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, 2015, 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
);
48 static char *ip2nets
= "";
49 module_param(ip2nets
, charp
, 0444);
50 MODULE_PARM_DESC(ip2nets
, "LNET network <- IP table");
52 static char *networks
= "";
53 module_param(networks
, charp
, 0444);
54 MODULE_PARM_DESC(networks
, "local networks");
56 static char *routes
= "";
57 module_param(routes
, charp
, 0444);
58 MODULE_PARM_DESC(routes
, "routes to non-local networks");
60 static int rnet_htable_size
= LNET_REMOTE_NETS_HASH_DEFAULT
;
61 module_param(rnet_htable_size
, int, 0444);
62 MODULE_PARM_DESC(rnet_htable_size
, "size of remote network hash table");
64 static int lnet_ping(lnet_process_id_t id
, int timeout_ms
,
65 lnet_process_id_t __user
*ids
, int n_ids
);
74 lnet_get_networks(void)
79 if (*networks
!= 0 && *ip2nets
!= 0) {
80 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or 'ip2nets' but not both at once\n");
85 rc
= lnet_parse_ip2nets(&nets
, ip2nets
);
86 return (rc
== 0) ? nets
: NULL
;
98 spin_lock_init(&the_lnet
.ln_eq_wait_lock
);
99 init_waitqueue_head(&the_lnet
.ln_eq_waitq
);
100 mutex_init(&the_lnet
.ln_lnd_mutex
);
101 mutex_init(&the_lnet
.ln_api_mutex
);
105 lnet_create_remote_nets_table(void)
108 struct list_head
*hash
;
110 LASSERT(!the_lnet
.ln_remote_nets_hash
);
111 LASSERT(the_lnet
.ln_remote_nets_hbits
> 0);
112 LIBCFS_ALLOC(hash
, LNET_REMOTE_NETS_HASH_SIZE
* sizeof(*hash
));
114 CERROR("Failed to create remote nets hash table\n");
118 for (i
= 0; i
< LNET_REMOTE_NETS_HASH_SIZE
; i
++)
119 INIT_LIST_HEAD(&hash
[i
]);
120 the_lnet
.ln_remote_nets_hash
= hash
;
125 lnet_destroy_remote_nets_table(void)
129 if (!the_lnet
.ln_remote_nets_hash
)
132 for (i
= 0; i
< LNET_REMOTE_NETS_HASH_SIZE
; i
++)
133 LASSERT(list_empty(&the_lnet
.ln_remote_nets_hash
[i
]));
135 LIBCFS_FREE(the_lnet
.ln_remote_nets_hash
,
136 LNET_REMOTE_NETS_HASH_SIZE
*
137 sizeof(the_lnet
.ln_remote_nets_hash
[0]));
138 the_lnet
.ln_remote_nets_hash
= NULL
;
142 lnet_destroy_locks(void)
144 if (the_lnet
.ln_res_lock
) {
145 cfs_percpt_lock_free(the_lnet
.ln_res_lock
);
146 the_lnet
.ln_res_lock
= NULL
;
149 if (the_lnet
.ln_net_lock
) {
150 cfs_percpt_lock_free(the_lnet
.ln_net_lock
);
151 the_lnet
.ln_net_lock
= NULL
;
156 lnet_create_locks(void)
160 the_lnet
.ln_res_lock
= cfs_percpt_lock_alloc(lnet_cpt_table());
161 if (!the_lnet
.ln_res_lock
)
164 the_lnet
.ln_net_lock
= cfs_percpt_lock_alloc(lnet_cpt_table());
165 if (!the_lnet
.ln_net_lock
)
171 lnet_destroy_locks();
175 static void lnet_assert_wire_constants(void)
178 * Wire protocol assertions generated by 'wirecheck'
179 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
180 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
181 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
185 CLASSERT(LNET_PROTO_TCP_MAGIC
== 0xeebc0ded);
186 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR
== 1);
187 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR
== 0);
188 CLASSERT(LNET_MSG_ACK
== 0);
189 CLASSERT(LNET_MSG_PUT
== 1);
190 CLASSERT(LNET_MSG_GET
== 2);
191 CLASSERT(LNET_MSG_REPLY
== 3);
192 CLASSERT(LNET_MSG_HELLO
== 4);
194 /* Checks for struct ptl_handle_wire_t */
195 CLASSERT((int)sizeof(lnet_handle_wire_t
) == 16);
196 CLASSERT((int)offsetof(lnet_handle_wire_t
, wh_interface_cookie
) == 0);
197 CLASSERT((int)sizeof(((lnet_handle_wire_t
*)0)->wh_interface_cookie
) == 8);
198 CLASSERT((int)offsetof(lnet_handle_wire_t
, wh_object_cookie
) == 8);
199 CLASSERT((int)sizeof(((lnet_handle_wire_t
*)0)->wh_object_cookie
) == 8);
201 /* Checks for struct lnet_magicversion_t */
202 CLASSERT((int)sizeof(lnet_magicversion_t
) == 8);
203 CLASSERT((int)offsetof(lnet_magicversion_t
, magic
) == 0);
204 CLASSERT((int)sizeof(((lnet_magicversion_t
*)0)->magic
) == 4);
205 CLASSERT((int)offsetof(lnet_magicversion_t
, version_major
) == 4);
206 CLASSERT((int)sizeof(((lnet_magicversion_t
*)0)->version_major
) == 2);
207 CLASSERT((int)offsetof(lnet_magicversion_t
, version_minor
) == 6);
208 CLASSERT((int)sizeof(((lnet_magicversion_t
*)0)->version_minor
) == 2);
210 /* Checks for struct lnet_hdr_t */
211 CLASSERT((int)sizeof(lnet_hdr_t
) == 72);
212 CLASSERT((int)offsetof(lnet_hdr_t
, dest_nid
) == 0);
213 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->dest_nid
) == 8);
214 CLASSERT((int)offsetof(lnet_hdr_t
, src_nid
) == 8);
215 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->src_nid
) == 8);
216 CLASSERT((int)offsetof(lnet_hdr_t
, dest_pid
) == 16);
217 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->dest_pid
) == 4);
218 CLASSERT((int)offsetof(lnet_hdr_t
, src_pid
) == 20);
219 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->src_pid
) == 4);
220 CLASSERT((int)offsetof(lnet_hdr_t
, type
) == 24);
221 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->type
) == 4);
222 CLASSERT((int)offsetof(lnet_hdr_t
