Whamcloud - gitweb
LU-12930 various: use schedule_timeout_*interruptible
[fs/lustre-release.git] / lnet / lnet / api-ni.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  */
32
33 #define DEBUG_SUBSYSTEM S_LNET
34
35 #include <linux/ctype.h>
36 #include <linux/log2.h>
37 #include <linux/ktime.h>
38 #include <linux/moduleparam.h>
39 #include <linux/uaccess.h>
40
41 #include <lnet/lib-lnet.h>
42
43 #define D_LNI D_CONSOLE
44
45 /*
46  * initialize ln_api_mutex statically, since it needs to be used in
47  * discovery_set callback. That module parameter callback can be called
48  * before module init completes. The mutex needs to be ready for use then.
49  */
50 struct lnet the_lnet = {
51         .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
52 };              /* THE state of the network */
53 EXPORT_SYMBOL(the_lnet);
54
55 static char *ip2nets = "";
56 module_param(ip2nets, charp, 0444);
57 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
58
59 static char *networks = "";
60 module_param(networks, charp, 0444);
61 MODULE_PARM_DESC(networks, "local networks");
62
63 static char *routes = "";
64 module_param(routes, charp, 0444);
65 MODULE_PARM_DESC(routes, "routes to non-local networks");
66
67 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
68 module_param(rnet_htable_size, int, 0444);
69 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
70
71 static int use_tcp_bonding = false;
72 module_param(use_tcp_bonding, int, 0444);
73 MODULE_PARM_DESC(use_tcp_bonding,
74                  "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
75
76 unsigned int lnet_numa_range = 0;
77 module_param(lnet_numa_range, uint, 0444);
78 MODULE_PARM_DESC(lnet_numa_range,
79                 "NUMA range to consider during Multi-Rail selection");
80
81 /*
82  * lnet_health_sensitivity determines by how much we decrement the health
83  * value on sending error. The value defaults to 100, which means health
84  * interface health is decremented by 100 points every failure.
85  */
86 unsigned int lnet_health_sensitivity = 100;
87 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
88 #ifdef HAVE_KERNEL_PARAM_OPS
89 static struct kernel_param_ops param_ops_health_sensitivity = {
90         .set = sensitivity_set,
91         .get = param_get_int,
92 };
93 #define param_check_health_sensitivity(name, p) \
94                 __param_check(name, p, int)
95 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
96 #else
97 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
98                   &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
99 #endif
100 MODULE_PARM_DESC(lnet_health_sensitivity,
101                 "Value to decrement the health value by on error");
102
103 /*
104  * lnet_recovery_interval determines how often we should perform recovery
105  * on unhealthy interfaces.
106  */
107 unsigned int lnet_recovery_interval = 1;
108 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
109 #ifdef HAVE_KERNEL_PARAM_OPS
110 static struct kernel_param_ops param_ops_recovery_interval = {
111         .set = recovery_interval_set,
112         .get = param_get_int,
113 };
114 #define param_check_recovery_interval(name, p) \
115                 __param_check(name, p, int)
116 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
117 #else
118 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
119                   &lnet_recovery_interval, S_IRUGO|S_IWUSR);
120 #endif
121 MODULE_PARM_DESC(lnet_recovery_interval,
122                 "Interval to recover unhealthy interfaces in seconds");
123
124 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
125 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
126
127 static struct kernel_param_ops param_ops_interfaces_max = {
128         .set = intf_max_set,
129         .get = param_get_int,
130 };
131
132 #define param_check_interfaces_max(name, p) \
133                 __param_check(name, p, int)
134
135 #ifdef HAVE_KERNEL_PARAM_OPS
136 module_param(lnet_interfaces_max, interfaces_max, 0644);
137 #else
138 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
139                   &param_ops_interfaces_max, 0644);
140 #endif
141 MODULE_PARM_DESC(lnet_interfaces_max,
142                 "Maximum number of interfaces in a node.");
143
144 unsigned lnet_peer_discovery_disabled = 0;
145 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
146
147 static struct kernel_param_ops param_ops_discovery_disabled = {
148         .set = discovery_set,
149         .get = param_get_int,
150 };
151
152 #define param_check_discovery_disabled(name, p) \
153                 __param_check(name, p, int)
154 #ifdef HAVE_KERNEL_PARAM_OPS
155 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
156 #else
157 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
158                   &param_ops_discovery_disabled, 0644);
159 #endif
160 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
161                 "Set to 1 to disable peer discovery on this node.");
162
163 unsigned int lnet_drop_asym_route;
164 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
165
166 static struct kernel_param_ops param_ops_drop_asym_route = {
167         .set = drop_asym_route_set,
168         .get = param_get_int,
169 };
170
171 #define param_check_drop_asym_route(name, p)    \
172         __param_check(name, p, int)
173 #ifdef HAVE_KERNEL_PARAM_OPS
174 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
175 #else
176 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
177                   &param_ops_drop_asym_route, 0644);
178 #endif
179 MODULE_PARM_DESC(lnet_drop_asym_route,
180                  "Set to 1 to drop asymmetrical route messages.");
181
182 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
183 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 50
184
185 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
186 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
187 #ifdef HAVE_KERNEL_PARAM_OPS
188 static struct kernel_param_ops param_ops_transaction_timeout = {
189         .set = transaction_to_set,
190         .get = param_get_int,
191 };
192
193 #define param_check_transaction_timeout(name, p) \
194                 __param_check(name, p, int)
195 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
196 #else
197 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
198                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
199 #endif
200 MODULE_PARM_DESC(lnet_transaction_timeout,
201                 "Maximum number of seconds to wait for a peer response.");
202
203 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 2
204 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
205 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
206 #ifdef HAVE_KERNEL_PARAM_OPS
207 static struct kernel_param_ops param_ops_retry_count = {
208         .set = retry_count_set,
209         .get = param_get_int,
210 };
211
212 #define param_check_retry_count(name, p) \
213                 __param_check(name, p, int)
214 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
215 #else
216 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
217                   &lnet_retry_count, S_IRUGO|S_IWUSR);
218 #endif
219 MODULE_PARM_DESC(lnet_retry_count,
220                  "Maximum number of times to retry transmitting a message");
221
222
223 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
224 unsigned int lnet_current_net_count;
225
226 /*
227  * This sequence number keeps track of how many times DLC was used to
228  * update the local NIs. It is incremented when a NI is added or
229  * removed and checked when sending a message to determine if there is
230  * a need to re-run the selection algorithm. See lnet_select_pathway()
231  * for more details on its usage.
232  */
233 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
234
235 static int lnet_ping(struct lnet_process_id id, signed long timeout,
236                      struct lnet_process_id __user *ids, int n_ids);
237
238 static int lnet_discover(struct lnet_process_id id, __u32 force,
239                          struct lnet_process_id __user *ids, int n_ids);
240
241 static int
242 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
243 {
244         int rc;
245         unsigned *sensitivity = (unsigned *)kp->arg;
246         unsigned long value;
247
248         rc = kstrtoul(val, 0, &value);
249         if (rc) {
250                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
251                 return rc;
252         }
253
254         /*
255          * The purpose of locking the api_mutex here is to ensure that
256          * the correct value ends up stored properly.
257          */
258         mutex_lock(&the_lnet.ln_api_mutex);
259
260         if (value > LNET_MAX_HEALTH_VALUE) {
261                 mutex_unlock(&the_lnet.ln_api_mutex);
262                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
263                        LNET_MAX_HEALTH_VALUE, value);
264                 return -EINVAL;
265         }
266
267         /*
268          * if we're turning on health then use the health timeout
269          * defaults.
270          */
271         if (*sensitivity == 0 && value != 0) {
272                 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
273                 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
274         /*
275          * if we're turning off health then use the no health timeout
276          * default.
277          */
278         } else if (*sensitivity != 0 && value == 0) {
279                 lnet_transaction_timeout =
280                         LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
281                 lnet_retry_count = 0;
282         }
283
284         *sensitivity = value;
285
286         mutex_unlock(&the_lnet.ln_api_mutex);
287
288         return 0;
289 }
290
291 static int
292 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
293 {
294         int rc;
295         unsigned *interval = (unsigned *)kp->arg;
296         unsigned long value;
297
298         rc = kstrtoul(val, 0, &value);
299         if (rc) {
300                 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
301                 return rc;
302         }
303
304         if (value < 1) {
305                 CERROR("lnet_recovery_interval must be at least 1 second\n");
306                 return -EINVAL;
307         }
308
309         /*
310          * The purpose of locking the api_mutex here is to ensure that
311          * the correct value ends up stored properly.
312          */
313         mutex_lock(&the_lnet.ln_api_mutex);
314
315         *interval = value;
316
317         mutex_unlock(&the_lnet.ln_api_mutex);
318
319         return 0;
320 }
321
322 static int
323 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
324 {
325         int rc;
326         unsigned *discovery = (unsigned *)kp->arg;
327         unsigned long value;
328         struct lnet_ping_buffer *pbuf;
329
330         rc = kstrtoul(val, 0, &value);
331         if (rc) {
332                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
333                 return rc;
334         }
335
336         value = (value) ? 1 : 0;
337
338         /*
339          * The purpose of locking the api_mutex here is to ensure that
340          * the correct value ends up stored properly.
341          */
342         mutex_lock(&the_lnet.ln_api_mutex);
343
344         if (value == *discovery) {
345                 mutex_unlock(&the_lnet.ln_api_mutex);
346                 return 0;
347         }
348
349         *discovery = value;
350
351         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
352                 mutex_unlock(&the_lnet.ln_api_mutex);
353                 return 0;
354         }
355
356         /* tell peers that discovery setting has changed */
357         lnet_net_lock(LNET_LOCK_EX);
358         pbuf = the_lnet.ln_ping_target;
359         if (value)
360                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
361         else
362                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
363         lnet_net_unlock(LNET_LOCK_EX);
364
365         lnet_push_update_to_peers(1);
366
367         mutex_unlock(&the_lnet.ln_api_mutex);
368
369         return 0;
370 }
371
372 static int
373 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
374 {
375         int rc;
376         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
377         unsigned long value;
378
379         rc = kstrtoul(val, 0, &value);
380         if (rc) {
381                 CERROR("Invalid module parameter value for "
382                        "'lnet_drop_asym_route'\n");
383                 return rc;
384         }
385
386         /*
387          * The purpose of locking the api_mutex here is to ensure that
388          * the correct value ends up stored properly.
389          */
390         mutex_lock(&the_lnet.ln_api_mutex);
391
392         if (value == *drop_asym_route) {
393                 mutex_unlock(&the_lnet.ln_api_mutex);
394                 return 0;
395         }
396
397         *drop_asym_route = value;
398
399         mutex_unlock(&the_lnet.ln_api_mutex);
400
401         return 0;
402 }
403
404 static int
405 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
406 {
407         int rc;
408         unsigned *transaction_to = (unsigned *)kp->arg;
409         unsigned long value;
410
411         rc = kstrtoul(val, 0, &value);
412         if (rc) {
413                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
414                 return rc;
415         }
416
417         /*
418          * The purpose of locking the api_mutex here is to ensure that
419          * the correct value ends up stored properly.
420          */
421         mutex_lock(&the_lnet.ln_api_mutex);
422
423         if (value < lnet_retry_count || value == 0) {
424                 mutex_unlock(&the_lnet.ln_api_mutex);
425                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
426                        "Has to be greater than lnet_retry_count (%u)\n",
427                        value, lnet_retry_count);
428                 return -EINVAL;
429         }
430
431         if (value == *transaction_to) {
432                 mutex_unlock(&the_lnet.ln_api_mutex);
433                 return 0;
434         }
435
436         *transaction_to = value;
437         if (lnet_retry_count == 0)
438                 lnet_lnd_timeout = value;
439         else
440                 lnet_lnd_timeout = value / lnet_retry_count;
441
442         mutex_unlock(&the_lnet.ln_api_mutex);
443
444         return 0;
445 }
446
447 static int
448 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
449 {
450         int rc;
451         unsigned *retry_count = (unsigned *)kp->arg;
452         unsigned long value;
453
454         rc = kstrtoul(val, 0, &value);
455         if (rc) {
456                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
457                 return rc;
458         }
459
460         /*
461          * The purpose of locking the api_mutex here is to ensure that
462          * the correct value ends up stored properly.
