Whamcloud - gitweb
LU-9859 libcfs: switch to cpumask_var_t
[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 struct list_head **
1104 lnet_create_array_of_queues(void)
1105 {
1106         struct list_head **qs;
1107         struct list_head *q;
1108         int i;
1109
1110         qs = cfs_percpt_alloc(lnet_cpt_table(),
1111                               sizeof(struct list_head));
1112         if (!qs) {
1113                 CERROR("Failed to allocate queues\n");
1114                 return NULL;
1115         }
1116
1117         cfs_percpt_for_each(q, i, qs)
1118                 INIT_LIST_HEAD(q);
1119
1120         return qs;
1121 }
1122
1123 static int lnet_unprepare(void);
1124
1125 static int
1126 lnet_prepare(lnet_pid_t requested_pid)
1127 {
1128         /* Prepare to bring up the network */
1129         struct lnet_res_container **recs;
1130         int                       rc = 0;
1131
1132         if (requested_pid == LNET_PID_ANY) {
1133                 /* Don't instantiate LNET just for me */
1134                 return -ENETDOWN;
1135         }
1136
1137         LASSERT(the_lnet.ln_refcount == 0);
1138
1139         the_lnet.ln_routing = 0;
1140
1141         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1142         the_lnet.ln_pid = requested_pid;
1143
1144         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1145         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1146         INIT_LIST_HEAD(&the_lnet.ln_nets);
1147         INIT_LIST_HEAD(&the_lnet.ln_routers);
1148         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1149         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1150         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1151         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1152         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1153         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1154         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1155         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1156         LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1157         init_completion(&the_lnet.ln_started);
1158
1159         rc = lnet_descriptor_setup();
1160         if (rc != 0)
1161                 goto failed;
1162
1163         rc = lnet_create_remote_nets_table();
1164         if (rc != 0)
1165                 goto failed;
1166
1167         /*
1168          * NB the interface cookie in wire handles guards against delayed
1169          * replies and ACKs appearing valid after reboot.
1170          */
1171         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1172
1173         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1174                                                 sizeof(struct lnet_counters));
1175         if (the_lnet.ln_counters == NULL) {
1176                 CERROR("Failed to allocate counters for LNet\n");
1177                 rc = -ENOMEM;
1178                 goto failed;
1179         }
1180
1181         rc = lnet_peer_tables_create();
1182         if (rc != 0)
1183                 goto failed;
1184
1185         rc = lnet_msg_containers_create();
1186         if (rc != 0)
1187                 goto failed;
1188
1189         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1190                                       LNET_COOKIE_TYPE_EQ);
1191         if (rc != 0)
1192                 goto failed;
1193
1194         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
1195         if (recs == NULL) {
1196                 rc = -ENOMEM;
1197                 goto failed;
1198         }
1199
1200         the_lnet.ln_me_containers = recs;
1201
1202         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1203         if (recs == NULL) {
1204                 rc = -ENOMEM;
1205                 goto failed;
1206         }
1207
1208         the_lnet.ln_md_containers = recs;
1209
1210         rc = lnet_portals_create();
1211         if (rc != 0) {
1212                 CERROR("Failed to create portals for LNet: %d\n", rc);
1213                 goto failed;
1214         }
1215
1216         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1217         if (!the_lnet.ln_mt_zombie_rstqs) {
1218                 rc = -ENOMEM;
1219                 goto failed;
1220         }
1221
1222         return 0;
1223
1224  failed:
1225         lnet_unprepare();
1226         return rc;
1227 }
1228
1229 static int
1230 lnet_unprepare (void)
1231 {
1232         int rc;
1233
1234         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1235          * have shut down already, so it is safe to unlink and free all
1236          * descriptors, even those that appear committed to a network op (eg MD
1237          * with non-zero pending count) */
1238
1239         lnet_fail_nid(LNET_NID_ANY, 0);
1240
1241         LASSERT(the_lnet.ln_refcount == 0);
1242         LASSERT(list_empty(&the_lnet.ln_test_peers));
1243         LASSERT(list_empty(&the_lnet.ln_nets));
1244
1245         if (the_lnet.ln_mt_zombie_rstqs) {
1246                 lnet_clean_zombie_rstqs();
1247                 the_lnet.ln_mt_zombie_rstqs = NULL;
1248         }
1249
1250         if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1251                 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1252                 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1253                 LASSERT(rc == 0);
1254         }
1255
1256         lnet_portals_destroy();
1257
1258         if (the_lnet.ln_md_containers != NULL) {
1259                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1260                 the_lnet.ln_md_containers = NULL;
1261         }
1262
1263         if (the_lnet.ln_me_containers != NULL) {
1264                 lnet_res_containers_destroy(the_lnet.ln_me_containers);
1265                 the_lnet.ln_me_containers = NULL;
1266         }
1267
1268         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1269
1270         lnet_msg_containers_destroy();
1271         lnet_peer_uninit();
1272         lnet_rtrpools_free(0);
1273
1274         if (the_lnet.ln_counters != NULL) {
1275                 cfs_percpt_free(the_lnet.ln_counters);
1276                 the_lnet.ln_counters = NULL;
1277         }
1278         lnet_destroy_remote_nets_table();
1279         lnet_descriptor_cleanup();
1280
1281         return 0;
1282 }
1283
1284 struct lnet_ni  *
1285 lnet_net2ni_locked(__u32 net_id, int cpt)
1286 {
1287         struct lnet_ni   *ni;
1288         struct lnet_net  *net;
1289
1290         LASSERT(cpt != LNET_LOCK_EX);
1291
1292         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1293                 if (net->net_id == net_id) {
1294                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1295                                         ni_netlist);
1296                         return ni;
1297                 }
1298         }
1299
1300         return NULL;
1301 }
1302
1303 struct lnet_ni *
1304 lnet_net2ni_addref(__u32 net)
1305 {
1306         struct lnet_ni *ni;
1307
1308         lnet_net_lock(0);
1309         ni = lnet_net2ni_locked(net, 0);
1310         if (ni)
1311                 lnet_ni_addref_locked(ni, 0);
1312         lnet_net_unlock(0);
1313
1314         return ni;
1315 }
1316 EXPORT_SYMBOL(lnet_net2ni_addref);
1317
1318 struct lnet_net *
1319 lnet_get_net_locked(__u32 net_id)
1320 {
1321         struct lnet_net  *net;
1322
1323         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1324                 if (net->net_id == net_id)
1325                         return net;
1326         }
1327
1328         return NULL;
1329 }
1330
1331 unsigned int
1332 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1333 {
1334         __u64           key = nid;
1335         unsigned int    val;
1336
1337         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1338
1339         if (number == 1)
1340                 return 0;
1341
1342         val = hash_long(key, LNET_CPT_BITS);
1343         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1344         if (val < number)
1345                 return val;
1346
1347         return (unsigned int)(key + val + (val >> 1)) % number;
1348 }
1349
1350 int
1351 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1352 {
1353         struct lnet_net *net;
1354
1355         /* must called with hold of lnet_net_lock */
1356         if (LNET_CPT_NUMBER == 1)
1357                 return 0; /* the only one */
1358
1359         /*
1360          * If NI is provided then use the CPT identified in the NI cpt
1361          * list if one exists. If one doesn't exist, then that NI is
1362          * associated with all CPTs and it follows that the net it belongs
1363          * to is implicitly associated with all CPTs, so just hash the nid
1364          * and return that.
1365          */
1366         if (ni != NULL) {
1367                 if (ni->ni_cpts != NULL)
1368                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1369                                                              ni->ni_ncpts)];
1370                 else
1371                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1372         }
1373
1374         /* no NI provided so look at the net */
1375         net = lnet_get_net_locked(LNET_NIDNET(nid));
1376
1377         if (net != NULL && net->net_cpts != NULL) {
1378                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1379         }
1380
1381         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1382 }
1383
1384 int
1385 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1386 {
1387         int     cpt;
1388         int     cpt2;
1389
1390         if (LNET_CPT_NUMBER == 1)
1391                 return 0; /* the only one */
1392
1393         cpt = lnet_net_lock_current();
1394
1395         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1396
1397         lnet_net_unlock(cpt);
1398
1399         return cpt2;
1400 }
1401 EXPORT_SYMBOL(lnet_cpt_of_nid);
1402
1403 int
1404 lnet_islocalnet_locked(__u32 net_id)
1405 {
1406         struct lnet_net *net;
1407         bool local;
1408
1409         net = lnet_get_net_locked(net_id);
1410
1411         local = net != NULL;
1412
1413         return local;
1414 }
1415
1416 int
1417 lnet_islocalnet(__u32 net_id)
1418 {
1419         int cpt;
1420         bool local;
1421
1422         cpt = lnet_net_lock_current();
1423
1424         local = lnet_islocalnet_locked(net_id);
1425
1426         lnet_net_unlock(cpt);
1427
1428         return local;
1429 }
1430
1431 struct lnet_ni  *
1432 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1433 {
1434         struct lnet_net  *net;
1435         struct lnet_ni   *ni;
1436
1437         LASSERT(cpt != LNET_LOCK_EX);
1438
1439         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1440                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1441                         if (ni->ni_nid == nid)
1442                                 return ni;
1443                 }
1444         }
1445
1446         return NULL;
1447 }
1448
1449 struct lnet_ni *
1450 lnet_nid2ni_addref(lnet_nid_t nid)
1451 {
1452         struct lnet_ni *ni;
1453
1454         lnet_net_lock(0);
1455         ni = lnet_nid2ni_locked(nid, 0);
1456         if (ni)
1457                 lnet_ni_addref_locked(ni, 0);
1458         lnet_net_unlock(0);
1459
1460         return ni;
1461 }
1462 EXPORT_SYMBOL(lnet_nid2ni_addref);
1463
1464 int
1465 lnet_islocalnid(lnet_nid_t nid)
1466 {
1467         struct lnet_ni  *ni;
1468         int             cpt;
1469
1470         cpt = lnet_net_lock_current();
1471         ni = lnet_nid2ni_locked(nid, cpt);
1472         lnet_net_unlock(cpt);
1473
1474         return ni != NULL;
1475 }
1476
1477 int
1478 lnet_count_acceptor_nets(void)
1479 {
1480         /* Return the # of NIs that need the acceptor. */
1481         int              count = 0;
1482         struct lnet_net  *net;
1483         int              cpt;
1484
1485         cpt = lnet_net_lock_current();
1486         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1487                 /* all socklnd type networks should have the acceptor
1488                  * thread started */
1489                 if (net->net_lnd->lnd_accept != NULL)
1490                         count++;
1491         }
1492
1493         lnet_net_unlock(cpt);
1494
1495         return count;
1496 }
1497
1498 struct lnet_ping_buffer *
1499 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1500 {
1501         struct lnet_ping_buffer *pbuf;
1502
1503         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1504         if (pbuf) {
1505                 pbuf->pb_nnis = nnis;
1506                 atomic_set(&pbuf->pb_refcnt, 1);
1507         }
1508
1509         return pbuf;
1510 }
1511
1512 void
1513 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1514 {
1515         LASSERT(lnet_ping_buffer_numref(pbuf) == 0);
1516         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1517 }
1518
1519 static struct lnet_ping_buffer *
1520 lnet_ping_target_create(int nnis)
1521 {
1522         struct lnet_ping_buffer *pbuf;
1523
1524         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1525         if (pbuf == NULL) {
1526                 CERROR("Can't allocate ping source [%d]\n", nnis);
1527                 return NULL;
1528         }
1529
1530         pbuf->pb_info.pi_nnis = nnis;
1531         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1532         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1533         pbuf->pb_info.pi_features =
1534                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1535
1536         return pbuf;
1537 }
1538
1539 static inline int
1540 lnet_get_net_ni_count_locked(struct lnet_net *net)
1541 {
1542         struct lnet_ni  *ni;
1543         int             count = 0;
1544
1545         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1546                 count++;
1547
1548         return count;
1549 }
1550
1551 static inline int
1552 lnet_get_net_ni_count_pre(struct lnet_net *net)
1553 {
1554         struct lnet_ni  *ni;
1555         int             count = 0;
1556
1557         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1558                 count++;
1559
1560         return count;
1561 }
1562
1563 static inline int
1564 lnet_get_ni_count(void)
1565 {
1566         struct lnet_ni  *ni;
1567         struct lnet_net *net;
1568         int             count = 0;
1569
1570         lnet_net_lock(0);
1571
1572         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1573                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1574                         count++;
1575         }
1576
1577         lnet_net_unlock(0);
1578
1579         return count;
1580 }
1581
1582 int
