projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
staging: lustre: remove wrappers for two atomic functions
[linux-2.6-block.git] / drivers / staging / lustre / lnet / klnds / socklnd / socklnd.c
CommitLineData
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 * lnet/klnds/socklnd/socklnd.c
37 *
38 * Author: Zach Brown <zab@zabbo.net>
39 * Author: Peter J. Braam <braam@clusterfs.com>
40 * Author: Phil Schwan <phil@clusterfs.com>
41 * Author: Eric Barton <eric@bartonsoftware.com>
42 */
43
44#include "socklnd.h"
45
46lnd_t the_ksocklnd;
47ksock_nal_data_t ksocknal_data;
48
49ksock_interface_t *
50ksocknal_ip2iface(lnet_ni_t *ni, __u32 ip)
51{
52 ksock_net_t *net = ni->ni_data;
53 int i;
54 ksock_interface_t *iface;
55
56 for (i = 0; i < net->ksnn_ninterfaces; i++) {
57 LASSERT(i < LNET_MAX_INTERFACES);
58 iface = &net->ksnn_interfaces[i];
59
60 if (iface->ksni_ipaddr == ip)
61 return (iface);
62 }
63
64 return (NULL);
65}
66
67ksock_route_t *
68ksocknal_create_route (__u32 ipaddr, int port)
69{
70 ksock_route_t *route;
71
72 LIBCFS_ALLOC (route, sizeof (*route));
73 if (route == NULL)
74 return (NULL);
75
76 atomic_set (&route->ksnr_refcount, 1);
77 route->ksnr_peer = NULL;
78 route->ksnr_retry_interval = 0; /* OK to connect at any time */
79 route->ksnr_ipaddr = ipaddr;
80 route->ksnr_port = port;
81 route->ksnr_scheduled = 0;
82 route->ksnr_connecting = 0;
83 route->ksnr_connected = 0;
84 route->ksnr_deleted = 0;
85 route->ksnr_conn_count = 0;
86 route->ksnr_share_count = 0;
87
88 return (route);
89}
90
91void
92ksocknal_destroy_route (ksock_route_t *route)
93{
94 LASSERT (atomic_read(&route->ksnr_refcount) == 0);
95
96 if (route->ksnr_peer != NULL)
97 ksocknal_peer_decref(route->ksnr_peer);
98
99 LIBCFS_FREE (route, sizeof (*route));
100}
101
102int
103ksocknal_create_peer (ksock_peer_t **peerp, lnet_ni_t *ni, lnet_process_id_t id)
104{
105 ksock_net_t *net = ni->ni_data;
106 ksock_peer_t *peer;
107
108 LASSERT (id.nid != LNET_NID_ANY);
109 LASSERT (id.pid != LNET_PID_ANY);
110 LASSERT (!in_interrupt());
111
112 LIBCFS_ALLOC (peer, sizeof (*peer));
113 if (peer == NULL)
114 return -ENOMEM;
115
116 memset (peer, 0, sizeof (*peer)); /* NULL pointers/clear flags etc */
117
118 peer->ksnp_ni = ni;
119 peer->ksnp_id = id;
120 atomic_set (&peer->ksnp_refcount, 1); /* 1 ref for caller */
121 peer->ksnp_closing = 0;
122 peer->ksnp_accepting = 0;
123 peer->ksnp_proto = NULL;
124 peer->ksnp_last_alive = 0;
125 peer->ksnp_zc_next_cookie = SOCKNAL_KEEPALIVE_PING + 1;
126
127 INIT_LIST_HEAD (&peer->ksnp_conns);
128 INIT_LIST_HEAD (&peer->ksnp_routes);
129 INIT_LIST_HEAD (&peer->ksnp_tx_queue);
130 INIT_LIST_HEAD (&peer->ksnp_zc_req_list);
131 spin_lock_init(&peer->ksnp_lock);
132
133 spin_lock_bh(&net->ksnn_lock);
134
135 if (net->ksnn_shutdown) {
136 spin_unlock_bh(&net->ksnn_lock);
137
138 LIBCFS_FREE(peer, sizeof(*peer));
139 CERROR("Can't create peer: network shutdown\n");
140 return -ESHUTDOWN;
141 }
142
143 net->ksnn_npeers++;
144
145 spin_unlock_bh(&net->ksnn_lock);
146
147 *peerp = peer;
148 return 0;
149}
150
151void
152ksocknal_destroy_peer (ksock_peer_t *peer)
153{
154 ksock_net_t *net = peer->ksnp_ni->ni_data;
155
156 CDEBUG (D_NET, "peer %s %p deleted\n",
157 libcfs_id2str(peer->ksnp_id), peer);
158
159 LASSERT (atomic_read (&peer->ksnp_refcount) == 0);
160 LASSERT (peer->ksnp_accepting == 0);
161 LASSERT (list_empty (&peer->ksnp_conns));
162 LASSERT (list_empty (&peer->ksnp_routes));
163 LASSERT (list_empty (&peer->ksnp_tx_queue));
164 LASSERT (list_empty (&peer->ksnp_zc_req_list));
165
166 LIBCFS_FREE (peer, sizeof (*peer));
167
168 /* NB a peer's connections and routes keep a reference on their peer
169 * until they are destroyed, so we can be assured that _all_ state to
170 * do with this peer has been cleaned up when its refcount drops to
171 * zero. */
172 spin_lock_bh(&net->ksnn_lock);
173 net->ksnn_npeers--;
174 spin_unlock_bh(&net->ksnn_lock);
175}
176
177ksock_peer_t *
178ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id)
179{
180 struct list_head *peer_list = ksocknal_nid2peerlist(id.nid);
181 struct list_head *tmp;
182 ksock_peer_t *peer;
183
184 list_for_each (tmp, peer_list) {
185
186 peer = list_entry (tmp, ksock_peer_t, ksnp_list);
187
188 LASSERT (!peer->ksnp_closing);
189
190 if (peer->ksnp_ni != ni)
191 continue;
192
193 if (peer->ksnp_id.nid != id.nid ||
194 peer->ksnp_id.pid != id.pid)
195 continue;
196
197 CDEBUG(D_NET, "got peer [%p] -> %s (%d)\n",
198 peer, libcfs_id2str(id),
199 atomic_read(&peer->ksnp_refcount));
200 return (peer);
201 }
202 return (NULL);
203}
204
205ksock_peer_t *
206ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id)
207{
208 ksock_peer_t *peer;
209
210 read_lock(&ksocknal_data.ksnd_global_lock);
211 peer = ksocknal_find_peer_locked(ni, id);
212 if (peer != NULL) /* +1 ref for caller? */
213 ksocknal_peer_addref(peer);
214 read_unlock(&ksocknal_data.ksnd_global_lock);
215
216 return (peer);
217}
218
219void
220ksocknal_unlink_peer_locked (ksock_peer_t *peer)
221{
222 int i;
223 __u32 ip;
224 ksock_interface_t *iface;
225
226 for (i = 0; i < peer->ksnp_n_passive_ips; i++) {
227 LASSERT (i < LNET_MAX_INTERFACES);
228 ip = peer->ksnp_passive_ips[i];
229
230 iface = ksocknal_ip2iface(peer->ksnp_ni, ip);
231 /* All IPs in peer->ksnp_passive_ips[] come from the
232 * interface list, therefore the call must succeed. */
233 LASSERT (iface != NULL);
234
235 CDEBUG(D_NET, "peer=%p iface=%p ksni_nroutes=%d\n",
236 peer, iface, iface->ksni_nroutes);
237 iface->ksni_npeers--;
238 }
239
240 LASSERT (list_empty(&peer->ksnp_conns));
241 LASSERT (list_empty(&peer->ksnp_routes));
242 LASSERT (!peer->ksnp_closing);
243 peer->ksnp_closing = 1;
244 list_del (&peer->ksnp_list);
245 /* lose peerlist's ref */
246 ksocknal_peer_decref(peer);
247}
248
249int
250ksocknal_get_peer_info (lnet_ni_t *ni, int index,
251 lnet_process_id_t *id, __u32 *myip, __u32 *peer_ip,
252 int *port, int *conn_count, int *share_count)
253{
254 ksock_peer_t *peer;
255 struct list_head *ptmp;
256 ksock_route_t *route;
257 struct list_head *rtmp;
258 int i;
259 int j;
260 int rc = -ENOENT;
261
262 read_lock(&ksocknal_data.ksnd_global_lock);
263
264 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
265
266 list_for_each (ptmp, &ksocknal_data.ksnd_peers[i]) {
267 peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
268
269 if (peer->ksnp_ni != ni)
270 continue;
271
272 if (peer->ksnp_n_passive_ips == 0 &&
273 list_empty(&peer->ksnp_routes)) {
274 if (index-- > 0)
275 continue;
276
277 *id = peer->ksnp_id;
278 *myip = 0;
279 *peer_ip = 0;
280 *port = 0;
281 *conn_count = 0;
282 *share_count = 0;
283 rc = 0;
284 goto out;
285 }
286
287 for (j = 0; j < peer->ksnp_n_passive_ips; j++) {
288 if (index-- > 0)
289 continue;
290
291 *id = peer->ksnp_id;
292 *myip = peer->ksnp_passive_ips[j];
293 *peer_ip = 0;
294 *port = 0;
295 *conn_count = 0;
296 *share_count = 0;
297 rc = 0;
298 goto out;
299 }
300
301 list_for_each (rtmp, &peer->ksnp_routes) {
302 if (index-- > 0)
303 continue;
304
305 route = list_entry(rtmp, ksock_route_t,
306 ksnr_list);
307
308 *id = peer->ksnp_id;
309 *myip = route->ksnr_myipaddr;
310 *peer_ip = route->ksnr_ipaddr;
311 *port = route->ksnr_port;
312 *conn_count = route->ksnr_conn_count;
313 *share_count = route->ksnr_share_count;
314 rc = 0;
315 goto out;
316 }
317 }
318 }
319 out:
320 read_unlock(&ksocknal_data.ksnd_global_lock);
321 return (rc);
322}
323
324void
325ksocknal_associate_route_conn_locked(ksock_route_t *route, ksock_conn_t *conn)
326{
327 ksock_peer_t *peer = route->ksnr_peer;
328 int type = conn->ksnc_type;
329 ksock_interface_t *iface;
330
331 conn->ksnc_route = route;
332 ksocknal_route_addref(route);
333
334 if (route->ksnr_myipaddr != conn->ksnc_myipaddr) {
335 if (route->ksnr_myipaddr == 0) {
336 /* route wasn't bound locally yet (the initial route) */
5e8f6920 337 CDEBUG(D_NET, "Binding %s %pI4h to %pI4h\n",
d7e09d03 338 libcfs_id2str(peer->ksnp_id),
5e8f6920
PT		 339 &route->ksnr_ipaddr,
340 &conn->ksnc_myipaddr);
d7e09d03 341 } else {
5e8f6920
PT		 342 CDEBUG(D_NET, "Rebinding %s %pI4h from "
343 "%pI4h to %pI4h\n",
d7e09d03 344 libcfs_id2str(peer->ksnp_id),
5e8f6920
PT		 345 &route->ksnr_ipaddr,
346 &route->ksnr_myipaddr,
347 &conn->ksnc_myipaddr);
d7e09d03
PT		 348
349 iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
350 route->ksnr_myipaddr);
351 if (iface != NULL)
352 iface->ksni_nroutes--;
353 }
354 route->ksnr_myipaddr = conn->ksnc_myipaddr;
355 iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
356 route->ksnr_myipaddr);
357 if (iface != NULL)
358 iface->ksni_nroutes++;
359 }
360
361 route->ksnr_connected |= (1<<type);
362 route->ksnr_conn_count++;
363
364 /* Successful connection => further attempts can
365 * proceed immediately */
366 route->ksnr_retry_interval = 0;
367}
368
369void
370ksocknal_add_route_locked (ksock_peer_t *peer, ksock_route_t *route)
371{
372 struct list_head *tmp;
373 ksock_conn_t *conn;
374 ksock_route_t *route2;
375
376 LASSERT (!peer->ksnp_closing);
377 LASSERT (route->ksnr_peer == NULL);
378 LASSERT (!route->ksnr_scheduled);
379 LASSERT (!route->ksnr_connecting);
380 LASSERT (route->ksnr_connected == 0);
381
382 /* LASSERT(unique) */
383 list_for_each(tmp, &peer->ksnp_routes) {
384 route2 = list_entry(tmp, ksock_route_t, ksnr_list);
385
386 if (route2->ksnr_ipaddr == route->ksnr_ipaddr) {
5e8f6920 387 CERROR("Duplicate route %s %pI4h\n",
d7e09d03 388 libcfs_id2str(peer->ksnp_id),
5e8f6920 389 &route->ksnr_ipaddr);
d7e09d03
PT		 390 LBUG();
391 }
392 }
393
394 route->ksnr_peer = peer;
395 ksocknal_peer_addref(peer);
396 /* peer's routelist takes over my ref on 'route' */
397 list_add_tail(&route->ksnr_list, &peer->ksnp_routes);
398
399 list_for_each(tmp, &peer->ksnp_conns) {
400 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
401