463          */
464         mutex_lock(&the_lnet.ln_api_mutex);
465
466         if (lnet_health_sensitivity == 0) {
467                 mutex_unlock(&the_lnet.ln_api_mutex);
468                 CERROR("Can not set retry_count when health feature is turned off\n");
469                 return -EINVAL;
470         }
471
472         if (value > lnet_transaction_timeout) {
473                 mutex_unlock(&the_lnet.ln_api_mutex);
474                 CERROR("Invalid value for lnet_retry_count (%lu). "
475                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
476                        value, lnet_transaction_timeout);
477                 return -EINVAL;
478         }
479
480         *retry_count = value;
481
482         if (value == 0)
483                 lnet_lnd_timeout = lnet_transaction_timeout;
484         else
485                 lnet_lnd_timeout = lnet_transaction_timeout / value;
486
487         mutex_unlock(&the_lnet.ln_api_mutex);
488
489         return 0;
490 }
491
492 static int
493 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
494 {
495         int value, rc;
496
497         rc = kstrtoint(val, 0, &value);
498         if (rc) {
499                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
500                 return rc;
501         }
502
503         if (value < LNET_INTERFACES_MIN) {
504                 CWARN("max interfaces provided are too small, setting to %d\n",
505                       LNET_INTERFACES_MAX_DEFAULT);
506                 value = LNET_INTERFACES_MAX_DEFAULT;
507         }
508
509         *(int *)kp->arg = value;
510
511         return 0;
512 }
513
514 static char *
515 lnet_get_routes(void)
516 {
517         return routes;
518 }
519
520 static char *
521 lnet_get_networks(void)
522 {
523         char   *nets;
524         int     rc;
525
526         if (*networks != 0 && *ip2nets != 0) {
527                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
528                                    "'ip2nets' but not both at once\n");
529                 return NULL;
530         }
531
532         if (*ip2nets != 0) {
533                 rc = lnet_parse_ip2nets(&nets, ip2nets);
534                 return (rc == 0) ? nets : NULL;
535         }
536
537         if (*networks != 0)
538                 return networks;
539
540         return "tcp";
541 }
542
543 static void
544 lnet_init_locks(void)
545 {
546         spin_lock_init(&the_lnet.ln_eq_wait_lock);
547         spin_lock_init(&the_lnet.ln_msg_resend_lock);
548         init_waitqueue_head(&the_lnet.ln_eq_waitq);
549         init_completion(&the_lnet.ln_mt_wait_complete);
550         mutex_init(&the_lnet.ln_lnd_mutex);
551 }
552
553 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
554 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
555                                             *  MDs kmem_cache */
556 struct kmem_cache *lnet_rspt_cachep;       /* response tracker cache */
557 struct kmem_cache *lnet_msg_cachep;
558
559 static int
560 lnet_slab_setup(void)
561 {
562         /* create specific kmem_cache for MEs and small MDs (i.e., originally
563          * allocated in <size-xxx> kmem_cache).
564          */
565         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
566                                             0, 0, NULL);
567         if (!lnet_mes_cachep)
568                 return -ENOMEM;
569
570         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
571                                                   LNET_SMALL_MD_SIZE, 0, 0,
572                                                   NULL);
573         if (!lnet_small_mds_cachep)
574                 return -ENOMEM;
575
576         lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
577                                             0, 0, NULL);
578         if (!lnet_rspt_cachep)
579                 return -ENOMEM;
580
581         lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
582                                             0, 0, NULL);
583         if (!lnet_msg_cachep)
584                 return -ENOMEM;
585
586         return 0;
587 }
588
589 static void
590 lnet_slab_cleanup(void)
591 {
592         if (lnet_msg_cachep) {
593                 kmem_cache_destroy(lnet_msg_cachep);
594                 lnet_msg_cachep = NULL;
595         }
596
597
598         if (lnet_rspt_cachep) {
599                 kmem_cache_destroy(lnet_rspt_cachep);
600                 lnet_rspt_cachep = NULL;
601         }
602
603         if (lnet_small_mds_cachep) {
604                 kmem_cache_destroy(lnet_small_mds_cachep);
605                 lnet_small_mds_cachep = NULL;
606         }
607
608         if (lnet_mes_cachep) {
609                 kmem_cache_destroy(lnet_mes_cachep);
610                 lnet_mes_cachep = NULL;
611         }
612 }
613
614 static int
615 lnet_create_remote_nets_table(void)
616 {
617         int               i;
618         struct list_head *hash;
619
620         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
621         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
622         LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
623         if (hash == NULL) {
624                 CERROR("Failed to create remote nets hash table\n");
625                 return -ENOMEM;
626         }
627
628         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
629                 INIT_LIST_HEAD(&hash[i]);
630         the_lnet.ln_remote_nets_hash = hash;
631         return 0;
632 }
633
634 static void
635 lnet_destroy_remote_nets_table(void)
636 {
637         int i;
638
639         if (the_lnet.ln_remote_nets_hash == NULL)
640                 return;
641
642         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
643                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
644
645         LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
646                     LNET_REMOTE_NETS_HASH_SIZE *
647                     sizeof(the_lnet.ln_remote_nets_hash[0]));
648         the_lnet.ln_remote_nets_hash = NULL;
649 }
650
651 static void
652 lnet_destroy_locks(void)
653 {
654         if (the_lnet.ln_res_lock != NULL) {
655                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
656                 the_lnet.ln_res_lock = NULL;
657         }
658
659         if (the_lnet.ln_net_lock != NULL) {
660                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
661                 the_lnet.ln_net_lock = NULL;
662         }
663 }
664
665 static int
666 lnet_create_locks(void)
667 {
668         lnet_init_locks();
669
670         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
671         if (the_lnet.ln_res_lock == NULL)
672                 goto failed;
673
674         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
675         if (the_lnet.ln_net_lock == NULL)
676                 goto failed;
677
678         return 0;
679
680  failed:
681         lnet_destroy_locks();
682         return -ENOMEM;
683 }
684
685 static void lnet_assert_wire_constants(void)
686 {
687         /* Wire protocol assertions generated by 'wirecheck'
688          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
689          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
690          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
691
692         /* Constants... */
693         BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
694         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
695         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
696         BUILD_BUG_ON(LNET_MSG_ACK != 0);
697         BUILD_BUG_ON(LNET_MSG_PUT != 1);
698         BUILD_BUG_ON(LNET_MSG_GET != 2);
699         BUILD_BUG_ON(LNET_MSG_REPLY != 3);
700         BUILD_BUG_ON(LNET_MSG_HELLO != 4);
701
702         /* Checks for struct lnet_handle_wire */
703         BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
704         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
705                                    wh_interface_cookie) != 0);
706         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
707         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
708                                    wh_object_cookie) != 8);
709         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
710
711         /* Checks for struct struct lnet_magicversion */
712         BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
713         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
714         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
715         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
716         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
717         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
718                                    version_minor) != 6);
719         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
720
721         /* Checks for struct struct lnet_hdr */
722         BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
723         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
724         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
725         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
726         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
727         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
728         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
729         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
730         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
731         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
732         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
733         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
734         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
735         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
736         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
737
738         /* Ack */
739         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
740         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
741         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
742         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
743         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
744         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
745
746         /* Put */
747         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
748         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
749         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
750         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
751         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
752         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
753         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
754         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
755         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
756         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
757
758         /* Get */
759         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
760         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
761         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
762         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
763         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
764         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
765         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
766         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
767         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
768         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
769
770         /* Reply */
771         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
772         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
773
774         /* Hello */
775         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
776         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
777         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
778         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
779
780         /* Checks for struct lnet_ni_status and related constants */
781         BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
782         BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
783         BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
784
785         /* Checks for struct lnet_ni_status */
786         BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
787         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
788         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
789         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
790         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
791         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
792         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
793
794         /* Checks for struct lnet_ping_info and related constants */
795         BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
796         BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
797         BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
798         BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
799         BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
800         BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
801         BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
802         BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
803
804         /* Checks for struct lnet_ping_info */
805         BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
806         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
807         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
808         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
809         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
810         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
811         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
812         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
813         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
814         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
815         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
816 }
817
818 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
819 {
820         const struct lnet_lnd *lnd;
821
822         /* holding lnd mutex */
823         if (type >= NUM_LNDS)
824                 return NULL;
825         lnd = the_lnet.ln_lnds[type];
826         LASSERT(!lnd || lnd->lnd_type == type);
827
828         return lnd;
829 }
830
831 unsigned int
832 lnet_get_lnd_timeout(void)
833 {
834         return lnet_lnd_timeout;
835 }
836 EXPORT_SYMBOL(lnet_get_lnd_timeout);
837
838 void
839 lnet_register_lnd(const struct lnet_lnd *lnd)
840 {
841         mutex_lock(&the_lnet.ln_lnd_mutex);
842
843         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
844         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
845
846         the_lnet.ln_lnds[lnd->lnd_type] = lnd;
847
848         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
849
850         mutex_unlock(&the_lnet.ln_lnd_mutex);
851 }
852 EXPORT_SYMBOL(lnet_register_lnd);
853
854 void
855 lnet_unregister_lnd(const struct lnet_lnd *lnd)
856 {
857         mutex_lock(&the_lnet.ln_lnd_mutex);
858
859         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
860
861         the_lnet.ln_lnds[lnd->lnd_type] = NULL;
862         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
863
864         mutex_unlock(&the_lnet.ln_lnd_mutex);
865 }
866 EXPORT_SYMBOL(lnet_unregister_lnd);
867
868 void
869 lnet_counters_get_common(struct lnet_counters_common *common)
870 {
871         struct lnet_counters *ctr;
872         int i;
873
874         memset(common, 0, sizeof(*common));
875
876         lnet_net_lock(LNET_LOCK_EX);
877
878         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
879                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
880                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
881                 common->lcc_errors       += ctr->lct_common.lcc_errors;
882                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
883                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
884                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
885                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
886                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
