4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LNET
38 #include <linux/log2.h>
39 #include <linux/ktime.h>
41 #include "../../include/linux/lnet/lib-lnet.h"
43 #define D_LNI D_CONSOLE
45 lnet_t the_lnet
; /* THE state of the network */
46 EXPORT_SYMBOL(the_lnet
);
49 static char *ip2nets
= "";
50 module_param(ip2nets
, charp
, 0444);
51 MODULE_PARM_DESC(ip2nets
, "LNET network <- IP table");
53 static char *networks
= "";
54 module_param(networks
, charp
, 0444);
55 MODULE_PARM_DESC(networks
, "local networks");
57 static char *routes
= "";
58 module_param(routes
, charp
, 0444);
59 MODULE_PARM_DESC(routes
, "routes to non-local networks");
61 static int rnet_htable_size
= LNET_REMOTE_NETS_HASH_DEFAULT
;
62 module_param(rnet_htable_size
, int, 0444);
63 MODULE_PARM_DESC(rnet_htable_size
, "size of remote network hash table");
72 lnet_get_networks(void)
77 if (*networks
!= 0 && *ip2nets
!= 0) {
78 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or 'ip2nets' but not both at once\n");
83 rc
= lnet_parse_ip2nets(&nets
, ip2nets
);
84 return (rc
== 0) ? nets
: NULL
;
96 spin_lock_init(&the_lnet
.ln_eq_wait_lock
);
97 init_waitqueue_head(&the_lnet
.ln_eq_waitq
);
98 mutex_init(&the_lnet
.ln_lnd_mutex
);
99 mutex_init(&the_lnet
.ln_api_mutex
);
103 lnet_create_remote_nets_table(void)
106 struct list_head
*hash
;
108 LASSERT(the_lnet
.ln_remote_nets_hash
== NULL
);
109 LASSERT(the_lnet
.ln_remote_nets_hbits
> 0);
110 LIBCFS_ALLOC(hash
, LNET_REMOTE_NETS_HASH_SIZE
* sizeof(*hash
));
112 CERROR("Failed to create remote nets hash table\n");
116 for (i
= 0; i
< LNET_REMOTE_NETS_HASH_SIZE
; i
++)
117 INIT_LIST_HEAD(&hash
[i
]);
118 the_lnet
.ln_remote_nets_hash
= hash
;
123 lnet_destroy_remote_nets_table(void)
127 if (the_lnet
.ln_remote_nets_hash
== NULL
)
130 for (i
= 0; i
< LNET_REMOTE_NETS_HASH_SIZE
; i
++)
131 LASSERT(list_empty(&the_lnet
.ln_remote_nets_hash
[i
]));
133 LIBCFS_FREE(the_lnet
.ln_remote_nets_hash
,
134 LNET_REMOTE_NETS_HASH_SIZE
*
135 sizeof(the_lnet
.ln_remote_nets_hash
[0]));
136 the_lnet
.ln_remote_nets_hash
= NULL
;
140 lnet_destroy_locks(void)
142 if (the_lnet
.ln_res_lock
!= NULL
) {
143 cfs_percpt_lock_free(the_lnet
.ln_res_lock
);
144 the_lnet
.ln_res_lock
= NULL
;
147 if (the_lnet
.ln_net_lock
!= NULL
) {
148 cfs_percpt_lock_free(the_lnet
.ln_net_lock
);
149 the_lnet
.ln_net_lock
= NULL
;
154 lnet_create_locks(void)
158 the_lnet
.ln_res_lock
= cfs_percpt_lock_alloc(lnet_cpt_table());
159 if (the_lnet
.ln_res_lock
== NULL
)
162 the_lnet
.ln_net_lock
= cfs_percpt_lock_alloc(lnet_cpt_table());
163 if (the_lnet
.ln_net_lock
== NULL
)
169 lnet_destroy_locks();
173 static void lnet_assert_wire_constants(void)
175 /* Wire protocol assertions generated by 'wirecheck'
176 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
177 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
178 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
181 CLASSERT(LNET_PROTO_TCP_MAGIC
== 0xeebc0ded);
182 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR
== 1);
183 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR
== 0);
184 CLASSERT(LNET_MSG_ACK
== 0);
185 CLASSERT(LNET_MSG_PUT
== 1);
186 CLASSERT(LNET_MSG_GET
== 2);
187 CLASSERT(LNET_MSG_REPLY
== 3);
188 CLASSERT(LNET_MSG_HELLO
== 4);
190 /* Checks for struct ptl_handle_wire_t */
191 CLASSERT((int)sizeof(lnet_handle_wire_t
) == 16);
192 CLASSERT((int)offsetof(lnet_handle_wire_t
, wh_interface_cookie
) == 0);
193 CLASSERT((int)sizeof(((lnet_handle_wire_t
*)0)->wh_interface_cookie
) == 8);
194 CLASSERT((int)offsetof(lnet_handle_wire_t
, wh_object_cookie
) == 8);
195 CLASSERT((int)sizeof(((lnet_handle_wire_t
*)0)->wh_object_cookie
) == 8);
197 /* Checks for struct lnet_magicversion_t */
198 CLASSERT((int)sizeof(lnet_magicversion_t
) == 8);
199 CLASSERT((int)offsetof(lnet_magicversion_t
, magic
) == 0);
200 CLASSERT((int)sizeof(((lnet_magicversion_t
*)0)->magic
) == 4);
201 CLASSERT((int)offsetof(lnet_magicversion_t
, version_major
) == 4);
202 CLASSERT((int)sizeof(((lnet_magicversion_t
*)0)->version_major
) == 2);
203 CLASSERT((int)offsetof(lnet_magicversion_t
, version_minor
) == 6);
204 CLASSERT((int)sizeof(((lnet_magicversion_t
*)0)->version_minor
) == 2);
206 /* Checks for struct lnet_hdr_t */
207 CLASSERT((int)sizeof(lnet_hdr_t
) == 72);
208 CLASSERT((int)offsetof(lnet_hdr_t
, dest_nid
) == 0);
209 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->dest_nid
) == 8);
210 CLASSERT((int)offsetof(lnet_hdr_t
, src_nid
) == 8);
211 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->src_nid
) == 8);
212 CLASSERT((int)offsetof(lnet_hdr_t
, dest_pid
) == 16);
213 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->dest_pid
) == 4);
214 CLASSERT((int)offsetof(lnet_hdr_t
, src_pid
) == 20);
215 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->src_pid
) == 4);
216 CLASSERT((int)offsetof(lnet_hdr_t
, type
) == 24);
217 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->type
) == 4);
218 CLASSERT((int)offsetof(lnet_hdr_t
, payload_length
) == 28);
219 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->payload_length
) == 4);
220 CLASSERT((int)offsetof(lnet_hdr_t
, msg
) == 32);
221 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
) == 40);
224 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.ack