402 if (conn->ksnc_ipaddr != route->ksnr_ipaddr)
403 continue;
404
405 ksocknal_associate_route_conn_locked(route, conn);
406 /* keep going (typed routes) */
407 }
408}
409
410void
411ksocknal_del_route_locked (ksock_route_t *route)
412{
413 ksock_peer_t *peer = route->ksnr_peer;
414 ksock_interface_t *iface;
415 ksock_conn_t *conn;
416 struct list_head *ctmp;
417 struct list_head *cnxt;
418
419 LASSERT (!route->ksnr_deleted);
420
421 /* Close associated conns */
422 list_for_each_safe (ctmp, cnxt, &peer->ksnp_conns) {
423 conn = list_entry(ctmp, ksock_conn_t, ksnc_list);
424
425 if (conn->ksnc_route != route)
426 continue;
427
428 ksocknal_close_conn_locked (conn, 0);
429 }
430
431 if (route->ksnr_myipaddr != 0) {
432 iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
433 route->ksnr_myipaddr);
434 if (iface != NULL)
435 iface->ksni_nroutes--;
436 }
437
438 route->ksnr_deleted = 1;
439 list_del (&route->ksnr_list);
440 ksocknal_route_decref(route); /* drop peer's ref */
441
442 if (list_empty (&peer->ksnp_routes) &&
443 list_empty (&peer->ksnp_conns)) {
444 /* I've just removed the last route to a peer with no active
445 * connections */
446 ksocknal_unlink_peer_locked (peer);
447 }
448}
449
450int
451ksocknal_add_peer (lnet_ni_t *ni, lnet_process_id_t id, __u32 ipaddr, int port)
452{
453 struct list_head *tmp;
454 ksock_peer_t *peer;
455 ksock_peer_t *peer2;
456 ksock_route_t *route;
457 ksock_route_t *route2;
458 int rc;
459
460 if (id.nid == LNET_NID_ANY ||
461 id.pid == LNET_PID_ANY)
462 return (-EINVAL);
463
464 /* Have a brand new peer ready... */
465 rc = ksocknal_create_peer(&peer, ni, id);
466 if (rc != 0)
467 return rc;
468
469 route = ksocknal_create_route (ipaddr, port);
470 if (route == NULL) {
471 ksocknal_peer_decref(peer);
472 return (-ENOMEM);
473 }
474
475 write_lock_bh(&ksocknal_data.ksnd_global_lock);
476
477 /* always called with a ref on ni, so shutdown can't have started */
478 LASSERT (((ksock_net_t *) ni->ni_data)->ksnn_shutdown == 0);
479
480 peer2 = ksocknal_find_peer_locked (ni, id);
481 if (peer2 != NULL) {
482 ksocknal_peer_decref(peer);
483 peer = peer2;
484 } else {
485 /* peer table takes my ref on peer */
486 list_add_tail (&peer->ksnp_list,
487 ksocknal_nid2peerlist (id.nid));
488 }
489
490 route2 = NULL;
491 list_for_each (tmp, &peer->ksnp_routes) {
492 route2 = list_entry(tmp, ksock_route_t, ksnr_list);
493
494 if (route2->ksnr_ipaddr == ipaddr)
495 break;
496
497 route2 = NULL;
498 }
499 if (route2 == NULL) {
500 ksocknal_add_route_locked(peer, route);
501 route->ksnr_share_count++;
502 } else {
503 ksocknal_route_decref(route);
504 route2->ksnr_share_count++;
505 }
506
507 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
508
509 return (0);
510}
511
512void
513ksocknal_del_peer_locked (ksock_peer_t *peer, __u32 ip)
514{
515 ksock_conn_t *conn;
516 ksock_route_t *route;
517 struct list_head *tmp;
518 struct list_head *nxt;
519 int nshared;
520
521 LASSERT (!peer->ksnp_closing);
522
523 /* Extra ref prevents peer disappearing until I'm done with it */
524 ksocknal_peer_addref(peer);
525
526 list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
527 route = list_entry(tmp, ksock_route_t, ksnr_list);
528
529 /* no match */
530 if (!(ip == 0 || route->ksnr_ipaddr == ip))
531 continue;
532
533 route->ksnr_share_count = 0;
534 /* This deletes associated conns too */
535 ksocknal_del_route_locked (route);
536 }
537
538 nshared = 0;
539 list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
540 route = list_entry(tmp, ksock_route_t, ksnr_list);
541 nshared += route->ksnr_share_count;
542 }
543
544 if (nshared == 0) {
545 /* remove everything else if there are no explicit entries
546 * left */
547
548 list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
549 route = list_entry(tmp, ksock_route_t, ksnr_list);
550
551 /* we should only be removing auto-entries */
552 LASSERT(route->ksnr_share_count == 0);
553 ksocknal_del_route_locked (route);
554 }
555
556 list_for_each_safe (tmp, nxt, &peer->ksnp_conns) {
557 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
558
559 ksocknal_close_conn_locked(conn, 0);
560 }
561 }
562
563 ksocknal_peer_decref(peer);
564 /* NB peer unlinks itself when last conn/route is removed */
565}
566
567int
568ksocknal_del_peer (lnet_ni_t *ni, lnet_process_id_t id, __u32 ip)
569{
570 LIST_HEAD (zombies);
571 struct list_head *ptmp;
572 struct list_head *pnxt;
573 ksock_peer_t *peer;
574 int lo;
575 int hi;
576 int i;
577 int rc = -ENOENT;
578
579 write_lock_bh(&ksocknal_data.ksnd_global_lock);
580
581 if (id.nid != LNET_NID_ANY)
582 lo = hi = (int)(ksocknal_nid2peerlist(id.nid) - ksocknal_data.ksnd_peers);
583 else {
584 lo = 0;
585 hi = ksocknal_data.ksnd_peer_hash_size - 1;
586 }
587
588 for (i = lo; i <= hi; i++) {
589 list_for_each_safe (ptmp, pnxt,
590 &ksocknal_data.ksnd_peers[i]) {
591 peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
592
593 if (peer->ksnp_ni != ni)
594 continue;
595
596 if (!((id.nid == LNET_NID_ANY || peer->ksnp_id.nid == id.nid) &&
597 (id.pid == LNET_PID_ANY || peer->ksnp_id.pid == id.pid)))
598 continue;
599
600 ksocknal_peer_addref(peer); /* a ref for me... */
601
602 ksocknal_del_peer_locked (peer, ip);
603
604 if (peer->ksnp_closing &&
605 !list_empty(&peer->ksnp_tx_queue)) {
606 LASSERT (list_empty(&peer->ksnp_conns));
607 LASSERT (list_empty(&peer->ksnp_routes));
608
609 list_splice_init(&peer->ksnp_tx_queue,
610 &zombies);
611 }
612
613 ksocknal_peer_decref(peer); /* ...till here */
614
615 rc = 0; /* matched! */
616 }
617 }
618
619 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
620
621 ksocknal_txlist_done(ni, &zombies, 1);
622
623 return (rc);
624}
625
626ksock_conn_t *
627ksocknal_get_conn_by_idx (lnet_ni_t *ni, int index)
628{
629 ksock_peer_t *peer;
630 struct list_head *ptmp;
631 ksock_conn_t *conn;
632 struct list_head *ctmp;
633 int i;
634
635 read_lock(&ksocknal_data.ksnd_global_lock);
636
637 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
638 list_for_each (ptmp, &ksocknal_data.ksnd_peers[i]) {
639 peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
640
641 LASSERT (!peer->ksnp_closing);
642
643 if (peer->ksnp_ni != ni)
644 continue;
645
646 list_for_each (ctmp, &peer->ksnp_conns) {
647 if (index-- > 0)
648 continue;
649
650 conn = list_entry (ctmp, ksock_conn_t,
651 ksnc_list);
652 ksocknal_conn_addref(conn);
653 read_unlock(&ksocknal_data. \
654 ksnd_global_lock);
655 return (conn);
656 }
657 }
658 }
659
660 read_unlock(&ksocknal_data.ksnd_global_lock);
661 return (NULL);
662}
663
664ksock_sched_t *
665ksocknal_choose_scheduler_locked(unsigned int cpt)
666{
667 struct ksock_sched_info *info = ksocknal_data.ksnd_sched_info[cpt];
668 ksock_sched_t *sched;
669 int i;
670
671 LASSERT(info->ksi_nthreads > 0);
672
673 sched = &info->ksi_scheds[0];
674 /*
675 * NB: it's safe so far, but info->ksi_nthreads could be changed
676 * at runtime when we have dynamic LNet configuration, then we
677 * need to take care of this.
678 */
679 for (i = 1; i < info->ksi_nthreads; i++) {
680 if (sched->kss_nconns > info->ksi_scheds[i].kss_nconns)
681 sched = &info->ksi_scheds[i];
682 }
683
684 return sched;
685}
686
687int
688ksocknal_local_ipvec (lnet_ni_t *ni, __u32 *ipaddrs)
689{
690 ksock_net_t *net = ni->ni_data;
691 int i;
692 int nip;
693
694 read_lock(&ksocknal_data.ksnd_global_lock);
695
696 nip = net->ksnn_ninterfaces;
697 LASSERT (nip <= LNET_MAX_INTERFACES);
698
699 /* Only offer interfaces for additional connections if I have
700 * more than one. */
701 if (nip < 2) {
702 read_unlock(&ksocknal_data.ksnd_global_lock);
703 return 0;
704 }
705
706 for (i = 0; i < nip; i++) {
707 ipaddrs[i] = net->ksnn_interfaces[i].ksni_ipaddr;
708 LASSERT (ipaddrs[i] != 0);
709 }
710
711 read_unlock(&ksocknal_data.ksnd_global_lock);
712 return (nip);
713}
714
715int
716ksocknal_match_peerip (ksock_interface_t *iface, __u32 *ips, int nips)
717{
718 int best_netmatch = 0;
719 int best_xor = 0;
720 int best = -1;
721 int this_xor;
722 int this_netmatch;
723 int i;
724
725 for (i = 0; i < nips; i++) {
726 if (ips[i] == 0)
727 continue;
728
729 this_xor = (ips[i] ^ iface->ksni_ipaddr);
730 this_netmatch = ((this_xor & iface->ksni_netmask) == 0) ? 1 : 0;
731
732 if (!(best < 0 ||
733 best_netmatch < this_netmatch ||
734 (best_netmatch == this_netmatch &&
735 best_xor > this_xor)))
736 continue;
737
738 best = i;
739 best_netmatch = this_netmatch;
740 best_xor = this_xor;
741 }
742
743 LASSERT (best >= 0);
744 return (best);
745}
746
747int
748ksocknal_select_ips(ksock_peer_t *peer, __u32 *peerips, int n_peerips)
749{
750 rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
751 ksock_net_t *net = peer->ksnp_ni->ni_data;
752 ksock_interface_t *iface;
753 ksock_interface_t *best_iface;
754 int n_ips;
755 int i;
756 int j;
757 int k;
758 __u32 ip;
759 __u32 xor;
760 int this_netmatch;
761 int best_netmatch;
762 int best_npeers;
763
764 /* CAVEAT EMPTOR: We do all our interface matching with an
765 * exclusive hold of global lock at IRQ priority. We're only
766 * expecting to be dealing with small numbers of interfaces, so the
767 * O(n**3)-ness shouldn't matter */
768
769 /* Also note that I'm not going to return more than n_peerips
770 * interfaces, even if I have more myself */
771
772 write_lock_bh(global_lock);
773
774 LASSERT (n_peerips <= LNET_MAX_INTERFACES);
775 LASSERT (net->ksnn_ninterfaces <= LNET_MAX_INTERFACES);
776
777 /* Only match interfaces for additional connections
778 * if I have > 1 interface */
779 n_ips = (net->ksnn_ninterfaces < 2) ? 0 :
780 MIN(n_peerips, net->ksnn_ninterfaces);
781
782 for (i = 0; peer->ksnp_n_passive_ips < n_ips; i++) {
783 /* ^ yes really... */
784
785 /* If we have any new interfaces, first tick off all the
786 * peer IPs that match old interfaces, then choose new
787 * interfaces to match the remaining peer IPS.