887                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
888                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
889                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
890         }
891         lnet_net_unlock(LNET_LOCK_EX);
892 }
893 EXPORT_SYMBOL(lnet_counters_get_common);
894
895 void
896 lnet_counters_get(struct lnet_counters *counters)
897 {
898         struct lnet_counters *ctr;
899         struct lnet_counters_health *health = &counters->lct_health;
900         int             i;
901
902         memset(counters, 0, sizeof(*counters));
903
904         lnet_counters_get_common(&counters->lct_common);
905
906         lnet_net_lock(LNET_LOCK_EX);
907
908         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
909                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
910                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
911                 health->lch_response_timeout_count +=
912                                 ctr->lct_health.lch_response_timeout_count;
913                 health->lch_local_interrupt_count +=
914                                 ctr->lct_health.lch_local_interrupt_count;
915                 health->lch_local_dropped_count +=
916                                 ctr->lct_health.lch_local_dropped_count;
917                 health->lch_local_aborted_count +=
918                                 ctr->lct_health.lch_local_aborted_count;
919                 health->lch_local_no_route_count +=
920                                 ctr->lct_health.lch_local_no_route_count;
921                 health->lch_local_timeout_count +=
922                                 ctr->lct_health.lch_local_timeout_count;
923                 health->lch_local_error_count +=
924                                 ctr->lct_health.lch_local_error_count;
925                 health->lch_remote_dropped_count +=
926                                 ctr->lct_health.lch_remote_dropped_count;
927                 health->lch_remote_error_count +=
928                                 ctr->lct_health.lch_remote_error_count;
929                 health->lch_remote_timeout_count +=
930                                 ctr->lct_health.lch_remote_timeout_count;
931                 health->lch_network_timeout_count +=
932                                 ctr->lct_health.lch_network_timeout_count;
933         }
934         lnet_net_unlock(LNET_LOCK_EX);
935 }
936 EXPORT_SYMBOL(lnet_counters_get);
937
938 void
939 lnet_counters_reset(void)
940 {
941         struct lnet_counters *counters;
942         int             i;
943
944         lnet_net_lock(LNET_LOCK_EX);
945
946         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
947                 memset(counters, 0, sizeof(struct lnet_counters));
948
949         lnet_net_unlock(LNET_LOCK_EX);
950 }
951
952 static char *
953 lnet_res_type2str(int type)
954 {
955         switch (type) {
956         default:
957                 LBUG();
958         case LNET_COOKIE_TYPE_MD:
959                 return "MD";
960         case LNET_COOKIE_TYPE_ME:
961                 return "ME";
962         case LNET_COOKIE_TYPE_EQ:
963                 return "EQ";
964         }
965 }
966
967 static void
968 lnet_res_container_cleanup(struct lnet_res_container *rec)
969 {
970         int     count = 0;
971
972         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
973                 return;
974
975         while (!list_empty(&rec->rec_active)) {
976                 struct list_head *e = rec->rec_active.next;
977
978                 list_del_init(e);
979                 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
980                         lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
981
982                 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
983                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
984
985                 } else { /* NB: Active MEs should be attached on portals */
986                         LBUG();
987                 }
988                 count++;
989         }
990
991         if (count > 0) {
992                 /* Found alive MD/ME/EQ, user really should unlink/free
993                  * all of them before finalize LNet, but if someone didn't,
994                  * we have to recycle garbage for him */
995                 CERROR("%d active elements on exit of %s container\n",
996                        count, lnet_res_type2str(rec->rec_type));
997         }
998
999         if (rec->rec_lh_hash != NULL) {
1000                 LIBCFS_FREE(rec->rec_lh_hash,
1001                             LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1002                 rec->rec_lh_hash = NULL;
1003         }
1004
1005         rec->rec_type = 0; /* mark it as finalized */
1006 }
1007
1008 static int
1009 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1010 {
1011         int     rc = 0;
1012         int     i;
1013
1014         LASSERT(rec->rec_type == 0);
1015
1016         rec->rec_type = type;
1017         INIT_LIST_HEAD(&rec->rec_active);
1018
1019         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1020
1021         /* Arbitrary choice of hash table size */
1022         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1023                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1024         if (rec->rec_lh_hash == NULL) {
1025                 rc = -ENOMEM;
1026                 goto out;
1027         }
1028
1029         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1030                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1031
1032         return 0;
1033
1034 out:
1035         CERROR("Failed to setup %s resource container\n",
1036                lnet_res_type2str(type));
1037         lnet_res_container_cleanup(rec);
1038         return rc;
1039 }
1040
1041 static void
1042 lnet_res_containers_destroy(struct lnet_res_container **recs)
1043 {
1044         struct lnet_res_container       *rec;
1045         int                             i;
1046
1047         cfs_percpt_for_each(rec, i, recs)
1048                 lnet_res_container_cleanup(rec);
1049
1050         cfs_percpt_free(recs);
1051 }
1052
1053 static struct lnet_res_container **
1054 lnet_res_containers_create(int type)
1055 {
1056         struct lnet_res_container       **recs;
1057         struct lnet_res_container       *rec;
1058         int                             rc;
1059         int                             i;
1060
1061         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1062         if (recs == NULL) {
1063                 CERROR("Failed to allocate %s resource containers\n",
1064                        lnet_res_type2str(type));
1065                 return NULL;
1066         }
1067
1068         cfs_percpt_for_each(rec, i, recs) {
1069                 rc = lnet_res_container_setup(rec, i, type);
1070                 if (rc != 0) {
1071                         lnet_res_containers_destroy(recs);
1072                         return NULL;
1073                 }
1074         }
1075
1076         return recs;
1077 }
1078
1079 struct lnet_libhandle *
1080 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1081 {
1082         /* ALWAYS called with lnet_res_lock held */
1083         struct list_head        *head;
1084         struct lnet_libhandle   *lh;
1085         unsigned int            hash;
1086
1087         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1088                 return NULL;
1089
1090         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1091         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1092
1093         list_for_each_entry(lh, head, lh_hash_chain) {
1094                 if (lh->lh_cookie == cookie)
1095                         return lh;
1096         }
1097
1098         return NULL;
1099 }
1100
1101 void
1102 lnet_res_lh_initialize(struct lnet_res_container *rec,
1103                        struct lnet_libhandle *lh)
1104 {
1105         /* ALWAYS called with lnet_res_lock held */
1106         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1107         unsigned int    hash;
1108
1109         lh->lh_cookie = rec->rec_lh_cookie;
1110         rec->rec_lh_cookie += 1 << ibits;
1111
1112         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1113
1114         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1115 }
1116
1117 struct list_head **
1118 lnet_create_array_of_queues(void)
1119 {
1120         struct list_head **qs;
1121         struct list_head *q;
1122         int i;
1123
1124         qs = cfs_percpt_alloc(lnet_cpt_table(),
1125                               sizeof(struct list_head));
1126         if (!qs) {
1127                 CERROR("Failed to allocate queues\n");
1128                 return NULL;
1129         }
1130
1131         cfs_percpt_for_each(q, i, qs)
1132                 INIT_LIST_HEAD(q);
1133
1134         return qs;
1135 }
1136
1137 static int lnet_unprepare(void);
1138
1139 static int
1140 lnet_prepare(lnet_pid_t requested_pid)
1141 {
1142         /* Prepare to bring up the network */
1143         struct lnet_res_container **recs;
1144         int                       rc = 0;
1145
1146         if (requested_pid == LNET_PID_ANY) {
1147                 /* Don't instantiate LNET just for me */
1148                 return -ENETDOWN;
1149         }
1150
1151         LASSERT(the_lnet.ln_refcount == 0);
1152
1153         the_lnet.ln_routing = 0;
1154
1155         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1156         the_lnet.ln_pid = requested_pid;
1157
1158         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1159         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1160         INIT_LIST_HEAD(&the_lnet.ln_nets);
1161         INIT_LIST_HEAD(&the_lnet.ln_routers);
1162         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1163         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1164         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1165         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1166         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1167         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1168         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1169         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1170         LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1171         init_completion(&the_lnet.ln_started);
1172
1173         rc = lnet_slab_setup();
1174         if (rc != 0)
1175                 goto failed;
1176
1177         rc = lnet_create_remote_nets_table();
1178         if (rc != 0)
1179                 goto failed;
1180
1181         /*
1182          * NB the interface cookie in wire handles guards against delayed
1183          * replies and ACKs appearing valid after reboot.
1184          */
1185         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1186
1187         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1188                                                 sizeof(struct lnet_counters));
1189         if (the_lnet.ln_counters == NULL) {
1190                 CERROR("Failed to allocate counters for LNet\n");
1191                 rc = -ENOMEM;
1192                 goto failed;
1193         }
1194
1195         rc = lnet_peer_tables_create();
1196         if (rc != 0)
1197                 goto failed;
1198
1199         rc = lnet_msg_containers_create();
1200         if (rc != 0)
1201                 goto failed;
1202
1203         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1204                                       LNET_COOKIE_TYPE_EQ);
1205         if (rc != 0)
1206                 goto failed;
1207
1208         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1209         if (recs == NULL) {
1210                 rc = -ENOMEM;
1211                 goto failed;
1212         }
1213
1214         the_lnet.ln_md_containers = recs;
1215
1216         rc = lnet_portals_create();
1217         if (rc != 0) {
1218                 CERROR("Failed to create portals for LNet: %d\n", rc);
1219                 goto failed;
1220         }
1221
1222         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1223         if (!the_lnet.ln_mt_zombie_rstqs) {
1224                 rc = -ENOMEM;
1225                 goto failed;
1226         }
1227
1228         return 0;
1229
1230  failed:
1231         lnet_unprepare();
1232         return rc;
1233 }
1234
1235 static int
1236 lnet_unprepare (void)
1237 {
1238         int rc;
1239
1240         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1241          * have shut down already, so it is safe to unlink and free all
1242          * descriptors, even those that appear committed to a network op (eg MD
1243          * with non-zero pending count) */
1244
1245         lnet_fail_nid(LNET_NID_ANY, 0);
1246
1247         LASSERT(the_lnet.ln_refcount == 0);
1248         LASSERT(list_empty(&the_lnet.ln_test_peers));
1249         LASSERT(list_empty(&the_lnet.ln_nets));
1250
1251         if (the_lnet.ln_mt_zombie_rstqs) {
1252                 lnet_clean_zombie_rstqs();
1253                 the_lnet.ln_mt_zombie_rstqs = NULL;
1254         }
1255
1256         if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1257                 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1258                 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1259                 LASSERT(rc == 0);
1260         }
1261
1262         lnet_portals_destroy();
1263
1264         if (the_lnet.ln_md_containers != NULL) {
1265                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1266                 the_lnet.ln_md_containers = NULL;
1267         }
1268
1269         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1270
1271         lnet_msg_containers_destroy();
1272         lnet_peer_uninit();
1273         lnet_rtrpools_free(0);
1274
1275         if (the_lnet.ln_counters != NULL) {
1276                 cfs_percpt_free(the_lnet.ln_counters);
1277                 the_lnet.ln_counters = NULL;
1278         }
1279         lnet_destroy_remote_nets_table();
1280         lnet_slab_cleanup();
1281
1282         return 0;
1283 }
1284
1285 struct lnet_ni  *
1286 lnet_net2ni_locked(__u32 net_id, int cpt)
1287 {
1288         struct lnet_ni   *ni;
1289         struct lnet_net  *net;
1290
1291         LASSERT(cpt != LNET_LOCK_EX);
1292
1293         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1294                 if (net->net_id == net_id) {
1295                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1296                                         ni_netlist);
1297                         return ni;
1298                 }
1299         }
1300
1301         return NULL;
1302 }
1303
1304 struct lnet_ni *
1305 lnet_net2ni_addref(__u32 net)
1306 {
1307         struct lnet_ni *ni;
1308
1309         lnet_net_lock(0);
1310         ni = lnet_net2ni_locked(net, 0);
1311         if (ni)
1312                 lnet_ni_addref_locked(ni, 0);
1313         lnet_net_unlock(0);
1314
1315         return ni;
1316 }
1317 EXPORT_SYMBOL(lnet_net2ni_addref);
1318
1319 struct lnet_net *
1320 lnet_get_net_locked(__u32 net_id)
1321 {
1322         struct lnet_net  *net;
1323
1324         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1325                 if (net->net_id == net_id)
1326                         return net;
1327         }
1328
1329         return NULL;
1330 }
1331
1332 unsigned int
1333 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1334 {
1335         __u64           key = nid;
1336         unsigned int    val;
1337
1338         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1339
1340         if (number == 1)
1341                 return 0;
1342
1343         val = hash_long(key, LNET_CPT_BITS);
1344         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1345         if (val < number)
1346                 return val;
1347
1348         return (unsigned int)(key + val + (val >> 1)) % number;
1349 }
1350
1351 int
1352 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1353 {
1354         struct lnet_net *net;
1355
1356         /* must called with hold of lnet_net_lock */
1357         if (LNET_CPT_NUMBER == 1)
1358                 return 0; /* the only one */
1359
1360         /*
1361          * If NI is provided then use the CPT identified in the NI cpt
1362          * list if one exists. If one doesn't exist, then that NI is
1363          * associated with all CPTs and it follows that the net it belongs
1364          * to is implicitly associated with all CPTs, so just hash the nid
1365          * and return that.
