Whamcloud - gitweb
LU-13569 lnet: Add health ping stats
[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  */
31
32 #define DEBUG_SUBSYSTEM S_LNET
33
34 #include <linux/ctype.h>
35 #include <linux/log2.h>
36 #include <linux/ktime.h>
37 #include <linux/moduleparam.h>
38 #include <linux/uaccess.h>
39 #ifdef HAVE_SCHED_HEADERS
40 #include <linux/sched/signal.h>
41 #endif
42
43 #include <lnet/udsp.h>
44 #include <lnet/lib-lnet.h>
45
46 #define D_LNI D_CONSOLE
47
48 /*
49  * initialize ln_api_mutex statically, since it needs to be used in
50  * discovery_set callback. That module parameter callback can be called
51  * before module init completes. The mutex needs to be ready for use then.
52  */
53 struct lnet the_lnet = {
54         .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
55 };              /* THE state of the network */
56 EXPORT_SYMBOL(the_lnet);
57
58 static char *ip2nets = "";
59 module_param(ip2nets, charp, 0444);
60 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
61
62 static char *networks = "";
63 module_param(networks, charp, 0444);
64 MODULE_PARM_DESC(networks, "local networks");
65
66 static char *routes = "";
67 module_param(routes, charp, 0444);
68 MODULE_PARM_DESC(routes, "routes to non-local networks");
69
70 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
71 module_param(rnet_htable_size, int, 0444);
72 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
73
74 static int use_tcp_bonding;
75 module_param(use_tcp_bonding, int, 0444);
76 MODULE_PARM_DESC(use_tcp_bonding,
77                  "use_tcp_bonding parameter has been removed");
78
79 unsigned int lnet_numa_range = 0;
80 module_param(lnet_numa_range, uint, 0444);
81 MODULE_PARM_DESC(lnet_numa_range,
82                 "NUMA range to consider during Multi-Rail selection");
83
84 /*
85  * lnet_health_sensitivity determines by how much we decrement the health
86  * value on sending error. The value defaults to 100, which means health
87  * interface health is decremented by 100 points every failure.
88  */
89 unsigned int lnet_health_sensitivity = 100;
90 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
91 #ifdef HAVE_KERNEL_PARAM_OPS
92 static struct kernel_param_ops param_ops_health_sensitivity = {
93         .set = sensitivity_set,
94         .get = param_get_int,
95 };
96 #define param_check_health_sensitivity(name, p) \
97                 __param_check(name, p, int)
98 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
99 #else
100 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
101                   &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
102 #endif
103 MODULE_PARM_DESC(lnet_health_sensitivity,
104                 "Value to decrement the health value by on error");
105
106 /*
107  * lnet_recovery_interval determines how often we should perform recovery
108  * on unhealthy interfaces.
109  */
110 unsigned int lnet_recovery_interval = 1;
111 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
112 #ifdef HAVE_KERNEL_PARAM_OPS
113 static struct kernel_param_ops param_ops_recovery_interval = {
114         .set = recovery_interval_set,
115         .get = param_get_int,
116 };
117 #define param_check_recovery_interval(name, p) \
118                 __param_check(name, p, int)
119 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
120 #else
121 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
122                   &lnet_recovery_interval, S_IRUGO|S_IWUSR);
123 #endif
124 MODULE_PARM_DESC(lnet_recovery_interval,
125                 "DEPRECATED - Interval to recover unhealthy interfaces in seconds");
126
127 unsigned int lnet_recovery_limit;
128 module_param(lnet_recovery_limit, uint, 0644);
129 MODULE_PARM_DESC(lnet_recovery_limit,
130                  "How long to attempt recovery of unhealthy peer interfaces in seconds. Set to 0 to allow indefinite recovery");
131
132 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
133 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
134
135 static struct kernel_param_ops param_ops_interfaces_max = {
136         .set = intf_max_set,
137         .get = param_get_int,
138 };
139
140 #define param_check_interfaces_max(name, p) \
141                 __param_check(name, p, int)
142
143 #ifdef HAVE_KERNEL_PARAM_OPS
144 module_param(lnet_interfaces_max, interfaces_max, 0644);
145 #else
146 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
147                   &param_ops_interfaces_max, 0644);
148 #endif
149 MODULE_PARM_DESC(lnet_interfaces_max,
150                 "Maximum number of interfaces in a node.");
151
152 unsigned lnet_peer_discovery_disabled = 0;
153 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
154
155 static struct kernel_param_ops param_ops_discovery_disabled = {
156         .set = discovery_set,
157         .get = param_get_int,
158 };
159
160 #define param_check_discovery_disabled(name, p) \
161                 __param_check(name, p, int)
162 #ifdef HAVE_KERNEL_PARAM_OPS
163 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
164 #else
165 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
166                   &param_ops_discovery_disabled, 0644);
167 #endif
168 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
169                 "Set to 1 to disable peer discovery on this node.");
170
171 unsigned int lnet_drop_asym_route;
172 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
173
174 static struct kernel_param_ops param_ops_drop_asym_route = {
175         .set = drop_asym_route_set,
176         .get = param_get_int,
177 };
178
179 #define param_check_drop_asym_route(name, p)    \
180         __param_check(name, p, int)
181 #ifdef HAVE_KERNEL_PARAM_OPS
182 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
183 #else
184 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
185                   &param_ops_drop_asym_route, 0644);
186 #endif
187 MODULE_PARM_DESC(lnet_drop_asym_route,
188                  "Set to 1 to drop asymmetrical route messages.");
189
190 #define LNET_TRANSACTION_TIMEOUT_DEFAULT 50
191 unsigned int lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_DEFAULT;
192 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
193 #ifdef HAVE_KERNEL_PARAM_OPS
194 static struct kernel_param_ops param_ops_transaction_timeout = {
195         .set = transaction_to_set,
196         .get = param_get_int,
197 };
198
199 #define param_check_transaction_timeout(name, p) \
200                 __param_check(name, p, int)
201 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
202 #else
203 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
204                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
205 #endif
206 MODULE_PARM_DESC(lnet_transaction_timeout,
207                 "Maximum number of seconds to wait for a peer response.");
208
209 #define LNET_RETRY_COUNT_DEFAULT 2
210 unsigned int lnet_retry_count = LNET_RETRY_COUNT_DEFAULT;
211 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
212 #ifdef HAVE_KERNEL_PARAM_OPS
213 static struct kernel_param_ops param_ops_retry_count = {
214         .set = retry_count_set,
215         .get = param_get_int,
216 };
217
218 #define param_check_retry_count(name, p) \
219                 __param_check(name, p, int)
220 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
221 #else
222 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
223                   &lnet_retry_count, S_IRUGO|S_IWUSR);
224 #endif
225 MODULE_PARM_DESC(lnet_retry_count,
226                  "Maximum number of times to retry transmitting a message");
227
228 unsigned int lnet_response_tracking = 3;
229 static int response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp);
230
231 #ifdef HAVE_KERNEL_PARAM_OPS
232 static struct kernel_param_ops param_ops_response_tracking = {
233         .set = response_tracking_set,
234         .get = param_get_int,
235 };
236
237 #define param_check_response_tracking(name, p)  \
238         __param_check(name, p, int)
239 module_param(lnet_response_tracking, response_tracking, 0644);
240 #else
241 module_param_call(lnet_response_tracking, response_tracking_set, param_get_int,
242                   &lnet_response_tracking, 0644);
243 #endif
244 MODULE_PARM_DESC(lnet_response_tracking,
245                  "(0|1|2|3) LNet Internal Only|GET Reply only|PUT ACK only|Full Tracking (default)");
246
247 #define LNET_LND_TIMEOUT_DEFAULT ((LNET_TRANSACTION_TIMEOUT_DEFAULT - 1) / \
248                                   (LNET_RETRY_COUNT_DEFAULT + 1))
249 unsigned int lnet_lnd_timeout = LNET_LND_TIMEOUT_DEFAULT;
250 static void lnet_set_lnd_timeout(void)
251 {
252         lnet_lnd_timeout = (lnet_transaction_timeout - 1) /
253                            (lnet_retry_count + 1);
254 }
255
256 unsigned int lnet_current_net_count;
257
258 /*
259  * This sequence number keeps track of how many times DLC was used to
260  * update the local NIs. It is incremented when a NI is added or
261  * removed and checked when sending a message to determine if there is
262  * a need to re-run the selection algorithm. See lnet_select_pathway()
263  * for more details on its usage.
264  */
265 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
266
267 static int lnet_ping(struct lnet_process_id id, signed long timeout,
268                      struct lnet_process_id __user *ids, int n_ids);
269
270 static int lnet_discover(struct lnet_process_id id, __u32 force,
271                          struct lnet_process_id __user *ids, int n_ids);
272
273 static int
274 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
275 {
276         int rc;
277         unsigned *sensitivity = (unsigned *)kp->arg;
278         unsigned long value;
279
280         rc = kstrtoul(val, 0, &value);
281         if (rc) {
282                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
283                 return rc;
284         }
285
286         /*
287          * The purpose of locking the api_mutex here is to ensure that
288          * the correct value ends up stored properly.
289          */
290         mutex_lock(&the_lnet.ln_api_mutex);
291
292         if (value > LNET_MAX_HEALTH_VALUE) {
293                 mutex_unlock(&the_lnet.ln_api_mutex);
294                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
295                        LNET_MAX_HEALTH_VALUE, value);
296                 return -EINVAL;
297         }
298
299         if (*sensitivity != 0 && value == 0 && lnet_retry_count != 0) {
300                 lnet_retry_count = 0;
301                 lnet_set_lnd_timeout();
302         }
303
304         *sensitivity = value;
305
306         mutex_unlock(&the_lnet.ln_api_mutex);
307
308         return 0;
309 }
310
311 static int
312 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
313 {
314         CWARN("'lnet_recovery_interval' has been deprecated\n");
315
316         return 0;
317 }
318
319 static int
320 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
321 {
322         int rc;
323         unsigned *discovery_off = (unsigned *)kp->arg;
324         unsigned long value;
325         struct lnet_ping_buffer *pbuf;
326
327         rc = kstrtoul(val, 0, &value);
328         if (rc) {
329                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
330                 return rc;
331         }
332
333         value = (value) ? 1 : 0;
334
335         /*
336          * The purpose of locking the api_mutex here is to ensure that
337          * the correct value ends up stored properly.
338          */
339         mutex_lock(&the_lnet.ln_api_mutex);
340
341         if (value == *discovery_off) {
342                 mutex_unlock(&the_lnet.ln_api_mutex);
343                 return 0;
344         }
345
346         /*
347          * We still want to set the discovery value even when LNet is not
348          * running. This is the case when LNet is being loaded and we want
349          * the module parameters to take effect. Otherwise if we're
350          * changing the value dynamically, we want to set it after
351          * updating the peers
352          */
353         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
354                 *discovery_off = value;
355                 mutex_unlock(&the_lnet.ln_api_mutex);
356                 return 0;
357         }
358
359         /* tell peers that discovery setting has changed */
360         lnet_net_lock(LNET_LOCK_EX);
361         pbuf = the_lnet.ln_ping_target;
362         if (value)
363                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
364         else
365                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
366         lnet_net_unlock(LNET_LOCK_EX);
367
368         /* only send a push when we're turning off discovery */
369         if (*discovery_off <= 0 && value > 0)
370                 lnet_push_update_to_peers(1);
371         *discovery_off = value;
372
373         mutex_unlock(&the_lnet.ln_api_mutex);
374
375         return 0;
376 }
377
378 static int
379 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
380 {
381         int rc;
382         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
383         unsigned long value;
384
385         rc = kstrtoul(val, 0, &value);
386         if (rc) {
387                 CERROR("Invalid module parameter value for "
388                        "'lnet_drop_asym_route'\n");
389                 return rc;
390         }
391
392         /*
393          * The purpose of locking the api_mutex here is to ensure that
394          * the correct value ends up stored properly.
395          */
396         mutex_lock(&the_lnet.ln_api_mutex);
397
398         if (value == *drop_asym_route) {
399                 mutex_unlock(&the_lnet.ln_api_mutex);
400                 return 0;
401         }
402
403         *drop_asym_route = value;
404
405         mutex_unlock(&the_lnet.ln_api_mutex);
406
407         return 0;
408 }
409
410 static int
411 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
412 {
413         int rc;
414         unsigned *transaction_to = (unsigned *)kp->arg;
415         unsigned long value;
416
417         rc = kstrtoul(val, 0, &value);
418         if (rc) {
419                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
420                 return rc;
421         }
422
423         /*
424          * The purpose of locking the api_mutex here is to ensure that
425          * the correct value ends up stored properly.