.dst_wmd
) == 32);
225 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.ack
.dst_wmd
) == 16);
226 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.ack
.match_bits
) == 48);
227 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.ack
.match_bits
) == 8);
228 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.ack
.mlength
) == 56);
229 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.ack
.mlength
) == 4);
232 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.ack_wmd
) == 32);
233 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.ack_wmd
) == 16);
234 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.match_bits
) == 48);
235 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.match_bits
) == 8);
236 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.hdr_data
) == 56);
237 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.hdr_data
) == 8);
238 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.ptl_index
) == 64);
239 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.ptl_index
) == 4);
240 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.offset
) == 68);
241 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.offset
) == 4);
244 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.return_wmd
) == 32);
245 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.return_wmd
) == 16);
246 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.match_bits
) == 48);
247 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.match_bits
) == 8);
248 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.ptl_index
) == 56);
249 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.ptl_index
) == 4);
250 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.src_offset
) == 60);
251 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.src_offset
) == 4);
252 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.sink_length
) == 64);
253 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.sink_length
) == 4);
256 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.reply
.dst_wmd
) == 32);
257 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.reply
.dst_wmd
) == 16);
260 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.hello
.incarnation
) == 32);
261 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.hello
.incarnation
) == 8);
262 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.hello
.type
) == 40);
263 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.hello
.type
) == 4);
267 lnet_find_lnd_by_type(int type
)
270 struct list_head
*tmp
;
272 /* holding lnd mutex */
273 list_for_each(tmp
, &the_lnet
.ln_lnds
) {
274 lnd
= list_entry(tmp
, lnd_t
, lnd_list
);
276 if ((int)lnd
->lnd_type
== type
)
284 lnet_register_lnd(lnd_t
*lnd
)
286 mutex_lock(&the_lnet
.ln_lnd_mutex
);
288 LASSERT(the_lnet
.ln_init
);
289 LASSERT(libcfs_isknown_lnd(lnd
->lnd_type
));
290 LASSERT(lnet_find_lnd_by_type(lnd
->lnd_type
) == NULL
);
292 list_add_tail(&lnd
->lnd_list
, &the_lnet
.ln_lnds
);
293 lnd
->lnd_refcount
= 0;
295 CDEBUG(D_NET
, "%s LND registered\n", libcfs_lnd2str(lnd
->lnd_type
));
297 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
299 EXPORT_SYMBOL(lnet_register_lnd
);
302 lnet_unregister_lnd(lnd_t
*lnd
)
304 mutex_lock(&the_lnet
.ln_lnd_mutex
);
306 LASSERT(the_lnet
.ln_init
);
307 LASSERT(lnet_find_lnd_by_type(lnd
->lnd_type
) == lnd
);
308 LASSERT(lnd
->lnd_refcount
== 0);
310 list_del(&lnd
->lnd_list
);
311 CDEBUG(D_NET
, "%s LND unregistered\n", libcfs_lnd2str(lnd
->lnd_type
));
313 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
315 EXPORT_SYMBOL(lnet_unregister_lnd
);
318 lnet_counters_get(lnet_counters_t
*counters
)
320 lnet_counters_t
*ctr
;
323 memset(counters
, 0, sizeof(*counters
));
325 lnet_net_lock(LNET_LOCK_EX
);
327 cfs_percpt_for_each(ctr
, i
, the_lnet
.ln_counters
) {
328 counters
->msgs_max
+= ctr
->msgs_max
;
329 counters
->msgs_alloc
+= ctr
->msgs_alloc
;
330 counters
->errors
+= ctr
->errors
;
331 counters
->send_count
+= ctr
->send_count
;
332 counters
->recv_count
+= ctr
->recv_count
;
333 counters
->route_count
+= ctr
->route_count
;
334 counters
->drop_count
+= ctr
->drop_count
;
335 counters
->send_length
+= ctr
->send_length
;
336 counters
->recv_length
+= ctr
->recv_length
;
337 counters
->route_length
+= ctr
->route_length
;
338 counters
->drop_length
+= ctr
->drop_length
;
341 lnet_net_unlock(LNET_LOCK_EX
);
343 EXPORT_SYMBOL(lnet_counters_get
);
346 lnet_counters_reset(void)
348 lnet_counters_t
*counters
;
351 lnet_net_lock(LNET_LOCK_EX
);
353 cfs_percpt_for_each(counters
, i
, the_lnet
.ln_counters
)
354 memset(counters
, 0, sizeof(lnet_counters_t
));
356 lnet_net_unlock(LNET_LOCK_EX
);
358 EXPORT_SYMBOL(lnet_counters_reset
);
361 lnet_create_interface_cookie(void)
363 /* NB the interface cookie in wire handles guards against delayed
364 * replies and ACKs appearing valid after reboot.
366 return ktime_get_ns();
370 lnet_res_type2str(int type
)
375 case LNET_COOKIE_TYPE_MD
:
377 case LNET_COOKIE_TYPE_ME
:
379 case LNET_COOKIE_TYPE_EQ
:
385 lnet_res_container_cleanup(struct lnet_res_container
*rec
)
389 if (rec
->rec_type
== 0) /* not set yet, it's uninitialized */
392 while (!list_empty(&rec
->rec_active
)) {
393 struct list_head
*e
= rec
->rec_active
.next
;
396 if (rec
->rec_type