, payload_length
) == 28);
223 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->payload_length
) == 4);
224 CLASSERT((int)offsetof(lnet_hdr_t
, msg
) == 32);
225 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
) == 40);
228 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.ack
.dst_wmd
) == 32);
229 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.ack
.dst_wmd
) == 16);
230 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.ack
.match_bits
) == 48);
231 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.ack
.match_bits
) == 8);
232 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.ack
.mlength
) == 56);
233 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.ack
.mlength
) == 4);
236 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.ack_wmd
) == 32);
237 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.ack_wmd
) == 16);
238 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.match_bits
) == 48);
239 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.match_bits
) == 8);
240 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.hdr_data
) == 56);
241 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.hdr_data
) == 8);
242 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.ptl_index
) == 64);
243 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.ptl_index
) == 4);
244 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.put
.offset
) == 68);
245 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.put
.offset
) == 4);
248 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.return_wmd
) == 32);
249 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.return_wmd
) == 16);
250 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.match_bits
) == 48);
251 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.match_bits
) == 8);
252 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.ptl_index
) == 56);
253 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.ptl_index
) == 4);
254 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.src_offset
) == 60);
255 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.src_offset
) == 4);
256 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.get
.sink_length
) == 64);
257 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.get
.sink_length
) == 4);
260 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.reply
.dst_wmd
) == 32);
261 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.reply
.dst_wmd
) == 16);
264 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.hello
.incarnation
) == 32);
265 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.hello
.incarnation
) == 8);
266 CLASSERT((int)offsetof(lnet_hdr_t
, msg
.hello
.type
) == 40);
267 CLASSERT((int)sizeof(((lnet_hdr_t
*)0)->msg
.hello
.type
) == 4);
271 lnet_find_lnd_by_type(__u32 type
)
274 struct list_head
*tmp
;
276 /* holding lnd mutex */
277 list_for_each(tmp
, &the_lnet
.ln_lnds
) {
278 lnd
= list_entry(tmp
, lnd_t
, lnd_list
);
280 if (lnd
->lnd_type
== type
)
288 lnet_register_lnd(lnd_t
*lnd
)
290 mutex_lock(&the_lnet
.ln_lnd_mutex
);
292 LASSERT(the_lnet
.ln_init
);
293 LASSERT(libcfs_isknown_lnd(lnd
->lnd_type
));
294 LASSERT(!lnet_find_lnd_by_type(lnd
->lnd_type
));
296 list_add_tail(&lnd
->lnd_list
, &the_lnet
.ln_lnds
);
297 lnd
->lnd_refcount
= 0;
299 CDEBUG(D_NET
, "%s LND registered\n", libcfs_lnd2str(lnd
->lnd_type
));
301 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
303 EXPORT_SYMBOL(lnet_register_lnd
);
306 lnet_unregister_lnd(lnd_t
*lnd
)
308 mutex_lock(&the_lnet
.ln_lnd_mutex
);
310 LASSERT(the_lnet
.ln_init
);
311 LASSERT(lnet_find_lnd_by_type(lnd
->lnd_type
) == lnd
);
312 LASSERT(lnd
->lnd_refcount
== 0);
314 list_del(&lnd
->lnd_list
);
315 CDEBUG(D_NET
, "%s LND unregistered\n", libcfs_lnd2str(lnd
->lnd_type
));
317 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
319 EXPORT_SYMBOL(lnet_unregister_lnd
);
322 lnet_counters_get(lnet_counters_t
*counters
)
324 lnet_counters_t
*ctr
;
327 memset(counters
, 0, sizeof(*counters
));
329 lnet_net_lock(LNET_LOCK_EX
);
331 cfs_percpt_for_each(ctr
, i
, the_lnet
.ln_counters
) {
332 counters
->msgs_max
+= ctr
->msgs_max
;
333 counters
->msgs_alloc
+= ctr
->msgs_alloc
;
334 counters
->errors
+= ctr
->errors
;
335 counters
->send_count
+= ctr
->send_count
;
336 counters
->recv_count
+= ctr
->recv_count
;
337 counters
->route_count
+= ctr
->route_count
;
338 counters
->drop_count
+= ctr
->drop_count
;
339 counters
->send_length
+= ctr
->send_length
;
340 counters
->recv_length
+= ctr
->recv_length
;
341 counters
->route_length
+= ctr
->route_length
;
342 counters
->drop_length
+= ctr
->drop_length
;
344 lnet_net_unlock(LNET_LOCK_EX
);
346 EXPORT_SYMBOL(lnet_counters_get
);
349 lnet_counters_reset(void)
351 lnet_counters_t
*counters
;
354 lnet_net_lock(LNET_LOCK_EX
);
356 cfs_percpt_for_each(counters
, i
, the_lnet
.ln_counters
)