426          */
427         mutex_lock(&the_lnet.ln_api_mutex);
428
429         if (value <= lnet_retry_count || value == 0) {
430                 mutex_unlock(&the_lnet.ln_api_mutex);
431                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
432                        "Has to be greater than lnet_retry_count (%u)\n",
433                        value, lnet_retry_count);
434                 return -EINVAL;
435         }
436
437         if (value == *transaction_to) {
438                 mutex_unlock(&the_lnet.ln_api_mutex);
439                 return 0;
440         }
441
442         *transaction_to = value;
443         /* Update the lnet_lnd_timeout now that we've modified the
444          * transaction timeout
445          */
446         lnet_set_lnd_timeout();
447
448         mutex_unlock(&the_lnet.ln_api_mutex);
449
450         return 0;
451 }
452
453 static int
454 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
455 {
456         int rc;
457         unsigned *retry_count = (unsigned *)kp->arg;
458         unsigned long value;
459
460         rc = kstrtoul(val, 0, &value);
461         if (rc) {
462                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
463                 return rc;
464         }
465
466         /*
467          * The purpose of locking the api_mutex here is to ensure that
468          * the correct value ends up stored properly.
469          */
470         mutex_lock(&the_lnet.ln_api_mutex);
471
472         if (lnet_health_sensitivity == 0 && value > 0) {
473                 mutex_unlock(&the_lnet.ln_api_mutex);
474                 CERROR("Can not set lnet_retry_count when health feature is turned off\n");
475                 return -EINVAL;
476         }
477
478         if (value > lnet_transaction_timeout) {
479                 mutex_unlock(&the_lnet.ln_api_mutex);
480                 CERROR("Invalid value for lnet_retry_count (%lu). "
481                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
482                        value, lnet_transaction_timeout);
483                 return -EINVAL;
484         }
485
486         *retry_count = value;
487
488         /* Update the lnet_lnd_timeout now that we've modified the
489          * retry count
490          */
491         lnet_set_lnd_timeout();
492
493         mutex_unlock(&the_lnet.ln_api_mutex);
494
495         return 0;
496 }
497
498 static int
499 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
500 {
501         int value, rc;
502
503         rc = kstrtoint(val, 0, &value);
504         if (rc) {
505                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
506                 return rc;
507         }
508
509         if (value < LNET_INTERFACES_MIN) {
510                 CWARN("max interfaces provided are too small, setting to %d\n",
511                       LNET_INTERFACES_MAX_DEFAULT);
512                 value = LNET_INTERFACES_MAX_DEFAULT;
513         }
514
515         *(int *)kp->arg = value;
516
517         return 0;
518 }
519
520 static int
521 response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp)
522 {
523         int rc;
524         unsigned long new_value;
525
526         rc = kstrtoul(val, 0, &new_value);
527         if (rc) {
528                 CERROR("Invalid value for 'lnet_response_tracking'\n");
529                 return -EINVAL;
530         }
531
532         if (new_value < 0 || new_value > 3) {
533                 CWARN("Invalid value (%lu) for 'lnet_response_tracking'\n",
534                       new_value);
535                 return -EINVAL;
536         }
537
538         lnet_response_tracking = new_value;
539
540         return 0;
541 }
542
543 static const char *
544 lnet_get_routes(void)
545 {
546         return routes;
547 }
548
549 static const char *
550 lnet_get_networks(void)
551 {
552         const char *nets;
553         int rc;
554
555         if (*networks != 0 && *ip2nets != 0) {
556                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
557                                    "'ip2nets' but not both at once\n");
558                 return NULL;
559         }
560
561         if (*ip2nets != 0) {
562                 rc = lnet_parse_ip2nets(&nets, ip2nets);
563                 return (rc == 0) ? nets : NULL;
564         }
565
566         if (*networks != 0)
567                 return networks;
568
569         return "tcp";
570 }
571
572 static void
573 lnet_init_locks(void)
574 {
575         spin_lock_init(&the_lnet.ln_eq_wait_lock);
576         spin_lock_init(&the_lnet.ln_msg_resend_lock);
577         init_completion(&the_lnet.ln_mt_wait_complete);
578         mutex_init(&the_lnet.ln_lnd_mutex);
579 }
580
581 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
582 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
583                                             *  MDs kmem_cache */
584 struct kmem_cache *lnet_udsp_cachep;       /* udsp cache */
585 struct kmem_cache *lnet_rspt_cachep;       /* response tracker cache */
586 struct kmem_cache *lnet_msg_cachep;
587
588 static int
589 lnet_slab_setup(void)
590 {
591         /* create specific kmem_cache for MEs and small MDs (i.e., originally
592          * allocated in <size-xxx> kmem_cache).
593          */
594         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
595                                             0, 0, NULL);
596         if (!lnet_mes_cachep)
597                 return -ENOMEM;
598
599         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
600                                                   LNET_SMALL_MD_SIZE, 0, 0,
601                                                   NULL);
602         if (!lnet_small_mds_cachep)
603                 return -ENOMEM;
604
605         lnet_udsp_cachep = kmem_cache_create("lnet_udsp",
606                                              sizeof(struct lnet_udsp),
607                                              0, 0, NULL);
608         if (!lnet_udsp_cachep)
609                 return -ENOMEM;
610
611         lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
612                                             0, 0, NULL);
613         if (!lnet_rspt_cachep)
614                 return -ENOMEM;
615
616         lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
617                                             0, 0, NULL);
618         if (!lnet_msg_cachep)
619                 return -ENOMEM;
620
621         return 0;
622 }
623
624 static void
625 lnet_slab_cleanup(void)
626 {
627         if (lnet_msg_cachep) {
628                 kmem_cache_destroy(lnet_msg_cachep);
629                 lnet_msg_cachep = NULL;
630         }
631
632         if (lnet_rspt_cachep) {
633                 kmem_cache_destroy(lnet_rspt_cachep);
634                 lnet_rspt_cachep = NULL;
635         }
636
637         if (lnet_udsp_cachep) {
638                 kmem_cache_destroy(lnet_udsp_cachep);
639                 lnet_udsp_cachep = NULL;
640         }
641
642         if (lnet_small_mds_cachep) {
643                 kmem_cache_destroy(lnet_small_mds_cachep);
644                 lnet_small_mds_cachep = NULL;
645         }
646
647         if (lnet_mes_cachep) {
648                 kmem_cache_destroy(lnet_mes_cachep);
649                 lnet_mes_cachep = NULL;
650         }
651 }
652
653 static int
654 lnet_create_remote_nets_table(void)
655 {
656         int               i;
657         struct list_head *hash;
658
659         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
660         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
661         CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
662         if (hash == NULL) {
663                 CERROR("Failed to create remote nets hash table\n");
664                 return -ENOMEM;
665         }
666
667         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
668                 INIT_LIST_HEAD(&hash[i]);
669         the_lnet.ln_remote_nets_hash = hash;
670         return 0;
671 }
672
673 static void
674 lnet_destroy_remote_nets_table(void)
675 {
676         int i;
677
678         if (the_lnet.ln_remote_nets_hash == NULL)
679                 return;
680
681         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
682                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
683
684         CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
685                            LNET_REMOTE_NETS_HASH_SIZE);
686         the_lnet.ln_remote_nets_hash = NULL;
687 }
688
689 static void
690 lnet_destroy_locks(void)
691 {
692         if (the_lnet.ln_res_lock != NULL) {
693                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
694                 the_lnet.ln_res_lock = NULL;
695         }
696
697         if (the_lnet.ln_net_lock != NULL) {
698                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
699                 the_lnet.ln_net_lock = NULL;
700         }
701 }
702
703 static int
704 lnet_create_locks(void)
705 {
706         lnet_init_locks();
707
708         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
709         if (the_lnet.ln_res_lock == NULL)
710                 goto failed;
711
712         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
713         if (the_lnet.ln_net_lock == NULL)
714                 goto failed;
715
716         return 0;
717
718  failed:
719         lnet_destroy_locks();
720         return -ENOMEM;
721 }
722
723 static void lnet_assert_wire_constants(void)
724 {
725         /* Wire protocol assertions generated by 'wirecheck'
726          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
727          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
728          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
729          */
730
731         /* Constants... */
732         BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
733         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
734         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
735         BUILD_BUG_ON(LNET_MSG_ACK != 0);
736         BUILD_BUG_ON(LNET_MSG_PUT != 1);
737         BUILD_BUG_ON(LNET_MSG_GET != 2);
738         BUILD_BUG_ON(LNET_MSG_REPLY != 3);
739         BUILD_BUG_ON(LNET_MSG_HELLO != 4);
740
741         BUILD_BUG_ON((int)sizeof(lnet_nid_t) != 8);
742         BUILD_BUG_ON((int)sizeof(lnet_pid_t) != 4);
743
744         /* Checks for struct lnet_process_id_packed */
745         BUILD_BUG_ON((int)sizeof(struct lnet_process_id_packed) != 12);
746         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, nid) != 0);
747         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->nid) != 8);
748         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, pid) != 8);
749         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->pid) != 4);
750
751         /* Checks for struct lnet_handle_wire */
752         BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
753         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
754                                    wh_interface_cookie) != 0);
755         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
756         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
757                                    wh_object_cookie) != 8);
758         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
759
760         /* Checks for struct struct lnet_magicversion */
761         BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
762         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
763         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
764         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
765         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
766         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
767                                    version_minor) != 6);
768         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
769
770         /* Checks for struct struct lnet_hdr */
771         BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
772         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
773         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
774         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
775         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
776         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
777         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
778         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
779         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
780         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
781         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
782         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
783         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
784         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
785         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
786
787         /* Ack */
788         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
789         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
790         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
791         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
792         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
793         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
794
795         /* Put */
796         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
797         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
798         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
799         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
800         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
801         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
802         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
803         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
804         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
805         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
806
807         /* Get */
808         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
809         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
810         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
811         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
812         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
813         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
814         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
815         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
816         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
817         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
818
819         /* Reply */
820         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
821         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
822
823         /* Hello */
824         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
825         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
826         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
827         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
828
829         /* Checks for struct lnet_ni_status and related constants */
830         BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
831         BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
832         BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
833
834         /* Checks for struct lnet_ni_status */
835         BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
836         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
837         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
838         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
839         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
840         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
841         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
842
843         /* Checks for struct lnet_ping_info and related constants */
844         BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
845         BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
846         BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
847         BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
848         BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
849         BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
850         BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
851         BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
852
853         /* Checks for struct lnet_ping_info */
854         BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
855         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
856         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
857         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
858         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
859         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
860         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
861         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
862         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
863         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
864         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
865
866         /* Acceptor connection request */
867         BUILD_BUG_ON(LNET_PROTO_ACCEPTOR_VERSION != 1);
868
869         /* Checks for struct lnet_acceptor_connreq */
870         BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq) != 16);
871         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_magic) != 0);
872         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_magic) != 4);
873         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_version) != 4);
874         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_version) != 4);
875         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_nid) != 8);
876         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_nid) != 8);
877
878         /* Checks for struct lnet_counters_common */
879         BUILD_BUG_ON((int)sizeof(struct lnet_counters_common) != 60);
880         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_alloc) != 0);
881         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_alloc) != 4);
882         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_max) != 4);
883         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_max) != 4);
884         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_errors) != 8);
885         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_errors) != 4);
886         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_count) != 12);
887         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_count) != 4);
888         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_count) != 16);
889         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_count) != 4);
890         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_count) != 20);
891         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_count) != 4);
892         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_count) != 24);
893         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_count) != 4);
894         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_length) != 28);
895         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_length) != 8);
896         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_length) != 36);
897         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_length) != 8);
898         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_length) != 44);
899         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_length) != 8);
900         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_length) != 52);
901         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_length) != 8);
902 }
903
904 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
905 {
906         const struct lnet_lnd *lnd;
907
908         /* holding lnd mutex */
909         if (type >= NUM_LNDS)
910                 return NULL;
911         lnd = the_lnet.ln_lnds[type];
912         LASSERT(!lnd || lnd->lnd_type == type);
913
914         return lnd;
915 }
916
917 unsigned int
918 lnet_get_lnd_timeout(void)
919 {
920         return lnet_lnd_timeout;
921 }
922 EXPORT_SYMBOL(lnet_get_lnd_timeout);
923
924 void
925 lnet_register_lnd(const struct lnet_lnd *lnd)
926 {
927         mutex_lock(&the_lnet.ln_lnd_mutex);
928
929         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
930         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
931
932         the_lnet.ln_lnds[lnd->lnd_type] = lnd;
933
934         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
935
936         mutex_unlock(&the_lnet.ln_lnd_mutex);
937 }
938 EXPORT_SYMBOL(lnet_register_lnd);
939
940 void
941 lnet_unregister_lnd(const struct lnet_lnd *lnd)
942 {
943         mutex_lock(&the_lnet.ln_lnd_mutex);
944
945         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
946
947         the_lnet.ln_lnds[lnd->lnd_type] = NULL;
948         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
949
950         mutex_unlock(&the_lnet.ln_lnd_mutex);
951 }
952 EXPORT_SYMBOL(lnet_unregister_lnd);
953
954 static void
955 lnet_counters_get_common_locked(struct lnet_counters_common *common)
956 {
957         struct lnet_counters *ctr;
958         int i;
959
960         /* FIXME !!! Their is no assert_lnet_net_locked() to ensure this
961          * actually called under the protection of the lnet_net_lock.