1583 lnet_get_net_count(void)
1584 {
1585         struct lnet_net *net;
1586         int count = 0;
1587
1588         lnet_net_lock(0);
1589
1590         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1591                 count++;
1592         }
1593
1594         lnet_net_unlock(0);
1595
1596         return count;
1597 }
1598
1599 void
1600 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1601 {
1602         struct lnet_ni_status *stat;
1603         int nnis;
1604         int i;
1605
1606         __swab32s(&pbuf->pb_info.pi_magic);
1607         __swab32s(&pbuf->pb_info.pi_features);
1608         __swab32s(&pbuf->pb_info.pi_pid);
1609         __swab32s(&pbuf->pb_info.pi_nnis);
1610         nnis = pbuf->pb_info.pi_nnis;
1611         if (nnis > pbuf->pb_nnis)
1612                 nnis = pbuf->pb_nnis;
1613         for (i = 0; i < nnis; i++) {
1614                 stat = &pbuf->pb_info.pi_ni[i];
1615                 __swab64s(&stat->ns_nid);
1616                 __swab32s(&stat->ns_status);
1617         }
1618         return;
1619 }
1620
1621 int
1622 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1623 {
1624         if (!pinfo)
1625                 return -EINVAL;
1626         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1627                 return -EPROTO;
1628         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1629                 return -EPROTO;
1630         /* Loopback is guaranteed to be present */
1631         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1632                 return -ERANGE;
1633         if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1634                 return -EPROTO;
1635         return 0;
1636 }
1637
1638 static void
1639 lnet_ping_target_destroy(void)
1640 {
1641         struct lnet_net *net;
1642         struct lnet_ni  *ni;
1643
1644         lnet_net_lock(LNET_LOCK_EX);
1645
1646         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1647                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1648                         lnet_ni_lock(ni);
1649                         ni->ni_status = NULL;
1650                         lnet_ni_unlock(ni);
1651                 }
1652         }
1653
1654         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1655         the_lnet.ln_ping_target = NULL;
1656
1657         lnet_net_unlock(LNET_LOCK_EX);
1658 }
1659
1660 static void
1661 lnet_ping_target_event_handler(struct lnet_event *event)
1662 {
1663         struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1664
1665         if (event->unlinked)
1666                 lnet_ping_buffer_decref(pbuf);
1667 }
1668
1669 static int
1670 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1671                        struct lnet_handle_md *ping_mdh,
1672                        int ni_count, bool set_eq)
1673 {
1674         struct lnet_process_id id = {
1675                 .nid = LNET_NID_ANY,
1676                 .pid = LNET_PID_ANY
1677         };
1678         struct lnet_handle_me me_handle;
1679         struct lnet_md md = { NULL };
1680         int rc, rc2;
1681
1682         if (set_eq) {
1683                 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1684                                  &the_lnet.ln_ping_target_eq);
1685                 if (rc != 0) {
1686                         CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1687                         return rc;
1688                 }
1689         }
1690
1691         *ppbuf = lnet_ping_target_create(ni_count);
1692         if (*ppbuf == NULL) {
1693                 rc = -ENOMEM;
1694                 goto fail_free_eq;
1695         }
1696
1697         /* Ping target ME/MD */
1698         rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1699                           LNET_PROTO_PING_MATCHBITS, 0,
1700                           LNET_UNLINK, LNET_INS_AFTER,
1701                           &me_handle);
1702         if (rc != 0) {
1703                 CERROR("Can't create ping target ME: %d\n", rc);
1704                 goto fail_decref_ping_buffer;
1705         }
1706
1707         /* initialize md content */
1708         md.start     = &(*ppbuf)->pb_info;
1709         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1710         md.threshold = LNET_MD_THRESH_INF;
1711         md.max_size  = 0;
1712         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1713                        LNET_MD_MANAGE_REMOTE;
1714         md.eq_handle = the_lnet.ln_ping_target_eq;
1715         md.user_ptr  = *ppbuf;
1716
1717         rc = LNetMDAttach(me_handle, md, LNET_RETAIN, ping_mdh);
1718         if (rc != 0) {
1719                 CERROR("Can't attach ping target MD: %d\n", rc);
1720                 goto fail_unlink_ping_me;
1721         }
1722         lnet_ping_buffer_addref(*ppbuf);
1723
1724         return 0;
1725
1726 fail_unlink_ping_me:
1727         rc2 = LNetMEUnlink(me_handle);
1728         LASSERT(rc2 == 0);
1729 fail_decref_ping_buffer:
1730         LASSERT(lnet_ping_buffer_numref(*ppbuf) == 1);
1731         lnet_ping_buffer_decref(*ppbuf);
1732         *ppbuf = NULL;
1733 fail_free_eq:
1734         if (set_eq) {
1735                 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1736                 LASSERT(rc2 == 0);
1737         }
1738         return rc;
1739 }
1740
1741 static void
1742 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1743                     struct lnet_handle_md *ping_mdh)
1744 {
1745         sigset_t        blocked = cfs_block_allsigs();
1746
1747         LNetMDUnlink(*ping_mdh);
1748         LNetInvalidateMDHandle(ping_mdh);
1749
1750         /* NB the MD could be busy; this just starts the unlink */
1751         while (lnet_ping_buffer_numref(pbuf) > 1) {
1752                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1753                 set_current_state(TASK_UNINTERRUPTIBLE);
1754                 schedule_timeout(cfs_time_seconds(1));
1755         }
1756
1757         cfs_restore_sigs(blocked);
1758 }
1759
1760 static void
1761 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1762 {
1763         struct lnet_ni          *ni;
1764         struct lnet_net         *net;
1765         struct lnet_ni_status *ns;
1766         int                     i;
1767         int                     rc;
1768
1769         i = 0;
1770         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1771                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1772                         LASSERT(i < pbuf->pb_nnis);
1773
1774                         ns = &pbuf->pb_info.pi_ni[i];
1775
1776                         ns->ns_nid = ni->ni_nid;
1777
1778                         lnet_ni_lock(ni);
1779                         ns->ns_status = (ni->ni_status != NULL) ?
1780                                          ni->ni_status->ns_status :
1781                                                 LNET_NI_STATUS_UP;
1782                         ni->ni_status = ns;
1783                         lnet_ni_unlock(ni);
1784
1785                         i++;
1786                 }
1787         }
1788         /*
1789          * We (ab)use the ns_status of the loopback interface to
1790          * transmit the sequence number. The first interface listed
1791          * must be the loopback interface.
1792          */
1793         rc = lnet_ping_info_validate(&pbuf->pb_info);
1794         if (rc) {
1795                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1796                 LBUG();
1797         }
1798         LNET_PING_BUFFER_SEQNO(pbuf) =
1799                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1800 }
1801
1802 static void
1803 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1804                         struct lnet_handle_md ping_mdh)
1805 {
1806         struct lnet_ping_buffer *old_pbuf = NULL;
1807         struct lnet_handle_md old_ping_md;
1808
1809         /* switch the NIs to point to the new ping info created */
1810         lnet_net_lock(LNET_LOCK_EX);
1811
1812         if (!the_lnet.ln_routing)
1813                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1814         if (!lnet_peer_discovery_disabled)
1815                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1816
1817         /* Ensure only known feature bits have been set. */
1818         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1819         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1820
1821         lnet_ping_target_install_locked(pbuf);
1822
1823         if (the_lnet.ln_ping_target) {
1824                 old_pbuf = the_lnet.ln_ping_target;
1825                 old_ping_md = the_lnet.ln_ping_target_md;
1826         }
1827         the_lnet.ln_ping_target_md = ping_mdh;
1828         the_lnet.ln_ping_target = pbuf;
1829
1830         lnet_net_unlock(LNET_LOCK_EX);
1831
1832         if (old_pbuf) {
1833                 /* unlink and free the old ping info */
1834                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1835                 lnet_ping_buffer_decref(old_pbuf);
1836         }
1837
1838         lnet_push_update_to_peers(0);
1839 }
1840
1841 static void
1842 lnet_ping_target_fini(void)
1843 {
1844         int             rc;
1845
1846         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1847                             &the_lnet.ln_ping_target_md);
1848
1849         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1850         LASSERT(rc == 0);
1851
1852         lnet_ping_target_destroy();
1853 }
1854
1855 /* Resize the push target. */
1856 int lnet_push_target_resize(void)
1857 {
1858         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1859         struct lnet_md md = { NULL };
1860         struct lnet_handle_me meh;
1861         struct lnet_handle_md mdh;
1862         struct lnet_handle_md old_mdh;
1863         struct lnet_ping_buffer *pbuf;
1864         struct lnet_ping_buffer *old_pbuf;
1865         int nnis = the_lnet.ln_push_target_nnis;
1866         int rc;
1867
1868         if (nnis <= 0) {
1869                 rc = -EINVAL;
1870                 goto fail_return;
1871         }
1872 again:
1873         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1874         if (!pbuf) {
1875                 rc = -ENOMEM;
1876                 goto fail_return;
1877         }
1878
1879         rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1880                           LNET_PROTO_PING_MATCHBITS, 0,
1881                           LNET_UNLINK, LNET_INS_AFTER,
1882                           &meh);
1883         if (rc) {
1884                 CERROR("Can't create push target ME: %d\n", rc);
1885                 goto fail_decref_pbuf;
1886         }
1887
1888         /* initialize md content */
1889         md.start     = &pbuf->pb_info;
1890         md.length    = LNET_PING_INFO_SIZE(nnis);
1891         md.threshold = LNET_MD_THRESH_INF;
1892         md.max_size  = 0;
1893         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1894                        LNET_MD_MANAGE_REMOTE;
1895         md.user_ptr  = pbuf;
1896         md.eq_handle = the_lnet.ln_push_target_eq;
1897
1898         rc = LNetMDAttach(meh, md, LNET_RETAIN, &mdh);
1899         if (rc) {
1900                 CERROR("Can't attach push MD: %d\n", rc);
1901                 goto fail_unlink_meh;
1902         }
1903         lnet_ping_buffer_addref(pbuf);
1904
1905         lnet_net_lock(LNET_LOCK_EX);
1906         old_pbuf = the_lnet.ln_push_target;
1907         old_mdh = the_lnet.ln_push_target_md;
1908         the_lnet.ln_push_target = pbuf;
1909         the_lnet.ln_push_target_md = mdh;
1910         lnet_net_unlock(LNET_LOCK_EX);
1911
1912         if (old_pbuf) {
1913                 LNetMDUnlink(old_mdh);
1914                 lnet_ping_buffer_decref(old_pbuf);
1915         }
1916
1917         if (nnis < the_lnet.ln_push_target_nnis)
1918                 goto again;
1919
1920         CDEBUG(D_NET, "nnis %d success\n", nnis);
1921
1922         return 0;
1923
1924 fail_unlink_meh:
1925         LNetMEUnlink(meh);
1926 fail_decref_pbuf:
1927         lnet_ping_buffer_decref(pbuf);
1928 fail_return:
1929         CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1930         return rc;
1931 }
1932
1933 static void lnet_push_target_event_handler(struct lnet_event *ev)
1934 {
1935         struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1936
1937         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1938                 lnet_swap_pinginfo(pbuf);
1939
1940         lnet_peer_push_event(ev);
1941         if (ev->unlinked)
1942                 lnet_ping_buffer_decref(pbuf);
1943 }
1944
1945 /* Initialize the push target. */
1946 static int lnet_push_target_init(void)
1947 {
1948         int rc;
1949
1950         if (the_lnet.ln_push_target)
1951                 return -EALREADY;
1952
1953         rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1954                          &the_lnet.ln_push_target_eq);
1955         if (rc) {
1956                 CERROR("Can't allocated push target EQ: %d\n", rc);
1957                 return rc;
1958         }
1959
1960         /* Start at the required minimum, we'll enlarge if required. */
1961         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1962
1963         rc = lnet_push_target_resize();
1964
1965         if (rc) {
1966                 LNetEQFree(the_lnet.ln_push_target_eq);
1967                 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1968         }
1969
1970         return rc;
1971 }
1972
1973 /* Clean up the push target. */
1974 static void lnet_push_target_fini(void)
1975 {
1976         if (!the_lnet.ln_push_target)
1977                 return;
1978
1979         /* Unlink and invalidate to prevent new references. */
1980         LNetMDUnlink(the_lnet.ln_push_target_md);
1981         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1982
1983         /* Wait for the unlink to complete. */
1984         while (lnet_ping_buffer_numref(the_lnet.ln_push_target) > 1) {
1985                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1986                 set_current_state(TASK_UNINTERRUPTIBLE);
1987                 schedule_timeout(cfs_time_seconds(1));
1988         }
1989
1990         lnet_ping_buffer_decref(the_lnet.ln_push_target);
1991         the_lnet.ln_push_target = NULL;
1992         the_lnet.ln_push_target_nnis = 0;
1993
1994         LNetEQFree(the_lnet.ln_push_target_eq);
1995         LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1996 }
1997
1998 static int
1999 lnet_ni_tq_credits(struct lnet_ni *ni)
2000 {
2001         int     credits;
2002
2003         LASSERT(ni->ni_ncpts >= 1);
2004
2005         if (ni->ni_ncpts == 1)
2006                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2007