== LNET_COOKIE_TYPE_EQ
) {
397 lnet_eq_free(list_entry(e
, lnet_eq_t
, eq_list
));
399 } else if (rec
->rec_type
== LNET_COOKIE_TYPE_MD
) {
400 lnet_md_free(list_entry(e
, lnet_libmd_t
, md_list
));
402 } else { /* NB: Active MEs should be attached on portals */
409 /* Found alive MD/ME/EQ, user really should unlink/free
410 * all of them before finalize LNet, but if someone didn't,
411 * we have to recycle garbage for him */
412 CERROR("%d active elements on exit of %s container\n",
413 count
, lnet_res_type2str(rec
->rec_type
));
416 if (rec
->rec_lh_hash
!= NULL
) {
417 LIBCFS_FREE(rec
->rec_lh_hash
,
418 LNET_LH_HASH_SIZE
* sizeof(rec
->rec_lh_hash
[0]));
419 rec
->rec_lh_hash
= NULL
;
422 rec
->rec_type
= 0; /* mark it as finalized */
426 lnet_res_container_setup(struct lnet_res_container
*rec
, int cpt
, int type
)
431 LASSERT(rec
->rec_type
== 0);
433 rec
->rec_type
= type
;
434 INIT_LIST_HEAD(&rec
->rec_active
);
435 rec
->rec_lh_cookie
= (cpt
<< LNET_COOKIE_TYPE_BITS
) | type
;
437 /* Arbitrary choice of hash table size */
438 LIBCFS_CPT_ALLOC(rec
->rec_lh_hash
, lnet_cpt_table(), cpt
,
439 LNET_LH_HASH_SIZE
* sizeof(rec
->rec_lh_hash
[0]));
440 if (rec
->rec_lh_hash
== NULL
) {
445 for (i
= 0; i
< LNET_LH_HASH_SIZE
; i
++)
446 INIT_LIST_HEAD(&rec
->rec_lh_hash
[i
]);
451 CERROR("Failed to setup %s resource container\n",
452 lnet_res_type2str(type
));
453 lnet_res_container_cleanup(rec
);
458 lnet_res_containers_destroy(struct lnet_res_container
**recs
)
460 struct lnet_res_container
*rec
;
463 cfs_percpt_for_each(rec
, i
, recs
)
464 lnet_res_container_cleanup(rec
);
466 cfs_percpt_free(recs
);
469 static struct lnet_res_container
**
470 lnet_res_containers_create(int type
)
472 struct lnet_res_container
**recs
;
473 struct lnet_res_container
*rec
;
477 recs
= cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec
));
479 CERROR("Failed to allocate %s resource containers\n",
480 lnet_res_type2str(type
));
484 cfs_percpt_for_each(rec
, i
, recs
) {
485 rc
= lnet_res_container_setup(rec
, i
, type
);
487 lnet_res_containers_destroy(recs
);
496 lnet_res_lh_lookup(struct lnet_res_container
*rec
, __u64 cookie
)
498 /* ALWAYS called with lnet_res_lock held */
499 struct list_head
*head
;
500 lnet_libhandle_t
*lh
;
503 if ((cookie
& LNET_COOKIE_MASK
) != rec
->rec_type
)
506 hash
= cookie
>> (LNET_COOKIE_TYPE_BITS
+ LNET_CPT_BITS
);
507 head
= &rec
->rec_lh_hash
[hash
& LNET_LH_HASH_MASK
];
509 list_for_each_entry(lh
, head
, lh_hash_chain
) {
510 if (lh
->lh_cookie
== cookie
)
518 lnet_res_lh_initialize(struct lnet_res_container
*rec
, lnet_libhandle_t
*lh
)
520 /* ALWAYS called with lnet_res_lock held */
521 unsigned int ibits
= LNET_COOKIE_TYPE_BITS
+ LNET_CPT_BITS
;
524 lh
->lh_cookie
= rec
->rec_lh_cookie
;
525 rec
->rec_lh_cookie
+= 1 << ibits
;
527 hash
= (lh
->lh_cookie
>> ibits
) & LNET_LH_HASH_MASK
;
529 list_add(&lh
->lh_hash_chain
, &rec
->rec_lh_hash
[hash
]);
533 int lnet_unprepare(void);
536 lnet_prepare(lnet_pid_t requested_pid
)
538 /* Prepare to bring up the network */
539 struct lnet_res_container
**recs
;
542 LASSERT(the_lnet
.ln_refcount
== 0);
544 the_lnet
.ln_routing
= 0;
546 LASSERT((requested_pid
& LNET_PID_USERFLAG
) == 0);
547 the_lnet
.ln_pid
= requested_pid
;
549 INIT_LIST_HEAD(&the_lnet
.ln_test_peers
);
550 INIT_LIST_HEAD(&the_lnet
.ln_nis
);
551 INIT_LIST_HEAD(&the_lnet
.ln_nis_cpt
);
552 INIT_LIST_HEAD(&the_lnet
.ln_nis_zombie
);
553 INIT_LIST_HEAD(&the_lnet
.ln_routers
);
555 rc
= lnet_create_remote_nets_table();
559 the_lnet
.ln_interface_cookie
= lnet_create_interface_cookie();
561 the_lnet
.ln_counters
= cfs_percpt_alloc(lnet_cpt_table(),
562 sizeof(lnet_counters_t
));
563 if (the_lnet
.ln_counters
== NULL
) {
564 CERROR("Failed to allocate counters for LNet\n");
569 rc
= lnet_peer_tables_create();
573 rc
= lnet_msg_containers_create();
577 rc
= lnet_res_container_setup(&the_lnet
.ln_eq_container
, 0,
578 LNET_COOKIE_TYPE_EQ
);
582 recs
= lnet_res_containers_create(LNET_COOKIE_TYPE_ME
);
588 the_lnet
.ln_me_containers
= recs
;
590 recs
= lnet_res_containers_create(LNET_COOKIE_TYPE_MD
);
596 the_lnet
.ln_md_containers
= recs
;
598 rc
= lnet_portals_create();
600 CERROR("Failed to create portals for LNet: %d\n", rc
);
614 /* NB no LNET_LOCK since this is the last reference. All LND instances
615 * have shut down already, so it is safe to unlink and free all
616 * descriptors, even those that appear committed to a network op (eg MD
617 * with non-zero pending count) */
619 lnet_fail_nid(LNET_NID_ANY
, 0);
621 LASSERT(the_lnet
.ln_refcount
== 0);
622 LASSERT(list_empty(&the_lnet
.ln_test_peers
));
623 LASSERT(list_empty(&the_lnet
.ln_nis
));
624 LASSERT(list_empty(&the_lnet
.ln_nis_cpt
));
625 LASSERT(list_empty(&the_lnet
.ln_nis_zombie
));
627 lnet_portals_destroy();
629 if (the_lnet
.ln_md_containers
!= NULL
) {
630 lnet_res_containers_destroy(the_lnet
.ln_md_containers
);
631 the_lnet
.ln_md_containers
= NULL
;
634 if (the_lnet
.ln_me_containers
!= NULL
) {
635 lnet_res_containers_destroy(the_lnet
.ln_me_containers
);
636 the_lnet
.ln_me_containers
= NULL
;
639 lnet_res_container_cleanup(&the_lnet
.ln_eq_container
);
641 lnet_msg_containers_destroy();
642 lnet_peer_tables_destroy();
643 lnet_rtrpools_free();
645 if (the_lnet
.ln_counters