357 memset(counters
, 0, sizeof(lnet_counters_t
));
359 lnet_net_unlock(LNET_LOCK_EX
);
363 lnet_res_type2str(int type
)
368 case LNET_COOKIE_TYPE_MD
:
370 case LNET_COOKIE_TYPE_ME
:
372 case LNET_COOKIE_TYPE_EQ
:
378 lnet_res_container_cleanup(struct lnet_res_container
*rec
)
382 if (rec
->rec_type
== 0) /* not set yet, it's uninitialized */
385 while (!list_empty(&rec
->rec_active
)) {
386 struct list_head
*e
= rec
->rec_active
.next
;
389 if (rec
->rec_type
== LNET_COOKIE_TYPE_EQ
) {
390 lnet_eq_free(list_entry(e
, lnet_eq_t
, eq_list
));
392 } else if (rec
->rec_type
== LNET_COOKIE_TYPE_MD
) {
393 lnet_md_free(list_entry(e
, lnet_libmd_t
, md_list
));
395 } else { /* NB: Active MEs should be attached on portals */
403 * Found alive MD/ME/EQ, user really should unlink/free
404 * all of them before finalize LNet, but if someone didn't,
405 * we have to recycle garbage for him
407 CERROR("%d active elements on exit of %s container\n",
408 count
, lnet_res_type2str(rec
->rec_type
));
411 if (rec
->rec_lh_hash
) {
412 LIBCFS_FREE(rec
->rec_lh_hash
,
413 LNET_LH_HASH_SIZE
* sizeof(rec
->rec_lh_hash
[0]));
414 rec
->rec_lh_hash
= NULL
;
417 rec
->rec_type
= 0; /* mark it as finalized */
421 lnet_res_container_setup(struct lnet_res_container
*rec
, int cpt
, int type
)
426 LASSERT(rec
->rec_type
== 0);
428 rec
->rec_type
= type
;
429 INIT_LIST_HEAD(&rec
->rec_active
);
430 rec
->rec_lh_cookie
= (cpt
<< LNET_COOKIE_TYPE_BITS
) | type
;
432 /* Arbitrary choice of hash table size */
433 LIBCFS_CPT_ALLOC(rec
->rec_lh_hash
, lnet_cpt_table(), cpt
,
434 LNET_LH_HASH_SIZE
* sizeof(rec
->rec_lh_hash
[0]));
435 if (!rec
->rec_lh_hash
) {
440 for (i
= 0; i
< LNET_LH_HASH_SIZE
; i
++)
441 INIT_LIST_HEAD(&rec
->rec_lh_hash
[i
]);
446 CERROR("Failed to setup %s resource container\n",
447 lnet_res_type2str(type
));
448 lnet_res_container_cleanup(rec
);
453 lnet_res_containers_destroy(struct lnet_res_container
**recs
)
455 struct lnet_res_container
*rec
;
458 cfs_percpt_for_each(rec
, i
, recs
)
459 lnet_res_container_cleanup(rec
);
461 cfs_percpt_free(recs
);
464 static struct lnet_res_container
**
465 lnet_res_containers_create(int type
)
467 struct lnet_res_container
**recs
;
468 struct lnet_res_container
*rec
;
472 recs
= cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec
));
474 CERROR("Failed to allocate %s resource containers\n",
475 lnet_res_type2str(type
));
479 cfs_percpt_for_each(rec
, i
, recs
) {
480 rc
= lnet_res_container_setup(rec
, i
, type
);
482 lnet_res_containers_destroy(recs
);
491 lnet_res_lh_lookup(struct lnet_res_container
*rec
, __u64 cookie
)
493 /* ALWAYS called with lnet_res_lock held */
494 struct list_head
*head
;
495 lnet_libhandle_t
*lh
;
498 if ((cookie
& LNET_COOKIE_MASK
) != rec
->rec_type
)
501 hash
= cookie
>> (LNET_COOKIE_TYPE_BITS
+ LNET_CPT_BITS
);
502 head
= &rec
->rec_lh_hash
[hash
& LNET_LH_HASH_MASK
];
504 list_for_each_entry(lh
, head
, lh_hash_chain
) {
505 if (lh
->lh_cookie
== cookie
)
513 lnet_res_lh_initialize(struct lnet_res_container
*rec
, lnet_libhandle_t
*lh
)
515 /* ALWAYS called with lnet_res_lock held */
516 unsigned int ibits
= LNET_COOKIE_TYPE_BITS
+ LNET_CPT_BITS
;
519 lh
->lh_cookie
= rec
->rec_lh_cookie
;
520 rec
->rec_lh_cookie
+= 1 << ibits
;
522 hash
= (lh
->lh_cookie
>> ibits
) & LNET_LH_HASH_MASK
;
524 list_add(&lh
->lh_hash_chain
, &rec
->rec_lh_hash
[hash
]);
527 int lnet_unprepare(void);
530 lnet_prepare(lnet_pid_t requested_pid
)
532 /* Prepare to bring up the network */
533 struct lnet_res_container
**recs
;
536 LASSERT(the_lnet
.ln_refcount
== 0);
538 the_lnet
.ln_routing
= 0;
540 LASSERT((requested_pid
& LNET_PID_USERFLAG
) == 0);
541 the_lnet
.ln_pid
= requested_pid
;
543 INIT_LIST_HEAD(&the_lnet
.ln_test_peers
);
544 INIT_LIST_HEAD(&the_lnet
.ln_nis
);
545 INIT_LIST_HEAD(&the_lnet
.ln_nis_cpt
);
546 INIT_LIST_HEAD(&the_lnet
.ln_nis_zombie
);
547 INIT_LIST_HEAD(&the_lnet
.ln_routers
);
549 rc
= lnet_create_remote_nets_table();
553 * NB the interface cookie in wire handles guards against delayed
554 * replies and ACKs appearing valid after reboot.
556 the_lnet
.ln_interface_cookie
= ktime_get_ns();
558 the_lnet
.ln_counters
= cfs_percpt_alloc(lnet_cpt_table(),
559 sizeof(lnet_counters_t
));
560 if (!the_lnet
.ln_counters
) {
561 CERROR("Failed to allocate counters for LNet\n");
566 rc
= lnet_peer_tables_create();
570 rc
= lnet_msg_containers_create();
574 rc
= lnet_res_container_setup(&the_lnet
.ln_eq_container
, 0,
575 LNET_COOKIE_TYPE_EQ
);
579 recs
= lnet_res_containers_create(LNET_COOKIE_TYPE_ME
);
585 the_lnet
.ln_me_containers
= recs
;
587 recs
= lnet_res_containers_create(LNET_COOKIE_TYPE_MD
);
593 the_lnet
.ln_md_containers
= recs
;
595 rc
= lnet_portals_create();
597 CERROR("Failed to create portals for LNet: %d\n", rc
);
612 * NB no LNET_LOCK since this is the last reference. All LND instances
613 * have shut down already, so it is safe to unlink and free all
