1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lnet/klnds/socklnd/socklnd.c
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>
47 ksock_nal_data_t ksocknal_data;
50 ksocknal_ip2iface(lnet_ni_t *ni, __u32 ip)
52 ksock_net_t *net = ni->ni_data;
54 ksock_interface_t *iface;
56 for (i = 0; i < net->ksnn_ninterfaces; i++) {
57 LASSERT(i < LNET_MAX_INTERFACES);
58 iface = &net->ksnn_interfaces[i];
60 if (iface->ksni_ipaddr == ip)
68 ksocknal_create_route (__u32 ipaddr, int port)
72 LIBCFS_ALLOC (route, sizeof (*route));
76 cfs_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;
92 ksocknal_destroy_route (ksock_route_t *route)
94 LASSERT (cfs_atomic_read(&route->ksnr_refcount) == 0);
96 if (route->ksnr_peer != NULL)
97 ksocknal_peer_decref(route->ksnr_peer);
99 LIBCFS_FREE (route, sizeof (*route));
103 ksocknal_create_peer (ksock_peer_t **peerp, lnet_ni_t *ni, lnet_process_id_t id)
105 ksock_net_t *net = ni->ni_data;
108 LASSERT (id.nid != LNET_NID_ANY);
109 LASSERT (id.pid != LNET_PID_ANY);
110 LASSERT (!cfs_in_interrupt());
112 LIBCFS_ALLOC (peer, sizeof (*peer));
116 memset (peer, 0, sizeof (*peer)); /* NULL pointers/clear flags etc */
120 cfs_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_zc_next_cookie = SOCKNAL_KEEPALIVE_PING + 1;
126 CFS_INIT_LIST_HEAD (&peer->ksnp_conns);
127 CFS_INIT_LIST_HEAD (&peer->ksnp_routes);
128 CFS_INIT_LIST_HEAD (&peer->ksnp_tx_queue);
129 CFS_INIT_LIST_HEAD (&peer->ksnp_zc_req_list);
130 cfs_spin_lock_init(&peer->ksnp_lock);
132 cfs_spin_lock_bh (&net->ksnn_lock);
134 if (net->ksnn_shutdown) {
135 cfs_spin_unlock_bh (&net->ksnn_lock);
137 LIBCFS_FREE(peer, sizeof(*peer));
138 CERROR("Can't create peer: network shutdown\n");
144 cfs_spin_unlock_bh (&net->ksnn_lock);
151 ksocknal_destroy_peer (ksock_peer_t *peer)
153 ksock_net_t *net = peer->ksnp_ni->ni_data;
155 CDEBUG (D_NET, "peer %s %p deleted\n",
156 libcfs_id2str(peer->ksnp_id), peer);
158 LASSERT (cfs_atomic_read (&peer->ksnp_refcount) == 0);
159 LASSERT (peer->ksnp_accepting == 0);
160 LASSERT (list_empty (&peer->ksnp_conns));
161 LASSERT (list_empty (&peer->ksnp_routes));
162 LASSERT (list_empty (&peer->ksnp_tx_queue));
163 LASSERT (list_empty (&peer->ksnp_zc_req_list));
165 LIBCFS_FREE (peer, sizeof (*peer));
167 /* NB a peer's connections and routes keep a reference on their peer
168 * until they are destroyed, so we can be assured that _all_ state to
169 * do with this peer has been cleaned up when its refcount drops to
171 cfs_spin_lock_bh (&net->ksnn_lock);
173 cfs_spin_unlock_bh (&net->ksnn_lock);
177 ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id)
179 struct list_head *peer_list = ksocknal_nid2peerlist(id.nid);
180 struct list_head *tmp;
183 list_for_each (tmp, peer_list) {
185 peer = list_entry (tmp, ksock_peer_t, ksnp_list);
187 LASSERT (!peer->ksnp_closing);
189 if (peer->ksnp_ni != ni)
192 if (peer->ksnp_id.nid != id.nid ||
193 peer->ksnp_id.pid != id.pid)
196 CDEBUG(D_NET, "got peer [%p] -> %s (%d)\n",
197 peer, libcfs_id2str(id),
198 cfs_atomic_read(&peer->ksnp_refcount));
205 ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id)
209 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
210 peer = ksocknal_find_peer_locked (ni, id);
211 if (peer != NULL) /* +1 ref for caller? */
