer_decref(pbuf);
1636 }
1637
1638 static int
1639 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1640                        struct lnet_handle_md *ping_mdh,
1641                        int ni_count, bool set_eq)
1642 {
1643         struct lnet_process_id id = {
1644                 .nid = LNET_NID_ANY,
1645                 .pid = LNET_PID_ANY
1646         };
1647         struct lnet_handle_me me_handle;
1648         struct lnet_md md = { NULL };
1649         int rc, rc2;
1650
1651         if (set_eq) {
1652                 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1653                                  &the_lnet.ln_ping_target_eq);
1654                 if (rc != 0) {
1655                         CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1656                         return rc;
1657                 }
1658         }
1659
1660         *ppbuf = lnet_ping_target_create(ni_count);
1661         if (*ppbuf == NULL) {
1662                 rc = -ENOMEM;
1663                 goto fail_free_eq;
1664         }
1665
1666         /* Ping target ME/MD */
1667         rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1668                           LNET_PROTO_PING_MATCHBITS, 0,
1669                           LNET_UNLINK, LNET_INS_AFTER,
1670                           &me_handle);
1671         if (rc != 0) {
1672                 CERROR("Can't create ping target ME: %d\n", rc);
1673                 goto fail_decref_ping_buffer;
1674         }
1675
1676         /* initialize md content */
1677         md.start     = &(*ppbuf)->pb_info;
1678         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1679         md.threshold = LNET_MD_THRESH_INF;
1680         md.max_size  = 0;
1681         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1682                        LNET_MD_MANAGE_REMOTE;
1683         md.eq_handle = the_lnet.ln_ping_target_eq;
1684         md.user_ptr  = *ppbuf;
1685
1686         rc = LNetMDAttach(me_handle, md, LNET_RETAIN, ping_mdh);
1687         if (rc != 0) {
1688                 CERROR("Can't attach ping target MD: %d\n", rc);
1689                 goto fail_unlink_ping_me;
1690         }
1691         lnet_ping_buffer_addref(*ppbuf);
1692
1693         return 0;
1694
1695 fail_unlink_ping_me:
1696         rc2 = LNetMEUnlink(me_handle);
1697         LASSERT(rc2 == 0);
1698 fail_decref_ping_buffer:
1699         LASSERT(lnet_ping_buffer_numref(*ppbuf) == 1);
1700         lnet_ping_buffer_decref(*ppbuf);
1701         *ppbuf = NULL;
1702 fail_free_eq:
1703         if (set_eq) {
1704                 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1705                 LASSERT(rc2 == 0);
1706         }
1707         return rc;
1708 }
1709
1710 static void
1711 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1712                     struct lnet_handle_md *ping_mdh)
1713 {
1714         sigset_t        blocked = cfs_block_allsigs();
1715
1716         LNetMDUnlink(*ping_mdh);
1717         LNetInvalidateMDHandle(ping_mdh);
1718
1719         /* NB the MD could be busy; this just starts the unlink */
1720         while (lnet_ping_buffer_numref(pbuf) > 1) {
1721                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1722                 set_current_state(TASK_UNINTERRUPTIBLE);
1723                 schedule_timeout(cfs_time_seconds(1));
1724         }
1725
1726         cfs_restore_sigs(blocked);
1727 }
1728
1729 static void
1730 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1731 {
1732         struct lnet_ni          *ni;
1733         struct lnet_net         *net;
1734         struct lnet_ni_status *ns;
1735         int                     i;
1736         int                     rc;
1737
1738         i = 0;
1739         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1740                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1741                         LASSERT(i < pbuf->pb_nnis);
1742
1743                         ns = &pbuf->pb_info.pi_ni[i];
1744
1745                         ns->ns_nid = ni->ni_nid;
1746
1747                         lnet_ni_lock(ni);
1748                         ns->ns_status = (ni->ni_status != NULL) ?
1749                                          ni->ni_status->ns_status :
1750                                                 LNET_NI_STATUS_UP;
1751                         ni->ni_status = ns;
1752                         lnet_ni_unlock(ni);
1753
1754                         i++;
1755                 }
1756         }
1757         /*
1758          * We (ab)use the ns_status of the loopback interface to
1759          * transmit the sequence number. The first interface listed
1760          * must be the loopback interface.
1761          */
1762         rc = lnet_ping_info_validate(&pbuf->pb_info);
1763         if (rc) {
1764                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1765                 LBUG();
1766         }
1767         LNET_PING_BUFFER_SEQNO(pbuf) =
1768                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1769 }
1770
1771 static void
1772 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1773                         struct lnet_handle_md ping_mdh)
1774 {
1775         struct lnet_ping_buffer *old_pbuf = NULL;
1776         struct lnet_handle_md old_ping_md;
1777
1778         /* switch the NIs to point to the new ping info created */
1779         lnet_net_lock(LNET_LOCK_EX);
1780
1781         if (!the_lnet.ln_routing)
1782                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1783         if (!lnet_peer_discovery_disabled)
1784                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1785
1786         /* Ensure only known feature bits have been set. */
1787         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1788         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1789
1790         lnet_ping_target_install_locked(pbuf);
1791
1792         if (the_lnet.ln_ping_target) {
1793                 old_pbuf = the_lnet.ln_ping_target;
1794                 old_ping_md = the_lnet.ln_ping_target_md;
1795         }
1796         the_lnet.ln_ping_target_md = ping_mdh;
1797         the_lnet.ln_ping_target = pbuf;
1798
1799         lnet_net_unlock(LNET_LOCK_EX);
1800
1801         if (old_pbuf) {
1802                 /* unlink and free the old ping info */
1803                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1804                 lnet_ping_buffer_decref(old_pbuf);
1805         }
1806
1807         lnet_push_update_to_peers(0);
1808 }
1809
1810 static void
1811 lnet_ping_target_fini(void)
1812 {
1813         int             rc;
1814
1815         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1816                             &the_lnet.ln_ping_target_md);
1817
1818         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1819         LASSERT(rc == 0);
1820
1821         lnet_ping_target_destroy();
1822 }
1823
1824 /* Resize the push target. */
1825 int lnet_push_target_resize(void)
1826 {
1827         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1828         struct lnet_md md = { NULL };
1829         struct lnet_handle_me meh;
1830         struct lnet_handle_md mdh;
1831         struct lnet_handle_md old_mdh;
1832         struct lnet_ping_buffer *pbuf;
1833         struct lnet_ping_buffer *old_pbuf;
1834         int nnis = the_lnet.ln_push_target_nnis;
1835         int rc;
1836
1837         if (nnis <= 0) {
1838                 rc = -EINVAL;
1839                 goto fail_return;
1840         }
1841 again:
1842         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1843         if (!pbuf) {
1844                 rc = -ENOMEM;
1845                 goto fail_return;
1846         }
1847
1848         rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1849                           LNET_PROTO_PING_MATCHBITS, 0,
1850                           LNET_UNLINK, LNET_INS_AFTER,
1851                           &meh);
1852         if (rc) {
1853                 CERROR("Can't create push target ME: %d\n", rc);
1854                 goto fail_decref_pbuf;
1855         }
1856
1857         /* initialize md content */
1858         md.start     = &pbuf->pb_info;
1859         md.length    = LNET_PING_INFO_SIZE(nnis);
1860         md.threshold = LNET_MD_THRESH_INF;
1861         md.max_size  = 0;
1862         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1863                        LNET_MD_MANAGE_REMOTE;
1864         md.user_ptr  = pbuf;
1865         md.eq_handle = the_lnet.ln_push_target_eq;
1866
1867         rc = LNetMDAttach(meh, md, LNET_RETAIN, &mdh);
1868         if (rc) {
1869                 CERROR("Can't attach push MD: %d\n", rc);
1870                 goto fail_unlink_meh;
1871         }
1872         lnet_ping_buffer_addref(pbuf);
1873
1874         lnet_net_lock(LNET_LOCK_EX);
1875         old_pbuf = the_lnet.ln_push_target;
1876         old_mdh = the_lnet.ln_push_target_md;
1877         the_lnet.ln_push_target = pbuf;
1878         the_lnet.ln_push_target_md = mdh;
1879         lnet_net_unlock(LNET_LOCK_EX);
1880
1881         if (old_pbuf) {
1882                 LNetMDUnlink(old_mdh);
1883                 lnet_ping_buffer_decref(old_pbuf);
1884         }
1885
1886         if (nnis < the_lnet.ln_push_target_nnis)
1887                 goto again;
1888
1889         CDEBUG(D_NET, "nnis %d success\n", nnis);
1890
1891         return 0;
1892
1893 fail_unlink_meh:
1894         LNetMEUnlink(meh);
1895 fail_decref_pbuf:
1896         lnet_ping_buffer_decref(pbuf);
1897 fail_return:
1898         CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1899         return rc;
1900 }
1901
1902 static void lnet_push_target_event_handler(struct lnet_event *ev)
1903 {
1904         struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1905
1906         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1907                 lnet_swap_pinginfo(pbuf);
1908
1909         lnet_peer_push_event(ev);
1910         if (ev->unlinked)
1911                 lnet_ping_buffer_decref(pbuf);
1912 }
1913
1914 /* Initialize the push target. */
1915 static int lnet_push_target_init(void)
1916 {
1917         int rc;
1918
1919         if (the_lnet.ln_push_target)
1920                 return -EALREADY;
1921
1922         rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1923                          &the_lnet.ln_push_target_eq);
1924         if (rc) {
1925                 CERROR("Can't allocated push target EQ: %d\n", rc);
1926                 return rc;
1927         }
1928
1929         /* Start at the required minimum, we'll enlarge if required. */
1930         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1931
1932         rc = lnet_push_target_resize();
1933
1934         if (rc) {
1935                 LNetEQFree(the_lnet.ln_push_target_eq);
1936                 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1937         }
1938
1939         return rc;
1940 }
1941
1942 /* Clean up the push target. */
1943 static void lnet_push_target_fini(void)
1944 {
1945         if (!the_lnet.ln_push_target)
1946                 return;
1947
1948         /* Unlink and invalidate to prevent new references. */
1949         LNetMDUnlink(the_lnet.ln_push_target_md);
1950         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1951
1952         /* Wait for the unlink to complete. */
1953         while (lnet_ping_buffer_numref(the_lnet.ln_push_target) > 1) {
1954                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1955                 set_current_state(TASK_UNINTERRUPTIBLE);
1956                 schedule_timeout(cfs_time_seconds(1));
1957         }
1958
1959         lnet_ping_buffer_decref(the_lnet.ln_push_target);
1960         the_lnet.ln_push_target = NULL;
1961         the_lnet.ln_push_target_nnis = 0;
1962
1963         LNetEQFree(the_lnet.ln_push_target_eq);
1964         LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1965 }