614 * descriptors, even those that appear committed to a network op (eg MD
615 * with non-zero pending count)
617 lnet_fail_nid(LNET_NID_ANY
, 0);
619 LASSERT(the_lnet
.ln_refcount
== 0);
620 LASSERT(list_empty(&the_lnet
.ln_test_peers
));
621 LASSERT(list_empty(&the_lnet
.ln_nis
));
622 LASSERT(list_empty(&the_lnet
.ln_nis_cpt
));
623 LASSERT(list_empty(&the_lnet
.ln_nis_zombie
));
625 lnet_portals_destroy();
627 if (the_lnet
.ln_md_containers
) {
628 lnet_res_containers_destroy(the_lnet
.ln_md_containers
);
629 the_lnet
.ln_md_containers
= NULL
;
632 if (the_lnet
.ln_me_containers
) {
633 lnet_res_containers_destroy(the_lnet
.ln_me_containers
);
634 the_lnet
.ln_me_containers
= NULL
;
637 lnet_res_container_cleanup(&the_lnet
.ln_eq_container
);
639 lnet_msg_containers_destroy();
640 lnet_peer_tables_destroy();
641 lnet_rtrpools_free();
643 if (the_lnet
.ln_counters
) {
644 cfs_percpt_free(the_lnet
.ln_counters
);
645 the_lnet
.ln_counters
= NULL
;
647 lnet_destroy_remote_nets_table();
653 lnet_net2ni_locked(__u32 net
, int cpt
)
655 struct list_head
*tmp
;
658 LASSERT(cpt
!= LNET_LOCK_EX
);
660 list_for_each(tmp
, &the_lnet
.ln_nis
) {
661 ni
= list_entry(tmp
, lnet_ni_t
, ni_list
);
663 if (LNET_NIDNET(ni
->ni_nid
) == net
) {
664 lnet_ni_addref_locked(ni
, cpt
);
673 lnet_net2ni(__u32 net
)
678 ni
= lnet_net2ni_locked(net
, 0);
683 EXPORT_SYMBOL(lnet_net2ni
);
686 lnet_nid_cpt_hash(lnet_nid_t nid
, unsigned int number
)
691 LASSERT(number
>= 1 && number
<= LNET_CPT_NUMBER
);
696 val
= hash_long(key
, LNET_CPT_BITS
);
697 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
701 return (unsigned int)(key
+ val
+ (val
>> 1)) % number
;
705 lnet_cpt_of_nid_locked(lnet_nid_t nid
)
709 /* must called with hold of lnet_net_lock */
710 if (LNET_CPT_NUMBER
== 1)
711 return 0; /* the only one */
713 /* take lnet_net_lock(any) would be OK */
714 if (!list_empty(&the_lnet
.ln_nis_cpt
)) {
715 list_for_each_entry(ni
, &the_lnet
.ln_nis_cpt
, ni_cptlist
) {
716 if (LNET_NIDNET(ni
->ni_nid
) != LNET_NIDNET(nid
))
719 LASSERT(ni
->ni_cpts
);
720 return ni
->ni_cpts
[lnet_nid_cpt_hash
721 (nid
, ni
->ni_ncpts
)];
725 return lnet_nid_cpt_hash(nid
, LNET_CPT_NUMBER
);
729 lnet_cpt_of_nid(lnet_nid_t nid
)
734 if (LNET_CPT_NUMBER
== 1)
735 return 0; /* the only one */
737 if (list_empty(&the_lnet
.ln_nis_cpt
))
738 return lnet_nid_cpt_hash(nid
, LNET_CPT_NUMBER
);
740 cpt
= lnet_net_lock_current();
741 cpt2
= lnet_cpt_of_nid_locked(nid
);
742 lnet_net_unlock(cpt
);
746 EXPORT_SYMBOL(lnet_cpt_of_nid
);
749 lnet_islocalnet(__u32 net
)
754 cpt
= lnet_net_lock_current();
756 ni
= lnet_net2ni_locked(net
, cpt
);
758 lnet_ni_decref_locked(ni
, cpt
);
760 lnet_net_unlock(cpt
);
766 lnet_nid2ni_locked(lnet_nid_t nid
, int cpt
)
769 struct list_head
*tmp
;
771 LASSERT(cpt
!= LNET_LOCK_EX
);
773 list_for_each(tmp
, &the_lnet
.ln_nis
) {
774 ni
= list_entry(tmp
, lnet_ni_t
, ni_list
);
776 if (ni
->ni_nid
== nid
) {
777 lnet_ni_addref_locked(ni
, cpt
);
786 lnet_islocalnid(lnet_nid_t nid
)
791 cpt
= lnet_net_lock_current();
792 ni
= lnet_nid2ni_locked(nid
, cpt
);
794 lnet_ni_decref_locked(ni
, cpt
);
795 lnet_net_unlock(cpt
);
801 lnet_count_acceptor_nis(void)
803 /* Return the # of NIs that need the acceptor. */
805 struct list_head
*tmp
;
809 cpt
= lnet_net_lock_current();
810 list_for_each(tmp
, &the_lnet
.ln_nis
) {
811 ni
= list_entry(tmp
, lnet_ni_t
, ni_list
);
813 if (ni
->ni_lnd
->lnd_accept
)
817 lnet_net_unlock(cpt
);
823 lnet_ni_tq_credits(lnet_ni_t
*ni
)
827 LASSERT(ni
->ni_ncpts
>= 1);
829 if (ni
->ni_ncpts
== 1)
830 return ni
->ni_maxtxcredits
;
832 credits
= ni
->ni_maxtxcredits
/ ni
->ni_ncpts
;
833 credits
= max(credits
, 8 * ni
->ni_peertxcredits
);
834 credits
= min(credits
, ni
->ni_maxtxcredits
);
840 lnet_shutdown_lndnis(void)
846 /* NB called holding the global mutex */
848 /* All quiet on the API front */
849 LASSERT(!the_lnet
.ln_shutdown
);
850 LASSERT(the_lnet
.ln_refcount
== 0);
851 LASSERT(list_empty(&the_lnet
.ln_nis_zombie
));
853 lnet_net_lock(LNET_LOCK_EX
);
854 the_lnet
.ln_shutdown
= 1; /* flag shutdown */
856 /* Unlink NIs from the global table */
857 while (!list_empty(&the_lnet
.ln_nis
)) {
858 ni
= list_entry(the_lnet
.ln_nis
.next
,
860 /* move it to zombie list and nobody can find it anymore */
861 list_move(&ni
->ni_list
, &the_lnet
.ln_nis_zombie
);
862 lnet_ni_decref_locked(ni
, 0); /* drop ln_nis' ref */
864 if (!list_empty(&ni
->ni_cptlist
)) {
865 list_del_init(&ni
->ni_cptlist
);
866 lnet_ni_decref_locked(ni
, 0);
870 /* Drop the cached eqwait NI. */
871 if (the_lnet
.ln_eq_waitni
) {
872 lnet_ni_decref_locked(the_lnet
.ln_eq_waitni
, 0);
873 the_lnet
.ln_eq_waitni
= NULL
;
876 /* Drop the cached loopback NI. */
877 if (the_lnet
.ln_loni
) {
878 lnet_ni_decref_locked(the_lnet
.ln_loni
, 0);
879 the_lnet
.ln_loni
= NULL
;
882 lnet_net_unlock(LNET_LOCK_EX
