projects / deliverable / linux.git / blob
? search:


06046b216c93380552d6026de0725717bb69dc58
[deliverable/linux.git] / drivers / staging / lustre / 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.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2015, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37 #define DEBUG_SUBSYSTEM S_LNET
38 #include <linux/log2.h>
39 #include <linux/ktime.h>
40
41 #include "../../include/linux/lnet/lib-lnet.h"
42
43 #define D_LNI D_CONSOLE
44
45 lnet_t the_lnet; /* THE state of the network */
46 EXPORT_SYMBOL(the_lnet);
47
48 static char *ip2nets = "";
49 module_param(ip2nets, charp, 0444);
50 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
51
52 static char *networks = "";
53 module_param(networks, charp, 0444);
54 MODULE_PARM_DESC(networks, "local networks");
55
56 static char *routes = "";
57 module_param(routes, charp, 0444);
58 MODULE_PARM_DESC(routes, "routes to non-local networks");
59
60 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
61 module_param(rnet_htable_size, int, 0444);
62 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
63
64 static int lnet_ping(lnet_process_id_t id, int timeout_ms,
65 lnet_process_id_t __user *ids, int n_ids);
66
67 static char *
68 lnet_get_routes(void)
69 {
70 return routes;
71 }
72
73 static char *
74 lnet_get_networks(void)
75 {
76 char *nets;
77 int rc;
78
79 if (*networks && *ip2nets) {
80 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or 'ip2nets' but not both at once\n");
81 return NULL;
82 }
83
84 if (*ip2nets) {
85 rc = lnet_parse_ip2nets(&nets, ip2nets);
86 return !rc ? nets : NULL;
87 }
88
89 if (*networks)
90 return networks;
91
92 return "tcp";
93 }
94
95 static void
96 lnet_init_locks(void)
97 {
98 spin_lock_init(&the_lnet.ln_eq_wait_lock);
99 init_waitqueue_head(&the_lnet.ln_eq_waitq);
100 mutex_init(&the_lnet.ln_lnd_mutex);
101 mutex_init(&the_lnet.ln_api_mutex);
102 }
103
104 static int
105 lnet_create_remote_nets_table(void)
106 {
107 int i;
108 struct list_head *hash;
109
110 LASSERT(!the_lnet.ln_remote_nets_hash);
111 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
112 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
113 if (!hash) {
114 CERROR("Failed to create remote nets hash table\n");
115 return -ENOMEM;
116 }
117
118 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
119 INIT_LIST_HEAD(&hash[i]);
120 the_lnet.ln_remote_nets_hash = hash;
121 return 0;
122 }
123
124 static void
125 lnet_destroy_remote_nets_table(void)
126 {
127 int i;
128
129 if (!the_lnet.ln_remote_nets_hash)
130 return;
131
132 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
133 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
134
135 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
136 LNET_REMOTE_NETS_HASH_SIZE *
137 sizeof(the_lnet.ln_remote_nets_hash[0]));
138 the_lnet.ln_remote_nets_hash = NULL;
139 }
140
141 static void
142 lnet_destroy_locks(void)
143 {
144 if (the_lnet.ln_res_lock) {
145 cfs_percpt_lock_free(the_lnet.ln_res_lock);
146 the_lnet.ln_res_lock = NULL;
147 }
148
149 if (the_lnet.ln_net_lock) {
150 cfs_percpt_lock_free(the_lnet.ln_net_lock);
151 the_lnet.ln_net_lock = NULL;
152 }
153 }
154
155 static int
156 lnet_create_locks(void)
157 {
158 lnet_init_locks();
159
160 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
161 if (!the_lnet.ln_res_lock)
162 goto failed;
163
164 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
165 if (!the_lnet.ln_net_lock)
166 goto failed;
167
168 return 0;
169
170 failed:
171 lnet_destroy_locks();
172 return -ENOMEM;
173 }
174
175 static void lnet_assert_wire_constants(void)
176 {
177 /*
178 * Wire protocol assertions generated by 'wirecheck'
179 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
180 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
181 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
182 */
183
184 /* Constants... */
185 CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
186 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
187 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
188 CLASSERT(LNET_MSG_ACK == 0);
189 CLASSERT(LNET_MSG_PUT == 1);
190 CLASSERT(LNET_MSG_GET == 2);
191 CLASSERT(LNET_MSG_REPLY == 3);
192 CLASSERT(LNET_MSG_HELLO == 4);
193
194 /* Checks for struct ptl_handle_wire_t */
195 CLASSERT((int)sizeof(lnet_handle_wire_t) == 16);
196 CLASSERT((int)offsetof(lnet_handle_wire_t, wh_interface_cookie) == 0);
197 CLASSERT((int)sizeof(((lnet_handle_wire_t *)0)->wh_interface_cookie) == 8);
198 CLASSERT((int)offsetof(lnet_handle_wire_t, wh_object_cookie) == 8);
199 CLASSERT((int)sizeof(((lnet_handle_wire_t *)0)->wh_object_cookie) == 8);
200
201 /* Checks for struct lnet_magicversion_t */
202 CLASSERT((int)sizeof(lnet_magicversion_t) == 8);
203 CLASSERT((int)offsetof(lnet_magicversion_t, magic) == 0);
204 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->magic) == 4);
205 CLASSERT((int)offsetof(lnet_magicversion_t, version_major) == 4);
206 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_major) == 2);
207 CLASSERT((int)offsetof(lnet_magicversion_t, version_minor) == 6);
208 CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_minor) == 2);
209
210 /* Checks for struct lnet_hdr_t */
211 CLASSERT((int)sizeof(lnet_hdr_t) == 72);
212 CLASSERT((int)offsetof(lnet_hdr_t, dest_nid) == 0);
213 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_nid) == 8);
214 CLASSERT((int)offsetof(lnet_hdr_t, src_nid) == 8);
215 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_nid) == 8);
216 CLASSERT((int)offsetof(lnet_hdr_t, dest_pid) == 16);
217 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_pid) == 4);
218 CLASSERT((int)offsetof(lnet_hdr_t, src_pid) == 20);
219 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_pid) == 4);
220 CLASSERT((int)offsetof(lnet_hdr_t, type) == 24);
221 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->type) == 4);
222 CLASSERT((int)offsetof(lnet_hdr_t, payload_length) == 28);
223 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->payload_length) == 4);
224 CLASSERT((int)offsetof(lnet_hdr_t, msg) == 32);
225 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg) == 40);
226
227 /* Ack */
228 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.dst_wmd) == 32);
229 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.dst_wmd) == 16);
230 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.match_bits) == 48);
231 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.match_bits) == 8);
232 CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.mlength) == 56);
233 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.mlength) == 4);
234
235 /* Put */
236 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ack_wmd) == 32);
237 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ack_wmd) == 16);
238 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.match_bits) == 48);
239 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.match_bits) == 8);
240 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.hdr_data) == 56);
241 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.hdr_data) == 8);
242 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ptl_index) == 64);
243 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ptl_index) == 4);
244 CLASSERT((int)offsetof(lnet_hdr_t, msg.put.offset) == 68);
245 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.offset) == 4);
246
247 /* Get */
248 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.return_wmd) == 32);
249 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.return_wmd) == 16);
250 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.match_bits) == 48);
251 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.match_bits) == 8);
252 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.ptl_index) == 56);
253 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.ptl_index) == 4);