788 * We don't forget interfaces we've stopped using; we might
789 * start using them again... */
790
791 if (i < peer->ksnp_n_passive_ips) {
792 /* Old interface. */
793 ip = peer->ksnp_passive_ips[i];
794 best_iface = ksocknal_ip2iface(peer->ksnp_ni, ip);
795
796 /* peer passive ips are kept up to date */
797 LASSERT(best_iface != NULL);
798 } else {
799 /* choose a new interface */
800 LASSERT (i == peer->ksnp_n_passive_ips);
801
802 best_iface = NULL;
803 best_netmatch = 0;
804 best_npeers = 0;
805
806 for (j = 0; j < net->ksnn_ninterfaces; j++) {
807 iface = &net->ksnn_interfaces[j];
808 ip = iface->ksni_ipaddr;
809
810 for (k = 0; k < peer->ksnp_n_passive_ips; k++)
811 if (peer->ksnp_passive_ips[k] == ip)
812 break;
813
814 if (k < peer->ksnp_n_passive_ips) /* using it already */
815 continue;
816
817 k = ksocknal_match_peerip(iface, peerips, n_peerips);
818 xor = (ip ^ peerips[k]);
819 this_netmatch = ((xor & iface->ksni_netmask) == 0) ? 1 : 0;
820
821 if (!(best_iface == NULL ||
822 best_netmatch < this_netmatch ||
823 (best_netmatch == this_netmatch &&
824 best_npeers > iface->ksni_npeers)))
825 continue;
826
827 best_iface = iface;
828 best_netmatch = this_netmatch;
829 best_npeers = iface->ksni_npeers;
830 }
831
832 best_iface->ksni_npeers++;
833 ip = best_iface->ksni_ipaddr;
834 peer->ksnp_passive_ips[i] = ip;
835 peer->ksnp_n_passive_ips = i+1;
836 }
837
838 LASSERT (best_iface != NULL);
839
840 /* mark the best matching peer IP used */
841 j = ksocknal_match_peerip(best_iface, peerips, n_peerips);
842 peerips[j] = 0;
843 }
844
845 /* Overwrite input peer IP addresses */
846 memcpy(peerips, peer->ksnp_passive_ips, n_ips * sizeof(*peerips));
847
848 write_unlock_bh(global_lock);
849
850 return (n_ips);
851}
852
853void
854ksocknal_create_routes(ksock_peer_t *peer, int port,
855 __u32 *peer_ipaddrs, int npeer_ipaddrs)
856{
857 ksock_route_t *newroute = NULL;
858 rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
859 lnet_ni_t *ni = peer->ksnp_ni;
860 ksock_net_t *net = ni->ni_data;
861 struct list_head *rtmp;
862 ksock_route_t *route;
863 ksock_interface_t *iface;
864 ksock_interface_t *best_iface;
865 int best_netmatch;
866 int this_netmatch;
867 int best_nroutes;
868 int i;
869 int j;
870
871 /* CAVEAT EMPTOR: We do all our interface matching with an
872 * exclusive hold of global lock at IRQ priority. We're only
873 * expecting to be dealing with small numbers of interfaces, so the
874 * O(n**3)-ness here shouldn't matter */
875
876 write_lock_bh(global_lock);
877
878 if (net->ksnn_ninterfaces < 2) {
879 /* Only create additional connections
880 * if I have > 1 interface */
881 write_unlock_bh(global_lock);
882 return;
883 }
884
885 LASSERT (npeer_ipaddrs <= LNET_MAX_INTERFACES);
886
887 for (i = 0; i < npeer_ipaddrs; i++) {
888 if (newroute != NULL) {
889 newroute->ksnr_ipaddr = peer_ipaddrs[i];
890 } else {
891 write_unlock_bh(global_lock);
892
893 newroute = ksocknal_create_route(peer_ipaddrs[i], port);
894 if (newroute == NULL)
895 return;
896
897 write_lock_bh(global_lock);
898 }
899
900 if (peer->ksnp_closing) {
901 /* peer got closed under me */
902 break;
903 }
904
905 /* Already got a route? */
906 route = NULL;
907 list_for_each(rtmp, &peer->ksnp_routes) {
908 route = list_entry(rtmp, ksock_route_t, ksnr_list);
909
910 if (route->ksnr_ipaddr == newroute->ksnr_ipaddr)
911 break;
912
913 route = NULL;
914 }
915 if (route != NULL)
916 continue;
917
918 best_iface = NULL;
919 best_nroutes = 0;
920 best_netmatch = 0;
921
922 LASSERT (net->ksnn_ninterfaces <= LNET_MAX_INTERFACES);
923
924 /* Select interface to connect from */
925 for (j = 0; j < net->ksnn_ninterfaces; j++) {
926 iface = &net->ksnn_interfaces[j];
927
928 /* Using this interface already? */
929 list_for_each(rtmp, &peer->ksnp_routes) {
930 route = list_entry(rtmp, ksock_route_t,
931 ksnr_list);
932
933 if (route->ksnr_myipaddr == iface->ksni_ipaddr)
934 break;
935
936 route = NULL;
937 }
938 if (route != NULL)
939 continue;
940
941 this_netmatch = (((iface->ksni_ipaddr ^
942 newroute->ksnr_ipaddr) &
943 iface->ksni_netmask) == 0) ? 1 : 0;
944
945 if (!(best_iface == NULL ||
946 best_netmatch < this_netmatch ||
947 (best_netmatch == this_netmatch &&
948 best_nroutes > iface->ksni_nroutes)))
949 continue;
950
951 best_iface = iface;
952 best_netmatch = this_netmatch;
953 best_nroutes = iface->ksni_nroutes;
954 }
955
956 if (best_iface == NULL)
957 continue;
958
959 newroute->ksnr_myipaddr = best_iface->ksni_ipaddr;
960 best_iface->ksni_nroutes++;
961
962 ksocknal_add_route_locked(peer, newroute);
963 newroute = NULL;
964 }
965
966 write_unlock_bh(global_lock);
967 if (newroute != NULL)
968 ksocknal_route_decref(newroute);
969}
970
971int
972ksocknal_accept (lnet_ni_t *ni, socket_t *sock)
973{
974 ksock_connreq_t *cr;
975 int rc;
976 __u32 peer_ip;
977 int peer_port;
978
979 rc = libcfs_sock_getaddr(sock, 1, &peer_ip, &peer_port);
980 LASSERT (rc == 0); /* we succeeded before */
981
982 LIBCFS_ALLOC(cr, sizeof(*cr));
983 if (cr == NULL) {
984 LCONSOLE_ERROR_MSG(0x12f, "Dropping connection request from "
5e8f6920
PT		 985 "%pI4h: memory exhausted\n",
986 &peer_ip);
d7e09d03
PT		 987 return -ENOMEM;
988 }
989
990 lnet_ni_addref(ni);
991 cr->ksncr_ni = ni;
992 cr->ksncr_sock = sock;
993
994 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
995
996 list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
997 wake_up(&ksocknal_data.ksnd_connd_waitq);
998
999 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
1000 return 0;
1001}
1002
1003int
1004ksocknal_connecting (ksock_peer_t *peer, __u32 ipaddr)
1005{
1006 ksock_route_t *route;
1007
1008 list_for_each_entry (route, &peer->ksnp_routes, ksnr_list) {
1009
1010 if (route->ksnr_ipaddr == ipaddr)
1011 return route->ksnr_connecting;
1012 }
1013 return 0;
1014}
1015
1016int
1017ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
1018 socket_t *sock, int type)
1019{
1020 rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
1021 LIST_HEAD (zombies);
1022 lnet_process_id_t peerid;
1023 struct list_head *tmp;
1024 __u64 incarnation;
1025 ksock_conn_t *conn;
1026 ksock_conn_t *conn2;
1027 ksock_peer_t *peer = NULL;
1028 ksock_peer_t *peer2;
1029 ksock_sched_t *sched;
1030 ksock_hello_msg_t *hello;
1031 int cpt;
1032 ksock_tx_t *tx;
1033 ksock_tx_t *txtmp;
1034 int rc;
1035 int active;
1036 char *warn = NULL;
1037
1038 active = (route != NULL);
1039
1040 LASSERT (active == (type != SOCKLND_CONN_NONE));
1041
1042 LIBCFS_ALLOC(conn, sizeof(*conn));
1043 if (conn == NULL) {
1044 rc = -ENOMEM;
1045 goto failed_0;
1046 }
1047
1048 memset (conn, 0, sizeof (*conn));
1049
1050 conn->ksnc_peer = NULL;
1051 conn->ksnc_route = NULL;
1052 conn->ksnc_sock = sock;
1053 /* 2 ref, 1 for conn, another extra ref prevents socket
1054 * being closed before establishment of connection */
1055 atomic_set (&conn->ksnc_sock_refcount, 2);
1056 conn->ksnc_type = type;
1057 ksocknal_lib_save_callback(sock, conn);
1058 atomic_set (&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
1059
1060 conn->ksnc_rx_ready = 0;
1061 conn->ksnc_rx_scheduled = 0;
1062
1063 INIT_LIST_HEAD (&conn->ksnc_tx_queue);
1064 conn->ksnc_tx_ready = 0;
1065 conn->ksnc_tx_scheduled = 0;
1066 conn->ksnc_tx_carrier = NULL;
1067 atomic_set (&conn->ksnc_tx_nob, 0);
1068
1069 LIBCFS_ALLOC(hello, offsetof(ksock_hello_msg_t,
1070 kshm_ips[LNET_MAX_INTERFACES]));
1071 if (hello == NULL) {
1072 rc = -ENOMEM;
1073 goto failed_1;
1074 }
1075
1076 /* stash conn's local and remote addrs */
1077 rc = ksocknal_lib_get_conn_addrs (conn);
1078 if (rc != 0)
1079 goto failed_1;
1080
1081 /* Find out/confirm peer's NID and connection type and get the
1082 * vector of interfaces she's willing to let me connect to.
1083 * Passive connections use the listener timeout since the peer sends
1084 * eagerly */
1085
1086 if (active) {
1087 peer = route->ksnr_peer;
1088 LASSERT(ni == peer->ksnp_ni);
1089
1090 /* Active connection sends HELLO eagerly */
1091 hello->kshm_nips = ksocknal_local_ipvec(ni, hello->kshm_ips);
1092 peerid = peer->ksnp_id;
1093
1094 write_lock_bh(global_lock);
1095 conn->ksnc_proto = peer->ksnp_proto;
1096 write_unlock_bh(global_lock);
1097
1098 if (conn->ksnc_proto == NULL) {
1099 conn->ksnc_proto = &ksocknal_protocol_v3x;
1100#if SOCKNAL_VERSION_DEBUG
1101 if (*ksocknal_tunables.ksnd_protocol == 2)
1102 conn->ksnc_proto = &ksocknal_protocol_v2x;
1103 else if (*ksocknal_tunables.ksnd_protocol == 1)
1104 conn->ksnc_proto = &ksocknal_protocol_v1x;
1105#endif
1106 }
1107
1108 rc = ksocknal_send_hello (ni, conn, peerid.nid, hello);
1109 if (rc != 0)
1110 goto failed_1;
1111 } else {
1112 peerid.nid = LNET_NID_ANY;
1113 peerid.pid = LNET_PID_ANY;
1114
1115 /* Passive, get protocol from peer */
1116 conn->ksnc_proto = NULL;
1117 }
1118
1119 rc = ksocknal_recv_hello (ni, conn, hello, &peerid, &incarnation);
1120 if (rc < 0)
1121 goto failed_1;
1122
1123 LASSERT (rc == 0 || active);
1124 LASSERT (conn->ksnc_proto != NULL);
1125 LASSERT (peerid.nid != LNET_NID_ANY);
1126
1127 cpt = lnet_cpt_of_nid(peerid.nid);
1128
1129 if (active) {
1130 ksocknal_peer_addref(peer);
1131 write_lock_bh(global_lock);
1132 } else {
1133 rc = ksocknal_create_peer(&peer, ni, peerid);
1134 if (rc != 0)
1135 goto failed_1;
1136
1137 write_lock_bh(global_lock);
1138
1139 /* called with a ref on ni, so shutdown can't have started */
1140 LASSERT (((ksock_net_t *) ni->ni_data)->ksnn_shutdown == 0);
1141
1142 peer2 = ksocknal_find_peer_locked(ni, peerid);
1143 if (peer2 == NULL) {
1144 /* NB this puts an "empty" peer in the peer
1145 * table (which takes my ref) */
1146 list_add_tail(&peer->ksnp_list,
1147 ksocknal_nid2peerlist(peerid.nid));
1148 } else {
1149 ksocknal_peer_decref(peer);
1150 peer = peer2;
1151 }
1152
1153 /* +1 ref for me */
1154 ksocknal_peer_addref(peer);
1155 peer->ksnp_accepting++;
1156
1157 /* Am I already connecting to this guy? Resolve in
1158 * favour of higher NID... */
1159 if (peerid.nid < ni->ni_nid &&
1160 ksocknal_connecting(peer, conn->ksnc_ipaddr)) {
1161 rc = EALREADY;
1162 warn = "connection race resolution";
1163 goto failed_2;
1164 }
1165 }
1166
1167 if (peer->ksnp_closing ||
1168 (active && route->ksnr_deleted)) {
1169 /* peer/route got closed under me */
1170 rc = -ESTALE;
1171 warn = "peer/route removed";
1172 goto failed_2;
1173 }
1174
1175 if (peer->ksnp_proto == NULL) {
1176 /* Never connected before.
1177 * NB recv_hello may have returned EPROTO to signal my peer
1178 * wants a different protocol than the one I asked for.
1179 */
1180 LASSERT (list_empty(&peer->ksnp_conns));
1181
1182 peer->ksnp_proto = conn->ksnc_proto;
1183 peer->ksnp_incarnation = incarnation;
1184 }
1185
1186 if (peer->ksnp_proto != conn->ksnc_proto ||
1187 peer->ksnp_incarnation != incarnation) {
1188 /* Peer rebooted or I've got the wrong protocol version */
1189 ksocknal_close_peer_conns_locked(peer, 0, 0);
1190
1191 peer->ksnp_proto = NULL;
1192 rc = ESTALE;
1193 warn = peer->ksnp_incarnation != incarnation ?