212 ksocknal_peer_addref(peer);
213 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
219 ksocknal_unlink_peer_locked (ksock_peer_t *peer)
223 ksock_interface_t *iface;
225 for (i = 0; i < peer->ksnp_n_passive_ips; i++) {
226 LASSERT (i < LNET_MAX_INTERFACES);
227 ip = peer->ksnp_passive_ips[i];
229 iface = ksocknal_ip2iface(peer->ksnp_ni, ip);
230 /* All IPs in peer->ksnp_passive_ips[] come from the
231 * interface list, therefore the call must succeed. */
232 LASSERT (iface != NULL);
234 CDEBUG(D_NET, "peer=%p iface=%p ksni_nroutes=%d\n",
235 peer, iface, iface->ksni_nroutes);
236 iface->ksni_npeers--;
239 LASSERT (list_empty(&peer->ksnp_conns));
240 LASSERT (list_empty(&peer->ksnp_routes));
241 LASSERT (!peer->ksnp_closing);
242 peer->ksnp_closing = 1;
243 list_del (&peer->ksnp_list);
244 /* lose peerlist's ref */
245 ksocknal_peer_decref(peer);
249 ksocknal_get_peer_info (lnet_ni_t *ni, int index,
250 lnet_process_id_t *id, __u32 *myip, __u32 *peer_ip, int *port,
251 int *conn_count, int *share_count)
254 struct list_head *ptmp;
255 ksock_route_t *route;
256 struct list_head *rtmp;
261 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
263 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
265 list_for_each (ptmp, &ksocknal_data.ksnd_peers[i]) {
266 peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
268 if (peer->ksnp_ni != ni)
271 if (peer->ksnp_n_passive_ips == 0 &&
272 list_empty(&peer->ksnp_routes)) {
286 for (j = 0; j < peer->ksnp_n_passive_ips; j++) {
291 *myip = peer->ksnp_passive_ips[j];
300 list_for_each (rtmp, &peer->ksnp_routes) {
304 route = list_entry(rtmp, ksock_route_t,
308 *myip = route->ksnr_myipaddr;
309 *peer_ip = route->ksnr_ipaddr;
310 *port = route->ksnr_port;
311 *conn_count = route->ksnr_conn_count;
312 *share_count = route->ksnr_share_count;
319 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
324 ksocknal_associate_route_conn_locked(ksock_route_t *route, ksock_conn_t *conn)
326 ksock_peer_t *peer = route->ksnr_peer;
327 int type = conn->ksnc_type;
328 ksock_interface_t *iface;
330 conn->ksnc_route = route;
331 ksocknal_route_addref(route);
333 if (route->ksnr_myipaddr != conn->ksnc_myipaddr) {
334 if (route->ksnr_myipaddr == 0) {
335 /* route wasn't bound locally yet (the initial route) */
336 CDEBUG(D_NET, "Binding %s %u.%u.%u.%u to %u.%u.%u.%u\n",
337 libcfs_id2str(peer->ksnp_id),
338 HIPQUAD(route->ksnr_ipaddr),
339 HIPQUAD(conn->ksnc_myipaddr));
341 CDEBUG(D_NET, "Rebinding %s %u.%u.%u.%u from "
342 "%u.%u.%u.%u to %u.%u.%u.%u\n",
343 libcfs_id2str(peer->ksnp_id),
344 HIPQUAD(route->ksnr_ipaddr),
345 HIPQUAD(route->ksnr_myipaddr),
346 HIPQUAD(conn->ksnc_myipaddr));
348 iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
349 route->ksnr_myipaddr);
351 iface->ksni_nroutes--;
353 route->ksnr_myipaddr = conn->ksnc_myipaddr;
354 iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
355 route->ksnr_myipaddr);
357 iface->ksni_nroutes++;
360 route->ksnr_connected |= (1<<type);
361 route->ksnr_conn_count++;
363 /* Successful connection => further attempts can
364 * proceed immediately */
365 route->ksnr_retry_interval = 0;
369 ksocknal_add_route_locked (ksock_peer_t *peer, ksock_route_t *route)
371 struct list_head *tmp;