254 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.src_offset) == 60);
255 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.src_offset) == 4);
256 CLASSERT((int)offsetof(lnet_hdr_t, msg.get.sink_length) == 64);
257 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.sink_length) == 4);
258
259 /* Reply */
260 CLASSERT((int)offsetof(lnet_hdr_t, msg.reply.dst_wmd) == 32);
261 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.reply.dst_wmd) == 16);
262
263 /* Hello */
264 CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.incarnation) == 32);
265 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.incarnation) == 8);
266 CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.type) == 40);
267 CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.type) == 4);
268 }
269
270 static lnd_t *
271 lnet_find_lnd_by_type(__u32 type)
272 {
273 lnd_t *lnd;
274 struct list_head *tmp;
275
276 /* holding lnd mutex */
277 list_for_each(tmp, &the_lnet.ln_lnds) {
278 lnd = list_entry(tmp, lnd_t, lnd_list);
279
280 if (lnd->lnd_type == type)
281 return lnd;
282 }
283
284 return NULL;
285 }
286
287 void
288 lnet_register_lnd(lnd_t *lnd)
289 {
290 mutex_lock(&the_lnet.ln_lnd_mutex);
291
292 LASSERT(the_lnet.ln_init);
293 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
294 LASSERT(!lnet_find_lnd_by_type(lnd->lnd_type));
295
296 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
297 lnd->lnd_refcount = 0;
298
299 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
300
301 mutex_unlock(&the_lnet.ln_lnd_mutex);
302 }
303 EXPORT_SYMBOL(lnet_register_lnd);
304
305 void
306 lnet_unregister_lnd(lnd_t *lnd)
307 {
308 mutex_lock(&the_lnet.ln_lnd_mutex);
309
310 LASSERT(the_lnet.ln_init);
311 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
312 LASSERT(!lnd->lnd_refcount);
313
314 list_del(&lnd->lnd_list);
315 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
316
317 mutex_unlock(&the_lnet.ln_lnd_mutex);
318 }
319 EXPORT_SYMBOL(lnet_unregister_lnd);
320
321 void
322 lnet_counters_get(lnet_counters_t *counters)
323 {
324 lnet_counters_t *ctr;
325 int i;
326
327 memset(counters, 0, sizeof(*counters));
328
329 lnet_net_lock(LNET_LOCK_EX);
330
331 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
332 counters->msgs_max += ctr->msgs_max;
333 counters->msgs_alloc += ctr->msgs_alloc;
334 counters->errors += ctr->errors;
335 counters->send_count += ctr->send_count;
336 counters->recv_count += ctr->recv_count;
337 counters->route_count += ctr->route_count;
338 counters->drop_count += ctr->drop_count;
339 counters->send_length += ctr->send_length;
340 counters->recv_length += ctr->recv_length;
341 counters->route_length += ctr->route_length;
342 counters->drop_length += ctr->drop_length;
343 }
344 lnet_net_unlock(LNET_LOCK_EX);
345 }
346 EXPORT_SYMBOL(lnet_counters_get);
347
348 void
349 lnet_counters_reset(void)
350 {
351 lnet_counters_t *counters;
352 int i;
353
354 lnet_net_lock(LNET_LOCK_EX);
355
356 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
357 memset(counters, 0, sizeof(lnet_counters_t));
358
359 lnet_net_unlock(LNET_LOCK_EX);
360 }
361
362 static char *
363 lnet_res_type2str(int type)
364 {
365 switch (type) {
366 default:
367 LBUG();
368 case LNET_COOKIE_TYPE_MD:
369 return "MD";
370 case LNET_COOKIE_TYPE_ME:
371 return "ME";
372 case LNET_COOKIE_TYPE_EQ:
373 return "EQ";
374 }
375 }
376
377 static void
378 lnet_res_container_cleanup(struct lnet_res_container *rec)
379 {
380 int count = 0;
381
382 if (!rec->rec_type) /* not set yet, it's uninitialized */
383 return;
384
385 while (!list_empty(&rec->rec_active)) {
386 struct list_head *e = rec->rec_active.next;
387
388 list_del_init(e);
389 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
390 lnet_eq_free(list_entry(e, lnet_eq_t, eq_list));
391
392 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
393 lnet_md_free(list_entry(e, lnet_libmd_t, md_list));
394
395 } else { /* NB: Active MEs should be attached on portals */
396 LBUG();
397 }
398 count++;
399 }
400
401 if (count > 0) {
402 /*
403 * Found alive MD/ME/EQ, user really should unlink/free
404 * all of them before finalize LNet, but if someone didn't,
405 * we have to recycle garbage for him
406 */
407 CERROR("%d active elements on exit of %s container\n",
408 count, lnet_res_type2str(rec->rec_type));
409 }
410
411 if (rec->rec_lh_hash) {
412 LIBCFS_FREE(rec->rec_lh_hash,
413 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
414 rec->rec_lh_hash = NULL;
415 }
416
417 rec->rec_type = 0; /* mark it as finalized */
418 }
419
420 static int
421 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
422 {
423 int rc = 0;
424 int i;
425
426 LASSERT(!rec->rec_type);
427
428 rec->rec_type = type;
429 INIT_LIST_HEAD(&rec->rec_active);
430 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
431
432 /* Arbitrary choice of hash table size */
433 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
434 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
435 if (!rec->rec_lh_hash) {
436 rc = -ENOMEM;
437 goto out;
438 }
439
440 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
441 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
442
443 return 0;
444
445 out:
446 CERROR("Failed to setup %s resource container\n",
447 lnet_res_type2str(type));
448 lnet_res_container_cleanup(rec);
449 return rc;
450 }
451
452 static void
453 lnet_res_containers_destroy(struct lnet_res_container **recs)
454 {
455 struct lnet_res_container *rec;
456 int i;
457
458 cfs_percpt_for_each(rec, i, recs)
459 lnet_res_container_cleanup(rec);
460
461 cfs_percpt_free(recs);
462 }
463
464 static struct lnet_res_container **
465 lnet_res_containers_create(int type)
466 {
467 struct lnet_res_container **recs;
468 struct lnet_res_container *rec;
469 int rc;
470 int i;
471
472 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
473 if (!recs) {
474 CERROR("Failed to allocate %s resource containers\n",
475 lnet_res_type2str(type));
476 return NULL;
477 }
478
479 cfs_percpt_for_each(rec, i, recs) {
480 rc = lnet_res_container_setup(rec, i, type);
481 if (rc) {
482 lnet_res_containers_destroy(recs);
483 return NULL;
484 }
485 }
486
487 return recs;
488 }
489
490 lnet_libhandle_t *
491 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
492 {
493 /* ALWAYS called with lnet_res_lock held */
494 struct list_head *head;
495 lnet_libhandle_t *lh;
496 unsigned int hash;
497
498 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
499 return NULL;
500
501 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
502 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
503
504 list_for_each_entry(lh, head, lh_hash_chain) {
505 if (lh->lh_cookie == cookie)
506 return lh;
507 }
508
509 return NULL;
510 }
511
512 void
513 lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh)
514 {
515 /* ALWAYS called with lnet_res_lock held */
516 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
517 unsigned int hash;
518
519 lh->lh_cookie = rec->rec_lh_cookie;
520 rec->rec_lh_cookie += 1 << ibits;
521
522 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
523
524 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
525 }
526
527 int lnet_unprepare(void);
528
529 static int
530 lnet_prepare(lnet_pid_t requested_pid)
531 {
532 /* Prepare to bring up the network */
533 struct lnet_res_container **recs;
534 int rc = 0;
535
536 LASSERT(!the_lnet.ln_refcount);
537
538 the_lnet.ln_routing = 0;
539
540 LASSERT(!(requested_pid & LNET_PID_USERFLAG));
541 the_lnet.ln_pid = requested_pid;
542
543 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
544 INIT_LIST_HEAD(&the_lnet.ln_nis);
545 INIT_LIST_HEAD(&the_lnet.ln_nis_cpt);
546 INIT_LIST_HEAD(&the_lnet.ln_nis_zombie);
547 INIT_LIST_HEAD(&the_lnet.ln_routers);
548
549 rc = lnet_create_remote_nets_table();
550 if (rc)
551 goto failed;
552 /*
553 * NB the interface cookie in wire handles guards against delayed
554 * replies and ACKs appearing valid after reboot.