2008         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2009         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2010         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2011
2012         return credits;
2013 }
2014
2015 static void
2016 lnet_ni_unlink_locked(struct lnet_ni *ni)
2017 {
2018         /* move it to zombie list and nobody can find it anymore */
2019         LASSERT(!list_empty(&ni->ni_netlist));
2020         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2021         lnet_ni_decref_locked(ni, 0);
2022 }
2023
2024 static void
2025 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2026 {
2027         int             i;
2028         int             islo;
2029         struct lnet_ni  *ni;
2030         struct list_head *zombie_list = &net->net_ni_zombie;
2031
2032         /*
2033          * Now wait for the NIs I just nuked to show up on the zombie
2034          * list and shut them down in guaranteed thread context
2035          */
2036         i = 2;
2037         while (!list_empty(zombie_list)) {
2038                 int     *ref;
2039                 int     j;
2040
2041                 ni = list_entry(zombie_list->next,
2042                                 struct lnet_ni, ni_netlist);
2043                 list_del_init(&ni->ni_netlist);
2044                 /* the ni should be in deleting state. If it's not it's
2045                  * a bug */
2046                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2047                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2048                         if (*ref == 0)
2049                                 continue;
2050                         /* still busy, add it back to zombie list */
2051                         list_add(&ni->ni_netlist, zombie_list);
2052                         break;
2053                 }
2054
2055                 if (!list_empty(&ni->ni_netlist)) {
2056                         lnet_net_unlock(LNET_LOCK_EX);
2057                         ++i;
2058                         if ((i & (-i)) == i) {
2059                                 CDEBUG(D_WARNING,
2060                                        "Waiting for zombie LNI %s\n",
2061                                        libcfs_nid2str(ni->ni_nid));
2062                         }
2063                         set_current_state(TASK_UNINTERRUPTIBLE);
2064                         schedule_timeout(cfs_time_seconds(1));
2065                         lnet_net_lock(LNET_LOCK_EX);
2066                         continue;
2067                 }
2068
2069                 lnet_net_unlock(LNET_LOCK_EX);
2070
2071                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2072
2073                 LASSERT(!in_interrupt());
2074                 (net->net_lnd->lnd_shutdown)(ni);
2075
2076                 if (!islo)
2077                         CDEBUG(D_LNI, "Removed LNI %s\n",
2078                               libcfs_nid2str(ni->ni_nid));
2079
2080                 lnet_ni_free(ni);
2081                 i = 2;
2082                 lnet_net_lock(LNET_LOCK_EX);
2083         }
2084 }
2085
2086 /* shutdown down the NI and release refcount */
2087 static void
2088 lnet_shutdown_lndni(struct lnet_ni *ni)
2089 {
2090         int i;
2091         struct lnet_net *net = ni->ni_net;
2092
2093         lnet_net_lock(LNET_LOCK_EX);
2094         lnet_ni_lock(ni);
2095         ni->ni_state = LNET_NI_STATE_DELETING;
2096         lnet_ni_unlock(ni);
2097         lnet_ni_unlink_locked(ni);
2098         lnet_incr_dlc_seq();
2099         lnet_net_unlock(LNET_LOCK_EX);
2100
2101         /* clear messages for this NI on the lazy portal */
2102         for (i = 0; i < the_lnet.ln_nportals; i++)
2103                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2104
2105         lnet_net_lock(LNET_LOCK_EX);
2106         lnet_clear_zombies_nis_locked(net);
2107         lnet_net_unlock(LNET_LOCK_EX);
2108 }
2109
2110 static void
2111 lnet_shutdown_lndnet(struct lnet_net *net)
2112 {
2113         struct lnet_ni *ni;
2114
2115         lnet_net_lock(LNET_LOCK_EX);
2116
2117         net->net_state = LNET_NET_STATE_DELETING;
2118
2119         list_del_init(&net->net_list);
2120
2121         while (!list_empty(&net->net_ni_list)) {
2122                 ni = list_entry(net->net_ni_list.next,
2123                                 struct lnet_ni, ni_netlist);
2124                 lnet_net_unlock(LNET_LOCK_EX);
2125                 lnet_shutdown_lndni(ni);
2126                 lnet_net_lock(LNET_LOCK_EX);
2127         }
2128
2129         lnet_net_unlock(LNET_LOCK_EX);
2130
2131         /* Do peer table cleanup for this net */
2132         lnet_peer_tables_cleanup(net);
2133
2134         lnet_net_lock(LNET_LOCK_EX);
2135         /*
2136          * decrement ref count on lnd only when the entire network goes
2137          * away
2138          */
2139         net->net_lnd->lnd_refcount--;
2140
2141         lnet_net_unlock(LNET_LOCK_EX);
2142
2143         lnet_net_free(net);
2144 }
2145
2146 static void
2147 lnet_shutdown_lndnets(void)
2148 {
2149         struct lnet_net *net;
2150         struct list_head resend;
2151         struct lnet_msg *msg, *tmp;
2152
2153         INIT_LIST_HEAD(&resend);
2154
2155         /* NB called holding the global mutex */
2156
2157         /* All quiet on the API front */
2158         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2159         LASSERT(the_lnet.ln_refcount == 0);
2160
2161         lnet_net_lock(LNET_LOCK_EX);
2162         the_lnet.ln_state = LNET_STATE_STOPPING;
2163
2164         while (!list_empty(&the_lnet.ln_nets)) {
2165                 /*
2166                  * move the nets to the zombie list to avoid them being
2167                  * picked up for new work. LONET is also included in the
2168                  * Nets that will be moved to the zombie list
2169                  */
2170                 net = list_entry(the_lnet.ln_nets.next,
2171                                  struct lnet_net, net_list);
2172                 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2173         }
2174
2175         /* Drop the cached loopback Net. */
2176         if (the_lnet.ln_loni != NULL) {
2177                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2178                 the_lnet.ln_loni = NULL;
2179         }
2180         lnet_net_unlock(LNET_LOCK_EX);
2181
2182         /* iterate through the net zombie list and delete each net */
2183         while (!list_empty(&the_lnet.ln_net_zombie)) {
2184                 net = list_entry(the_lnet.ln_net_zombie.next,
2185                                  struct lnet_net, net_list);
2186                 lnet_shutdown_lndnet(net);
2187         }
2188
2189         spin_lock(&the_lnet.ln_msg_resend_lock);
2190         list_splice(&the_lnet.ln_msg_resend, &resend);
2191         spin_unlock(&the_lnet.ln_msg_resend_lock);
2192
2193         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2194                 list_del_init(&msg->msg_list);
2195                 msg->msg_no_resend = true;
2196                 lnet_finalize(msg, -ECANCELED);
2197         }
2198
2199         lnet_net_lock(LNET_LOCK_EX);
2200         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2201         lnet_net_unlock(LNET_LOCK_EX);
2202 }
2203
2204 static int
2205 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2206 {
2207         int                     rc = -EINVAL;
2208         struct lnet_tx_queue    *tq;
2209         int                     i;
2210         struct lnet_net         *net = ni->ni_net;
2211
2212         mutex_lock(&the_lnet.ln_lnd_mutex);
2213
2214         if (tun) {
2215                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2216                 ni->ni_lnd_tunables_set = true;
2217         }
2218
2219         rc = (net->net_lnd->lnd_startup)(ni);
2220
2221         mutex_unlock(&the_lnet.ln_lnd_mutex);
2222
2223         if (rc != 0) {
2224                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2225                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2226                 lnet_net_lock(LNET_LOCK_EX);
2227                 net->net_lnd->lnd_refcount--;
2228                 lnet_net_unlock(LNET_LOCK_EX);
2229                 goto failed0;
2230         }
2231
2232         lnet_ni_lock(ni);
2233         ni->ni_state = LNET_NI_STATE_ACTIVE;
2234         lnet_ni_unlock(ni);
2235
2236         /* We keep a reference on the loopback net through the loopback NI */
2237         if (net->net_lnd->lnd_type == LOLND) {
2238                 lnet_ni_addref(ni);
2239                 LASSERT(the_lnet.ln_loni == NULL);
2240                 the_lnet.ln_loni = ni;
2241                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2242                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2243                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2244                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2245                 return 0;
2246         }
2247
2248         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2249             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2250                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2251                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2252                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2253                                         "" : "per-peer ");
2254                 /* shutdown the NI since if we get here then it must've already
2255                  * been started
2256                  */
2257                 lnet_shutdown_lndni(ni);
2258                 return -EINVAL;
2259         }
2260
2261         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2262                 tq->tq_credits_min =
2263                 tq->tq_credits_max =
2264                 tq->tq_credits = lnet_ni_tq_credits(ni);
2265         }
2266
2267         atomic_set(&ni->ni_tx_credits,
2268                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2269         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2270
2271         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2272                 libcfs_nid2str(ni->ni_nid),
2273                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2274                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2275                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2276                 ni->ni_net->net_tunables.lct_peer_timeout);
2277
2278         return 0;
2279 failed0:
2280         lnet_ni_free(ni);
2281         return rc;
2282 }
2283
2284 static int
2285 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2286 {
2287         struct lnet_ni *ni;
2288         struct lnet_net *net_l = NULL;
2289         struct list_head        local_ni_list;
2290         int                     rc;
2291         int                     ni_count = 0;
2292         __u32                   lnd_type;
2293         struct lnet_lnd *lnd;
2294         int                     peer_timeout =
2295                 net->net_tunables.lct_peer_timeout;
2296         int                     maxtxcredits =
2297                 net->net_tunables.lct_max_tx_credits;
2298         int                     peerrtrcredits =
2299                 net->net_tunables.lct_peer_rtr_credits;
2300
2301         INIT_LIST_HEAD(&local_ni_list);
2302
2303         /*
2304          * make sure that this net is unique. If it isn't then
2305          * we are adding interfaces to an already existing network, and
2306          * 'net' is just a convenient way to pass in the list.
2307          * if it is unique we need to find the LND and load it if
2308          * necessary.
2309          */
2310         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2311                 lnd_type = LNET_NETTYP(net->net_id);
2312
2313                 mutex_lock(&the_lnet.ln_lnd_mutex);
2314                 lnd = lnet_find_lnd_by_type(lnd_type);
2315
2316                 if (lnd == NULL) {
2317                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2318                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2319                         mutex_lock(&the_lnet.ln_lnd_mutex);
2320
2321                         lnd = lnet_find_lnd_by_type(lnd_type);
2322                         if (lnd == NULL) {
2323                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2324                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2325                                 libcfs_lnd2str(lnd_type),
2326                                 libcfs_lnd2modname(lnd_type), rc);
2327 #ifndef HAVE_MODULE_LOADING_SUPPORT
2328                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2329                                                 "compiled with kernel module "
2330                                                 "loading support.");
2331 #endif
2332                                 rc = -EINVAL;
2333                                 goto failed0;
2334                         }
2335                 }
2336
2337                 lnet_net_lock(LNET_LOCK_EX);
2338                 lnd->lnd_refcount++;
2339                 lnet_net_unlock(LNET_LOCK_EX);
2340
2341                 net->net_lnd = lnd;
2342
2343                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2344
2345                 net_l = net;
2346         }
2347
2348         /*
2349          * net_l: if the network being added is unique then net_l
2350          *        will point to that network
2351          *        if the network being added is not unique then
2352          *        net_l points to the existing network.
2353          *
2354          * When we enter the loop below, we'll pick NIs off he
2355          * network beign added and start them up, then add them to
2356          * a local ni list. Once we've successfully started all
2357          * the NIs then we join the local NI list (of started up
2358          * networks) with the net_l->net_ni_list, which should
2359          * point to the correct network to add the new ni list to
2360          *
2361          * If any of the new NIs fail to start up, then we want to
2362          * iterate through the local ni list, which should include
2363          * any NIs which were successfully started up, and shut
2364          * them down.
2365          *
2366          * After than we want to delete the network being added,
2367          * to avoid a memory leak.
2368          */
2369
2370         /*
2371          * When a network uses TCP bonding then all its interfaces
2372          * must be specified when the network is first defined: the
2373          * TCP bonding code doesn't allow for interfaces to be added
2374          * or removed.