962          */
963         memset(common, 0, sizeof(*common));
964
965         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
966                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
967                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
968                 common->lcc_errors       += ctr->lct_common.lcc_errors;
969                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
970                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
971                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
972                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
973                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
974                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
975                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
976                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
977         }
978 }
979
980 void
981 lnet_counters_get_common(struct lnet_counters_common *common)
982 {
983         lnet_net_lock(LNET_LOCK_EX);
984         lnet_counters_get_common_locked(common);
985         lnet_net_unlock(LNET_LOCK_EX);
986 }
987 EXPORT_SYMBOL(lnet_counters_get_common);
988
989 int
990 lnet_counters_get(struct lnet_counters *counters)
991 {
992         struct lnet_counters *ctr;
993         struct lnet_counters_health *health = &counters->lct_health;
994         int i, rc = 0;
995
996         memset(counters, 0, sizeof(*counters));
997
998         lnet_net_lock(LNET_LOCK_EX);
999
1000         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1001                 GOTO(out_unlock, rc = -ENODEV);
1002
1003         lnet_counters_get_common_locked(&counters->lct_common);
1004
1005         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
1006                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
1007                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
1008                 health->lch_response_timeout_count +=
1009                                 ctr->lct_health.lch_response_timeout_count;
1010                 health->lch_local_interrupt_count +=
1011                                 ctr->lct_health.lch_local_interrupt_count;
1012                 health->lch_local_dropped_count +=
1013                                 ctr->lct_health.lch_local_dropped_count;
1014                 health->lch_local_aborted_count +=
1015                                 ctr->lct_health.lch_local_aborted_count;
1016                 health->lch_local_no_route_count +=
1017                                 ctr->lct_health.lch_local_no_route_count;
1018                 health->lch_local_timeout_count +=
1019                                 ctr->lct_health.lch_local_timeout_count;
1020                 health->lch_local_error_count +=
1021                                 ctr->lct_health.lch_local_error_count;
1022                 health->lch_remote_dropped_count +=
1023                                 ctr->lct_health.lch_remote_dropped_count;
1024                 health->lch_remote_error_count +=
1025                                 ctr->lct_health.lch_remote_error_count;
1026                 health->lch_remote_timeout_count +=
1027                                 ctr->lct_health.lch_remote_timeout_count;
1028                 health->lch_network_timeout_count +=
1029                                 ctr->lct_health.lch_network_timeout_count;
1030         }
1031 out_unlock:
1032         lnet_net_unlock(LNET_LOCK_EX);
1033         return rc;
1034 }
1035 EXPORT_SYMBOL(lnet_counters_get);
1036
1037 void
1038 lnet_counters_reset(void)
1039 {
1040         struct lnet_counters *counters;
1041         int             i;
1042
1043         lnet_net_lock(LNET_LOCK_EX);
1044
1045         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1046                 goto avoid_reset;
1047
1048         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
1049                 memset(counters, 0, sizeof(struct lnet_counters));
1050 avoid_reset:
1051         lnet_net_unlock(LNET_LOCK_EX);
1052 }
1053
1054 static char *
1055 lnet_res_type2str(int type)
1056 {
1057         switch (type) {
1058         default:
1059                 LBUG();
1060         case LNET_COOKIE_TYPE_MD:
1061                 return "MD";
1062         case LNET_COOKIE_TYPE_ME:
1063                 return "ME";
1064         case LNET_COOKIE_TYPE_EQ:
1065                 return "EQ";
1066         }
1067 }
1068
1069 static void
1070 lnet_res_container_cleanup(struct lnet_res_container *rec)
1071 {
1072         int     count = 0;
1073
1074         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
1075                 return;
1076
1077         while (!list_empty(&rec->rec_active)) {
1078                 struct list_head *e = rec->rec_active.next;
1079
1080                 list_del_init(e);
1081                 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1082                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1083
1084                 } else { /* NB: Active MEs should be attached on portals */
1085                         LBUG();
1086                 }
1087                 count++;
1088         }
1089
1090         if (count > 0) {
1091                 /* Found alive MD/ME/EQ, user really should unlink/free
1092                  * all of them before finalize LNet, but if someone didn't,
1093                  * we have to recycle garbage for him */
1094                 CERROR("%d active elements on exit of %s container\n",
1095                        count, lnet_res_type2str(rec->rec_type));
1096         }
1097
1098         if (rec->rec_lh_hash != NULL) {
1099                 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1100                 rec->rec_lh_hash = NULL;
1101         }
1102
1103         rec->rec_type = 0; /* mark it as finalized */
1104 }
1105
1106 static int
1107 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1108 {
1109         int     rc = 0;
1110         int     i;
1111
1112         LASSERT(rec->rec_type == 0);
1113
1114         rec->rec_type = type;
1115         INIT_LIST_HEAD(&rec->rec_active);
1116
1117         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1118
1119         /* Arbitrary choice of hash table size */
1120         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1121                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1122         if (rec->rec_lh_hash == NULL) {
1123                 rc = -ENOMEM;
1124                 goto out;
1125         }
1126
1127         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1128                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1129
1130         return 0;
1131
1132 out:
1133         CERROR("Failed to setup %s resource container\n",
1134                lnet_res_type2str(type));
1135         lnet_res_container_cleanup(rec);
1136         return rc;
1137 }
1138
1139 static void
1140 lnet_res_containers_destroy(struct lnet_res_container **recs)
1141 {
1142         struct lnet_res_container       *rec;
1143         int                             i;
1144
1145         cfs_percpt_for_each(rec, i, recs)
1146                 lnet_res_container_cleanup(rec);
1147
1148         cfs_percpt_free(recs);
1149 }
1150
1151 static struct lnet_res_container **
1152 lnet_res_containers_create(int type)
1153 {
1154         struct lnet_res_container       **recs;
1155         struct lnet_res_container       *rec;
1156         int                             rc;
1157         int                             i;
1158
1159         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1160         if (recs == NULL) {
1161                 CERROR("Failed to allocate %s resource containers\n",
1162                        lnet_res_type2str(type));
1163                 return NULL;
1164         }
1165
1166         cfs_percpt_for_each(rec, i, recs) {
1167                 rc = lnet_res_container_setup(rec, i, type);
1168                 if (rc != 0) {
1169                         lnet_res_containers_destroy(recs);
1170                         return NULL;
1171                 }
1172         }
1173
1174         return recs;
1175 }
1176
1177 struct lnet_libhandle *
1178 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1179 {
1180         /* ALWAYS called with lnet_res_lock held */
1181         struct list_head        *head;
1182         struct lnet_libhandle   *lh;
1183         unsigned int            hash;
1184
1185         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1186                 return NULL;
1187
1188         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1189         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1190
1191         list_for_each_entry(lh, head, lh_hash_chain) {
1192                 if (lh->lh_cookie == cookie)
1193                         return lh;
1194         }
1195
1196         return NULL;
1197 }
1198
1199 void
1200 lnet_res_lh_initialize(struct lnet_res_container *rec,
1201                        struct lnet_libhandle *lh)
1202 {
1203         /* ALWAYS called with lnet_res_lock held */
1204         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1205         unsigned int    hash;
1206
1207         lh->lh_cookie = rec->rec_lh_cookie;
1208         rec->rec_lh_cookie += 1 << ibits;
1209
1210         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1211
1212         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1213 }
1214
1215 struct list_head **
1216 lnet_create_array_of_queues(void)
1217 {
1218         struct list_head **qs;
1219         struct list_head *q;
1220         int i;
1221
1222         qs = cfs_percpt_alloc(lnet_cpt_table(),
1223                               sizeof(struct list_head));
1224         if (!qs) {
1225                 CERROR("Failed to allocate queues\n");
1226                 return NULL;
1227         }
1228
1229         cfs_percpt_for_each(q, i, qs)
1230                 INIT_LIST_HEAD(q);
1231
1232         return qs;
1233 }
1234
1235 static int lnet_unprepare(void);
1236
1237 static int
1238 lnet_prepare(lnet_pid_t requested_pid)
1239 {
1240         /* Prepare to bring up the network */
1241         struct lnet_res_container **recs;
1242         int                       rc = 0;
1243
1244         if (requested_pid == LNET_PID_ANY) {
1245                 /* Don't instantiate LNET just for me */
1246                 return -ENETDOWN;
1247         }
1248
1249         LASSERT(the_lnet.ln_refcount == 0);
1250
1251         the_lnet.ln_routing = 0;
1252
1253         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1254         the_lnet.ln_pid = requested_pid;
1255
1256         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1257         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1258         INIT_LIST_HEAD(&the_lnet.ln_nets);
1259         INIT_LIST_HEAD(&the_lnet.ln_routers);
1260         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1261         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1262         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1263         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1264         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1265         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1266         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1267         INIT_LIST_HEAD(&the_lnet.ln_udsp_list);
1268         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1269         the_lnet.ln_mt_handler = NULL;
1270         init_completion(&the_lnet.ln_started);
1271
1272         rc = lnet_slab_setup();
1273         if (rc != 0)
1274                 goto failed;
1275
1276         rc = lnet_create_remote_nets_table();
1277         if (rc != 0)
1278                 goto failed;
1279
1280         /*
1281          * NB the interface cookie in wire handles guards against delayed
1282          * replies and ACKs appearing valid after reboot.
1283          */
1284         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1285
1286         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1287                                                 sizeof(struct lnet_counters));
1288         if (the_lnet.ln_counters == NULL) {
1289                 CERROR("Failed to allocate counters for LNet\n");
1290                 rc = -ENOMEM;
1291                 goto failed;
1292         }
1293
1294         rc = lnet_peer_tables_create();
1295         if (rc != 0)
1296                 goto failed;
1297
1298         rc = lnet_msg_containers_create();
1299         if (rc != 0)
1300                 goto failed;
1301
1302         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1303                                       LNET_COOKIE_TYPE_EQ);
1304         if (rc != 0)
1305                 goto failed;
1306
1307         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1308         if (recs == NULL) {
1309                 rc = -ENOMEM;
1310                 goto failed;
1311         }
1312
1313         the_lnet.ln_md_containers = recs;
1314
1315         rc = lnet_portals_create();
1316         if (rc != 0) {
1317                 CERROR("Failed to create portals for LNet: %d\n", rc);
1318                 goto failed;
1319         }
1320
1321         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1322         if (!the_lnet.ln_mt_zombie_rstqs) {
1323                 rc = -ENOMEM;
1324                 goto failed;
1325         }
1326
1327         return 0;
1328
1329  failed:
1330         lnet_unprepare();
1331         return rc;
1332 }
1333
1334 static int
1335 lnet_unprepare (void)
1336 {
1337         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1338          * have shut down already, so it is safe to unlink and free all
1339          * descriptors, even those that appear committed to a network op (eg MD
1340          * with non-zero pending count) */
1341
1342         lnet_fail_nid(LNET_NID_ANY, 0);
1343
1344         LASSERT(the_lnet.ln_refcount == 0);
1345         LASSERT(list_empty(&the_lnet.ln_test_peers));
1346         LASSERT(list_empty(&the_lnet.ln_nets));
1347
1348         if (the_lnet.ln_mt_zombie_rstqs) {
1349                 lnet_clean_zombie_rstqs();
1350                 the_lnet.ln_mt_zombie_rstqs = NULL;
1351         }
1352
1353         lnet_assert_handler_unused(the_lnet.ln_mt_handler);
1354         the_lnet.ln_mt_handler = NULL;
1355
1356         lnet_portals_destroy();
1357
1358         if (the_lnet.ln_md_containers != NULL) {
1359                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1360                 the_lnet.ln_md_containers = NULL;
1361         }
1362
1363         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1364
1365         lnet_msg_containers_destroy();
1366         lnet_peer_uninit();
1367         lnet_rtrpools_free(0);
1368
1369         if (the_lnet.ln_counters != NULL) {
1370                 cfs_percpt_free(the_lnet.ln_counters);
1371                 the_lnet.ln_counters = NULL;
1372         }
1373         lnet_destroy_remote_nets_table();
1374         lnet_udsp_destroy(true);
1375         lnet_slab_cleanup();
1376
1377         return 0;
1378 }
1379
1380 struct lnet_ni  *
1381 lnet_net2ni_locked(__u32 net_id, int cpt)
1382 {
1383         struct lnet_ni   *ni;
1384         struct lnet_net  *net;
1385
1386         LASSERT(cpt != LNET_LOCK_EX);
1387
1388         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1389                 if (net->net_id == net_id) {
1390                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1391                                         ni_netlist);
1392                         return ni;
1393                 }
1394         }
1395
1396         return NULL;
1397 }
1398
1399 struct lnet_ni *
1400 lnet_net2ni_addref(__u32 net)
1401 {
1402         struct lnet_ni *ni;
1403
1404         lnet_net_lock(0);
1405         ni = lnet_net2ni_locked(net, 0);
1406         if (ni)
1407                 lnet_ni_addref_locked(ni, 0);
1408         lnet_net_unlock(0);
1409
1410         return ni;
1411 }
1412 EXPORT_SYMBOL(lnet_net2ni_addref);
1413
1414 struct lnet_net *
1415 lnet_get_net_locked(__u32 net_id)
1416 {
1417         struct lnet_net  *net;
1418
1419         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1420                 if (net->net_id == net_id)
1421                         return net;
1422         }
1423
1424         return NULL;
1425 }
1426
1427 void
1428 lnet_net_clr_pref_rtrs(struct lnet_net *net)
1429 {
1430         struct list_head zombies;
1431         struct lnet_nid_list *ne;
1432         struct lnet_nid_list *tmp;
1433
1434         INIT_LIST_HEAD(&zombies);
1435
1436         lnet_net_lock(LNET_LOCK_EX);
1437         list_splice_init(&net->net_rtr_pref_nids, &zombies);
1438         lnet_net_unlock(LNET_LOCK_EX);
1439
1440         list_for_each_entry_safe(ne, tmp, &zombies, nl_list) {
1441                 list_del_init(&ne->nl_list);
1442                 LIBCFS_FREE(ne, sizeof(*ne));
1443         }
1444 }
1445
1446 int
1447 lnet_net_add_pref_rtr(struct lnet_net *net,
1448                       lnet_nid_t gw_nid)
1449 __must_hold(&the_lnet.ln_api_mutex)
1450 {
1451         struct lnet_nid_list *ne;
1452
1453         /* This function is called with api_mutex held. When the api_mutex
1454          * is held the list can not be modified, as it is only modified as
1455          * a result of applying a UDSP and that happens under api_mutex
1456          * lock.