!= NULL
) {
646 cfs_percpt_free(the_lnet
.ln_counters
);
647 the_lnet
.ln_counters
= NULL
;
649 lnet_destroy_remote_nets_table();
655 lnet_net2ni_locked(__u32 net
, int cpt
)
657 struct list_head
*tmp
;
660 LASSERT(cpt
!= LNET_LOCK_EX
);
662 list_for_each(tmp
, &the_lnet
.ln_nis
) {
663 ni
= list_entry(tmp
, lnet_ni_t
, ni_list
);
665 if (LNET_NIDNET(ni
->ni_nid
) == net
) {
666 lnet_ni_addref_locked(ni
, cpt
);
675 lnet_net2ni(__u32 net
)
680 ni
= lnet_net2ni_locked(net
, 0);
685 EXPORT_SYMBOL(lnet_net2ni
);
688 lnet_nid_cpt_hash(lnet_nid_t nid
, unsigned int number
)
693 LASSERT(number
>= 1 && number
<= LNET_CPT_NUMBER
);
698 val
= hash_long(key
, LNET_CPT_BITS
);
699 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
703 return (unsigned int)(key
+ val
+ (val
>> 1)) % number
;
707 lnet_cpt_of_nid_locked(lnet_nid_t nid
)
711 /* must called with hold of lnet_net_lock */
712 if (LNET_CPT_NUMBER
== 1)
713 return 0; /* the only one */
715 /* take lnet_net_lock(any) would be OK */
716 if (!list_empty(&the_lnet
.ln_nis_cpt
)) {
717 list_for_each_entry(ni
, &the_lnet
.ln_nis_cpt
, ni_cptlist
) {
718 if (LNET_NIDNET(ni
->ni_nid
) != LNET_NIDNET(nid
))
721 LASSERT(ni
->ni_cpts
!= NULL
);
722 return ni
->ni_cpts
[lnet_nid_cpt_hash
723 (nid
, ni
->ni_ncpts
)];
727 return lnet_nid_cpt_hash(nid
, LNET_CPT_NUMBER
);
731 lnet_cpt_of_nid(lnet_nid_t nid
)
736 if (LNET_CPT_NUMBER
== 1)
737 return 0; /* the only one */
739 if (list_empty(&the_lnet
.ln_nis_cpt
))
740 return lnet_nid_cpt_hash(nid
, LNET_CPT_NUMBER
);
742 cpt
= lnet_net_lock_current();
743 cpt2
= lnet_cpt_of_nid_locked(nid
);
744 lnet_net_unlock(cpt
);
748 EXPORT_SYMBOL(lnet_cpt_of_nid
);
751 lnet_islocalnet(__u32 net
)
756 cpt
= lnet_net_lock_current();
758 ni
= lnet_net2ni_locked(net
, cpt
);
760 lnet_ni_decref_locked(ni
, cpt
);
762 lnet_net_unlock(cpt
);
768 lnet_nid2ni_locked(lnet_nid_t nid
, int cpt
)
771 struct list_head
*tmp
;
773 LASSERT(cpt
!= LNET_LOCK_EX
);
775 list_for_each(tmp
, &the_lnet
.ln_nis
) {
776 ni
= list_entry(tmp
, lnet_ni_t
, ni_list
);
778 if (ni
->ni_nid
== nid
) {
779 lnet_ni_addref_locked(ni
, cpt
);
788 lnet_islocalnid(lnet_nid_t nid
)
793 cpt
= lnet_net_lock_current();
794 ni
= lnet_nid2ni_locked(nid
, cpt
);
796 lnet_ni_decref_locked(ni
, cpt
);
797 lnet_net_unlock(cpt
);
803 lnet_count_acceptor_nis(void)
805 /* Return the # of NIs that need the acceptor. */
807 struct list_head
*tmp
;
811 cpt
= lnet_net_lock_current();
812 list_for_each(tmp
, &the_lnet
.ln_nis
) {
813 ni
= list_entry(tmp
, lnet_ni_t
, ni_list
);
815 if (ni
->ni_lnd
->lnd_accept
!= NULL
)
819 lnet_net_unlock(cpt
);
825 lnet_ni_tq_credits(lnet_ni_t
*ni
)
829 LASSERT(ni
->ni_ncpts
>= 1);
831 if (ni
->ni_ncpts
== 1)
832 return ni
->ni_maxtxcredits
;
834 credits
= ni
->ni_maxtxcredits
/ ni
->ni_ncpts
;
835 credits
= max(credits
, 8 * ni
->ni_peertxcredits
);
836 credits
= min(credits
, ni
->ni_maxtxcredits
);
842 lnet_shutdown_lndnis(void)
848 /* NB called holding the global mutex */
850 /* All quiet on the API front */
851 LASSERT(!the_lnet
.ln_shutdown
);
852 LASSERT(the_lnet
.ln_refcount
== 0);
853 LASSERT(list_empty(&the_lnet
.ln_nis_zombie
));
855 lnet_net_lock(LNET_LOCK_EX
);
856 the_lnet
.ln_shutdown
= 1; /* flag shutdown */
858 /* Unlink NIs from the global table */
859 while (!list_empty(&the_lnet
.ln_nis
)) {
860 ni
= list_entry(the_lnet
.ln_nis
.next
,
862 /* move it to zombie list and nobody can find it anymore */
863 list_move(&ni
->ni_list
, &the_lnet
.ln_nis_zombie
);
864 lnet_ni_decref_locked(ni
, 0); /* drop ln_nis' ref */
866 if (!list_empty(&ni
->ni_cptlist
)) {
867 list_del_init(&ni
->ni_cptlist
);
868 lnet_ni_decref_locked(ni
, 0);
872 /* Drop the cached eqwait NI. */
873 if (the_lnet
.ln_eq_waitni
!= NULL
) {
874 lnet_ni_decref_locked(the_lnet
.ln_eq_waitni
, 0);
875 the_lnet
.ln_eq_waitni
= NULL
;
878 /* Drop the cached loopback NI. */
879 if (the_lnet
.ln_loni
!= NULL
) {
880 lnet_ni_decref_locked(the_lnet
.ln_loni
, 0);
881 the_lnet
.ln_loni
= NULL
;
884 lnet_net_unlock(LNET_LOCK_EX
);
886 /* Clear lazy portals and drop delayed messages which hold refs
887 * on their lnet_msg_t::msg_rxpeer */
888 for (i
= 0; i
< the_lnet
.ln_nportals
; i
++)
889 LNetClearLazyPortal(i
);
891 /* Clear the peer table and wait for all peers to go (they hold refs on
893 lnet_peer_tables_cleanup();
895 lnet_net_lock(LNET_LOCK_EX
);
896 /* Now wait for the NI's I just nuked to show up on ln_zombie_nis
897 * and shut them down in guaranteed thread context */
899 while (!list_empty(&the_lnet
.ln_nis_zombie
)) {
903 ni
= list_entry(the_lnet
.ln_nis_zombie
.next
,
905 list_del_init(&ni
->ni_list
);
906 cfs_percpt_for_each(ref
, j
, ni
->ni_refs
) {
909 /* still busy, add it back to zombie list */
910 list_add(&ni
->ni_list
, &the_lnet
.ln_nis_zombie
);
914 if (!list_empty(&ni
->ni_list
)) {
915 lnet_net_unlock(LNET_LOCK_EX
);
917 if ((i
& (-i
)) == i
) {
918 CDEBUG(D_WARNING
, "Waiting for zombie LNI %s\n",
919 libcfs_nid2str(ni
->ni_nid
));
921 set_current_state(TASK_UNINTERRUPTIBLE
);
922 schedule_timeout(cfs_time_seconds(1));
923 lnet_net_lock(LNET_LOCK_EX
);
927 ni
->ni_lnd
->lnd_refcount
--;
928 lnet_net_unlock(LNET_LOCK_EX
);
930 islo
= ni
->ni_lnd
->lnd_type
== LOLND
;
932 LASSERT(!in_interrupt());
933 (ni
->ni_lnd
->lnd_shutdown