2375          */
2376         if (net_l != net && net_l != NULL && use_tcp_bonding &&
2377             LNET_NETTYP(net_l->net_id) == SOCKLND) {
2378                 rc = -EINVAL;
2379                 goto failed0;
2380         }
2381
2382         while (!list_empty(&net->net_ni_added)) {
2383                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2384                                 ni_netlist);
2385                 list_del_init(&ni->ni_netlist);
2386
2387                 /* make sure that the the NI we're about to start
2388                  * up is actually unique. if it's not fail. */
2389                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2390                                         ni->ni_interfaces[0])) {
2391                         rc = -EEXIST;
2392                         goto failed1;
2393                 }
2394
2395                 /* adjust the pointer the parent network, just in case it
2396                  * the net is a duplicate */
2397                 ni->ni_net = net_l;
2398
2399                 rc = lnet_startup_lndni(ni, tun);
2400
2401                 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
2402                         ni->ni_net->net_lnd->lnd_query != NULL);
2403
2404                 if (rc < 0)
2405                         goto failed1;
2406
2407                 lnet_ni_addref(ni);
2408                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2409
2410                 ni_count++;
2411         }
2412
2413         lnet_net_lock(LNET_LOCK_EX);
2414         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2415         lnet_incr_dlc_seq();
2416         lnet_net_unlock(LNET_LOCK_EX);
2417
2418         /* if the network is not unique then we don't want to keep
2419          * it around after we're done. Free it. Otherwise add that
2420          * net to the global the_lnet.ln_nets */
2421         if (net_l != net && net_l != NULL) {
2422                 /*
2423                  * TODO - note. currently the tunables can not be updated
2424                  * once added
2425                  */
2426                 lnet_net_free(net);
2427         } else {
2428                 net->net_state = LNET_NET_STATE_ACTIVE;
2429                 /*
2430                  * restore tunables after it has been overwitten by the
2431                  * lnd
2432                  */
2433                 if (peer_timeout != -1)
2434                         net->net_tunables.lct_peer_timeout = peer_timeout;
2435                 if (maxtxcredits != -1)
2436                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2437                 if (peerrtrcredits != -1)
2438                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2439
2440                 lnet_net_lock(LNET_LOCK_EX);
2441                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2442                 lnet_net_unlock(LNET_LOCK_EX);
2443         }
2444
2445         /* update net count */
2446         lnet_current_net_count = lnet_get_net_count();
2447
2448         return ni_count;
2449
2450 failed1:
2451         /*
2452          * shutdown the new NIs that are being started up
2453          * free the NET being started
2454          */
2455         while (!list_empty(&local_ni_list)) {
2456                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2457                                 ni_netlist);
2458
2459                 lnet_shutdown_lndni(ni);
2460         }
2461
2462 failed0:
2463         lnet_net_free(net);
2464
2465         return rc;
2466 }
2467
2468 static int
2469 lnet_startup_lndnets(struct list_head *netlist)
2470 {
2471         struct lnet_net         *net;
2472         int                     rc;
2473         int                     ni_count = 0;
2474
2475         /*
2476          * Change to running state before bringing up the LNDs. This
2477          * allows lnet_shutdown_lndnets() to assert that we've passed
2478          * through here.
2479          */
2480         lnet_net_lock(LNET_LOCK_EX);
2481         the_lnet.ln_state = LNET_STATE_RUNNING;
2482         lnet_net_unlock(LNET_LOCK_EX);
2483
2484         while (!list_empty(netlist)) {
2485                 net = list_entry(netlist->next, struct lnet_net, net_list);
2486                 list_del_init(&net->net_list);
2487
2488                 rc = lnet_startup_lndnet(net, NULL);
2489
2490                 if (rc < 0)
2491                         goto failed;
2492
2493                 ni_count += rc;
2494         }
2495
2496         return ni_count;
2497 failed:
2498         lnet_shutdown_lndnets();
2499
2500         return rc;
2501 }
2502
2503 /**
2504  * Initialize LNet library.
2505  *
2506  * Automatically called at module loading time. Caller has to call
2507  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2508  * latter returned 0. It must be called exactly once.
2509  *
2510  * \retval 0 on success
2511  * \retval -ve on failures.
2512  */
2513 int lnet_lib_init(void)
2514 {
2515         int rc;
2516
2517         lnet_assert_wire_constants();
2518
2519         /* refer to global cfs_cpt_table for now */
2520         the_lnet.ln_cpt_table   = cfs_cpt_table;
2521         the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
2522
2523         LASSERT(the_lnet.ln_cpt_number > 0);
2524         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2525                 /* we are under risk of consuming all lh_cookie */
2526                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2527                        "please change setting of CPT-table and retry\n",
2528                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2529                 return -E2BIG;
2530         }
2531
2532         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2533                 the_lnet.ln_cpt_bits++;
2534
2535         rc = lnet_create_locks();
2536         if (rc != 0) {
2537                 CERROR("Can't create LNet global locks: %d\n", rc);
2538                 return rc;
2539         }
2540
2541         the_lnet.ln_refcount = 0;
2542         INIT_LIST_HEAD(&the_lnet.ln_lnds);
2543         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2544         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2545
2546         /* The hash table size is the number of bits it takes to express the set
2547          * ln_num_routes, minus 1 (better to under estimate than over so we
2548          * don't waste memory). */
2549         if (rnet_htable_size <= 0)
2550                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2551         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2552                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2553         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2554                                            order_base_2(rnet_htable_size) - 1);
2555
2556         /* All LNDs apart from the LOLND are in separate modules.  They
2557          * register themselves when their module loads, and unregister
2558          * themselves when their module is unloaded. */
2559         lnet_register_lnd(&the_lolnd);
2560         return 0;
2561 }
2562
2563 /**
2564  * Finalize LNet library.
2565  *
2566  * \pre lnet_lib_init() called with success.
2567  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2568  */
2569 void lnet_lib_exit(void)
2570 {
2571         LASSERT(the_lnet.ln_refcount == 0);
2572
2573         while (!list_empty(&the_lnet.ln_lnds))
2574                 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
2575                                                struct lnet_lnd, lnd_list));
2576         lnet_destroy_locks();
2577 }
2578
2579 /**
2580  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2581  *
2582  * Users must call this function at least once before any other functions.
2583  * For each successful call there must be a corresponding call to
2584  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2585  * ignored.
2586  *
2587  * The PID used by LNet may be different from the one requested.
2588  * See LNetGetId().
2589  *
2590  * \param requested_pid PID requested by the caller.
2591  *
2592  * \return >= 0 on success, and < 0 error code on failures.
2593  */
2594 int
2595 LNetNIInit(lnet_pid_t requested_pid)
2596 {
2597         int                     im_a_router = 0;
2598         int                     rc;
2599         int                     ni_count;
2600         struct lnet_ping_buffer *pbuf;
2601         struct lnet_handle_md   ping_mdh;
2602         struct list_head        net_head;
2603         struct lnet_net         *net;
2604
2605         INIT_LIST_HEAD(&net_head);
2606
2607         mutex_lock(&the_lnet.ln_api_mutex);
2608
2609         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2610
2611         if (the_lnet.ln_refcount > 0) {
2612                 rc = the_lnet.ln_refcount++;
2613                 mutex_unlock(&the_lnet.ln_api_mutex);
2614                 return rc;
2615         }
2616
2617         rc = lnet_prepare(requested_pid);
2618         if (rc != 0) {
2619                 mutex_unlock(&the_lnet.ln_api_mutex);
2620                 return rc;
2621         }
2622
2623         /* create a network for Loopback network */
2624         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2625         if (net == NULL) {
2626                 rc = -ENOMEM;
2627                 goto err_empty_list;
2628         }
2629
2630         /* Add in the loopback NI */
2631         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2632                 rc = -ENOMEM;
2633                 goto err_empty_list;
2634         }
2635
2636         /* If LNet is being initialized via DLC it is possible
2637          * that the user requests not to load module parameters (ones which
2638          * are supported by DLC) on initialization.  Therefore, make sure not
2639          * to load networks, routes and forwarding from module parameters
2640          * in this case.  On cleanup in case of failure only clean up
2641          * routes if it has been loaded */
2642         if (!the_lnet.ln_nis_from_mod_params) {
2643                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2644                                          use_tcp_bonding);
2645                 if (rc < 0)
2646                         goto err_empty_list;
2647         }
2648
2649         ni_count = lnet_startup_lndnets(&net_head);
2650         if (ni_count < 0) {
2651                 rc = ni_count;
2652                 goto err_empty_list;
2653         }
2654
2655         if (!the_lnet.ln_nis_from_mod_params) {
2656                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2657                 if (rc != 0)
2658                         goto err_shutdown_lndnis;
2659
2660                 rc = lnet_rtrpools_alloc(im_a_router);
2661                 if (rc != 0)
2662                         goto err_destroy_routes;
2663         }
2664
2665         rc = lnet_acceptor_start();
2666         if (rc != 0)
2667                 goto err_destroy_routes;
2668
2669         the_lnet.ln_refcount = 1;
2670         /* Now I may use my own API functions... */
2671
2672         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2673         if (rc != 0)
2674                 goto err_acceptor_stop;
2675
2676         lnet_ping_target_update(pbuf, ping_mdh);
2677
2678         rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2679         if (rc != 0) {
2680                 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2681                 goto err_stop_ping;
2682         }
2683
2684         rc = lnet_push_target_init();
2685         if (rc != 0)
2686                 goto err_stop_ping;
2687
2688         rc = lnet_peer_discovery_start();
2689         if (rc != 0)
2690                 goto err_destroy_push_target;
2691
2692         rc = lnet_monitor_thr_start();
2693         if (rc != 0)
2694                 goto err_stop_discovery_thr;
2695
2696         lnet_fault_init();
2697         lnet_router_debugfs_init();
2698
2699         mutex_unlock(&the_lnet.ln_api_mutex);
2700
2701         complete_all(&the_lnet.ln_started);
2702
2703         /* wait for all routers to start */
2704         lnet_wait_router_start();
2705
2706         return 0;
2707
2708 err_stop_discovery_thr:
2709         lnet_peer_discovery_stop();
2710 err_destroy_push_target:
2711         lnet_push_target_fini();
2712 err_stop_ping:
2713         lnet_ping_target_fini();
2714 err_acceptor_stop:
2715         the_lnet.ln_refcount = 0;
2716         lnet_acceptor_stop();
2717 err_destroy_routes:
2718         if (!the_lnet.ln_nis_from_mod_params)
2719                 lnet_destroy_routes();
2720 err_shutdown_lndnis:
2721         lnet_shutdown_lndnets();
2722 err_empty_list:
2723         lnet_unprepare();
2724         LASSERT(rc < 0);
2725         mutex_unlock(&the_lnet.ln_api_mutex);
2726         while (!list_empty(&net_head)) {
2727                 struct lnet_net *net;
2728
2729                 net = list_entry(net_head.next, struct lnet_net, net_list);
2730                 list_del_init(&net->net_list);
2731                 lnet_net_free(net);
2732         }
2733         return rc;
2734 }
2735 EXPORT_SYMBOL(LNetNIInit);
2736
2737 /**
2738  * Stop LNet interfaces, routing, and forwarding.
2739  *
2740  * Users must call this function once for each successful call to LNetNIInit().
2741  * Once the LNetNIFini() operation has been started, the results of pending
2742  * API operations are undefined.
2743  *
2744  * \return always 0 for current implementation.