);
885 * Clear lazy portals and drop delayed messages which hold refs
886 * on their lnet_msg_t::msg_rxpeer
888 for (i
= 0; i
< the_lnet
.ln_nportals
; i
++)
889 LNetClearLazyPortal(i
);
892 * Clear the peer table and wait for all peers to go (they hold refs on
895 lnet_peer_tables_cleanup();
897 lnet_net_lock(LNET_LOCK_EX
);
899 * Now wait for the NI's I just nuked to show up on ln_zombie_nis
900 * and shut them down in guaranteed thread context
903 while (!list_empty(&the_lnet
.ln_nis_zombie
)) {
907 ni
= list_entry(the_lnet
.ln_nis_zombie
.next
,
909 list_del_init(&ni
->ni_list
);
910 cfs_percpt_for_each(ref
, j
, ni
->ni_refs
) {
913 /* still busy, add it back to zombie list */
914 list_add(&ni
->ni_list
, &the_lnet
.ln_nis_zombie
);
918 if (!list_empty(&ni
->ni_list
)) {
919 lnet_net_unlock(LNET_LOCK_EX
);
921 if ((i
& (-i
)) == i
) {
922 CDEBUG(D_WARNING
, "Waiting for zombie LNI %s\n",
923 libcfs_nid2str(ni
->ni_nid
));
925 set_current_state(TASK_UNINTERRUPTIBLE
);
926 schedule_timeout(cfs_time_seconds(1));
927 lnet_net_lock(LNET_LOCK_EX
);
931 ni
->ni_lnd
->lnd_refcount
--;
932 lnet_net_unlock(LNET_LOCK_EX
);
934 islo
= ni
->ni_lnd
->lnd_type
== LOLND
;
936 LASSERT(!in_interrupt());
937 ni
->ni_lnd
->lnd_shutdown(ni
);
940 * can't deref lnd anymore now; it might have unregistered
944 CDEBUG(D_LNI
, "Removed LNI %s\n",
945 libcfs_nid2str(ni
->ni_nid
));
950 lnet_net_lock(LNET_LOCK_EX
);
953 the_lnet
.ln_shutdown
= 0;
954 lnet_net_unlock(LNET_LOCK_EX
);
956 if (the_lnet
.ln_network_tokens
) {
957 LIBCFS_FREE(the_lnet
.ln_network_tokens
,
958 the_lnet
.ln_network_tokens_nob
);
959 the_lnet
.ln_network_tokens
= NULL
;
964 lnet_startup_lndnis(void)
968 struct lnet_tx_queue
*tq
;
969 struct list_head nilist
;
974 char *nets
= lnet_get_networks();
976 INIT_LIST_HEAD(&nilist
);
981 rc
= lnet_parse_networks(&nilist
, nets
);
985 while (!list_empty(&nilist
)) {
986 ni
= list_entry(nilist
.next
, lnet_ni_t
, ni_list
);
987 lnd_type
= LNET_NETTYP(LNET_NIDNET(ni
->ni_nid
));
989 LASSERT(libcfs_isknown_lnd(lnd_type
));
991 if (lnd_type
== CIBLND
||
992 lnd_type
== OPENIBLND
||
993 lnd_type
== IIBLND
||
994 lnd_type
== VIBLND
) {
995 CERROR("LND %s obsoleted\n",
996 libcfs_lnd2str(lnd_type
));
1000 mutex_lock(&the_lnet
.ln_lnd_mutex
);
1001 lnd
= lnet_find_lnd_by_type(lnd_type
);
1004 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
1005 rc
= request_module("%s",
1006 libcfs_lnd2modname(lnd_type
));
1007 mutex_lock(&the_lnet
.ln_lnd_mutex
);
1009 lnd
= lnet_find_lnd_by_type(lnd_type
);
1011 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
1012 CERROR("Can't load LND %s, module %s, rc=%d\n",
1013 libcfs_lnd2str(lnd_type
),
1014 libcfs_lnd2modname(lnd_type
), rc
);
1019 lnet_net_lock(LNET_LOCK_EX
);
1020 lnd
->lnd_refcount
++;
1021 lnet_net_unlock(LNET_LOCK_EX
);
1025 rc
= lnd
->lnd_startup(ni
);
1027 mutex_unlock(&the_lnet
.ln_lnd_mutex
);
1030 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1031 rc
, libcfs_lnd2str(lnd
->lnd_type
));
1032 lnet_net_lock(LNET_LOCK_EX
);
1033 lnd
->lnd_refcount
--;
1034 lnet_net_unlock(LNET_LOCK_EX
);
1038 LASSERT(ni
->ni_peertimeout
<= 0 || lnd
->lnd_query
);
1040 list_del(&ni
->ni_list
);
1042 lnet_net_lock(LNET_LOCK_EX
);
1043 /* refcount for ln_nis */
1044 lnet_ni_addref_locked(ni
, 0);
1045 list_add_tail(&ni
->ni_list
, &the_lnet
.ln_nis
);
1047 list_add_tail(&ni
->ni_cptlist
,
1048 &the_lnet
.ln_nis_cpt
);
1049 lnet_ni_addref_locked(ni
, 0);
1052 lnet_net_unlock(LNET_LOCK_EX
);
1054 if (lnd
->lnd_type
== LOLND
) {
1056 LASSERT(!the_lnet
.ln_loni
);
1057 the_lnet
.ln_loni
= ni
;
1061 if (ni
->ni_peertxcredits
== 0 ||
1062 ni
->ni_maxtxcredits
== 0) {
1063 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1064 libcfs_lnd2str(lnd
->lnd_type
),
1065 ni
->ni_peertxcredits
== 0 ?
1070 cfs_percpt_for_each(tq
, i
, ni
->ni_tx_queues
) {
1071 tq
->tq_credits_min
=
1072 tq
->tq_credits_max
=
1073 tq
->tq_credits
= lnet_ni_tq_credits(ni
);
1076 CDEBUG(D_LNI
, "Added LNI %s [%d/%d/%d/%d]\n",
1077 libcfs_nid2str(ni
->ni_nid
), ni
->ni_peertxcredits
,
1078 lnet_ni_tq_credits(ni
) * LNET_CPT_NUMBER
,
1079 ni
->ni_peerrtrcredits
, ni
->ni_peertimeout
);
1084 if (the_lnet
.ln_eq_waitni
&& nicount
> 1) {
1085 lnd_type
= the_lnet
.ln_eq_waitni
->ni_lnd
->lnd_type
;
1086 LCONSOLE_ERROR_MSG(0x109, "LND %s can only run single-network\n",
1087 libcfs_lnd2str(lnd_type
));
1094 lnet_shutdown_lndnis();
1096 while (!list_empty(&nilist
)) {
1097 ni
= list_entry(nilist
.next
, lnet_ni_t
, ni_list
);
1098 list_del(&ni
->ni_list
);
1106 * Initialize LNet library.
1108 * Only userspace program needs to call this function - it's automatically
1109 * called in the kernel at module loading time. Caller has to call lnet_fini()
1110 * after a call to lnet_init(), if and only if the latter returned 0. It must
1111 * be called exactly once.
1113 * \return 0 on success, and -ve on failures.