555 */
556 the_lnet.ln_interface_cookie = ktime_get_ns();
557
558 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
559 sizeof(lnet_counters_t));
560 if (!the_lnet.ln_counters) {
561 CERROR("Failed to allocate counters for LNet\n");
562 rc = -ENOMEM;
563 goto failed;
564 }
565
566 rc = lnet_peer_tables_create();
567 if (rc)
568 goto failed;
569
570 rc = lnet_msg_containers_create();
571 if (rc)
572 goto failed;
573
574 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
575 LNET_COOKIE_TYPE_EQ);
576 if (rc)
577 goto failed;
578
579 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
580 if (!recs) {
581 rc = -ENOMEM;
582 goto failed;
583 }
584
585 the_lnet.ln_me_containers = recs;
586
587 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
588 if (!recs) {
589 rc = -ENOMEM;
590 goto failed;
591 }
592
593 the_lnet.ln_md_containers = recs;
594
595 rc = lnet_portals_create();
596 if (rc) {
597 CERROR("Failed to create portals for LNet: %d\n", rc);
598 goto failed;
599 }
600
601 return 0;
602
603 failed:
604 lnet_unprepare();
605 return rc;
606 }
607
608 int
609 lnet_unprepare(void)
610 {
611 /*
612 * NB no LNET_LOCK since this is the last reference. All LND instances
613 * have shut down already, so it is safe to unlink and free all
614 * descriptors, even those that appear committed to a network op (eg MD
615 * with non-zero pending count)
616 */
617 lnet_fail_nid(LNET_NID_ANY, 0);
618
619 LASSERT(!the_lnet.ln_refcount);
620 LASSERT(list_empty(&the_lnet.ln_test_peers));
621 LASSERT(list_empty(&the_lnet.ln_nis));
622 LASSERT(list_empty(&the_lnet.ln_nis_cpt));
623 LASSERT(list_empty(&the_lnet.ln_nis_zombie));
624
625 lnet_portals_destroy();
626
627 if (the_lnet.ln_md_containers) {
628 lnet_res_containers_destroy(the_lnet.ln_md_containers);
629 the_lnet.ln_md_containers = NULL;
630 }
631
632 if (the_lnet.ln_me_containers) {
633 lnet_res_containers_destroy(the_lnet.ln_me_containers);
634 the_lnet.ln_me_containers = NULL;
635 }
636
637 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
638
639 lnet_msg_containers_destroy();
640 lnet_peer_tables_destroy();
641 lnet_rtrpools_free(0);
642
643 if (the_lnet.ln_counters) {
644 cfs_percpt_free(the_lnet.ln_counters);
645 the_lnet.ln_counters = NULL;
646 }
647 lnet_destroy_remote_nets_table();
648
649 return 0;
650 }
651
652 lnet_ni_t *
653 lnet_net2ni_locked(__u32 net, int cpt)
654 {
655 struct list_head *tmp;
656 lnet_ni_t *ni;
657
658 LASSERT(cpt != LNET_LOCK_EX);
659
660 list_for_each(tmp, &the_lnet.ln_nis) {
661 ni = list_entry(tmp, lnet_ni_t, ni_list);
662
663 if (LNET_NIDNET(ni->ni_nid) == net) {
664 lnet_ni_addref_locked(ni, cpt);
665 return ni;
666 }
667 }
668
669 return NULL;
670 }
671
672 lnet_ni_t *
673 lnet_net2ni(__u32 net)
674 {
675 lnet_ni_t *ni;
676
677 lnet_net_lock(0);
678 ni = lnet_net2ni_locked(net, 0);
679 lnet_net_unlock(0);
680
681 return ni;
682 }
683 EXPORT_SYMBOL(lnet_net2ni);
684
685 static unsigned int
686 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
687 {
688 __u64 key = nid;
689 unsigned int val;
690
691 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
692
693 if (number == 1)
694 return 0;
695
696 val = hash_long(key, LNET_CPT_BITS);
697 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
698 if (val < number)
699 return val;
700
701 return (unsigned int)(key + val + (val >> 1)) % number;
702 }
703
704 int
705 lnet_cpt_of_nid_locked(lnet_nid_t nid)
706 {
707 struct lnet_ni *ni;
708
709 /* must called with hold of lnet_net_lock */
710 if (LNET_CPT_NUMBER == 1)
711 return 0; /* the only one */
712
713 /* take lnet_net_lock(any) would be OK */
714 if (!list_empty(&the_lnet.ln_nis_cpt)) {
715 list_for_each_entry(ni, &the_lnet.ln_nis_cpt, ni_cptlist) {
716 if (LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid))
717 continue;
718
719 LASSERT(ni->ni_cpts);
720 return ni->ni_cpts[lnet_nid_cpt_hash
721 (nid, ni->ni_ncpts)];
722 }
723 }
724
725 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
726 }
727
728 int
729 lnet_cpt_of_nid(lnet_nid_t nid)
730 {
731 int cpt;
732 int cpt2;
733
734 if (LNET_CPT_NUMBER == 1)
735 return 0; /* the only one */
736
737 if (list_empty(&the_lnet.ln_nis_cpt))
738 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
739
740 cpt = lnet_net_lock_current();
741 cpt2 = lnet_cpt_of_nid_locked(nid);
742 lnet_net_unlock(cpt);
743
744 return cpt2;
745 }
746 EXPORT_SYMBOL(lnet_cpt_of_nid);
747
748 int
749 lnet_islocalnet(__u32 net)
750 {