2745  */
2746 int
2747 LNetNIFini()
2748 {
2749         mutex_lock(&the_lnet.ln_api_mutex);
2750
2751         LASSERT(the_lnet.ln_refcount > 0);
2752
2753         if (the_lnet.ln_refcount != 1) {
2754                 the_lnet.ln_refcount--;
2755         } else {
2756                 LASSERT(!the_lnet.ln_niinit_self);
2757
2758                 lnet_fault_fini();
2759
2760                 lnet_router_debugfs_fini();
2761                 lnet_monitor_thr_stop();
2762                 lnet_peer_discovery_stop();
2763                 lnet_push_target_fini();
2764                 lnet_ping_target_fini();
2765
2766                 /* Teardown fns that use my own API functions BEFORE here */
2767                 the_lnet.ln_refcount = 0;
2768
2769                 lnet_acceptor_stop();
2770                 lnet_destroy_routes();
2771                 lnet_shutdown_lndnets();
2772                 lnet_unprepare();
2773         }
2774
2775         mutex_unlock(&the_lnet.ln_api_mutex);
2776         return 0;
2777 }
2778 EXPORT_SYMBOL(LNetNIFini);
2779
2780 /**
2781  * Grabs the ni data from the ni structure and fills the out
2782  * parameters
2783  *
2784  * \param[in] ni network        interface structure
2785  * \param[out] cfg_ni           NI config information
2786  * \param[out] tun              network and LND tunables
2787  */
2788 static void
2789 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2790                    struct lnet_ioctl_config_lnd_tunables *tun,
2791                    struct lnet_ioctl_element_stats *stats,
2792                    __u32 tun_size)
2793 {
2794         size_t min_size = 0;
2795         int i;
2796
2797         if (!ni || !cfg_ni || !tun)
2798                 return;
2799
2800         if (ni->ni_interfaces[0] != NULL) {
2801                 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2802                         if (ni->ni_interfaces[i] != NULL) {
2803                                 strncpy(cfg_ni->lic_ni_intf[i],
2804                                         ni->ni_interfaces[i],
2805                                         sizeof(cfg_ni->lic_ni_intf[i]));
2806                         }
2807                 }
2808         }
2809
2810         cfg_ni->lic_nid = ni->ni_nid;
2811         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2812                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2813         else
2814                 cfg_ni->lic_status = ni->ni_status->ns_status;
2815         cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2816         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2817
2818         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2819
2820         if (stats) {
2821                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2822                                                        LNET_STATS_TYPE_SEND);
2823                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2824                                                        LNET_STATS_TYPE_RECV);
2825                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2826                                                        LNET_STATS_TYPE_DROP);
2827         }
2828
2829         /*
2830          * tun->lt_tun will always be present, but in order to be
2831          * backwards compatible, we need to deal with the cases when
2832          * tun->lt_tun is smaller than what the kernel has, because it
2833          * comes from an older version of a userspace program, then we'll
2834          * need to copy as much information as we have available space.
2835          */
2836         min_size = tun_size - sizeof(tun->lt_cmn);
2837         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2838
2839         /* copy over the cpts */
2840         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2841             ni->ni_cpts == NULL)  {
2842                 for (i = 0; i < ni->ni_ncpts; i++)
2843                         cfg_ni->lic_cpts[i] = i;
2844         } else {
2845                 for (i = 0;
2846                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2847                      i < LNET_MAX_SHOW_NUM_CPT;
2848                      i++)
2849                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2850         }
2851         cfg_ni->lic_ncpts = ni->ni_ncpts;
2852 }
2853
2854 /**
2855  * NOTE: This is a legacy function left in the code to be backwards
2856  * compatible with older userspace programs. It should eventually be
2857  * removed.
2858  *
2859  * Grabs the ni data from the ni structure and fills the out
2860  * parameters
2861  *
2862  * \param[in] ni network        interface structure
2863  * \param[out] config           config information
2864  */
2865 static void
2866 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2867                          struct lnet_ioctl_config_data *config)
2868 {
2869         struct lnet_ioctl_net_config *net_config;
2870         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2871         size_t min_size, tunable_size = 0;
2872         int i;
2873
2874         if (!ni || !config)
2875                 return;
2876
2877         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2878         if (!net_config)
2879                 return;
2880
2881         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2882                      ARRAY_SIZE(net_config->ni_interfaces));
2883
2884         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2885                 if (!ni->ni_interfaces[i])
2886                         break;
2887
2888                 strncpy(net_config->ni_interfaces[i],
2889                         ni->ni_interfaces[i],
2890                         sizeof(net_config->ni_interfaces[i]));
2891         }
2892
2893         config->cfg_nid = ni->ni_nid;
2894         config->cfg_config_u.cfg_net.net_peer_timeout =
2895                 ni->ni_net->net_tunables.lct_peer_timeout;
2896         config->cfg_config_u.cfg_net.net_max_tx_credits =
2897                 ni->ni_net->net_tunables.lct_max_tx_credits;
2898         config->cfg_config_u.cfg_net.net_peer_tx_credits =
2899                 ni->ni_net->net_tunables.lct_peer_tx_credits;
2900         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2901                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2902
2903         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2904                 net_config->ni_status = LNET_NI_STATUS_UP;
2905         else
2906                 net_config->ni_status = ni->ni_status->ns_status;
2907
2908         if (ni->ni_cpts) {
2909                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2910
2911                 for (i = 0; i < num_cpts; i++)
2912                         net_config->ni_cpts[i] = ni->ni_cpts[i];
2913
2914                 config->cfg_ncpts = num_cpts;
2915         }
2916
2917         /*
2918          * See if user land tools sent in a newer and larger version
2919          * of struct lnet_tunables than what the kernel uses.
2920          */
2921         min_size = sizeof(*config) + sizeof(*net_config);
2922
2923         if (config->cfg_hdr.ioc_len > min_size)
2924                 tunable_size = config->cfg_hdr.ioc_len - min_size;
2925
2926         /* Don't copy too much data to user space */
2927         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2928         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2929
2930         if (lnd_cfg && min_size) {
2931                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2932                 config->cfg_config_u.cfg_net.net_interface_count = 1;
2933
2934                 /* Tell user land that kernel side has less data */
2935                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2936                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2937                         config->cfg_hdr.ioc_len -= min_size;
2938                 }
2939         }
2940 }
2941
2942 struct lnet_ni *
2943 lnet_get_ni_idx_locked(int idx)
2944 {
2945         struct lnet_ni          *ni;
2946         struct lnet_net         *net;
2947
2948         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2949                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2950                         if (idx-- == 0)
2951                                 return ni;
2952                 }
2953         }
2954
2955         return NULL;
2956 }
2957
2958 struct lnet_ni *
2959 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2960 {
2961         struct lnet_ni          *ni;
2962         struct lnet_net         *net = mynet;
2963
2964         /*
2965          * It is possible that the net has been cleaned out while there is
2966          * a message being sent. This function accessed the net without
2967          * checking if the list is empty
2968          */
2969         if (prev == NULL) {
2970                 if (net == NULL)
2971                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2972                                         net_list);
2973                 if (list_empty(&net->net_ni_list))
2974                         return NULL;
2975                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2976                                 ni_netlist);
2977
2978                 return ni;
2979         }
2980
2981         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2982                 /* if you reached the end of the ni list and the net is
2983                  * specified, then there are no more nis in that net */
2984                 if (net != NULL)
2985                         return NULL;
2986
2987                 /* we reached the end of this net ni list. move to the
2988                  * next net */
2989                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2990                         /* no more nets and no more NIs. */
2991                         return NULL;
2992
2993                 /* get the next net */
2994                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2995                                  net_list);
2996                 if (list_empty(&net->net_ni_list))
2997                         return NULL;
2998                 /* get the ni on it */
2999                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3000                                 ni_netlist);
3001
3002                 return ni;
3003         }
3004
3005         if (list_empty(&prev->ni_netlist))
3006                 return NULL;
3007
3008         /* there are more nis left */
3009         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3010
3011         return ni;
3012 }
3013
3014 int
3015 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3016 {
3017         struct lnet_ni *ni;
3018         int cpt;
3019         int rc = -ENOENT;
3020         int idx = config->cfg_count;
3021
3022         cpt = lnet_net_lock_current();
3023
3024         ni = lnet_get_ni_idx_locked(idx);
3025
3026         if (ni != NULL) {
3027                 rc = 0;
3028                 lnet_ni_lock(ni);
3029                 lnet_fill_ni_info_legacy(ni, config);
3030                 lnet_ni_unlock(ni);
3031         }
3032
3033         lnet_net_unlock(cpt);
3034         return rc;
3035 }
3036
3037 int
3038 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3039                    struct lnet_ioctl_config_lnd_tunables *tun,
3040                    struct lnet_ioctl_element_stats *stats,
3041                    __u32 tun_size)
3042 {
3043         struct lnet_ni          *ni;
3044         int                     cpt;
3045         int                     rc = -ENOENT;
3046
3047         if (!cfg_ni || !tun || !stats)
3048                 return -EINVAL;
3049
3050         cpt = lnet_net_lock_current();
3051
3052         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3053
3054         if (ni) {
3055                 rc = 0;
3056                 lnet_ni_lock(ni);
3057                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3058                 lnet_ni_unlock(ni);
3059         }
3060
3061         lnet_net_unlock(cpt);
3062         return rc;
3063 }
3064
3065 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3066 {
3067         struct lnet_ni *ni;
3068         int cpt;
3069         int rc = -ENOENT;
3070
3071         if (!msg_stats)
3072                 return -EINVAL;
3073
3074         cpt = lnet_net_lock_current();
3075
3076         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3077
3078         if (ni) {
3079                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3080                 rc = 0;
3081         }
3082
3083         lnet_net_unlock(cpt);
3084
3085         return rc;
3086 }
3087
3088 static int lnet_add_net_common(struct lnet_net *net,
3089                                struct lnet_ioctl_config_lnd_tunables *tun)
3090 {
3091         __u32                   net_id;
3092         struct lnet_ping_buffer *pbuf;
3093         struct lnet_handle_md   ping_mdh;
3094         int                     rc;
3095         struct lnet_remotenet *rnet;
3096         int                     net_ni_count;
3097         int                     num_acceptor_nets;
3098
3099         lnet_net_lock(LNET_LOCK_EX);
3100         rnet = lnet_find_rnet_locked(net->net_id);
3101         lnet_net_unlock(LNET_LOCK_EX);
3102         /*
3103          * make sure that the net added doesn't invalidate the current
3104          * configuration LNet is keeping
3105          */
3106         if (rnet) {
3107                 CERROR("Adding net %s will invalidate routing configuration\n",
3108                        libcfs_net2str(net->net_id));
3109                 lnet_net_free(net);
3110                 return -EUSERS;
3111         }
3112
3113         /*
3114          * make sure you calculate the correct number of slots in the ping
3115          * buffer. Since the ping info is a flattened list of all the NIs,
3116          * we should allocate enough slots to accomodate the number of NIs
3117          * which will be added.
3118          *
3119          * since ni hasn't been configured yet, use
3120          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3121          */
3122         net_ni_count = lnet_get_net_ni_count_pre(net);
3123
3124         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3125                                     net_ni_count + lnet_get_ni_count(),
3126                                     false);
3127         if (rc < 0) {
3128                 lnet_net_free(net);
3129                 return rc;
3130         }
3131
3132         if (tun)
3133                 memcpy(&net->net_tunables,
3134                        &tun->lt_cmn, sizeof(net->net_tunables));
3135         else
3136                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3137
3138         /*
3139          * before starting this network get a count of the current TCP
3140          * networks which require the acceptor thread running. If that
3141          * count is == 0 before we start up this network, then we'd want to
3142          * start up the acceptor thread after starting up this network
3143          */
3144         num_acceptor_nets = lnet_count_acceptor_nets();
3145
3146         net_id = net->net_id;
3147
3148         rc = lnet_startup_lndnet(net,
3149                                  (tun) ? &tun->lt_tun : NULL);
3150         if (rc < 0)
3151                 goto failed;
3152
3153         lnet_net_lock(LNET_LOCK_EX);
3154         net = lnet_get_net_locked(net_id);
3155         lnet_net_unlock(LNET_LOCK_EX);
3156
3157         LASSERT(net);
3158
3159         /*
3160          * Start the acceptor thread if this is the first network
3161          * being added that requires the thread.