1457          */
1458         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1459                 if (ne->nl_nid == gw_nid)
1460                         return -EEXIST;
1461         }
1462
1463         LIBCFS_ALLOC(ne, sizeof(*ne));
1464         if (!ne)
1465                 return -ENOMEM;
1466
1467         ne->nl_nid = gw_nid;
1468
1469         /* Lock the cpt to protect against addition and checks in the
1470          * selection algorithm
1471          */
1472         lnet_net_lock(LNET_LOCK_EX);
1473         list_add(&ne->nl_list, &net->net_rtr_pref_nids);
1474         lnet_net_unlock(LNET_LOCK_EX);
1475
1476         return 0;
1477 }
1478
1479 bool
1480 lnet_net_is_pref_rtr_locked(struct lnet_net *net, lnet_nid_t rtr_nid)
1481 {
1482         struct lnet_nid_list *ne;
1483
1484         CDEBUG(D_NET, "%s: rtr pref emtpy: %d\n",
1485                libcfs_net2str(net->net_id),
1486                list_empty(&net->net_rtr_pref_nids));
1487
1488         if (list_empty(&net->net_rtr_pref_nids))
1489                 return false;
1490
1491         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1492                 CDEBUG(D_NET, "Comparing pref %s with gw %s\n",
1493                        libcfs_nid2str(ne->nl_nid),
1494                        libcfs_nid2str(rtr_nid));
1495                 if (rtr_nid == ne->nl_nid)
1496                         return true;
1497         }
1498
1499         return false;
1500 }
1501
1502 unsigned int
1503 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1504 {
1505         __u64           key = nid;
1506         unsigned int    val;
1507
1508         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1509
1510         if (number == 1)
1511                 return 0;
1512
1513         val = hash_long(key, LNET_CPT_BITS);
1514         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1515         if (val < number)
1516                 return val;
1517
1518         return (unsigned int)(key + val + (val >> 1)) % number;
1519 }
1520
1521 int
1522 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1523 {
1524         struct lnet_net *net;
1525
1526         /* must called with hold of lnet_net_lock */
1527         if (LNET_CPT_NUMBER == 1)
1528                 return 0; /* the only one */
1529
1530         /*
1531          * If NI is provided then use the CPT identified in the NI cpt
1532          * list if one exists. If one doesn't exist, then that NI is
1533          * associated with all CPTs and it follows that the net it belongs
1534          * to is implicitly associated with all CPTs, so just hash the nid
1535          * and return that.
1536          */
1537         if (ni != NULL) {
1538                 if (ni->ni_cpts != NULL)
1539                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1540                                                              ni->ni_ncpts)];
1541                 else
1542                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1543         }
1544
1545         /* no NI provided so look at the net */
1546         net = lnet_get_net_locked(LNET_NIDNET(nid));
1547
1548         if (net != NULL && net->net_cpts != NULL) {
1549                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1550         }
1551
1552         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1553 }
1554
1555 int
1556 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1557 {
1558         int     cpt;
1559         int     cpt2;
1560
1561         if (LNET_CPT_NUMBER == 1)
1562                 return 0; /* the only one */
1563
1564         cpt = lnet_net_lock_current();
1565
1566         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1567
1568         lnet_net_unlock(cpt);
1569
1570         return cpt2;
1571 }
1572 EXPORT_SYMBOL(lnet_cpt_of_nid);
1573
1574 int
1575 lnet_islocalnet_locked(__u32 net_id)
1576 {
1577         struct lnet_net *net;
1578         bool local;
1579
1580         net = lnet_get_net_locked(net_id);
1581
1582         local = net != NULL;
1583
1584         return local;
1585 }
1586
1587 int
1588 lnet_islocalnet(__u32 net_id)
1589 {
1590         int cpt;
1591         bool local;
1592
1593         cpt = lnet_net_lock_current();
1594
1595         local = lnet_islocalnet_locked(net_id);
1596
1597         lnet_net_unlock(cpt);
1598
1599         return local;
1600 }
1601
1602 struct lnet_ni  *
1603 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1604 {
1605         struct lnet_net  *net;
1606         struct lnet_ni   *ni;
1607
1608         LASSERT(cpt != LNET_LOCK_EX);
1609
1610         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1611                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1612                         if (ni->ni_nid == nid)
1613                                 return ni;
1614                 }
1615         }
1616
1617         return NULL;
1618 }
1619
1620 struct lnet_ni *
1621 lnet_nid2ni_addref(lnet_nid_t nid)
1622 {
1623         struct lnet_ni *ni;
1624
1625         lnet_net_lock(0);
1626         ni = lnet_nid2ni_locked(nid, 0);
1627         if (ni)
1628                 lnet_ni_addref_locked(ni, 0);
1629         lnet_net_unlock(0);
1630
1631         return ni;
1632 }
1633 EXPORT_SYMBOL(lnet_nid2ni_addref);
1634
1635 int
1636 lnet_islocalnid(lnet_nid_t nid)
1637 {
1638         struct lnet_ni  *ni;
1639         int             cpt;
1640
1641         cpt = lnet_net_lock_current();
1642         ni = lnet_nid2ni_locked(nid, cpt);
1643         lnet_net_unlock(cpt);
1644
1645         return ni != NULL;
1646 }
1647
1648 int
1649 lnet_count_acceptor_nets(void)
1650 {
1651         /* Return the # of NIs that need the acceptor. */
1652         int              count = 0;
1653         struct lnet_net  *net;
1654         int              cpt;
1655
1656         cpt = lnet_net_lock_current();
1657         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1658                 /* all socklnd type networks should have the acceptor
1659                  * thread started */
1660                 if (net->net_lnd->lnd_accept != NULL)
1661                         count++;
1662         }
1663
1664         lnet_net_unlock(cpt);
1665
1666         return count;
1667 }
1668
1669 struct lnet_ping_buffer *
1670 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1671 {
1672         struct lnet_ping_buffer *pbuf;
1673
1674         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1675         if (pbuf) {
1676                 pbuf->pb_nnis = nnis;
1677                 pbuf->pb_needs_post = false;
1678                 atomic_set(&pbuf->pb_refcnt, 1);
1679         }
1680
1681         return pbuf;
1682 }
1683
1684 void
1685 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1686 {
1687         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1688         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1689 }
1690
1691 static struct lnet_ping_buffer *
1692 lnet_ping_target_create(int nnis)
1693 {
1694         struct lnet_ping_buffer *pbuf;
1695
1696         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1697         if (pbuf == NULL) {
1698                 CERROR("Can't allocate ping source [%d]\n", nnis);
1699                 return NULL;
1700         }
1701
1702         pbuf->pb_info.pi_nnis = nnis;
1703         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1704         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1705         pbuf->pb_info.pi_features =
1706                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1707
1708         return pbuf;
1709 }
1710
1711 static inline int
1712 lnet_get_net_ni_count_locked(struct lnet_net *net)
1713 {
1714         struct lnet_ni  *ni;
1715         int             count = 0;
1716
1717         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1718                 count++;
1719
1720         return count;
1721 }
1722
1723 static inline int
1724 lnet_get_net_ni_count_pre(struct lnet_net *net)
1725 {
1726         struct lnet_ni  *ni;
1727         int             count = 0;
1728
1729         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1730                 count++;
1731
1732         return count;
1733 }
1734
1735 static inline int
1736 lnet_get_ni_count(void)
1737 {
1738         struct lnet_ni  *ni;
1739         struct lnet_net *net;
1740         int             count = 0;
1741
1742         lnet_net_lock(0);
1743
1744         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1745                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1746                         count++;
1747         }
1748
1749         lnet_net_unlock(0);
1750
1751         return count;
1752 }
1753
1754 int
1755 lnet_get_net_count(void)
1756 {
1757         struct lnet_net *net;
1758         int count = 0;
1759
1760         lnet_net_lock(0);
1761
1762         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1763                 count++;
1764         }
1765
1766         lnet_net_unlock(0);
1767
1768         return count;
1769 }
1770
1771 void
1772 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1773 {
1774         struct lnet_ni_status *stat;
1775         int nnis;
1776         int i;
1777
1778         __swab32s(&pbuf->pb_info.pi_magic);
1779         __swab32s(&pbuf->pb_info.pi_features);
1780         __swab32s(&pbuf->pb_info.pi_pid);
1781         __swab32s(&pbuf->pb_info.pi_nnis);
1782         nnis = pbuf->pb_info.pi_nnis;
1783         if (nnis > pbuf->pb_nnis)
1784                 nnis = pbuf->pb_nnis;
1785         for (i = 0; i < nnis; i++) {
1786                 stat = &pbuf->pb_info.pi_ni[i];
1787                 __swab64s(&stat->ns_nid);
1788                 __swab32s(&stat->ns_status);
1789         }
1790 }
1791
1792 int
1793 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1794 {
1795         if (!pinfo)
1796                 return -EINVAL;
1797         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1798                 return -EPROTO;
1799         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1800                 return -EPROTO;
1801         /* Loopback is guaranteed to be present */
1802         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1803                 return -ERANGE;
1804         if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1805                 return -EPROTO;
1806         return 0;
1807 }
1808
1809 static void
1810 lnet_ping_target_destroy(void)
1811 {
1812         struct lnet_net *net;
1813         struct lnet_ni  *ni;
1814
1815         lnet_net_lock(LNET_LOCK_EX);
1816
1817         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1818                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1819                         lnet_ni_lock(ni);
1820                         ni->ni_status = NULL;
1821                         lnet_ni_unlock(ni);
1822                 }
1823         }
1824
1825         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1826         the_lnet.ln_ping_target = NULL;
1827
1828         lnet_net_unlock(LNET_LOCK_EX);
1829 }
1830
1831 static void
1832 lnet_ping_target_event_handler(struct lnet_event *event)
1833 {
1834         struct lnet_ping_buffer *pbuf = event->md_user_ptr;
1835
1836         if (event->unlinked)
1837                 lnet_ping_buffer_decref(pbuf);
1838 }
1839
1840 static int
1841 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1842                        struct lnet_handle_md *ping_mdh,
1843                        int ni_count, bool set_eq)
1844 {
1845         struct lnet_process_id id = {
1846                 .nid = LNET_NID_ANY,
1847                 .pid = LNET_PID_ANY
1848         };
1849         struct lnet_me *me;
1850         struct lnet_md md = { NULL };
1851         int rc;
1852
1853         if (set_eq)
1854                 the_lnet.ln_ping_target_handler =
1855                         lnet_ping_target_event_handler;
1856
1857         *ppbuf = lnet_ping_target_create(ni_count);
1858         if (*ppbuf == NULL) {
1859                 rc = -ENOMEM;
1860                 goto fail_free_eq;
1861         }
1862
1863         /* Ping target ME/MD */
1864         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1865                           LNET_PROTO_PING_MATCHBITS, 0,
1866                           LNET_UNLINK, LNET_INS_AFTER);
1867         if (IS_ERR(me)) {
1868                 rc = PTR_ERR(me);
1869                 CERROR("Can't create ping target ME: %d\n", rc);
1870                 goto fail_decref_ping_buffer;
1871         }
1872
1873         /* initialize md content */
1874         md.start     = &(*ppbuf)->pb_info;
1875         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1876         md.threshold = LNET_MD_THRESH_INF;
1877         md.max_size  = 0;
1878         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1879                        LNET_MD_MANAGE_REMOTE;
1880         md.handler   = the_lnet.ln_ping_target_handler;
1881         md.user_ptr  = *ppbuf;
1882
1883         rc = LNetMDAttach(me, &md, LNET_RETAIN, ping_mdh);
1884         if (rc != 0) {
1885                 CERROR("Can't attach ping target MD: %d\n", rc);
1886                 goto fail_decref_ping_buffer;
1887         }
1888         lnet_ping_buffer_addref(*ppbuf);
1889
1890         return 0;
1891
1892 fail_decref_ping_buffer:
1893         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1894         lnet_ping_buffer_decref(*ppbuf);
1895         *ppbuf = NULL;
1896 fail_free_eq:
1897         return rc;
1898 }
1899
1900 static void
1901 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1902                     struct lnet_handle_md *ping_mdh)
1903 {
1904         LNetMDUnlink(*ping_mdh);
1905         LNetInvalidateMDHandle(ping_mdh);
1906
1907         /* NB the MD could be busy; this just starts the unlink */
1908         wait_var_event_warning(&pbuf->pb_refcnt,
1909                                atomic_read(&pbuf->pb_refcnt) <= 1,
1910                                "Still waiting for ping data MD to unlink\n");
1911 }
1912
1913 static void
1914 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1915 {
1916         struct lnet_ni          *ni;
1917         struct lnet_net         *net;
1918         struct lnet_ni_status *ns;
1919         int                     i;
1920         int                     rc;
1921
1922         i = 0;
1923         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1924                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1925                         LASSERT(i < pbuf->pb_nnis);
1926
1927                         ns = &pbuf->pb_info.pi_ni[i];
1928
1929                         ns->ns_nid = ni->ni_nid;
1930
1931                         lnet_ni_lock(ni);
1932                         ns->ns_status = (ni->ni_status != NULL) ?