1966
1967 static int
1968 lnet_ni_tq_credits(struct lnet_ni *ni)
1969 {
1970         int     credits;
1971
1972         LASSERT(ni->ni_ncpts >= 1);
1973
1974         if (ni->ni_ncpts == 1)
1975                 return ni->ni_net->net_tunables.lct_max_tx_credits;
1976
1977         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
1978         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
1979         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
1980
1981         return credits;
1982 }
1983
1984 static void
1985 lnet_ni_unlink_locked(struct lnet_ni *ni)
1986 {
1987         /* move it to zombie list and nobody can find it anymore */
1988         LASSERT(!list_empty(&ni->ni_netlist));
1989         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
1990         lnet_ni_decref_locked(ni, 0);
1991 }
1992
1993 static void
1994 lnet_clear_zombies_nis_locked(struct lnet_net *net)
1995 {
1996         int             i;
1997         int             islo;
1998         struct lnet_ni  *ni;
1999         struct list_head *zombie_list = &net->net_ni_zombie;
2000
2001         /*
2002          * Now wait for the NIs I just nuked to show up on the zombie
2003          * list and shut them down in guaranteed thread context
2004          */
2005         i = 2;
2006         while (!list_empty(zombie_list)) {
2007                 int     *ref;
2008                 int     j;
2009
2010                 ni = list_entry(zombie_list->next,
2011                                 struct lnet_ni, ni_netlist);
2012                 list_del_init(&ni->ni_netlist);
2013                 /* the ni should be in deleting state. If it's not it's
2014                  * a bug */
2015                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2016                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2017                         if (*ref == 0)
2018                                 continue;
2019                         /* still busy, add it back to zombie list */
2020                         list_add(&ni->ni_netlist, zombie_list);
2021                         break;
2022                 }
2023
2024                 if (!list_empty(&ni->ni_netlist)) {
2025                         lnet_net_unlock(LNET_LOCK_EX);
2026                         ++i;
2027                         if ((i & (-i)) == i) {
2028                                 CDEBUG(D_WARNING,
2029                                        "Waiting for zombie LNI %s\n",
2030                                        libcfs_nid2str(ni->ni_nid));
2031                         }
2032                         set_current_state(TASK_UNINTERRUPTIBLE);
2033                         schedule_timeout(cfs_time_seconds(1));
2034                         lnet_net_lock(LNET_LOCK_EX);
2035                         continue;
2036                 }
2037
2038                 lnet_net_unlock(LNET_LOCK_EX);
2039
2040                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2041
2042                 LASSERT(!in_interrupt());
2043                 (net->net_lnd->lnd_shutdown)(ni);
2044
2045                 if (!islo)
2046                         CDEBUG(D_LNI, "Removed LNI %s\n",
2047                               libcfs_nid2str(ni->ni_nid));
2048
2049                 lnet_ni_free(ni);
2050                 i = 2;
2051                 lnet_net_lock(LNET_LOCK_EX);
2052         }
2053 }
2054
2055 /* shutdown down the NI and release refcount */
2056 static void
2057 lnet_shutdown_lndni(struct lnet_ni *ni)
2058 {
2059         int i;
2060         struct lnet_net *net = ni->ni_net;
2061
2062         lnet_net_lock(LNET_LOCK_EX);
2063         lnet_ni_lock(ni);
2064         ni->ni_state = LNET_NI_STATE_DELETING;
2065         lnet_ni_unlock(ni);
2066         lnet_ni_unlink_locked(ni);
2067         lnet_incr_dlc_seq();
2068         lnet_net_unlock(LNET_LOCK_EX);
2069
2070         /* clear messages for this NI on the lazy portal */
2071         for (i = 0; i < the_lnet.ln_nportals; i++)
2072                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2073
2074         lnet_net_lock(LNET_LOCK_EX);
2075         lnet_clear_zombies_nis_locked(net);
2076         lnet_net_unlock(LNET_LOCK_EX);
2077 }
2078
2079 static void
2080 lnet_shutdown_lndnet(struct lnet_net *net)
2081 {
2082         struct lnet_ni *ni;
2083
2084         lnet_net_lock(LNET_LOCK_EX);
2085
2086         net->net_state = LNET_NET_STATE_DELETING;
2087
2088         list_del_init(&net->net_list);
2089
2090         while (!list_empty(&net->net_ni_list)) {
2091                 ni = list_entry(net->net_ni_list.next,
2092                                 struct lnet_ni, ni_netlist);
2093                 lnet_net_unlock(LNET_LOCK_EX);
2094                 lnet_shutdown_lndni(ni);
2095                 lnet_net_lock(LNET_LOCK_EX);
2096         }
2097
2098         lnet_net_unlock(LNET_LOCK_EX);
2099
2100         /* Do peer table cleanup for this net */
2101         lnet_peer_tables_cleanup(net);
2102
2103         lnet_net_lock(LNET_LOCK_EX);
2104         /*
2105          * decrement ref count on lnd only when the entire network goes
2106          * away
2107          */
2108         net->net_lnd->lnd_refcount--;
2109
2110         lnet_net_unlock(LNET_LOCK_EX);
2111
2112         lnet_net_free(net);
2113 }
2114
2115 static void
2116 lnet_shutdown_lndnets(void)
2117 {
2118         struct lnet_net *net;
2119         struct list_head resend;
2120         struct lnet_msg *msg, *tmp;
2121
2122         INIT_LIST_HEAD(&resend);
2123
2124         /* NB called holding the global mutex */
2125
2126         /* All quiet on the API front */
2127         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2128         LASSERT(the_lnet.ln_refcount == 0);
2129
2130         lnet_net_lock(LNET_LOCK_EX);
2131         the_lnet.ln_state = LNET_STATE_STOPPING;
2132
2133         while (!list_empty(&the_lnet.ln_nets)) {
2134                 /*
2135                  * move the nets to the zombie list to avoid them being
2136                  * picked up for new work. LONET is also included in the
2137                  * Nets that will be moved to the zombie list
2138                  */
2139                 net = list_entry(the_lnet.ln_nets.next,
2140                                  struct lnet_net, net_list);
2141                 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2142         }
2143
2144         /* Drop the cached loopback Net. */
2145         if (the_lnet.ln_loni != NULL) {
2146                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2147                 the_lnet.ln_loni = NULL;
2148         }
2149         lnet_net_unlock(LNET_LOCK_EX);
2150
2151         /* iterate through the net zombie list and delete each net */
2152         while (!list_empty(&the_lnet.ln_net_zombie)) {
2153                 net = list_entry(the_lnet.ln_net_zombie.next,
2154                                  struct lnet_net, net_list);
2155                 lnet_shutdown_lndnet(net);
2156         }
2157
2158         spin_lock(&the_lnet.ln_msg_resend_lock);
2159         list_splice(&the_lnet.ln_msg_resend, &resend);
2160         spin_unlock(&the_lnet.ln_msg_resend_lock);
2161
2162         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2163                 list_del_init(&msg->msg_list);
2164                 msg->msg_no_resend = true;
2165                 lnet_finalize(msg, -ECANCELED);
2166         }
2167
2168         lnet_net_lock(LNET_LOCK_EX);
2169         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2170         lnet_net_unlock(LNET_LOCK_EX);
2171 }
2172
2173 static int
2174 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2175 {
2176         int                     rc = -EINVAL;
2177         struct lnet_tx_queue    *tq;
2178         int                     i;
2179         struct lnet_net         *net = ni->ni_net;
2180
2181         mutex_lock(&the_lnet.ln_lnd_mutex);
2182
2183         if (tun) {
2184                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2185                 ni->ni_lnd_tunables_set = true;
2186         }
2187
2188         rc = (net->net_lnd->lnd_startup)(ni);
2189
2190         mutex_unlock(&the_lnet.ln_lnd_mutex);
2191
2192         if (rc != 0) {
2193                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2194                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2195                 lnet_net_lock(LNET_LOCK_EX);
2196                 net->net_lnd->lnd_refcount--;
2197                 lnet_net_unlock(LNET_LOCK_EX);
2198                 goto failed0;
2199         }
2200
2201         lnet_ni_lock(ni);
2202         ni->ni_state = LNET_NI_STATE_ACTIVE;
2203         lnet_ni_unlock(ni);
2204
2205         /* We keep a reference on the loopback net through the loopback NI */
2206         if (net->net_lnd->lnd_type == LOLND) {
2207                 lnet_ni_addref(ni);
2208                 LASSERT(the_lnet.ln_loni == NULL);
2209                 the_lnet.ln_loni = ni;
2210                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2211                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2212                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2213                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2214                 return 0;
2215         }
2216
2217         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2218             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2219                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2220                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2221                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2222                                         "" : "per-peer ");
2223                 /* shutdown the NI since if we get here then it must've already
2224                  * been started
2225                  */
2226                 lnet_shutdown_lndni(ni);
2227                 return -EINVAL;
2228         }
2229
2230         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2231                 tq->tq_credits_min =
2232                 tq->tq_credits_max =
2233                 tq->tq_credits = lnet_ni_tq_credits(ni);
2234         }
2235
2236         atomic_set(&ni->ni_tx_credits,
2237                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2238         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2239
2240         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2241                 libcfs_nid2str(ni->ni_nid),
2242                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2243                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2244                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2245                 ni->ni_net->net_tunables.lct_peer_timeout);
2246
2247         return 0;
2248 failed0:
2249         lnet_ni_free(ni);
2250         return rc;
2251 }
2252
2253 static int
2254 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2255 {
2256         struct lnet_ni *ni;
2257         struct lnet_net *net_l = NULL;
2258         struct list_head        local_ni_list;
2259         int                     rc;
2260         int                     ni_count = 0;
2261         __u32                   lnd_type;
2262         struct lnet_lnd *lnd;
2263         int                     peer_timeout =
2264                 net->net_tunables.lct_peer_timeout;
2265         int                     maxtxcredits =
2266                 net->net_tunables.lct_max_tx_credits;
2267         int                     peerrtrcredits =
2268                 net->net_tunables.lct_peer_rtr_credits;
2269
2270         INIT_LIST_HEAD(&local_ni_list);
2271
2272         /*
2273          * make sure that this net is unique. If it isn't then
2274          * we are adding interfaces to an already existing network, and
2275          * 'net' is just a convenient way to pass in the list.