1366          */
1367         if (ni != NULL) {
1368                 if (ni->ni_cpts != NULL)
1369                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1370                                                              ni->ni_ncpts)];
1371                 else
1372                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1373         }
1374
1375         /* no NI provided so look at the net */
1376         net = lnet_get_net_locked(LNET_NIDNET(nid));
1377
1378         if (net != NULL && net->net_cpts != NULL) {
1379                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1380         }
1381
1382         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1383 }
1384
1385 int
1386 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1387 {
1388         int     cpt;
1389         int     cpt2;
1390
1391         if (LNET_CPT_NUMBER == 1)
1392                 return 0; /* the only one */
1393
1394         cpt = lnet_net_lock_current();
1395
1396         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1397
1398         lnet_net_unlock(cpt);
1399
1400         return cpt2;
1401 }
1402 EXPORT_SYMBOL(lnet_cpt_of_nid);
1403
1404 int
1405 lnet_islocalnet_locked(__u32 net_id)
1406 {
1407         struct lnet_net *net;
1408         bool local;
1409
1410         net = lnet_get_net_locked(net_id);
1411
1412         local = net != NULL;
1413
1414         return local;
1415 }
1416
1417 int
1418 lnet_islocalnet(__u32 net_id)
1419 {
1420         int cpt;
1421         bool local;
1422
1423         cpt = lnet_net_lock_current();
1424
1425         local = lnet_islocalnet_locked(net_id);
1426
1427         lnet_net_unlock(cpt);
1428
1429         return local;
1430 }
1431
1432 struct lnet_ni  *
1433 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1434 {
1435         struct lnet_net  *net;
1436         struct lnet_ni   *ni;
1437
1438         LASSERT(cpt != LNET_LOCK_EX);
1439
1440         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1441                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1442                         if (ni->ni_nid == nid)
1443                                 return ni;
1444                 }
1445         }
1446
1447         return NULL;
1448 }
1449
1450 struct lnet_ni *
1451 lnet_nid2ni_addref(lnet_nid_t nid)
1452 {
1453         struct lnet_ni *ni;
1454
1455         lnet_net_lock(0);
1456         ni = lnet_nid2ni_locked(nid, 0);
1457         if (ni)
1458                 lnet_ni_addref_locked(ni, 0);
1459         lnet_net_unlock(0);
1460
1461         return ni;
1462 }
1463 EXPORT_SYMBOL(lnet_nid2ni_addref);
1464
1465 int
1466 lnet_islocalnid(lnet_nid_t nid)
1467 {
1468         struct lnet_ni  *ni;
1469         int             cpt;
1470
1471         cpt = lnet_net_lock_current();
1472         ni = lnet_nid2ni_locked(nid, cpt);
1473         lnet_net_unlock(cpt);
1474
1475         return ni != NULL;
1476 }
1477
1478 int
1479 lnet_count_acceptor_nets(void)
1480 {
1481         /* Return the # of NIs that need the acceptor. */
1482         int              count = 0;
1483         struct lnet_net  *net;
1484         int              cpt;
1485
1486         cpt = lnet_net_lock_current();
1487         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1488                 /* all socklnd type networks should have the acceptor
1489                  * thread started */
1490                 if (net->net_lnd->lnd_accept != NULL)
1491                         count++;
1492         }
1493
1494         lnet_net_unlock(cpt);
1495
1496         return count;
1497 }
1498
1499 struct lnet_ping_buffer *
1500 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1501 {
1502         struct lnet_ping_buffer *pbuf;
1503
1504         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1505         if (pbuf) {
1506                 pbuf->pb_nnis = nnis;
1507                 atomic_set(&pbuf->pb_refcnt, 1);
1508         }
1509
1510         return pbuf;
1511 }
1512
1513 void
1514 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1515 {
1516         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1517         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1518 }
1519
1520 static struct lnet_ping_buffer *
1521 lnet_ping_target_create(int nnis)
1522 {
1523         struct lnet_ping_buffer *pbuf;
1524
1525         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1526         if (pbuf == NULL) {
1527                 CERROR("Can't allocate ping source [%d]\n", nnis);
1528                 return NULL;
1529         }
1530
1531         pbuf->pb_info.pi_nnis = nnis;
1532         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1533         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1534         pbuf->pb_info.pi_features =
1535                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1536
1537         return pbuf;
1538 }
1539
1540 static inline int
1541 lnet_get_net_ni_count_locked(struct lnet_net *net)
1542 {
1543         struct lnet_ni  *ni;
1544         int             count = 0;
1545
1546         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1547                 count++;
1548
1549         return count;
1550 }
1551
1552 static inline int
1553 lnet_get_net_ni_count_pre(struct lnet_net *net)
1554 {
1555         struct lnet_ni  *ni;
1556         int             count = 0;
1557
1558         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1559                 count++;
1560
1561         return count;
1562 }
1563
1564 static inline int
1565 lnet_get_ni_count(void)
1566 {
1567         struct lnet_ni  *ni;
1568         struct lnet_net *net;
1569         int             count = 0;
1570
1571         lnet_net_lock(0);
1572
1573         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1574                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1575                         count++;
1576         }
1577
1578         lnet_net_unlock(0);
1579
1580         return count;
1581 }
1582
1583 int
1584 lnet_get_net_count(void)
1585 {
1586         struct lnet_net *net;
1587         int count = 0;
1588
1589         lnet_net_lock(0);
1590
1591         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1592                 count++;
1593         }
1594
1595         lnet_net_unlock(0);
1596
1597         return count;
1598 }
1599
1600 void
1601 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1602 {
1603         struct lnet_ni_status *stat;
1604         int nnis;
1605         int i;
1606
1607         __swab32s(&pbuf->pb_info.pi_magic);
1608         __swab32s(&pbuf->pb_info.pi_features);
1609         __swab32s(&pbuf->pb_info.pi_pid);
1610         __swab32s(&pbuf->pb_info.pi_nnis);
1611         nnis = pbuf->pb_info.pi_nnis;
1612         if (nnis > pbuf->pb_nnis)
1613                 nnis = pbuf->pb_nnis;
1614         for (i = 0; i < nnis; i++) {
1615                 stat = &pbuf->pb_info.pi_ni[i];
1616                 __swab64s(&stat->ns_nid);
1617                 __swab32s(&stat->ns_status);
1618         }
1619 }
1620
1621 int
1622 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1623 {
1624         if (!pinfo)
1625                 return -EINVAL;
1626         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1627                 return -EPROTO;
1628         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1629                 return -EPROTO;
1630         /* Loopback is guaranteed to be present */
1631         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1632                 return -ERANGE;
1633         if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1634                 return -EPROTO;
1635         return 0;
1636 }
1637
1638 static void
1639 lnet_ping_target_destroy(void)
1640 {
1641         struct lnet_net *net;
1642         struct lnet_ni  *ni;
1643
1644         lnet_net_lock(LNET_LOCK_EX);
1645
1646         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1647                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1648                         lnet_ni_lock(ni);
1649                         ni->ni_status = NULL;
1650                         lnet_ni_unlock(ni);
1651                 }
1652         }
1653
1654         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1655         the_lnet.ln_ping_target = NULL;
1656
1657         lnet_net_unlock(LNET_LOCK_EX);
1658 }
1659
1660 static void
1661 lnet_ping_target_event_handler(struct lnet_event *event)
1662 {
1663         struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1664
1665         if (event->unlinked)
1666                 lnet_ping_buffer_decref(pbuf);
1667 }
1668
1669 static int
1670 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1671                        struct lnet_handle_md *ping_mdh,
1672                        int ni_count, bool set_eq)
1673 {
1674         struct lnet_process_id id = {
1675                 .nid = LNET_NID_ANY,
1676                 .pid = LNET_PID_ANY
1677         };
1678         struct lnet_me *me;
1679         struct lnet_md md = { NULL };
1680         int rc, rc2;
1681
1682         if (set_eq) {
1683                 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1684                                  &the_lnet.ln_ping_target_eq);
1685                 if (rc != 0) {
1686                         CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1687                         return rc;
1688                 }
1689         }
1690
1691         *ppbuf = lnet_ping_target_create(ni_count);
1692         if (*ppbuf == NULL) {
1693                 rc = -ENOMEM;
1694                 goto fail_free_eq;
1695         }
1696
1697         /* Ping target ME/MD */
1698         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1699                           LNET_PROTO_PING_MATCHBITS, 0,
1700                           LNET_UNLINK, LNET_INS_AFTER);
1701         if (IS_ERR(me)) {
1702                 rc = PTR_ERR(me);
1703                 CERROR("Can't create ping target ME: %d\n", rc);
1704                 goto fail_decref_ping_buffer;
1705         }
1706
1707         /* initialize md content */
1708         md.start     = &(*ppbuf)->pb_info;
1709         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1710         md.threshold = LNET_MD_THRESH_INF;
1711         md.max_size  = 0;
1712         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1713                        LNET_MD_MANAGE_REMOTE;
1714         md.eq_handle = the_lnet.ln_ping_target_eq;
1715         md.user_ptr  = *ppbuf;
1716
1717         rc = LNetMDAttach(me, md, LNET_RETAIN, ping_mdh);
1718         if (rc != 0) {
1719                 CERROR("Can't attach ping target MD: %d\n", rc);
1720                 goto fail_unlink_ping_me;
1721         }
1722         lnet_ping_buffer_addref(*ppbuf);
1723
1724         return 0;
1725
1726 fail_unlink_ping_me:
1727         LNetMEUnlink(me);
1728 fail_decref_ping_buffer:
1729         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1730         lnet_ping_buffer_decref(*ppbuf);
1731         *ppbuf = NULL;
1732 fail_free_eq:
1733         if (set_eq) {
1734                 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1735                 LASSERT(rc2 == 0);
1736         }
1737         return rc;
1738 }
1739
1740 static void
1741 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1742                     struct lnet_handle_md *ping_mdh)
1743 {
1744         sigset_t        blocked = cfs_block_allsigs();
1745
1746         LNetMDUnlink(*ping_mdh);
1747         LNetInvalidateMDHandle(ping_mdh);
1748
1749         /* NB the MD could be busy; this just starts the unlink */
1750         while (atomic_read(&pbuf->pb_refcnt) > 1) {
1751                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1752                 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1753         }
1754
1755         cfs_restore_sigs(blocked);
1756 }
1757
1758 static void
1759 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1760 {
1761         struct lnet_ni          *ni;
1762         struct lnet_net         *net;
1763         struct lnet_ni_status *ns;
1764         int                     i;
1765         int                     rc;
1766
1767         i = 0;
1768         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1769                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1770                         LASSERT(i < pbuf->pb_nnis);
1771
1772                         ns = &pbuf->pb_info.pi_ni[i];
1773
1774                         ns->ns_nid = ni->ni_nid;
1775
1776                         lnet_ni_lock(ni);
1777                         ns->ns_status = (ni->ni_status != NULL) ?
1778                                          ni->ni_status->ns_status :
1779                                                 LNET_NI_STATUS_UP;
1780                         ni->ni_status = ns;
1781                         lnet_ni_unlock(ni);
1782
1783                         i++;
1784                 }
1785         }
1786         /*
1787          * We (ab)use the ns_status of the loopback interface to
1788          * transmit the sequence number. The first interface listed
1789          * must be the loopback interface.