1933                                          ni->ni_status->ns_status :
1934                                                 LNET_NI_STATUS_UP;
1935                         ni->ni_status = ns;
1936                         lnet_ni_unlock(ni);
1937
1938                         i++;
1939                 }
1940         }
1941         /*
1942          * We (ab)use the ns_status of the loopback interface to
1943          * transmit the sequence number. The first interface listed
1944          * must be the loopback interface.
1945          */
1946         rc = lnet_ping_info_validate(&pbuf->pb_info);
1947         if (rc) {
1948                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1949                 LBUG();
1950         }
1951         LNET_PING_BUFFER_SEQNO(pbuf) =
1952                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1953 }
1954
1955 static void
1956 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1957                         struct lnet_handle_md ping_mdh)
1958 {
1959         struct lnet_ping_buffer *old_pbuf = NULL;
1960         struct lnet_handle_md old_ping_md;
1961
1962         /* switch the NIs to point to the new ping info created */
1963         lnet_net_lock(LNET_LOCK_EX);
1964
1965         if (!the_lnet.ln_routing)
1966                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1967         if (!lnet_peer_discovery_disabled)
1968                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1969
1970         /* Ensure only known feature bits have been set. */
1971         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1972         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1973
1974         lnet_ping_target_install_locked(pbuf);
1975
1976         if (the_lnet.ln_ping_target) {
1977                 old_pbuf = the_lnet.ln_ping_target;
1978                 old_ping_md = the_lnet.ln_ping_target_md;
1979         }
1980         the_lnet.ln_ping_target_md = ping_mdh;
1981         the_lnet.ln_ping_target = pbuf;
1982
1983         lnet_net_unlock(LNET_LOCK_EX);
1984
1985         if (old_pbuf) {
1986                 /* unlink and free the old ping info */
1987                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1988                 lnet_ping_buffer_decref(old_pbuf);
1989         }
1990
1991         lnet_push_update_to_peers(0);
1992 }
1993
1994 static void
1995 lnet_ping_target_fini(void)
1996 {
1997         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1998                             &the_lnet.ln_ping_target_md);
1999
2000         lnet_assert_handler_unused(the_lnet.ln_ping_target_handler);
2001         lnet_ping_target_destroy();
2002 }
2003
2004 /* Resize the push target. */
2005 int lnet_push_target_resize(void)
2006 {
2007         struct lnet_handle_md mdh;
2008         struct lnet_handle_md old_mdh;
2009         struct lnet_ping_buffer *pbuf;
2010         struct lnet_ping_buffer *old_pbuf;
2011         int nnis;
2012         int rc;
2013
2014 again:
2015         nnis = the_lnet.ln_push_target_nnis;
2016         if (nnis <= 0) {
2017                 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
2018                 return -EINVAL;
2019         }
2020
2021         /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
2022          * dropped when we need to resize again (see "old_pbuf" below) or when
2023          * LNet is shutdown (see lnet_push_target_fini())
2024          */
2025         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
2026         if (!pbuf) {
2027                 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
2028                 return -ENOMEM;
2029         }
2030
2031         rc = lnet_push_target_post(pbuf, &mdh);
2032         if (rc) {
2033                 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
2034                 lnet_ping_buffer_decref(pbuf);
2035                 return rc;
2036         }
2037
2038         lnet_net_lock(LNET_LOCK_EX);
2039         old_pbuf = the_lnet.ln_push_target;
2040         old_mdh = the_lnet.ln_push_target_md;
2041         the_lnet.ln_push_target = pbuf;
2042         the_lnet.ln_push_target_md = mdh;
2043         lnet_net_unlock(LNET_LOCK_EX);
2044
2045         if (old_pbuf) {
2046                 LNetMDUnlink(old_mdh);
2047                 /* Drop ref set by lnet_ping_buffer_alloc() */
2048                 lnet_ping_buffer_decref(old_pbuf);
2049         }
2050
2051         /* Received another push or reply that requires a larger buffer */
2052         if (nnis < the_lnet.ln_push_target_nnis)
2053                 goto again;
2054
2055         CDEBUG(D_NET, "nnis %d success\n", nnis);
2056         return 0;
2057 }
2058
2059 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
2060                           struct lnet_handle_md *mdhp)
2061 {
2062         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
2063         struct lnet_md md = { NULL };
2064         struct lnet_me *me;
2065         int rc;
2066
2067         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
2068                           LNET_PROTO_PING_MATCHBITS, 0,
2069                           LNET_UNLINK, LNET_INS_AFTER);
2070         if (IS_ERR(me)) {
2071                 rc = PTR_ERR(me);
2072                 CERROR("Can't create push target ME: %d\n", rc);
2073                 return rc;
2074         }
2075
2076         pbuf->pb_needs_post = false;
2077
2078         /* This reference is dropped by lnet_push_target_event_handler() */
2079         lnet_ping_buffer_addref(pbuf);
2080
2081         /* initialize md content */
2082         md.start     = &pbuf->pb_info;
2083         md.length    = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2084         md.threshold = 1;
2085         md.max_size  = 0;
2086         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
2087         md.user_ptr  = pbuf;
2088         md.handler   = the_lnet.ln_push_target_handler;
2089
2090         rc = LNetMDAttach(me, &md, LNET_UNLINK, mdhp);
2091         if (rc) {
2092                 CERROR("Can't attach push MD: %d\n", rc);
2093                 lnet_ping_buffer_decref(pbuf);
2094                 pbuf->pb_needs_post = true;
2095                 return rc;
2096         }
2097
2098         CDEBUG(D_NET, "posted push target %p\n", pbuf);
2099
2100         return 0;
2101 }
2102
2103 static void lnet_push_target_event_handler(struct lnet_event *ev)
2104 {
2105         struct lnet_ping_buffer *pbuf = ev->md_user_ptr;
2106
2107         CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
2108                ev->unlinked);
2109
2110         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2111                 lnet_swap_pinginfo(pbuf);
2112
2113         if (ev->type == LNET_EVENT_UNLINK) {
2114                 /* Drop ref added by lnet_push_target_post() */
2115                 lnet_ping_buffer_decref(pbuf);
2116                 return;
2117         }
2118
2119         lnet_peer_push_event(ev);
2120         if (ev->unlinked)
2121                 /* Drop ref added by lnet_push_target_post */
2122                 lnet_ping_buffer_decref(pbuf);
2123 }
2124
2125 /* Initialize the push target. */
2126 static int lnet_push_target_init(void)
2127 {
2128         int rc;
2129
2130         if (the_lnet.ln_push_target)
2131                 return -EALREADY;
2132
2133         the_lnet.ln_push_target_handler =
2134                 lnet_push_target_event_handler;
2135
2136         rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2137         LASSERT(rc == 0);
2138
2139         /* Start at the required minimum, we'll enlarge if required. */
2140         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2141
2142         rc = lnet_push_target_resize();
2143
2144         if (rc) {
2145                 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2146                 the_lnet.ln_push_target_handler = NULL;
2147         }
2148
2149         return rc;
2150 }
2151
2152 /* Clean up the push target. */
2153 static void lnet_push_target_fini(void)
2154 {
2155         if (!the_lnet.ln_push_target)
2156                 return;
2157
2158         /* Unlink and invalidate to prevent new references. */
2159         LNetMDUnlink(the_lnet.ln_push_target_md);
2160         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2161
2162         /* Wait for the unlink to complete. */
2163         wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2164                                atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2165                                "Still waiting for ping data MD to unlink\n");
2166
2167         /* Drop ref set by lnet_ping_buffer_alloc() */
2168         lnet_ping_buffer_decref(the_lnet.ln_push_target);
2169         the_lnet.ln_push_target = NULL;
2170         the_lnet.ln_push_target_nnis = 0;
2171
2172         LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2173         lnet_assert_handler_unused(the_lnet.ln_push_target_handler);
2174         the_lnet.ln_push_target_handler = NULL;
2175 }
2176
2177 static int
2178 lnet_ni_tq_credits(struct lnet_ni *ni)
2179 {
2180         int     credits;
2181
2182         LASSERT(ni->ni_ncpts >= 1);
2183
2184         if (ni->ni_ncpts == 1)
2185                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2186
2187         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2188         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2189         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2190
2191         return credits;
2192 }
2193
2194 static void
2195 lnet_ni_unlink_locked(struct lnet_ni *ni)
2196 {
2197         /* move it to zombie list and nobody can find it anymore */
2198         LASSERT(!list_empty(&ni->ni_netlist));
2199         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2200         lnet_ni_decref_locked(ni, 0);
2201 }
2202
2203 static void
2204 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2205 {
2206         int             i;
2207         int             islo;
2208         struct lnet_ni  *ni;
2209         struct list_head *zombie_list = &net->net_ni_zombie;
2210
2211         /*
2212          * Now wait for the NIs I just nuked to show up on the zombie
2213          * list and shut them down in guaranteed thread context
2214          */
2215         i = 2;
2216         while (!list_empty(zombie_list)) {
2217                 int     *ref;
2218                 int     j;
2219
2220                 ni = list_entry(zombie_list->next,
2221                                 struct lnet_ni, ni_netlist);
2222                 list_del_init(&ni->ni_netlist);
2223                 /* the ni should be in deleting state. If it's not it's
2224                  * a bug */
2225                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2226                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2227                         if (*ref == 0)
2228                                 continue;
2229                         /* still busy, add it back to zombie list */
2230                         list_add(&ni->ni_netlist, zombie_list);
2231                         break;
2232                 }
2233
2234                 if (!list_empty(&ni->ni_netlist)) {
2235                         /* Unlock mutex while waiting to allow other
2236                          * threads to read the LNet state and fall through
2237                          * to avoid deadlock
2238                          */
2239                         lnet_net_unlock(LNET_LOCK_EX);
2240                         mutex_unlock(&the_lnet.ln_api_mutex);
2241
2242                         ++i;
2243                         if ((i & (-i)) == i) {
2244                                 CDEBUG(D_WARNING,
2245                                        "Waiting for zombie LNI %s\n",
2246                                        libcfs_nid2str(ni->ni_nid));
2247                         }
2248                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2249
2250                         mutex_lock(&the_lnet.ln_api_mutex);
2251                         lnet_net_lock(LNET_LOCK_EX);
2252                         continue;
2253                 }
2254
2255                 lnet_net_unlock(LNET_LOCK_EX);
2256
2257                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2258
2259                 LASSERT(!in_interrupt());
2260                 /* Holding the mutex makes it safe for lnd_shutdown
2261                  * to call module_put(). Module unload cannot finish
2262                  * until lnet_unregister_lnd() completes, and that
2263                  * requires the mutex.