1194 "peer rebooted" :
1195 "wrong proto version";
1196 goto failed_2;
1197 }
1198
1199 switch (rc) {
1200 default:
1201 LBUG();
1202 case 0:
1203 break;
1204 case EALREADY:
1205 warn = "lost conn race";
1206 goto failed_2;
1207 case EPROTO:
1208 warn = "retry with different protocol version";
1209 goto failed_2;
1210 }
1211
1212 /* Refuse to duplicate an existing connection, unless this is a
1213 * loopback connection */
1214 if (conn->ksnc_ipaddr != conn->ksnc_myipaddr) {
1215 list_for_each(tmp, &peer->ksnp_conns) {
1216 conn2 = list_entry(tmp, ksock_conn_t, ksnc_list);
1217
1218 if (conn2->ksnc_ipaddr != conn->ksnc_ipaddr ||
1219 conn2->ksnc_myipaddr != conn->ksnc_myipaddr ||
1220 conn2->ksnc_type != conn->ksnc_type)
1221 continue;
1222
1223 /* Reply on a passive connection attempt so the peer
1224 * realises we're connected. */
1225 LASSERT (rc == 0);
1226 if (!active)
1227 rc = EALREADY;
1228
1229 warn = "duplicate";
1230 goto failed_2;
1231 }
1232 }
1233
1234 /* If the connection created by this route didn't bind to the IP
1235 * address the route connected to, the connection/route matching
1236 * code below probably isn't going to work. */
1237 if (active &&
1238 route->ksnr_ipaddr != conn->ksnc_ipaddr) {
5e8f6920 1239 CERROR("Route %s %pI4h connected to %pI4h\n",
d7e09d03 1240 libcfs_id2str(peer->ksnp_id),
5e8f6920
PT		 1241 &route->ksnr_ipaddr,
1242 &conn->ksnc_ipaddr);
d7e09d03
PT		 1243 }
1244
1245 /* Search for a route corresponding to the new connection and
1246 * create an association. This allows incoming connections created
1247 * by routes in my peer to match my own route entries so I don't
1248 * continually create duplicate routes. */
1249 list_for_each (tmp, &peer->ksnp_routes) {
1250 route = list_entry(tmp, ksock_route_t, ksnr_list);
1251
1252 if (route->ksnr_ipaddr != conn->ksnc_ipaddr)
1253 continue;
1254
1255 ksocknal_associate_route_conn_locked(route, conn);
1256 break;
1257 }
1258
1259 conn->ksnc_peer = peer; /* conn takes my ref on peer */
1260 peer->ksnp_last_alive = cfs_time_current();
1261 peer->ksnp_send_keepalive = 0;
1262 peer->ksnp_error = 0;
1263
1264 sched = ksocknal_choose_scheduler_locked(cpt);
1265 sched->kss_nconns++;
1266 conn->ksnc_scheduler = sched;
1267
1268 conn->ksnc_tx_last_post = cfs_time_current();
1269 /* Set the deadline for the outgoing HELLO to drain */
1270 conn->ksnc_tx_bufnob = cfs_sock_wmem_queued(sock);
1271 conn->ksnc_tx_deadline = cfs_time_shift(*ksocknal_tunables.ksnd_timeout);
1272 mb(); /* order with adding to peer's conn list */
1273
1274 list_add (&conn->ksnc_list, &peer->ksnp_conns);
1275 ksocknal_conn_addref(conn);
1276
1277 ksocknal_new_packet(conn, 0);
1278
1279 conn->ksnc_zc_capable = ksocknal_lib_zc_capable(conn);
1280
1281 /* Take packets blocking for this connection. */
1282 list_for_each_entry_safe(tx, txtmp, &peer->ksnp_tx_queue, tx_list) {
1283 if (conn->ksnc_proto->pro_match_tx(conn, tx, tx->tx_nonblk) == SOCKNAL_MATCH_NO)
1284 continue;
1285
1286 list_del (&tx->tx_list);
1287 ksocknal_queue_tx_locked (tx, conn);
1288 }
1289
1290 write_unlock_bh(global_lock);
1291
1292 /* We've now got a new connection. Any errors from here on are just
1293 * like "normal" comms errors and we close the connection normally.
1294 * NB (a) we still have to send the reply HELLO for passive
1295 * connections,
1296 * (b) normal I/O on the conn is blocked until I setup and call the
1297 * socket callbacks.
1298 */
1299
5e8f6920 1300 CDEBUG(D_NET, "New conn %s p %d.x %pI4h -> %pI4h/%d"
d7e09d03
PT		 1301 " incarnation:"LPD64" sched[%d:%d]\n",
1302 libcfs_id2str(peerid), conn->ksnc_proto->pro_version,
5e8f6920 1303 &conn->ksnc_myipaddr, &conn->ksnc_ipaddr,
d7e09d03
PT		 1304 conn->ksnc_port, incarnation, cpt,
1305 (int)(sched - &sched->kss_info->ksi_scheds[0]));
1306
1307 if (active) {
1308 /* additional routes after interface exchange? */
1309 ksocknal_create_routes(peer, conn->ksnc_port,
1310 hello->kshm_ips, hello->kshm_nips);
1311 } else {
1312 hello->kshm_nips = ksocknal_select_ips(peer, hello->kshm_ips,
1313 hello->kshm_nips);
1314 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
1315 }
1316
1317 LIBCFS_FREE(hello, offsetof(ksock_hello_msg_t,
1318 kshm_ips[LNET_MAX_INTERFACES]));
1319
1320 /* setup the socket AFTER I've received hello (it disables
1321 * SO_LINGER). I might call back to the acceptor who may want
1322 * to send a protocol version response and then close the
1323 * socket; this ensures the socket only tears down after the
1324 * response has been sent. */
1325 if (rc == 0)
1326 rc = ksocknal_lib_setup_sock(sock);
1327
1328 write_lock_bh(global_lock);
1329
1330 /* NB my callbacks block while I hold ksnd_global_lock */
1331 ksocknal_lib_set_callback(sock, conn);
1332
1333 if (!active)
1334 peer->ksnp_accepting--;
1335
1336 write_unlock_bh(global_lock);
1337
1338 if (rc != 0) {
1339 write_lock_bh(global_lock);
1340 if (!conn->ksnc_closing) {
1341 /* could be closed by another thread */
1342 ksocknal_close_conn_locked(conn, rc);
1343 }
1344 write_unlock_bh(global_lock);
1345 } else if (ksocknal_connsock_addref(conn) == 0) {
1346 /* Allow I/O to proceed. */
1347 ksocknal_read_callback(conn);
1348 ksocknal_write_callback(conn);
1349 ksocknal_connsock_decref(conn);
1350 }
1351
1352 ksocknal_connsock_decref(conn);
1353 ksocknal_conn_decref(conn);
1354 return rc;
1355
1356 failed_2:
1357 if (!peer->ksnp_closing &&
1358 list_empty (&peer->ksnp_conns) &&
1359 list_empty (&peer->ksnp_routes)) {
1360 list_add(&zombies, &peer->ksnp_tx_queue);
1361 list_del_init(&peer->ksnp_tx_queue);
1362 ksocknal_unlink_peer_locked(peer);
1363 }
1364
1365 write_unlock_bh(global_lock);
1366
1367 if (warn != NULL) {
1368 if (rc < 0)
1369 CERROR("Not creating conn %s type %d: %s\n",
1370 libcfs_id2str(peerid), conn->ksnc_type, warn);
1371 else
1372 CDEBUG(D_NET, "Not creating conn %s type %d: %s\n",
1373 libcfs_id2str(peerid), conn->ksnc_type, warn);
1374 }
1375
1376 if (!active) {
1377 if (rc > 0) {
1378 /* Request retry by replying with CONN_NONE
1379 * ksnc_proto has been set already */
1380 conn->ksnc_type = SOCKLND_CONN_NONE;
1381 hello->kshm_nips = 0;
1382 ksocknal_send_hello(ni, conn, peerid.nid, hello);
1383 }
1384
1385 write_lock_bh(global_lock);
1386 peer->ksnp_accepting--;
1387 write_unlock_bh(global_lock);
1388 }
1389
1390 ksocknal_txlist_done(ni, &zombies, 1);
1391 ksocknal_peer_decref(peer);
1392
1393 failed_1:
1394 if (hello != NULL)
1395 LIBCFS_FREE(hello, offsetof(ksock_hello_msg_t,
1396 kshm_ips[LNET_MAX_INTERFACES]));
1397
1398 LIBCFS_FREE (conn, sizeof(*conn));
1399
1400 failed_0:
1401 libcfs_sock_release(sock);
1402 return rc;
1403}
1404
1405void
1406ksocknal_close_conn_locked (ksock_conn_t *conn, int error)
1407{
1408 /* This just does the immmediate housekeeping, and queues the
1409 * connection for the reaper to terminate.
1410 * Caller holds ksnd_global_lock exclusively in irq context */
1411 ksock_peer_t *peer = conn->ksnc_peer;
1412 ksock_route_t *route;
1413 ksock_conn_t *conn2;
1414 struct list_head *tmp;
1415
1416 LASSERT (peer->ksnp_error == 0);
1417 LASSERT (!conn->ksnc_closing);
1418 conn->ksnc_closing = 1;
1419
1420 /* ksnd_deathrow_conns takes over peer's ref */
1421 list_del (&conn->ksnc_list);
1422
1423 route = conn->ksnc_route;
1424 if (route != NULL) {
1425 /* dissociate conn from route... */
1426 LASSERT (!route->ksnr_deleted);
1427 LASSERT ((route->ksnr_connected & (1 << conn->ksnc_type)) != 0);
1428
1429 conn2 = NULL;
1430 list_for_each(tmp, &peer->ksnp_conns) {
1431 conn2 = list_entry(tmp, ksock_conn_t, ksnc_list);
1432
1433 if (conn2->ksnc_route == route &&
1434 conn2->ksnc_type == conn->ksnc_type)
1435 break;
1436
1437 conn2 = NULL;
1438 }
1439 if (conn2 == NULL)
1440 route->ksnr_connected &= ~(1 << conn->ksnc_type);
1441
1442 conn->ksnc_route = NULL;
1443
1444#if 0 /* irrelevent with only eager routes */
1445 /* make route least favourite */
1446 list_del (&route->ksnr_list);
1447 list_add_tail (&route->ksnr_list, &peer->ksnp_routes);
1448#endif
1449 ksocknal_route_decref(route); /* drop conn's ref on route */
1450 }
1451
1452 if (list_empty (&peer->ksnp_conns)) {
1453 /* No more connections to this peer */
1454
1455 if (!list_empty(&peer->ksnp_tx_queue)) {
1456 ksock_tx_t *tx;
1457
1458 LASSERT (conn->ksnc_proto == &ksocknal_protocol_v3x);
1459
1460 /* throw them to the last connection...,
1461 * these TXs will be send to /dev/null by scheduler */
1462 list_for_each_entry(tx, &peer->ksnp_tx_queue,
1463 tx_list)
1464 ksocknal_tx_prep(conn, tx);
1465
1466 spin_lock_bh(&conn->ksnc_scheduler->kss_lock);
1467 list_splice_init(&peer->ksnp_tx_queue,
1468 &conn->ksnc_tx_queue);
1469 spin_unlock_bh(&conn->ksnc_scheduler->kss_lock);
1470 }
1471
1472 peer->ksnp_proto = NULL; /* renegotiate protocol version */
1473 peer->ksnp_error = error; /* stash last conn close reason */
1474
1475 if (list_empty (&peer->ksnp_routes)) {
1476 /* I've just closed last conn belonging to a
1477 * peer with no routes to it */
1478 ksocknal_unlink_peer_locked (peer);
1479 }
1480 }
1481
1482 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1483
1484 list_add_tail(&conn->ksnc_list,
1485 &ksocknal_data.ksnd_deathrow_conns);
1486 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1487
1488 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1489}
1490
1491void
1492ksocknal_peer_failed (ksock_peer_t *peer)
1493{
1494 int notify = 0;
1495 cfs_time_t last_alive = 0;
1496
1497 /* There has been a connection failure or comms error; but I'll only
1498 * tell LNET I think the peer is dead if it's to another kernel and
1499 * there are no connections or connection attempts in existance. */
1500
1501 read_lock(&ksocknal_data.ksnd_global_lock);
1502
1503 if ((peer->ksnp_id.pid & LNET_PID_USERFLAG) == 0 &&
1504 list_empty(&peer->ksnp_conns) &&
1505 peer->ksnp_accepting == 0 &&
1506 ksocknal_find_connecting_route_locked(peer) == NULL) {
1507 notify = 1;
1508 last_alive = peer->ksnp_last_alive;
1509 }
1510
1511 read_unlock(&ksocknal_data.ksnd_global_lock);
1512
1513 if (notify)
1514 lnet_notify (peer->ksnp_ni, peer->ksnp_id.nid, 0,
1515 last_alive);
1516}
1517
1518void
1519ksocknal_finalize_zcreq(ksock_conn_t *conn)
1520{
1521 ksock_peer_t *peer = conn->ksnc_peer;
1522 ksock_tx_t *tx;
1523 ksock_tx_t *tmp;
1524 LIST_HEAD (zlist);
1525
1526 /* NB safe to finalize TXs because closing of socket will
1527 * abort all buffered data */
1528 LASSERT (conn->ksnc_sock == NULL);
1529
1530 spin_lock(&peer->ksnp_lock);
1531
1532 list_for_each_entry_safe(tx, tmp, &peer->ksnp_zc_req_list, tx_zc_list) {
1533 if (tx->tx_conn != conn)
1534 continue;
1535
1536 LASSERT (tx->tx_msg.ksm_zc_cookies[0] != 0);
1537
1538 tx->tx_msg.ksm_zc_cookies[0] = 0;
1539 tx->tx_zc_aborted = 1; /* mark it as not-acked */
1540 list_del(&tx->tx_zc_list);
1541 list_add(&tx->tx_zc_list, &zlist);
1542 }
1543
1544 spin_unlock(&peer->ksnp_lock);
1545
1546 while (!list_empty(&zlist)) {
1547 tx = list_entry(zlist.next, ksock_tx_t, tx_zc_list);
1548
1549 list_del(&tx->tx_zc_list);
1550 ksocknal_tx_decref(tx);
1551 }
1552}
1553
1554void
1555ksocknal_terminate_conn (ksock_conn_t *conn)
1556{
1557 /* This gets called by the reaper (guaranteed thread context) to
1558 * disengage the socket from its callbacks and close it.
