Commit | Line | Data |
---|---|---|
d7e09d03 PT |
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, 2012, 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 | |
9fdaf8c0 | 38 | #include "../../include/linux/lnet/lib-lnet.h" |
d7e09d03 PT |
39 | #include <linux/log2.h> |
40 | ||
41 | #define D_LNI D_CONSOLE | |
42 | ||
43 | lnet_t the_lnet; /* THE state of the network */ | |
44 | EXPORT_SYMBOL(the_lnet); | |
45 | ||
46 | ||
47 | static char *ip2nets = ""; | |
8cc7b4b9 PT |
48 | module_param(ip2nets, charp, 0444); |
49 | MODULE_PARM_DESC(ip2nets, "LNET network <- IP table"); | |
d7e09d03 PT |
50 | |
51 | static char *networks = ""; | |
8cc7b4b9 PT |
52 | module_param(networks, charp, 0444); |
53 | MODULE_PARM_DESC(networks, "local networks"); | |
d7e09d03 PT |
54 | |
55 | static char *routes = ""; | |
8cc7b4b9 PT |
56 | module_param(routes, charp, 0444); |
57 | MODULE_PARM_DESC(routes, "routes to non-local networks"); | |
d7e09d03 PT |
58 | |
59 | static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT; | |
8cc7b4b9 PT |
60 | module_param(rnet_htable_size, int, 0444); |
61 | MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table"); | |
d7e09d03 | 62 | |
51bd8814 | 63 | static char * |
d7e09d03 PT |
64 | lnet_get_routes(void) |
65 | { | |
66 | return routes; | |
67 | } | |
68 | ||
51bd8814 | 69 | static char * |
d7e09d03 PT |
70 | lnet_get_networks(void) |
71 | { | |
72 | char *nets; | |
73 | int rc; | |
74 | ||
75 | if (*networks != 0 && *ip2nets != 0) { | |
76 | LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or " | |
77 | "'ip2nets' but not both at once\n"); | |
78 | return NULL; | |
79 | } | |
80 | ||
81 | if (*ip2nets != 0) { | |
82 | rc = lnet_parse_ip2nets(&nets, ip2nets); | |
83 | return (rc == 0) ? nets : NULL; | |
84 | } | |
85 | ||
86 | if (*networks != 0) | |
87 | return networks; | |
88 | ||
89 | return "tcp"; | |
90 | } | |
91 | ||
51bd8814 | 92 | static void |
d7e09d03 PT |
93 | lnet_init_locks(void) |
94 | { | |
95 | spin_lock_init(&the_lnet.ln_eq_wait_lock); | |
96 | init_waitqueue_head(&the_lnet.ln_eq_waitq); | |
97 | mutex_init(&the_lnet.ln_lnd_mutex); | |
98 | mutex_init(&the_lnet.ln_api_mutex); | |
99 | } | |
100 | ||
51bd8814 | 101 | static void |
d7e09d03 PT |
102 | lnet_fini_locks(void) |
103 | { | |
104 | } | |
105 | ||
106 | ||
107 | static int | |
108 | lnet_create_remote_nets_table(void) | |
109 | { | |
110 | int i; | |
111 | struct list_head *hash; | |
112 | ||
113 | LASSERT(the_lnet.ln_remote_nets_hash == NULL); | |
114 | LASSERT(the_lnet.ln_remote_nets_hbits > 0); | |
115 | LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash)); | |
116 | if (hash == NULL) { | |
117 | CERROR("Failed to create remote nets hash table\n"); | |
118 | return -ENOMEM; | |
119 | } | |
120 | ||
121 | for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) | |
122 | INIT_LIST_HEAD(&hash[i]); | |
123 | the_lnet.ln_remote_nets_hash = hash; | |
124 | return 0; | |
125 | } | |
126 | ||
127 | static void | |
128 | lnet_destroy_remote_nets_table(void) | |
129 | { | |
1ae2bdb2 | 130 | int i; |
d7e09d03 PT |
131 | |
132 | if (the_lnet.ln_remote_nets_hash == NULL) | |
133 | return; | |
134 | ||
135 | for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) | |
136 | LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i])); | |
137 | ||
138 | LIBCFS_FREE(the_lnet.ln_remote_nets_hash, | |
1ae2bdb2 DE |
139 | LNET_REMOTE_NETS_HASH_SIZE * |
140 | sizeof(the_lnet.ln_remote_nets_hash[0])); | |
d7e09d03 PT |
141 | the_lnet.ln_remote_nets_hash = NULL; |
142 | } | |
143 | ||
144 | static void | |
145 | lnet_destroy_locks(void) | |
146 | { | |
147 | if (the_lnet.ln_res_lock != NULL) { | |
148 | cfs_percpt_lock_free(the_lnet.ln_res_lock); | |
149 | the_lnet.ln_res_lock = NULL; | |
150 | } | |
151 | ||
152 | if (the_lnet.ln_net_lock != NULL) { | |
153 | cfs_percpt_lock_free(the_lnet.ln_net_lock); | |
154 | the_lnet.ln_net_lock = NULL; | |
155 | } | |
156 | ||
157 | lnet_fini_locks(); | |
158 | } | |
159 | ||
160 | static int | |
161 | lnet_create_locks(void) | |
162 | { | |
163 | lnet_init_locks(); | |
164 | ||
165 | the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table()); | |
166 | if (the_lnet.ln_res_lock == NULL) | |
167 | goto failed; | |
168 | ||
169 | the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table()); | |
170 | if (the_lnet.ln_net_lock == NULL) | |
171 | goto failed; | |
172 | ||
173 | return 0; | |
174 | ||
175 | failed: | |
176 | lnet_destroy_locks(); | |
177 | return -ENOMEM; | |
178 | } | |
179 | ||
51bd8814 | 180 | static void lnet_assert_wire_constants (void) |
d7e09d03 PT |
181 | { |
182 | /* Wire protocol assertions generated by 'wirecheck' | |
183 | * running on Linux robert.bartonsoftware.com 2.6.8-1.521 | |
184 | * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux | |
185 | * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */ | |
186 | ||
187 | /* Constants... */ | |
188 | CLASSERT (LNET_PROTO_TCP_MAGIC == 0xeebc0ded); | |
189 | CLASSERT (LNET_PROTO_TCP_VERSION_MAJOR == 1); | |
190 | CLASSERT (LNET_PROTO_TCP_VERSION_MINOR == 0); | |
191 | CLASSERT (LNET_MSG_ACK == 0); | |
192 | CLASSERT (LNET_MSG_PUT == 1); | |
193 | CLASSERT (LNET_MSG_GET == 2); | |
194 | CLASSERT (LNET_MSG_REPLY == 3); | |
195 | CLASSERT (LNET_MSG_HELLO == 4); | |
196 | ||
197 | /* Checks for struct ptl_handle_wire_t */ | |
198 | CLASSERT ((int)sizeof(lnet_handle_wire_t) == 16); | |
199 | CLASSERT ((int)offsetof(lnet_handle_wire_t, wh_interface_cookie) == 0); | |
200 | CLASSERT ((int)sizeof(((lnet_handle_wire_t *)0)->wh_interface_cookie) == 8); | |
201 | CLASSERT ((int)offsetof(lnet_handle_wire_t, wh_object_cookie) == 8); | |
202 | CLASSERT ((int)sizeof(((lnet_handle_wire_t *)0)->wh_object_cookie) == 8); | |
203 | ||
204 | /* Checks for struct lnet_magicversion_t */ | |
205 | CLASSERT ((int)sizeof(lnet_magicversion_t) == 8); | |
206 | CLASSERT ((int)offsetof(lnet_magicversion_t, magic) == 0); | |
207 | CLASSERT ((int)sizeof(((lnet_magicversion_t *)0)->magic) == 4); | |
208 | CLASSERT ((int)offsetof(lnet_magicversion_t, version_major) == 4); | |
209 | CLASSERT ((int)sizeof(((lnet_magicversion_t *)0)->version_major) == 2); | |
210 | CLASSERT ((int)offsetof(lnet_magicversion_t, version_minor) == 6); | |
211 | CLASSERT ((int)sizeof(((lnet_magicversion_t *)0)->version_minor) == 2); | |
212 | ||
213 | /* Checks for struct lnet_hdr_t */ | |
214 | CLASSERT ((int)sizeof(lnet_hdr_t) == 72); | |
215 | CLASSERT ((int)offsetof(lnet_hdr_t, dest_nid) == 0); | |
216 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->dest_nid) == 8); | |
217 | CLASSERT ((int)offsetof(lnet_hdr_t, src_nid) == 8); | |
218 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->src_nid) == 8); | |
219 | CLASSERT ((int)offsetof(lnet_hdr_t, dest_pid) == 16); | |
220 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->dest_pid) == 4); | |
221 | CLASSERT ((int)offsetof(lnet_hdr_t, src_pid) == 20); | |
222 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->src_pid) == 4); | |
223 | CLASSERT ((int)offsetof(lnet_hdr_t, type) == 24); | |
224 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->type) == 4); | |
225 | CLASSERT ((int)offsetof(lnet_hdr_t, payload_length) == 28); | |
226 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->payload_length) == 4); | |
227 | CLASSERT ((int)offsetof(lnet_hdr_t, msg) == 32); | |
228 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg) == 40); | |
229 | ||
230 | /* Ack */ | |
231 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.ack.dst_wmd) == 32); | |
232 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.ack.dst_wmd) == 16); | |
233 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.ack.match_bits) == 48); | |
234 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.ack.match_bits) == 8); | |
235 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.ack.mlength) == 56); | |
236 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.ack.mlength) == 4); | |