2264                  */
2265                 mutex_lock(&the_lnet.ln_lnd_mutex);
2266                 (net->net_lnd->lnd_shutdown)(ni);
2267                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2268
2269                 if (!islo)
2270                         CDEBUG(D_LNI, "Removed LNI %s\n",
2271                               libcfs_nid2str(ni->ni_nid));
2272
2273                 lnet_ni_free(ni);
2274                 i = 2;
2275                 lnet_net_lock(LNET_LOCK_EX);
2276         }
2277 }
2278
2279 /* shutdown down the NI and release refcount */
2280 static void
2281 lnet_shutdown_lndni(struct lnet_ni *ni)
2282 {
2283         int i;
2284         struct lnet_net *net = ni->ni_net;
2285
2286         lnet_net_lock(LNET_LOCK_EX);
2287         lnet_ni_lock(ni);
2288         ni->ni_state = LNET_NI_STATE_DELETING;
2289         lnet_ni_unlock(ni);
2290         lnet_ni_unlink_locked(ni);
2291         lnet_incr_dlc_seq();
2292         lnet_net_unlock(LNET_LOCK_EX);
2293
2294         /* clear messages for this NI on the lazy portal */
2295         for (i = 0; i < the_lnet.ln_nportals; i++)
2296                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2297
2298         lnet_net_lock(LNET_LOCK_EX);
2299         lnet_clear_zombies_nis_locked(net);
2300         lnet_net_unlock(LNET_LOCK_EX);
2301 }
2302
2303 static void
2304 lnet_shutdown_lndnet(struct lnet_net *net)
2305 {
2306         struct lnet_ni *ni;
2307
2308         lnet_net_lock(LNET_LOCK_EX);
2309
2310         list_del_init(&net->net_list);
2311
2312         while (!list_empty(&net->net_ni_list)) {
2313                 ni = list_entry(net->net_ni_list.next,
2314                                 struct lnet_ni, ni_netlist);
2315                 lnet_net_unlock(LNET_LOCK_EX);
2316                 lnet_shutdown_lndni(ni);
2317                 lnet_net_lock(LNET_LOCK_EX);
2318         }
2319
2320         lnet_net_unlock(LNET_LOCK_EX);
2321
2322         /* Do peer table cleanup for this net */
2323         lnet_peer_tables_cleanup(net);
2324
2325         lnet_net_free(net);
2326 }
2327
2328 static void
2329 lnet_shutdown_lndnets(void)
2330 {
2331         struct lnet_net *net;
2332         LIST_HEAD(resend);
2333         struct lnet_msg *msg, *tmp;
2334
2335         /* NB called holding the global mutex */
2336
2337         /* All quiet on the API front */
2338         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2339         LASSERT(the_lnet.ln_refcount == 0);
2340
2341         lnet_net_lock(LNET_LOCK_EX);
2342         the_lnet.ln_state = LNET_STATE_STOPPING;
2343
2344         /*
2345          * move the nets to the zombie list to avoid them being
2346          * picked up for new work. LONET is also included in the
2347          * Nets that will be moved to the zombie list
2348          */
2349         list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2350
2351         /* Drop the cached loopback Net. */
2352         if (the_lnet.ln_loni != NULL) {
2353                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2354                 the_lnet.ln_loni = NULL;
2355         }
2356         lnet_net_unlock(LNET_LOCK_EX);
2357
2358         /* iterate through the net zombie list and delete each net */
2359         while (!list_empty(&the_lnet.ln_net_zombie)) {
2360                 net = list_entry(the_lnet.ln_net_zombie.next,
2361                                  struct lnet_net, net_list);
2362                 lnet_shutdown_lndnet(net);
2363         }
2364
2365         spin_lock(&the_lnet.ln_msg_resend_lock);
2366         list_splice(&the_lnet.ln_msg_resend, &resend);
2367         spin_unlock(&the_lnet.ln_msg_resend_lock);
2368
2369         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2370                 list_del_init(&msg->msg_list);
2371                 msg->msg_no_resend = true;
2372                 lnet_finalize(msg, -ECANCELED);
2373         }
2374
2375         lnet_net_lock(LNET_LOCK_EX);
2376         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2377         lnet_net_unlock(LNET_LOCK_EX);
2378 }
2379
2380 static int
2381 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2382 {
2383         int                     rc = -EINVAL;
2384         struct lnet_tx_queue    *tq;
2385         int                     i;
2386         struct lnet_net         *net = ni->ni_net;
2387
2388         mutex_lock(&the_lnet.ln_lnd_mutex);
2389
2390         if (tun) {
2391                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2392                 ni->ni_lnd_tunables_set = true;
2393         }
2394
2395         rc = (net->net_lnd->lnd_startup)(ni);
2396
2397         mutex_unlock(&the_lnet.ln_lnd_mutex);
2398
2399         if (rc != 0) {
2400                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2401                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2402                 goto failed0;
2403         }
2404
2405         lnet_ni_lock(ni);
2406         ni->ni_state = LNET_NI_STATE_ACTIVE;
2407         lnet_ni_unlock(ni);
2408
2409         /* We keep a reference on the loopback net through the loopback NI */
2410         if (net->net_lnd->lnd_type == LOLND) {
2411                 lnet_ni_addref(ni);
2412                 LASSERT(the_lnet.ln_loni == NULL);
2413                 the_lnet.ln_loni = ni;
2414                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2415                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2416                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2417                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2418                 return 0;
2419         }
2420
2421         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2422             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2423                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2424                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2425                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2426                                         "" : "per-peer ");
2427                 /* shutdown the NI since if we get here then it must've already
2428                  * been started
2429                  */
2430                 lnet_shutdown_lndni(ni);
2431                 return -EINVAL;
2432         }
2433
2434         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2435                 tq->tq_credits_min =
2436                 tq->tq_credits_max =
2437                 tq->tq_credits = lnet_ni_tq_credits(ni);
2438         }
2439
2440         atomic_set(&ni->ni_tx_credits,
2441                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2442         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2443
2444         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2445                 libcfs_nid2str(ni->ni_nid),
2446                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2447                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2448                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2449                 ni->ni_net->net_tunables.lct_peer_timeout);
2450
2451         return 0;
2452 failed0:
2453         lnet_ni_free(ni);
2454         return rc;
2455 }
2456
2457 static int
2458 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2459 {
2460         struct lnet_ni *ni;
2461         struct lnet_net *net_l = NULL;
2462         LIST_HEAD(local_ni_list);
2463         int rc;
2464         int ni_count = 0;
2465         __u32 lnd_type;
2466         const struct lnet_lnd  *lnd;
2467         int peer_timeout =
2468                 net->net_tunables.lct_peer_timeout;
2469         int maxtxcredits =
2470                 net->net_tunables.lct_max_tx_credits;
2471         int peerrtrcredits =
2472                 net->net_tunables.lct_peer_rtr_credits;
2473
2474         /*
2475          * make sure that this net is unique. If it isn't then
2476          * we are adding interfaces to an already existing network, and
2477          * 'net' is just a convenient way to pass in the list.
2478          * if it is unique we need to find the LND and load it if
2479          * necessary.
2480          */
2481         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2482                 lnd_type = LNET_NETTYP(net->net_id);
2483
2484                 mutex_lock(&the_lnet.ln_lnd_mutex);
2485                 lnd = lnet_find_lnd_by_type(lnd_type);
2486
2487                 if (lnd == NULL) {
2488                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2489                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2490                         mutex_lock(&the_lnet.ln_lnd_mutex);
2491
2492                         lnd = lnet_find_lnd_by_type(lnd_type);
2493                         if (lnd == NULL) {
2494                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2495                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2496                                 libcfs_lnd2str(lnd_type),
2497                                 libcfs_lnd2modname(lnd_type), rc);
2498 #ifndef HAVE_MODULE_LOADING_SUPPORT
2499                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2500                                                 "compiled with kernel module "
2501                                                 "loading support.");
2502 #endif
2503                                 rc = -EINVAL;
2504                                 goto failed0;
2505                         }
2506                 }
2507
2508                 net->net_lnd = lnd;
2509
2510                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2511
2512                 net_l = net;
2513         }
2514
2515         /*
2516          * net_l: if the network being added is unique then net_l
2517          *        will point to that network
2518          *        if the network being added is not unique then
2519          *        net_l points to the existing network.
2520          *
2521          * When we enter the loop below, we'll pick NIs off he
2522          * network beign added and start them up, then add them to
2523          * a local ni list. Once we've successfully started all
2524          * the NIs then we join the local NI list (of started up
2525          * networks) with the net_l->net_ni_list, which should
2526          * point to the correct network to add the new ni list to
2527          *
2528          * If any of the new NIs fail to start up, then we want to
2529          * iterate through the local ni list, which should include
2530          * any NIs which were successfully started up, and shut
2531          * them down.
2532          *
2533          * After than we want to delete the network being added,
2534          * to avoid a memory leak.
2535          */
2536         while (!list_empty(&net->net_ni_added)) {
2537                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2538                                 ni_netlist);
2539                 list_del_init(&ni->ni_netlist);
2540
2541                 /* make sure that the the NI we're about to start
2542                  * up is actually unique. if it's not fail. */
2543                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2544                                         ni->ni_interface)) {
2545                         rc = -EEXIST;
2546                         goto failed1;
2547                 }
2548
2549                 /* adjust the pointer the parent network, just in case it
2550                  * the net is a duplicate */
2551                 ni->ni_net = net_l;
2552
2553                 rc = lnet_startup_lndni(ni, tun);
2554
2555                 if (rc < 0)
2556                         goto failed1;
2557
2558                 lnet_ni_addref(ni);
2559                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2560
2561                 ni_count++;
2562         }
2563
2564         lnet_net_lock(LNET_LOCK_EX);
2565         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2566         lnet_incr_dlc_seq();
2567         lnet_net_unlock(LNET_LOCK_EX);
2568
2569         /* if the network is not unique then we don't want to keep
2570          * it around after we're done. Free it. Otherwise add that
2571          * net to the global the_lnet.ln_nets */
2572         if (net_l != net && net_l != NULL) {
2573                 /*
2574                  * TODO - note. currently the tunables can not be updated
2575                  * once added
2576                  */
2577                 lnet_net_free(net);
2578         } else {
2579                 /*
2580                  * restore tunables after it has been overwitten by the
2581                  * lnd
2582                  */
2583                 if (peer_timeout != -1)
2584                         net->net_tunables.lct_peer_timeout = peer_timeout;
2585                 if (maxtxcredits != -1)
2586                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2587                 if (peerrtrcredits != -1)
2588                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2589
2590                 lnet_net_lock(LNET_LOCK_EX);
2591                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2592                 lnet_net_unlock(LNET_LOCK_EX);
2593         }
2594
2595         /* update net count */
2596         lnet_current_net_count = lnet_get_net_count();
2597
2598         return ni_count;
2599
2600 failed1:
2601         /*
2602          * shutdown the new NIs that are being started up
2603          * free the NET being started
2604          */
2605         while (!list_empty(&local_ni_list)) {
2606                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2607                                 ni_netlist);
2608
2609                 lnet_shutdown_lndni(ni);
2610         }
2611
2612 failed0:
2613         lnet_net_free(net);
2614
2615         return rc;
2616 }
2617
2618 static int
2619 lnet_startup_lndnets(struct list_head *netlist)
2620 {
2621         struct lnet_net         *net;
2622         int                     rc;
2623         int                     ni_count = 0;
2624
2625         /*
2626          * Change to running state before bringing up the LNDs. This
2627          * allows lnet_shutdown_lndnets() to assert that we've passed
2628          * through here.
2629          */
2630         lnet_net_lock(LNET_LOCK_EX);
2631         the_lnet.ln_state = LNET_STATE_RUNNING;
2632         lnet_net_unlock(LNET_LOCK_EX);
2633
2634         while (!list_empty(netlist)) {
2635                 net = list_entry(netlist->next, struct lnet_net, net_list);
2636                 list_del_init(&net->net_list);
2637
2638                 rc = lnet_startup_lndnet(net, NULL);
2639
2640                 if (rc < 0)
2641                         goto failed;
2642
2643                 ni_count += rc;
2644         }
2645
2646         return ni_count;
2647 failed:
2648         lnet_shutdown_lndnets();
2649
2650         return rc;
2651 }
2652
2653 /**
2654  * Initialize LNet library.
2655  *
2656  * Automatically called at module loading time. Caller has to call
2657  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2658  * latter returned 0. It must be called exactly once.
2659  *
2660  * \retval 0 on success
2661  * \retval -ve on failures.
2662  */
2663 int lnet_lib_init(void)
2664 {
2665         int rc;
2666
2667         lnet_assert_wire_constants();
2668
2669         /* refer to global cfs_cpt_table for now */
2670         the_lnet.ln_cpt_table = cfs_cpt_tab;
2671         the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2672
2673         LASSERT(the_lnet.ln_cpt_number > 0);
2674         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2675                 /* we are under risk of consuming all lh_cookie */
2676                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2677                        "please change setting of CPT-table and retry\n",
2678                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2679                 return -E2BIG;
2680         }
2681
2682         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2683                 the_lnet.ln_cpt_bits++;
2684
2685         rc = lnet_create_locks();
2686         if (rc != 0) {
2687                 CERROR("Can't create LNet global locks: %d\n", rc);
2688                 return rc;
2689         }
2690
2691         the_lnet.ln_refcount = 0;
2692         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2693         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2694
2695         /* The hash table size is the number of bits it takes to express the set
2696          * ln_num_routes, minus 1 (better to under estimate than over so we
2697          * don't waste memory). */
2698         if (rnet_htable_size <= 0)
2699                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2700         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2701                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2702         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2703                                            order_base_2(rnet_htable_size) - 1);
2704
2705         /* All LNDs apart from the LOLND are in separate modules.  They
2706          * register themselves when their module loads, and unregister
2707          * themselves when their module is unloaded. */
2708         lnet_register_lnd(&the_lolnd);
2709         return 0;
2710 }
2711
2712 /**
2713  * Finalize LNet library.