1559 * ksnc_refcount will eventually hit zero, and then the reaper will
1560 * destroy it. */
1561 ksock_peer_t *peer = conn->ksnc_peer;
1562 ksock_sched_t *sched = conn->ksnc_scheduler;
1563 int failed = 0;
1564
1565 LASSERT(conn->ksnc_closing);
1566
1567 /* wake up the scheduler to "send" all remaining packets to /dev/null */
1568 spin_lock_bh(&sched->kss_lock);
1569
1570 /* a closing conn is always ready to tx */
1571 conn->ksnc_tx_ready = 1;
1572
1573 if (!conn->ksnc_tx_scheduled &&
1574 !list_empty(&conn->ksnc_tx_queue)){
1575 list_add_tail (&conn->ksnc_tx_list,
1576 &sched->kss_tx_conns);
1577 conn->ksnc_tx_scheduled = 1;
1578 /* extra ref for scheduler */
1579 ksocknal_conn_addref(conn);
1580
1581 wake_up (&sched->kss_waitq);
1582 }
1583
1584 spin_unlock_bh(&sched->kss_lock);
1585
1586 /* serialise with callbacks */
1587 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1588
1589 ksocknal_lib_reset_callback(conn->ksnc_sock, conn);
1590
1591 /* OK, so this conn may not be completely disengaged from its
1592 * scheduler yet, but it _has_ committed to terminate... */
1593 conn->ksnc_scheduler->kss_nconns--;
1594
1595 if (peer->ksnp_error != 0) {
1596 /* peer's last conn closed in error */
1597 LASSERT (list_empty (&peer->ksnp_conns));
1598 failed = 1;
1599 peer->ksnp_error = 0; /* avoid multiple notifications */
1600 }
1601
1602 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1603
1604 if (failed)
1605 ksocknal_peer_failed(peer);
1606
1607 /* The socket is closed on the final put; either here, or in
1608 * ksocknal_{send,recv}msg(). Since we set up the linger2 option
1609 * when the connection was established, this will close the socket
1610 * immediately, aborting anything buffered in it. Any hung
1611 * zero-copy transmits will therefore complete in finite time. */
1612 ksocknal_connsock_decref(conn);
1613}
1614
1615void
1616ksocknal_queue_zombie_conn (ksock_conn_t *conn)
1617{
1618 /* Queue the conn for the reaper to destroy */
1619
1620 LASSERT(atomic_read(&conn->ksnc_conn_refcount) == 0);
1621 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1622
1623 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
1624 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1625
1626 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1627}
1628
1629void
1630ksocknal_destroy_conn (ksock_conn_t *conn)
1631{
1632 cfs_time_t last_rcv;
1633
1634 /* Final coup-de-grace of the reaper */
1635 CDEBUG (D_NET, "connection %p\n", conn);
1636
1637 LASSERT (atomic_read (&conn->ksnc_conn_refcount) == 0);
1638 LASSERT (atomic_read (&conn->ksnc_sock_refcount) == 0);
1639 LASSERT (conn->ksnc_sock == NULL);
1640 LASSERT (conn->ksnc_route == NULL);
1641 LASSERT (!conn->ksnc_tx_scheduled);
1642 LASSERT (!conn->ksnc_rx_scheduled);
1643 LASSERT (list_empty(&conn->ksnc_tx_queue));
1644
1645 /* complete current receive if any */
1646 switch (conn->ksnc_rx_state) {
1647 case SOCKNAL_RX_LNET_PAYLOAD:
1648 last_rcv = conn->ksnc_rx_deadline -
1649 cfs_time_seconds(*ksocknal_tunables.ksnd_timeout);
1650 CERROR("Completing partial receive from %s[%d]"
5e8f6920 1651 ", ip %pI4h:%d, with error, wanted: %d, left: %d, "
d7e09d03
PT		 1652 "last alive is %ld secs ago\n",
1653 libcfs_id2str(conn->ksnc_peer->ksnp_id), conn->ksnc_type,
5e8f6920 1654 &conn->ksnc_ipaddr, conn->ksnc_port,
d7e09d03
PT		 1655 conn->ksnc_rx_nob_wanted, conn->ksnc_rx_nob_left,
1656 cfs_duration_sec(cfs_time_sub(cfs_time_current(),
1657 last_rcv)));
1658 lnet_finalize (conn->ksnc_peer->ksnp_ni,
1659 conn->ksnc_cookie, -EIO);
1660 break;
1661 case SOCKNAL_RX_LNET_HEADER:
1662 if (conn->ksnc_rx_started)
1663 CERROR("Incomplete receive of lnet header from %s"
5e8f6920 1664 ", ip %pI4h:%d, with error, protocol: %d.x.\n",
d7e09d03 1665 libcfs_id2str(conn->ksnc_peer->ksnp_id),
5e8f6920 1666 &conn->ksnc_ipaddr, conn->ksnc_port,
d7e09d03
PT		 1667 conn->ksnc_proto->pro_version);
1668 break;
1669 case SOCKNAL_RX_KSM_HEADER:
1670 if (conn->ksnc_rx_started)
1671 CERROR("Incomplete receive of ksock message from %s"
5e8f6920 1672 ", ip %pI4h:%d, with error, protocol: %d.x.\n",
d7e09d03 1673 libcfs_id2str(conn->ksnc_peer->ksnp_id),
5e8f6920 1674 &conn->ksnc_ipaddr, conn->ksnc_port,
d7e09d03
PT		 1675 conn->ksnc_proto->pro_version);
1676 break;
1677 case SOCKNAL_RX_SLOP:
1678 if (conn->ksnc_rx_started)
1679 CERROR("Incomplete receive of slops from %s"
5e8f6920 1680 ", ip %pI4h:%d, with error\n",
d7e09d03 1681 libcfs_id2str(conn->ksnc_peer->ksnp_id),
5e8f6920 1682 &conn->ksnc_ipaddr, conn->ksnc_port);
d7e09d03
PT		 1683 break;
1684 default:
1685 LBUG ();
1686 break;
1687 }
1688
1689 ksocknal_peer_decref(conn->ksnc_peer);
1690
1691 LIBCFS_FREE (conn, sizeof (*conn));
1692}
1693
1694int
1695ksocknal_close_peer_conns_locked (ksock_peer_t *peer, __u32 ipaddr, int why)
1696{
1697 ksock_conn_t *conn;
1698 struct list_head *ctmp;
1699 struct list_head *cnxt;
1700 int count = 0;
1701
1702 list_for_each_safe (ctmp, cnxt, &peer->ksnp_conns) {
1703 conn = list_entry (ctmp, ksock_conn_t, ksnc_list);
1704
1705 if (ipaddr == 0 ||
1706 conn->ksnc_ipaddr == ipaddr) {
1707 count++;
1708 ksocknal_close_conn_locked (conn, why);
1709 }
1710 }
1711
1712 return (count);
1713}
1714
1715int
1716ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why)
1717{
1718 ksock_peer_t *peer = conn->ksnc_peer;
1719 __u32 ipaddr = conn->ksnc_ipaddr;
1720 int count;
1721
1722 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1723
1724 count = ksocknal_close_peer_conns_locked (peer, ipaddr, why);
1725
1726 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1727
1728 return (count);
1729}
1730
1731int
1732ksocknal_close_matching_conns (lnet_process_id_t id, __u32 ipaddr)
1733{
1734 ksock_peer_t *peer;
1735 struct list_head *ptmp;
1736 struct list_head *pnxt;
1737 int lo;
1738 int hi;
1739 int i;
1740 int count = 0;
1741
1742 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1743
1744 if (id.nid != LNET_NID_ANY)
1745 lo = hi = (int)(ksocknal_nid2peerlist(id.nid) - ksocknal_data.ksnd_peers);
1746 else {
1747 lo = 0;
1748 hi = ksocknal_data.ksnd_peer_hash_size - 1;
1749 }
1750
1751 for (i = lo; i <= hi; i++) {
1752 list_for_each_safe (ptmp, pnxt,
1753 &ksocknal_data.ksnd_peers[i]) {
1754
1755 peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
1756
1757 if (!((id.nid == LNET_NID_ANY || id.nid == peer->ksnp_id.nid) &&
1758 (id.pid == LNET_PID_ANY || id.pid == peer->ksnp_id.pid)))
1759 continue;
1760
1761 count += ksocknal_close_peer_conns_locked (peer, ipaddr, 0);
1762 }
1763 }
1764
1765 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1766
1767 /* wildcards always succeed */
1768 if (id.nid == LNET_NID_ANY || id.pid == LNET_PID_ANY || ipaddr == 0)
1769 return (0);
1770
1771 return (count == 0 ? -ENOENT : 0);
1772}
1773
1774void
1775ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive)
1776{
1777 /* The router is telling me she's been notified of a change in
1778 * gateway state.... */
1779 lnet_process_id_t id = {0};
1780
1781 id.nid = gw_nid;
1782 id.pid = LNET_PID_ANY;
1783
1784 CDEBUG (D_NET, "gw %s %s\n", libcfs_nid2str(gw_nid),
1785 alive ? "up" : "down");
1786
1787 if (!alive) {
1788 /* If the gateway crashed, close all open connections... */
1789 ksocknal_close_matching_conns (id, 0);
1790 return;
1791 }
1792
1793 /* ...otherwise do nothing. We can only establish new connections
1794 * if we have autroutes, and these connect on demand. */
1795}
1796
1797void
1798ksocknal_query (lnet_ni_t *ni, lnet_nid_t nid, cfs_time_t *when)
1799{
1800 int connect = 1;
1801 cfs_time_t last_alive = 0;
1802 cfs_time_t now = cfs_time_current();
1803 ksock_peer_t *peer = NULL;
1804 rwlock_t *glock = &ksocknal_data.ksnd_global_lock;
1805 lnet_process_id_t id = {.nid = nid, .pid = LUSTRE_SRV_LNET_PID};
1806
1807 read_lock(glock);
1808
1809 peer = ksocknal_find_peer_locked(ni, id);
1810 if (peer != NULL) {
1811 struct list_head *tmp;
1812 ksock_conn_t *conn;
1813 int bufnob;
1814
1815 list_for_each (tmp, &peer->ksnp_conns) {
1816 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
1817 bufnob = cfs_sock_wmem_queued(conn->ksnc_sock);
1818
1819 if (bufnob < conn->ksnc_tx_bufnob) {
1820 /* something got ACKed */
1821 conn->ksnc_tx_deadline =
1822 cfs_time_shift(*ksocknal_tunables.ksnd_timeout);
1823 peer->ksnp_last_alive = now;
1824 conn->ksnc_tx_bufnob = bufnob;
1825 }
1826 }
1827
1828 last_alive = peer->ksnp_last_alive;
1829 if (ksocknal_find_connectable_route_locked(peer) == NULL)
1830 connect = 0;
1831 }
1832
1833 read_unlock(glock);
1834
1835 if (last_alive != 0)
1836 *when = last_alive;
1837
1838 CDEBUG(D_NET, "Peer %s %p, alive %ld secs ago, connect %d\n",
1839 libcfs_nid2str(nid), peer,
1840 last_alive ? cfs_duration_sec(now - last_alive) : -1,
1841 connect);
1842
1843 if (!connect)
1844 return;
1845
1846 ksocknal_add_peer(ni, id, LNET_NIDADDR(nid), lnet_acceptor_port());
1847
1848 write_lock_bh(glock);
1849
1850 peer = ksocknal_find_peer_locked(ni, id);
1851 if (peer != NULL)
1852 ksocknal_launch_all_connections_locked(peer);
1853
1854 write_unlock_bh(glock);
1855 return;
1856}
1857
1858void
1859ksocknal_push_peer (ksock_peer_t *peer)
1860{
1861 int index;
1862 int i;
1863 struct list_head *tmp;
1864 ksock_conn_t *conn;
1865
1866 for (index = 0; ; index++) {
1867 read_lock(&ksocknal_data.ksnd_global_lock);
1868