2276          * if it is unique we need to find the LND and load it if
2277          * necessary.
2278          */
2279         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2280                 lnd_type = LNET_NETTYP(net->net_id);
2281
2282                 mutex_lock(&the_lnet.ln_lnd_mutex);
2283                 lnd = lnet_find_lnd_by_type(lnd_type);
2284
2285                 if (lnd == NULL) {
2286                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2287                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2288                         mutex_lock(&the_lnet.ln_lnd_mutex);
2289
2290                         lnd = lnet_find_lnd_by_type(lnd_type);
2291                         if (lnd == NULL) {
2292                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2293                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2294                                 libcfs_lnd2str(lnd_type),
2295                                 libcfs_lnd2modname(lnd_type), rc);
2296 #ifndef HAVE_MODULE_LOADING_SUPPORT
2297                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2298                                                 "compiled with kernel module "
2299                                                 "loading support.");
2300 #endif
2301                                 rc = -EINVAL;
2302                                 goto failed0;
2303                         }
2304                 }
2305
2306                 lnet_net_lock(LNET_LOCK_EX);
2307                 lnd->lnd_refcount++;
2308                 lnet_net_unlock(LNET_LOCK_EX);
2309
2310                 net->net_lnd = lnd;
2311
2312                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2313
2314                 net_l = net;
2315         }
2316
2317         /*
2318          * net_l: if the network being added is unique then net_l
2319          *        will point to that network
2320          *        if the network being added is not unique then
2321          *        net_l points to the existing network.
2322          *
2323          * When we enter the loop below, we'll pick NIs off he
2324          * network beign added and start them up, then add them to
2325          * a local ni list. Once we've successfully started all
2326          * the NIs then we join the local NI list (of started up
2327          * networks) with the net_l->net_ni_list, which should
2328          * point to the correct network to add the new ni list to
2329          *
2330          * If any of the new NIs fail to start up, then we want to
2331          * iterate through the local ni list, which should include
2332          * any NIs which were successfully started up, and shut
2333          * them down.
2334          *
2335          * After than we want to delete the network being added,
2336          * to avoid a memory leak.
2337          */
2338
2339         /*
2340          * When a network uses TCP bonding then all its interfaces
2341          * must be specified when the network is first defined: the
2342          * TCP bonding code doesn't allow for interfaces to be added
2343          * or removed.
2344          */
2345         if (net_l != net && net_l != NULL && use_tcp_bonding &&
2346             LNET_NETTYP(net_l->net_id) == SOCKLND) {
2347                 rc = -EINVAL;
2348                 goto failed0;
2349         }
2350
2351         while (!list_empty(&net->net_ni_added)) {
2352                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2353                                 ni_netlist);
2354                 list_del_init(&ni->ni_netlist);
2355
2356                 /* make sure that the the NI we're about to start
2357                  * up is actually unique. if it's not fail. */
2358                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2359                                         ni->ni_interfaces[0])) {
2360                         rc = -EINVAL;
2361                         goto failed1;
2362                 }
2363
2364                 /* adjust the pointer the parent network, just in case it
2365                  * the net is a duplicate */
2366                 ni->ni_net = net_l;
2367
2368                 rc = lnet_startup_lndni(ni, tun);
2369
2370                 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
2371                         ni->ni_net->net_lnd->lnd_query != NULL);
2372
2373                 if (rc < 0)
2374                         goto failed1;
2375
2376                 lnet_ni_addref(ni);
2377                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2378
2379                 ni_count++;
2380         }
2381
2382         lnet_net_lock(LNET_LOCK_EX);
2383         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2384         lnet_incr_dlc_seq();
2385         lnet_net_unlock(LNET_LOCK_EX);
2386
2387         /* if the network is not unique then we don't want to keep
2388          * it around after we're done. Free it. Otherwise add that
2389          * net to the global the_lnet.ln_nets */
2390         if (net_l != net && net_l != NULL) {
2391                 /*
2392                  * TODO - note. currently the tunables can not be updated
2393                  * once added
2394                  */
2395                 lnet_net_free(net);
2396         } else {
2397                 net->net_state = LNET_NET_STATE_ACTIVE;
2398                 /*
2399                  * restore tunables after it has been overwitten by the
2400                  * lnd
2401                  */
2402                 if (peer_timeout != -1)
2403                         net->net_tunables.lct_peer_timeout = peer_timeout;
2404                 if (maxtxcredits != -1)
2405                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2406                 if (peerrtrcredits != -1)
2407                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2408
2409                 lnet_net_lock(LNET_LOCK_EX);
2410                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2411                 lnet_net_unlock(LNET_LOCK_EX);
2412         }
2413
2414         /* update net count */
2415         lnet_current_net_count = lnet_get_net_count();
2416
2417         return ni_count;
2418
2419 failed1:
2420         /*
2421          * shutdown the new NIs that are being started up
2422          * free the NET being started
2423          */
2424         while (!list_empty(&local_ni_list)) {
2425                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2426                                 ni_netlist);
2427
2428                 lnet_shutdown_lndni(ni);
2429         }
2430
2431 failed0:
2432         lnet_net_free(net);
2433
2434         return rc;
2435 }
2436
2437 static int
2438 lnet_startup_lndnets(struct list_head *netlist)
2439 {
2440         struct lnet_net         *net;
2441         int                     rc;
2442         int                     ni_count = 0;
2443
2444         /*
2445          * Change to running state before bringing up the LNDs. This
2446          * allows lnet_shutdown_lndnets() to assert that we've passed
2447          * through here.
2448          */
2449         lnet_net_lock(LNET_LOCK_EX);
2450         the_lnet.ln_state = LNET_STATE_RUNNING;
2451         lnet_net_unlock(LNET_LOCK_EX);
2452
2453         while (!list_empty(netlist)) {
2454                 net = list_entry(netlist->next, struct lnet_net, net_list);
2455                 list_del_init(&net->net_list);
2456
2457                 rc = lnet_startup_lndnet(net, NULL);
2458
2459                 if (rc < 0)
2460                         goto failed;
2461
2462                 ni_count += rc;
2463         }
2464
2465         return ni_count;
2466 failed:
2467         lnet_shutdown_lndnets();
2468
2469         return rc;
2470 }
2471
2472 /**
2473  * Initialize LNet library.
2474  *
2475  * Automatically called at module loading time. Caller has to call
2476  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2477  * latter returned 0. It must be called exactly once.
2478  *
2479  * \retval 0 on success
2480  * \retval -ve on failures.
2481  */
2482 int lnet_lib_init(void)
2483 {
2484         int rc;
2485
2486         lnet_assert_wire_constants();
2487
2488         /* refer to global cfs_cpt_table for now */
2489         the_lnet.ln_cpt_table   = cfs_cpt_table;
2490         the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
2491
2492         LASSERT(the_lnet.ln_cpt_number > 0);
2493         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2494                 /* we are under risk of consuming all lh_cookie */
2495                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2496                        "please change setting of CPT-table and retry\n",
2497                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2498                 return -E2BIG;
2499         }
2500
2501         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2502                 the_lnet.ln_cpt_bits++;
2503
2504         rc = lnet_create_locks();
2505         if (rc != 0) {
2506                 CERROR("Can't create LNet global locks: %d\n", rc);
2507                 return rc;
2508         }
2509
2510         the_lnet.ln_refcount = 0;
2511         INIT_LIST_HEAD(&the_lnet.ln_lnds);
2512         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2513         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2514
2515         /* The hash table size is the number of bits it takes to express the set
2516          * ln_num_routes, minus 1 (better to under estimate than over so we
2517          * don't waste memory). */
2518         if (rnet_htable_size <= 0)
2519                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2520         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2521                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2522         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2523                                            order_base_2(rnet_htable_size) - 1);
2524
2525         /* All LNDs apart from the LOLND are in separate modules.  They
2526          * register themselves when their module loads, and unregister
2527          * themselves when their module is unloaded. */
2528         lnet_register_lnd(&the_lolnd);
2529         return 0;
2530 }
2531
2532 /**
2533  * Finalize LNet library.