2714  *
2715  * \pre lnet_lib_init() called with success.
2716  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2717  *
2718  * As this happens at module-unload, all lnds must already be unloaded,
2719  * so they must already be unregistered.
2720  */
2721 void lnet_lib_exit(void)
2722 {
2723         int i;
2724
2725         LASSERT(the_lnet.ln_refcount == 0);
2726         lnet_unregister_lnd(&the_lolnd);
2727         for (i = 0; i < NUM_LNDS; i++)
2728                 LASSERT(!the_lnet.ln_lnds[i]);
2729         lnet_destroy_locks();
2730 }
2731
2732 /**
2733  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2734  *
2735  * Users must call this function at least once before any other functions.
2736  * For each successful call there must be a corresponding call to
2737  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2738  * ignored.
2739  *
2740  * The PID used by LNet may be different from the one requested.
2741  * See LNetGetId().
2742  *
2743  * \param requested_pid PID requested by the caller.
2744  *
2745  * \return >= 0 on success, and < 0 error code on failures.
2746  */
2747 int
2748 LNetNIInit(lnet_pid_t requested_pid)
2749 {
2750         int                     im_a_router = 0;
2751         int                     rc;
2752         int                     ni_count;
2753         struct lnet_ping_buffer *pbuf;
2754         struct lnet_handle_md   ping_mdh;
2755         LIST_HEAD(net_head);
2756         struct lnet_net         *net;
2757
2758         mutex_lock(&the_lnet.ln_api_mutex);
2759
2760         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2761
2762         if (the_lnet.ln_refcount > 0) {
2763                 rc = the_lnet.ln_refcount++;
2764                 mutex_unlock(&the_lnet.ln_api_mutex);
2765                 return rc;
2766         }
2767
2768         rc = lnet_prepare(requested_pid);
2769         if (rc != 0) {
2770                 mutex_unlock(&the_lnet.ln_api_mutex);
2771                 return rc;
2772         }
2773
2774         /* create a network for Loopback network */
2775         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2776         if (net == NULL) {
2777                 rc = -ENOMEM;
2778                 goto err_empty_list;
2779         }
2780
2781         /* Add in the loopback NI */
2782         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2783                 rc = -ENOMEM;
2784                 goto err_empty_list;
2785         }
2786
2787         if (use_tcp_bonding)
2788                 CWARN("use_tcp_bonding has been removed. Use Multi-Rail and Dynamic Discovery instead, see LU-13641\n");
2789
2790         /* If LNet is being initialized via DLC it is possible
2791          * that the user requests not to load module parameters (ones which
2792          * are supported by DLC) on initialization.  Therefore, make sure not
2793          * to load networks, routes and forwarding from module parameters
2794          * in this case.  On cleanup in case of failure only clean up
2795          * routes if it has been loaded */
2796         if (!the_lnet.ln_nis_from_mod_params) {
2797                 rc = lnet_parse_networks(&net_head, lnet_get_networks());
2798                 if (rc < 0)
2799                         goto err_empty_list;
2800         }
2801
2802         ni_count = lnet_startup_lndnets(&net_head);
2803         if (ni_count < 0) {
2804                 rc = ni_count;
2805                 goto err_empty_list;
2806         }
2807
2808         if (!the_lnet.ln_nis_from_mod_params) {
2809                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2810                 if (rc != 0)
2811                         goto err_shutdown_lndnis;
2812
2813                 rc = lnet_rtrpools_alloc(im_a_router);
2814                 if (rc != 0)
2815                         goto err_destroy_routes;
2816         }
2817
2818         rc = lnet_acceptor_start();
2819         if (rc != 0)
2820                 goto err_destroy_routes;
2821
2822         the_lnet.ln_refcount = 1;
2823         /* Now I may use my own API functions... */
2824
2825         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2826         if (rc != 0)
2827                 goto err_acceptor_stop;
2828
2829         lnet_ping_target_update(pbuf, ping_mdh);
2830
2831         the_lnet.ln_mt_handler = lnet_mt_event_handler;
2832
2833         rc = lnet_push_target_init();
2834         if (rc != 0)
2835                 goto err_stop_ping;
2836
2837         rc = lnet_peer_discovery_start();
2838         if (rc != 0)
2839                 goto err_destroy_push_target;
2840
2841         rc = lnet_monitor_thr_start();
2842         if (rc != 0)
2843                 goto err_stop_discovery_thr;
2844
2845         lnet_fault_init();
2846         lnet_router_debugfs_init();
2847
2848         mutex_unlock(&the_lnet.ln_api_mutex);
2849
2850         complete_all(&the_lnet.ln_started);
2851
2852         /* wait for all routers to start */
2853         lnet_wait_router_start();
2854
2855         return 0;
2856
2857 err_stop_discovery_thr:
2858         lnet_peer_discovery_stop();
2859 err_destroy_push_target:
2860         lnet_push_target_fini();
2861 err_stop_ping:
2862         lnet_ping_target_fini();
2863 err_acceptor_stop:
2864         the_lnet.ln_refcount = 0;
2865         lnet_acceptor_stop();
2866 err_destroy_routes:
2867         if (!the_lnet.ln_nis_from_mod_params)
2868                 lnet_destroy_routes();
2869 err_shutdown_lndnis:
2870         lnet_shutdown_lndnets();
2871 err_empty_list:
2872         lnet_unprepare();
2873         LASSERT(rc < 0);
2874         mutex_unlock(&the_lnet.ln_api_mutex);
2875         while (!list_empty(&net_head)) {
2876                 struct lnet_net *net;
2877
2878                 net = list_entry(net_head.next, struct lnet_net, net_list);
2879                 list_del_init(&net->net_list);
2880                 lnet_net_free(net);
2881         }
2882         return rc;
2883 }
2884 EXPORT_SYMBOL(LNetNIInit);
2885
2886 /**
2887  * Stop LNet interfaces, routing, and forwarding.
2888  *
2889  * Users must call this function once for each successful call to LNetNIInit().
2890  * Once the LNetNIFini() operation has been started, the results of pending
2891  * API operations are undefined.
2892  *
2893  * \return always 0 for current implementation.
2894  */
2895 int
2896 LNetNIFini(void)
2897 {
2898         mutex_lock(&the_lnet.ln_api_mutex);
2899
2900         LASSERT(the_lnet.ln_refcount > 0);
2901
2902         if (the_lnet.ln_refcount != 1) {
2903                 the_lnet.ln_refcount--;
2904         } else {
2905                 LASSERT(!the_lnet.ln_niinit_self);
2906
2907                 lnet_fault_fini();
2908
2909                 lnet_router_debugfs_fini();
2910                 lnet_monitor_thr_stop();
2911                 lnet_peer_discovery_stop();
2912                 lnet_push_target_fini();
2913                 lnet_ping_target_fini();
2914
2915                 /* Teardown fns that use my own API functions BEFORE here */
2916                 the_lnet.ln_refcount = 0;
2917
2918                 lnet_acceptor_stop();
2919                 lnet_destroy_routes();
2920                 lnet_shutdown_lndnets();
2921                 lnet_unprepare();
2922         }
2923
2924         mutex_unlock(&the_lnet.ln_api_mutex);
2925         return 0;
2926 }
2927 EXPORT_SYMBOL(LNetNIFini);
2928
2929 /**
2930  * Grabs the ni data from the ni structure and fills the out
2931  * parameters
2932  *
2933  * \param[in] ni network        interface structure
2934  * \param[out] cfg_ni           NI config information
2935  * \param[out] tun              network and LND tunables
2936  */
2937 static void
2938 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2939                    struct lnet_ioctl_config_lnd_tunables *tun,
2940                    struct lnet_ioctl_element_stats *stats,
2941                    __u32 tun_size)
2942 {
2943         size_t min_size = 0;
2944         int i;
2945
2946         if (!ni || !cfg_ni || !tun)
2947                 return;
2948
2949         if (ni->ni_interface != NULL) {
2950                 strncpy(cfg_ni->lic_ni_intf,
2951                         ni->ni_interface,
2952                         sizeof(cfg_ni->lic_ni_intf));
2953         }
2954
2955         cfg_ni->lic_nid = ni->ni_nid;
2956         if (ni->ni_nid == LNET_NID_LO_0)
2957                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2958         else
2959                 cfg_ni->lic_status = ni->ni_status->ns_status;
2960         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2961
2962         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2963
2964         if (stats) {
2965                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2966                                                        LNET_STATS_TYPE_SEND);
2967                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2968                                                        LNET_STATS_TYPE_RECV);
2969                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2970                                                        LNET_STATS_TYPE_DROP);
2971         }
2972
2973         /*
2974          * tun->lt_tun will always be present, but in order to be
2975          * backwards compatible, we need to deal with the cases when
2976          * tun->lt_tun is smaller than what the kernel has, because it
2977          * comes from an older version of a userspace program, then we'll
2978          * need to copy as much information as we have available space.
2979          */
2980         min_size = tun_size - sizeof(tun->lt_cmn);
2981         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2982
2983         /* copy over the cpts */
2984         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2985             ni->ni_cpts == NULL)  {
2986                 for (i = 0; i < ni->ni_ncpts; i++)
2987                         cfg_ni->lic_cpts[i] = i;
2988         } else {
2989                 for (i = 0;
2990                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2991                      i < LNET_MAX_SHOW_NUM_CPT;
2992                      i++)
2993                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2994         }
2995         cfg_ni->lic_ncpts = ni->ni_ncpts;
2996 }
2997
2998 /**
2999  * NOTE: This is a legacy function left in the code to be backwards
3000  * compatible with older userspace programs. It should eventually be
3001  * removed.
3002  *
3003  * Grabs the ni data from the ni structure and fills the out
3004  * parameters
3005  *
3006  * \param[in] ni network        interface structure
3007  * \param[out] config           config information
3008  */
3009 static void
3010 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
3011                          struct lnet_ioctl_config_data *config)
3012 {
3013         struct lnet_ioctl_net_config *net_config;
3014         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
3015         size_t min_size, tunable_size = 0;
3016         int i;
3017
3018         if (!ni || !config)
3019                 return;
3020
3021         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
3022         if (!net_config)
3023                 return;
3024
3025         if (!ni->ni_interface)
3026                 return;
3027
3028         strncpy(net_config->ni_interface,
3029                 ni->ni_interface,
3030                 sizeof(net_config->ni_interface));
3031
3032         config->cfg_nid = ni->ni_nid;
3033         config->cfg_config_u.cfg_net.net_peer_timeout =
3034                 ni->ni_net->net_tunables.lct_peer_timeout;
3035         config->cfg_config_u.cfg_net.net_max_tx_credits =
3036                 ni->ni_net->net_tunables.lct_max_tx_credits;
3037         config->cfg_config_u.cfg_net.net_peer_tx_credits =
3038                 ni->ni_net->net_tunables.lct_peer_tx_credits;
3039         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
3040                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
3041
3042         if (ni->ni_nid == LNET_NID_LO_0)
3043                 net_config->ni_status = LNET_NI_STATUS_UP;
3044         else
3045                 net_config->ni_status = ni->ni_status->ns_status;
3046
3047         if (ni->ni_cpts) {
3048                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
3049
3050                 for (i = 0; i < num_cpts; i++)
3051                         net_config->ni_cpts[i] = ni->ni_cpts[i];
3052
3053                 config->cfg_ncpts = num_cpts;
3054         }
3055
3056         /*
3057          * See if user land tools sent in a newer and larger version
3058          * of struct lnet_tunables than what the kernel uses.