1120 lnet_assert_wire_constants();
1121 LASSERT(!the_lnet
.ln_init
);
1123 memset(&the_lnet
, 0, sizeof(the_lnet
));
1125 /* refer to global cfs_cpt_table for now */
1126 the_lnet
.ln_cpt_table
= cfs_cpt_table
;
1127 the_lnet
.ln_cpt_number
= cfs_cpt_number(cfs_cpt_table
);
1129 LASSERT(the_lnet
.ln_cpt_number
> 0);
1130 if (the_lnet
.ln_cpt_number
> LNET_CPT_MAX
) {
1131 /* we are under risk of consuming all lh_cookie */
1132 CERROR("Can't have %d CPTs for LNet (max allowed is %d), please change setting of CPT-table and retry\n",
1133 the_lnet
.ln_cpt_number
, LNET_CPT_MAX
);
1137 while ((1 << the_lnet
.ln_cpt_bits
) < the_lnet
.ln_cpt_number
)
1138 the_lnet
.ln_cpt_bits
++;
1140 rc
= lnet_create_locks();
1142 CERROR("Can't create LNet global locks: %d\n", rc
);
1146 the_lnet
.ln_refcount
= 0;
1147 the_lnet
.ln_init
= 1;
1148 LNetInvalidateHandle(&the_lnet
.ln_rc_eqh
);
1149 INIT_LIST_HEAD(&the_lnet
.ln_lnds
);
1150 INIT_LIST_HEAD(&the_lnet
.ln_rcd_zombie
);
1151 INIT_LIST_HEAD(&the_lnet
.ln_rcd_deathrow
);
1154 * The hash table size is the number of bits it takes to express the set
1155 * ln_num_routes, minus 1 (better to under estimate than over so we
1156 * don't waste memory).
1158 if (rnet_htable_size
<= 0)
1159 rnet_htable_size
= LNET_REMOTE_NETS_HASH_DEFAULT
;
1160 else if (rnet_htable_size
> LNET_REMOTE_NETS_HASH_MAX
)
1161 rnet_htable_size
= LNET_REMOTE_NETS_HASH_MAX
;
1162 the_lnet
.ln_remote_nets_hbits
= max_t(int, 1,
1163 order_base_2(rnet_htable_size
) - 1);
1166 * All LNDs apart from the LOLND are in separate modules. They
1167 * register themselves when their module loads, and unregister
1168 * themselves when their module is unloaded.
1170 lnet_register_lnd(&the_lolnd
);
1175 * Finalize LNet library.
1177 * Only userspace program needs to call this function. It can be called
1180 * \pre lnet_init() called with success.
1181 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1186 LASSERT(the_lnet
.ln_init
);
1187 LASSERT(the_lnet
.ln_refcount
== 0);
1189 while (!list_empty(&the_lnet
.ln_lnds
))
1190 lnet_unregister_lnd(list_entry(the_lnet
.ln_lnds
.next
,
1192 lnet_destroy_locks();
1194 the_lnet
.ln_init
= 0;
1198 * Set LNet PID and start LNet interfaces, routing, and forwarding.
1200 * Userspace program should call this after a successful call to lnet_init().
1201 * Users must call this function at least once before any other functions.
1202 * For each successful call there must be a corresponding call to
1203 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1206 * The PID used by LNet may be different from the one requested.
1209 * \param requested_pid PID requested by the caller.
1211 * \return >= 0 on success, and < 0 error code on failures.
1214 LNetNIInit(lnet_pid_t requested_pid
)
1216 int im_a_router
= 0;
1219 mutex_lock(&the_lnet
.ln_api_mutex
);
1221 LASSERT(the_lnet
.ln_init
);
1222 CDEBUG(D_OTHER
, "refs %d\n", the_lnet
.ln_refcount
);
1224 if (the_lnet
.ln_refcount
> 0) {
1225 rc
= the_lnet
.ln_refcount
++;
1229 if (requested_pid
== LNET_PID_ANY
) {
1230 /* Don't instantiate LNET just for me */
1235 rc
= lnet_prepare(requested_pid
);
1239 rc
= lnet_startup_lndnis();
1243 rc
= lnet_parse_routes(lnet_get_routes(), &im_a_router
);
1247 rc
= lnet_check_routes();
1251 rc
= lnet_rtrpools_alloc(im_a_router
);
1255 rc
= lnet_acceptor_start();
1259 the_lnet
.ln_refcount
= 1;
1260 /* Now I may use my own API functions... */
1263 * NB router checker needs the_lnet.ln_ping_info in
1264 * lnet_router_checker -> lnet_update_ni_status_locked
1266 rc
= lnet_ping_target_init();
1270 rc
= lnet_router_checker_start();
1274 lnet_router_debugfs_init();
1278 lnet_ping_target_fini();
1280 the_lnet
.ln_refcount
= 0;
1281 lnet_acceptor_stop();
1283 lnet_destroy_routes();
1284 lnet_shutdown_lndnis();
1290 mutex_unlock(&the_lnet
.ln_api_mutex
);
1293 EXPORT_SYMBOL(LNetNIInit
);
1296 * Stop LNet interfaces, routing, and forwarding.
1298 * Users must call this function once for each successful call to LNetNIInit().
1299 * Once the LNetNIFini() operation has been started, the results of pending
1300 * API operations are undefined.
1302 * \return always 0 for current implementation.
1307 mutex_lock(&the_lnet
.ln_api_mutex
);
1309 LASSERT(the_lnet
.ln_init
);
1310 LASSERT(the_lnet
.ln_refcount
> 0);
1312 if (the_lnet
.ln_refcount
!= 1) {
1313 the_lnet
.ln_refcount
--;
1315 LASSERT(!the_lnet
.ln_niinit_self
);
1317 lnet_router_debugfs_fini();
1318 lnet_router_checker_stop();
1319 lnet_ping_target_fini();
1321 /* Teardown fns that use my own API functions BEFORE here */
1322 the_lnet
.ln_refcount
= 0;
1324 lnet_acceptor_stop();
1325 lnet_destroy_routes();
1326 lnet_shutdown_lndnis();
1330 mutex_unlock(&the_lnet
.ln_api_mutex
);
1333 EXPORT_SYMBOL(LNetNIFini
);
1336 * LNet ioctl handler.