1790          */
1791         rc = lnet_ping_info_validate(&pbuf->pb_info);
1792         if (rc) {
1793                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1794                 LBUG();
1795         }
1796         LNET_PING_BUFFER_SEQNO(pbuf) =
1797                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1798 }
1799
1800 static void
1801 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1802                         struct lnet_handle_md ping_mdh)
1803 {
1804         struct lnet_ping_buffer *old_pbuf = NULL;
1805         struct lnet_handle_md old_ping_md;
1806
1807         /* switch the NIs to point to the new ping info created */
1808         lnet_net_lock(LNET_LOCK_EX);
1809
1810         if (!the_lnet.ln_routing)
1811                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1812         if (!lnet_peer_discovery_disabled)
1813                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1814
1815         /* Ensure only known feature bits have been set. */
1816         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1817         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1818
1819         lnet_ping_target_install_locked(pbuf);
1820
1821         if (the_lnet.ln_ping_target) {
1822                 old_pbuf = the_lnet.ln_ping_target;
1823                 old_ping_md = the_lnet.ln_ping_target_md;
1824         }
1825         the_lnet.ln_ping_target_md = ping_mdh;
1826         the_lnet.ln_ping_target = pbuf;
1827
1828         lnet_net_unlock(LNET_LOCK_EX);
1829
1830         if (old_pbuf) {
1831                 /* unlink and free the old ping info */
1832                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1833                 lnet_ping_buffer_decref(old_pbuf);
1834         }
1835
1836         lnet_push_update_to_peers(0);
1837 }
1838
1839 static void
1840 lnet_ping_target_fini(void)
1841 {
1842         int             rc;
1843
1844         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1845                             &the_lnet.ln_ping_target_md);
1846
1847         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1848         LASSERT(rc == 0);
1849
1850         lnet_ping_target_destroy();
1851 }
1852
1853 /* Resize the push target. */
1854 int lnet_push_target_resize(void)
1855 {
1856         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1857         struct lnet_md md = { NULL };
1858         struct lnet_me *me;
1859         struct lnet_handle_md mdh;
1860         struct lnet_handle_md old_mdh;
1861         struct lnet_ping_buffer *pbuf;
1862         struct lnet_ping_buffer *old_pbuf;
1863         int nnis = the_lnet.ln_push_target_nnis;
1864         int rc;
1865
1866         if (nnis <= 0) {
1867                 rc = -EINVAL;
1868                 goto fail_return;
1869         }
1870 again:
1871         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1872         if (!pbuf) {
1873                 rc = -ENOMEM;
1874                 goto fail_return;
1875         }
1876
1877         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1878                           LNET_PROTO_PING_MATCHBITS, 0,
1879                           LNET_UNLINK, LNET_INS_AFTER);
1880
1881         if (IS_ERR(me)) {
1882                 rc = PTR_ERR(me);
1883                 CERROR("Can't create push target ME: %d\n", rc);
1884                 goto fail_decref_pbuf;
1885         }
1886
1887         /* initialize md content */
1888         md.start     = &pbuf->pb_info;
1889         md.length    = LNET_PING_INFO_SIZE(nnis);
1890         md.threshold = LNET_MD_THRESH_INF;
1891         md.max_size  = 0;
1892         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1893                        LNET_MD_MANAGE_REMOTE;
1894         md.user_ptr  = pbuf;
1895         md.eq_handle = the_lnet.ln_push_target_eq;
1896
1897         rc = LNetMDAttach(me, md, LNET_RETAIN, &mdh);
1898         if (rc) {
1899                 CERROR("Can't attach push MD: %d\n", rc);
1900                 goto fail_unlink_me;
1901         }
1902         lnet_ping_buffer_addref(pbuf);
1903
1904         lnet_net_lock(LNET_LOCK_EX);
1905         old_pbuf = the_lnet.ln_push_target;
1906         old_mdh = the_lnet.ln_push_target_md;
1907         the_lnet.ln_push_target = pbuf;
1908         the_lnet.ln_push_target_md = mdh;
1909         lnet_net_unlock(LNET_LOCK_EX);
1910
1911         if (old_pbuf) {
1912                 LNetMDUnlink(old_mdh);
1913                 lnet_ping_buffer_decref(old_pbuf);
1914         }
1915
1916         if (nnis < the_lnet.ln_push_target_nnis)
1917                 goto again;
1918
1919         CDEBUG(D_NET, "nnis %d success\n", nnis);
1920
1921         return 0;
1922
1923 fail_unlink_me:
1924         LNetMEUnlink(me);
1925 fail_decref_pbuf:
1926         lnet_ping_buffer_decref(pbuf);
1927 fail_return:
1928         CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1929         return rc;
1930 }
1931
1932 static void lnet_push_target_event_handler(struct lnet_event *ev)
1933 {
1934         struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1935
1936         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1937                 lnet_swap_pinginfo(pbuf);
1938
1939         lnet_peer_push_event(ev);
1940         if (ev->unlinked)
1941                 lnet_ping_buffer_decref(pbuf);
1942 }
1943
1944 /* Initialize the push target. */
1945 static int lnet_push_target_init(void)
1946 {
1947         int rc;
1948
1949         if (the_lnet.ln_push_target)
1950                 return -EALREADY;
1951
1952         rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1953                          &the_lnet.ln_push_target_eq);
1954         if (rc) {
1955                 CERROR("Can't allocated push target EQ: %d\n", rc);
1956                 return rc;
1957         }
1958
1959         /* Start at the required minimum, we'll enlarge if required. */
1960         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1961
1962         rc = lnet_push_target_resize();
1963
1964         if (rc) {
1965                 LNetEQFree(the_lnet.ln_push_target_eq);
1966                 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1967         }
1968
1969         return rc;
1970 }
1971
1972 /* Clean up the push target. */
1973 static void lnet_push_target_fini(void)
1974 {
1975         if (!the_lnet.ln_push_target)
1976                 return;
1977
1978         /* Unlink and invalidate to prevent new references. */
1979         LNetMDUnlink(the_lnet.ln_push_target_md);
1980         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1981
1982         /* Wait for the unlink to complete. */
1983         while (atomic_read(&the_lnet.ln_push_target->pb_refcnt) > 1) {
1984                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1985                 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1986         }
1987
1988         lnet_ping_buffer_decref(the_lnet.ln_push_target);
1989         the_lnet.ln_push_target = NULL;
1990         the_lnet.ln_push_target_nnis = 0;
1991
1992         LNetEQFree(the_lnet.ln_push_target_eq);
1993         LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1994 }
1995
1996 static int
1997 lnet_ni_tq_credits(struct lnet_ni *ni)
1998 {
1999         int     credits;
2000
2001         LASSERT(ni->ni_ncpts >= 1);
2002
2003         if (ni->ni_ncpts == 1)
2004                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2005
2006         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2007         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2008         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2009
2010         return credits;
2011 }
2012
2013 static void
2014 lnet_ni_unlink_locked(struct lnet_ni *ni)
2015 {
2016         /* move it to zombie list and nobody can find it anymore */
2017         LASSERT(!list_empty(&ni->ni_netlist));
2018         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2019         lnet_ni_decref_locked(ni, 0);
2020 }
2021
2022 static void
2023 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2024 {
2025         int             i;
2026         int             islo;
2027         struct lnet_ni  *ni;
2028         struct list_head *zombie_list = &net->net_ni_zombie;
2029
2030         /*
2031          * Now wait for the NIs I just nuked to show up on the zombie
2032          * list and shut them down in guaranteed thread context
2033          */
2034         i = 2;
2035         while (!list_empty(zombie_list)) {
2036                 int     *ref;
2037                 int     j;
2038
2039                 ni = list_entry(zombie_list->next,
2040                                 struct lnet_ni, ni_netlist);
2041                 list_del_init(&ni->ni_netlist);
2042                 /* the ni should be in deleting state. If it's not it's
2043                  * a bug */
2044                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2045                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2046                         if (*ref == 0)
2047                                 continue;
2048                         /* still busy, add it back to zombie list */
2049                         list_add(&ni->ni_netlist, zombie_list);
2050                         break;
2051                 }
2052
2053                 if (!list_empty(&ni->ni_netlist)) {
2054                         lnet_net_unlock(LNET_LOCK_EX);
2055                         ++i;
2056                         if ((i & (-i)) == i) {
2057                                 CDEBUG(D_WARNING,
2058                                        "Waiting for zombie LNI %s\n",
2059                                        libcfs_nid2str(ni->ni_nid));
2060                         }
2061                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2062                         lnet_net_lock(LNET_LOCK_EX);
2063                         continue;
2064                 }
2065
2066                 lnet_net_unlock(LNET_LOCK_EX);
2067
2068                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2069
2070                 LASSERT(!in_interrupt());
2071                 /* Holding the mutex makes it safe for lnd_shutdown
2072                  * to call module_put(). Module unload cannot finish
2073                  * until lnet_unregister_lnd() completes, and that
2074                  * requires the mutex.
2075                  */
2076                 mutex_lock(&the_lnet.ln_lnd_mutex);
2077                 (net->net_lnd->lnd_shutdown)(ni);
2078                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2079
2080                 if (!islo)
2081                         CDEBUG(D_LNI, "Removed LNI %s\n",
2082                               libcfs_nid2str(ni->ni_nid));
2083
2084                 lnet_ni_free(ni);
2085                 i = 2;
2086                 lnet_net_lock(LNET_LOCK_EX);
2087         }
2088 }
2089
2090 /* shutdown down the NI and release refcount */
2091 static void
2092 lnet_shutdown_lndni(struct lnet_ni *ni)
2093 {
2094         int i;
2095         struct lnet_net *net = ni->ni_net;
2096
2097         lnet_net_lock(LNET_LOCK_EX);
2098         lnet_ni_lock(ni);
2099         ni->ni_state = LNET_NI_STATE_DELETING;
2100         lnet_ni_unlock(ni);
2101         lnet_ni_unlink_locked(ni);
2102         lnet_incr_dlc_seq();
2103         lnet_net_unlock(LNET_LOCK_EX);
2104
2105         /* clear messages for this NI on the lazy portal */
2106         for (i = 0; i < the_lnet.ln_nportals; i++)
2107                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2108
2109         lnet_net_lock(LNET_LOCK_EX);
2110         lnet_clear_zombies_nis_locked(net);
2111         lnet_net_unlock(LNET_LOCK_EX);
2112 }
2113
2114 static void
2115 lnet_shutdown_lndnet(struct lnet_net *net)
2116 {
2117         struct lnet_ni *ni;
2118
2119         lnet_net_lock(LNET_LOCK_EX);
2120
2121         list_del_init(&net->net_list);
2122
2123         while (!list_empty(&net->net_ni_list)) {
2124                 ni = list_entry(net->net_ni_list.next,
2125                                 struct lnet_ni, ni_netlist);
2126                 lnet_net_unlock(LNET_LOCK_EX);
2127                 lnet_shutdown_lndni(ni);
2128                 lnet_net_lock(LNET_LOCK_EX);
2129         }
2130
2131         lnet_net_unlock(LNET_LOCK_EX);
2132
2133         /* Do peer table cleanup for this net */
2134         lnet_peer_tables_cleanup(net);
2135
2136         lnet_net_free(net);
2137 }
2138
2139 static void
2140 lnet_shutdown_lndnets(void)
2141 {
2142         struct lnet_net *net;
2143         LIST_HEAD(resend);
2144         struct lnet_msg *msg, *tmp;
2145
2146         /* NB called holding the global mutex */
2147
2148         /* All quiet on the API front */
2149         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2150         LASSERT(the_lnet.ln_refcount == 0);
2151
2152         lnet_net_lock(LNET_LOCK_EX);
2153         the_lnet.ln_state = LNET_STATE_STOPPING;
2154
2155         while (!list_empty(&the_lnet.ln_nets)) {
2156                 /*
2157                  * move the nets to the zombie list to avoid them being
2158                  * picked up for new work. LONET is also included in the
2159                  * Nets that will be moved to the zombie list
2160                  */
2161                 net = list_entry(the_lnet.ln_nets.next,
2162                                  struct lnet_net, net_list);
2163                 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2164         }
2165
2166         /* Drop the cached loopback Net. */
2167         if (the_lnet.ln_loni != NULL) {
2168                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2169                 the_lnet.ln_loni = NULL;
2170         }
2171         lnet_net_unlock(LNET_LOCK_EX);
2172
2173         /* iterate through the net zombie list and delete each net */
2174         while (!list_empty(&the_lnet.ln_net_zombie)) {
2175                 net = list_entry(the_lnet.ln_net_zombie.next,
2176                                  struct lnet_net, net_list);
2177                 lnet_shutdown_lndnet(net);
2178         }
2179
2180         spin_lock(&the_lnet.ln_msg_resend_lock);
2181         list_splice(&the_lnet.ln_msg_resend, &resend);
2182         spin_unlock(&the_lnet.ln_msg_resend_lock);
2183
2184         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2185                 list_del_init(&msg->msg_list);
2186                 msg->msg_no_resend = true;
2187                 lnet_finalize(msg, -ECANCELED);
2188         }
2189
2190         lnet_net_lock(LNET_LOCK_EX);
2191         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2192         lnet_net_unlock(LNET_LOCK_EX);
2193 }
2194
2195 static int
2196 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2197 {
2198         int                     rc = -EINVAL;
2199         struct lnet_tx_queue    *tq;
2200         int                     i;
2201         struct lnet_net         *net = ni->ni_net;
2202
2203         mutex_lock(&the_lnet.ln_lnd_mutex);
2204
2205         if (tun) {
2206                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2207                 ni->ni_lnd_tunables_set = true;
2208         }
2209
2210         rc = (net->net_lnd->lnd_startup)(ni);
2211
2212         mutex_unlock(&the_lnet.ln_lnd_mutex);
2213
2214         if (rc != 0) {
2215                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2216                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2217                 goto failed0;
2218         }
2219
2220         lnet_ni_lock(ni);
2221         ni->ni_state = LNET_NI_STATE_ACTIVE;
2222         lnet_ni_unlock(ni);
2223
2224         /* We keep a reference on the loopback net through the loopback NI */
2225         if (net->net_lnd->lnd_type == LOLND) {
2226                 lnet_ni_addref(ni);
2227                 LASSERT(the_lnet.ln_loni == NULL);
2228                 the_lnet.ln_loni = ni;
2229                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2230                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2231                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2232                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2233                 return 0;
2234         }
2235
2236         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2237             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2238                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2239                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2240                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2241                                         "" : "per-peer ");
2242                 /* shutdown the NI since if we get here then it must've already
2243                  * been started
2244                  */
2245                 lnet_shutdown_lndni(ni);
2246                 return -EINVAL;
2247         }
2248
2249         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2250                 tq->tq_credits_min =
2251                 tq->tq_credits_max =
2252                 tq->tq_credits = lnet_ni_tq_credits(ni);
2253         }
2254
2255         atomic_set(&ni->ni_tx_credits,
2256                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2257         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2258
2259         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2260                 libcfs_nid2str(ni->ni_nid),
2261                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2262                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2263                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2264                 ni->ni_net->net_tunables.lct_peer_timeout);
2265
2266         return 0;
2267 failed0:
2268         lnet_ni_free(ni);
2269         return rc;
2270 }
2271
2272 static int
2273 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2274 {
2275         struct lnet_ni *ni;
2276         struct lnet_net *net_l = NULL;
2277         LIST_HEAD(local_ni_list);
2278         int rc;
2279         int ni_count = 0;
2280         __u32 lnd_type;
2281         const struct lnet_lnd  *lnd;
2282         int peer_timeout =
2283                 net->net_tunables.lct_peer_timeout;
2284         int maxtxcredits =
2285                 net->net_tunables.lct_max_tx_credits;
2286         int peerrtrcredits =
2287                 net->net_tunables.lct_peer_rtr_credits;
2288
2289         /*
2290          * make sure that this net is unique. If it isn't then
2291          * we are adding interfaces to an already existing network, and
2292          * 'net' is just a convenient way to pass in the list.