3162          */
3163         if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3164                 rc = lnet_acceptor_start();
3165                 if (rc < 0) {
3166                         /* shutdown the net that we just started */
3167                         CERROR("Failed to start up acceptor thread\n");
3168                         lnet_shutdown_lndnet(net);
3169                         goto failed;
3170                 }
3171         }
3172
3173         lnet_net_lock(LNET_LOCK_EX);
3174         lnet_peer_net_added(net);
3175         lnet_net_unlock(LNET_LOCK_EX);
3176
3177         lnet_ping_target_update(pbuf, ping_mdh);
3178
3179         return 0;
3180
3181 failed:
3182         lnet_ping_md_unlink(pbuf, &ping_mdh);
3183         lnet_ping_buffer_decref(pbuf);
3184         return rc;
3185 }
3186
3187 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3188                                       struct lnet_ioctl_config_lnd_tunables *tun)
3189 {
3190         struct lnet_net *net;
3191         char *nets;
3192         int rc;
3193         struct list_head net_head;
3194
3195         INIT_LIST_HEAD(&net_head);
3196
3197         rc = lnet_parse_ip2nets(&nets, ip2nets);
3198         if (rc < 0)
3199                 return rc;
3200
3201         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3202         if (rc < 0)
3203                 return rc;
3204
3205         mutex_lock(&the_lnet.ln_api_mutex);
3206         while (!list_empty(&net_head)) {
3207                 net = list_entry(net_head.next, struct lnet_net, net_list);
3208                 list_del_init(&net->net_list);
3209                 rc = lnet_add_net_common(net, tun);
3210                 if (rc < 0)
3211                         goto out;
3212         }
3213
3214 out:
3215         mutex_unlock(&the_lnet.ln_api_mutex);
3216
3217         while (!list_empty(&net_head)) {
3218                 net = list_entry(net_head.next, struct lnet_net, net_list);
3219                 list_del_init(&net->net_list);
3220                 lnet_net_free(net);
3221         }
3222         return rc;
3223 }
3224
3225 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3226 {
3227         struct lnet_net *net;
3228         struct lnet_ni *ni;
3229         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3230         int rc, i;
3231         __u32 net_id, lnd_type;
3232
3233         /* get the tunables if they are available */
3234         if (conf->lic_cfg_hdr.ioc_len >=
3235             sizeof(*conf) + sizeof(*tun))
3236                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3237                         conf->lic_bulk;
3238
3239         /* handle legacy ip2nets from DLC */
3240         if (conf->lic_legacy_ip2nets[0] != '\0')
3241                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3242                                                   tun);
3243
3244         net_id = LNET_NIDNET(conf->lic_nid);
3245         lnd_type = LNET_NETTYP(net_id);
3246
3247         if (!libcfs_isknown_lnd(lnd_type)) {
3248                 CERROR("No valid net and lnd information provided\n");
3249                 return -EINVAL;
3250         }
3251
3252         net = lnet_net_alloc(net_id, NULL);
3253         if (!net)
3254                 return -ENOMEM;
3255
3256         for (i = 0; i < conf->lic_ncpts; i++) {
3257                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3258                         return -EINVAL;
3259         }
3260
3261         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3262                                        conf->lic_ni_intf[0]);
3263         if (!ni)
3264                 return -ENOMEM;
3265
3266         mutex_lock(&the_lnet.ln_api_mutex);
3267
3268         rc = lnet_add_net_common(net, tun);
3269
3270         mutex_unlock(&the_lnet.ln_api_mutex);
3271
3272         return rc;
3273 }
3274
3275 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3276 {
3277         struct lnet_net  *net;
3278         struct lnet_ni *ni;
3279         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3280         struct lnet_ping_buffer *pbuf;
3281         struct lnet_handle_md  ping_mdh;
3282         int               rc;
3283         int               net_count;
3284         __u32             addr;
3285
3286         /* don't allow userspace to shutdown the LOLND */
3287         if (LNET_NETTYP(net_id) == LOLND)
3288                 return -EINVAL;
3289
3290         mutex_lock(&the_lnet.ln_api_mutex);
3291
3292         lnet_net_lock(0);
3293
3294         net = lnet_get_net_locked(net_id);
3295         if (!net) {
3296                 CERROR("net %s not found\n",
3297                        libcfs_net2str(net_id));
3298                 rc = -ENOENT;
3299                 goto unlock_net;
3300         }
3301
3302         addr = LNET_NIDADDR(conf->lic_nid);
3303         if (addr == 0) {
3304                 /* remove the entire net */
3305                 net_count = lnet_get_net_ni_count_locked(net);
3306
3307                 lnet_net_unlock(0);
3308
3309                 /* create and link a new ping info, before removing the old one */
3310                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3311                                         lnet_get_ni_count() - net_count,
3312                                         false);
3313                 if (rc != 0)
3314                         goto unlock_api_mutex;
3315
3316                 lnet_shutdown_lndnet(net);
3317
3318                 if (lnet_count_acceptor_nets() == 0)
3319                         lnet_acceptor_stop();
3320
3321                 lnet_ping_target_update(pbuf, ping_mdh);
3322
3323                 goto unlock_api_mutex;
3324         }
3325
3326         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3327         if (!ni) {
3328                 CERROR("nid %s not found\n",
3329                        libcfs_nid2str(conf->lic_nid));
3330                 rc = -ENOENT;
3331                 goto unlock_net;
3332         }
3333
3334         net_count = lnet_get_net_ni_count_locked(net);
3335
3336         lnet_net_unlock(0);
3337
3338         /* create and link a new ping info, before removing the old one */
3339         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3340                                   lnet_get_ni_count() - 1, false);
3341         if (rc != 0)
3342                 goto unlock_api_mutex;
3343
3344         lnet_shutdown_lndni(ni);
3345
3346         if (lnet_count_acceptor_nets() == 0)
3347                 lnet_acceptor_stop();
3348
3349         lnet_ping_target_update(pbuf, ping_mdh);
3350
3351         /* check if the net is empty and remove it if it is */
3352         if (net_count == 1)
3353                 lnet_shutdown_lndnet(net);
3354
3355         goto unlock_api_mutex;
3356
3357 unlock_net:
3358         lnet_net_unlock(0);
3359 unlock_api_mutex:
3360         mutex_unlock(&the_lnet.ln_api_mutex);
3361
3362         return rc;
3363 }
3364
3365 /*
3366  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3367  * They are only expected to be called for unique networks.
3368  * That can be as a result of older DLC library
3369  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3370  */
3371 int
3372 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3373 {
3374         struct lnet_net         *net;
3375         struct list_head        net_head;
3376         int                     rc;
3377         struct lnet_ioctl_config_lnd_tunables tun;
3378         char *nets = conf->cfg_config_u.cfg_net.net_intf;
3379
3380         INIT_LIST_HEAD(&net_head);
3381
3382         /* Create a net/ni structures for the network string */
3383         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3384         if (rc <= 0)
3385                 return rc == 0 ? -EINVAL : rc;
3386
3387         mutex_lock(&the_lnet.ln_api_mutex);
3388
3389         if (rc > 1) {
3390                 rc = -EINVAL; /* only add one network per call */
3391                 goto out_unlock_clean;
3392         }
3393
3394         net = list_entry(net_head.next, struct lnet_net, net_list);
3395         list_del_init(&net->net_list);
3396
3397         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3398
3399         memset(&tun, 0, sizeof(tun));
3400
3401         tun.lt_cmn.lct_peer_timeout =
3402           conf->cfg_config_u.cfg_net.net_peer_timeout;
3403         tun.lt_cmn.lct_peer_tx_credits =
3404           conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3405         tun.lt_cmn.lct_peer_rtr_credits =
3406           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3407         tun.lt_cmn.lct_max_tx_credits =
3408           conf->cfg_config_u.cfg_net.net_max_tx_credits;
3409
3410         rc = lnet_add_net_common(net, &tun);
3411
3412 out_unlock_clean:
3413         mutex_unlock(&the_lnet.ln_api_mutex);
3414         while (!list_empty(&net_head)) {
3415                 /* net_head list is empty in success case */
3416                 net = list_entry(net_head.next, struct lnet_net, net_list);
3417                 list_del_init(&net->net_list);
3418                 lnet_net_free(net);
3419         }
3420         return rc;
3421 }
3422
3423 int
3424 lnet_dyn_del_net(__u32 net_id)
3425 {
3426         struct lnet_net  *net;
3427         struct lnet_ping_buffer *pbuf;
3428         struct lnet_handle_md ping_mdh;
3429         int               rc;
3430         int               net_ni_count;
3431
3432         /* don't allow userspace to shutdown the LOLND */
3433         if (LNET_NETTYP(net_id) == LOLND)
3434                 return -EINVAL;
3435
3436         mutex_lock(&the_lnet.ln_api_mutex);
3437
3438         lnet_net_lock(0);
3439
3440         net = lnet_get_net_locked(net_id);
3441         if (net == NULL) {
3442                 lnet_net_unlock(0);
3443                 rc = -EINVAL;
3444                 goto out;
3445         }
3446
3447         net_ni_count = lnet_get_net_ni_count_locked(net);
3448
3449         lnet_net_unlock(0);
3450
3451         /* create and link a new ping info, before removing the old one */
3452         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3453                                     lnet_get_ni_count() - net_ni_count, false);
3454         if (rc != 0)
3455                 goto out;
3456
3457         lnet_shutdown_lndnet(net);
3458
3459         if (lnet_count_acceptor_nets() == 0)
3460                 lnet_acceptor_stop();
3461
3462         lnet_ping_target_update(pbuf, ping_mdh);
3463
3464 out:
3465         mutex_unlock(&the_lnet.ln_api_mutex);
3466
3467         return rc;
3468 }
3469
3470 void lnet_incr_dlc_seq(void)
3471 {
3472         atomic_inc(&lnet_dlc_seq_no);
3473 }
3474
3475 __u32 lnet_get_dlc_seq_locked(void)
3476 {
3477         return atomic_read(&lnet_dlc_seq_no);
3478 }
3479
3480 static void
3481 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3482 {
3483         struct lnet_net *net;
3484         struct lnet_ni *ni;
3485
3486         lnet_net_lock(LNET_LOCK_EX);
3487         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3488                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3489                         if (ni->ni_nid == nid || all) {
3490                                 atomic_set(&ni->ni_healthv, value);
3491                                 if (list_empty(&ni->ni_recovery) &&
3492                                     value < LNET_MAX_HEALTH_VALUE) {
3493                                         CERROR("manually adding local NI %s to recovery\n",
3494                                                libcfs_nid2str(ni->ni_nid));
3495                                         list_add_tail(&ni->ni_recovery,
3496                                                       &the_lnet.ln_mt_localNIRecovq);
3497                                         lnet_ni_addref_locked(ni, 0);
3498                                 }
3499                                 if (!all) {
3500                                         lnet_net_unlock(LNET_LOCK_EX);
3501                                         return;
3502                                 }
3503                         }
3504                 }
3505         }
3506         lnet_net_unlock(LNET_LOCK_EX);
3507 }
3508
3509 static int
3510 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3511 {
3512         int cpt, rc = 0;
3513         struct lnet_ni *ni;
3514         lnet_nid_t nid = stats->hlni_nid;
3515
3516         cpt = lnet_net_lock_current();
3517         ni = lnet_nid2ni_locked(nid, cpt);
3518
3519         if (!ni) {
3520                 rc = -ENOENT;
3521                 goto unlock;
3522         }
3523
3524         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3525         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3526         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3527         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3528         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3529         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3530         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3531
3532 unlock:
3533         lnet_net_unlock(cpt);
3534
3535         return rc;
3536 }
3537
3538 static int
3539 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3540 {
3541         struct lnet_ni *ni;
3542         int i = 0;
3543
3544         lnet_net_lock(LNET_LOCK_EX);
3545         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3546                 list->rlst_nid_array[i] = ni->ni_nid;
3547                 i++;
3548                 if (i >= LNET_MAX_SHOW_NUM_NID)
3549                         break;
3550         }
3551         lnet_net_unlock(LNET_LOCK_EX);
3552         list->rlst_num_nids = i;
3553
3554         return 0;
3555 }
3556
3557 static int
3558 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3559 {
3560         struct lnet_peer_ni *lpni;
3561         int i = 0;
3562
3563         lnet_net_lock(LNET_LOCK_EX);
3564         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3565                 list->rlst_nid_array[i] = lpni->lpni_nid;
3566                 i++;
3567                 if (i >= LNET_MAX_SHOW_NUM_NID)
3568                         break;
3569         }
3570         lnet_net_unlock(LNET_LOCK_EX);
3571         list->rlst_num_nids = i;
3572
3573         return 0;
3574 }
3575
3576 /**
3577  * LNet ioctl handler.
3578  *
3579  */
3580 int
3581 LNetCtl(unsigned int cmd, void *arg)
3582 {
3583         struct libcfs_ioctl_data *data = arg;
3584         struct lnet_ioctl_config_data *config;
3585         struct lnet_process_id    id = {0};
3586         struct lnet_ni           *ni;
3587         int                       rc;
3588
3589         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3590                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3591
3592         switch (cmd) {
3593         case IOC_LIBCFS_GET_NI:
3594                 rc = LNetGetId(data->ioc_count, &id);
3595                 data->ioc_nid = id.nid;
3596                 return rc;
3597
3598         case IOC_LIBCFS_FAIL_NID:
3599                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3600
3601         case IOC_LIBCFS_ADD_ROUTE: {
3602                 /* default router sensitivity to 1 */
3603                 unsigned int sensitivity = 1;
3604                 config = arg;
3605
3606                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3607                         return -EINVAL;
3608
3609                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3610                         sensitivity =
3611                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3612                 }
3613
3614                 mutex_lock(&the_lnet.ln_api_mutex);
3615                 rc = lnet_add_route(config->cfg_net,
3616                                     config->cfg_config_u.cfg_route.rtr_hop,
3617                                     config->cfg_nid,
3618                                     config->cfg_config_u.cfg_route.