)(ni
);
935 /* can't deref lnd anymore now; it might have unregistered
939 CDEBUG(D_LNI
, "Removed LNI %s\n",
940 libcfs_nid2str(ni
->ni_nid
));
945 lnet_net_lock(LNET_LOCK_EX
);
948 the_lnet
.ln_shutdown
= 0;
949 lnet_net_unlock(LNET_LOCK_EX
);
951 if (the_lnet
.ln_network_tokens
!= NULL
) {
952 LIBCFS_FREE(the_lnet
.ln_network_tokens
,
953 the_lnet
.ln_network_tokens_nob
);
954 the_lnet
.ln_network_tokens
= NULL
;
959 lnet_startup_lndnis(void)
963 struct lnet_tx_queue
*tq
;
964 struct list_head nilist
;
969 char *nets
= lnet_get_networks();
971 INIT_LIST_HEAD(&nilist
);
976 rc
= lnet_parse_networks(&nilist
, nets
);
980 while (!list_empty(&nilist
)) {
981 ni
= list_entry(nilist
.next
, lnet_ni_t
, ni_list
);
982 lnd_type
= LNET_NETTYP(LNET_NIDNET(ni
->ni_nid
));
984 LASSERT(libcfs_isknown_lnd(lnd_type
));
986 if (lnd_type
== CIBLND
||
987 lnd_type
== OPENIBLND
||
988 lnd_type
== IIBLND
||
989 lnd_type
== VIBLND
) {
990 CERROR("LND %s obsoleted\n",
991 libcfs_lnd2str(lnd_type
));
995 mutex_lock(&the_lnet
.ln_lnd_mutex
);
996 lnd
= lnet_find_lnd_by_type(lnd_type
);
999 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
1000 rc
= request_module("%s",
1001 libcfs_lnd2modname(lnd_type
));
1002 mutex_lock(&the_lnet
.ln_lnd_mutex
);
1004 lnd
= lnet_find_lnd_by_type(lnd_type
);
1006 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
1007 CERROR("Can't load LND %s, module %s, rc=%d\n",
1008 libcfs_lnd2str(lnd_type
),
1009 libcfs_lnd2modname(lnd_type
), rc
);
1014 lnet_net_lock(LNET_LOCK_EX
);
1015 lnd
->lnd_refcount
++;
1016 lnet_net_unlock(LNET_LOCK_EX
);
1020 rc
= (lnd
->lnd_startup
)(ni
);
1022 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
1025 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1026 rc
, libcfs_lnd2str(lnd
->lnd_type
));
1027 lnet_net_lock(LNET_LOCK_EX
);
1028 lnd
->lnd_refcount
--;
1029 lnet_net_unlock(LNET_LOCK_EX
);
1033 LASSERT(ni
->ni_peertimeout
<= 0 || lnd
->lnd_query
!= NULL
);
1035 list_del(&ni
->ni_list
);
1037 lnet_net_lock(LNET_LOCK_EX
);
1038 /* refcount for ln_nis */
1039 lnet_ni_addref_locked(ni
, 0);
1040 list_add_tail(&ni
->ni_list
, &the_lnet
.ln_nis
);
1041 if (ni
->ni_cpts
!= NULL
) {
1042 list_add_tail(&ni
->ni_cptlist
,
1043 &the_lnet
.ln_nis_cpt
);
1044 lnet_ni_addref_locked(ni
, 0);
1047 lnet_net_unlock(LNET_LOCK_EX
);
1049 if (lnd
->lnd_type
== LOLND
) {
1051 LASSERT(the_lnet
.ln_loni
== NULL
);
1052 the_lnet
.ln_loni
= ni
;
1056 if (ni
->ni_peertxcredits
== 0 ||
1057 ni
->ni_maxtxcredits
== 0) {
1058 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1059 libcfs_lnd2str(lnd
->lnd_type
),
1060 ni
->ni_peertxcredits
== 0 ?
1065 cfs_percpt_for_each(tq
, i
, ni
->ni_tx_queues
) {
1066 tq
->tq_credits_min
=
1067 tq
->tq_credits_max
=
1068 tq
->tq_credits
= lnet_ni_tq_credits(ni
);
1071 CDEBUG(D_LNI
, "Added LNI %s [%d/%d/%d/%d]\n",
1072 libcfs_nid2str(ni
->ni_nid
), ni
->ni_peertxcredits
,
1073 lnet_ni_tq_credits(ni
) * LNET_CPT_NUMBER
,
1074 ni
->ni_peerrtrcredits
, ni
->ni_peertimeout
);
1079 if (the_lnet
.ln_eq_waitni
!= NULL
&& nicount
> 1) {
1080 lnd_type
= the_lnet
.ln_eq_waitni
->ni_lnd
->lnd_type
;
1081 LCONSOLE_ERROR_MSG(0x109, "LND %s can only run single-network\n",
1082 libcfs_lnd2str(lnd_type
));
1089 lnet_shutdown_lndnis();
1091 while (!list_empty(&nilist
)) {
1092 ni
= list_entry(nilist
.next
, lnet_ni_t
, ni_list
);
1093 list_del(&ni
->ni_list
);
1101 * Initialize LNet library.
1103 * Only userspace program needs to call this function - it's automatically
1104 * called in the kernel at module loading time. Caller has to call lnet_fini()
1105 * after a call to lnet_init(), if and only if the latter returned 0. It must
1106 * be called exactly once.
1108 * \return 0 on success, and -ve on failures.
1115 lnet_assert_wire_constants();
1116 LASSERT(!the_lnet
.ln_init
);
1118 memset(&the_lnet
, 0, sizeof(the_lnet
));
1120 /* refer to global cfs_cpt_table for now */
1121 the_lnet
.ln_cpt_table
= cfs_cpt_table
;
1122 the_lnet
.ln_cpt_number
= cfs_cpt_number(cfs_cpt_table
);
1124 LASSERT(the_lnet
.ln_cpt_number
> 0);
1125 if (the_lnet
.ln_cpt_number
> LNET_CPT_MAX
) {
1126 /* we are under risk of consuming all lh_cookie */
1127 CERROR("Can't have %d CPTs for LNet (max allowed is %d), please change setting of CPT-table and retry\n",
1128 the_lnet
.ln_cpt_number
, LNET_CPT_MAX
);
1132 while ((1 << the_lnet
.ln_cpt_bits
) < the_lnet
.ln_cpt_number
)
1133 the_lnet
.ln_cpt_bits
++;
1135 rc
= lnet_create_locks();
1137 CERROR("Can't create LNet global locks: %d\n", rc
);
1141 the_lnet
.ln_refcount
= 0;
1142 the_lnet
.ln_init
= 1;
1143 LNetInvalidateHandle(&the_lnet
.ln_rc_eqh
);
1144 INIT_LIST_HEAD(&the_lnet
.ln_lnds
);
1145 INIT_LIST_HEAD(&the_lnet
.ln_rcd_zombie
);
1146 INIT_LIST_HEAD(&the_lnet
.ln_rcd_deathrow
);
1148 /* The hash table size is the number of bits it takes to express the set
1149 * ln_num_routes, minus 1 (better to under estimate than over so we
1150 * don't waste memory). */
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);
1158 /* All LNDs apart from the LOLND are in separate modules. They
1159 * register themselves when their module loads, and unregister
1160 * themselves when their module is unloaded. */
1161 lnet_register_lnd(&the_lolnd
);
1164 EXPORT_SYMBOL(lnet_init
);
1167 * Finalize LNet library.