2534  *
2535  * \pre lnet_lib_init() called with success.
2536  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2537  */
2538 void lnet_lib_exit(void)
2539 {
2540         LASSERT(the_lnet.ln_refcount == 0);
2541
2542         while (!list_empty(&the_lnet.ln_lnds))
2543                 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
2544                                                struct lnet_lnd, lnd_list));
2545         lnet_destroy_locks();
2546 }
2547
2548 /**
2549  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2550  *
2551  * Users must call this function at least once before any other functions.
2552  * For each successful call there must be a corresponding call to
2553  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2554  * ignored.
2555  *
2556  * The PID used by LNet may be different from the one requested.
2557  * See LNetGetId().
2558  *
2559  * \param requested_pid PID requested by the caller.
2560  *
2561  * \return >= 0 on success, and < 0 error code on failures.
2562  */
2563 int
2564 LNetNIInit(lnet_pid_t requested_pid)
2565 {
2566         int                     im_a_router = 0;
2567         int                     rc;
2568         int                     ni_count;
2569         struct lnet_ping_buffer *pbuf;
2570         struct lnet_handle_md   ping_mdh;
2571         struct list_head        net_head;
2572         struct lnet_net         *net;
2573
2574         INIT_LIST_HEAD(&net_head);
2575
2576         mutex_lock(&the_lnet.ln_api_mutex);
2577
2578         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2579
2580         if (the_lnet.ln_refcount > 0) {
2581                 rc = the_lnet.ln_refcount++;
2582                 mutex_unlock(&the_lnet.ln_api_mutex);
2583                 return rc;
2584         }
2585
2586         rc = lnet_prepare(requested_pid);
2587         if (rc != 0) {
2588                 mutex_unlock(&the_lnet.ln_api_mutex);
2589                 return rc;
2590         }
2591
2592         /* create a network for Loopback network */
2593         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2594         if (net == NULL) {
2595                 rc = -ENOMEM;
2596                 goto err_empty_list;
2597         }
2598
2599         /* Add in the loopback NI */
2600         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2601                 rc = -ENOMEM;
2602                 goto err_empty_list;
2603         }
2604
2605         /* If LNet is being initialized via DLC it is possible
2606          * that the user requests not to load module parameters (ones which
2607          * are supported by DLC) on initialization.  Therefore, make sure not
2608          * to load networks, routes and forwarding from module parameters
2609          * in this case.  On cleanup in case of failure only clean up
2610          * routes if it has been loaded */
2611         if (!the_lnet.ln_nis_from_mod_params) {
2612                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2613                                          use_tcp_bonding);
2614                 if (rc < 0)
2615                         goto err_empty_list;
2616         }
2617
2618         ni_count = lnet_startup_lndnets(&net_head);
2619         if (ni_count < 0) {
2620                 rc = ni_count;
2621                 goto err_empty_list;
2622         }
2623
2624         if (!the_lnet.ln_nis_from_mod_params) {
2625                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2626                 if (rc != 0)
2627                         goto err_shutdown_lndnis;
2628
2629                 rc = lnet_rtrpools_alloc(im_a_router);
2630                 if (rc != 0)
2631                         goto err_destroy_routes;
2632         }
2633
2634         rc = lnet_acceptor_start();
2635         if (rc != 0)
2636                 goto err_destroy_routes;
2637
2638         the_lnet.ln_refcount = 1;
2639         /* Now I may use my own API functions... */
2640
2641         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2642         if (rc != 0)
2643                 goto err_acceptor_stop;
2644
2645         lnet_ping_target_update(pbuf, ping_mdh);
2646
2647         rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2648         if (rc != 0) {
2649                 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2650                 goto err_stop_ping;
2651         }
2652
2653         rc = lnet_push_target_init();
2654         if (rc != 0)
2655                 goto err_stop_ping;
2656
2657         rc = lnet_peer_discovery_start();
2658         if (rc != 0)
2659                 goto err_destroy_push_target;
2660
2661         rc = lnet_monitor_thr_start();
2662         if (rc != 0)
2663                 goto err_stop_discovery_thr;
2664
2665         lnet_fault_init();
2666         lnet_router_debugfs_init();
2667
2668         mutex_unlock(&the_lnet.ln_api_mutex);
2669
2670         complete_all(&the_lnet.ln_started);
2671
2672         /* wait for all routers to start */
2673         lnet_wait_router_start();
2674
2675         return 0;
2676
2677 err_stop_discovery_thr:
2678         lnet_peer_discovery_stop();
2679 err_destroy_push_target:
2680         lnet_push_target_fini();
2681 err_stop_ping:
2682         lnet_ping_target_fini();
2683 err_acceptor_stop:
2684         the_lnet.ln_refcount = 0;
2685         lnet_acceptor_stop();
2686 err_destroy_routes:
2687         if (!the_lnet.ln_nis_from_mod_params)
2688                 lnet_destroy_routes();
2689 err_shutdown_lndnis:
2690         lnet_shutdown_lndnets();
2691 err_empty_list:
2692         lnet_unprepare();
2693         LASSERT(rc < 0);
2694         mutex_unlock(&the_lnet.ln_api_mutex);
2695         while (!list_empty(&net_head)) {
2696                 struct lnet_net *net;
2697
2698                 net = list_entry(net_head.next, struct lnet_net, net_list);
2699                 list_del_init(&net->net_list);
2700                 lnet_net_free(net);
2701         }
2702         return rc;
2703 }
2704 EXPORT_SYMBOL(LNetNIInit);
2705
2706 /**
2707  * Stop LNet interfaces, routing, and forwarding.
2708  *
2709  * Users must call this function once for each successful call to LNetNIInit().
2710  * Once the LNetNIFini() operation has been started, the results of pending
2711  * API operations are undefined.
2712  *
2713  * \return always 0 for current implementation.
2714  */
2715 int
2716 LNetNIFini()
2717 {
2718         mutex_lock(&the_lnet.ln_api_mutex);
2719
2720         LASSERT(the_lnet.ln_refcount > 0);
2721
2722         if (the_lnet.ln_refcount != 1) {
2723                 the_lnet.ln_refcount--;
2724         } else {
2725                 LASSERT(!the_lnet.ln_niinit_self);
2726
2727                 lnet_fault_fini();
2728
2729                 lnet_router_debugfs_fini();
2730                 lnet_monitor_thr_stop();
2731                 lnet_peer_discovery_stop();
2732                 lnet_push_target_fini();
2733                 lnet_ping_target_fini();
2734
2735                 /* Teardown fns that use my own API functions BEFORE here */
2736                 the_lnet.ln_refcount = 0;
2737
2738                 lnet_acceptor_stop();
2739                 lnet_destroy_routes();
2740                 lnet_shutdown_lndnets();
2741                 lnet_unprepare();
2742         }
2743
2744         mutex_unlock(&the_lnet.ln_api_mutex);
2745         return 0;
2746 }
2747 EXPORT_SYMBOL(LNetNIFini);
2748
2749 /**
2750  * Grabs the ni data from the ni structure and fills the out
2751  * parameters
2752  *
2753  * \param[in] ni network        interface structure
2754  * \param[out] cfg_ni           NI config information
2755  * \param[out] tun              network and LND tunables
2756  */
2757 static void
2758 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2759                    struct lnet_ioctl_config_lnd_tunables *tun,
2760                    struct lnet_ioctl_element_stats *stats,
2761                    __u32 tun_size)
2762 {
2763         size_t min_size = 0;
2764         int i;
2765
2766         if (!ni || !cfg_ni || !tun)
2767                 return;
2768
2769         if (ni->ni_interfaces[0] != NULL) {
2770                 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2771                         if (ni->ni_interfaces[i] != NULL) {
2772                                 strncpy(cfg_ni->lic_ni_intf[i],
2773                                         ni->ni_interfaces[i],
2774                                         sizeof(cfg_ni->lic_ni_intf[i]));
2775                         }
2776                 }
2777         }
2778
2779         cfg_ni->lic_nid = ni->ni_nid;
2780         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2781                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2782         else
2783                 cfg_ni->lic_status = ni->ni_status->ns_status;
2784         cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2785         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2786
2787         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2788
2789         if (stats) {
2790                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2791                                                        LNET_STATS_TYPE_SEND);
2792                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2793                                                        LNET_STATS_TYPE_RECV);
2794                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2795                                                        LNET_STATS_TYPE_DROP);
2796         }
2797
2798         /*
2799          * tun->lt_tun will always be present, but in order to be
2800          * backwards compatible, we need to deal with the cases when
2801          * tun->lt_tun is smaller than what the kernel has, because it
2802          * comes from an older version of a userspace program, then we'll
2803          * need to copy as much information as we have available space.
2804          */
2805         min_size = tun_size - sizeof(tun->lt_cmn);
2806         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2807
2808         /* copy over the cpts */
2809         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2810             ni->ni_cpts == NULL)  {
2811                 for (i = 0; i < ni->ni_ncpts; i++)
2812                         cfg_ni->lic_cpts[i] = i;
2813         } else {
2814                 for (i = 0;
2815                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2816                      i < LNET_MAX_SHOW_NUM_CPT;
2817                      i++)
2818                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2819         }
2820         cfg_ni->lic_ncpts = ni->ni_ncpts;
2821 }
2822
2823 /**
2824  * NOTE: This is a legacy function left in the code to be backwards
2825  * compatible with older userspace programs. It should eventually be
2826  * removed.