237 | ||
238 | /* Put */ | |
239 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.ack_wmd) == 32); | |
240 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.ack_wmd) == 16); | |
241 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.match_bits) == 48); | |
242 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.match_bits) == 8); | |
243 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.hdr_data) == 56); | |
244 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.hdr_data) == 8); | |
245 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.ptl_index) == 64); | |
246 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.ptl_index) == 4); | |
247 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.put.offset) == 68); | |
248 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.put.offset) == 4); | |
249 | ||
250 | /* Get */ | |
251 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.return_wmd) == 32); | |
252 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.return_wmd) == 16); | |
253 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.match_bits) == 48); | |
254 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.match_bits) == 8); | |
255 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.ptl_index) == 56); | |
256 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.ptl_index) == 4); | |
257 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.src_offset) == 60); | |
258 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.src_offset) == 4); | |
259 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.get.sink_length) == 64); | |
260 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.get.sink_length) == 4); | |
261 | ||
262 | /* Reply */ | |
263 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.reply.dst_wmd) == 32); | |
264 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.reply.dst_wmd) == 16); | |
265 | ||
266 | /* Hello */ | |
267 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.hello.incarnation) == 32); | |
268 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.hello.incarnation) == 8); | |
269 | CLASSERT ((int)offsetof(lnet_hdr_t, msg.hello.type) == 40); | |
270 | CLASSERT ((int)sizeof(((lnet_hdr_t *)0)->msg.hello.type) == 4); | |
271 | } | |
272 | ||
51bd8814 | 273 | static lnd_t * |
d7e09d03 PT |
274 | lnet_find_lnd_by_type (int type) |
275 | { | |
276 | lnd_t *lnd; | |
277 | struct list_head *tmp; | |
278 | ||
279 | /* holding lnd mutex */ | |
280 | list_for_each (tmp, &the_lnet.ln_lnds) { | |
281 | lnd = list_entry(tmp, lnd_t, lnd_list); | |
282 | ||
283 | if ((int)lnd->lnd_type == type) | |
284 | return lnd; | |
285 | } | |
286 | ||
287 | return NULL; | |
288 | } | |
289 | ||
290 | void | |
291 | lnet_register_lnd (lnd_t *lnd) | |
292 | { | |
293 | LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex); | |
294 | ||
295 | LASSERT (the_lnet.ln_init); | |
296 | LASSERT (libcfs_isknown_lnd(lnd->lnd_type)); | |
297 | LASSERT (lnet_find_lnd_by_type(lnd->lnd_type) == NULL); | |
298 | ||
299 | list_add_tail (&lnd->lnd_list, &the_lnet.ln_lnds); | |
300 | lnd->lnd_refcount = 0; | |
301 | ||
302 | CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type)); | |
303 | ||
304 | LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); | |
305 | } | |
306 | EXPORT_SYMBOL(lnet_register_lnd); | |
307 | ||
308 | void | |
309 | lnet_unregister_lnd (lnd_t *lnd) | |
310 | { | |
311 | LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex); | |
312 | ||
313 | LASSERT (the_lnet.ln_init); | |
314 | LASSERT (lnet_find_lnd_by_type(lnd->lnd_type) == lnd); | |
315 | LASSERT (lnd->lnd_refcount == 0); | |
316 | ||
317 | list_del (&lnd->lnd_list); | |
318 | CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type)); | |
319 | ||
320 | LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); | |
321 | } | |
322 | EXPORT_SYMBOL(lnet_unregister_lnd); | |
323 | ||
324 | void | |
325 | lnet_counters_get(lnet_counters_t *counters) | |
326 | { | |
327 | lnet_counters_t *ctr; | |
328 | int i; | |
329 | ||
330 | memset(counters, 0, sizeof(*counters)); | |
331 | ||
332 | lnet_net_lock(LNET_LOCK_EX); | |
333 | ||
334 | cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) { | |
335 | counters->msgs_max += ctr->msgs_max; | |
336 | counters->msgs_alloc += ctr->msgs_alloc; | |
337 | counters->errors += ctr->errors; | |
338 | counters->send_count += ctr->send_count; | |
339 | counters->recv_count += ctr->recv_count; | |
340 | counters->route_count += ctr->route_count; | |
b3d0dfe5 | 341 | counters->drop_count += ctr->drop_count; |
d7e09d03 PT |
342 | counters->send_length += ctr->send_length; |
343 | counters->recv_length += ctr->recv_length; | |
344 | counters->route_length += ctr->route_length; | |
345 | counters->drop_length += ctr->drop_length; | |
346 | ||
347 | } | |
348 | lnet_net_unlock(LNET_LOCK_EX); | |
349 | } | |
350 | EXPORT_SYMBOL(lnet_counters_get); | |
351 | ||
352 | void | |
353 | lnet_counters_reset(void) | |
354 | { | |
355 | lnet_counters_t *counters; | |
356 | int i; | |
357 | ||
358 | lnet_net_lock(LNET_LOCK_EX); | |
359 | ||
360 | cfs_percpt_for_each(counters, i, the_lnet.ln_counters) | |
361 | memset(counters, 0, sizeof(lnet_counters_t)); | |
362 | ||
363 | lnet_net_unlock(LNET_LOCK_EX); | |
364 | } | |
365 | EXPORT_SYMBOL(lnet_counters_reset); | |
366 | ||
367 | #ifdef LNET_USE_LIB_FREELIST | |
368 | ||
369 | int | |
370 | lnet_freelist_init (lnet_freelist_t *fl, int n, int size) | |
371 | { | |
372 | char *space; | |
373 | ||
374 | LASSERT (n > 0); | |
375 | ||
376 | size += offsetof (lnet_freeobj_t, fo_contents); | |
377 | ||
378 | LIBCFS_ALLOC(space, n * size); | |
379 | if (space == NULL) | |
380 | return (-ENOMEM); | |
381 | ||
382 | INIT_LIST_HEAD (&fl->fl_list); | |
383 | fl->fl_objs = space; | |
384 | fl->fl_nobjs = n; | |
385 | fl->fl_objsize = size; | |
386 | ||
387 | do | |
388 | { | |
389 | memset (space, 0, size); | |
390 | list_add ((struct list_head *)space, &fl->fl_list); | |
391 | space += size; | |
392 | } while (--n != 0); | |
393 | ||
394 | return (0); | |
395 | } | |
396 | ||
397 | void | |
398 | lnet_freelist_fini (lnet_freelist_t *fl) | |
399 | { | |
400 | struct list_head *el; | |
401 | int count; | |
402 | ||
403 | if (fl->fl_nobjs == 0) | |
404 | return; | |
405 | ||
406 | count = 0; | |
407 | for (el = fl->fl_list.next; el != &fl->fl_list; el = el->next) | |
408 | count++; | |
409 | ||
410 | LASSERT (count == fl->fl_nobjs); | |
411 | ||
412 | LIBCFS_FREE(fl->fl_objs, fl->fl_nobjs * fl->fl_objsize); | |
413 | memset (fl, 0, sizeof (*fl)); | |
414 | } | |
415 | ||
416 | #endif /* LNET_USE_LIB_FREELIST */ | |
417 | ||
51bd8814 | 418 | static __u64 |
d7e09d03 PT |
419 | lnet_create_interface_cookie (void) |
420 | { | |
421 | /* NB the interface cookie in wire handles guards against delayed | |
422 | * replies and ACKs appearing valid after reboot. Initialisation time, | |
423 | * even if it's only implemented to millisecond resolution is probably | |
424 | * easily good enough. */ | |
425 | struct timeval tv; | |
426 | __u64 cookie; | |
427 | do_gettimeofday(&tv); | |
428 | cookie = tv.tv_sec; | |
429 | cookie *= 1000000; | |
430 | cookie += tv.tv_usec; | |
431 | return cookie; | |
432 | } | |
433 | ||
434 | static char * | |
435 | lnet_res_type2str(int type) | |
436 | { | |
437 | switch (type) { | |
438 | default: | |
439 | LBUG(); | |
440 | case LNET_COOKIE_TYPE_MD: | |
441 | return "MD"; | |
442 | case LNET_COOKIE_TYPE_ME: | |
443 | return "ME"; | |
444 | case LNET_COOKIE_TYPE_EQ: | |
445 | return "EQ"; | |
446 | } | |
447 | } | |
448 | ||
51bd8814 | 449 | static void |
d7e09d03 PT |
450 | lnet_res_container_cleanup(struct lnet_res_container *rec) |
451 | { | |
452 | int count = 0; | |
453 | ||
454 | if (rec->rec_type == 0) /* not set yet, it's uninitialized */ | |
455 | return; | |
456 | ||
457 | while (!list_empty(&rec->rec_active)) { | |