1869 i = 0;
1870 conn = NULL;
1871
1872 list_for_each (tmp, &peer->ksnp_conns) {
1873 if (i++ == index) {
1874 conn = list_entry (tmp, ksock_conn_t,
1875 ksnc_list);
1876 ksocknal_conn_addref(conn);
1877 break;
1878 }
1879 }
1880
1881 read_unlock(&ksocknal_data.ksnd_global_lock);
1882
1883 if (conn == NULL)
1884 break;
1885
1886 ksocknal_lib_push_conn (conn);
1887 ksocknal_conn_decref(conn);
1888 }
1889}
1890
1891int
1892ksocknal_push (lnet_ni_t *ni, lnet_process_id_t id)
1893{
1894 ksock_peer_t *peer;
1895 struct list_head *tmp;
1896 int index;
1897 int i;
1898 int j;
1899 int rc = -ENOENT;
1900
1901 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
1902 for (j = 0; ; j++) {
1903 read_lock(&ksocknal_data.ksnd_global_lock);
1904
1905 index = 0;
1906 peer = NULL;
1907
1908 list_for_each (tmp, &ksocknal_data.ksnd_peers[i]) {
1909 peer = list_entry(tmp, ksock_peer_t,
1910 ksnp_list);
1911
1912 if (!((id.nid == LNET_NID_ANY ||
1913 id.nid == peer->ksnp_id.nid) &&
1914 (id.pid == LNET_PID_ANY ||
1915 id.pid == peer->ksnp_id.pid))) {
1916 peer = NULL;
1917 continue;
1918 }
1919
1920 if (index++ == j) {
1921 ksocknal_peer_addref(peer);
1922 break;
1923 }
1924 }
1925
1926 read_unlock(&ksocknal_data.ksnd_global_lock);
1927
1928 if (peer != NULL) {
1929 rc = 0;
1930 ksocknal_push_peer (peer);
1931 ksocknal_peer_decref(peer);
1932 }
1933 }
1934
1935 }
1936
1937 return (rc);
1938}
1939
1940int
1941ksocknal_add_interface(lnet_ni_t *ni, __u32 ipaddress, __u32 netmask)
1942{
1943 ksock_net_t *net = ni->ni_data;
1944 ksock_interface_t *iface;
1945 int rc;
1946 int i;
1947 int j;
1948 struct list_head *ptmp;
1949 ksock_peer_t *peer;
1950 struct list_head *rtmp;
1951 ksock_route_t *route;
1952
1953 if (ipaddress == 0 ||
1954 netmask == 0)
1955 return (-EINVAL);
1956
1957 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1958
1959 iface = ksocknal_ip2iface(ni, ipaddress);
1960 if (iface != NULL) {
1961 /* silently ignore dups */
1962 rc = 0;
1963 } else if (net->ksnn_ninterfaces == LNET_MAX_INTERFACES) {
1964 rc = -ENOSPC;
1965 } else {
1966 iface = &net->ksnn_interfaces[net->ksnn_ninterfaces++];
1967
1968 iface->ksni_ipaddr = ipaddress;
1969 iface->ksni_netmask = netmask;
1970 iface->ksni_nroutes = 0;
1971 iface->ksni_npeers = 0;
1972
1973 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
1974 list_for_each(ptmp, &ksocknal_data.ksnd_peers[i]) {
1975 peer = list_entry(ptmp, ksock_peer_t,
1976 ksnp_list);
1977
1978 for (j = 0; j < peer->ksnp_n_passive_ips; j++)
1979 if (peer->ksnp_passive_ips[j] == ipaddress)
1980 iface->ksni_npeers++;
1981
1982 list_for_each(rtmp, &peer->ksnp_routes) {
1983 route = list_entry(rtmp,
1984 ksock_route_t,
1985 ksnr_list);
1986
1987 if (route->ksnr_myipaddr == ipaddress)
1988 iface->ksni_nroutes++;
1989 }
1990 }
1991 }
1992
1993 rc = 0;
1994 /* NB only new connections will pay attention to the new interface! */
1995 }
1996
1997 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1998
1999 return (rc);
2000}
2001
2002void
2003ksocknal_peer_del_interface_locked(ksock_peer_t *peer, __u32 ipaddr)
2004{
2005 struct list_head *tmp;
2006 struct list_head *nxt;
2007 ksock_route_t *route;
2008 ksock_conn_t *conn;
2009 int i;
2010 int j;
2011
2012 for (i = 0; i < peer->ksnp_n_passive_ips; i++)
2013 if (peer->ksnp_passive_ips[i] == ipaddr) {
2014 for (j = i+1; j < peer->ksnp_n_passive_ips; j++)
2015 peer->ksnp_passive_ips[j-1] =
2016 peer->ksnp_passive_ips[j];
2017 peer->ksnp_n_passive_ips--;
2018 break;
2019 }
2020
2021 list_for_each_safe(tmp, nxt, &peer->ksnp_routes) {
2022 route = list_entry (tmp, ksock_route_t, ksnr_list);
2023
2024 if (route->ksnr_myipaddr != ipaddr)
2025 continue;
2026
2027 if (route->ksnr_share_count != 0) {
2028 /* Manually created; keep, but unbind */
2029 route->ksnr_myipaddr = 0;
2030 } else {
2031 ksocknal_del_route_locked(route);
2032 }
2033 }
2034
2035 list_for_each_safe(tmp, nxt, &peer->ksnp_conns) {
2036 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
2037
2038 if (conn->ksnc_myipaddr == ipaddr)
2039 ksocknal_close_conn_locked (conn, 0);
2040 }
2041}
2042
2043int
2044ksocknal_del_interface(lnet_ni_t *ni, __u32 ipaddress)
2045{
2046 ksock_net_t *net = ni->ni_data;
2047 int rc = -ENOENT;
2048 struct list_head *tmp;
2049 struct list_head *nxt;
2050 ksock_peer_t *peer;
2051 __u32 this_ip;
2052 int i;
2053 int j;
2054
2055 write_lock_bh(&ksocknal_data.ksnd_global_lock);
2056
2057 for (i = 0; i < net->ksnn_ninterfaces; i++) {
2058 this_ip = net->ksnn_interfaces[i].ksni_ipaddr;
2059
2060 if (!(ipaddress == 0 ||
2061 ipaddress == this_ip))
2062 continue;
2063
2064 rc = 0;
2065
2066 for (j = i+1; j < net->ksnn_ninterfaces; j++)
2067 net->ksnn_interfaces[j-1] =
2068 net->ksnn_interfaces[j];
2069
2070 net->ksnn_ninterfaces--;
2071
2072 for (j = 0; j < ksocknal_data.ksnd_peer_hash_size; j++) {
2073 list_for_each_safe(tmp, nxt,
2074 &ksocknal_data.ksnd_peers[j]) {
2075 peer = list_entry(tmp, ksock_peer_t,
2076 ksnp_list);
2077
2078 if (peer->ksnp_ni != ni)
2079 continue;
2080
2081 ksocknal_peer_del_interface_locked(peer, this_ip);
2082 }
2083 }
2084 }
2085
2086 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
2087
2088 return (rc);
2089}
2090
2091int
2092ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
2093{
2094 lnet_process_id_t id = {0};
2095 struct libcfs_ioctl_data *data = arg;
2096 int rc;
2097
2098 switch(cmd) {
2099 case IOC_LIBCFS_GET_INTERFACE: {
2100 ksock_net_t *net = ni->ni_data;
2101 ksock_interface_t *iface;
2102
2103 read_lock(&ksocknal_data.ksnd_global_lock);
2104
2105 if (data->ioc_count >= (__u32)net->ksnn_ninterfaces) {
2106 rc = -ENOENT;
2107 } else {
2108 rc = 0;
2109 iface = &net->ksnn_interfaces[data->ioc_count];
2110
2111 data->ioc_u32[0] = iface->ksni_ipaddr;
2112 data->ioc_u32[1] = iface->ksni_netmask;
2113 data->ioc_u32[2] = iface->ksni_npeers;
2114 data->ioc_u32[3] = iface->ksni_nroutes;
2115 }
2116
2117 read_unlock(&ksocknal_data.ksnd_global_lock);
2118 return rc;
2119 }
2120
2121 case IOC_LIBCFS_ADD_INTERFACE:
2122 return ksocknal_add_interface(ni,
2123 data->ioc_u32[0], /* IP address */
2124 data->ioc_u32[1]); /* net mask */
2125
2126 case IOC_LIBCFS_DEL_INTERFACE:
2127 return ksocknal_del_interface(ni,
2128 data->ioc_u32[0]); /* IP address */
2129
2130 case IOC_LIBCFS_GET_PEER: {
2131 __u32 myip = 0;
2132 __u32 ip = 0;
2133 int port = 0;
2134 int conn_count = 0;
2135 int share_count = 0;
2136
2137 rc = ksocknal_get_peer_info(ni, data->ioc_count,
2138 &id, &myip, &ip, &port,
2139 &conn_count, &share_count);
2140 if (rc != 0)
2141 return rc;
2142
2143 data->ioc_nid = id.nid;
2144 data->ioc_count = share_count;
2145 data->ioc_u32[0] = ip;
2146 data->ioc_u32[1] = port;
2147 data->ioc_u32[2] = myip;
2148 data->ioc_u32[3] = conn_count;
2149 data->ioc_u32[4] = id.pid;
2150 return 0;
2151 }
2152
2153 case IOC_LIBCFS_ADD_PEER:
2154 id.nid = data->ioc_nid;
2155 id.pid = LUSTRE_SRV_LNET_PID;
2156 return ksocknal_add_peer (ni, id,
2157 data->ioc_u32[0], /* IP */
2158 data->ioc_u32[1]); /* port */
2159
2160 case IOC_LIBCFS_DEL_PEER:
2161 id.nid = data->ioc_nid;
2162 id.pid = LNET_PID_ANY;
2163 return ksocknal_del_peer (ni, id,
2164 data->ioc_u32[0]); /* IP */
2165
2166 case IOC_LIBCFS_GET_CONN: {
2167 int txmem;
2168 int rxmem;
2169 int nagle;
2170 ksock_conn_t *conn = ksocknal_get_conn_by_idx (ni, data->ioc_count);
2171
2172 if (conn == NULL)
2173 return -ENOENT;
2174
2175 ksocknal_lib_get_conn_tunables(conn, &txmem, &rxmem, &nagle);
2176
2177 data->ioc_count = txmem;
2178 data->ioc_nid = conn->ksnc_peer->ksnp_id.nid;
2179 data->ioc_flags = nagle;
2180 data->ioc_u32[0] = conn->ksnc_ipaddr;
2181 data->ioc_u32[1] = conn->ksnc_port;
2182 data->ioc_u32[2] = conn->ksnc_myipaddr;
2183 data->ioc_u32[3] = conn->ksnc_type;
2184 data->ioc_u32[4] = conn->ksnc_scheduler->kss_info->ksi_cpt;
2185 data->ioc_u32[5] = rxmem;
2186 data->ioc_u32[6] = conn->ksnc_peer->ksnp_id.pid;
2187 ksocknal_conn_decref(conn);
2188 return 0;
2189 }
2190
2191 case IOC_LIBCFS_CLOSE_CONNECTION:
2192 id.nid = data->ioc_nid;
2193 id.pid = LNET_PID_ANY;
2194 return ksocknal_close_matching_conns (id,
2195 data->ioc_u32[0]);
2196
2197 case IOC_LIBCFS_REGISTER_MYNID:
2198 /* Ignore if this is a noop */
2199 if (data->ioc_nid == ni->ni_nid)
2200 return 0;
2201
2202 CERROR("obsolete IOC_LIBCFS_REGISTER_MYNID: %s(%s)\n",
2203 libcfs_nid2str(data->ioc_nid),
2204 libcfs_nid2str(ni->ni_nid));
2205 return -EINVAL;
2206
2207 case IOC_LIBCFS_PUSH_CONNECTION:
2208 id.nid = data->ioc_nid;
2209 id.pid = LNET_PID_ANY;
2210 return ksocknal_push(ni, id);
2211
2212 default:
2213 return -EINVAL;
2214 }
2215 /* not reached */
2216}
2217
2218void
2219ksocknal_free_buffers (void)
2220{
2221 LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
2222
2223 if (ksocknal_data.ksnd_sched_info != NULL) {
2224 struct ksock_sched_info *info;
2225 int i;
2226
2227 cfs_percpt_for_each(info, i, ksocknal_data.ksnd_sched_info) {
2228 if (info->ksi_scheds != NULL) {
2229 LIBCFS_FREE(info->ksi_scheds,
2230 info->ksi_nthreads_max *
2231 sizeof(info->ksi_scheds[0]));
2232 }
2233 }
2234 cfs_percpt_free(ksocknal_data.ksnd_sched_info);
2235 }
2236
2237 LIBCFS_FREE (ksocknal_data.ksnd_peers,
2238 sizeof (struct list_head) *
2239 ksocknal_data.ksnd_peer_hash_size);
2240
2241 spin_lock(&ksocknal_data.ksnd_tx_lock);
2242