2827  *
2828  * Grabs the ni data from the ni structure and fills the out
2829  * parameters
2830  *
2831  * \param[in] ni network        interface structure
2832  * \param[out] config           config information
2833  */
2834 static void
2835 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2836                          struct lnet_ioctl_config_data *config)
2837 {
2838         struct lnet_ioctl_net_config *net_config;
2839         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2840         size_t min_size, tunable_size = 0;
2841         int i;
2842
2843         if (!ni || !config)
2844                 return;
2845
2846         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2847         if (!net_config)
2848                 return;
2849
2850         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2851                      ARRAY_SIZE(net_config->ni_interfaces));
2852
2853         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2854                 if (!ni->ni_interfaces[i])
2855                         break;
2856
2857                 strncpy(net_config->ni_interfaces[i],
2858                         ni->ni_interfaces[i],
2859                         sizeof(net_config->ni_interfaces[i]));
2860         }
2861
2862         config->cfg_nid = ni->ni_nid;
2863         config->cfg_config_u.cfg_net.net_peer_timeout =
2864                 ni->ni_net->net_tunables.lct_peer_timeout;
2865         config->cfg_config_u.cfg_net.net_max_tx_credits =
2866                 ni->ni_net->net_tunables.lct_max_tx_credits;
2867         config->cfg_config_u.cfg_net.net_peer_tx_credits =
2868                 ni->ni_net->net_tunables.lct_peer_tx_credits;
2869         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2870                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2871
2872         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2873                 net_config->ni_status = LNET_NI_STATUS_UP;
2874         else
2875                 net_config->ni_status = ni->ni_status->ns_status;
2876
2877         if (ni->ni_cpts) {
2878                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2879
2880                 for (i = 0; i < num_cpts; i++)
2881                         net_config->ni_cpts[i] = ni->ni_cpts[i];
2882
2883                 config->cfg_ncpts = num_cpts;
2884         }
2885
2886         /*
2887          * See if user land tools sent in a newer and larger version
2888          * of struct lnet_tunables than what the kernel uses.
2889          */
2890         min_size = sizeof(*config) + sizeof(*net_config);
2891
2892         if (config->cfg_hdr.ioc_len > min_size)
2893                 tunable_size = config->cfg_hdr.ioc_len - min_size;
2894
2895         /* Don't copy too much data to user space */
2896         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2897         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2898
2899         if (lnd_cfg && min_size) {
2900                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2901                 config->cfg_config_u.cfg_net.net_interface_count = 1;
2902
2903                 /* Tell user land that kernel side has less data */
2904                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2905                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2906                         config->cfg_hdr.ioc_len -= min_size;
2907                 }
2908         }
2909 }
2910
2911 struct lnet_ni *
2912 lnet_get_ni_idx_locked(int idx)
2913 {
2914         struct lnet_ni          *ni;
2915         struct lnet_net         *net;
2916
2917         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2918                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2919                         if (idx-- == 0)
2920                                 return ni;
2921                 }
2922         }
2923
2924         return NULL;
2925 }
2926
2927 struct lnet_ni *
2928 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2929 {
2930         struct lnet_ni          *ni;
2931         struct lnet_net         *net = mynet;
2932
2933         /*
2934          * It is possible that the net has been cleaned out while there is
2935          * a message being sent. This function accessed the net without
2936          * checking if the list is empty
2937          */
2938         if (prev == NULL) {
2939                 if (net == NULL)
2940                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2941                                         net_list);
2942                 if (list_empty(&net->net_ni_list))
2943                         return NULL;
2944                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2945                                 ni_netlist);
2946
2947                 return ni;
2948         }
2949
2950         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2951                 /* if you reached the end of the ni list and the net is
2952                  * specified, then there are no more nis in that net */
2953                 if (net != NULL)
2954                         return NULL;
2955
2956                 /* we reached the end of this net ni list. move to the
2957                  * next net */
2958                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2959                         /* no more nets and no more NIs. */
2960                         return NULL;
2961
2962                 /* get the next net */
2963                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2964                                  net_list);
2965                 if (list_empty(&net->net_ni_list))
2966                         return NULL;
2967                 /* get the ni on it */
2968                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2969                                 ni_netlist);
2970
2971                 return ni;
2972         }
2973
2974         if (list_empty(&prev->ni_netlist))
2975                 return NULL;
2976
2977         /* there are more nis left */
2978         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2979
2980         return ni;
2981 }
2982
2983 int
2984 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2985 {
2986         struct lnet_ni *ni;
2987         int cpt;
2988         int rc = -ENOENT;
2989         int idx = config->cfg_count;
2990
2991         cpt = lnet_net_lock_current();
2992
2993         ni = lnet_get_ni_idx_locked(idx);
2994
2995         if (ni != NULL) {
2996                 rc = 0;
2997                 lnet_ni_lock(ni);
2998                 lnet_fill_ni_info_legacy(ni, config);
2999                 lnet_ni_unlock(ni);
3000         }
3001
3002         lnet_net_unlock(cpt);
3003         return rc;
3004 }
3005
3006 int
3007 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3008                    struct lnet_ioctl_config_lnd_tunables *tun,
3009                    struct lnet_ioctl_element_stats *stats,
3010                    __u32 tun_size)
3011 {
3012         struct lnet_ni          *ni;
3013         int                     cpt;
3014         int                     rc = -ENOENT;
3015
3016         if (!cfg_ni || !tun || !stats)
3017                 return -EINVAL;
3018
3019         cpt = lnet_net_lock_current();
3020
3021         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3022
3023         if (ni) {
3024                 rc = 0;
3025                 lnet_ni_lock(ni);
3026                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3027                 lnet_ni_unlock(ni);
3028         }
3029
3030         lnet_net_unlock(cpt);
3031         return rc;
3032 }
3033
3034 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3035 {
3036         struct lnet_ni *ni;
3037         int cpt;
3038         int rc = -ENOENT;
3039
3040         if (!msg_stats)
3041                 return -EINVAL;
3042
3043         cpt = lnet_net_lock_current();
3044
3045         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3046
3047         if (ni) {
3048                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3049                 rc = 0;
3050         }
3051
3052         lnet_net_unlock(cpt);
3053
3054         return rc;
3055 }
3056
3057 static int lnet_add_net_common(struct lnet_net *net,
3058                                struct lnet_ioctl_config_lnd_tunables *tun)
3059 {
3060         __u32                   net_id;
3061         struct lnet_ping_buffer *pbuf;
3062         struct lnet_handle_md   ping_mdh;
3063         int                     rc;
3064         struct lnet_remotenet *rnet;
3065         int                     net_ni_count;
3066         int                     num_acceptor_nets;
3067
3068         lnet_net_lock(LNET_LOCK_EX);
3069         rnet = lnet_find_rnet_locked(net->net_id);
3070         lnet_net_unlock(LNET_LOCK_EX);
3071         /*
3072          * make sure that the net added doesn't invalidate the current
3073          * configuration LNet is keeping
3074          */
3075         if (rnet) {
3076                 CERROR("Adding net %s will invalidate routing configuration\n",
3077                        libcfs_net2str(net->net_id));
3078                 lnet_net_free(net);
3079                 return -EUSERS;
3080         }
3081
3082         /*
3083          * make sure you calculate the correct number of slots in the ping
3084          * buffer. Since the ping info is a flattened list of all the NIs,
3085          * we should allocate enough slots to accomodate the number of NIs
3086          * which will be added.
3087          *
3088          * since ni hasn't been configured yet, use
3089          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3090          */
3091         net_ni_count = lnet_get_net_ni_count_pre(net);
3092
3093         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3094                                     net_ni_count + lnet_get_ni_count(),
3095                                     false);
3096         if (rc < 0) {
3097                 lnet_net_free(net);
3098                 return rc;
3099         }
3100
3101         if (tun)
3102                 memcpy(&net->net_tunables,
3103                        &tun->lt_cmn, sizeof(net->net_tunables));
3104         else
3105                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3106
3107         /*
3108          * before starting this network get a count of the current TCP
3109          * networks which require the acceptor thread running. If that
3110          * count is == 0 before we start up this network, then we'd want to
3111          * start up the acceptor thread after starting up this network
3112          */
3113         num_acceptor_nets = lnet_count_acceptor_nets();
3114
3115         net_id = net->net_id;
3116
3117         rc = lnet_startup_lndnet(net,
3118                                  (tun) ? &tun->lt_tun : NULL);
3119         if (rc < 0)
3120                 goto failed;
3121
3122         lnet_net_lock(LNET_LOCK_EX);
3123         net = lnet_get_net_locked(net_id);
3124         lnet_net_unlock(LNET_LOCK_EX);
3125
3126         LASSERT(net);
3127
3128         /*
3129          * Start the acceptor thread if this is the first network
3130          * being added that requires the thread.