1169 * Only userspace program needs to call this function. It can be called
1172 * \pre lnet_init() called with success.
1173 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1178 LASSERT(the_lnet
.ln_init
);
1179 LASSERT(the_lnet
.ln_refcount
== 0);
1181 while (!list_empty(&the_lnet
.ln_lnds
))
1182 lnet_unregister_lnd(list_entry(the_lnet
.ln_lnds
.next
,
1184 lnet_destroy_locks();
1186 the_lnet
.ln_init
= 0;
1188 EXPORT_SYMBOL(lnet_fini
);
1191 * Set LNet PID and start LNet interfaces, routing, and forwarding.
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
1199 * The PID used by LNet may be different from the one requested.
1202 * \param requested_pid PID requested by the caller.
1204 * \return >= 0 on success, and < 0 error code on failures.
1207 LNetNIInit(lnet_pid_t requested_pid
)
1209 int im_a_router
= 0;
1212 mutex_lock(&the_lnet
.ln_api_mutex
);
1214 LASSERT(the_lnet
.ln_init
);
1215 CDEBUG(D_OTHER
, "refs %d\n", the_lnet
.ln_refcount
);
1217 if (the_lnet
.ln_refcount
> 0) {
1218 rc
= the_lnet
.ln_refcount
++;
1222 if (requested_pid
== LNET_PID_ANY
) {
1223 /* Don't instantiate LNET just for me */
1228 rc
= lnet_prepare(requested_pid
);
1232 rc
= lnet_startup_lndnis();
1236 rc
= lnet_parse_routes(lnet_get_routes(), &im_a_router
);
1240 rc
= lnet_check_routes();
1244 rc
= lnet_rtrpools_alloc(im_a_router
);
1248 rc
= lnet_acceptor_start();
1252 the_lnet
.ln_refcount
= 1;
1253 /* Now I may use my own API functions... */
1255 /* NB router checker needs the_lnet.ln_ping_info in
1256 * lnet_router_checker -> lnet_update_ni_status_locked */
1257 rc
= lnet_ping_target_init();
1261 rc
= lnet_router_checker_start();
1265 lnet_router_debugfs_init();
1269 lnet_ping_target_fini();
1271 the_lnet
.ln_refcount
= 0;
1272 lnet_acceptor_stop();
1274 lnet_destroy_routes();
1275 lnet_shutdown_lndnis();
1281 mutex_unlock(&the_lnet
.ln_api_mutex
);
1284 EXPORT_SYMBOL(LNetNIInit
);
1287 * Stop LNet interfaces, routing, and forwarding.
1289 * Users must call this function once for each successful call to LNetNIInit().
1290 * Once the LNetNIFini() operation has been started, the results of pending
1291 * API operations are undefined.
1293 * \return always 0 for current implementation.
1298 mutex_lock(&the_lnet
.ln_api_mutex
);
1300 LASSERT(the_lnet
.ln_init
);
1301 LASSERT(the_lnet
.ln_refcount
> 0);
1303 if (the_lnet
.ln_refcount
!= 1) {
1304 the_lnet
.ln_refcount
--;
1306 LASSERT(!the_lnet
.ln_niinit_self
);
1308 lnet_router_debugfs_fini();
1309 lnet_router_checker_stop();
1310 lnet_ping_target_fini();
1312 /* Teardown fns that use my own API functions BEFORE here */
1313 the_lnet
.ln_refcount
= 0;
1315 lnet_acceptor_stop();
1316 lnet_destroy_routes();
1317 lnet_shutdown_lndnis();
1321 mutex_unlock(&the_lnet
.ln_api_mutex
);
1324 EXPORT_SYMBOL(LNetNIFini
);
1327 * This is an ugly hack to export IOC_LIBCFS_DEBUG_PEER and
1328 * IOC_LIBCFS_PORTALS_COMPATIBILITY commands to users, by tweaking the LNet
1329 * internal ioctl handler.
1331 * IOC_LIBCFS_PORTALS_COMPATIBILITY is now deprecated, don't use it.
1333 * \param cmd IOC_LIBCFS_DEBUG_PEER to print debugging data about a peer.
1334 * The data will be printed to system console. Don't use it excessively.
1335 * \param arg A pointer to lnet_process_id_t, process ID of the peer.
1337 * \return Always return 0 when called by users directly (i.e., not via ioctl).