1340 LNetCtl(unsigned int cmd
, void *arg
)
1342 struct libcfs_ioctl_data
*data
= arg
;
1343 lnet_process_id_t id
= {0};
1346 unsigned long secs_passed
;
1348 LASSERT(the_lnet
.ln_init
);
1349 LASSERT(the_lnet
.ln_refcount
> 0);
1352 case IOC_LIBCFS_GET_NI
:
1353 rc
= LNetGetId(data
->ioc_count
, &id
);
1354 data
->ioc_nid
= id
.nid
;
1357 case IOC_LIBCFS_FAIL_NID
:
1358 return lnet_fail_nid(data
->ioc_nid
, data
->ioc_count
);
1360 case IOC_LIBCFS_ADD_ROUTE
:
1361 rc
= lnet_add_route(data
->ioc_net
, data
->ioc_count
,
1362 data
->ioc_nid
, data
->ioc_priority
);
1363 return (rc
!= 0) ? rc
: lnet_check_routes();
1365 case IOC_LIBCFS_DEL_ROUTE
:
1366 return lnet_del_route(data
->ioc_net
, data
->ioc_nid
);
1368 case IOC_LIBCFS_GET_ROUTE
:
1369 return lnet_get_route(data
->ioc_count
,
1370 &data
->ioc_net
, &data
->ioc_count
,
1371 &data
->ioc_nid
, &data
->ioc_flags
,
1372 &data
->ioc_priority
);
1373 case IOC_LIBCFS_NOTIFY_ROUTER
:
1374 secs_passed
= (ktime_get_real_seconds() - data
->ioc_u64
[0]);
1375 return lnet_notify(NULL
, data
->ioc_nid
, data
->ioc_flags
,
1376 jiffies
- secs_passed
* HZ
);
1378 case IOC_LIBCFS_LNET_DIST
:
1379 rc
= LNetDist(data
->ioc_nid
, &data
->ioc_nid
, &data
->ioc_u32
[1]);
1380 if (rc
< 0 && rc
!= -EHOSTUNREACH
)
1383 data
->ioc_u32
[0] = rc
;
1386 case IOC_LIBCFS_TESTPROTOCOMPAT
:
1387 lnet_net_lock(LNET_LOCK_EX
);
1388 the_lnet
.ln_testprotocompat
= data
->ioc_flags
;
1389 lnet_net_unlock(LNET_LOCK_EX
);
1392 case IOC_LIBCFS_PING
:
1393 id
.nid
= data
->ioc_nid
;
1394 id
.pid
= data
->ioc_u32
[0];
1395 rc
= lnet_ping(id
, data
->ioc_u32
[1], /* timeout */
1397 data
->ioc_plen1
/ sizeof(lnet_process_id_t
));
1400 data
->ioc_count
= rc
;
1404 ni
= lnet_net2ni(data
->ioc_net
);
1408 if (!ni
->ni_lnd
->lnd_ctl
)
1411 rc
= ni
->ni_lnd
->lnd_ctl(ni
, cmd
, arg
);
1418 EXPORT_SYMBOL(LNetCtl
);
1420 void LNetDebugPeer(lnet_process_id_t id
)
1422 lnet_debug_peer(id
.nid
);
1424 EXPORT_SYMBOL(LNetDebugPeer
);
1427 * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
1428 * all interfaces share a same PID, as requested by LNetNIInit().
1430 * \param index Index of the interface to look up.
1431 * \param id On successful return, this location will hold the
1432 * lnet_process_id_t ID of the interface.
1434 * \retval 0 If an interface exists at \a index.
1435 * \retval -ENOENT If no interface has been found.
1438 LNetGetId(unsigned int index
, lnet_process_id_t
*id
)
1441 struct list_head
*tmp
;
1445 LASSERT(the_lnet
.ln_init
);
1447 /* LNetNI initilization failed? */
1448 if (the_lnet
.ln_refcount
== 0)
1451 cpt
= lnet_net_lock_current();
1453 list_for_each(tmp
, &the_lnet
.ln_nis
) {
1457 ni
= list_entry(tmp
, lnet_ni_t
, ni_list
);
1459 id
->nid
= ni
->ni_nid
;
1460 id
->pid
= the_lnet
.ln_pid
;
1465 lnet_net_unlock(cpt
);
1468 EXPORT_SYMBOL(LNetGetId
);
1471 * Print a string representation of handle \a h into buffer \a str of
1475 LNetSnprintHandle(char *str
, int len
, lnet_handle_any_t h
)
1477 snprintf(str
, len
, "%#llx", h
.cookie
);
1479 EXPORT_SYMBOL(LNetSnprintHandle
);
1482 lnet_create_ping_info(void)
1487 unsigned int infosz
;
1489 lnet_process_id_t id
;
1490 lnet_ping_info_t
*pinfo
;
1492 for (n
= 0; ; n
++) {
1493 rc
= LNetGetId(n
, &id
);
1500 infosz
= offsetof(lnet_ping_info_t
, pi_ni
[n
]);
1501 LIBCFS_ALLOC(pinfo
, infosz
);
1503 CERROR("Can't allocate ping info[%d]\n", n
);
1508 pinfo
->pi_pid
= the_lnet
.ln_pid
;
1509 pinfo
->pi_magic
= LNET_PROTO_PING_MAGIC
;
1510 pinfo
->pi_features
= LNET_PING_FEAT_NI_STATUS
;
1512 for (i
= 0; i
< n
; i
++) {
1513 lnet_ni_status_t
*ns
= &pinfo
->pi_ni
[i
];
1515 rc
= LNetGetId(i
, &id
);
1518 ns
->ns_nid
= id
.nid
;
1519 ns
->ns_status
= LNET_NI_STATUS_UP
;
1523 ni
= lnet_nid2ni_locked(id
.nid
, 0);
1527 LASSERT(!ni
->ni_status
);
1531 lnet_ni_decref_locked(ni
, 0);
1535 the_lnet
.ln_ping_info
= pinfo
;
1540 lnet_destroy_ping_info(void)
1546 list_for_each_entry(ni
, &the_lnet
.ln_nis
, ni_list
) {
1548 ni
->ni_status
= NULL
;
1554 LIBCFS_FREE(the_lnet
.ln_ping_info
,
1555 offsetof(lnet_ping_info_t
,
1556 pi_ni
[the_lnet
.ln_ping_info
->pi_nnis
]));
1557 the_lnet
.ln_ping_info
= NULL
;
1561 lnet_ping_target_init(void)
1563 lnet_md_t md
= { NULL
};
1564 lnet_handle_me_t meh
;
1565 lnet_process_id_t id
;
1570 rc
= lnet_create_ping_info();
1575 * We can have a tiny EQ since we only need to see the unlink event on
1576 * teardown, which by definition is the last one!