458 | struct list_head *e = rec->rec_active.next; | |
459 | ||
460 | list_del_init(e); | |
461 | if (rec->rec_type == LNET_COOKIE_TYPE_EQ) { | |
462 | lnet_eq_free(list_entry(e, lnet_eq_t, eq_list)); | |
463 | ||
464 | } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) { | |
465 | lnet_md_free(list_entry(e, lnet_libmd_t, md_list)); | |
466 | ||
467 | } else { /* NB: Active MEs should be attached on portals */ | |
468 | LBUG(); | |
469 | } | |
470 | count++; | |
471 | } | |
472 | ||
473 | if (count > 0) { | |
474 | /* Found alive MD/ME/EQ, user really should unlink/free | |
475 | * all of them before finalize LNet, but if someone didn't, | |
476 | * we have to recycle garbage for him */ | |
477 | CERROR("%d active elements on exit of %s container\n", | |
478 | count, lnet_res_type2str(rec->rec_type)); | |
479 | } | |
480 | ||
481 | #ifdef LNET_USE_LIB_FREELIST | |
482 | lnet_freelist_fini(&rec->rec_freelist); | |
483 | #endif | |
484 | if (rec->rec_lh_hash != NULL) { | |
485 | LIBCFS_FREE(rec->rec_lh_hash, | |
486 | LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0])); | |
487 | rec->rec_lh_hash = NULL; | |
488 | } | |
489 | ||
490 | rec->rec_type = 0; /* mark it as finalized */ | |
491 | } | |
492 | ||
51bd8814 | 493 | static int |
d7e09d03 PT |
494 | lnet_res_container_setup(struct lnet_res_container *rec, |
495 | int cpt, int type, int objnum, int objsz) | |
496 | { | |
497 | int rc = 0; | |
498 | int i; | |
499 | ||
500 | LASSERT(rec->rec_type == 0); | |
501 | ||
502 | rec->rec_type = type; | |
503 | INIT_LIST_HEAD(&rec->rec_active); | |
504 | ||
505 | #ifdef LNET_USE_LIB_FREELIST | |
506 | memset(&rec->rec_freelist, 0, sizeof(rec->rec_freelist)); | |
507 | rc = lnet_freelist_init(&rec->rec_freelist, objnum, objsz); | |
508 | if (rc != 0) | |
509 | goto out; | |
510 | #endif | |
511 | rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type; | |
512 | ||
513 | /* Arbitrary choice of hash table size */ | |
514 | LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt, | |
515 | LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0])); | |
516 | if (rec->rec_lh_hash == NULL) { | |
517 | rc = -ENOMEM; | |
518 | goto out; | |
519 | } | |
520 | ||
521 | for (i = 0; i < LNET_LH_HASH_SIZE; i++) | |
522 | INIT_LIST_HEAD(&rec->rec_lh_hash[i]); | |
523 | ||
524 | return 0; | |
525 | ||
526 | out: | |
527 | CERROR("Failed to setup %s resource container\n", | |
528 | lnet_res_type2str(type)); | |
529 | lnet_res_container_cleanup(rec); | |
530 | return rc; | |
531 | } | |
532 | ||
533 | static void | |
534 | lnet_res_containers_destroy(struct lnet_res_container **recs) | |
535 | { | |
536 | struct lnet_res_container *rec; | |
537 | int i; | |
538 | ||
539 | cfs_percpt_for_each(rec, i, recs) | |
540 | lnet_res_container_cleanup(rec); | |
541 | ||
542 | cfs_percpt_free(recs); | |
543 | } | |
544 | ||
545 | static struct lnet_res_container ** | |
546 | lnet_res_containers_create(int type, int objnum, int objsz) | |
547 | { | |
548 | struct lnet_res_container **recs; | |
549 | struct lnet_res_container *rec; | |
550 | int rc; | |
551 | int i; | |
552 | ||
553 | recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec)); | |
554 | if (recs == NULL) { | |
555 | CERROR("Failed to allocate %s resource containers\n", | |
556 | lnet_res_type2str(type)); | |
557 | return NULL; | |
558 | } | |
559 | ||
560 | cfs_percpt_for_each(rec, i, recs) { | |
561 | rc = lnet_res_container_setup(rec, i, type, objnum, objsz); | |
562 | if (rc != 0) { | |
563 | lnet_res_containers_destroy(recs); | |
564 | return NULL; | |
565 | } | |
566 | } | |
567 | ||
568 | return recs; | |
569 | } | |
570 | ||
571 | lnet_libhandle_t * | |
572 | lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie) | |
573 | { | |
574 | /* ALWAYS called with lnet_res_lock held */ | |
575 | struct list_head *head; | |
576 | lnet_libhandle_t *lh; | |
577 | unsigned int hash; | |
578 | ||
579 | if ((cookie & LNET_COOKIE_MASK) != rec->rec_type) | |
580 | return NULL; | |
581 | ||
582 | hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS); | |
583 | head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK]; | |
584 | ||
585 | list_for_each_entry(lh, head, lh_hash_chain) { | |
586 | if (lh->lh_cookie == cookie) | |
587 | return lh; | |
588 | } | |
589 | ||
590 | return NULL; | |
591 | } | |
592 | ||
593 | void | |
594 | lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh) | |
595 | { | |
596 | /* ALWAYS called with lnet_res_lock held */ | |
597 | unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS; | |
598 | unsigned int hash; | |
599 | ||
600 | lh->lh_cookie = rec->rec_lh_cookie; | |
601 | rec->rec_lh_cookie += 1 << ibits; | |
602 | ||
603 | hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK; | |
604 | ||
605 | list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]); | |
606 | } | |
607 | ||
608 | ||
609 | int lnet_unprepare(void); | |
610 | ||
51bd8814 | 611 | static int |
d7e09d03 PT |
612 | lnet_prepare(lnet_pid_t requested_pid) |
613 | { | |
614 | /* Prepare to bring up the network */ | |
615 | struct lnet_res_container **recs; | |
616 | int rc = 0; | |
617 | ||
618 | LASSERT (the_lnet.ln_refcount == 0); | |
619 | ||
620 | the_lnet.ln_routing = 0; | |
621 | ||
622 | LASSERT ((requested_pid & LNET_PID_USERFLAG) == 0); | |
623 | the_lnet.ln_pid = requested_pid; | |
624 | ||
625 | INIT_LIST_HEAD(&the_lnet.ln_test_peers); | |
626 | INIT_LIST_HEAD(&the_lnet.ln_nis); | |
627 | INIT_LIST_HEAD(&the_lnet.ln_nis_cpt); | |
628 | INIT_LIST_HEAD(&the_lnet.ln_nis_zombie); | |
629 | INIT_LIST_HEAD(&the_lnet.ln_routers); | |
630 | ||
631 | rc = lnet_create_remote_nets_table(); | |
632 | if (rc != 0) | |
633 | goto failed; | |
634 | ||
635 | the_lnet.ln_interface_cookie = lnet_create_interface_cookie(); | |
636 | ||
637 | the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(), | |
638 | sizeof(lnet_counters_t)); | |
639 | if (the_lnet.ln_counters == NULL) { | |
640 | CERROR("Failed to allocate counters for LNet\n"); | |
641 | rc = -ENOMEM; | |
642 | goto failed; | |
643 | } | |
644 | ||
645 | rc = lnet_peer_tables_create(); | |
646 | if (rc != 0) | |
647 | goto failed; | |
648 | ||
649 | rc = lnet_msg_containers_create(); | |
650 | if (rc != 0) | |
651 | goto failed; | |
652 | ||
653 | rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0, | |
654 | LNET_COOKIE_TYPE_EQ, LNET_FL_MAX_EQS, | |
655 | sizeof(lnet_eq_t)); | |
656 | if (rc != 0) | |
657 | goto failed; | |
658 | ||
659 | recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME, LNET_FL_MAX_MES, | |
660 | sizeof(lnet_me_t)); | |
661 | if (recs == NULL) | |
662 | goto failed; | |
663 | ||
664 | the_lnet.ln_me_containers = recs; | |
665 | ||
666 | recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD, LNET_FL_MAX_MDS, | |
667 | sizeof(lnet_libmd_t)); | |
668 | if (recs == NULL) | |
669 | goto failed; | |
670 | ||
671 | the_lnet.ln_md_containers = recs; | |
672 | ||
673 | rc = lnet_portals_create(); | |
674 | if (rc != 0) { | |
675 | CERROR("Failed to create portals for LNet: %d\n", rc); | |
676 | goto failed; | |
677 | } | |
678 | ||
679 | return 0; | |
680 | ||
681 | failed: | |
682 | lnet_unprepare(); | |
683 | return rc; | |
684 | } | |
685 | ||
686 | int | |
687 | lnet_unprepare (void) | |
688 | { | |
689 | /* NB no LNET_LOCK since this is the last reference. All LND instances | |
690 | * have shut down already, so it is safe to unlink and free all | |
691 | * descriptors, even those that appear committed to a network op (eg MD | |
692 | * with non-zero pending count) */ | |
693 | ||
694 | lnet_fail_nid(LNET_NID_ANY, 0); | |
695 | ||
696 | LASSERT(the_lnet.ln_refcount == 0); | |
697 | LASSERT(list_empty(&the_lnet.ln_test_peers)); | |
698 | LASSERT(list_empty(&the_lnet.ln_nis)); | |
699 | LASSERT(list_empty(&the_lnet.ln_nis_cpt)); | |
700 | LASSERT(list_empty(&the_lnet.ln_nis_zombie)); | |
701 | ||
702 | lnet_portals_destroy(); | |
703 | ||