751 struct lnet_ni *ni;
752 int cpt;
753
754 cpt = lnet_net_lock_current();
755
756 ni = lnet_net2ni_locked(net, cpt);
757 if (ni)
758 lnet_ni_decref_locked(ni, cpt);
759
760 lnet_net_unlock(cpt);
761
762 return !!ni;
763 }
764
765 lnet_ni_t *
766 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
767 {
768 struct lnet_ni *ni;
769 struct list_head *tmp;
770
771 LASSERT(cpt != LNET_LOCK_EX);
772
773 list_for_each(tmp, &the_lnet.ln_nis) {
774 ni = list_entry(tmp, lnet_ni_t, ni_list);
775
776 if (ni->ni_nid == nid) {
777 lnet_ni_addref_locked(ni, cpt);
778 return ni;
779 }
780 }
781
782 return NULL;
783 }
784
785 int
786 lnet_islocalnid(lnet_nid_t nid)
787 {
788 struct lnet_ni *ni;
789 int cpt;
790
791 cpt = lnet_net_lock_current();
792 ni = lnet_nid2ni_locked(nid, cpt);
793 if (ni)
794 lnet_ni_decref_locked(ni, cpt);
795 lnet_net_unlock(cpt);
796
797 return !!ni;
798 }
799
800 int
801 lnet_count_acceptor_nis(void)
802 {
803 /* Return the # of NIs that need the acceptor. */
804 int count = 0;
805 struct list_head *tmp;
806 struct lnet_ni *ni;
807 int cpt;
808
809 cpt = lnet_net_lock_current();
810 list_for_each(tmp, &the_lnet.ln_nis) {
811 ni = list_entry(tmp, lnet_ni_t, ni_list);
812
813 if (ni->ni_lnd->lnd_accept)
814 count++;
815 }
816
817 lnet_net_unlock(cpt);
818
819 return count;
820 }
821
822 static int
823 lnet_ni_tq_credits(lnet_ni_t *ni)
824 {
825 int credits;
826
827 LASSERT(ni->ni_ncpts >= 1);
828
829 if (ni->ni_ncpts == 1)
830 return ni->ni_maxtxcredits;
831
832 credits = ni->ni_maxtxcredits / ni->ni_ncpts;
833 credits = max(credits, 8 * ni->ni_peertxcredits);
834 credits = min(credits, ni->ni_maxtxcredits);
835
836 return credits;
837 }
838
839 static void
840 lnet_shutdown_lndnis(void)
841 {
842 int i;
843 int islo;
844 lnet_ni_t *ni;
845
846 /* NB called holding the global mutex */
847
848 /* All quiet on the API front */
849 LASSERT(!the_lnet.ln_shutdown);
850 LASSERT(!the_lnet.ln_refcount);
851 LASSERT(list_empty(&the_lnet.ln_nis_zombie));
852
853 lnet_net_lock(LNET_LOCK_EX);
854 the_lnet.ln_shutdown = 1; /* flag shutdown */
855
856 /* Unlink NIs from the global table */
857 while (!list_empty(&the_lnet.ln_nis)) {
858 ni = list_entry(the_lnet.ln_nis.next,
859 lnet_ni_t, ni_list);
860 /* move it to zombie list and nobody can find it anymore */
861 list_move(&ni->ni_list, &the_lnet.ln_nis_zombie);
862 lnet_ni_decref_locked(ni, 0); /* drop ln_nis' ref */
863
864 if (!list_empty(&ni->ni_cptlist)) {
865 list_del_init(&ni->ni_cptlist);
866 lnet_ni_decref_locked(ni, 0);
867 }
868 }
869
870 /* Drop the cached eqwait NI. */
871 if (the_lnet.ln_eq_waitni) {
872 lnet_ni_decref_locked(the_lnet.ln_eq_waitni, 0);
873 the_lnet.ln_eq_waitni = NULL;
874 }
875
876 /* Drop the cached loopback NI. */
877 if (the_lnet.ln_loni) {
878 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
879 the_lnet.ln_loni = NULL;
880 }
881
882 lnet_net_unlock(LNET_LOCK_EX);
883
884 /*
885 * Clear lazy portals and drop delayed messages which hold refs
886 * on their lnet_msg_t::msg_rxpeer
887 */
888 for (i = 0; i < the_lnet.ln_nportals; i++)
889 LNetClearLazyPortal(i);
890
891 /*
892 * Clear the peer table and wait for all peers to go (they hold refs on
893 * their NIs)
894 */
895 lnet_peer_tables_cleanup(NULL);
896
897 lnet_net_lock(LNET_LOCK_EX);
898 /*
899 * Now wait for the NI's I just nuked to show up on ln_zombie_nis
900 * and shut them down in guaranteed thread context
901 */
902 i = 2;
903 while (!list_empty(&the_lnet.ln_nis_zombie)) {
904 int *ref;
905 int j;
906
907 ni = list_entry(the_lnet.ln_nis_zombie.next,
908 lnet_ni_t, ni_list);
909 list_del_init(&ni->ni_list);
910 cfs_percpt_for_each(ref, j, ni->ni_refs) {
911 if (!*ref)
912 continue;
913 /* still busy, add it back to zombie list */
914 list_add(&ni->ni_list, &the_lnet.ln_nis_zombie);
915 break;
916 }
917
918 if (!list_empty(&ni->ni_list)) {
919 lnet_net_unlock(LNET_LOCK_EX);
920 ++i;
921 if ((i & (-i)) == i) {
922 CDEBUG(D_WARNING, "Waiting for zombie LNI %s\n",
923 libcfs_nid2str(ni->ni_nid));
924 }
925 set_current_state(TASK_UNINTERRUPTIBLE);
926 schedule_timeout(cfs_time_seconds(1));
927 lnet_net_lock(LNET_LOCK_EX);
928 continue;
929 }
930
931 ni->ni_lnd->lnd_refcount--;
932 lnet_net_unlock(LNET_LOCK_EX);
933
934 islo = ni->ni_lnd->lnd_type == LOLND;
935
936 LASSERT(!in_interrupt());
937 ni->ni_lnd->lnd_shutdown(ni);
938
939 /*
940 * can't deref lnd anymore now; it might have unregistered
941 * itself...