3619                                         rtr_priority, sensitivity);
3620                 mutex_unlock(&the_lnet.ln_api_mutex);
3621                 return rc;
3622         }
3623
3624         case IOC_LIBCFS_DEL_ROUTE:
3625                 config = arg;
3626
3627                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3628                         return -EINVAL;
3629
3630                 mutex_lock(&the_lnet.ln_api_mutex);
3631                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3632                 mutex_unlock(&the_lnet.ln_api_mutex);
3633                 return rc;
3634
3635         case IOC_LIBCFS_GET_ROUTE:
3636                 config = arg;
3637
3638                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3639                         return -EINVAL;
3640
3641                 mutex_lock(&the_lnet.ln_api_mutex);
3642                 rc = lnet_get_route(config->cfg_count,
3643                                     &config->cfg_net,
3644                                     &config->cfg_config_u.cfg_route.rtr_hop,
3645                                     &config->cfg_nid,
3646                                     &config->cfg_config_u.cfg_route.rtr_flags,
3647                                     &config->cfg_config_u.cfg_route.
3648                                         rtr_priority,
3649                                     &config->cfg_config_u.cfg_route.
3650                                         rtr_sensitivity);
3651                 mutex_unlock(&the_lnet.ln_api_mutex);
3652                 return rc;
3653
3654         case IOC_LIBCFS_GET_LOCAL_NI: {
3655                 struct lnet_ioctl_config_ni *cfg_ni;
3656                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3657                 struct lnet_ioctl_element_stats *stats;
3658                 __u32 tun_size;
3659
3660                 cfg_ni = arg;
3661
3662                 /* get the tunables if they are available */
3663                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3664                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3665                         return -EINVAL;
3666
3667                 stats = (struct lnet_ioctl_element_stats *)
3668                         cfg_ni->lic_bulk;
3669                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3670                                 (cfg_ni->lic_bulk + sizeof(*stats));
3671
3672                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3673                         sizeof(*stats);
3674
3675                 mutex_lock(&the_lnet.ln_api_mutex);
3676                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3677                 mutex_unlock(&the_lnet.ln_api_mutex);
3678                 return rc;
3679         }
3680
3681         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3682                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3683
3684                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3685                         return -EINVAL;
3686
3687                 mutex_lock(&the_lnet.ln_api_mutex);
3688                 rc = lnet_get_ni_stats(msg_stats);
3689                 mutex_unlock(&the_lnet.ln_api_mutex);
3690
3691                 return rc;
3692         }
3693
3694         case IOC_LIBCFS_GET_NET: {
3695                 size_t total = sizeof(*config) +
3696                                sizeof(struct lnet_ioctl_net_config);
3697                 config = arg;
3698
3699                 if (config->cfg_hdr.ioc_len < total)
3700                         return -EINVAL;
3701
3702                 mutex_lock(&the_lnet.ln_api_mutex);
3703                 rc = lnet_get_net_config(config);
3704                 mutex_unlock(&the_lnet.ln_api_mutex);
3705                 return rc;
3706         }
3707
3708         case IOC_LIBCFS_GET_LNET_STATS:
3709         {
3710                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3711
3712                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3713                         return -EINVAL;
3714
3715                 mutex_lock(&the_lnet.ln_api_mutex);
3716                 lnet_counters_get(&lnet_stats->st_cntrs);
3717                 mutex_unlock(&the_lnet.ln_api_mutex);
3718                 return 0;
3719         }
3720
3721         case IOC_LIBCFS_CONFIG_RTR:
3722                 config = arg;
3723
3724                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3725                         return -EINVAL;
3726
3727                 mutex_lock(&the_lnet.ln_api_mutex);
3728                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3729                         rc = lnet_rtrpools_enable();
3730                         mutex_unlock(&the_lnet.ln_api_mutex);
3731                         return rc;
3732                 }
3733                 lnet_rtrpools_disable();
3734                 mutex_unlock(&the_lnet.ln_api_mutex);
3735                 return 0;
3736
3737         case IOC_LIBCFS_ADD_BUF:
3738                 config = arg;
3739
3740                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3741                         return -EINVAL;
3742
3743                 mutex_lock(&the_lnet.ln_api_mutex);
3744                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3745                                                 buf_tiny,
3746                                           config->cfg_config_u.cfg_buffers.
3747                                                 buf_small,
3748                                           config->cfg_config_u.cfg_buffers.
3749                                                 buf_large);
3750                 mutex_unlock(&the_lnet.ln_api_mutex);
3751                 return rc;
3752
3753         case IOC_LIBCFS_SET_NUMA_RANGE: {
3754                 struct lnet_ioctl_set_value *numa;
3755                 numa = arg;
3756                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3757                         return -EINVAL;
3758                 lnet_net_lock(LNET_LOCK_EX);
3759                 lnet_numa_range = numa->sv_value;
3760                 lnet_net_unlock(LNET_LOCK_EX);
3761                 return 0;
3762         }
3763
3764         case IOC_LIBCFS_GET_NUMA_RANGE: {
3765                 struct lnet_ioctl_set_value *numa;
3766                 numa = arg;
3767                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3768                         return -EINVAL;
3769                 numa->sv_value = lnet_numa_range;
3770                 return 0;
3771         }
3772
3773         case IOC_LIBCFS_GET_BUF: {
3774                 struct lnet_ioctl_pool_cfg *pool_cfg;
3775                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3776
3777                 config = arg;
3778
3779                 if (config->cfg_hdr.ioc_len < total)
3780                         return -EINVAL;
3781
3782                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3783
3784                 mutex_lock(&the_lnet.ln_api_mutex);
3785                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3786                 mutex_unlock(&the_lnet.ln_api_mutex);
3787                 return rc;
3788         }
3789
3790         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3791                 struct lnet_ioctl_local_ni_hstats *stats = arg;
3792
3793                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3794                         return -EINVAL;
3795
3796                 mutex_lock(&the_lnet.ln_api_mutex);
3797                 rc = lnet_get_local_ni_hstats(stats);
3798                 mutex_unlock(&the_lnet.ln_api_mutex);
3799
3800                 return rc;
3801         }
3802
3803         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3804                 struct lnet_ioctl_recovery_list *list = arg;
3805                 if (list->rlst_hdr.ioc_len < sizeof(*list))
3806                         return -EINVAL;
3807
3808                 mutex_lock(&the_lnet.ln_api_mutex);
3809                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3810                         rc = lnet_get_local_ni_recovery_list(list);
3811                 else
3812                         rc = lnet_get_peer_ni_recovery_list(list);
3813                 mutex_unlock(&the_lnet.ln_api_mutex);
3814                 return rc;
3815         }
3816
3817         case IOC_LIBCFS_ADD_PEER_NI: {
3818                 struct lnet_ioctl_peer_cfg *cfg = arg;
3819
3820                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3821                         return -EINVAL;
3822
3823                 mutex_lock(&the_lnet.ln_api_mutex);
3824                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3825                                       cfg->prcfg_cfg_nid,
3826                                       cfg->prcfg_mr);
3827                 mutex_unlock(&the_lnet.ln_api_mutex);
3828                 return rc;
3829         }
3830
3831         case IOC_LIBCFS_DEL_PEER_NI: {
3832                 struct lnet_ioctl_peer_cfg *cfg = arg;
3833
3834                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3835                         return -EINVAL;
3836
3837                 mutex_lock(&the_lnet.ln_api_mutex);
3838                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3839                                       cfg->prcfg_cfg_nid);
3840                 mutex_unlock(&the_lnet.ln_api_mutex);
3841                 return rc;
3842         }
3843
3844         case IOC_LIBCFS_GET_PEER_INFO: {
3845                 struct lnet_ioctl_peer *peer_info = arg;
3846
3847                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3848                         return -EINVAL;
3849
3850                 mutex_lock(&the_lnet.ln_api_mutex);
3851                 rc = lnet_get_peer_ni_info(
3852                    peer_info->pr_count,
3853                    &peer_info->pr_nid,
3854                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3855                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3856                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3857                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3858                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3859                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3860                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3861                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3862                 mutex_unlock(&the_lnet.ln_api_mutex);
3863                 return rc;
3864         }
3865
3866         case IOC_LIBCFS_GET_PEER_NI: {
3867                 struct lnet_ioctl_peer_cfg *cfg = arg;
3868
3869                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3870                         return -EINVAL;
3871
3872                 mutex_lock(&the_lnet.ln_api_mutex);
3873                 rc = lnet_get_peer_info(cfg,
3874                                         (void __user *)cfg->prcfg_bulk);
3875                 mutex_unlock(&the_lnet.ln_api_mutex);
3876                 return rc;
3877         }
3878
3879         case IOC_LIBCFS_GET_PEER_LIST: {
3880                 struct lnet_ioctl_peer_cfg *cfg = arg;
3881
3882                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3883                         return -EINVAL;
3884
3885                 mutex_lock(&the_lnet.ln_api_mutex);
3886                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3887                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3888                 mutex_unlock(&the_lnet.ln_api_mutex);
3889                 return rc;
3890         }
3891
3892         case IOC_LIBCFS_SET_HEALHV: {
3893                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3894                 int value;
3895                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3896                         return -EINVAL;
3897                 if (cfg->rh_value < 0 ||
3898                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3899                         value = LNET_MAX_HEALTH_VALUE;
3900                 else
3901                         value = cfg->rh_value;
3902                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3903                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3904                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3905                 mutex_lock(&the_lnet.ln_api_mutex);
3906                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3907                         lnet_ni_set_healthv(cfg->rh_nid, value,
3908                                              cfg->rh_all);
3909                 else
3910                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3911                                                   cfg->rh_all);
3912                 mutex_unlock(&the_lnet.ln_api_mutex);
3913                 return 0;
3914         }
3915
3916         case IOC_LIBCFS_NOTIFY_ROUTER: {
3917                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3918
3919                 /* The deadline passed in by the user should be some time in
3920                  * seconds in the future since the UNIX epoch. We have to map
3921                  * that deadline to the wall clock.
3922                  */
3923                 deadline += ktime_get_seconds();
3924                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3925                                    deadline);
3926         }
3927
3928         case IOC_LIBCFS_LNET_DIST:
3929                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3930                 if (rc < 0 && rc != -EHOSTUNREACH)
3931                         return rc;
3932
3933                 data->ioc_u32[0] = rc;
3934                 return 0;
3935
3936         case IOC_LIBCFS_TESTPROTOCOMPAT:
3937                 lnet_net_lock(LNET_LOCK_EX);
3938                 the_lnet.ln_testprotocompat = data->ioc_flags;
3939                 lnet_net_unlock(LNET_LOCK_EX);
3940                 return 0;
3941
3942         case IOC_LIBCFS_LNET_FAULT:
3943                 return lnet_fault_ctl(data->ioc_flags, data);
3944
3945         case IOC_LIBCFS_PING: {
3946                 signed long timeout;
3947
3948                 id.nid = data->ioc_nid;
3949                 id.pid = data->ioc_u32[0];
3950
3951                 /* If timeout is negative then set default of 3 minutes */
3952                 if (((s32)data->ioc_u32[1] <= 0) ||
3953                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3954                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3955                 else
3956                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3957
3958                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3959                                data->ioc_plen1 / sizeof(struct lnet_process_id));
3960
3961                 if (rc < 0)
3962                         return rc;
3963
3964                 data->ioc_count = rc;
3965                 return 0;
3966         }
3967
3968         case IOC_LIBCFS_PING_PEER: {
3969                 struct lnet_ioctl_ping_data *ping = arg;
3970                 struct lnet_peer *lp;
3971                 signed long timeout;
3972
3973                 /* If timeout is negative then set default of 3 minutes */
3974                 if (((s32)ping->op_param) <= 0 ||
3975                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3976                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3977                 else
3978                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3979
3980                 rc = lnet_ping(ping->ping_id, timeout,
3981                                ping->ping_buf,
3982                                ping->ping_count);
3983                 if (rc < 0)
3984                         return rc;
3985
3986                 mutex_lock(&the_lnet.ln_api_mutex);
3987                 lp = lnet_find_peer(ping->ping_id.nid);
3988                 if (lp) {
3989                         ping->ping_id.nid = lp->lp_primary_nid;
3990                         ping->mr_info = lnet_peer_is_multi_rail(lp);
3991                         lnet_peer_decref_locked(lp);
3992                 }
3993                 mutex_unlock(&the_lnet.ln_api_mutex);
3994
3995                 ping->ping_count = rc;
3996                 return 0;
3997         }
3998
3999         case IOC_LIBCFS_DISCOVER: {
4000                 struct lnet_ioctl_ping_data *discover = arg;
4001                 struct lnet_peer *lp;
4002
4003                 rc = lnet_discover(discover->ping_id, discover->op_param,
4004                                    discover->ping_buf,
4005                                    discover->ping_count);
4006                 if (rc < 0)
4007                         return rc;
4008
4009                 mutex_lock(&the_lnet.ln_api_mutex);
4010                 lp = lnet_find_peer(discover->ping_id.nid);
4011                 if (lp) {
4012                         discover->ping_id.nid = lp->lp_primary_nid;
4013                         discover->mr_info = lnet_peer_is_multi_rail(lp);
4014                         lnet_peer_decref_locked(lp);
4015                 }
4016                 mutex_unlock(&the_lnet.ln_api_mutex);
4017
4018                 discover->ping_count = rc;
4019                 return 0;
4020         }
4021
4022         default:
4023                 ni = lnet_net2ni_addref(data->ioc_net);
4024                 if (ni == NULL)
4025                         return -EINVAL;
4026
4027                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4028                         rc = -EINVAL;
4029                 else
4030                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4031
4032                 lnet_ni_decref(ni);
4033                 return rc;
4034         }
4035         /* not reached */
4036 }
4037 EXPORT_SYMBOL(LNetCtl);
4038
4039 void LNetDebugPeer(struct lnet_process_id id)
4040 {
4041         lnet_debug_peer(id.nid);
4042 }
4043 EXPORT_SYMBOL(LNetDebugPeer);
4044
4045 /**
4046  * Determine if the specified peer \a nid is on the local node.