2243 if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
2244 struct list_head zlist;
2245 ksock_tx_t *tx;
2246
2247 list_add(&zlist, &ksocknal_data.ksnd_idle_noop_txs);
2248 list_del_init(&ksocknal_data.ksnd_idle_noop_txs);
2249 spin_unlock(&ksocknal_data.ksnd_tx_lock);
2250
2251 while (!list_empty(&zlist)) {
2252 tx = list_entry(zlist.next, ksock_tx_t, tx_list);
2253 list_del(&tx->tx_list);
2254 LIBCFS_FREE(tx, tx->tx_desc_size);
2255 }
2256 } else {
2257 spin_unlock(&ksocknal_data.ksnd_tx_lock);
2258 }
2259}
2260
2261void
2262ksocknal_base_shutdown(void)
2263{
2264 struct ksock_sched_info *info;
2265 ksock_sched_t *sched;
2266 int i;
2267 int j;
2268
2269 CDEBUG(D_MALLOC, "before NAL cleanup: kmem %d\n",
2270 atomic_read (&libcfs_kmemory));
2271 LASSERT (ksocknal_data.ksnd_nnets == 0);
2272
2273 switch (ksocknal_data.ksnd_init) {
2274 default:
2275 LASSERT (0);
2276
2277 case SOCKNAL_INIT_ALL:
2278 case SOCKNAL_INIT_DATA:
2279 LASSERT (ksocknal_data.ksnd_peers != NULL);
2280 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
2281 LASSERT (list_empty (&ksocknal_data.ksnd_peers[i]));
2282 }
2283
2284 LASSERT(list_empty(&ksocknal_data.ksnd_nets));
2285 LASSERT (list_empty (&ksocknal_data.ksnd_enomem_conns));
2286 LASSERT (list_empty (&ksocknal_data.ksnd_zombie_conns));
2287 LASSERT (list_empty (&ksocknal_data.ksnd_connd_connreqs));
2288 LASSERT (list_empty (&ksocknal_data.ksnd_connd_routes));
2289
2290 if (ksocknal_data.ksnd_sched_info != NULL) {
2291 cfs_percpt_for_each(info, i,
2292 ksocknal_data.ksnd_sched_info) {
2293 if (info->ksi_scheds == NULL)
2294 continue;
2295
2296 for (j = 0; j < info->ksi_nthreads_max; j++) {
2297
2298 sched = &info->ksi_scheds[j];
2299 LASSERT(list_empty(&sched->\
2300 kss_tx_conns));
2301 LASSERT(list_empty(&sched->\
2302 kss_rx_conns));
2303 LASSERT(list_empty(&sched-> \
2304 kss_zombie_noop_txs));
2305 LASSERT(sched->kss_nconns == 0);
2306 }
2307 }
2308 }
2309
2310 /* flag threads to terminate; wake and wait for them to die */
2311 ksocknal_data.ksnd_shuttingdown = 1;
2312 wake_up_all(&ksocknal_data.ksnd_connd_waitq);
2313 wake_up_all(&ksocknal_data.ksnd_reaper_waitq);
2314
2315 if (ksocknal_data.ksnd_sched_info != NULL) {
2316 cfs_percpt_for_each(info, i,
2317 ksocknal_data.ksnd_sched_info) {
2318 if (info->ksi_scheds == NULL)
2319 continue;
2320
2321 for (j = 0; j < info->ksi_nthreads_max; j++) {
2322 sched = &info->ksi_scheds[j];
2323 wake_up_all(&sched->kss_waitq);
2324 }
2325 }
2326 }
2327
2328 i = 4;
2329 read_lock(&ksocknal_data.ksnd_global_lock);
2330 while (ksocknal_data.ksnd_nthreads != 0) {
2331 i++;
2332 CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
2333 "waiting for %d threads to terminate\n",
2334 ksocknal_data.ksnd_nthreads);
2335 read_unlock(&ksocknal_data.ksnd_global_lock);
2336 cfs_pause(cfs_time_seconds(1));
2337 read_lock(&ksocknal_data.ksnd_global_lock);
2338 }
2339 read_unlock(&ksocknal_data.ksnd_global_lock);
2340
2341 ksocknal_free_buffers();
2342
2343 ksocknal_data.ksnd_init = SOCKNAL_INIT_NOTHING;
2344 break;
2345 }
2346
2347 CDEBUG(D_MALLOC, "after NAL cleanup: kmem %d\n",
2348 atomic_read (&libcfs_kmemory));
2349
2350 module_put(THIS_MODULE);
2351}
2352
2353__u64
2354ksocknal_new_incarnation (void)
2355{
2356 struct timeval tv;
2357
2358 /* The incarnation number is the time this module loaded and it
2359 * identifies this particular instance of the socknal. Hopefully
2360 * we won't be able to reboot more frequently than 1MHz for the
2361 * forseeable future :) */
2362
2363 do_gettimeofday(&tv);
2364
2365 return (((__u64)tv.tv_sec) * 1000000) + tv.tv_usec;
2366}
2367
2368int
2369ksocknal_base_startup(void)
2370{
2371 struct ksock_sched_info *info;
2372 int rc;
2373 int i;
2374
2375 LASSERT (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
2376 LASSERT (ksocknal_data.ksnd_nnets == 0);
2377
2378 memset (&ksocknal_data, 0, sizeof (ksocknal_data)); /* zero pointers */
2379
2380 ksocknal_data.ksnd_peer_hash_size = SOCKNAL_PEER_HASH_SIZE;
2381 LIBCFS_ALLOC (ksocknal_data.ksnd_peers,
2382 sizeof (struct list_head) *
2383 ksocknal_data.ksnd_peer_hash_size);
2384 if (ksocknal_data.ksnd_peers == NULL)
2385 return -ENOMEM;
2386
2387 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++)
2388 INIT_LIST_HEAD(&ksocknal_data.ksnd_peers[i]);
2389
2390 rwlock_init(&ksocknal_data.ksnd_global_lock);
2391 INIT_LIST_HEAD(&ksocknal_data.ksnd_nets);
2392
2393 spin_lock_init(&ksocknal_data.ksnd_reaper_lock);
2394 INIT_LIST_HEAD (&ksocknal_data.ksnd_enomem_conns);
2395 INIT_LIST_HEAD (&ksocknal_data.ksnd_zombie_conns);
2396 INIT_LIST_HEAD (&ksocknal_data.ksnd_deathrow_conns);
2397 init_waitqueue_head(&ksocknal_data.ksnd_reaper_waitq);
2398
2399 spin_lock_init(&ksocknal_data.ksnd_connd_lock);
2400 INIT_LIST_HEAD (&ksocknal_data.ksnd_connd_connreqs);
2401 INIT_LIST_HEAD (&ksocknal_data.ksnd_connd_routes);
2402 init_waitqueue_head(&ksocknal_data.ksnd_connd_waitq);
2403
2404 spin_lock_init(&ksocknal_data.ksnd_tx_lock);
2405 INIT_LIST_HEAD (&ksocknal_data.ksnd_idle_noop_txs);
2406
2407 /* NB memset above zeros whole of ksocknal_data */
2408
2409 /* flag lists/ptrs/locks initialised */
2410 ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
2411 try_module_get(THIS_MODULE);
2412
2413 ksocknal_data.ksnd_sched_info = cfs_percpt_alloc(lnet_cpt_table(),
2414 sizeof(*info));
2415 if (ksocknal_data.ksnd_sched_info == NULL)
2416 goto failed;
2417
2418 cfs_percpt_for_each(info, i, ksocknal_data.ksnd_sched_info) {
2419 ksock_sched_t *sched;
2420 int nthrs;
2421
2422 nthrs = cfs_cpt_weight(lnet_cpt_table(), i);
2423 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2424 nthrs = min(nthrs, *ksocknal_tunables.ksnd_nscheds);
2425 } else {
2426 /* max to half of CPUs, assume another half should be
2427 * reserved for upper layer modules */
2428 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2429 }
2430
2431 info->ksi_nthreads_max = nthrs;
2432 info->ksi_cpt = i;
2433
2434 LIBCFS_CPT_ALLOC(info->ksi_scheds, lnet_cpt_table(), i,
2435 info->ksi_nthreads_max * sizeof(*sched));
2436 if (info->ksi_scheds == NULL)
2437 goto failed;
2438
2439 for (; nthrs > 0; nthrs--) {
2440 sched = &info->ksi_scheds[nthrs - 1];
2441
2442 sched->kss_info = info;
2443 spin_lock_init(&sched->kss_lock);
2444 INIT_LIST_HEAD(&sched->kss_rx_conns);
2445 INIT_LIST_HEAD(&sched->kss_tx_conns);
2446 INIT_LIST_HEAD(&sched->kss_zombie_noop_txs);
2447 init_waitqueue_head(&sched->kss_waitq);
2448 }
2449 }
2450
2451 ksocknal_data.ksnd_connd_starting = 0;
2452 ksocknal_data.ksnd_connd_failed_stamp = 0;
2453 ksocknal_data.ksnd_connd_starting_stamp = cfs_time_current_sec();
2454 /* must have at least 2 connds to remain responsive to accepts while
2455 * connecting */
2456 if (*ksocknal_tunables.ksnd_nconnds < SOCKNAL_CONND_RESV + 1)
2457 *ksocknal_tunables.ksnd_nconnds = SOCKNAL_CONND_RESV + 1;
2458
2459 if (*ksocknal_tunables.ksnd_nconnds_max <
2460 *ksocknal_tunables.ksnd_nconnds) {
2461 ksocknal_tunables.ksnd_nconnds_max =
2462 ksocknal_tunables.ksnd_nconnds;
2463 }
2464
2465 for (i = 0; i < *ksocknal_tunables.ksnd_nconnds; i++) {
2466 char name[16];
2467 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2468 ksocknal_data.ksnd_connd_starting++;
2469 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2470
2471
2472 snprintf(name, sizeof(name), "socknal_cd%02d", i);
2473 rc = ksocknal_thread_start(ksocknal_connd,
2474 (void *)((ulong_ptr_t)i), name);
2475 if (rc != 0) {
2476 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2477 ksocknal_data.ksnd_connd_starting--;
2478 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2479 CERROR("Can't spawn socknal connd: %d\n", rc);
2480 goto failed;
2481 }
2482 }
2483
2484 rc = ksocknal_thread_start(ksocknal_reaper, NULL, "socknal_reaper");
2485 if (rc != 0) {
2486 CERROR ("Can't spawn socknal reaper: %d\n", rc);
2487 goto failed;
2488 }
2489
2490 /* flag everything initialised */
2491 ksocknal_data.ksnd_init = SOCKNAL_INIT_ALL;
2492
2493 return 0;
2494
2495 failed:
2496 ksocknal_base_shutdown();
2497 return -ENETDOWN;
2498}
2499
2500void
2501ksocknal_debug_peerhash (lnet_ni_t *ni)
2502{
2503 ksock_peer_t *peer = NULL;
2504 struct list_head *tmp;
2505 int i;
2506
2507 read_lock(&ksocknal_data.ksnd_global_lock);
2508
2509 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
2510 list_for_each (tmp, &ksocknal_data.ksnd_peers[i]) {
2511 peer = list_entry (tmp, ksock_peer_t, ksnp_list);
2512
2513 if (peer->ksnp_ni == ni) break;
2514
2515 peer = NULL;
2516 }
2517 }
2518
2519 if (peer != NULL) {
2520 ksock_route_t *route;
2521 ksock_conn_t *conn;
2522
2523 CWARN ("Active peer on shutdown: %s, ref %d, scnt %d, "
2524 "closing %d, accepting %d, err %d, zcookie "LPU64", "
2525 "txq %d, zc_req %d\n", libcfs_id2str(peer->ksnp_id),
2526 atomic_read(&peer->ksnp_refcount),
2527 peer->ksnp_sharecount, peer->ksnp_closing,
2528 peer->ksnp_accepting, peer->ksnp_error,
2529 peer->ksnp_zc_next_cookie,
2530 !list_empty(&peer->ksnp_tx_queue),
2531 !list_empty(&peer->ksnp_zc_req_list));
2532
2533 list_for_each (tmp, &peer->ksnp_routes) {
2534 route = list_entry(tmp, ksock_route_t, ksnr_list);
2535 CWARN ("Route: ref %d, schd %d, conn %d, cnted %d, "
2536 "del %d\n", atomic_read(&route->ksnr_refcount),
2537 route->ksnr_scheduled, route->ksnr_connecting,
2538 route->ksnr_connected, route->ksnr_deleted);