1340 LNetCtl(unsigned int cmd
, void *arg
)
1342 struct libcfs_ioctl_data
*data
= arg
;
1343 lnet_process_id_t id
= {0};
1347 LASSERT(the_lnet
.ln_init
);
1348 LASSERT(the_lnet
.ln_refcount
> 0);
1351 case IOC_LIBCFS_GET_NI
:
1352 rc
= LNetGetId(data
->ioc_count
, &id
);
1353 data
->ioc_nid
= id
.nid
;
1356 case IOC_LIBCFS_FAIL_NID
:
1357 return lnet_fail_nid(data
->ioc_nid
, data
->ioc_count
);
1359 case IOC_LIBCFS_ADD_ROUTE
:
1360 rc
= lnet_add_route(data
->ioc_net
, data
->ioc_count
,
1361 data
->ioc_nid
, data
->ioc_priority
);
1362 return (rc
!= 0) ? rc
: lnet_check_routes();
1364 case IOC_LIBCFS_DEL_ROUTE
:
1365 return lnet_del_route(data
->ioc_net
, data
->ioc_nid
);
1367 case IOC_LIBCFS_GET_ROUTE
:
1368 return lnet_get_route(data
->ioc_count
,
1369 &data
->ioc_net
, &data
->ioc_count
,
1370 &data
->ioc_nid
, &data
->ioc_flags
,
1371 &data
->ioc_priority
);
1372 case IOC_LIBCFS_NOTIFY_ROUTER
:
1373 return lnet_notify(NULL
, data
->ioc_nid
, data
->ioc_flags
,
1374 cfs_time_current() -
1375 cfs_time_seconds(get_seconds() -
1376 (time_t)data
->ioc_u64
[0]));
1378 case IOC_LIBCFS_PORTALS_COMPATIBILITY
:
1379 /* This can be removed once lustre stops calling it */
1382 case IOC_LIBCFS_LNET_DIST
:
1383 rc
= LNetDist(data
->ioc_nid
, &data
->ioc_nid
, &data
->ioc_u32
[1]);
1384 if (rc
< 0 && rc
!= -EHOSTUNREACH
)
1387 data
->ioc_u32
[0] = rc
;
1390 case IOC_LIBCFS_TESTPROTOCOMPAT
:
1391 lnet_net_lock(LNET_LOCK_EX
);
1392 the_lnet
.ln_testprotocompat
= data
->ioc_flags
;
1393 lnet_net_unlock(LNET_LOCK_EX
);
1396 case IOC_LIBCFS_PING
:
1397 id
.nid
= data
->ioc_nid
;
1398 id
.pid
= data
->ioc_u32
[0];
1399 rc
= lnet_ping(id
, data
->ioc_u32
[1], /* timeout */
1400 (lnet_process_id_t
*)data
->ioc_pbuf1
,
1401 data
->ioc_plen1
/sizeof(lnet_process_id_t
));
1404 data
->ioc_count
= rc
;
1407 case IOC_LIBCFS_DEBUG_PEER
: {
1408 /* CAVEAT EMPTOR: this one designed for calling directly; not
1410 id
= *((lnet_process_id_t
*) arg
);
1412 lnet_debug_peer(id
.nid
);
1414 ni
= lnet_net2ni(LNET_NIDNET(id
.nid
));
1416 CDEBUG(D_WARNING
, "No NI for %s\n", libcfs_id2str(id
));
1418 if (ni
->ni_lnd
->lnd_ctl
== NULL
) {
1419 CDEBUG(D_WARNING
, "No ctl for %s\n",
1422 (void)ni
->ni_lnd
->lnd_ctl(ni
, cmd
, arg
);
1431 ni
= lnet_net2ni(data
->ioc_net
);
1435 if (ni
->ni_lnd
->lnd_ctl
== NULL
)
1438 rc
= ni
->ni_lnd
->lnd_ctl(ni
, cmd
, arg
);
1445 EXPORT_SYMBOL(LNetCtl
);
1448 * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
1449 * all interfaces share a same PID, as requested by LNetNIInit().
1451 * \param index Index of the interface to look up.
1452 * \param id On successful return, this location will hold the
1453 * lnet_process_id_t ID of the interface.
1455 * \retval 0 If an interface exists at \a index.
1456 * \retval -ENOENT If no interface has been found.
1459 LNetGetId(unsigned int index
, lnet_process_id_t
*id
)
1462 struct list_head
*tmp
;
1466 LASSERT(the_lnet
.ln_init
);
1468 /* LNetNI initilization failed? */
1469 if (the_lnet
.ln_refcount
== 0)
1472 cpt
= lnet_net_lock_current();
1474 list_for_each(tmp
, &the_lnet
.ln_nis
) {
1478 ni
= list_entry(tmp
, lnet_ni_t
, ni_list
);
1480 id
->nid
= ni
->ni_nid
;
1481 id
->pid
= the_lnet
.ln_pid
;
1486 lnet_net_unlock(cpt
);
1489 EXPORT_SYMBOL(LNetGetId
);
1492 * Print a string representation of handle \a h into buffer \a str of
1496 LNetSnprintHandle(char *str
, int len
, lnet_handle_any_t h
)
1498 snprintf(str
, len
, "%#llx", h
.cookie
);
1500 EXPORT_SYMBOL(LNetSnprintHandle
);
1503 lnet_create_ping_info(void)
1508 unsigned int infosz
;
1510 lnet_process_id_t id
;
1511 lnet_ping_info_t
*pinfo
;
1513 for (n
= 0; ; n
++) {
1514 rc
= LNetGetId(n
, &id
);
1521 infosz
= offsetof(lnet_ping_info_t
, pi_ni
[n
]);
1522 LIBCFS_ALLOC(pinfo
, infosz
);
1523 if (pinfo
== NULL
) {
1524 CERROR("Can't allocate ping info[%d]\n", n
);
1529 pinfo
->pi_pid
= the_lnet
.ln_pid
;
1530 pinfo
->pi_magic
= LNET_PROTO_PING_MAGIC
;
1531 pinfo
->pi_features
= LNET_PING_FEAT_NI_STATUS
;
1533 for (i
= 0; i
< n
; i
++) {
1534 lnet_ni_status_t
*ns
= &pinfo
->pi_ni
[i
];
1536 rc
= LNetGetId(i
, &id
);
1539 ns
->ns_nid
= id
.nid
;
1540 ns
->ns_status
= LNET_NI_STATUS_UP
;
1544 ni
= lnet_nid2ni_locked(id
.nid
, 0);
1545 LASSERT(ni
!= NULL
);
1548 LASSERT(ni
->ni_status
== NULL
);
1552 lnet_ni_decref_locked(ni
, 0);
1556 the_lnet
.ln_ping_info
= pinfo
;
1561 lnet_destroy_ping_info(void)
1567 list_for_each_entry(ni
, &the_lnet
.ln_nis
, ni_list
) {
1569 ni
->ni_status
= NULL
;
1575 LIBCFS_FREE(the_lnet
.ln_ping_info
,
1576 offsetof(lnet_ping_info_t
,
1577 pi_ni
[the_lnet
.ln_ping_info
->pi_nnis
]));
1578 the_lnet
.ln_ping_info
= NULL
;
1582 lnet_ping_target_init(void)
1584 lnet_md_t md
= { NULL
};
1585 lnet_handle_me_t meh
;
1586 lnet_process_id_t id
;
1591 rc
= lnet_create_ping_info();
1595 /* We can have a tiny EQ since we only need to see the unlink event on
1596 * teardown, which by definition is the last one! */
1597 rc
= LNetEQAlloc(2, LNET_EQ_HANDLER_NONE
, &the_lnet
.ln_ping_target_eq
);
1599 CERROR("Can't allocate ping EQ: %d\n", rc
);