704 | if (the_lnet.ln_md_containers != NULL) { | |
705 | lnet_res_containers_destroy(the_lnet.ln_md_containers); | |
706 | the_lnet.ln_md_containers = NULL; | |
707 | } | |
708 | ||
709 | if (the_lnet.ln_me_containers != NULL) { | |
710 | lnet_res_containers_destroy(the_lnet.ln_me_containers); | |
711 | the_lnet.ln_me_containers = NULL; | |
712 | } | |
713 | ||
714 | lnet_res_container_cleanup(&the_lnet.ln_eq_container); | |
715 | ||
716 | lnet_msg_containers_destroy(); | |
717 | lnet_peer_tables_destroy(); | |
718 | lnet_rtrpools_free(); | |
719 | ||
720 | if (the_lnet.ln_counters != NULL) { | |
721 | cfs_percpt_free(the_lnet.ln_counters); | |
722 | the_lnet.ln_counters = NULL; | |
723 | } | |
724 | lnet_destroy_remote_nets_table(); | |
725 | ||
726 | return 0; | |
727 | } | |
728 | ||
729 | lnet_ni_t * | |
730 | lnet_net2ni_locked(__u32 net, int cpt) | |
731 | { | |
732 | struct list_head *tmp; | |
733 | lnet_ni_t *ni; | |
734 | ||
735 | LASSERT(cpt != LNET_LOCK_EX); | |
736 | ||
737 | list_for_each(tmp, &the_lnet.ln_nis) { | |
738 | ni = list_entry(tmp, lnet_ni_t, ni_list); | |
739 | ||
740 | if (LNET_NIDNET(ni->ni_nid) == net) { | |
741 | lnet_ni_addref_locked(ni, cpt); | |
742 | return ni; | |
743 | } | |
744 | } | |
745 | ||
746 | return NULL; | |
747 | } | |
748 | ||
749 | lnet_ni_t * | |
750 | lnet_net2ni(__u32 net) | |
751 | { | |
752 | lnet_ni_t *ni; | |
753 | ||
754 | lnet_net_lock(0); | |
755 | ni = lnet_net2ni_locked(net, 0); | |
756 | lnet_net_unlock(0); | |
757 | ||
758 | return ni; | |
759 | } | |
760 | EXPORT_SYMBOL(lnet_net2ni); | |
761 | ||
762 | static unsigned int | |
763 | lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number) | |
764 | { | |
765 | __u64 key = nid; | |
766 | unsigned int val; | |
767 | ||
768 | LASSERT(number >= 1 && number <= LNET_CPT_NUMBER); | |
769 | ||
770 | if (number == 1) | |
771 | return 0; | |
772 | ||
72c0824a | 773 | val = hash_long(key, LNET_CPT_BITS); |
d7e09d03 PT |
774 | /* NB: LNET_CP_NUMBER doesn't have to be PO2 */ |
775 | if (val < number) | |
776 | return val; | |
777 | ||
778 | return (unsigned int)(key + val + (val >> 1)) % number; | |
779 | } | |
780 | ||
781 | int | |
782 | lnet_cpt_of_nid_locked(lnet_nid_t nid) | |
783 | { | |
784 | struct lnet_ni *ni; | |
785 | ||
786 | /* must called with hold of lnet_net_lock */ | |
787 | if (LNET_CPT_NUMBER == 1) | |
788 | return 0; /* the only one */ | |
789 | ||
790 | /* take lnet_net_lock(any) would be OK */ | |
791 | if (!list_empty(&the_lnet.ln_nis_cpt)) { | |
792 | list_for_each_entry(ni, &the_lnet.ln_nis_cpt, ni_cptlist) { | |
793 | if (LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid)) | |
794 | continue; | |
795 | ||
796 | LASSERT(ni->ni_cpts != NULL); | |
797 | return ni->ni_cpts[lnet_nid_cpt_hash | |
798 | (nid, ni->ni_ncpts)]; | |
799 | } | |
800 | } | |
801 | ||
802 | return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER); | |
803 | } | |
804 | ||
805 | int | |
806 | lnet_cpt_of_nid(lnet_nid_t nid) | |
807 | { | |
808 | int cpt; | |
809 | int cpt2; | |
810 | ||
811 | if (LNET_CPT_NUMBER == 1) | |
812 | return 0; /* the only one */ | |
813 | ||
814 | if (list_empty(&the_lnet.ln_nis_cpt)) | |
815 | return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER); | |
816 | ||
817 | cpt = lnet_net_lock_current(); | |
818 | cpt2 = lnet_cpt_of_nid_locked(nid); | |
819 | lnet_net_unlock(cpt); | |
820 | ||
821 | return cpt2; | |
822 | } | |
823 | EXPORT_SYMBOL(lnet_cpt_of_nid); | |
824 | ||
825 | int | |
826 | lnet_islocalnet(__u32 net) | |
827 | { | |
828 | struct lnet_ni *ni; | |
829 | int cpt; | |
830 | ||
831 | cpt = lnet_net_lock_current(); | |
832 | ||
833 | ni = lnet_net2ni_locked(net, cpt); | |
834 | if (ni != NULL) | |
835 | lnet_ni_decref_locked(ni, cpt); | |
836 | ||
837 | lnet_net_unlock(cpt); | |
838 | ||
839 | return ni != NULL; | |
840 | } | |
841 | ||
842 | lnet_ni_t * | |
843 | lnet_nid2ni_locked(lnet_nid_t nid, int cpt) | |
844 | { | |
845 | struct lnet_ni *ni; | |
846 | struct list_head *tmp; | |
847 | ||
848 | LASSERT(cpt != LNET_LOCK_EX); | |
849 | ||
850 | list_for_each(tmp, &the_lnet.ln_nis) { | |
851 | ni = list_entry(tmp, lnet_ni_t, ni_list); | |
852 | ||
853 | if (ni->ni_nid == nid) { | |
854 | lnet_ni_addref_locked(ni, cpt); | |
855 | return ni; | |
856 | } | |
857 | } | |
858 | ||
859 | return NULL; | |
860 | } | |
861 | ||
862 | int | |
863 | lnet_islocalnid(lnet_nid_t nid) | |
864 | { | |
865 | struct lnet_ni *ni; | |
866 | int cpt; | |
867 | ||
868 | cpt = lnet_net_lock_current(); | |
869 | ni = lnet_nid2ni_locked(nid, cpt); | |
870 | if (ni != NULL) | |
871 | lnet_ni_decref_locked(ni, cpt); | |
872 | lnet_net_unlock(cpt); | |
873 | ||
874 | return ni != NULL; | |
875 | } | |
876 | ||
877 | int | |
878 | lnet_count_acceptor_nis (void) | |
879 | { | |
880 | /* Return the # of NIs that need the acceptor. */ | |
881 | int count = 0; | |
882 | struct list_head *tmp; | |
883 | struct lnet_ni *ni; | |
884 | int cpt; | |
885 | ||
886 | cpt = lnet_net_lock_current(); | |
887 | list_for_each(tmp, &the_lnet.ln_nis) { | |
888 | ni = list_entry(tmp, lnet_ni_t, ni_list); | |
889 | ||
890 | if (ni->ni_lnd->lnd_accept != NULL) | |
891 | count++; | |
892 | } | |
893 | ||
894 | lnet_net_unlock(cpt); | |
895 | ||
896 | return count; | |
897 | } | |
898 | ||
899 | static int | |
900 | lnet_ni_tq_credits(lnet_ni_t *ni) | |
901 | { | |
902 | int credits; | |
903 | ||
904 | LASSERT(ni->ni_ncpts >= 1); | |
905 | ||
906 | if (ni->ni_ncpts == 1) | |
907 | return ni->ni_maxtxcredits; | |
908 | ||
909 | credits = ni->ni_maxtxcredits / ni->ni_ncpts; | |
910 | credits = max(credits, 8 * ni->ni_peertxcredits); | |
911 | credits = min(credits, ni->ni_maxtxcredits); | |
912 | ||
913 | return credits; | |
914 | } | |
915 | ||
51bd8814 | 916 | static void |
d7e09d03 PT |
917 | lnet_shutdown_lndnis (void) |
918 | { | |
919 | int i; | |
920 | int islo; | |
921 | lnet_ni_t *ni; | |
922 | ||
923 | /* NB called holding the global mutex */ | |
924 | ||
925 | /* All quiet on the API front */ | |
926 | LASSERT(!the_lnet.ln_shutdown); | |
927 | LASSERT(the_lnet.ln_refcount == 0); | |
928 | LASSERT(list_empty(&the_lnet.ln_nis_zombie)); | |
929 | ||
930 | lnet_net_lock(LNET_LOCK_EX); | |
931 | the_lnet.ln_shutdown = 1; /* flag shutdown */ | |
932 | ||
933 | /* Unlink NIs from the global table */ | |
934 | while (!list_empty(&the_lnet.ln_nis)) { | |
935 | ni = list_entry(the_lnet.ln_nis.next, | |
936 | lnet_ni_t, ni_list); | |
937 | /* move it to zombie list and nobody can find it anymore */ | |
938 | list_move(&ni->ni_list, &the_lnet.ln_nis_zombie); | |
939 | lnet_ni_decref_locked(ni, 0); /* drop ln_nis' ref */ | |
940 | ||
941 | if (!list_empty(&ni->ni_cptlist)) { | |
942 | list_del_init(&ni->ni_cptlist); | |
943 | lnet_ni_decref_locked(ni, 0); | |
944 | } | |
945 | } | |
946 | ||
947 | /* Drop the cached eqwait NI. */ | |
948 | if (the_lnet.ln_eq_waitni != NULL) { | |
949 | lnet_ni_decref_locked(the_lnet.ln_eq_waitni, 0); | |
950 | the_lnet.ln_eq_waitni = NULL; | |
951 | } | |
952 | ||
953 | /* Drop the cached loopback NI. */ | |
954 | if (the_lnet.ln_loni != NULL) { | |
955 | lnet_ni_decref_locked(the_lnet.ln_loni, 0); | |
956 | the_lnet.ln_loni = NULL; | |
957 | } | |
958 | ||
959 | lnet_net_unlock(LNET_LOCK_EX); | |
960 | ||
961 | /* Clear lazy portals and drop delayed messages which hold refs | |
962 | * on their lnet_msg_t::msg_rxpeer */ | |
963 | for (i = 0; i < the_lnet.ln_nportals; i++) | |
964 | LNetClearLazyPortal(i); | |
965 | ||
966 | /* Clear the peer table and wait for all peers to go (they hold refs on | |
967 | * their NIs) */ | |
968 | lnet_peer_tables_cleanup(); | |
969 | ||
970 | lnet_net_lock(LNET_LOCK_EX); | |
971 | /* Now wait for the NI's I just nuked to show up on ln_zombie_nis | |
972 | * and shut them down in guaranteed thread context */ | |
973 | i = 2; | |
974 | while (!list_empty(&the_lnet.ln_nis_zombie)) { | |
975 | int *ref; | |