373 ksock_route_t *route2;
375 LASSERT (!peer->ksnp_closing);
376 LASSERT (route->ksnr_peer == NULL);
377 LASSERT (!route->ksnr_scheduled);
378 LASSERT (!route->ksnr_connecting);
379 LASSERT (route->ksnr_connected == 0);
381 /* LASSERT(unique) */
382 list_for_each(tmp, &peer->ksnp_routes) {
383 route2 = list_entry(tmp, ksock_route_t, ksnr_list);
385 if (route2->ksnr_ipaddr == route->ksnr_ipaddr) {
386 CERROR ("Duplicate route %s %u.%u.%u.%u\n",
387 libcfs_id2str(peer->ksnp_id),
388 HIPQUAD(route->ksnr_ipaddr));
393 route->ksnr_peer = peer;
394 ksocknal_peer_addref(peer);
395 /* peer's routelist takes over my ref on 'route' */
396 list_add_tail(&route->ksnr_list, &peer->ksnp_routes);
398 list_for_each(tmp, &peer->ksnp_conns) {
399 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
401 if (conn->ksnc_ipaddr != route->ksnr_ipaddr)
404 ksocknal_associate_route_conn_locked(route, conn);
405 /* keep going (typed routes) */
410 ksocknal_del_route_locked (ksock_route_t *route)
412 ksock_peer_t *peer = route->ksnr_peer;
413 ksock_interface_t *iface;
415 struct list_head *ctmp;
416 struct list_head *cnxt;
418 LASSERT (!route->ksnr_deleted);
420 /* Close associated conns */
421 list_for_each_safe (ctmp, cnxt, &peer->ksnp_conns) {
422 conn = list_entry(ctmp, ksock_conn_t, ksnc_list);
424 if (conn->ksnc_route != route)
427 ksocknal_close_conn_locked (conn, 0);
430 if (route->ksnr_myipaddr != 0) {
431 iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
432 route->ksnr_myipaddr);
434 iface->ksni_nroutes--;
437 route->ksnr_deleted = 1;
438 list_del (&route->ksnr_list);
439 ksocknal_route_decref(route); /* drop peer's ref */
441 if (list_empty (&peer->ksnp_routes) &&
442 list_empty (&peer->ksnp_conns)) {
443 /* I've just removed the last route to a peer with no active
445 ksocknal_unlink_peer_locked (peer);
450 ksocknal_add_peer (lnet_ni_t *ni, lnet_process_id_t id, __u32 ipaddr, int port)
452 struct list_head *tmp;
455 ksock_route_t *route;
456 ksock_route_t *route2;
459 if (id.nid == LNET_NID_ANY ||
460 id.pid == LNET_PID_ANY)
463 /* Have a brand new peer ready... */
464 rc = ksocknal_create_peer(&peer, ni, id);
468 route = ksocknal_create_route (ipaddr, port);
470 ksocknal_peer_decref(peer);
474 cfs_write_lock_bh (&ksocknal_data.ksnd_global_lock);
476 /* always called with a ref on ni, so shutdown can't have started */
477 LASSERT (((ksock_net_t *) ni->ni_data)->ksnn_shutdown == 0);
479 peer2 = ksocknal_find_peer_locked (ni, id);
481 ksocknal_peer_decref(peer);
484 /* peer table takes my ref on peer */
485 list_add_tail (&peer->ksnp_list,
486 ksocknal_nid2peerlist (id.nid));
490 list_for_each (tmp, &peer->ksnp_routes) {
491 route2 = list_entry(tmp, ksock_route_t, ksnr_list);
493 if (route2->ksnr_ipaddr == ipaddr)
498 if (route2 == NULL) {
499 ksocknal_add_route_locked(peer, route);
500 route->ksnr_share_count++;
502 ksocknal_route_decref(route);
503 route2->ksnr_share_count++;
506 cfs_write_unlock_bh (&ksocknal_data.ksnd_global_lock);
512 ksocknal_del_peer_locked (ksock_peer_t *peer, __u32 ip)
515 ksock_route_t *route;
516 struct list_head *tmp;
517 struct list_head *nxt;
520 LASSERT (!peer->ksnp_closing);
522 /* Extra ref prevents peer disappearing until I'm done with it */
523 ksocknal_peer_addref(peer);