1603 memset(&id
, 0, sizeof(lnet_process_id_t
));
1604 id
.nid
= LNET_NID_ANY
;
1605 id
.pid
= LNET_PID_ANY
;
1607 rc
= LNetMEAttach(LNET_RESERVED_PORTAL
, id
,
1608 LNET_PROTO_PING_MATCHBITS
, 0,
1609 LNET_UNLINK
, LNET_INS_AFTER
,
1612 CERROR("Can't create ping ME: %d\n", rc
);
1616 /* initialize md content */
1617 infosz
= offsetof(lnet_ping_info_t
,
1618 pi_ni
[the_lnet
.ln_ping_info
->pi_nnis
]);
1619 md
.start
= the_lnet
.ln_ping_info
;
1621 md
.threshold
= LNET_MD_THRESH_INF
;
1623 md
.options
= LNET_MD_OP_GET
| LNET_MD_TRUNCATE
|
1624 LNET_MD_MANAGE_REMOTE
;
1626 md
.eq_handle
= the_lnet
.ln_ping_target_eq
;
1628 rc
= LNetMDAttach(meh
, md
,
1630 &the_lnet
.ln_ping_target_md
);
1632 CERROR("Can't attach ping MD: %d\n", rc
);
1639 rc2
= LNetMEUnlink(meh
);
1642 rc2
= LNetEQFree(the_lnet
.ln_ping_target_eq
);
1645 lnet_destroy_ping_info();
1650 lnet_ping_target_fini(void)
1655 int timeout_ms
= 1000;
1656 sigset_t blocked
= cfs_block_allsigs();
1658 LNetMDUnlink(the_lnet
.ln_ping_target_md
);
1659 /* NB md could be busy; this just starts the unlink */
1662 rc
= LNetEQPoll(&the_lnet
.ln_ping_target_eq
, 1,
1663 timeout_ms
, &event
, &which
);
1665 /* I expect overflow... */
1666 LASSERT(rc
>= 0 || rc
== -EOVERFLOW
);
1669 /* timed out: provide a diagnostic */
1670 CWARN("Still waiting for ping MD to unlink\n");
1675 /* Got a valid event */
1680 rc
= LNetEQFree(the_lnet
.ln_ping_target_eq
);
1682 lnet_destroy_ping_info();
1683 cfs_restore_sigs(blocked
);
1687 lnet_ping(lnet_process_id_t id
, int timeout_ms
, lnet_process_id_t
*ids
, int n_ids
)
1689 lnet_handle_eq_t eqh
;
1690 lnet_handle_md_t mdh
;
1692 lnet_md_t md
= { NULL
};
1696 const int a_long_time
= 60000; /* mS */
1697 int infosz
= offsetof(lnet_ping_info_t
, pi_ni
[n_ids
]);
1698 lnet_ping_info_t
*info
;
1699 lnet_process_id_t tmpid
;
1707 id
.nid
== LNET_NID_ANY
||
1708 timeout_ms
> 500000 || /* arbitrary limit! */
1709 n_ids
> 20) /* arbitrary limit! */
1712 if (id
.pid
== LNET_PID_ANY
)
1713 id
.pid
= LUSTRE_SRV_LNET_PID
;
1715 LIBCFS_ALLOC(info
, infosz
);
1719 /* NB 2 events max (including any unlink event) */
1720 rc
= LNetEQAlloc(2, LNET_EQ_HANDLER_NONE
, &eqh
);
1722 CERROR("Can't allocate EQ: %d\n", rc
);
1726 /* initialize md content */
1729 md
.threshold
= 2; /*GET/REPLY*/
1731 md
.options
= LNET_MD_TRUNCATE
;
1735 rc
= LNetMDBind(md
, LNET_UNLINK
, &mdh
);
1737 CERROR("Can't bind MD: %d\n", rc
);
1741 rc
= LNetGet(LNET_NID_ANY
, mdh
, id
,
1742 LNET_RESERVED_PORTAL
,
1743 LNET_PROTO_PING_MATCHBITS
, 0);
1746 /* Don't CERROR; this could be deliberate! */
1748 rc2
= LNetMDUnlink(mdh
);
1751 /* NB must wait for the UNLINK event below... */
1753 timeout_ms
= a_long_time
;
1757 /* MUST block for unlink to complete */
1759 blocked
= cfs_block_allsigs();
1761 rc2
= LNetEQPoll(&eqh
, 1, timeout_ms
, &event
, &which
);
1764 cfs_restore_sigs(blocked
);
1766 CDEBUG(D_NET
, "poll %d(%d %d)%s\n", rc2
,
1767 (rc2
<= 0) ? -1 : event
.type
,
1768 (rc2
<= 0) ? -1 : event
.status
,
1769 (rc2
> 0 && event
.unlinked
) ? " unlinked" : "");
1771 LASSERT(rc2
!= -EOVERFLOW
); /* can't miss anything */
1773 if (rc2
<= 0 || event
.status
!= 0) {
1774 /* timeout or error */
1775 if (!replied
&& rc
== 0)
1776 rc
= (rc2
< 0) ? rc2
:
1777 (rc2
== 0) ? -ETIMEDOUT
:
1781 /* Ensure completion in finite time... */
1783 /* No assertion (racing with network) */
1785 timeout_ms
= a_long_time
;
1786 } else if (rc2
== 0) {
1787 /* timed out waiting for unlink */
1788 CWARN("ping %s: late network completion\n",
1791 } else if (event
.type
== LNET_EVENT_REPLY
) {
1796 } while (rc2
<= 0 || !event
.unlinked
);
1800 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
1807 LASSERT(nob
>= 0 && nob
<= infosz
);
1809 rc
= -EPROTO
; /* if I can't parse... */
1812 /* can't check magic/version */
1813 CERROR("%s: ping info too short %d\n",
1814 libcfs_id2str(id
), nob
);
1818 if (info
->pi_magic
== __swab32(LNET_PROTO_PING_MAGIC
)) {
1819 lnet_swap_pinginfo(info
);
1820 } else if (info
->pi_magic
!= LNET_PROTO_PING_MAGIC
) {
1821 CERROR("%s: Unexpected magic %08x\n",
1822 libcfs_id2str(id
), info
->pi_magic
);
1826 if ((info
->pi_features
& LNET_PING_FEAT_NI_STATUS
) == 0) {
1827 CERROR("%s: ping w/o NI status: 0x%x\n",
1828 libcfs_id2str(id
), info
->pi_features
);
1832 if (nob
< offsetof(lnet_ping_info_t
, pi_ni
[0])) {
1833 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id
),
1834 nob
, (int)offsetof(lnet_ping_info_t
, pi_ni
[0]));
1838 if (info
->pi_nnis
< n_ids
)
1839 n_ids
= info
->pi_nnis
;
1841 if (nob
< offsetof(lnet_ping_info_t
, pi_ni
[n_ids
])) {
1842 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id
),
1843 nob
, (int)offsetof(lnet_ping_info_t
, pi_ni
[n_ids
]));
1847 rc
= -EFAULT
; /* If I SEGV... */
1849 memset(&tmpid
, 0, sizeof(tmpid
));
1850 for (i
= 0; i
< n_ids
; i
++) {
1851 tmpid
.pid
= info
->pi_pid
;
1852 tmpid
.nid
= info
->pi_ni
[i
].ns_nid
;
1853 if (copy_to_user(&ids
[i
], &tmpid
, sizeof(tmpid
)))
1859 rc2
= LNetEQFree(eqh
);
1861 CERROR("rc2 %d\n", rc2
);
1865 LIBCFS_FREE(info
, infosz
);