976 | int j; | |
977 | ||
978 | ni = list_entry(the_lnet.ln_nis_zombie.next, | |
979 | lnet_ni_t, ni_list); | |
980 | list_del_init(&ni->ni_list); | |
981 | cfs_percpt_for_each(ref, j, ni->ni_refs) { | |
982 | if (*ref == 0) | |
983 | continue; | |
984 | /* still busy, add it back to zombie list */ | |
985 | list_add(&ni->ni_list, &the_lnet.ln_nis_zombie); | |
986 | break; | |
987 | } | |
988 | ||
526cdb4f | 989 | if (!list_empty(&ni->ni_list)) { |
d7e09d03 PT |
990 | lnet_net_unlock(LNET_LOCK_EX); |
991 | ++i; | |
992 | if ((i & (-i)) == i) { | |
526cdb4f | 993 | CDEBUG(D_WARNING, "Waiting for zombie LNI %s\n", |
d7e09d03 PT |
994 | libcfs_nid2str(ni->ni_nid)); |
995 | } | |
d3caf4d5 PT |
996 | set_current_state(TASK_UNINTERRUPTIBLE); |
997 | schedule_timeout(cfs_time_seconds(1)); | |
d7e09d03 PT |
998 | lnet_net_lock(LNET_LOCK_EX); |
999 | continue; | |
1000 | } | |
1001 | ||
1002 | ni->ni_lnd->lnd_refcount--; | |
1003 | lnet_net_unlock(LNET_LOCK_EX); | |
1004 | ||
1005 | islo = ni->ni_lnd->lnd_type == LOLND; | |
1006 | ||
1007 | LASSERT (!in_interrupt ()); | |
1008 | (ni->ni_lnd->lnd_shutdown)(ni); | |
1009 | ||
1010 | /* can't deref lnd anymore now; it might have unregistered | |
1011 | * itself... */ | |
1012 | ||
1013 | if (!islo) | |
1014 | CDEBUG(D_LNI, "Removed LNI %s\n", | |
1015 | libcfs_nid2str(ni->ni_nid)); | |
1016 | ||
1017 | lnet_ni_free(ni); | |
526cdb4f LZ |
1018 | i = 2; |
1019 | ||
d7e09d03 PT |
1020 | lnet_net_lock(LNET_LOCK_EX); |
1021 | } | |
1022 | ||
1023 | the_lnet.ln_shutdown = 0; | |
1024 | lnet_net_unlock(LNET_LOCK_EX); | |
1025 | ||
1026 | if (the_lnet.ln_network_tokens != NULL) { | |
1027 | LIBCFS_FREE(the_lnet.ln_network_tokens, | |
1028 | the_lnet.ln_network_tokens_nob); | |
1029 | the_lnet.ln_network_tokens = NULL; | |
1030 | } | |
1031 | } | |
1032 | ||
51bd8814 | 1033 | static int |
d7e09d03 PT |
1034 | lnet_startup_lndnis (void) |
1035 | { | |
1036 | lnd_t *lnd; | |
1037 | struct lnet_ni *ni; | |
1038 | struct lnet_tx_queue *tq; | |
1039 | struct list_head nilist; | |
1040 | int i; | |
1041 | int rc = 0; | |
1042 | int lnd_type; | |
1043 | int nicount = 0; | |
1044 | char *nets = lnet_get_networks(); | |
1045 | ||
1046 | INIT_LIST_HEAD(&nilist); | |
1047 | ||
1048 | if (nets == NULL) | |
1049 | goto failed; | |
1050 | ||
1051 | rc = lnet_parse_networks(&nilist, nets); | |
1052 | if (rc != 0) | |
1053 | goto failed; | |
1054 | ||
1055 | while (!list_empty(&nilist)) { | |
1056 | ni = list_entry(nilist.next, lnet_ni_t, ni_list); | |
1057 | lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid)); | |
1058 | ||
1059 | LASSERT (libcfs_isknown_lnd(lnd_type)); | |
1060 | ||
1061 | if (lnd_type == CIBLND || | |
1062 | lnd_type == OPENIBLND || | |
1063 | lnd_type == IIBLND || | |
1064 | lnd_type == VIBLND) { | |
1065 | CERROR("LND %s obsoleted\n", | |
1066 | libcfs_lnd2str(lnd_type)); | |
1067 | goto failed; | |
1068 | } | |
1069 | ||
1070 | LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex); | |
1071 | lnd = lnet_find_lnd_by_type(lnd_type); | |
1072 | ||
1073 | if (lnd == NULL) { | |
1074 | LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); | |
1075 | rc = request_module("%s", | |
1076 | libcfs_lnd2modname(lnd_type)); | |
1077 | LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex); | |
1078 | ||
1079 | lnd = lnet_find_lnd_by_type(lnd_type); | |
1080 | if (lnd == NULL) { | |
1081 | LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); | |
1082 | CERROR("Can't load LND %s, module %s, rc=%d\n", | |
1083 | libcfs_lnd2str(lnd_type), | |
1084 | libcfs_lnd2modname(lnd_type), rc); | |
1085 | goto failed; | |
1086 | } | |
1087 | } | |
1088 | ||
1089 | lnet_net_lock(LNET_LOCK_EX); | |
1090 | lnd->lnd_refcount++; | |
1091 | lnet_net_unlock(LNET_LOCK_EX); | |
1092 | ||
1093 | ni->ni_lnd = lnd; | |
1094 | ||
1095 | rc = (lnd->lnd_startup)(ni); | |
1096 | ||
1097 | LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); | |
1098 | ||
1099 | if (rc != 0) { | |
1100 | LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s" | |
1101 | "\n", | |
1102 | rc, libcfs_lnd2str(lnd->lnd_type)); | |
1103 | lnet_net_lock(LNET_LOCK_EX); | |
1104 | lnd->lnd_refcount--; | |
1105 | lnet_net_unlock(LNET_LOCK_EX); | |
1106 | goto failed; | |
1107 | } | |
1108 | ||
1109 | LASSERT (ni->ni_peertimeout <= 0 || lnd->lnd_query != NULL); | |
1110 | ||
1111 | list_del(&ni->ni_list); | |
1112 | ||
1113 | lnet_net_lock(LNET_LOCK_EX); | |
1114 | /* refcount for ln_nis */ | |
1115 | lnet_ni_addref_locked(ni, 0); | |
1116 | list_add_tail(&ni->ni_list, &the_lnet.ln_nis); | |
1117 | if (ni->ni_cpts != NULL) { | |
1118 | list_add_tail(&ni->ni_cptlist, | |
1119 | &the_lnet.ln_nis_cpt); | |
1120 | lnet_ni_addref_locked(ni, 0); | |
1121 | } | |
1122 | ||
1123 | lnet_net_unlock(LNET_LOCK_EX); | |
1124 | ||
1125 | if (lnd->lnd_type == LOLND) { | |
1126 | lnet_ni_addref(ni); | |
1127 | LASSERT (the_lnet.ln_loni == NULL); | |
1128 | the_lnet.ln_loni = ni; | |
1129 | continue; | |
1130 | } | |
1131 | ||
1132 | if (ni->ni_peertxcredits == 0 || | |
1133 | ni->ni_maxtxcredits == 0) { | |
1134 | LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n", | |
1135 | libcfs_lnd2str(lnd->lnd_type), | |
1136 | ni->ni_peertxcredits == 0 ? | |
1137 | "" : "per-peer "); | |
1138 | goto failed; | |
1139 | } | |
1140 | ||
1141 | cfs_percpt_for_each(tq, i, ni->ni_tx_queues) { | |
1142 | tq->tq_credits_min = | |
1143 | tq->tq_credits_max = | |
1144 | tq->tq_credits = lnet_ni_tq_credits(ni); | |
1145 | } | |
1146 | ||
1147 | CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n", | |
1148 | libcfs_nid2str(ni->ni_nid), ni->ni_peertxcredits, | |
1149 | lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER, | |
1150 | ni->ni_peerrtrcredits, ni->ni_peertimeout); | |
1151 | ||
1152 | nicount++; | |
1153 | } | |
1154 | ||
1155 | if (the_lnet.ln_eq_waitni != NULL && nicount > 1) { | |
1156 | lnd_type = the_lnet.ln_eq_waitni->ni_lnd->lnd_type; | |
1157 | LCONSOLE_ERROR_MSG(0x109, "LND %s can only run single-network" | |
1158 | "\n", | |
1159 | libcfs_lnd2str(lnd_type)); | |
1160 | goto failed; | |
1161 | } | |
1162 | ||
1163 | return 0; | |
1164 | ||
1165 | failed: | |
1166 | lnet_shutdown_lndnis(); | |
1167 | ||
1168 | while (!list_empty(&nilist)) { | |
1169 | ni = list_entry(nilist.next, lnet_ni_t, ni_list); | |
1170 | list_del(&ni->ni_list); | |
1171 | lnet_ni_free(ni); | |
1172 | } | |
1173 | ||
1174 | return -ENETDOWN; | |
1175 | } | |
1176 | ||
1177 | /** | |
1178 | * Initialize LNet library. | |
1179 | * | |
1180 | * Only userspace program needs to call this function - it's automatically | |
1181 | * called in the kernel at module loading time. Caller has to call LNetFini() | |
1182 | * after a call to LNetInit(), if and only if the latter returned 0. It must | |
1183 | * be called exactly once. | |
1184 | * | |
1185 | * \return 0 on success, and -ve on failures. | |
1186 | */ | |
1187 | int | |
1188 | LNetInit(void) | |
1189 | { | |
1190 | int rc; | |
1191 | ||
1192 | lnet_assert_wire_constants(); | |
1193 | LASSERT(!the_lnet.ln_init); | |
1194 | ||
1195 | memset(&the_lnet, 0, sizeof(the_lnet)); | |
1196 | ||
1197 | /* refer to global cfs_cpt_table for now */ | |
1198 | the_lnet.ln_cpt_table = cfs_cpt_table; | |
1199 | the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table); | |
1200 | ||
1201 | LASSERT(the_lnet.ln_cpt_number > 0); | |
1202 | if (the_lnet.ln_cpt_number > LNET_CPT_MAX) { | |
1203 | /* we are under risk of consuming all lh_cookie */ | |
1204 | CERROR("Can't have %d CPTs for LNet (max allowed is %d), " | |
1205 | "please change setting of CPT-table and retry\n", | |
1206 | the_lnet.ln_cpt_number, LNET_CPT_MAX); | |
1207 | return -1; | |
1208 | } | |
1209 | ||
1210 | while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number) | |
1211 | the_lnet.ln_cpt_bits++; | |
1212 | ||
1213 | rc = lnet_create_locks(); | |
1214 | if (rc != 0) { | |