942 */
943 if (!islo)
944 CDEBUG(D_LNI, "Removed LNI %s\n",
945 libcfs_nid2str(ni->ni_nid));
946
947 lnet_ni_free(ni);
948 i = 2;
949
950 lnet_net_lock(LNET_LOCK_EX);
951 }
952
953 the_lnet.ln_shutdown = 0;
954 lnet_net_unlock(LNET_LOCK_EX);
955 }
956
957 static int
958 lnet_startup_lndnis(void)
959 {
960 lnd_t *lnd;
961 struct lnet_ni *ni;
962 struct lnet_tx_queue *tq;
963 struct list_head nilist;
964 int i;
965 int rc = 0;
966 __u32 lnd_type;
967 int nicount = 0;
968 char *nets = lnet_get_networks();
969
970 INIT_LIST_HEAD(&nilist);
971
972 if (!nets)
973 goto failed;
974
975 rc = lnet_parse_networks(&nilist, nets);
976 if (rc)
977 goto failed;
978
979 while (!list_empty(&nilist)) {
980 ni = list_entry(nilist.next, lnet_ni_t, ni_list);
981 lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));
982
983 LASSERT(libcfs_isknown_lnd(lnd_type));
984
985 if (lnd_type == CIBLND ||
986 lnd_type == OPENIBLND ||
987 lnd_type == IIBLND ||
988 lnd_type == VIBLND) {
989 CERROR("LND %s obsoleted\n",
990 libcfs_lnd2str(lnd_type));
991 goto failed;
992 }
993
994 mutex_lock(&the_lnet.ln_lnd_mutex);
995 lnd = lnet_find_lnd_by_type(lnd_type);
996
997 if (!lnd) {
998 mutex_unlock(&the_lnet.ln_lnd_mutex);
999 rc = request_module("%s",
1000 libcfs_lnd2modname(lnd_type));
1001 mutex_lock(&the_lnet.ln_lnd_mutex);
1002
1003 lnd = lnet_find_lnd_by_type(lnd_type);
1004 if (!lnd) {
1005 mutex_unlock(&the_lnet.ln_lnd_mutex);
1006 CERROR("Can't load LND %s, module %s, rc=%d\n",
1007 libcfs_lnd2str(lnd_type),
1008 libcfs_lnd2modname(lnd_type), rc);
1009 goto failed;
1010 }
1011 }
1012
1013 lnet_net_lock(LNET_LOCK_EX);
1014 lnd->lnd_refcount++;
1015 lnet_net_unlock(LNET_LOCK_EX);
1016
1017 ni->ni_lnd = lnd;
1018
1019 rc = lnd->lnd_startup(ni);
1020
1021 mutex_unlock(&the_lnet.ln_lnd_mutex);
1022
1023 if (rc) {
1024 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1025 rc, libcfs_lnd2str(lnd->lnd_type));
1026 lnet_net_lock(LNET_LOCK_EX);
1027 lnd->lnd_refcount--;
1028 lnet_net_unlock(LNET_LOCK_EX);
1029 goto failed;
1030 }
1031
1032 LASSERT(ni->ni_peertimeout <= 0 || lnd->lnd_query);
1033
1034 list_del(&ni->ni_list);
1035
1036 lnet_net_lock(LNET_LOCK_EX);
1037 /* refcount for ln_nis */
1038 lnet_ni_addref_locked(ni, 0);
1039 list_add_tail(&ni->ni_list, &the_lnet.ln_nis);
1040 if (ni->ni_cpts) {
1041 list_add_tail(&ni->ni_cptlist,
1042 &the_lnet.ln_nis_cpt);
1043 lnet_ni_addref_locked(ni, 0);
1044 }
1045
1046 lnet_net_unlock(LNET_LOCK_EX);
1047
1048 if (lnd->lnd_type == LOLND) {
1049 lnet_ni_addref(ni);
1050 LASSERT(!the_lnet.ln_loni);
1051 the_lnet.ln_loni = ni;
1052 continue;
1053 }
1054
1055 if (!ni->ni_peertxcredits || !ni->ni_maxtxcredits) {
1056 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1057 libcfs_lnd2str(lnd->lnd_type),
1058 !ni->ni_peertxcredits ?
1059 "" : "per-peer ");
1060 goto failed;
1061 }
1062
1063 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
1064 tq->tq_credits_min =
1065 tq->tq_credits_max =
1066 tq->tq_credits = lnet_ni_tq_credits(ni);
1067 }
1068
1069 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
1070 libcfs_nid2str(ni->ni_nid), ni->ni_peertxcredits,
1071 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
1072 ni->ni_peerrtrcredits, ni->ni_peertimeout);
1073
1074 nicount++;
1075 }
1076
1077 if (the_lnet.ln_eq_waitni && nicount > 1) {
1078 lnd_type = the_lnet.ln_eq_waitni->ni_lnd->lnd_type;
1079 LCONSOLE_ERROR_MSG(0x109, "LND %s can only run single-network\n",
1080 libcfs_lnd2str(lnd_type));
1081 goto failed;
1082 }
1083
1084 return 0;
1085
1086 failed:
1087 lnet_shutdown_lndnis();
1088
1089 while (!list_empty(&nilist)) {
1090 ni = list_entry(nilist.next, lnet_ni_t, ni_list);
1091 list_del(&ni->ni_list);
1092 lnet_ni_free(ni);
1093 }
1094
1095 return -ENETDOWN;
1096 }
1097
1098 /**
1099 * Initialize LNet library.
1100 *
1101 * Only userspace program needs to call this function - it's automatically
1102 * called in the kernel at module loading time. Caller has to call lnet_fini()
1103 * after a call to lnet_init(), if and only if the latter returned 0. It must
1104 * be called exactly once.
1105 *
1106 * \return 0 on success, and -ve on failures.
1107 */
1108 int
1109 lnet_init(void)
1110 {
1111 int rc;
1112
1113 lnet_assert_wire_constants();
1114 LASSERT(!the_lnet.ln_init);
1115
1116 memset(&the_lnet, 0, sizeof(the_lnet));
1117
1118 /* refer to global cfs_cpt_table for now */
1119 the_lnet.ln_cpt_table = cfs_cpt_table;
1120 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
1121
1122 LASSERT(the_lnet.ln_cpt_number > 0);
1123 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
1124 /* we are under risk of consuming all lh_cookie */
1125 CERROR("Can't have %d CPTs for LNet (max allowed is %d), please change setting of CPT-table and retry\n",
1126 the_lnet.ln_cpt_number, LNET_CPT_MAX);
1127 return -1;
1128 }
1129
1130 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
1131 the_lnet.ln_cpt_bits++;
1132
1133 rc = lnet_create_locks();
1134 if (rc) {
1135 CERROR("Can't create LNet global locks: %d\n", rc);
1136 return -1;
1137 }
1138
1139 the_lnet.ln_refcount = 0;
1140 the_lnet.ln_init = 1;
1141 LNetInvalidateHandle(&the_lnet.ln_rc_eqh);
1142 INIT_LIST_HEAD(&the_lnet.ln_lnds);
1143 INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
1144 INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
1145
1146 /*
1147 * The hash table size is the number of bits it takes to express the set
1148 * ln_num_routes, minus 1 (better to under estimate than over so we
1149 * don't waste memory).
1150 */
1151 if (rnet_htable_size <= 0)
1152 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
1153 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
1154 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
1155 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
1156 order_base_2(rnet_htable_size) - 1);
1157
1158 /*
1159 * All LNDs apart from the LOLND are in separate modules. They
1160 * register themselves when their module loads, and unregister
1161 * themselves when their module is unloaded.
1162 */
1163 lnet_register_lnd(&the_lolnd);
1164 return 0;
1165 }
1166
1167 /**
1168 * Finalize LNet library.
1169 *
1170 * Only userspace program needs to call this function. It can be called
1171 * at most once.
1172 *
1173 * \pre lnet_init() called with success.
1174 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1175 */
1176 void
1177 lnet_fini(void)
1178 {
1179 LASSERT(the_lnet.ln_init);
1180 LASSERT(!the_lnet.ln_refcount);
1181
1182 while (!list_empty(&the_lnet.ln_lnds))
1183 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
1184 lnd_t, lnd_list));
1185 lnet_destroy_locks();
1186
1187 the_lnet.ln_init = 0;
1188 }
1189
1190 /**
1191 * Set LNet PID and start LNet interfaces, routing, and forwarding.
1192 *
1193 * Userspace program should call this after a successful call to lnet_init().