525 list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
526 route = list_entry(tmp, ksock_route_t, ksnr_list);
529 if (!(ip == 0 || route->ksnr_ipaddr == ip))
532 route->ksnr_share_count = 0;
533 /* This deletes associated conns too */
534 ksocknal_del_route_locked (route);
538 list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
539 route = list_entry(tmp, ksock_route_t, ksnr_list);
540 nshared += route->ksnr_share_count;
544 /* remove everything else if there are no explicit entries
547 list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
548 route = list_entry(tmp, ksock_route_t, ksnr_list);
550 /* we should only be removing auto-entries */
551 LASSERT(route->ksnr_share_count == 0);
552 ksocknal_del_route_locked (route);
555 list_for_each_safe (tmp, nxt, &peer->ksnp_conns) {
556 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
558 ksocknal_close_conn_locked(conn, 0);
562 ksocknal_peer_decref(peer);
563 /* NB peer unlinks itself when last conn/route is removed */
567 ksocknal_del_peer (lnet_ni_t *ni, lnet_process_id_t id, __u32 ip)
569 CFS_LIST_HEAD (zombies);
570 struct list_head *ptmp;
571 struct list_head *pnxt;
578 cfs_write_lock_bh (&ksocknal_data.ksnd_global_lock);
580 if (id.nid != LNET_NID_ANY)
581 lo = hi = (int)(ksocknal_nid2peerlist(id.nid) - ksocknal_data.ksnd_peers);
584 hi = ksocknal_data.ksnd_peer_hash_size - 1;
587 for (i = lo; i <= hi; i++) {
588 list_for_each_safe (ptmp, pnxt, &ksocknal_data.ksnd_peers[i]) {
589 peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
591 if (peer->ksnp_ni != ni)
594 if (!((id.nid == LNET_NID_ANY || peer->ksnp_id.nid == id.nid) &&
595 (id.pid == LNET_PID_ANY || peer->ksnp_id.pid == id.pid)))
598 ksocknal_peer_addref(peer); /* a ref for me... */
600 ksocknal_del_peer_locked (peer, ip);
602 if (peer->ksnp_closing && !list_empty(&peer->ksnp_tx_queue)) {
603 LASSERT (list_empty(&peer->ksnp_conns));
604 LASSERT (list_empty(&peer->ksnp_routes));
606 list_splice_init(&peer->ksnp_tx_queue, &zombies);
609 ksocknal_peer_decref(peer); /* ...till here */
611 rc = 0; /* matched! */
615 cfs_write_unlock_bh (&ksocknal_data.ksnd_global_lock);
617 ksocknal_txlist_done(ni, &zombies, 1);
623 ksocknal_get_conn_by_idx (lnet_ni_t *ni, int index)
626 struct list_head *ptmp;
628 struct list_head *ctmp;
631 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
633 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
634 list_for_each (ptmp, &ksocknal_data.ksnd_peers[i]) {
635 peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
637 LASSERT (!peer->ksnp_closing);
639 if (peer->ksnp_ni != ni)
642 list_for_each (ctmp, &peer->ksnp_conns) {
646 conn = list_entry (ctmp, ksock_conn_t, ksnc_list);
647 ksocknal_conn_addref(conn);
648 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
654 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
659 ksocknal_choose_scheduler_locked (unsigned int irq)
661 ksock_sched_t *sched;
662 ksock_irqinfo_t *info;
665 LASSERT (irq < NR_IRQS);
666 info = &ksocknal_data.ksnd_irqinfo[irq];
668 if (irq != 0 && /* hardware NIC */
669 info->ksni_valid) { /* already set up */
670 return (&ksocknal_data.ksnd_schedulers[info->ksni_sched]);
673 /* software NIC (irq == 0) || not associated with a scheduler yet.