2539 }
2540
2541 list_for_each (tmp, &peer->ksnp_conns) {
2542 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
2543 CWARN ("Conn: ref %d, sref %d, t %d, c %d\n",
2544 atomic_read(&conn->ksnc_conn_refcount),
2545 atomic_read(&conn->ksnc_sock_refcount),
2546 conn->ksnc_type, conn->ksnc_closing);
2547 }
2548 }
2549
2550 read_unlock(&ksocknal_data.ksnd_global_lock);
2551 return;
2552}
2553
2554void
2555ksocknal_shutdown (lnet_ni_t *ni)
2556{
2557 ksock_net_t *net = ni->ni_data;
2558 int i;
2559 lnet_process_id_t anyid = {0};
2560
2561 anyid.nid = LNET_NID_ANY;
2562 anyid.pid = LNET_PID_ANY;
2563
2564 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL);
2565 LASSERT(ksocknal_data.ksnd_nnets > 0);
2566
2567 spin_lock_bh(&net->ksnn_lock);
2568 net->ksnn_shutdown = 1; /* prevent new peers */
2569 spin_unlock_bh(&net->ksnn_lock);
2570
2571 /* Delete all peers */
2572 ksocknal_del_peer(ni, anyid, 0);
2573
2574 /* Wait for all peer state to clean up */
2575 i = 2;
2576 spin_lock_bh(&net->ksnn_lock);
2577 while (net->ksnn_npeers != 0) {
2578 spin_unlock_bh(&net->ksnn_lock);
2579
2580 i++;
2581 CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
2582 "waiting for %d peers to disconnect\n",
2583 net->ksnn_npeers);
2584 cfs_pause(cfs_time_seconds(1));
2585
2586 ksocknal_debug_peerhash(ni);
2587
2588 spin_lock_bh(&net->ksnn_lock);
2589 }
2590 spin_unlock_bh(&net->ksnn_lock);
2591
2592 for (i = 0; i < net->ksnn_ninterfaces; i++) {
2593 LASSERT (net->ksnn_interfaces[i].ksni_npeers == 0);
2594 LASSERT (net->ksnn_interfaces[i].ksni_nroutes == 0);
2595 }
2596
2597 list_del(&net->ksnn_list);
2598 LIBCFS_FREE(net, sizeof(*net));
2599
2600 ksocknal_data.ksnd_nnets--;
2601 if (ksocknal_data.ksnd_nnets == 0)
2602 ksocknal_base_shutdown();
2603}
2604
2605int
2606ksocknal_enumerate_interfaces(ksock_net_t *net)
2607{
2608 char **names;
2609 int i;
2610 int j;
2611 int rc;
2612 int n;
2613
2614 n = libcfs_ipif_enumerate(&names);
2615 if (n <= 0) {
2616 CERROR("Can't enumerate interfaces: %d\n", n);
2617 return n;
2618 }
2619
2620 for (i = j = 0; i < n; i++) {
2621 int up;
2622 __u32 ip;
2623 __u32 mask;
2624
2625 if (!strcmp(names[i], "lo")) /* skip the loopback IF */
2626 continue;
2627
2628 rc = libcfs_ipif_query(names[i], &up, &ip, &mask);
2629 if (rc != 0) {
2630 CWARN("Can't get interface %s info: %d\n",
2631 names[i], rc);
2632 continue;
2633 }
2634
2635 if (!up) {
2636 CWARN("Ignoring interface %s (down)\n",
2637 names[i]);
2638 continue;
2639 }
2640
2641 if (j == LNET_MAX_INTERFACES) {
2642 CWARN("Ignoring interface %s (too many interfaces)\n",
2643 names[i]);
2644 continue;
2645 }
2646
2647 net->ksnn_interfaces[j].ksni_ipaddr = ip;
2648 net->ksnn_interfaces[j].ksni_netmask = mask;
2649 strncpy(&net->ksnn_interfaces[j].ksni_name[0],
2650 names[i], IFNAMSIZ);
2651 j++;
2652 }
2653
2654 libcfs_ipif_free_enumeration(names, n);
2655
2656 if (j == 0)
2657 CERROR("Can't find any usable interfaces\n");
2658
2659 return j;
2660}
2661
2662int
2663ksocknal_search_new_ipif(ksock_net_t *net)
2664{
2665 int new_ipif = 0;
2666 int i;
2667
2668 for (i = 0; i < net->ksnn_ninterfaces; i++) {
2669 char *ifnam = &net->ksnn_interfaces[i].ksni_name[0];
2670 char *colon = strchr(ifnam, ':');
2671 int found = 0;
2672 ksock_net_t *tmp;
2673 int j;
2674
2675 if (colon != NULL) /* ignore alias device */
2676 *colon = 0;
2677
2678 list_for_each_entry(tmp, &ksocknal_data.ksnd_nets,
2679 ksnn_list) {
2680 for (j = 0; !found && j < tmp->ksnn_ninterfaces; j++) {
2681 char *ifnam2 = &tmp->ksnn_interfaces[j].\
2682 ksni_name[0];
2683 char *colon2 = strchr(ifnam2, ':');
2684
2685 if (colon2 != NULL)
2686 *colon2 = 0;
2687
2688 found = strcmp(ifnam, ifnam2) == 0;
2689 if (colon2 != NULL)
2690 *colon2 = ':';
2691 }
2692 if (found)
2693 break;
2694 }
2695
2696 new_ipif += !found;
2697 if (colon != NULL)
2698 *colon = ':';
2699 }
2700
2701 return new_ipif;
2702}
2703
2704int
2705ksocknal_start_schedulers(struct ksock_sched_info *info)
2706{
2707 int nthrs;
2708 int rc = 0;
2709 int i;
2710
2711 if (info->ksi_nthreads == 0) {
2712 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2713 nthrs = info->ksi_nthreads_max;
2714 } else {
2715 nthrs = cfs_cpt_weight(lnet_cpt_table(),
2716 info->ksi_cpt);
2717 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2718 nthrs = min(SOCKNAL_NSCHEDS_HIGH, nthrs);
2719 }
2720 nthrs = min(nthrs, info->ksi_nthreads_max);
2721 } else {
2722 LASSERT(info->ksi_nthreads <= info->ksi_nthreads_max);
2723 /* increase two threads if there is new interface */
2724 nthrs = min(2, info->ksi_nthreads_max - info->ksi_nthreads);
2725 }
2726
2727 for (i = 0; i < nthrs; i++) {
2728 long id;
2729 char name[20];
2730 ksock_sched_t *sched;
2731 id = KSOCK_THREAD_ID(info->ksi_cpt, info->ksi_nthreads + i);
2732 sched = &info->ksi_scheds[KSOCK_THREAD_SID(id)];
2733 snprintf(name, sizeof(name), "socknal_sd%02d_%02d",
2734 info->ksi_cpt, (int)(sched - &info->ksi_scheds[0]));
2735
2736 rc = ksocknal_thread_start(ksocknal_scheduler,
2737 (void *)id, name);
2738 if (rc == 0)
2739 continue;
2740
2741 CERROR("Can't spawn thread %d for scheduler[%d]: %d\n",
2742 info->ksi_cpt, info->ksi_nthreads + i, rc);
2743 break;
2744 }
2745
2746 info->ksi_nthreads += i;
2747 return rc;
2748}
2749
2750int
2751ksocknal_net_start_threads(ksock_net_t *net, __u32 *cpts, int ncpts)
2752{
2753 int newif = ksocknal_search_new_ipif(net);
2754 int rc;
2755 int i;
2756
2757 LASSERT(ncpts > 0 && ncpts <= cfs_cpt_number(lnet_cpt_table()));
2758
2759 for (i = 0; i < ncpts; i++) {
2760 struct ksock_sched_info *info;
2761 int cpt = (cpts == NULL) ? i : cpts[i];
2762
2763 LASSERT(cpt < cfs_cpt_number(lnet_cpt_table()));
2764 info = ksocknal_data.ksnd_sched_info[cpt];
2765
2766 if (!newif && info->ksi_nthreads > 0)
2767 continue;
2768
2769 rc = ksocknal_start_schedulers(info);
2770 if (rc != 0)
2771 return rc;
2772 }
2773 return 0;
2774}
2775
2776int
2777ksocknal_startup (lnet_ni_t *ni)
2778{
2779 ksock_net_t *net;
2780 int rc;
2781 int i;
2782
2783 LASSERT (ni->ni_lnd == &the_ksocklnd);
2784
2785 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING) {
2786 rc = ksocknal_base_startup();
2787 if (rc != 0)
2788 return rc;
2789 }
2790
2791 LIBCFS_ALLOC(net, sizeof(*net));
2792 if (net == NULL)
2793 goto fail_0;
2794
2795 spin_lock_init(&net->ksnn_lock);
2796 net->ksnn_incarnation = ksocknal_new_incarnation();
2797 ni->ni_data = net;
2798 ni->ni_peertimeout = *ksocknal_tunables.ksnd_peertimeout;
2799 ni->ni_maxtxcredits = *ksocknal_tunables.ksnd_credits;
2800 ni->ni_peertxcredits = *ksocknal_tunables.ksnd_peertxcredits;
2801 ni->ni_peerrtrcredits = *ksocknal_tunables.ksnd_peerrtrcredits;
2802
2803 if (ni->ni_interfaces[0] == NULL) {
2804 rc = ksocknal_enumerate_interfaces(net);
2805 if (rc <= 0)
2806 goto fail_1;
2807
2808 net->ksnn_ninterfaces = 1;
2809 } else {
2810 for (i = 0; i < LNET_MAX_INTERFACES; i++) {
2811 int up;
2812
2813 if (ni->ni_interfaces[i] == NULL)
2814 break;
2815
2816 rc = libcfs_ipif_query(
2817 ni->ni_interfaces[i], &up,
2818 &net->ksnn_interfaces[i].ksni_ipaddr,
2819 &net->ksnn_interfaces[i].ksni_netmask);
2820
2821 if (rc != 0) {
2822 CERROR("Can't get interface %s info: %d\n",
2823 ni->ni_interfaces[i], rc);
2824 goto fail_1;
2825 }
2826
2827 if (!up) {
2828 CERROR("Interface %s is down\n",
2829 ni->ni_interfaces[i]);
2830 goto fail_1;
2831 }
2832
2833 strncpy(&net->ksnn_interfaces[i].ksni_name[0],
2834 ni->ni_interfaces[i], IFNAMSIZ);
2835 }
2836 net->ksnn_ninterfaces = i;
2837 }
2838
2839 /* call it before add it to ksocknal_data.ksnd_nets */
2840 rc = ksocknal_net_start_threads(net, ni->ni_cpts, ni->ni_ncpts);
2841 if (rc != 0)
2842 goto fail_1;
2843
2844 ni->ni_nid = LNET_MKNID(LNET_NIDNET(ni->ni_nid),
2845 net->ksnn_interfaces[0].ksni_ipaddr);
2846 list_add(&net->ksnn_list, &ksocknal_data.ksnd_nets);
2847
2848 ksocknal_data.ksnd_nnets++;
2849
2850 return 0;
2851
2852 fail_1:
2853 LIBCFS_FREE(net, sizeof(*net));
2854 fail_0:
2855 if (ksocknal_data.ksnd_nnets == 0)
2856 ksocknal_base_shutdown();
2857
2858 return -ENETDOWN;
2859}
2860
2861
2862void __exit
2863ksocknal_module_fini (void)
2864{
2865 lnet_unregister_lnd(&the_ksocklnd);
2866 ksocknal_tunables_fini();
2867}
2868
2869int __init
2870ksocknal_module_init (void)
2871{
2872 int rc;
2873
2874 /* check ksnr_connected/connecting field large enough */
2875 CLASSERT (SOCKLND_CONN_NTYPES <= 4);
2876 CLASSERT (SOCKLND_CONN_ACK == SOCKLND_CONN_BULK_IN);
2877
2878 /* initialize the_ksocklnd */
2879 the_ksocklnd.lnd_type = SOCKLND;
2880 the_ksocklnd.lnd_startup = ksocknal_startup;
2881 the_ksocklnd.lnd_shutdown = ksocknal_shutdown;
2882 the_ksocklnd.lnd_ctl = ksocknal_ctl;
2883 the_ksocklnd.lnd_send = ksocknal_send;
2884 the_ksocklnd.lnd_recv = ksocknal_recv;
2885 the_ksocklnd.lnd_notify = ksocknal_notify;
2886 the_ksocklnd.lnd_query = ksocknal_query;
2887 the_ksocklnd.lnd_accept = ksocknal_accept;
2888
2889 rc = ksocknal_tunables_init();
2890 if (rc != 0)
2891 return rc;
2892
2893 lnet_register_lnd(&the_ksocklnd);
2894
2895 return 0;
2896}
2897
2898MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2899MODULE_DESCRIPTION("Kernel TCP Socket LND v3.0.0");
2900MODULE_LICENSE("GPL");
2901
2902cfs_module(ksocknal, "3.0.0", ksocknal_module_init, ksocknal_module_fini);
Linux 6.x block layer and io_uring tree(s)