1215 | CERROR("Can't create LNet global locks: %d\n", rc); | |
1216 | return -1; | |
1217 | } | |
1218 | ||
1219 | the_lnet.ln_refcount = 0; | |
1220 | the_lnet.ln_init = 1; | |
1221 | LNetInvalidateHandle(&the_lnet.ln_rc_eqh); | |
1222 | INIT_LIST_HEAD(&the_lnet.ln_lnds); | |
1223 | INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie); | |
1224 | INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow); | |
1225 | ||
1226 | /* The hash table size is the number of bits it takes to express the set | |
1227 | * ln_num_routes, minus 1 (better to under estimate than over so we | |
1228 | * don't waste memory). */ | |
1229 | if (rnet_htable_size <= 0) | |
1230 | rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT; | |
1231 | else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX) | |
1232 | rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX; | |
1233 | the_lnet.ln_remote_nets_hbits = max_t(int, 1, | |
1234 | order_base_2(rnet_htable_size) - 1); | |
1235 | ||
1236 | /* All LNDs apart from the LOLND are in separate modules. They | |
1237 | * register themselves when their module loads, and unregister | |
1238 | * themselves when their module is unloaded. */ | |
1239 | lnet_register_lnd(&the_lolnd); | |
1240 | return 0; | |
1241 | } | |
1242 | EXPORT_SYMBOL(LNetInit); | |
1243 | ||
1244 | /** | |
1245 | * Finalize LNet library. | |
1246 | * | |
1247 | * Only userspace program needs to call this function. It can be called | |
1248 | * at most once. | |
1249 | * | |
1250 | * \pre LNetInit() called with success. | |
1251 | * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls. | |
1252 | */ | |
1253 | void | |
1254 | LNetFini(void) | |
1255 | { | |
1256 | LASSERT(the_lnet.ln_init); | |
1257 | LASSERT(the_lnet.ln_refcount == 0); | |
1258 | ||
1259 | while (!list_empty(&the_lnet.ln_lnds)) | |
1260 | lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next, | |
1261 | lnd_t, lnd_list)); | |
1262 | lnet_destroy_locks(); | |
1263 | ||
1264 | the_lnet.ln_init = 0; | |
1265 | } | |
1266 | EXPORT_SYMBOL(LNetFini); | |
1267 | ||
1268 | /** | |
1269 | * Set LNet PID and start LNet interfaces, routing, and forwarding. | |
1270 | * | |
1271 | * Userspace program should call this after a successful call to LNetInit(). | |
1272 | * Users must call this function at least once before any other functions. | |
1273 | * For each successful call there must be a corresponding call to | |
1274 | * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is | |
1275 | * ignored. | |
1276 | * | |
1277 | * The PID used by LNet may be different from the one requested. | |
1278 | * See LNetGetId(). | |
1279 | * | |
1280 | * \param requested_pid PID requested by the caller. | |
1281 | * | |
1282 | * \return >= 0 on success, and < 0 error code on failures. | |
1283 | */ | |
1284 | int | |
1285 | LNetNIInit(lnet_pid_t requested_pid) | |
1286 | { | |
1287 | int im_a_router = 0; | |
1288 | int rc; | |
1289 | ||
1290 | LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex); | |
1291 | ||
1292 | LASSERT (the_lnet.ln_init); | |
1293 | CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount); | |
1294 | ||
1295 | if (the_lnet.ln_refcount > 0) { | |
1296 | rc = the_lnet.ln_refcount++; | |
1297 | goto out; | |
1298 | } | |
1299 | ||
1300 | lnet_get_tunables(); | |
1301 | ||
1302 | if (requested_pid == LNET_PID_ANY) { | |
1303 | /* Don't instantiate LNET just for me */ | |
1304 | rc = -ENETDOWN; | |
1305 | goto failed0; | |
1306 | } | |
1307 | ||
1308 | rc = lnet_prepare(requested_pid); | |
1309 | if (rc != 0) | |
1310 | goto failed0; | |
1311 | ||
1312 | rc = lnet_startup_lndnis(); | |
1313 | if (rc != 0) | |
1314 | goto failed1; | |
1315 | ||
1316 | rc = lnet_parse_routes(lnet_get_routes(), &im_a_router); | |
1317 | if (rc != 0) | |
1318 | goto failed2; | |
1319 | ||
1320 | rc = lnet_check_routes(); | |
1321 | if (rc != 0) | |
1322 | goto failed2; | |
1323 | ||
1324 | rc = lnet_rtrpools_alloc(im_a_router); | |
1325 | if (rc != 0) | |
1326 | goto failed2; | |
1327 | ||
1328 | rc = lnet_acceptor_start(); | |
1329 | if (rc != 0) | |
1330 | goto failed2; | |
1331 | ||
1332 | the_lnet.ln_refcount = 1; | |
1333 | /* Now I may use my own API functions... */ | |
1334 | ||
1335 | /* NB router checker needs the_lnet.ln_ping_info in | |
1336 | * lnet_router_checker -> lnet_update_ni_status_locked */ | |
1337 | rc = lnet_ping_target_init(); | |
1338 | if (rc != 0) | |
1339 | goto failed3; | |
1340 | ||
1341 | rc = lnet_router_checker_start(); | |
1342 | if (rc != 0) | |
1343 | goto failed4; | |
1344 | ||
1345 | lnet_proc_init(); | |
1346 | goto out; | |
1347 | ||
1348 | failed4: | |
1349 | lnet_ping_target_fini(); | |
1350 | failed3: | |
1351 | the_lnet.ln_refcount = 0; | |
1352 | lnet_acceptor_stop(); | |
1353 | failed2: | |
1354 | lnet_destroy_routes(); | |
1355 | lnet_shutdown_lndnis(); | |
1356 | failed1: | |
1357 | lnet_unprepare(); | |
1358 | failed0: | |
1359 | LASSERT (rc < 0); | |
1360 | out: | |
1361 | LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex); | |
1362 | return rc; | |
1363 | } | |
1364 | EXPORT_SYMBOL(LNetNIInit); | |
1365 | ||
1366 | /** | |
1367 | * Stop LNet interfaces, routing, and forwarding. | |
1368 | * | |
1369 | * Users must call this function once for each successful call to LNetNIInit(). | |
1370 | * Once the LNetNIFini() operation has been started, the results of pending | |
1371 | * API operations are undefined. | |
1372 | * | |
1373 | * \return always 0 for current implementation. | |
1374 | */ | |
1375 | int | |
7d46a21a | 1376 | LNetNIFini(void) |
d7e09d03 PT |
1377 | { |
1378 | LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex); | |
1379 | ||
1380 | LASSERT (the_lnet.ln_init); | |
1381 | LASSERT (the_lnet.ln_refcount > 0); | |
1382 | ||
1383 | if (the_lnet.ln_refcount != 1) { | |
1384 | the_lnet.ln_refcount--; | |
1385 | } else { | |
1386 | LASSERT (!the_lnet.ln_niinit_self); | |
1387 | ||
1388 | lnet_proc_fini(); | |
1389 | lnet_router_checker_stop(); | |
1390 | lnet_ping_target_fini(); | |
1391 | ||
1392 | /* Teardown fns that use my own API functions BEFORE here */ | |
1393 | the_lnet.ln_refcount = 0; | |
1394 | ||
1395 | lnet_acceptor_stop(); | |
1396 | lnet_destroy_routes(); | |
1397 | lnet_shutdown_lndnis(); | |
1398 | lnet_unprepare(); | |
1399 | } | |
1400 | ||
1401 | LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex); | |
1402 | return 0; | |
1403 | } | |
1404 | EXPORT_SYMBOL(LNetNIFini); | |
1405 | ||
1406 | /** | |
1407 | * This is an ugly hack to export IOC_LIBCFS_DEBUG_PEER and | |
1408 | * IOC_LIBCFS_PORTALS_COMPATIBILITY commands to users, by tweaking the LNet | |
1409 | * internal ioctl handler. | |
1410 | * | |
1411 | * IOC_LIBCFS_PORTALS_COMPATIBILITY is now deprecated, don't use it. | |
1412 | * | |
1413 | * \param cmd IOC_LIBCFS_DEBUG_PEER to print debugging data about a peer. | |
1414 | * The data will be printed to system console. Don't use it excessively. | |
1415 | * \param arg A pointer to lnet_process_id_t, process ID of the peer. | |
1416 | * | |
1417 | * \return Always return 0 when called by users directly (i.e., not via ioctl). | |
1418 | */ | |
1419 | int | |
1420 | LNetCtl(unsigned int cmd, void *arg) | |
1421 | { | |
1422 | struct libcfs_ioctl_data *data = arg; | |
1423 | lnet_process_id_t id = {0}; | |
1424 | lnet_ni_t *ni; | |
1425 | int rc; | |
1426 | ||
1427 | LASSERT (the_lnet.ln_init); | |
1428 | LASSERT (the_lnet.ln_refcount > 0); | |
1429 | ||
1430 | switch (cmd) { | |
1431 | case IOC_LIBCFS_GET_NI: | |
1432 | rc = LNetGetId(data->ioc_count, &id); | |
1433 | data->ioc_nid = id.nid; | |
1434 | return rc; | |
1435 | ||
1436 | case IOC_LIBCFS_FAIL_NID: | |
1437 | return lnet_fail_nid(data->ioc_nid, data->ioc_count); | |
1438 | ||
1439 | case IOC_LIBCFS_ADD_ROUTE: | |
1440 | rc = lnet_add_route(data->ioc_net, data->ioc_count, | |
e75fb87f | 1441 | data->ioc_nid, data->ioc_priority); |
d7e09d03 PT |
1442 | return (rc != 0) ? rc : lnet_check_routes(); |