3131          */
3132         if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3133                 rc = lnet_acceptor_start();
3134                 if (rc < 0) {
3135                         /* shutdown the net that we just started */
3136                         CERROR("Failed to start up acceptor thread\n");
3137                         lnet_shutdown_lndnet(net);
3138                         goto failed;
3139                 }
3140         }
3141
3142         lnet_net_lock(LNET_LOCK_EX);
3143         lnet_peer_net_added(net);
3144         lnet_net_unlock(LNET_LOCK_EX);
3145
3146         lnet_ping_target_update(pbuf, ping_mdh);
3147
3148         return 0;
3149
3150 failed:
3151         lnet_ping_md_unlink(pbuf, &ping_mdh);
3152         lnet_ping_buffer_decref(pbuf);
3153         return rc;
3154 }
3155
3156 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3157                                       struct lnet_ioctl_config_lnd_tunables *tun)
3158 {
3159         struct lnet_net *net;
3160         char *nets;
3161         int rc;
3162         struct list_head net_head;
3163
3164         INIT_LIST_HEAD(&net_head);
3165
3166         rc = lnet_parse_ip2nets(&nets, ip2nets);
3167         if (rc < 0)
3168                 return rc;
3169
3170         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3171         if (rc < 0)
3172                 return rc;
3173
3174         mutex_lock(&the_lnet.ln_api_mutex);
3175         while (!list_empty(&net_head)) {
3176                 net = list_entry(net_head.next, struct lnet_net, net_list);
3177                 list_del_init(&net->net_list);
3178                 rc = lnet_add_net_common(net, tun);
3179                 if (rc < 0)
3180                         goto out;
3181         }
3182
3183 out:
3184         mutex_unlock(&the_lnet.ln_api_mutex);
3185
3186         while (!list_empty(&net_head)) {
3187                 net = list_entry(net_head.next, struct lnet_net, net_list);
3188                 list_del_init(&net->net_list);
3189                 lnet_net_free(net);
3190         }
3191         return rc;
3192 }
3193
3194 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3195 {
3196         struct lnet_net *net;
3197         struct lnet_ni *ni;
3198         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3199         int rc, i;
3200         __u32 net_id, lnd_type;
3201
3202         /* get the tunables if they are available */
3203         if (conf->lic_cfg_hdr.ioc_len >=
3204             sizeof(*conf) + sizeof(*tun))
3205                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3206                         conf->lic_bulk;
3207
3208         /* handle legacy ip2nets from DLC */
3209         if (conf->lic_legacy_ip2nets[0] != '\0')
3210                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3211                                                   tun);
3212
3213         net_id = LNET_NIDNET(conf->lic_nid);
3214         lnd_type = LNET_NETTYP(net_id);
3215
3216         if (!libcfs_isknown_lnd(lnd_type)) {
3217                 CERROR("No valid net and lnd information provided\n");
3218                 return -EINVAL;
3219         }
3220
3221         net = lnet_net_alloc(net_id, NULL);
3222         if (!net)
3223                 return -ENOMEM;
3224
3225         for (i = 0; i < conf->lic_ncpts; i++) {
3226                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3227                         return -EINVAL;
3228         }
3229
3230         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3231                                        conf->lic_ni_intf[0]);
3232         if (!ni)
3233                 return -ENOMEM;
3234
3235         mutex_lock(&the_lnet.ln_api_mutex);
3236
3237         rc = lnet_add_net_common(net, tun);
3238
3239         mutex_unlock(&the_lnet.ln_api_mutex);
3240
3241         return rc;
3242 }
3243
3244 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3245 {
3246         struct lnet_net  *net;
3247         struct lnet_ni *ni;
3248         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3249         struct lnet_ping_buffer *pbuf;
3250         struct lnet_handle_md  ping_mdh;
3251         int               rc;
3252         int               net_count;
3253         __u32             addr;
3254
3255         /* don't allow userspace to shutdown the LOLND */
3256         if (LNET_NETTYP(net_id) == LOLND)
3257                 return -EINVAL;
3258
3259         mutex_lock(&the_lnet.ln_api_mutex);
3260
3261         lnet_net_lock(0);
3262
3263         net = lnet_get_net_locked(net_id);
3264         if (!net) {
3265                 CERROR("net %s not found\n",
3266                        libcfs_net2str(net_id));
3267                 rc = -ENOENT;
3268                 goto unlock_net;
3269         }
3270
3271         addr = LNET_NIDADDR(conf->lic_nid);
3272         if (addr == 0) {
3273                 /* remove the entire net */
3274                 net_count = lnet_get_net_ni_count_locked(net);
3275
3276                 lnet_net_unlock(0);
3277
3278                 /* create and link a new ping info, before removing the old one */
3279                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3280                                         lnet_get_ni_count() - net_count,
3281                                         false);
3282                 if (rc != 0)
3283                         goto unlock_api_mutex;
3284
3285                 lnet_shutdown_lndnet(net);
3286
3287                 if (lnet_count_acceptor_nets() == 0)
3288                         lnet_acceptor_stop();
3289
3290                 lnet_ping_target_update(pbuf, ping_mdh);
3291
3292                 goto unlock_api_mutex;
3293         }
3294
3295         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3296         if (!ni) {
3297                 CERROR("nid %s not found\n",
3298                        libcfs_nid2str(conf->lic_nid));
3299                 rc = -ENOENT;
3300                 goto unlock_net;
3301         }
3302
3303         net_count = lnet_get_net_ni_count_locked(net);
3304
3305         lnet_net_unlock(0);
3306
3307         /* create and link a new ping info, before removing the old one */
3308         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3309                                   lnet_get_ni_count() - 1, false);
3310         if (rc != 0)
3311                 goto unlock_api_mutex;
3312
3313         lnet_shutdown_lndni(ni);
3314
3315         if (lnet_count_acceptor_nets() == 0)
3316                 lnet_acceptor_stop();
3317
3318         lnet_ping_target_update(pbuf, ping_mdh);
3319
3320         /* check if the net is empty and remove it if it is */
3321         if (net_count == 1)
3322                 lnet_shutdown_lndnet(net);
3323
3324         goto unlock_api_mutex;
3325
3326 unlock_net:
3327         lnet_net_unlock(0);
3328 unlock_api_mutex:
3329         mutex_unlock(&the_lnet.ln_api_mutex);
3330
3331         return rc;
3332 }
3333
3334 /*
3335  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3336  * They are only expected to be called for unique networks.
3337  * That can be as a result of older DLC library
3338  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3339  */
3340 int
3341 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3342 {
3343         struct lnet_net         *net;
3344         struct list_head        net_head;
3345         int                     rc;
3346         struct lnet_ioctl_config_lnd_tunables tun;
3347         char *nets = conf->cfg_config_u.cfg_net.net_intf;
3348
3349         INIT_LIST_HEAD(&net_head);
3350
3351         /* Create a net/ni structures for the network string */
3352         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3353         if (rc <= 0)
3354                 return rc == 0 ? -EINVAL : rc;
3355
3356         mutex_lock(&the_lnet.ln_api_mutex);
3357
3358         if (rc > 1) {
3359                 rc = -EINVAL; /* only add one network per call */
3360                 goto out_unlock_clean;
3361         }
3362
3363         net = list_entry(net_head.next, struct lnet_net, net_list);
3364         list_del_init(&net->net_list);
3365
3366         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3367
3368         memset(&tun, 0, sizeof(tun));
3369
3370         tun.lt_cmn.lct_peer_timeout =
3371           conf->cfg_config_u.cfg_net.net_peer_timeout;
3372         tun.lt_cmn.lct_peer_tx_credits =
3373           conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3374         tun.lt_cmn.lct_peer_rtr_credits =
3375           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3376         tun.lt_cmn.lct_max_tx_credits =
3377           conf->cfg_config_u.cfg_net.net_max_tx_credits;
3378
3379         rc = lnet_add_net_common(net, &tun);
3380
3381 out_unlock_clean:
3382         mutex_unlock(&the_lnet.ln_api_mutex);
3383         while (!list_empty(&net_head)) {
3384                 /* net_head list is empty in success case */
3385                 net = list_entry(net_head.next, struct lnet_net, net_list);
3386                 list_del_init(&net->net_list);
3387                 lnet_net_free(net);
3388         }
3389         return rc;
3390 }
3391
3392 int
3393 lnet_dyn_del_net(__u32 net_id)
3394 {
3395         struct lnet_net  *net;
3396         struct lnet_ping_buffer *pbuf;
3397         struct lnet_handle_md ping_mdh;
3398         int               rc;
3399         int               net_ni_count;
3400
3401         /* don't allow userspace to shutdown the LOLND */
3402         if (LNET_NETTYP(net_id) == LOLND)
3403                 return -EINVAL;
3404
3405         mutex_lock(&the_lnet.ln_api_mutex);
3406
3407         lnet_net_lock(0);
3408
3409         net = lnet_get_net_locked(net_id);
3410         if (net == NULL) {
3411                 lnet_net_unlock(0);
3412                 rc = -EINVAL;
3413                 goto out;
3414         }
3415
3416         net_ni_count = lnet_get_net_ni_count_locked(net);
3417
3418         lnet_net_unlock(0);
3419
3420         /* create and link a new ping info, before removing the old one */
3421         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3422                                     lnet_get_ni_count() - net_ni_count, false);
3423         if (rc != 0)
3424                 goto out;
3425
3426         lnet_shutdown_lndnet(net);
3427
3428         if (lnet_count_acceptor_nets() == 0)
3429                 lnet_acceptor_stop();
3430
3431         lnet_ping_target_update(pbuf, ping_mdh);
3432
3433 out:
3434         mutex_unlock(&the_lnet.ln_api_mutex);
3435
3436         return rc;
3437 }
3438
3439 void lnet_incr_dlc_seq(void)
3440 {
3441         atomic_inc(&lnet_dlc_seq_no);
3442 }
3443
3444 __u32 lnet_get_dlc_seq_locked(void)
3445 {
3446         return atomic_read(&lnet_dlc_seq_no);
3447 }
3448
3449 static void
3450 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3451 {
3452         struct lnet_net *net;
3453         struct lnet_ni *ni;
3454
3455         lnet_net_lock(LNET_LOCK_EX);
3456         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3457                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3458                         if (ni->ni_nid == nid || all) {
3459                                 atomic_set(&ni->ni_healthv, value);
3460                                 if (list_empty(&ni->ni_recovery) &&
3461                                     value < LNET_MAX_HEALTH_VALUE) {
3462                                         CERROR("manually adding local NI %s to recovery\n",
3463                                                libcfs_nid2str(ni->ni_nid));
3464                                         list_add_tail(&ni->ni_recovery,
3465                                                       &the_lnet.ln_mt_localNIRecovq);
3466                                         lnet_ni_addref_locked(ni, 0);
3467                                 }
3468                                 if (!all) {
3469                                         lnet_net_unlock(LNET_LOCK_EX);
3470                                         return;
3471                                 }
3472                         }
3473                 }
3474         }
3475         lnet_net_unlock(LNET_LOCK_EX);
3476 }
3477
3478 static int
3479 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3480 {
3481         int cpt, rc = 0;
3482         struct lnet_ni *ni;
3483         lnet_nid_t nid = stats->hlni_nid;
3484
3485         cpt = lnet_net_lock_current();
3486         ni = lnet_nid2ni_locked(nid, cpt);
3487
3488         if (!ni) {
3489                 rc = -ENOENT;
3490                 goto unlock;
3491         }
3492
3493         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3494         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3495         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3496         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3497         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3498         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3499         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3500
3501 unlock:
3502         lnet_net_unlock(cpt);
3503
3504         return rc;
3505 }
3506
3507 static int
3508 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3509 {
3510         struct lnet_ni *ni;
3511         int i = 0;
3512
3513         lnet_net_lock(LNET_LOCK_EX);
3514         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3515                 list->rlst_nid_array[i] = ni->ni_nid;
3516                 i++;
3517                 if (i >= LNET_MAX_SHOW_NUM_NID)
3518                         break;
3519         }
3520         lnet_net_unlock(LNET_LOCK_EX);
3521         list->rlst_num_nids = i;
3522
3523         return 0;
3524 }
3525
3526 static int
3527 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3528 {
3529         struct lnet_peer_ni *lpni;
3530         int i = 0;
3531
3532         lnet_net_lock(LNET_LOCK_EX);
3533         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3534                 list->rlst_nid_array[i] = lpni->lpni_nid;
3535                 i++;
3536                 if (i >= LNET_MAX_SHOW_NUM_NID)
3537                         break;
3538         }
3539         lnet_net_unlock(LNET_LOCK_EX);
3540         list->rlst_num_nids = i;
3541
3542         return 0;
3543 }
3544
3545 /**
3546  * LNet ioctl handler.