1194 * Users must call this function at least once before any other functions.
1195 * For each successful call there must be a corresponding call to
1196 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1197 * ignored.
1198 *
1199 * The PID used by LNet may be different from the one requested.
1200 * See LNetGetId().
1201 *
1202 * \param requested_pid PID requested by the caller.
1203 *
1204 * \return >= 0 on success, and < 0 error code on failures.
1205 */
1206 int
1207 LNetNIInit(lnet_pid_t requested_pid)
1208 {
1209 int im_a_router = 0;
1210 int rc;
1211
1212 mutex_lock(&the_lnet.ln_api_mutex);
1213
1214 LASSERT(the_lnet.ln_init);
1215 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
1216
1217 if (the_lnet.ln_refcount > 0) {
1218 rc = the_lnet.ln_refcount++;
1219 goto out;
1220 }
1221
1222 if (requested_pid == LNET_PID_ANY) {
1223 /* Don't instantiate LNET just for me */
1224 rc = -ENETDOWN;
1225 goto failed0;
1226 }
1227
1228 rc = lnet_prepare(requested_pid);
1229 if (rc)
1230 goto failed0;
1231
1232 rc = lnet_startup_lndnis();
1233 if (rc)
1234 goto failed1;
1235
1236 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
1237 if (rc)
1238 goto failed2;
1239
1240 rc = lnet_check_routes();
1241 if (rc)
1242 goto failed2;
1243
1244 rc = lnet_rtrpools_alloc(im_a_router);
1245 if (rc)
1246 goto failed2;
1247
1248 rc = lnet_acceptor_start();
1249 if (rc)
1250 goto failed2;
1251
1252 the_lnet.ln_refcount = 1;
1253 /* Now I may use my own API functions... */
1254
1255 /*
1256 * NB router checker needs the_lnet.ln_ping_info in
1257 * lnet_router_checker -> lnet_update_ni_status_locked
1258 */
1259 rc = lnet_ping_target_init();
1260 if (rc)
1261 goto failed3;
1262
1263 rc = lnet_router_checker_start();
1264 if (rc)
1265 goto failed4;
1266
1267 lnet_router_debugfs_init();
1268 goto out;
1269
1270 failed4:
1271 lnet_ping_target_fini();
1272 failed3:
1273 the_lnet.ln_refcount = 0;
1274 lnet_acceptor_stop();
1275 failed2:
1276 lnet_destroy_routes();
1277 lnet_shutdown_lndnis();
1278 failed1:
1279 lnet_unprepare();
1280 failed0:
1281 LASSERT(rc < 0);
1282 out:
1283 mutex_unlock(&the_lnet.ln_api_mutex);
1284 return rc;
1285 }
1286 EXPORT_SYMBOL(LNetNIInit);
1287
1288 /**
1289 * Stop LNet interfaces, routing, and forwarding.
1290 *
1291 * Users must call this function once for each successful call to LNetNIInit().
1292 * Once the LNetNIFini() operation has been started, the results of pending
1293 * API operations are undefined.
1294 *
1295 * \return always 0 for current implementation.
1296 */
1297 int
1298 LNetNIFini(void)
1299 {
1300 mutex_lock(&the_lnet.ln_api_mutex);
1301
1302 LASSERT(the_lnet.ln_init);
1303 LASSERT(the_lnet.ln_refcount > 0);
1304
1305 if (the_lnet.ln_refcount != 1) {
1306 the_lnet.ln_refcount--;
1307 } else {
1308 LASSERT(!the_lnet.ln_niinit_self);
1309
1310 lnet_router_debugfs_fini();
1311 lnet_router_checker_stop();
1312 lnet_ping_target_fini();
1313
1314 /* Teardown fns that use my own API functions BEFORE here */
1315 the_lnet.ln_refcount = 0;
1316
1317 lnet_acceptor_stop();
1318 lnet_destroy_routes();
1319 lnet_shutdown_lndnis();
1320 lnet_unprepare();
1321 }
1322
1323 mutex_unlock(&the_lnet.ln_api_mutex);
1324 return 0;
1325 }
1326 EXPORT_SYMBOL(LNetNIFini);
1327
1328 /**
1329 * LNet ioctl handler.
1330 *
1331 */
1332 int
1333 LNetCtl(unsigned int cmd, void *arg)
1334 {
1335 struct libcfs_ioctl_data *data = arg;
1336 lnet_process_id_t id = {0};
1337 lnet_ni_t *ni;
1338 int rc;
1339 unsigned long secs_passed;
1340
1341 LASSERT(the_lnet.ln_init);
1342 LASSERT(the_lnet.ln_refcount > 0);
1343
1344 switch (cmd) {
1345 case IOC_LIBCFS_GET_NI:
1346 rc = LNetGetId(data->ioc_count, &id);
1347 data->ioc_nid = id.nid;
1348 return rc;
1349
1350 case IOC_LIBCFS_FAIL_NID:
1351 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
1352
1353 case IOC_LIBCFS_ADD_ROUTE:
1354 rc = lnet_add_route(data->ioc_net, data->ioc_count,
1355 data->ioc_nid, data->ioc_priority);
1356 return (rc) ? rc : lnet_check_routes();
1357
1358 case IOC_LIBCFS_DEL_ROUTE:
1359 return lnet_del_route(data->ioc_net, data->ioc_nid);
1360
1361 case IOC_LIBCFS_GET_ROUTE:
1362 return lnet_get_route(data->ioc_count,
1363 &data->ioc_net, &data->ioc_count,
1364 &data->ioc_nid, &data->ioc_flags,
1365 &data->ioc_priority);
1366 case IOC_LIBCFS_NOTIFY_ROUTER:
1367 secs_passed = (ktime_get_real_seconds() - data->ioc_u64[0]);
1368 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
1369 jiffies - secs_passed * HZ);
1370
1371 case IOC_LIBCFS_LNET_DIST:
1372 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
1373 if (rc < 0 && rc != -EHOSTUNREACH)
1374 return rc;
1375
1376 data->ioc_u32[0] = rc;
1377 return 0;
1378
1379 case IOC_LIBCFS_TESTPROTOCOMPAT:
1380 lnet_net_lock(LNET_LOCK_EX);
1381 the_lnet.ln_testprotocompat = data->ioc_flags;
1382 lnet_net_unlock(LNET_LOCK_EX);
1383 return 0;
1384
1385 case IOC_LIBCFS_PING:
1386 id.nid = data->ioc_nid;
1387 id.pid = data->ioc_u32[0];
1388 rc = lnet_ping(id, data->ioc_u32[1], /* timeout */
1389 data->ioc_pbuf1,
1390 data->ioc_plen1 / sizeof(lnet_process_id_t));
1391 if (rc < 0)
1392 return rc;
1393 data->ioc_count = rc;
1394 return 0;
1395
1396 default:
1397 ni = lnet_net2ni(data->ioc_net);
1398 if (!ni)
1399 return -EINVAL;
1400
1401 if (!ni->ni_lnd->lnd_ctl)
1402 rc = -EINVAL;
1403 else
1404 rc = ni->ni_lnd->lnd_ctl(ni, cmd, arg);
1405
1406 lnet_ni_decref(ni);
1407 return rc;
1408 }
1409 /* not reached */
1410 }
1411 EXPORT_SYMBOL(LNetCtl);
1412
1413 void LNetDebugPeer(lnet_process_id_t id)
1414 {
1415 lnet_debug_peer(id.nid);
1416 }
1417 EXPORT_SYMBOL(LNetDebugPeer);
1418
1419 /**
1420 * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
1421 * all interfaces share a same PID, as requested by LNetNIInit().