1443 | ||
1444 | case IOC_LIBCFS_DEL_ROUTE: | |
1445 | return lnet_del_route(data->ioc_net, data->ioc_nid); | |
1446 | ||
1447 | case IOC_LIBCFS_GET_ROUTE: | |
1448 | return lnet_get_route(data->ioc_count, | |
1449 | &data->ioc_net, &data->ioc_count, | |
e75fb87f DO |
1450 | &data->ioc_nid, &data->ioc_flags, |
1451 | &data->ioc_priority); | |
d7e09d03 PT |
1452 | case IOC_LIBCFS_NOTIFY_ROUTER: |
1453 | return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, | |
1454 | cfs_time_current() - | |
7264b8a5 | 1455 | cfs_time_seconds(get_seconds() - |
d7e09d03 PT |
1456 | (time_t)data->ioc_u64[0])); |
1457 | ||
1458 | case IOC_LIBCFS_PORTALS_COMPATIBILITY: | |
1459 | /* This can be removed once lustre stops calling it */ | |
1460 | return 0; | |
1461 | ||
1462 | case IOC_LIBCFS_LNET_DIST: | |
1463 | rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]); | |
1464 | if (rc < 0 && rc != -EHOSTUNREACH) | |
1465 | return rc; | |
1466 | ||
1467 | data->ioc_u32[0] = rc; | |
1468 | return 0; | |
1469 | ||
1470 | case IOC_LIBCFS_TESTPROTOCOMPAT: | |
1471 | lnet_net_lock(LNET_LOCK_EX); | |
1472 | the_lnet.ln_testprotocompat = data->ioc_flags; | |
1473 | lnet_net_unlock(LNET_LOCK_EX); | |
1474 | return 0; | |
1475 | ||
1476 | case IOC_LIBCFS_PING: | |
1477 | id.nid = data->ioc_nid; | |
1478 | id.pid = data->ioc_u32[0]; | |
1479 | rc = lnet_ping(id, data->ioc_u32[1], /* timeout */ | |
1480 | (lnet_process_id_t *)data->ioc_pbuf1, | |
1481 | data->ioc_plen1/sizeof(lnet_process_id_t)); | |
1482 | if (rc < 0) | |
1483 | return rc; | |
1484 | data->ioc_count = rc; | |
1485 | return 0; | |
1486 | ||
1487 | case IOC_LIBCFS_DEBUG_PEER: { | |
1488 | /* CAVEAT EMPTOR: this one designed for calling directly; not | |
1489 | * via an ioctl */ | |
1490 | id = *((lnet_process_id_t *) arg); | |
1491 | ||
1492 | lnet_debug_peer(id.nid); | |
1493 | ||
1494 | ni = lnet_net2ni(LNET_NIDNET(id.nid)); | |
1495 | if (ni == NULL) { | |
1496 | CDEBUG(D_WARNING, "No NI for %s\n", libcfs_id2str(id)); | |
1497 | } else { | |
1498 | if (ni->ni_lnd->lnd_ctl == NULL) { | |
1499 | CDEBUG(D_WARNING, "No ctl for %s\n", | |
1500 | libcfs_id2str(id)); | |
1501 | } else { | |
1502 | (void)ni->ni_lnd->lnd_ctl(ni, cmd, arg); | |
1503 | } | |
1504 | ||
1505 | lnet_ni_decref(ni); | |
1506 | } | |
1507 | return 0; | |
1508 | } | |
1509 | ||
1510 | default: | |
1511 | ni = lnet_net2ni(data->ioc_net); | |
1512 | if (ni == NULL) | |
1513 | return -EINVAL; | |
1514 | ||
1515 | if (ni->ni_lnd->lnd_ctl == NULL) | |
1516 | rc = -EINVAL; | |
1517 | else | |
1518 | rc = ni->ni_lnd->lnd_ctl(ni, cmd, arg); | |
1519 | ||
1520 | lnet_ni_decref(ni); | |
1521 | return rc; | |
1522 | } | |
1523 | /* not reached */ | |
1524 | } | |
1525 | EXPORT_SYMBOL(LNetCtl); | |
1526 | ||
1527 | /** | |
1528 | * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that | |
1529 | * all interfaces share a same PID, as requested by LNetNIInit(). | |
1530 | * | |
1531 | * \param index Index of the interface to look up. | |
1532 | * \param id On successful return, this location will hold the | |
1533 | * lnet_process_id_t ID of the interface. | |
1534 | * | |
1535 | * \retval 0 If an interface exists at \a index. | |
1536 | * \retval -ENOENT If no interface has been found. | |
1537 | */ | |
1538 | int | |
1539 | LNetGetId(unsigned int index, lnet_process_id_t *id) | |
1540 | { | |
1541 | struct lnet_ni *ni; | |
1542 | struct list_head *tmp; | |
1543 | int cpt; | |
1544 | int rc = -ENOENT; | |
1545 | ||
1546 | LASSERT(the_lnet.ln_init); | |
4b1e84ed PT |
1547 | |
1548 | /* LNetNI initilization failed? */ | |
1549 | if (the_lnet.ln_refcount == 0) | |
1550 | return rc; | |
d7e09d03 PT |
1551 | |
1552 | cpt = lnet_net_lock_current(); | |
1553 | ||
1554 | list_for_each(tmp, &the_lnet.ln_nis) { | |
1555 | if (index-- != 0) | |
1556 | continue; | |
1557 | ||
1558 | ni = list_entry(tmp, lnet_ni_t, ni_list); | |
1559 | ||
1560 | id->nid = ni->ni_nid; | |
1561 | id->pid = the_lnet.ln_pid; | |
1562 | rc = 0; | |
1563 | break; | |
1564 | } | |
1565 | ||
1566 | lnet_net_unlock(cpt); | |
1567 | return rc; | |
1568 | } | |
1569 | EXPORT_SYMBOL(LNetGetId); | |
1570 | ||
1571 | /** | |
1572 | * Print a string representation of handle \a h into buffer \a str of | |
1573 | * \a len bytes. | |
1574 | */ | |
1575 | void | |
1576 | LNetSnprintHandle(char *str, int len, lnet_handle_any_t h) | |
1577 | { | |
55f5a824 | 1578 | snprintf(str, len, "%#llx", h.cookie); |
d7e09d03 PT |
1579 | } |
1580 | EXPORT_SYMBOL(LNetSnprintHandle); | |
1581 | ||
1582 | static int | |
1583 | lnet_create_ping_info(void) | |
1584 | { | |
1585 | int i; | |
1586 | int n; | |
1587 | int rc; | |
1588 | unsigned int infosz; | |
1589 | lnet_ni_t *ni; | |
1590 | lnet_process_id_t id; | |
1591 | lnet_ping_info_t *pinfo; | |
1592 | ||
1593 | for (n = 0; ; n++) { | |
1594 | rc = LNetGetId(n, &id); | |
1595 | if (rc == -ENOENT) | |
1596 | break; | |
1597 | ||
1598 | LASSERT (rc == 0); | |
1599 | } | |
1600 | ||
1601 | infosz = offsetof(lnet_ping_info_t, pi_ni[n]); | |
1602 | LIBCFS_ALLOC(pinfo, infosz); | |
1603 | if (pinfo == NULL) { | |
1604 | CERROR("Can't allocate ping info[%d]\n", n); | |
1605 | return -ENOMEM; | |
1606 | } | |
1607 | ||
1608 | pinfo->pi_nnis = n; | |
1609 | pinfo->pi_pid = the_lnet.ln_pid; | |
1610 | pinfo->pi_magic = LNET_PROTO_PING_MAGIC; | |
1611 | pinfo->pi_features = LNET_PING_FEAT_NI_STATUS; | |
1612 | ||
1613 | for (i = 0; i < n; i++) { | |
1614 | lnet_ni_status_t *ns = &pinfo->pi_ni[i]; | |
1615 | ||
1616 | rc = LNetGetId(i, &id); | |
1617 | LASSERT (rc == 0); | |
1618 | ||
1619 | ns->ns_nid = id.nid; | |
1620 | ns->ns_status = LNET_NI_STATUS_UP; | |
1621 | ||
1622 | lnet_net_lock(0); | |
1623 | ||
1624 | ni = lnet_nid2ni_locked(id.nid, 0); | |
1625 | LASSERT(ni != NULL); | |
1626 | ||
1627 | lnet_ni_lock(ni); | |
1628 | LASSERT(ni->ni_status == NULL); | |
1629 | ni->ni_status = ns; | |
1630 | lnet_ni_unlock(ni); | |
1631 | ||
1632 | lnet_ni_decref_locked(ni, 0); | |
1633 | lnet_net_unlock(0); | |
1634 | } | |
1635 | ||
1636 | the_lnet.ln_ping_info = pinfo; | |
1637 | return 0; | |
1638 | } | |
1639 | ||
1640 | static void | |
1641 | lnet_destroy_ping_info(void) | |
1642 | { | |
1643 | struct lnet_ni *ni; | |
1644 | ||
1645 | lnet_net_lock(0); | |
1646 | ||
1647 | list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) { | |
1648 | lnet_ni_lock(ni); | |
1649 | ni->ni_status = NULL; | |
1650 | lnet_ni_unlock(ni); | |
1651 | } | |
1652 | ||
1653 | lnet_net_unlock(0); | |
1654 | ||
1655 | LIBCFS_FREE(the_lnet.ln_ping_info, | |
1656 | offsetof(lnet_ping_info_t, | |
1657 | pi_ni[the_lnet.ln_ping_info->pi_nnis])); | |
1658 | the_lnet.ln_ping_info = NULL; | |
1659 | return; | |
1660 | } | |
1661 | ||
1662 | int | |
1663 | lnet_ping_target_init(void) | |
1664 | { | |
1665 | lnet_md_t md = {0}; | |
1666 | lnet_handle_me_t meh; | |
1667 | lnet_process_id_t id; | |
1668 | int rc; | |
1669 | int rc2; | |
1670 | int infosz; | |
1671 | ||
1672 | rc = lnet_create_ping_info(); | |
1673 | if (rc != 0) | |
1674 | return rc; | |
1675 | ||
1676 | /* We can have a tiny EQ since we only need to see the unlink event on | |
1677 | * teardown, which by definition is the last one! */ | |
1678 | rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &the_lnet.ln_ping_target_eq); | |
1679 | if (rc != 0) { | |
1680 | CERROR("Can't allocate ping EQ: %d\n", rc); | |
1681 | goto failed_0; | |
1682 | } | |
1683 | ||
1684 | memset(&id, 0, sizeof(lnet_process_id_t)); | |
1685 | id.nid = LNET_NID_ANY; | |
1686 | id.pid = LNET_PID_ANY; | |
1687 | ||
1688 | rc = LNetMEAttach(LNET_RESERVED_PORTAL, id, | |
1689 | LNET_PROTO_PING_MATCHBITS, 0, | |
1690 | LNET_UNLINK, LNET_INS_AFTER, | |
1691 | &meh); | |
1692 | if (rc != 0) { | |
1693 | CERROR("Can't create ping ME: %d\n", rc); | |
1694 | goto failed_1; | |
1695 | } | |
1696 | ||
1697 | /* initialize md content */ | |
1698 | infosz = offsetof(lnet_ping_info_t, | |