2293          * if it is unique we need to find the LND and load it if
2294          * necessary.
2295          */
2296         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2297                 lnd_type = LNET_NETTYP(net->net_id);
2298
2299                 mutex_lock(&the_lnet.ln_lnd_mutex);
2300                 lnd = lnet_find_lnd_by_type(lnd_type);
2301
2302                 if (lnd == NULL) {
2303                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2304                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2305                         mutex_lock(&the_lnet.ln_lnd_mutex);
2306
2307                         lnd = lnet_find_lnd_by_type(lnd_type);
2308                         if (lnd == NULL) {
2309                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2310                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2311                                 libcfs_lnd2str(lnd_type),
2312                                 libcfs_lnd2modname(lnd_type), rc);
2313 #ifndef HAVE_MODULE_LOADING_SUPPORT
2314                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2315                                                 "compiled with kernel module "
2316                                                 "loading support.");
2317 #endif
2318                                 rc = -EINVAL;
2319                                 goto failed0;
2320                         }
2321                 }
2322
2323                 net->net_lnd = lnd;
2324
2325                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2326
2327                 net_l = net;
2328         }
2329
2330         /*
2331          * net_l: if the network being added is unique then net_l
2332          *        will point to that network
2333          *        if the network being added is not unique then
2334          *        net_l points to the existing network.
2335          *
2336          * When we enter the loop below, we'll pick NIs off he
2337          * network beign added and start them up, then add them to
2338          * a local ni list. Once we've successfully started all
2339          * the NIs then we join the local NI list (of started up
2340          * networks) with the net_l->net_ni_list, which should
2341          * point to the correct network to add the new ni list to
2342          *
2343          * If any of the new NIs fail to start up, then we want to
2344          * iterate through the local ni list, which should include
2345          * any NIs which were successfully started up, and shut
2346          * them down.
2347          *
2348          * After than we want to delete the network being added,
2349          * to avoid a memory leak.
2350          */
2351
2352         /*
2353          * When a network uses TCP bonding then all its interfaces
2354          * must be specified when the network is first defined: the
2355          * TCP bonding code doesn't allow for interfaces to be added
2356          * or removed.
2357          */
2358         if (net_l != net && net_l != NULL && use_tcp_bonding &&
2359             LNET_NETTYP(net_l->net_id) == SOCKLND) {
2360                 rc = -EINVAL;
2361                 goto failed0;
2362         }
2363
2364         while (!list_empty(&net->net_ni_added)) {
2365                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2366                                 ni_netlist);
2367                 list_del_init(&ni->ni_netlist);
2368
2369                 /* make sure that the the NI we're about to start
2370                  * up is actually unique. if it's not fail. */
2371                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2372                                         ni->ni_interfaces[0])) {
2373                         rc = -EEXIST;
2374                         goto failed1;
2375                 }
2376
2377                 /* adjust the pointer the parent network, just in case it
2378                  * the net is a duplicate */
2379                 ni->ni_net = net_l;
2380
2381                 rc = lnet_startup_lndni(ni, tun);
2382
2383                 if (rc < 0)
2384                         goto failed1;
2385
2386                 lnet_ni_addref(ni);
2387                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2388
2389                 ni_count++;
2390         }
2391
2392         lnet_net_lock(LNET_LOCK_EX);
2393         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2394         lnet_incr_dlc_seq();
2395         lnet_net_unlock(LNET_LOCK_EX);
2396
2397         /* if the network is not unique then we don't want to keep
2398          * it around after we're done. Free it. Otherwise add that
2399          * net to the global the_lnet.ln_nets */
2400         if (net_l != net && net_l != NULL) {
2401                 /*
2402                  * TODO - note. currently the tunables can not be updated
2403                  * once added
2404                  */
2405                 lnet_net_free(net);
2406         } else {
2407                 /*
2408                  * restore tunables after it has been overwitten by the
2409                  * lnd
2410                  */
2411                 if (peer_timeout != -1)
2412                         net->net_tunables.lct_peer_timeout = peer_timeout;
2413                 if (maxtxcredits != -1)
2414                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2415                 if (peerrtrcredits != -1)
2416                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2417
2418                 lnet_net_lock(LNET_LOCK_EX);
2419                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2420                 lnet_net_unlock(LNET_LOCK_EX);
2421         }
2422
2423         /* update net count */
2424         lnet_current_net_count = lnet_get_net_count();
2425
2426         return ni_count;
2427
2428 failed1:
2429         /*
2430          * shutdown the new NIs that are being started up
2431          * free the NET being started
2432          */
2433         while (!list_empty(&local_ni_list)) {
2434                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2435                                 ni_netlist);
2436
2437                 lnet_shutdown_lndni(ni);
2438         }
2439
2440 failed0:
2441         lnet_net_free(net);
2442
2443         return rc;
2444 }
2445
2446 static int
2447 lnet_startup_lndnets(struct list_head *netlist)
2448 {
2449         struct lnet_net         *net;
2450         int                     rc;
2451         int                     ni_count = 0;
2452
2453         /*
2454          * Change to running state before bringing up the LNDs. This
2455          * allows lnet_shutdown_lndnets() to assert that we've passed
2456          * through here.
2457          */
2458         lnet_net_lock(LNET_LOCK_EX);
2459         the_lnet.ln_state = LNET_STATE_RUNNING;
2460         lnet_net_unlock(LNET_LOCK_EX);
2461
2462         while (!list_empty(netlist)) {
2463                 net = list_entry(netlist->next, struct lnet_net, net_list);
2464                 list_del_init(&net->net_list);
2465
2466                 rc = lnet_startup_lndnet(net, NULL);
2467
2468                 if (rc < 0)
2469                         goto failed;
2470
2471                 ni_count += rc;
2472         }
2473
2474         return ni_count;
2475 failed:
2476         lnet_shutdown_lndnets();
2477
2478         return rc;
2479 }
2480
2481 /**
2482  * Initialize LNet library.
2483  *
2484  * Automatically called at module loading time. Caller has to call
2485  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2486  * latter returned 0. It must be called exactly once.
2487  *
2488  * \retval 0 on success
2489  * \retval -ve on failures.
2490  */
2491 int lnet_lib_init(void)
2492 {
2493         int rc;
2494
2495         lnet_assert_wire_constants();
2496
2497         /* refer to global cfs_cpt_table for now */
2498         the_lnet.ln_cpt_table   = cfs_cpt_table;
2499         the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
2500
2501         LASSERT(the_lnet.ln_cpt_number > 0);
2502         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2503                 /* we are under risk of consuming all lh_cookie */
2504                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2505                        "please change setting of CPT-table and retry\n",
2506                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2507                 return -E2BIG;
2508         }
2509
2510         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2511                 the_lnet.ln_cpt_bits++;
2512
2513         rc = lnet_create_locks();
2514         if (rc != 0) {
2515                 CERROR("Can't create LNet global locks: %d\n", rc);
2516                 return rc;
2517         }
2518
2519         the_lnet.ln_refcount = 0;
2520         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2521         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2522
2523         /* The hash table size is the number of bits it takes to express the set
2524          * ln_num_routes, minus 1 (better to under estimate than over so we
2525          * don't waste memory). */
2526         if (rnet_htable_size <= 0)
2527                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2528         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2529                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2530         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2531                                            order_base_2(rnet_htable_size) - 1);
2532
2533         /* All LNDs apart from the LOLND are in separate modules.  They
2534          * register themselves when their module loads, and unregister
2535          * themselves when their module is unloaded. */
2536         lnet_register_lnd(&the_lolnd);
2537         return 0;
2538 }
2539
2540 /**
2541  * Finalize LNet library.
2542  *
2543  * \pre lnet_lib_init() called with success.
2544  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2545  *
2546  * As this happens at module-unload, all lnds must already be unloaded,
2547  * so they must already be unregistered.
2548  */
2549 void lnet_lib_exit(void)
2550 {
2551         int i;
2552
2553         LASSERT(the_lnet.ln_refcount == 0);
2554         lnet_unregister_lnd(&the_lolnd);
2555         for (i = 0; i < NUM_LNDS; i++)
2556                 LASSERT(!the_lnet.ln_lnds[i]);
2557         lnet_destroy_locks();
2558 }
2559
2560 /**
2561  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2562  *
2563  * Users must call this function at least once before any other functions.
2564  * For each successful call there must be a corresponding call to
2565  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2566  * ignored.
2567  *
2568  * The PID used by LNet may be different from the one requested.
2569  * See LNetGetId().
2570  *
2571  * \param requested_pid PID requested by the caller.
2572  *
2573  * \return >= 0 on success, and < 0 error code on failures.
2574  */
2575 int
2576 LNetNIInit(lnet_pid_t requested_pid)
2577 {
2578         int                     im_a_router = 0;
2579         int                     rc;
2580         int                     ni_count;
2581         struct lnet_ping_buffer *pbuf;
2582         struct lnet_handle_md   ping_mdh;
2583         LIST_HEAD(net_head);
2584         struct lnet_net         *net;
2585
2586         mutex_lock(&the_lnet.ln_api_mutex);
2587
2588         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2589
2590         if (the_lnet.ln_refcount > 0) {
2591                 rc = the_lnet.ln_refcount++;
2592                 mutex_unlock(&the_lnet.ln_api_mutex);
2593                 return rc;
2594         }
2595
2596         rc = lnet_prepare(requested_pid);
2597         if (rc != 0) {
2598                 mutex_unlock(&the_lnet.ln_api_mutex);
2599                 return rc;
2600         }
2601
2602         /* create a network for Loopback network */
2603         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2604         if (net == NULL) {
2605                 rc = -ENOMEM;
2606                 goto err_empty_list;
2607         }
2608
2609         /* Add in the loopback NI */
2610         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2611                 rc = -ENOMEM;
2612                 goto err_empty_list;
2613         }
2614
2615         /* If LNet is being initialized via DLC it is possible
2616          * that the user requests not to load module parameters (ones which
2617          * are supported by DLC) on initialization.  Therefore, make sure not
2618          * to load networks, routes and forwarding from module parameters
2619          * in this case.  On cleanup in case of failure only clean up
2620          * routes if it has been loaded */
2621         if (!the_lnet.ln_nis_from_mod_params) {
2622                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2623                                          use_tcp_bonding);
2624                 if (rc < 0)
2625                         goto err_empty_list;
2626         }
2627
2628         ni_count = lnet_startup_lndnets(&net_head);
2629         if (ni_count < 0) {
2630                 rc = ni_count;
2631                 goto err_empty_list;
2632         }
2633
2634         if (!the_lnet.ln_nis_from_mod_params) {
2635                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2636                 if (rc != 0)
2637                         goto err_shutdown_lndnis;
2638
2639                 rc = lnet_rtrpools_alloc(im_a_router);
2640                 if (rc != 0)
2641                         goto err_destroy_routes;
2642         }
2643
2644         rc = lnet_acceptor_start();
2645         if (rc != 0)
2646                 goto err_destroy_routes;
2647
2648         the_lnet.ln_refcount = 1;
2649         /* Now I may use my own API functions... */
2650
2651         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2652         if (rc != 0)
2653                 goto err_acceptor_stop;
2654
2655         lnet_ping_target_update(pbuf, ping_mdh);
2656
2657         rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2658         if (rc != 0) {
2659                 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2660                 goto err_stop_ping;
2661         }
2662
2663         rc = lnet_push_target_init();
2664         if (rc != 0)
2665                 goto err_stop_ping;
2666
2667         rc = lnet_peer_discovery_start();
2668         if (rc != 0)
2669                 goto err_destroy_push_target;
2670
2671         rc = lnet_monitor_thr_start();
2672         if (rc != 0)
2673                 goto err_stop_discovery_thr;
2674
2675         lnet_fault_init();
2676         lnet_router_debugfs_init();
2677
2678         mutex_unlock(&the_lnet.ln_api_mutex);
2679
2680         complete_all(&the_lnet.ln_started);
2681
2682         /* wait for all routers to start */
2683         lnet_wait_router_start();
2684
2685         return 0;
2686
2687 err_stop_discovery_thr:
2688         lnet_peer_discovery_stop();
2689 err_destroy_push_target:
2690         lnet_push_target_fini();
2691 err_stop_ping:
2692         lnet_ping_target_fini();
2693 err_acceptor_stop:
2694         the_lnet.ln_refcount = 0;
2695         lnet_acceptor_stop();
2696 err_destroy_routes:
2697         if (!the_lnet.ln_nis_from_mod_params)
2698                 lnet_destroy_routes();
2699 err_shutdown_lndnis:
2700         lnet_shutdown_lndnets();
2701 err_empty_list:
2702         lnet_unprepare();
2703         LASSERT(rc < 0);
2704         mutex_unlock(&the_lnet.ln_api_mutex);
2705         while (!list_empty(&net_head)) {
2706                 struct lnet_net *net;
2707
2708                 net = list_entry(net_head.next, struct lnet_net, net_list);
2709                 list_del_init(&net->net_list);
2710                 lnet_net_free(net);
2711         }
2712         return rc;
2713 }
2714 EXPORT_SYMBOL(LNetNIInit);
2715
2716 /**
2717  * Stop LNet interfaces, routing, and forwarding.