1422 *
1423 * \param index Index of the interface to look up.
1424 * \param id On successful return, this location will hold the
1425 * lnet_process_id_t ID of the interface.
1426 *
1427 * \retval 0 If an interface exists at \a index.
1428 * \retval -ENOENT If no interface has been found.
1429 */
1430 int
1431 LNetGetId(unsigned int index, lnet_process_id_t *id)
1432 {
1433 struct lnet_ni *ni;
1434 struct list_head *tmp;
1435 int cpt;
1436 int rc = -ENOENT;
1437
1438 LASSERT(the_lnet.ln_init);
1439
1440 /* LNetNI initilization failed? */
1441 if (!the_lnet.ln_refcount)
1442 return rc;
1443
1444 cpt = lnet_net_lock_current();
1445
1446 list_for_each(tmp, &the_lnet.ln_nis) {
1447 if (index--)
1448 continue;
1449
1450 ni = list_entry(tmp, lnet_ni_t, ni_list);
1451
1452 id->nid = ni->ni_nid;
1453 id->pid = the_lnet.ln_pid;
1454 rc = 0;
1455 break;
1456 }
1457
1458 lnet_net_unlock(cpt);
1459 return rc;
1460 }
1461 EXPORT_SYMBOL(LNetGetId);
1462
1463 /**
1464 * Print a string representation of handle \a h into buffer \a str of
1465 * \a len bytes.
1466 */
1467 void
1468 LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
1469 {
1470 snprintf(str, len, "%#llx", h.cookie);
1471 }
1472 EXPORT_SYMBOL(LNetSnprintHandle);
1473
1474 static int
1475 lnet_create_ping_info(void)
1476 {
1477 int i;
1478 int n;
1479 int rc;
1480 unsigned int infosz;
1481 lnet_ni_t *ni;
1482 lnet_process_id_t id;
1483 lnet_ping_info_t *pinfo;
1484
1485 for (n = 0; ; n++) {
1486 rc = LNetGetId(n, &id);
1487 if (rc == -ENOENT)
1488 break;
1489
1490 LASSERT(!rc);
1491 }
1492
1493 infosz = offsetof(lnet_ping_info_t, pi_ni[n]);
1494 LIBCFS_ALLOC(pinfo, infosz);
1495 if (!pinfo) {
1496 CERROR("Can't allocate ping info[%d]\n", n);
1497 return -ENOMEM;
1498 }
1499
1500 pinfo->pi_nnis = n;
1501 pinfo->pi_pid = the_lnet.ln_pid;
1502 pinfo->pi_magic = LNET_PROTO_PING_MAGIC;
1503 pinfo->pi_features = LNET_PING_FEAT_NI_STATUS;
1504 if (!the_lnet.ln_routing)
1505 pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1506
1507 for (i = 0; i < n; i++) {
1508 lnet_ni_status_t *ns = &pinfo->pi_ni[i];
1509
1510 rc = LNetGetId(i, &id);
1511 LASSERT(!rc);
1512
1513 ns->ns_nid = id.nid;
1514 ns->ns_status = LNET_NI_STATUS_UP;
1515
1516 lnet_net_lock(0);
1517
1518 ni = lnet_nid2ni_locked(id.nid, 0);
1519 LASSERT(ni);
1520
1521 lnet_ni_lock(ni);
1522 LASSERT(!ni->ni_status);
1523 ni->ni_status = ns;
1524 lnet_ni_unlock(ni);
1525
1526 lnet_ni_decref_locked(ni, 0);
1527 lnet_net_unlock(0);
1528 }
1529
1530 the_lnet.ln_ping_info = pinfo;
1531 return 0;
1532 }
1533
1534 static void
1535 lnet_destroy_ping_info(void)
1536 {
1537 struct lnet_ni *ni;
1538
1539 lnet_net_lock(0);
1540
1541 list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
1542 lnet_ni_lock(ni);
1543 ni->ni_status = NULL;
1544 lnet_ni_unlock(ni);
1545 }
1546
1547 lnet_net_unlock(0);
1548
1549 LIBCFS_FREE(the_lnet.ln_ping_info,
1550 offsetof(lnet_ping_info_t,
1551 pi_ni[the_lnet.ln_ping_info->pi_nnis]));
1552 the_lnet.ln_ping_info = NULL;
1553 }
1554
1555 int
1556 lnet_ping_target_init(void)
1557 {
1558 lnet_md_t md = { NULL };
1559 lnet_handle_me_t meh;
1560 lnet_process_id_t id;
1561 int rc;
1562 int rc2;
1563 int infosz;
1564
1565 rc = lnet_create_ping_info();
1566 if (rc)
1567 return rc;
1568
1569 /*
1570 * We can have a tiny EQ since we only need to see the unlink event on
1571 * teardown, which by definition is the last one!