3547  *
3548  */
3549 int
3550 LNetCtl(unsigned int cmd, void *arg)
3551 {
3552         struct libcfs_ioctl_data *data = arg;
3553         struct lnet_ioctl_config_data *config;
3554         struct lnet_process_id    id = {0};
3555         struct lnet_ni           *ni;
3556         int                       rc;
3557
3558         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3559                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3560
3561         switch (cmd) {
3562         case IOC_LIBCFS_GET_NI:
3563                 rc = LNetGetId(data->ioc_count, &id);
3564                 data->ioc_nid = id.nid;
3565                 return rc;
3566
3567         case IOC_LIBCFS_FAIL_NID:
3568                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3569
3570         case IOC_LIBCFS_ADD_ROUTE: {
3571                 /* default router sensitivity to 1 */
3572                 unsigned int sensitivity = 1;
3573                 config = arg;
3574
3575                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3576                         return -EINVAL;
3577
3578                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3579                         sensitivity =
3580                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3581                 }
3582
3583                 mutex_lock(&the_lnet.ln_api_mutex);
3584                 rc = lnet_add_route(config->cfg_net,
3585                                     config->cfg_config_u.cfg_route.rtr_hop,
3586                                     config->cfg_nid,
3587                                     config->cfg_config_u.cfg_route.
3588                                         rtr_priority, sensitivity);
3589                 mutex_unlock(&the_lnet.ln_api_mutex);
3590                 return rc;
3591         }
3592
3593         case IOC_LIBCFS_DEL_ROUTE:
3594                 config = arg;
3595
3596                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3597                         return -EINVAL;
3598
3599                 mutex_lock(&the_lnet.ln_api_mutex);
3600                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3601                 mutex_unlock(&the_lnet.ln_api_mutex);
3602                 return rc;
3603
3604         case IOC_LIBCFS_GET_ROUTE:
3605                 config = arg;
3606
3607                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3608                         return -EINVAL;
3609
3610                 mutex_lock(&the_lnet.ln_api_mutex);
3611                 rc = lnet_get_route(config->cfg_count,
3612                                     &config->cfg_net,
3613                                     &config->cfg_config_u.cfg_route.rtr_hop,
3614                                     &config->cfg_nid,
3615                                     &config->cfg_config_u.cfg_route.rtr_flags,
3616                                     &config->cfg_config_u.cfg_route.
3617                                         rtr_priority,
3618                                     &config->cfg_config_u.cfg_route.
3619                                         rtr_sensitivity);
3620                 mutex_unlock(&the_lnet.ln_api_mutex);
3621                 return rc;
3622
3623         case IOC_LIBCFS_GET_LOCAL_NI: {
3624                 struct lnet_ioctl_config_ni *cfg_ni;
3625                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3626                 struct lnet_ioctl_element_stats *stats;
3627                 __u32 tun_size;
3628
3629                 cfg_ni = arg;
3630
3631                 /* get the tunables if they are available */
3632                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3633                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3634                         return -EINVAL;
3635
3636                 stats = (struct lnet_ioctl_element_stats *)
3637                         cfg_ni->lic_bulk;
3638                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3639                                 (cfg_ni->lic_bulk + sizeof(*stats));
3640
3641                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3642                         sizeof(*stats);
3643
3644                 mutex_lock(&the_lnet.ln_api_mutex);
3645                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3646                 mutex_unlock(&the_lnet.ln_api_mutex);
3647                 return rc;
3648         }
3649
3650         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3651                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3652
3653                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3654                         return -EINVAL;
3655
3656                 mutex_lock(&the_lnet.ln_api_mutex);
3657                 rc = lnet_get_ni_stats(msg_stats);
3658                 mutex_unlock(&the_lnet.ln_api_mutex);
3659
3660                 return rc;
3661         }
3662
3663         case IOC_LIBCFS_GET_NET: {
3664                 size_t total = sizeof(*config) +
3665                                sizeof(struct lnet_ioctl_net_config);
3666                 config = arg;
3667
3668                 if (config->cfg_hdr.ioc_len < total)
3669                         return -EINVAL;
3670
3671                 mutex_lock(&the_lnet.ln_api_mutex);
3672                 rc = lnet_get_net_config(config);
3673                 mutex_unlock(&the_lnet.ln_api_mutex);
3674                 return rc;
3675         }
3676
3677         case IOC_LIBCFS_GET_LNET_STATS:
3678         {
3679                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3680
3681                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3682                         return -EINVAL;
3683
3684                 mutex_lock(&the_lnet.ln_api_mutex);
3685                 lnet_counters_get(&lnet_stats->st_cntrs);
3686                 mutex_unlock(&the_lnet.ln_api_mutex);
3687                 return 0;
3688         }
3689
3690         case IOC_LIBCFS_CONFIG_RTR:
3691                 config = arg;
3692
3693                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3694                         return -EINVAL;
3695
3696                 mutex_lock(&the_lnet.ln_api_mutex);
3697                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3698                         rc = lnet_rtrpools_enable();
3699                         mutex_unlock(&the_lnet.ln_api_mutex);
3700                         return rc;
3701                 }
3702                 lnet_rtrpools_disable();
3703                 mutex_unlock(&the_lnet.ln_api_mutex);
3704                 return 0;
3705
3706         case IOC_LIBCFS_ADD_BUF:
3707                 config = arg;
3708
3709                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3710                         return -EINVAL;
3711
3712                 mutex_lock(&the_lnet.ln_api_mutex);
3713                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3714                                                 buf_tiny,
3715                                           config->cfg_config_u.cfg_buffers.
3716                                                 buf_small,
3717                                           config->cfg_config_u.cfg_buffers.