1699 | pi_ni[the_lnet.ln_ping_info->pi_nnis]); | |
1700 | md.start = the_lnet.ln_ping_info; | |
1701 | md.length = infosz; | |
1702 | md.threshold = LNET_MD_THRESH_INF; | |
1703 | md.max_size = 0; | |
1704 | md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE | | |
1705 | LNET_MD_MANAGE_REMOTE; | |
1706 | md.user_ptr = NULL; | |
1707 | md.eq_handle = the_lnet.ln_ping_target_eq; | |
1708 | ||
1709 | rc = LNetMDAttach(meh, md, | |
1710 | LNET_RETAIN, | |
1711 | &the_lnet.ln_ping_target_md); | |
1712 | if (rc != 0) { | |
1713 | CERROR("Can't attach ping MD: %d\n", rc); | |
1714 | goto failed_2; | |
1715 | } | |
1716 | ||
1717 | return 0; | |
1718 | ||
1719 | failed_2: | |
1720 | rc2 = LNetMEUnlink(meh); | |
1721 | LASSERT (rc2 == 0); | |
1722 | failed_1: | |
1723 | rc2 = LNetEQFree(the_lnet.ln_ping_target_eq); | |
1724 | LASSERT (rc2 == 0); | |
1725 | failed_0: | |
1726 | lnet_destroy_ping_info(); | |
1727 | return rc; | |
1728 | } | |
1729 | ||
1730 | void | |
1731 | lnet_ping_target_fini(void) | |
1732 | { | |
1733 | lnet_event_t event; | |
1734 | int rc; | |
1735 | int which; | |
1736 | int timeout_ms = 1000; | |
1737 | sigset_t blocked = cfs_block_allsigs(); | |
1738 | ||
1739 | LNetMDUnlink(the_lnet.ln_ping_target_md); | |
1740 | /* NB md could be busy; this just starts the unlink */ | |
1741 | ||
1742 | for (;;) { | |
1743 | rc = LNetEQPoll(&the_lnet.ln_ping_target_eq, 1, | |
1744 | timeout_ms, &event, &which); | |
1745 | ||
1746 | /* I expect overflow... */ | |
1747 | LASSERT (rc >= 0 || rc == -EOVERFLOW); | |
1748 | ||
1749 | if (rc == 0) { | |
1750 | /* timed out: provide a diagnostic */ | |
1751 | CWARN("Still waiting for ping MD to unlink\n"); | |
1752 | timeout_ms *= 2; | |
1753 | continue; | |
1754 | } | |
1755 | ||
1756 | /* Got a valid event */ | |
1757 | if (event.unlinked) | |
1758 | break; | |
1759 | } | |
1760 | ||
1761 | rc = LNetEQFree(the_lnet.ln_ping_target_eq); | |
1762 | LASSERT (rc == 0); | |
1763 | lnet_destroy_ping_info(); | |
1764 | cfs_restore_sigs(blocked); | |
1765 | } | |
1766 | ||
1767 | int | |
1768 | lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_ids) | |
1769 | { | |
1770 | lnet_handle_eq_t eqh; | |
1771 | lnet_handle_md_t mdh; | |
1772 | lnet_event_t event; | |
1773 | lnet_md_t md = {0}; | |
1774 | int which; | |
1775 | int unlinked = 0; | |
1776 | int replied = 0; | |
1777 | const int a_long_time = 60000; /* mS */ | |
1778 | int infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]); | |
1779 | lnet_ping_info_t *info; | |
1780 | lnet_process_id_t tmpid; | |
1781 | int i; | |
1782 | int nob; | |
1783 | int rc; | |
1784 | int rc2; | |
1785 | sigset_t blocked; | |
1786 | ||
1787 | if (n_ids <= 0 || | |
1788 | id.nid == LNET_NID_ANY || | |
1789 | timeout_ms > 500000 || /* arbitrary limit! */ | |
1790 | n_ids > 20) /* arbitrary limit! */ | |
1791 | return -EINVAL; | |
1792 | ||
1793 | if (id.pid == LNET_PID_ANY) | |
1794 | id.pid = LUSTRE_SRV_LNET_PID; | |
1795 | ||
1796 | LIBCFS_ALLOC(info, infosz); | |
1797 | if (info == NULL) | |
1798 | return -ENOMEM; | |
1799 | ||
1800 | /* NB 2 events max (including any unlink event) */ | |
1801 | rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh); | |
1802 | if (rc != 0) { | |
1803 | CERROR("Can't allocate EQ: %d\n", rc); | |
1804 | goto out_0; | |
1805 | } | |
1806 | ||
1807 | /* initialize md content */ | |
1808 | md.start = info; | |
1809 | md.length = infosz; | |
1810 | md.threshold = 2; /*GET/REPLY*/ | |
1811 | md.max_size = 0; | |
1812 | md.options = LNET_MD_TRUNCATE; | |
1813 | md.user_ptr = NULL; | |
1814 | md.eq_handle = eqh; | |
1815 | ||
1816 | rc = LNetMDBind(md, LNET_UNLINK, &mdh); | |
1817 | if (rc != 0) { | |
1818 | CERROR("Can't bind MD: %d\n", rc); | |
1819 | goto out_1; | |
1820 | } | |
1821 | ||
1822 | rc = LNetGet(LNET_NID_ANY, mdh, id, | |
1823 | LNET_RESERVED_PORTAL, | |
1824 | LNET_PROTO_PING_MATCHBITS, 0); | |
1825 | ||
1826 | if (rc != 0) { | |
1827 | /* Don't CERROR; this could be deliberate! */ | |
1828 | ||
1829 | rc2 = LNetMDUnlink(mdh); | |
1830 | LASSERT (rc2 == 0); | |
1831 | ||
1832 | /* NB must wait for the UNLINK event below... */ | |
1833 | unlinked = 1; | |
1834 | timeout_ms = a_long_time; | |
1835 | } | |
1836 | ||
1837 | do { | |
1838 | /* MUST block for unlink to complete */ | |
1839 | if (unlinked) | |
1840 | blocked = cfs_block_allsigs(); | |
1841 | ||
1842 | rc2 = LNetEQPoll(&eqh, 1, timeout_ms, &event, &which); | |
1843 | ||
1844 | if (unlinked) | |
1845 | cfs_restore_sigs(blocked); | |
1846 | ||
1847 | CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2, | |
1848 | (rc2 <= 0) ? -1 : event.type, | |
1849 | (rc2 <= 0) ? -1 : event.status, | |
1850 | (rc2 > 0 && event.unlinked) ? " unlinked" : ""); | |
1851 | ||
1852 | LASSERT (rc2 != -EOVERFLOW); /* can't miss anything */ | |
1853 | ||
1854 | if (rc2 <= 0 || event.status != 0) { | |
1855 | /* timeout or error */ | |
1856 | if (!replied && rc == 0) | |
1857 | rc = (rc2 < 0) ? rc2 : | |
1858 | (rc2 == 0) ? -ETIMEDOUT : | |
1859 | event.status; | |
1860 | ||
1861 | if (!unlinked) { | |
1862 | /* Ensure completion in finite time... */ | |
1863 | LNetMDUnlink(mdh); | |
1864 | /* No assertion (racing with network) */ | |
1865 | unlinked = 1; | |
1866 | timeout_ms = a_long_time; | |
1867 | } else if (rc2 == 0) { | |
1868 | /* timed out waiting for unlink */ | |
1869 | CWARN("ping %s: late network completion\n", | |
1870 | libcfs_id2str(id)); | |
1871 | } | |
1872 | } else if (event.type == LNET_EVENT_REPLY) { | |
1873 | replied = 1; | |
1874 | rc = event.mlength; | |
1875 | } | |
1876 | ||
1877 | } while (rc2 <= 0 || !event.unlinked); | |
1878 | ||
1879 | if (!replied) { | |
1880 | if (rc >= 0) | |
1881 | CWARN("%s: Unexpected rc >= 0 but no reply!\n", | |
1882 | libcfs_id2str(id)); | |
1883 | rc = -EIO; | |
1884 | goto out_1; | |
1885 | } | |
1886 | ||
1887 | nob = rc; | |
1888 | LASSERT (nob >= 0 && nob <= infosz); | |
1889 | ||
1890 | rc = -EPROTO; /* if I can't parse... */ | |
1891 | ||
1892 | if (nob < 8) { | |
1893 | /* can't check magic/version */ | |
1894 | CERROR("%s: ping info too short %d\n", | |
1895 | libcfs_id2str(id), nob); | |
1896 | goto out_1; | |
1897 | } | |
1898 | ||
1899 | if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) { | |
1900 | lnet_swap_pinginfo(info); | |
1901 | } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) { | |
1902 | CERROR("%s: Unexpected magic %08x\n", | |
1903 | libcfs_id2str(id), info->pi_magic); | |
1904 | goto out_1; | |
1905 | } | |
1906 | ||
1907 | if ((info->pi_features & LNET_PING_FEAT_NI_STATUS) == 0) { | |
1908 | CERROR("%s: ping w/o NI status: 0x%x\n", | |
1909 | libcfs_id2str(id), info->pi_features); | |
1910 | goto out_1; | |
1911 | } | |
1912 | ||
1913 | if (nob < offsetof(lnet_ping_info_t, pi_ni[0])) { | |
1914 | CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id), | |
1915 | nob, (int)offsetof(lnet_ping_info_t, pi_ni[0])); | |
1916 | goto out_1; | |
1917 | } | |
1918 | ||
1919 | if (info->pi_nnis < n_ids) | |
1920 | n_ids = info->pi_nnis; | |
1921 | ||
1922 | if (nob < offsetof(lnet_ping_info_t, pi_ni[n_ids])) { | |
1923 | CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id), | |
1924 | nob, (int)offsetof(lnet_ping_info_t, pi_ni[n_ids])); | |
1925 | goto out_1; | |
1926 | } | |
1927 | ||
1928 | rc = -EFAULT; /* If I SEGV... */ | |
1929 | ||
751a624a | 1930 | memset(&tmpid, 0, sizeof(tmpid)); |
d7e09d03 PT |
1931 | for (i = 0; i < n_ids; i++) { |
1932 | tmpid.pid = info->pi_pid; | |
1933 | tmpid.nid = info->pi_ni[i].ns_nid; | |
1934 | if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid))) | |
1935 | goto out_1; | |
1936 | } | |
1937 | rc = info->pi_nnis; | |
1938 | ||
1939 | out_1: | |
1940 | rc2 = LNetEQFree(eqh); | |
1941 | if (rc2 != 0) | |
1942 | CERROR("rc2 %d\n", rc2); | |
1943 | LASSERT (rc2 == 0); | |
1944 | ||
1945 | out_0: | |
1946 | LIBCFS_FREE(info, infosz); | |
1947 | return rc; | |
1948 | } |