1572 */
1573 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &the_lnet.ln_ping_target_eq);
1574 if (rc) {
1575 CERROR("Can't allocate ping EQ: %d\n", rc);
1576 goto failed_0;
1577 }
1578
1579 memset(&id, 0, sizeof(lnet_process_id_t));
1580 id.nid = LNET_NID_ANY;
1581 id.pid = LNET_PID_ANY;
1582
1583 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1584 LNET_PROTO_PING_MATCHBITS, 0,
1585 LNET_UNLINK, LNET_INS_AFTER,
1586 &meh);
1587 if (rc) {
1588 CERROR("Can't create ping ME: %d\n", rc);
1589 goto failed_1;
1590 }
1591
1592 /* initialize md content */
1593 infosz = offsetof(lnet_ping_info_t,
1594 pi_ni[the_lnet.ln_ping_info->pi_nnis]);
1595 md.start = the_lnet.ln_ping_info;
1596 md.length = infosz;
1597 md.threshold = LNET_MD_THRESH_INF;
1598 md.max_size = 0;
1599 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1600 LNET_MD_MANAGE_REMOTE;
1601 md.user_ptr = NULL;
1602 md.eq_handle = the_lnet.ln_ping_target_eq;
1603
1604 rc = LNetMDAttach(meh, md,
1605 LNET_RETAIN,
1606 &the_lnet.ln_ping_target_md);
1607 if (rc) {
1608 CERROR("Can't attach ping MD: %d\n", rc);
1609 goto failed_2;
1610 }
1611
1612 return 0;
1613
1614 failed_2:
1615 rc2 = LNetMEUnlink(meh);
1616 LASSERT(!rc2);
1617 failed_1:
1618 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1619 LASSERT(!rc2);
1620 failed_0:
1621 lnet_destroy_ping_info();
1622 return rc;
1623 }
1624
1625 void
1626 lnet_ping_target_fini(void)
1627 {
1628 lnet_event_t event;
1629 int rc;
1630 int which;
1631 int timeout_ms = 1000;
1632 sigset_t blocked = cfs_block_allsigs();
1633
1634 LNetMDUnlink(the_lnet.ln_ping_target_md);
1635 /* NB md could be busy; this just starts the unlink */
1636
1637 for (;;) {
1638 rc = LNetEQPoll(&the_lnet.ln_ping_target_eq, 1,
1639 timeout_ms, &event, &which);
1640
1641 /* I expect overflow... */
1642 LASSERT(rc >= 0 || rc == -EOVERFLOW);
1643
1644 if (!rc) {
1645 /* timed out: provide a diagnostic */
1646 CWARN("Still waiting for ping MD to unlink\n");
1647 timeout_ms *= 2;
1648 continue;
1649 }
1650
1651 /* Got a valid event */
1652 if (event.unlinked)
1653 break;
1654 }
1655
1656 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1657 LASSERT(!rc);
1658 lnet_destroy_ping_info();
1659 cfs_restore_sigs(blocked);
1660 }
1661
1662 static int lnet_ping(lnet_process_id_t id, int timeout_ms,
1663 lnet_process_id_t __user *ids, int n_ids)
1664 {
1665 lnet_handle_eq_t eqh;
1666 lnet_handle_md_t mdh;
1667 lnet_event_t event;
1668 lnet_md_t md = { NULL };
1669 int which;
1670 int unlinked = 0;
1671 int replied = 0;
1672 const int a_long_time = 60000; /* mS */
1673 int infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
1674 lnet_ping_info_t *info;
1675 lnet_process_id_t tmpid;
1676 int i;
1677 int nob;
1678 int rc;
1679 int rc2;
1680 sigset_t blocked;
1681
1682 if (n_ids <= 0 ||
1683 id.nid == LNET_NID_ANY ||
1684 timeout_ms > 500000 || /* arbitrary limit! */
1685 n_ids > 20) /* arbitrary limit! */
1686 return -EINVAL;
1687
1688 if (id.pid == LNET_PID_ANY)
1689 id.pid = LUSTRE_SRV_LNET_PID;
1690
1691 LIBCFS_ALLOC(info, infosz);
1692 if (!info)
1693 return -ENOMEM;
1694
1695 /* NB 2 events max (including any unlink event) */
1696 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
1697 if (rc) {
1698 CERROR("Can't allocate EQ: %d\n", rc);
1699 goto out_0;
1700 }
1701
1702 /* initialize md content */
1703 md.start = info;
1704 md.length = infosz;
1705 md.threshold = 2; /*GET/REPLY*/
1706 md.max_size = 0;
1707 md.options = LNET_MD_TRUNCATE;
1708 md.user_ptr = NULL;
1709 md.eq_handle = eqh;
1710
1711 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
1712 if (rc) {
1713 CERROR("Can't bind MD: %d\n", rc);
1714 goto out_1;
1715 }
1716
1717 rc = LNetGet(LNET_NID_ANY, mdh, id,
1718 LNET_RESERVED_PORTAL,
1719 LNET_PROTO_PING_MATCHBITS, 0);
1720
1721 if (rc) {
1722 /* Don't CERROR; this could be deliberate! */
1723
1724 rc2 = LNetMDUnlink(mdh);
1725 LASSERT(!rc2);
1726
1727 /* NB must wait for the UNLINK event below... */
1728 unlinked = 1;
1729 timeout_ms = a_long_time;
1730 }
1731
1732 do {
1733 /* MUST block for unlink to complete */
1734 if (unlinked)
1735 blocked = cfs_block_allsigs();
1736
1737 rc2 = LNetEQPoll(&eqh, 1, timeout_ms, &event, &which);
1738
1739 if (unlinked)
1740 cfs_restore_sigs(blocked);
1741
1742 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
1743 (rc2 <= 0) ? -1 : event.type,
1744 (rc2 <= 0) ? -1 : event.status,
1745 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
1746
1747 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
1748
1749 if (rc2 <= 0 || event.status) {
1750 /* timeout or error */
1751 if (!replied && !rc)
1752 rc = (rc2 < 0) ? rc2 :
1753 !rc2 ? -ETIMEDOUT :
1754 event.status;
1755
1756 if (!unlinked) {
1757 /* Ensure completion in finite time... */
1758 LNetMDUnlink(mdh);
1759 /* No assertion (racing with network) */
1760 unlinked = 1;
1761 timeout_ms = a_long_time;
1762 } else if (!rc2) {
1763 /* timed out waiting for unlink */
1764 CWARN("ping %s: late network completion\n",
1765 libcfs_id2str(id));
1766 }
1767 } else if (event.type == LNET_EVENT_REPLY) {
1768 replied = 1;
1769 rc = event.mlength;
1770 }
1771
1772 } while (rc2 <= 0 || !event.unlinked);
1773
1774 if (!replied) {
1775 if (rc >= 0)
1776 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
1777 libcfs_id2str(id));
1778 rc = -EIO;
1779 goto out_1;
1780 }
1781
1782 nob = rc;
1783 LASSERT(nob >= 0 && nob <= infosz);
1784
1785 rc = -EPROTO; /* if I can't parse... */
1786
1787 if (nob < 8) {
1788 /* can't check magic/version */
1789 CERROR("%s: ping info too short %d\n",
1790 libcfs_id2str(id), nob);
1791 goto out_1;
1792 }
1793
1794 if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
1795 lnet_swap_pinginfo(info);
1796 } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
1797 CERROR("%s: Unexpected magic %08x\n",
1798 libcfs_id2str(id), info->pi_magic);
1799 goto out_1;
1800 }
1801
1802 if (!(info->pi_features & LNET_PING_FEAT_NI_STATUS)) {
1803 CERROR("%s: ping w/o NI status: 0x%x\n",
1804 libcfs_id2str(id), info->pi_features);
1805 goto out_1;
1806 }
1807
1808 if (nob < offsetof(lnet_ping_info_t, pi_ni[0])) {
1809 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
1810 nob, (int)offsetof(lnet_ping_info_t, pi_ni[0]));
1811 goto out_1;
1812 }
1813
1814 if (info->pi_nnis < n_ids)
1815 n_ids = info->pi_nnis;
1816
1817 if (nob < offsetof(lnet_ping_info_t, pi_ni[n_ids])) {
1818 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
1819 nob, (int)offsetof(lnet_ping_info_t, pi_ni[n_ids]));
1820 goto out_1;
1821 }
1822
1823 rc = -EFAULT; /* If I SEGV... */
1824
1825 memset(&tmpid, 0, sizeof(tmpid));
1826 for (i = 0; i < n_ids; i++) {
1827 tmpid.pid = info->pi_pid;
1828 tmpid.nid = info->pi_ni[i].ns_nid;
1829 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
1830 goto out_1;
1831 }
1832 rc = info->pi_nnis;
1833
1834 out_1:
1835 rc2 = LNetEQFree(eqh);
1836 if (rc2)
1837 CERROR("rc2 %d\n", rc2);
1838 LASSERT(!rc2);
1839
1840 out_0:
1841 LIBCFS_FREE(info, infosz);
1842 return rc;
1843 }
Linux kernel - Deliverables
This page took 0.070149 seconds and 4 git commands to generate.