2718  *
2719  * Users must call this function once for each successful call to LNetNIInit().
2720  * Once the LNetNIFini() operation has been started, the results of pending
2721  * API operations are undefined.
2722  *
2723  * \return always 0 for current implementation.
2724  */
2725 int
2726 LNetNIFini(void)
2727 {
2728         mutex_lock(&the_lnet.ln_api_mutex);
2729
2730         LASSERT(the_lnet.ln_refcount > 0);
2731
2732         if (the_lnet.ln_refcount != 1) {
2733                 the_lnet.ln_refcount--;
2734         } else {
2735                 LASSERT(!the_lnet.ln_niinit_self);
2736
2737                 lnet_fault_fini();
2738
2739                 lnet_router_debugfs_fini();
2740                 lnet_monitor_thr_stop();
2741                 lnet_peer_discovery_stop();
2742                 lnet_push_target_fini();
2743                 lnet_ping_target_fini();
2744
2745                 /* Teardown fns that use my own API functions BEFORE here */
2746                 the_lnet.ln_refcount = 0;
2747
2748                 lnet_acceptor_stop();
2749                 lnet_destroy_routes();
2750                 lnet_shutdown_lndnets();
2751                 lnet_unprepare();
2752         }
2753
2754         mutex_unlock(&the_lnet.ln_api_mutex);
2755         return 0;
2756 }
2757 EXPORT_SYMBOL(LNetNIFini);
2758
2759 /**
2760  * Grabs the ni data from the ni structure and fills the out
2761  * parameters
2762  *
2763  * \param[in] ni network        interface structure
2764  * \param[out] cfg_ni           NI config information
2765  * \param[out] tun              network and LND tunables
2766  */
2767 static void
2768 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2769                    struct lnet_ioctl_config_lnd_tunables *tun,
2770                    struct lnet_ioctl_element_stats *stats,
2771                    __u32 tun_size)
2772 {
2773         size_t min_size = 0;
2774         int i;
2775
2776         if (!ni || !cfg_ni || !tun)
2777                 return;
2778
2779         if (ni->ni_interfaces[0] != NULL) {
2780                 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2781                         if (ni->ni_interfaces[i] != NULL) {
2782                                 strncpy(cfg_ni->lic_ni_intf[i],
2783                                         ni->ni_interfaces[i],
2784                                         sizeof(cfg_ni->lic_ni_intf[i]));
2785                         }
2786                 }
2787         }
2788
2789         cfg_ni->lic_nid = ni->ni_nid;
2790         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2791                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2792         else
2793                 cfg_ni->lic_status = ni->ni_status->ns_status;
2794         cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2795         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2796
2797         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2798
2799         if (stats) {
2800                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2801                                                        LNET_STATS_TYPE_SEND);
2802                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2803                                                        LNET_STATS_TYPE_RECV);
2804                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2805                                                        LNET_STATS_TYPE_DROP);
2806         }
2807
2808         /*
2809          * tun->lt_tun will always be present, but in order to be
2810          * backwards compatible, we need to deal with the cases when
2811          * tun->lt_tun is smaller than what the kernel has, because it
2812          * comes from an older version of a userspace program, then we'll
2813          * need to copy as much information as we have available space.
2814          */
2815         min_size = tun_size - sizeof(tun->lt_cmn);
2816         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2817
2818         /* copy over the cpts */
2819         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2820             ni->ni_cpts == NULL)  {
2821                 for (i = 0; i < ni->ni_ncpts; i++)
2822                         cfg_ni->lic_cpts[i] = i;
2823         } else {
2824                 for (i = 0;
2825                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2826                      i < LNET_MAX_SHOW_NUM_CPT;
2827                      i++)
2828                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2829         }
2830         cfg_ni->lic_ncpts = ni->ni_ncpts;
2831 }
2832
2833 /**
2834  * NOTE: This is a legacy function left in the code to be backwards
2835  * compatible with older userspace programs. It should eventually be
2836  * removed.
2837  *
2838  * Grabs the ni data from the ni structure and fills the out
2839  * parameters
2840  *
2841  * \param[in] ni network        interface structure
2842  * \param[out] config           config information
2843  */
2844 static void
2845 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2846                          struct lnet_ioctl_config_data *config)
2847 {
2848         struct lnet_ioctl_net_config *net_config;
2849         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2850         size_t min_size, tunable_size = 0;
2851         int i;
2852
2853         if (!ni || !config)
2854                 return;
2855
2856         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2857         if (!net_config)
2858                 return;
2859
2860         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2861                      ARRAY_SIZE(net_config->ni_interfaces));
2862
2863         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2864                 if (!ni->ni_interfaces[i])
2865                         break;
2866
2867                 strncpy(net_config->ni_interfaces[i],
2868                         ni->ni_interfaces[i],
2869                         sizeof(net_config->ni_interfaces[i]));
2870         }
2871
2872         config->cfg_nid = ni->ni_nid;
2873         config->cfg_config_u.cfg_net.net_peer_timeout =
2874                 ni->ni_net->net_tunables.lct_peer_timeout;
2875         config->cfg_config_u.cfg_net.net_max_tx_credits =
2876                 ni->ni_net->net_tunables.lct_max_tx_credits;
2877         config->cfg_config_u.cfg_net.net_peer_tx_credits =
2878                 ni->ni_net->net_tunables.lct_peer_tx_credits;
2879         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2880                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2881
2882         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2883                 net_config->ni_status = LNET_NI_STATUS_UP;
2884         else
2885                 net_config->ni_status = ni->ni_status->ns_status;
2886
2887         if (ni->ni_cpts) {
2888                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2889
2890                 for (i = 0; i < num_cpts; i++)
2891                         net_config->ni_cpts[i] = ni->ni_cpts[i];
2892
2893                 config->cfg_ncpts = num_cpts;
2894         }
2895
2896         /*
2897          * See if user land tools sent in a newer and larger version
2898          * of struct lnet_tunables than what the kernel uses.
2899          */
2900         min_size = sizeof(*config) + sizeof(*net_config);
2901
2902         if (config->cfg_hdr.ioc_len > min_size)
2903                 tunable_size = config->cfg_hdr.ioc_len - min_size;
2904
2905         /* Don't copy too much data to user space */
2906         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2907         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2908
2909         if (lnd_cfg && min_size) {
2910                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2911                 config->cfg_config_u.cfg_net.net_interface_count = 1;
2912
2913                 /* Tell user land that kernel side has less data */
2914                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2915                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2916                         config->cfg_hdr.ioc_len -= min_size;
2917                 }
2918         }
2919 }
2920
2921 struct lnet_ni *
2922 lnet_get_ni_idx_locked(int idx)
2923 {
2924         struct lnet_ni          *ni;
2925         struct lnet_net         *net;
2926
2927         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2928                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2929                         if (idx-- == 0)
2930                                 return ni;
2931                 }
2932         }
2933
2934         return NULL;
2935 }
2936
2937 struct lnet_ni *
2938 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2939 {
2940         struct lnet_ni          *ni;
2941         struct lnet_net         *net = mynet;
2942
2943         /*
2944          * It is possible that the net has been cleaned out while there is
2945          * a message being sent. This function accessed the net without
2946          * checking if the list is empty
2947          */
2948         if (prev == NULL) {
2949                 if (net == NULL)
2950                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2951                                         net_list);
2952                 if (list_empty(&net->net_ni_list))
2953                         return NULL;
2954                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2955                                 ni_netlist);
2956
2957                 return ni;
2958         }
2959
2960         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2961                 /* if you reached the end of the ni list and the net is
2962                  * specified, then there are no more nis in that net */
2963                 if (net != NULL)
2964                         return NULL;
2965
2966                 /* we reached the end of this net ni list. move to the
2967                  * next net */
2968                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2969                         /* no more nets and no more NIs. */
2970                         return NULL;
2971
2972                 /* get the next net */
2973                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2974                                  net_list);
2975                 if (list_empty(&net->net_ni_list))
2976                         return NULL;
2977                 /* get the ni on it */
2978                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2979                                 ni_netlist);
2980
2981                 return ni;
2982         }
2983
2984         if (list_empty(&prev->ni_netlist))
2985                 return NULL;
2986
2987         /* there are more nis left */
2988         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2989
2990         return ni;
2991 }
2992
2993 int
2994 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2995 {
2996         struct lnet_ni *ni;
2997         int cpt;
2998         int rc = -ENOENT;
2999         int idx = config->cfg_count;
3000
3001         cpt = lnet_net_lock_current();
3002
3003         ni = lnet_get_ni_idx_locked(idx);
3004
3005         if (ni != NULL) {
3006                 rc = 0;
3007                 lnet_ni_lock(ni);
3008                 lnet_fill_ni_info_legacy(ni, config);
3009                 lnet_ni_unlock(ni);
3010         }
3011
3012         lnet_net_unlock(cpt);
3013         return rc;
3014 }
3015
3016 int
3017 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3018                    struct lnet_ioctl_config_lnd_tunables *tun,
3019                    struct lnet_ioctl_element_stats *stats,
3020                    __u32 tun_size)
3021 {
3022         struct lnet_ni          *ni;
3023         int                     cpt;
3024         int                     rc = -ENOENT;
3025
3026         if (!cfg_ni || !tun || !stats)
3027                 return -EINVAL;
3028
3029         cpt = lnet_net_lock_current();
3030
3031         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3032
3033         if (ni) {
3034                 rc = 0;
3035                 lnet_ni_lock(ni);
3036                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3037                 lnet_ni_unlock(ni);
3038         }
3039
3040         lnet_net_unlock(cpt);
3041         return rc;
3042 }
3043
3044 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3045 {
3046         struct lnet_ni *ni;
3047         int cpt;
3048         int rc = -ENOENT;
3049
3050         if (!msg_stats)
3051                 return -EINVAL;
3052
3053         cpt = lnet_net_lock_current();
3054
3055         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3056
3057         if (ni) {
3058                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3059                 rc = 0;
3060         }
3061
3062         lnet_net_unlock(cpt);
3063
3064         return rc;
3065 }
3066
3067 static int lnet_add_net_common(struct lnet_net *net,
3068                                struct lnet_ioctl_config_lnd_tunables *tun)
3069 {
3070         __u32                   net_id;
3071         struct lnet_ping_buffer *pbuf;
3072         struct lnet_handle_md   ping_mdh;
3073         int                     rc;
3074         struct lnet_remotenet *rnet;
3075         int                     net_ni_count;
3076         int                     num_acceptor_nets;
3077
3078         lnet_net_lock(LNET_LOCK_EX);
3079         rnet = lnet_find_rnet_locked(net->net_id);
3080         lnet_net_unlock(LNET_LOCK_EX);
3081         /*
3082          * make sure that the net added doesn't invalidate the current
3083          * configuration LNet is keeping
3084          */
3085         if (rnet) {
3086                 CERROR("Adding net %s will invalidate routing configuration\n",
3087                        libcfs_net2str(net->net_id));
3088                 lnet_net_free(net);
3089                 return -EUSERS;
3090         }
3091
3092         /*
3093          * make sure you calculate the correct number of slots in the ping
3094          * buffer. Since the ping info is a flattened list of all the NIs,
3095          * we should allocate enough slots to accomodate the number of NIs
3096          * which will be added.