1578 rc
= LNetEQAlloc(2, LNET_EQ_HANDLER_NONE
, &the_lnet
.ln_ping_target_eq
);
1580 CERROR("Can't allocate ping EQ: %d\n", rc
);
1584 memset(&id
, 0, sizeof(lnet_process_id_t
));
1585 id
.nid
= LNET_NID_ANY
;
1586 id
.pid
= LNET_PID_ANY
;
1588 rc
= LNetMEAttach(LNET_RESERVED_PORTAL
, id
,
1589 LNET_PROTO_PING_MATCHBITS
, 0,
1590 LNET_UNLINK
, LNET_INS_AFTER
,
1593 CERROR("Can't create ping ME: %d\n", rc
);
1597 /* initialize md content */
1598 infosz
= offsetof(lnet_ping_info_t
,
1599 pi_ni
[the_lnet
.ln_ping_info
->pi_nnis
]);
1600 md
.start
= the_lnet
.ln_ping_info
;
1602 md
.threshold
= LNET_MD_THRESH_INF
;
1604 md
.options
= LNET_MD_OP_GET
| LNET_MD_TRUNCATE
|
1605 LNET_MD_MANAGE_REMOTE
;
1607 md
.eq_handle
= the_lnet
.ln_ping_target_eq
;
1609 rc
= LNetMDAttach(meh
, md
,
1611 &the_lnet
.ln_ping_target_md
);
1613 CERROR("Can't attach ping MD: %d\n", rc
);
1620 rc2
= LNetMEUnlink(meh
);
1623 rc2
= LNetEQFree(the_lnet
.ln_ping_target_eq
);
1626 lnet_destroy_ping_info();
1631 lnet_ping_target_fini(void)
1636 int timeout_ms
= 1000;
1637 sigset_t blocked
= cfs_block_allsigs();
1639 LNetMDUnlink(the_lnet
.ln_ping_target_md
);
1640 /* NB md could be busy; this just starts the unlink */
1643 rc
= LNetEQPoll(&the_lnet
.ln_ping_target_eq
, 1,
1644 timeout_ms
, &event
, &which
);
1646 /* I expect overflow... */
1647 LASSERT(rc
>= 0 || rc
== -EOVERFLOW
);
1650 /* timed out: provide a diagnostic */
1651 CWARN("Still waiting for ping MD to unlink\n");
1656 /* Got a valid event */
1661 rc
= LNetEQFree(the_lnet
.ln_ping_target_eq
);
1663 lnet_destroy_ping_info();
1664 cfs_restore_sigs(blocked
);
1667 static int lnet_ping(lnet_process_id_t id
, int timeout_ms
,
1668 lnet_process_id_t __user
*ids
, int n_ids
)
1670 lnet_handle_eq_t eqh
;
1671 lnet_handle_md_t mdh
;
1673 lnet_md_t md
= { NULL
};
1677 const int a_long_time
= 60000; /* mS */
1678 int infosz
= offsetof(lnet_ping_info_t
, pi_ni
[n_ids
]);
1679 lnet_ping_info_t
*info
;
1680 lnet_process_id_t tmpid
;
1688 id
.nid
== LNET_NID_ANY
||
1689 timeout_ms
> 500000 || /* arbitrary limit! */
1690 n_ids
> 20) /* arbitrary limit! */
1693 if (id
.pid
== LNET_PID_ANY
)
1694 id
.pid
= LUSTRE_SRV_LNET_PID
;
1696 LIBCFS_ALLOC(info
, infosz
);
1700 /* NB 2 events max (including any unlink event) */
1701 rc
= LNetEQAlloc(2, LNET_EQ_HANDLER_NONE
, &eqh
);
1703 CERROR("Can't allocate EQ: %d\n", rc
);
1707 /* initialize md content */
1710 md
.threshold
= 2; /*GET/REPLY*/
1712 md
.options
= LNET_MD_TRUNCATE
;
1716 rc
= LNetMDBind(md
, LNET_UNLINK
, &mdh
);
1718 CERROR("Can't bind MD: %d\n", rc
);
1722 rc
= LNetGet(LNET_NID_ANY
, mdh
, id
,
1723 LNET_RESERVED_PORTAL
,
1724 LNET_PROTO_PING_MATCHBITS
, 0);
1727 /* Don't CERROR; this could be deliberate! */
1729 rc2
= LNetMDUnlink(mdh
);
1732 /* NB must wait for the UNLINK event below... */
1734 timeout_ms
= a_long_time
;
1738 /* MUST block for unlink to complete */
1740 blocked
= cfs_block_allsigs();
1742 rc2
= LNetEQPoll(&eqh
, 1, timeout_ms
, &event
, &which
);
1745 cfs_restore_sigs(blocked
);
1747 CDEBUG(D_NET
, "poll %d(%d %d)%s\n", rc2
,
1748 (rc2
<= 0) ? -1 : event
.type
,
1749 (rc2
<= 0) ? -1 : event
.status
,
1750 (rc2
> 0 && event
.unlinked
) ? " unlinked" : "");
1752 LASSERT(rc2
!= -EOVERFLOW
); /* can't miss anything */
1754 if (rc2
<= 0 || event
.status
!= 0) {
1755 /* timeout or error */
1756 if (!replied
&& rc
== 0)
1757 rc
= (rc2
< 0) ? rc2
:
1758 (rc2
== 0) ? -ETIMEDOUT
:
1762 /* Ensure completion in finite time... */
1764 /* No assertion (racing with network) */
1766 timeout_ms
= a_long_time
;
1767 } else if (rc2
== 0) {
1768 /* timed out waiting for unlink */
1769 CWARN("ping %s: late network completion\n",
1772 } else if (event
.type
== LNET_EVENT_REPLY
) {
1777 } while (rc2
<= 0 || !event
.unlinked
);
1781 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
1788 LASSERT(nob
>= 0 && nob
<= infosz
);
1790 rc
= -EPROTO
; /* if I can't parse... */
1793 /* can't check magic/version */
1794 CERROR("%s: ping info too short %d\n",
1795 libcfs_id2str(id
), nob
);
1799 if (info
->pi_magic
== __swab32(LNET_PROTO_PING_MAGIC
)) {
1800 lnet_swap_pinginfo(info
);
1801 } else if (info
->pi_magic
!= LNET_PROTO_PING_MAGIC
) {
1802 CERROR("%s: Unexpected magic %08x\n",
1803 libcfs_id2str(id
), info
->pi_magic
);
1807 if ((info
->pi_features
& LNET_PING_FEAT_NI_STATUS
) == 0) {
1808 CERROR("%s: ping w/o NI status: 0x%x\n",
1809 libcfs_id2str(id
), info
->pi_features
);
1813 if (nob
< offsetof(lnet_ping_info_t
, pi_ni
[0])) {
1814 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id
),
1815 nob
, (int)offsetof(lnet_ping_info_t
, pi_ni
[0]));
1819 if (info
->pi_nnis
< n_ids
)
1820 n_ids
= info
->pi_nnis
;
1822 if (nob
< offsetof(lnet_ping_info_t
, pi_ni
[n_ids
])) {
1823 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id
),
1824 nob
, (int)offsetof(lnet_ping_info_t
, pi_ni
[n_ids
]));
1828 rc
= -EFAULT
; /* If I SEGV... */
1830 memset(&tmpid
, 0, sizeof(tmpid
));
1831 for (i
= 0; i
< n_ids
; i
++) {
1832 tmpid
.pid
= info
->pi_pid
;
1833 tmpid
.nid
= info
->pi_ni
[i
].ns_nid
;
1834 if (copy_to_user(&ids
[i
], &tmpid
, sizeof(tmpid
)))
1840 rc2
= LNetEQFree(eqh
);
1842 CERROR("rc2 %d\n", rc2
);
1846 LIBCFS_FREE(info
, infosz
);