3059          */
3060         min_size = sizeof(*config) + sizeof(*net_config);
3061
3062         if (config->cfg_hdr.ioc_len > min_size)
3063                 tunable_size = config->cfg_hdr.ioc_len - min_size;
3064
3065         /* Don't copy too much data to user space */
3066         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
3067         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
3068
3069         if (lnd_cfg && min_size) {
3070                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
3071                 config->cfg_config_u.cfg_net.net_interface_count = 1;
3072
3073                 /* Tell user land that kernel side has less data */
3074                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
3075                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
3076                         config->cfg_hdr.ioc_len -= min_size;
3077                 }
3078         }
3079 }
3080
3081 struct lnet_ni *
3082 lnet_get_ni_idx_locked(int idx)
3083 {
3084         struct lnet_ni          *ni;
3085         struct lnet_net         *net;
3086
3087         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3088                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3089                         if (idx-- == 0)
3090                                 return ni;
3091                 }
3092         }
3093
3094         return NULL;
3095 }
3096
3097 int lnet_get_net_healthv_locked(struct lnet_net *net)
3098 {
3099         struct lnet_ni *ni;
3100         int best_healthv = 0;
3101         int healthv;
3102
3103         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3104                 healthv = atomic_read(&ni->ni_healthv);
3105                 if (healthv > best_healthv)
3106                         best_healthv = healthv;
3107         }
3108
3109         return best_healthv;
3110 }
3111
3112 struct lnet_ni *
3113 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
3114 {
3115         struct lnet_ni          *ni;
3116         struct lnet_net         *net = mynet;
3117
3118         /*
3119          * It is possible that the net has been cleaned out while there is
3120          * a message being sent. This function accessed the net without
3121          * checking if the list is empty
3122          */
3123         if (prev == NULL) {
3124                 if (net == NULL)
3125                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3126                                         net_list);
3127                 if (list_empty(&net->net_ni_list))
3128                         return NULL;
3129                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3130                                 ni_netlist);
3131
3132                 return ni;
3133         }
3134
3135         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3136                 /* if you reached the end of the ni list and the net is
3137                  * specified, then there are no more nis in that net */
3138                 if (net != NULL)
3139                         return NULL;
3140
3141                 /* we reached the end of this net ni list. move to the
3142                  * next net */
3143                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3144                         /* no more nets and no more NIs. */
3145                         return NULL;
3146
3147                 /* get the next net */
3148                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3149                                  net_list);
3150                 if (list_empty(&net->net_ni_list))
3151                         return NULL;
3152                 /* get the ni on it */
3153                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3154                                 ni_netlist);
3155
3156                 return ni;
3157         }
3158
3159         if (list_empty(&prev->ni_netlist))
3160                 return NULL;
3161
3162         /* there are more nis left */
3163         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3164
3165         return ni;
3166 }
3167
3168 int
3169 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3170 {
3171         struct lnet_ni *ni;
3172         int cpt;
3173         int rc = -ENOENT;
3174         int idx = config->cfg_count;
3175
3176         cpt = lnet_net_lock_current();
3177
3178         ni = lnet_get_ni_idx_locked(idx);
3179
3180         if (ni != NULL) {
3181                 rc = 0;
3182                 lnet_ni_lock(ni);
3183                 lnet_fill_ni_info_legacy(ni, config);
3184                 lnet_ni_unlock(ni);
3185         }
3186
3187         lnet_net_unlock(cpt);
3188         return rc;
3189 }
3190
3191 int
3192 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3193                    struct lnet_ioctl_config_lnd_tunables *tun,
3194                    struct lnet_ioctl_element_stats *stats,
3195                    __u32 tun_size)
3196 {
3197         struct lnet_ni          *ni;
3198         int                     cpt;
3199         int                     rc = -ENOENT;
3200
3201         if (!cfg_ni || !tun || !stats)
3202                 return -EINVAL;
3203
3204         cpt = lnet_net_lock_current();
3205
3206         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3207
3208         if (ni) {
3209                 rc = 0;
3210                 lnet_ni_lock(ni);
3211                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3212                 lnet_ni_unlock(ni);
3213         }
3214
3215         lnet_net_unlock(cpt);
3216         return rc;
3217 }
3218
3219 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3220 {
3221         struct lnet_ni *ni;
3222         int cpt;
3223         int rc = -ENOENT;
3224
3225         if (!msg_stats)
3226                 return -EINVAL;
3227
3228         cpt = lnet_net_lock_current();
3229
3230         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3231
3232         if (ni) {
3233                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3234                 rc = 0;
3235         }
3236
3237         lnet_net_unlock(cpt);
3238
3239         return rc;
3240 }
3241
3242 static int lnet_add_net_common(struct lnet_net *net,
3243                                struct lnet_ioctl_config_lnd_tunables *tun)
3244 {
3245         struct lnet_handle_md ping_mdh;
3246         struct lnet_ping_buffer *pbuf;
3247         struct lnet_remotenet *rnet;
3248         struct lnet_ni *ni;
3249         int net_ni_count;
3250         __u32 net_id;
3251         int rc;
3252
3253         lnet_net_lock(LNET_LOCK_EX);
3254         rnet = lnet_find_rnet_locked(net->net_id);
3255         lnet_net_unlock(LNET_LOCK_EX);
3256         /*
3257          * make sure that the net added doesn't invalidate the current
3258          * configuration LNet is keeping
3259          */
3260         if (rnet) {
3261                 CERROR("Adding net %s will invalidate routing configuration\n",
3262                        libcfs_net2str(net->net_id));
3263                 lnet_net_free(net);
3264                 return -EUSERS;
3265         }
3266
3267         /*
3268          * make sure you calculate the correct number of slots in the ping
3269          * buffer. Since the ping info is a flattened list of all the NIs,
3270          * we should allocate enough slots to accomodate the number of NIs
3271          * which will be added.
3272          *
3273          * since ni hasn't been configured yet, use
3274          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3275          */
3276         net_ni_count = lnet_get_net_ni_count_pre(net);
3277
3278         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3279                                     net_ni_count + lnet_get_ni_count(),
3280                                     false);
3281         if (rc < 0) {
3282                 lnet_net_free(net);
3283                 return rc;
3284         }
3285
3286         if (tun)
3287                 memcpy(&net->net_tunables,
3288                        &tun->lt_cmn, sizeof(net->net_tunables));
3289         else
3290                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3291
3292         net_id = net->net_id;
3293
3294         rc = lnet_startup_lndnet(net,
3295                                  (tun) ? &tun->lt_tun : NULL);
3296         if (rc < 0)
3297                 goto failed;
3298
3299         lnet_net_lock(LNET_LOCK_EX);
3300         net = lnet_get_net_locked(net_id);
3301         LASSERT(net);
3302
3303         /* apply the UDSPs */
3304         rc = lnet_udsp_apply_policies_on_net(net);
3305         if (rc)
3306                 CERROR("Failed to apply UDSPs on local net %s\n",
3307                        libcfs_net2str(net->net_id));
3308
3309         /* At this point we lost track of which NI was just added, so we
3310          * just re-apply the policies on all of the NIs on this net
3311          */
3312         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3313                 rc = lnet_udsp_apply_policies_on_ni(ni);
3314                 if (rc)
3315                         CERROR("Failed to apply UDSPs on ni %s\n",
3316                                libcfs_nid2str(ni->ni_nid));
3317         }
3318         lnet_net_unlock(LNET_LOCK_EX);
3319
3320         /*
3321          * Start the acceptor thread if this is the first network
3322          * being added that requires the thread.
3323          */
3324         if (net->net_lnd->lnd_accept) {
3325                 rc = lnet_acceptor_start();
3326                 if (rc < 0) {
3327                         /* shutdown the net that we just started */
3328                         CERROR("Failed to start up acceptor thread\n");
3329                         lnet_shutdown_lndnet(net);
3330                         goto failed;
3331                 }
3332         }
3333
3334         lnet_net_lock(LNET_LOCK_EX);
3335         lnet_peer_net_added(net);
3336         lnet_net_unlock(LNET_LOCK_EX);
3337
3338         lnet_ping_target_update(pbuf, ping_mdh);
3339
3340         return 0;
3341
3342 failed:
3343         lnet_ping_md_unlink(pbuf, &ping_mdh);
3344         lnet_ping_buffer_decref(pbuf);
3345         return rc;
3346 }
3347
3348 static void
3349 lnet_set_tune_defaults(struct lnet_ioctl_config_lnd_tunables *tun)
3350 {
3351         if (tun) {
3352                 if (!tun->lt_cmn.lct_peer_timeout)
3353                         tun->lt_cmn.lct_peer_timeout = DEFAULT_PEER_TIMEOUT;
3354                 if (!tun->lt_cmn.lct_peer_tx_credits)
3355                         tun->lt_cmn.lct_peer_tx_credits = DEFAULT_PEER_CREDITS;
3356                 if (!tun->lt_cmn.lct_max_tx_credits)
3357                         tun->lt_cmn.lct_max_tx_credits = DEFAULT_CREDITS;
3358         }
3359 }
3360
3361 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3362                                       struct lnet_ioctl_config_lnd_tunables *tun)
3363 {
3364         struct lnet_net *net;
3365         const char *nets;
3366         int rc;
3367         LIST_HEAD(net_head);
3368
3369         rc = lnet_parse_ip2nets(&nets, ip2nets);
3370         if (rc < 0)
3371                 return rc;
3372
3373         rc = lnet_parse_networks(&net_head, nets);
3374         if (rc < 0)
3375                 return rc;
3376
3377         lnet_set_tune_defaults(tun);
3378
3379         mutex_lock(&the_lnet.ln_api_mutex);
3380         while (!list_empty(&net_head)) {
3381                 net = list_entry(net_head.next, struct lnet_net, net_list);
3382                 list_del_init(&net->net_list);
3383                 rc = lnet_add_net_common(net, tun);
3384                 if (rc < 0)
3385                         goto out;
3386         }
3387
3388 out:
3389         mutex_unlock(&the_lnet.ln_api_mutex);
3390
3391         while (!list_empty(&net_head)) {
3392                 net = list_entry(net_head.next, struct lnet_net, net_list);
3393                 list_del_init(&net->net_list);
3394                 lnet_net_free(net);
3395         }
3396         return rc;
3397 }
3398
3399 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3400 {
3401         struct lnet_net *net;
3402         struct lnet_ni *ni;
3403         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3404         int rc, i;
3405         __u32 net_id, lnd_type;
3406
3407         /* get the tunables if they are available */
3408         if (conf->lic_cfg_hdr.ioc_len >=
3409             sizeof(*conf) + sizeof(*tun))
3410                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3411                         conf->lic_bulk;
3412
3413         /* handle legacy ip2nets from DLC */
3414         if (conf->lic_legacy_ip2nets[0] != '\0')
3415                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3416                                                   tun);
3417
3418         net_id = LNET_NIDNET(conf->lic_nid);
3419         lnd_type = LNET_NETTYP(net_id);
3420
3421         if (!libcfs_isknown_lnd(lnd_type)) {
3422                 CERROR("No valid net and lnd information provided\n");
3423                 return -EINVAL;
3424         }
3425
3426         net = lnet_net_alloc(net_id, NULL);
3427         if (!net)
3428                 return -ENOMEM;
3429
3430         for (i = 0; i < conf->lic_ncpts; i++) {
3431                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3432                         return -EINVAL;
3433         }
3434
3435         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3436                                        conf->lic_ni_intf);
3437         if (!ni)
3438                 return -ENOMEM;
3439
3440         lnet_set_tune_defaults(tun);
3441
3442         mutex_lock(&the_lnet.ln_api_mutex);
3443
3444         rc = lnet_add_net_common(net, tun);
3445
3446         mutex_unlock(&the_lnet.ln_api_mutex);
3447
3448         return rc;
3449 }
3450
3451 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3452 {
3453         struct lnet_net  *net;
3454         struct lnet_ni *ni;
3455         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3456         struct lnet_ping_buffer *pbuf;
3457         struct lnet_handle_md  ping_mdh;
3458         int               rc;
3459         int               net_count;
3460         __u32             addr;
3461
3462         /* don't allow userspace to shutdown the LOLND */
3463         if (LNET_NETTYP(net_id) == LOLND)
3464                 return -EINVAL;
3465
3466         mutex_lock(&the_lnet.ln_api_mutex);
3467
3468         lnet_net_lock(0);
3469
3470         net = lnet_get_net_locked(net_id);
3471         if (!net) {
3472                 CERROR("net %s not found\n",
3473                        libcfs_net2str(net_id));
3474                 rc = -ENOENT;
3475                 goto unlock_net;
3476         }
3477
3478         addr = LNET_NIDADDR(conf->lic_nid);
3479         if (addr == 0) {
3480                 /* remove the entire net */
3481                 net_count = lnet_get_net_ni_count_locked(net);
3482
3483                 lnet_net_unlock(0);
3484
3485                 /* create and link a new ping info, before removing the old one */
3486                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3487                                         lnet_get_ni_count() - net_count,
3488                                         false);
3489                 if (rc != 0)
3490                         goto unlock_api_mutex;
3491
3492                 lnet_shutdown_lndnet(net);
3493
3494                 lnet_acceptor_stop();
3495
3496                 lnet_ping_target_update(pbuf, ping_mdh);
3497
3498                 goto unlock_api_mutex;
3499         }
3500
3501         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3502         if (!ni) {
3503                 CERROR("nid %s not found\n",
3504                        libcfs_nid2str(conf->lic_nid));
3505                 rc = -ENOENT;
3506                 goto unlock_net;
3507         }
3508
3509         net_count = lnet_get_net_ni_count_locked(net);
3510
3511         lnet_net_unlock(0);
3512
3513         /* create and link a new ping info, before removing the old one */
3514         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3515                                   lnet_get_ni_count() - 1, false);
3516         if (rc != 0)