4047  *
4048  * \param nid   peer nid to check
4049  *
4050  * \retval true         If peer NID is on the local node.
4051  * \retval false        If peer NID is not on the local node.
4052  */
4053 bool LNetIsPeerLocal(lnet_nid_t nid)
4054 {
4055         struct lnet_net *net;
4056         struct lnet_ni *ni;
4057         int cpt;
4058
4059         cpt = lnet_net_lock_current();
4060         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4061                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4062                         if (ni->ni_nid == nid) {
4063                                 lnet_net_unlock(cpt);
4064                                 return true;
4065                         }
4066                 }
4067         }
4068         lnet_net_unlock(cpt);
4069
4070         return false;
4071 }
4072 EXPORT_SYMBOL(LNetIsPeerLocal);
4073
4074 /**
4075  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4076  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4077  *
4078  * \param index Index of the interface to look up.
4079  * \param id On successful return, this location will hold the
4080  * struct lnet_process_id ID of the interface.
4081  *
4082  * \retval 0 If an interface exists at \a index.
4083  * \retval -ENOENT If no interface has been found.
4084  */
4085 int
4086 LNetGetId(unsigned int index, struct lnet_process_id *id)
4087 {
4088         struct lnet_ni   *ni;
4089         struct lnet_net  *net;
4090         int               cpt;
4091         int               rc = -ENOENT;
4092
4093         LASSERT(the_lnet.ln_refcount > 0);
4094
4095         cpt = lnet_net_lock_current();
4096
4097         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4098                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4099                         if (index-- != 0)
4100                                 continue;
4101
4102                         id->nid = ni->ni_nid;
4103                         id->pid = the_lnet.ln_pid;
4104                         rc = 0;
4105                         break;
4106                 }
4107         }
4108
4109         lnet_net_unlock(cpt);
4110         return rc;
4111 }
4112 EXPORT_SYMBOL(LNetGetId);
4113
4114 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4115                      struct lnet_process_id __user *ids, int n_ids)
4116 {
4117         struct lnet_handle_eq eqh;
4118         struct lnet_handle_md mdh;
4119         struct lnet_event event;
4120         struct lnet_md md = { NULL };
4121         int which;
4122         int unlinked = 0;
4123         int replied = 0;
4124         const signed long a_long_time = cfs_time_seconds(60);
4125         struct lnet_ping_buffer *pbuf;
4126         struct lnet_process_id tmpid;
4127         int i;
4128         int nob;
4129         int rc;
4130         int rc2;
4131         sigset_t blocked;
4132
4133         /* n_ids limit is arbitrary */
4134         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4135                 return -EINVAL;
4136
4137         /*
4138          * if the user buffer has more space than the lnet_interfaces_max
4139          * then only fill it up to lnet_interfaces_max
4140          */
4141         if (n_ids > lnet_interfaces_max)
4142                 n_ids = lnet_interfaces_max;
4143
4144         if (id.pid == LNET_PID_ANY)
4145                 id.pid = LNET_PID_LUSTRE;
4146
4147         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4148         if (!pbuf)
4149                 return -ENOMEM;
4150
4151         /* NB 2 events max (including any unlink event) */
4152         rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
4153         if (rc != 0) {
4154                 CERROR("Can't allocate EQ: %d\n", rc);
4155                 goto fail_ping_buffer_decref;
4156         }
4157
4158         /* initialize md content */
4159         md.start     = &pbuf->pb_info;
4160         md.length    = LNET_PING_INFO_SIZE(n_ids);
4161         md.threshold = 2; /* GET/REPLY */
4162         md.max_size  = 0;
4163         md.options   = LNET_MD_TRUNCATE;
4164         md.user_ptr  = NULL;
4165         md.eq_handle = eqh;
4166
4167         rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4168         if (rc != 0) {
4169                 CERROR("Can't bind MD: %d\n", rc);
4170                 goto fail_free_eq;
4171         }
4172
4173         rc = LNetGet(LNET_NID_ANY, mdh, id,
4174                      LNET_RESERVED_PORTAL,
4175                      LNET_PROTO_PING_MATCHBITS, 0, false);
4176
4177         if (rc != 0) {
4178                 /* Don't CERROR; this could be deliberate! */
4179                 rc2 = LNetMDUnlink(mdh);
4180                 LASSERT(rc2 == 0);
4181
4182                 /* NB must wait for the UNLINK event below... */
4183                 unlinked = 1;
4184                 timeout = a_long_time;
4185         }
4186
4187         do {
4188                 /* MUST block for unlink to complete */
4189                 if (unlinked)
4190                         blocked = cfs_block_allsigs();
4191
4192                 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
4193
4194                 if (unlinked)
4195                         cfs_restore_sigs(blocked);
4196
4197                 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4198                        (rc2 <= 0) ? -1 : event.type,
4199                        (rc2 <= 0) ? -1 : event.status,
4200                        (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4201
4202                 LASSERT(rc2 != -EOVERFLOW);     /* can't miss anything */
4203
4204                 if (rc2 <= 0 || event.status != 0) {
4205                         /* timeout or error */
4206                         if (!replied && rc == 0)
4207                                 rc = (rc2 < 0) ? rc2 :
4208                                      (rc2 == 0) ? -ETIMEDOUT :
4209                                      event.status;
4210
4211                         if (!unlinked) {
4212                                 /* Ensure completion in finite time... */
4213                                 LNetMDUnlink(mdh);
4214                                 /* No assertion (racing with network) */
4215                                 unlinked = 1;
4216                                 timeout = a_long_time;
4217                         } else if (rc2 == 0) {
4218                                 /* timed out waiting for unlink */
4219                                 CWARN("ping %s: late network completion\n",
4220                                       libcfs_id2str(id));
4221                         }
4222                 } else if (event.type == LNET_EVENT_REPLY) {
4223                         replied = 1;
4224                         rc = event.mlength;
4225                 }
4226         } while (rc2 <= 0 || !event.unlinked);
4227
4228         if (!replied) {
4229                 if (rc >= 0)
4230                         CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4231                               libcfs_id2str(id));
4232                 rc = -EIO;
4233                 goto fail_free_eq;
4234         }
4235
4236         nob = rc;
4237         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4238
4239         rc = -EPROTO;           /* if I can't parse... */
4240
4241         if (nob < 8) {
4242                 CERROR("%s: ping info too short %d\n",
4243                        libcfs_id2str(id), nob);
4244                 goto fail_free_eq;
4245         }
4246
4247         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4248                 lnet_swap_pinginfo(pbuf);
4249         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4250                 CERROR("%s: Unexpected magic %08x\n",
4251                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4252                 goto fail_free_eq;
4253         }
4254
4255         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4256                 CERROR("%s: ping w/o NI status: 0x%x\n",
4257                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4258                 goto fail_free_eq;
4259         }
4260
4261         if (nob < LNET_PING_INFO_SIZE(0)) {
4262                 CERROR("%s: Short reply %d(%d min)\n",
4263                        libcfs_id2str(id),
4264                        nob, (int)LNET_PING_INFO_SIZE(0));
4265                 goto fail_free_eq;
4266         }
4267
4268         if (pbuf->pb_info.pi_nnis < n_ids)
4269                 n_ids = pbuf->pb_info.pi_nnis;
4270
4271         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4272                 CERROR("%s: Short reply %d(%d expected)\n",
4273                        libcfs_id2str(id),
4274                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4275                 goto fail_free_eq;
4276         }
4277
4278         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4279
4280         memset(&tmpid, 0, sizeof(tmpid));
4281         for (i = 0; i < n_ids; i++) {
4282                 tmpid.pid = pbuf->pb_info.pi_pid;
4283                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4284                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4285                         goto fail_free_eq;
4286         }
4287         rc = pbuf->pb_info.pi_nnis;
4288
4289  fail_free_eq:
4290         rc2 = LNetEQFree(eqh);
4291         if (rc2 != 0)
4292                 CERROR("rc2 %d\n", rc2);
4293         LASSERT(rc2 == 0);
4294
4295  fail_ping_buffer_decref:
4296         lnet_ping_buffer_decref(pbuf);
4297         return rc;
4298 }
4299
4300 static int
4301 lnet_discover(struct lnet_process_id id, __u32 force,
4302               struct lnet_process_id __user *ids, int n_ids)
4303 {
4304         struct lnet_peer_ni *lpni;
4305         struct lnet_peer_ni *p;
4306         struct lnet_peer *lp;
4307         struct lnet_process_id *buf;
4308         int cpt;
4309         int i;
4310         int rc;
4311         int max_intf = lnet_interfaces_max;
4312         size_t buf_size;
4313
4314         if (n_ids <= 0 ||
4315             id.nid == LNET_NID_ANY)
4316                 return -EINVAL;
4317
4318         if (id.pid == LNET_PID_ANY)
4319                 id.pid = LNET_PID_LUSTRE;
4320
4321         /*
4322          * if the user buffer has more space than the max_intf
4323          * then only fill it up to max_intf
4324          */
4325         if (n_ids > max_intf)
4326                 n_ids = max_intf;
4327
4328         buf_size = n_ids * sizeof(*buf);
4329
4330         LIBCFS_ALLOC(buf, buf_size);
4331         if (!buf)
4332                 return -ENOMEM;
4333
4334         cpt = lnet_net_lock_current();
4335         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4336         if (IS_ERR(lpni)) {
4337                 rc = PTR_ERR(lpni);
4338                 goto out;
4339         }
4340
4341         /*
4342          * Clearing the NIDS_UPTODATE flag ensures the peer will
4343          * be discovered, provided discovery has not been disabled.
4344          */
4345         lp = lpni->lpni_peer_net->lpn_peer;
4346         spin_lock(&lp->lp_lock);
4347         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4348         /* If the force flag is set, force a PING and PUSH as well. */
4349         if (force)
4350                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4351         spin_unlock(&lp->lp_lock);
4352         rc = lnet_discover_peer_locked(lpni, cpt, true);
4353         if (rc)
4354                 goto out_decref;
4355
4356         /* Peer may have changed. */
4357         lp = lpni->lpni_peer_net->lpn_peer;
4358         if (lp->lp_nnis < n_ids)
4359                 n_ids = lp->lp_nnis;
4360
4361         i = 0;
4362         p = NULL;
4363         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4364                 buf[i].pid = id.pid;
4365                 buf[i].nid = p->lpni_nid;
4366                 if (++i >= n_ids)
4367                         break;
4368         }
4369
4370         lnet_net_unlock(cpt);
4371
4372         rc = -EFAULT;
4373         if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4374                 goto out_relock;
4375         rc = n_ids;
4376 out_relock:
4377         lnet_net_lock(cpt);
4378 out_decref:
4379         lnet_peer_ni_decref_locked(lpni);
4380 out:
4381         lnet_net_unlock(cpt);
4382
4383         LIBCFS_FREE(buf, buf_size);
4384
4385         return rc;
4386 }
4387
4388 /**
4389  * Retrieve peer discovery status.
4390  *
4391  * \retval 1 if lnet_peer_discovery_disabled is 0
4392  * \retval 0 if lnet_peer_discovery_disabled is 1
4393  */
4394 int
4395 LNetGetPeerDiscoveryStatus(void)
4396 {
4397         return !lnet_peer_discovery_disabled;
4398 }
4399 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);