3097          *
3098          * since ni hasn't been configured yet, use
3099          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3100          */
3101         net_ni_count = lnet_get_net_ni_count_pre(net);
3102
3103         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3104                                     net_ni_count + lnet_get_ni_count(),
3105                                     false);
3106         if (rc < 0) {
3107                 lnet_net_free(net);
3108                 return rc;
3109         }
3110
3111         if (tun)
3112                 memcpy(&net->net_tunables,
3113                        &tun->lt_cmn, sizeof(net->net_tunables));
3114         else
3115                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3116
3117         /*
3118          * before starting this network get a count of the current TCP
3119          * networks which require the acceptor thread running. If that
3120          * count is == 0 before we start up this network, then we'd want to
3121          * start up the acceptor thread after starting up this network
3122          */
3123         num_acceptor_nets = lnet_count_acceptor_nets();
3124
3125         net_id = net->net_id;
3126
3127         rc = lnet_startup_lndnet(net,
3128                                  (tun) ? &tun->lt_tun : NULL);
3129         if (rc < 0)
3130                 goto failed;
3131
3132         lnet_net_lock(LNET_LOCK_EX);
3133         net = lnet_get_net_locked(net_id);
3134         lnet_net_unlock(LNET_LOCK_EX);
3135
3136         LASSERT(net);
3137
3138         /*
3139          * Start the acceptor thread if this is the first network
3140          * being added that requires the thread.
3141          */
3142         if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3143                 rc = lnet_acceptor_start();
3144                 if (rc < 0) {
3145                         /* shutdown the net that we just started */
3146                         CERROR("Failed to start up acceptor thread\n");
3147                         lnet_shutdown_lndnet(net);
3148                         goto failed;
3149                 }
3150         }
3151
3152         lnet_net_lock(LNET_LOCK_EX);
3153         lnet_peer_net_added(net);
3154         lnet_net_unlock(LNET_LOCK_EX);
3155
3156         lnet_ping_target_update(pbuf, ping_mdh);
3157
3158         return 0;
3159
3160 failed:
3161         lnet_ping_md_unlink(pbuf, &ping_mdh);
3162         lnet_ping_buffer_decref(pbuf);
3163         return rc;
3164 }
3165
3166 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3167                                       struct lnet_ioctl_config_lnd_tunables *tun)
3168 {
3169         struct lnet_net *net;
3170         char *nets;
3171         int rc;
3172         LIST_HEAD(net_head);
3173
3174         rc = lnet_parse_ip2nets(&nets, ip2nets);
3175         if (rc < 0)
3176                 return rc;
3177
3178         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3179         if (rc < 0)
3180                 return rc;
3181
3182         mutex_lock(&the_lnet.ln_api_mutex);
3183         while (!list_empty(&net_head)) {
3184                 net = list_entry(net_head.next, struct lnet_net, net_list);
3185                 list_del_init(&net->net_list);
3186                 rc = lnet_add_net_common(net, tun);
3187                 if (rc < 0)
3188                         goto out;
3189         }
3190
3191 out:
3192         mutex_unlock(&the_lnet.ln_api_mutex);
3193
3194         while (!list_empty(&net_head)) {
3195                 net = list_entry(net_head.next, struct lnet_net, net_list);
3196                 list_del_init(&net->net_list);
3197                 lnet_net_free(net);
3198         }
3199         return rc;
3200 }
3201
3202 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3203 {
3204         struct lnet_net *net;
3205         struct lnet_ni *ni;
3206         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3207         int rc, i;
3208         __u32 net_id, lnd_type;
3209
3210         /* get the tunables if they are available */
3211         if (conf->lic_cfg_hdr.ioc_len >=
3212             sizeof(*conf) + sizeof(*tun))
3213                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3214                         conf->lic_bulk;
3215
3216         /* handle legacy ip2nets from DLC */
3217         if (conf->lic_legacy_ip2nets[0] != '\0')
3218                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3219                                                   tun);
3220
3221         net_id = LNET_NIDNET(conf->lic_nid);
3222         lnd_type = LNET_NETTYP(net_id);
3223
3224         if (!libcfs_isknown_lnd(lnd_type)) {
3225                 CERROR("No valid net and lnd information provided\n");
3226                 return -EINVAL;
3227         }
3228
3229         net = lnet_net_alloc(net_id, NULL);
3230         if (!net)
3231                 return -ENOMEM;
3232
3233         for (i = 0; i < conf->lic_ncpts; i++) {
3234                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3235                         return -EINVAL;
3236         }
3237
3238         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3239                                        conf->lic_ni_intf[0]);
3240         if (!ni)
3241                 return -ENOMEM;
3242
3243         mutex_lock(&the_lnet.ln_api_mutex);
3244
3245         rc = lnet_add_net_common(net, tun);
3246
3247         mutex_unlock(&the_lnet.ln_api_mutex);
3248
3249         return rc;
3250 }
3251
3252 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3253 {
3254         struct lnet_net  *net;
3255         struct lnet_ni *ni;
3256         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3257         struct lnet_ping_buffer *pbuf;
3258         struct lnet_handle_md  ping_mdh;
3259         int               rc;
3260         int               net_count;
3261         __u32             addr;
3262
3263         /* don't allow userspace to shutdown the LOLND */
3264         if (LNET_NETTYP(net_id) == LOLND)
3265                 return -EINVAL;
3266
3267         mutex_lock(&the_lnet.ln_api_mutex);
3268
3269         lnet_net_lock(0);
3270
3271         net = lnet_get_net_locked(net_id);
3272         if (!net) {
3273                 CERROR("net %s not found\n",
3274                        libcfs_net2str(net_id));
3275                 rc = -ENOENT;
3276                 goto unlock_net;
3277         }
3278
3279         addr = LNET_NIDADDR(conf->lic_nid);
3280         if (addr == 0) {
3281                 /* remove the entire net */
3282                 net_count = lnet_get_net_ni_count_locked(net);
3283
3284                 lnet_net_unlock(0);
3285
3286                 /* create and link a new ping info, before removing the old one */
3287                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3288                                         lnet_get_ni_count() - net_count,
3289                                         false);
3290                 if (rc != 0)
3291                         goto unlock_api_mutex;
3292
3293                 lnet_shutdown_lndnet(net);
3294
3295                 if (lnet_count_acceptor_nets() == 0)
3296                         lnet_acceptor_stop();
3297
3298                 lnet_ping_target_update(pbuf, ping_mdh);
3299
3300                 goto unlock_api_mutex;
3301         }
3302
3303         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3304         if (!ni) {
3305                 CERROR("nid %s not found\n",
3306                        libcfs_nid2str(conf->lic_nid));
3307                 rc = -ENOENT;
3308                 goto unlock_net;
3309         }
3310
3311         net_count = lnet_get_net_ni_count_locked(net);
3312
3313         lnet_net_unlock(0);
3314
3315         /* create and link a new ping info, before removing the old one */
3316         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3317                                   lnet_get_ni_count() - 1, false);
3318         if (rc != 0)
3319                 goto unlock_api_mutex;
3320
3321         lnet_shutdown_lndni(ni);
3322
3323         if (lnet_count_acceptor_nets() == 0)
3324                 lnet_acceptor_stop();
3325
3326         lnet_ping_target_update(pbuf, ping_mdh);
3327
3328         /* check if the net is empty and remove it if it is */
3329         if (net_count == 1)
3330                 lnet_shutdown_lndnet(net);
3331
3332         goto unlock_api_mutex;
3333
3334 unlock_net:
3335         lnet_net_unlock(0);
3336 unlock_api_mutex:
3337         mutex_unlock(&the_lnet.ln_api_mutex);
3338
3339         return rc;
3340 }
3341
3342 /*
3343  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3344  * They are only expected to be called for unique networks.
3345  * That can be as a result of older DLC library
3346  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3347  */
3348 int
3349 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3350 {
3351         struct lnet_net *net;
3352         LIST_HEAD(net_head);
3353         int rc;
3354         struct lnet_ioctl_config_lnd_tunables tun;
3355         char *nets = conf->cfg_config_u.cfg_net.net_intf;
3356
3357         /* Create a net/ni structures for the network string */
3358         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3359         if (rc <= 0)
3360                 return rc == 0 ? -EINVAL : rc;
3361
3362         mutex_lock(&the_lnet.ln_api_mutex);
3363
3364         if (rc > 1) {
3365                 rc = -EINVAL; /* only add one network per call */
3366                 goto out_unlock_clean;
3367         }
3368
3369         net = list_entry(net_head.next, struct lnet_net, net_list);
3370         list_del_init(&net->net_list);
3371
3372         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3373
3374         memset(&tun, 0, sizeof(tun));
3375
3376         tun.lt_cmn.lct_peer_timeout =
3377           conf->cfg_config_u.cfg_net.net_peer_timeout;
3378         tun.lt_cmn.lct_peer_tx_credits =
3379           conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3380         tun.lt_cmn.lct_peer_rtr_credits =
3381           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3382         tun.lt_cmn.lct_max_tx_credits =
3383           conf->cfg_config_u.cfg_net.net_max_tx_credits;
3384
3385         rc = lnet_add_net_common(net, &tun);
3386
3387 out_unlock_clean:
3388         mutex_unlock(&the_lnet.ln_api_mutex);
3389         while (!list_empty(&net_head)) {
3390                 /* net_head list is empty in success case */
3391                 net = list_entry(net_head.next, struct lnet_net, net_list);
3392                 list_del_init(&net->net_list);
3393                 lnet_net_free(net);
3394         }
3395         return rc;
3396 }
3397
3398 int
3399 lnet_dyn_del_net(__u32 net_id)
3400 {
3401         struct lnet_net  *net;
3402         struct lnet_ping_buffer *pbuf;
3403         struct lnet_handle_md ping_mdh;
3404         int               rc;
3405         int               net_ni_count;
3406
3407         /* don't allow userspace to shutdown the LOLND */
3408         if (LNET_NETTYP(net_id) == LOLND)
3409                 return -EINVAL;
3410
3411         mutex_lock(&the_lnet.ln_api_mutex);
3412
3413         lnet_net_lock(0);
3414
3415         net = lnet_get_net_locked(net_id);
3416         if (net == NULL) {
3417                 lnet_net_unlock(0);
3418                 rc = -EINVAL;
3419                 goto out;
3420         }
3421
3422         net_ni_count = lnet_get_net_ni_count_locked(net);
3423
3424         lnet_net_unlock(0);
3425
3426         /* create and link a new ping info, before removing the old one */
3427         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3428                                     lnet_get_ni_count() - net_ni_count, false);
3429         if (rc != 0)
3430                 goto out;
3431
3432         lnet_shutdown_lndnet(net);
3433
3434         if (lnet_count_acceptor_nets() == 0)
3435                 lnet_acceptor_stop();
3436
3437         lnet_ping_target_update(pbuf, ping_mdh);
3438
3439 out:
3440         mutex_unlock(&the_lnet.ln_api_mutex);
3441
3442         return rc;
3443 }
3444
3445 void lnet_incr_dlc_seq(void)
3446 {
3447         atomic_inc(&lnet_dlc_seq_no);
3448 }
3449
3450 __u32 lnet_get_dlc_seq_locked(void)
3451 {
3452         return atomic_read(&lnet_dlc_seq_no);
3453 }
3454
3455 static void
3456 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3457 {
3458         struct lnet_net *net;
3459         struct lnet_ni *ni;
3460
3461         lnet_net_lock(LNET_LOCK_EX);
3462         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3463                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3464                         if (ni->ni_nid == nid || all) {
3465                                 atomic_set(&ni->ni_healthv, value);
3466                                 if (list_empty(&ni->ni_recovery) &&
3467                                     value < LNET_MAX_HEALTH_VALUE) {
3468                                         CERROR("manually adding local NI %s to recovery\n",
3469                                                libcfs_nid2str(ni->ni_nid));
3470                                         list_add_tail(&ni->ni_recovery,
3471                                                       &the_lnet.ln_mt_localNIRecovq);
3472                                         lnet_ni_addref_locked(ni, 0);
3473                                 }
3474                                 if (!all) {
3475                                         lnet_net_unlock(LNET_LOCK_EX);
3476                                         return;
3477                                 }
3478                         }
3479                 }
3480         }
3481         lnet_net_unlock(LNET_LOCK_EX);
3482 }
3483
3484 static int
3485 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3486 {
3487         int cpt, rc = 0;
3488         struct lnet_ni *ni;
3489         lnet_nid_t nid = stats->hlni_nid;
3490
3491         cpt = lnet_net_lock_current();
3492         ni = lnet_nid2ni_locked(nid, cpt);
3493
3494         if (!ni) {
3495                 rc = -ENOENT;
3496                 goto unlock;
3497         }
3498
3499         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3500         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3501         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3502         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3503         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3504         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3505         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3506
3507 unlock:
3508         lnet_net_unlock(cpt);
3509
3510         return rc;
3511 }
3512
3513 static int
3514 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3515 {
3516         struct lnet_ni *ni;
3517         int i = 0;
3518