3718                                                 buf_large);
3719                 mutex_unlock(&the_lnet.ln_api_mutex);
3720                 return rc;
3721
3722         case IOC_LIBCFS_SET_NUMA_RANGE: {
3723                 struct lnet_ioctl_set_value *numa;
3724                 numa = arg;
3725                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3726                         return -EINVAL;
3727                 lnet_net_lock(LNET_LOCK_EX);
3728                 lnet_numa_range = numa->sv_value;
3729                 lnet_net_unlock(LNET_LOCK_EX);
3730                 return 0;
3731         }
3732
3733         case IOC_LIBCFS_GET_NUMA_RANGE: {
3734                 struct lnet_ioctl_set_value *numa;
3735                 numa = arg;
3736                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3737                         return -EINVAL;
3738                 numa->sv_value = lnet_numa_range;
3739                 return 0;
3740         }
3741
3742         case IOC_LIBCFS_GET_BUF: {
3743                 struct lnet_ioctl_pool_cfg *pool_cfg;
3744                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3745
3746                 config = arg;
3747
3748                 if (config->cfg_hdr.ioc_len < total)
3749                         return -EINVAL;
3750
3751                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3752
3753                 mutex_lock(&the_lnet.ln_api_mutex);
3754                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3755                 mutex_unlock(&the_lnet.ln_api_mutex);
3756                 return rc;
3757         }
3758
3759         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3760                 struct lnet_ioctl_local_ni_hstats *stats = arg;
3761
3762                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3763                         return -EINVAL;
3764
3765                 mutex_lock(&the_lnet.ln_api_mutex);
3766                 rc = lnet_get_local_ni_hstats(stats);
3767                 mutex_unlock(&the_lnet.ln_api_mutex);
3768
3769                 return rc;
3770         }
3771
3772         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3773                 struct lnet_ioctl_recovery_list *list = arg;
3774                 if (list->rlst_hdr.ioc_len < sizeof(*list))
3775                         return -EINVAL;
3776
3777                 mutex_lock(&the_lnet.ln_api_mutex);
3778                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3779                         rc = lnet_get_local_ni_recovery_list(list);
3780                 else
3781                         rc = lnet_get_peer_ni_recovery_list(list);
3782                 mutex_unlock(&the_lnet.ln_api_mutex);
3783                 return rc;
3784         }
3785
3786         case IOC_LIBCFS_ADD_PEER_NI: {
3787                 struct lnet_ioctl_peer_cfg *cfg = arg;
3788
3789                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3790                         return -EINVAL;
3791
3792                 mutex_lock(&the_lnet.ln_api_mutex);
3793                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3794                                       cfg->prcfg_cfg_nid,
3795                                       cfg->prcfg_mr);
3796                 mutex_unlock(&the_lnet.ln_api_mutex);
3797                 return rc;
3798         }
3799
3800         case IOC_LIBCFS_DEL_PEER_NI: {
3801                 struct lnet_ioctl_peer_cfg *cfg = arg;
3802
3803                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3804                         return -EINVAL;
3805
3806                 mutex_lock(&the_lnet.ln_api_mutex);
3807                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3808                                       cfg->prcfg_cfg_nid);
3809                 mutex_unlock(&the_lnet.ln_api_mutex);
3810                 return rc;
3811         }
3812
3813         case IOC_LIBCFS_GET_PEER_INFO: {
3814                 struct lnet_ioctl_peer *peer_info = arg;
3815
3816                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3817                         return -EINVAL;
3818
3819                 mutex_lock(&the_lnet.ln_api_mutex);
3820                 rc = lnet_get_peer_ni_info(
3821                    peer_info->pr_count,
3822                    &peer_info->pr_nid,
3823                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3824                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3825                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3826                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3827                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3828                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3829                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3830                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3831                 mutex_unlock(&the_lnet.ln_api_mutex);
3832                 return rc;
3833         }
3834
3835         case IOC_LIBCFS_GET_PEER_NI: {
3836                 struct lnet_ioctl_peer_cfg *cfg = arg;
3837
3838                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3839                         return -EINVAL;
3840
3841                 mutex_lock(&the_lnet.ln_api_mutex);
3842                 rc = lnet_get_peer_info(cfg,
3843                                         (void __user *)cfg->prcfg_bulk);
3844                 mutex_unlock(&the_lnet.ln_api_mutex);
3845                 return rc;
3846         }
3847
3848         case IOC_LIBCFS_GET_PEER_LIST: {
3849                 struct lnet_ioctl_peer_cfg *cfg = arg;
3850
3851                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3852                         return -EINVAL;
3853
3854                 mutex_lock(&the_lnet.ln_api_mutex);
3855                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3856                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3857                 mutex_unlock(&the_lnet.ln_api_mutex);
3858                 return rc;
3859         }
3860
3861         case IOC_LIBCFS_SET_HEALHV: {
3862                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3863                 int value;
3864                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3865                         return -EINVAL;
3866                 if (cfg->rh_value < 0 ||
3867                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3868                         value = LNET_MAX_HEALTH_VALUE;
3869                 else
3870                         value = cfg->rh_value;
3871                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3872                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3873                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3874                 mutex_lock(&the_lnet.ln_api_mutex);
3875                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3876                         lnet_ni_set_healthv(cfg->rh_nid, value,
3877                                              cfg->rh_all);
3878                 else
3879                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3880                                                   cfg->rh_all);
3881                 mutex_unlock(&the_lnet.ln_api_mutex);
3882                 return 0;
3883         }
3884
3885         case IOC_LIBCFS_NOTIFY_ROUTER: {
3886                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3887
3888                 /* The deadline passed in by the user should be some time in
3889                  * seconds in the future since the UNIX epoch. We have to map
3890                  * that deadline to the wall clock.
3891                  */
3892                 deadline += ktime_get_seconds();
3893                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3894                                    deadline);
3895         }
3896
3897         case IOC_LIBCFS_LNET_DIST:
3898                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3899                 if (rc < 0 && rc != -EHOSTUNREACH)
3900                         return rc;
3901
3902                 data->ioc_u32[0] = rc;
3903                 return 0;
3904
3905         case IOC_LIBCFS_TESTPROTOCOMPAT:
3906                 lnet_net_lock(LNET_LOCK_EX);
3907                 the_lnet.ln_testprotocompat = data->ioc_flags;
3908                 lnet_net_unlock(LNET_LOCK_EX);
3909                 return 0;
3910
3911         case IOC_LIBCFS_LNET_FAULT:
3912                 return lnet_fault_ctl(data->ioc_flags, data);
3913
3914         case IOC_LIBCFS_PING: {
3915                 signed long timeout;
3916
3917                 id.nid = data->ioc_nid;
3918                 id.pid = data->ioc_u32[0];
3919
3920                 /* If timeout is negative then set default of 3 minutes */
3921                 if (((s32)data->ioc_u32[1] <= 0) ||
3922                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3923                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3924                 else
3925                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3926
3927                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3928                                data->ioc_plen1 / sizeof(struct lnet_process_id));
3929
3930                 if (rc < 0)
3931                         return rc;
3932
3933                 data->ioc_count = rc;
3934                 return 0;
3935         }
3936
3937         case IOC_LIBCFS_PING_PEER: {
3938                 struct lnet_ioctl_ping_data *ping = arg;
3939                 struct lnet_peer *lp;
3940                 signed long timeout;
3941
3942                 /* If timeout is negative then set default of 3 minutes */
3943                 if (((s32)ping->op_param) <= 0 ||
3944                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3945                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3946                 else
3947                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3948
3949                 rc = lnet_ping(ping->ping_id, timeout,
3950                                ping->ping_buf,
3951                                ping->ping_count);
3952                 if (rc < 0)
3953                         return rc;
3954
3955                 mutex_lock(&the_lnet.ln_api_mutex);
3956                 lp = lnet_find_peer(ping->ping_id.nid);
3957                 if (lp) {
3958                         ping->ping_id.nid = lp->lp_primary_nid;
3959                         ping->mr_info = lnet_peer_is_multi_rail(lp);
3960                         lnet_peer_decref_locked(lp);
3961                 }
3962                 mutex_unlock(&the_lnet.ln_api_mutex);
3963
3964                 ping->ping_count = rc;
3965                 return 0;
3966         }
3967
3968         case IOC_LIBCFS_DISCOVER: {
3969                 struct lnet_ioctl_ping_data *discover = arg;
3970                 struct lnet_peer *lp;
3971
3972                 rc = lnet_discover(discover->ping_id, discover->op_param,
3973                                    discover->ping_buf,
3974                                    discover->ping_count);
3975                 if (rc < 0)
3976                         return rc;
3977
3978                 mutex_lock(&the_lnet.ln_api_mutex);
3979                 lp = lnet_find_peer(discover->ping_id.nid);
3980                 if (lp) {
3981                         discover->ping_id.nid = lp->lp_primary_nid;
3982                         discover->mr_info = lnet_peer_is_multi_rail(lp);
3983                         lnet_peer_decref_locked(lp);
3984                 }
3985                 mutex_unlock(&the_lnet.ln_api_mutex);
3986
3987                 discover->ping_count = rc;
3988                 return 0;
3989         }
3990
3991         default:
3992                 ni = lnet_net2ni_addref(data->ioc_net);
3993                 if (ni == NULL)
3994                         return -EINVAL;
3995
3996                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
3997                         rc = -EINVAL;
3998                 else
3999                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4000
4001                 lnet_ni_decref(ni);
4002                 return rc;
4003         }
4004         /* not reached */
4005 }
4006 EXPORT_SYMBOL(LNetCtl);
4007
4008 void LNetDebugPeer(struct lnet_process_id id)
4009 {
4010         lnet_debug_peer(id.nid);
4011 }
4012 EXPORT_SYMBOL(LNetDebugPeer);
4013
4014 /**
4015  * Determine if the specified peer \a nid is on the local node.
4016  *
4017  * \param nid   peer nid to check
4018  *
4019  * \retval true         If peer NID is on the local node.
4020  * \retval false        If peer NID is not on the local node.
4021  */
4022 bool LNetIsPeerLocal(lnet_nid_t nid)
4023 {
4024         struct lnet_net *net;
4025         struct lnet_ni *ni;
4026         int cpt;
4027
4028         cpt = lnet_net_lock_current();
4029         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4030                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4031                         if (ni->ni_nid == nid) {
4032                                 lnet_net_unlock(cpt);
4033                                 return true;
4034                         }
4035                 }
4036         }
4037         lnet_net_unlock(cpt);
4038
4039         return false;
4040 }
4041 EXPORT_SYMBOL(LNetIsPeerLocal);
4042
4043 /**
4044  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4045  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4046  *
4047  * \param index Index of the interface to look up.
4048  * \param id On successful return, this location will hold the
4049  * struct lnet_process_id ID of the interface.
4050  *
4051  * \retval 0 If an interface exists at \a index.
4052  * \retval -ENOENT If no interface has been found.
4053  */
4054 int
4055 LNetGetId(unsigned int index, struct lnet_process_id *id)
4056 {
4057         struct lnet_ni   *ni;
4058         struct lnet_net  *net;
4059         int               cpt;
4060         int               rc = -ENOENT;
4061
4062         LASSERT(the_lnet.ln_refcount > 0);
4063
4064         cpt = lnet_net_lock_current();
4065
4066         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4067                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4068                         if (index-- != 0)
4069                                 continue;
4070
4071                         id->nid = ni->ni_nid;
4072                         id->pid = the_lnet.ln_pid;
4073                         rc = 0;
4074   &n