674 * Choose the CPU with the fewest connections... */
675 sched = &ksocknal_data.ksnd_schedulers[0];
676 for (i = 1; i < ksocknal_data.ksnd_nschedulers; i++)
677 if (sched->kss_nconns >
678 ksocknal_data.ksnd_schedulers[i].kss_nconns)
679 sched = &ksocknal_data.ksnd_schedulers[i];
681 if (irq != 0) { /* Hardware NIC */
682 info->ksni_valid = 1;
683 info->ksni_sched = (unsigned int)(sched - ksocknal_data.ksnd_schedulers);
686 LASSERT (info->ksni_sched == (unsigned int)(sched - ksocknal_data.ksnd_schedulers));
693 ksocknal_local_ipvec (lnet_ni_t *ni, __u32 *ipaddrs)
695 ksock_net_t *net = ni->ni_data;
699 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
701 nip = net->ksnn_ninterfaces;
702 LASSERT (nip <= LNET_MAX_INTERFACES);
704 /* Only offer interfaces for additional connections if I have
707 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
711 for (i = 0; i < nip; i++) {
712 ipaddrs[i] = net->ksnn_interfaces[i].ksni_ipaddr;
713 LASSERT (ipaddrs[i] != 0);
716 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
721 ksocknal_match_peerip (ksock_interface_t *iface, __u32 *ips, int nips)
723 int best_netmatch = 0;
730 for (i = 0; i < nips; i++) {
734 this_xor = (ips[i] ^ iface->ksni_ipaddr);
735 this_netmatch = ((this_xor & iface->ksni_netmask) == 0) ? 1 : 0;
738 best_netmatch < this_netmatch ||
739 (best_netmatch == this_netmatch &&
740 best_xor > this_xor)))
744 best_netmatch = this_netmatch;
753 ksocknal_select_ips(ksock_peer_t *peer, __u32 *peerips, int n_peerips)
755 cfs_rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
756 ksock_net_t *net = peer->ksnp_ni->ni_data;
757 ksock_interface_t *iface;
758 ksock_interface_t *best_iface;
769 /* CAVEAT EMPTOR: We do all our interface matching with an
770 * exclusive hold of global lock at IRQ priority. We're only
771 * expecting to be dealing with small numbers of interfaces, so the
772 * O(n**3)-ness shouldn't matter */
774 /* Also note that I'm not going to return more than n_peerips
775 * interfaces, even if I have more myself */
777 cfs_write_lock_bh (global_lock);
779 LASSERT (n_peerips <= LNET_MAX_INTERFACES);
780 LASSERT (net->ksnn_ninterfaces <= LNET_MAX_INTERFACES);
782 /* Only match interfaces for additional connections
783 * if I have > 1 interface */
784 n_ips = (net->ksnn_ninterfaces < 2) ? 0 :
785 MIN(n_peerips, net->ksnn_ninterfaces);
787 for (i = 0; peer->ksnp_n_passive_ips < n_ips; i++) {
788 /* ^ yes really... */
790 /* If we have any new interfaces, first tick off all the
791 * peer IPs that match old interfaces, then choose new
792 * interfaces to match the remaining peer IPS.
793 * We don't forget interfaces we've stopped using; we might
794 * start using them again... */
796 if (i < peer->ksnp_n_passive_ips) {
798 ip = peer->ksnp_passive_ips[i];
799 best_iface = ksocknal_ip2iface(peer->ksnp_ni, ip);
801 /* peer passive ips are kept up to date */
802 LASSERT(best_iface != NULL);
804 /* choose a new interface */
805 LASSERT (i == peer->ksnp_n_passive_ips);
811 for (j = 0; j < net->ksnn_ninterfaces; j++) {
812 iface = &net->ksnn_interfaces[j];
813 ip = iface->ksni_ipaddr;
815 for (k = 0; k < peer->ksnp_n_passive_ips; k++)
816 if (peer->ksnp_passive_ips[k] == ip)
819 if (k < peer->ksnp_n_passive_ips) /* using it already */
822 k = ksocknal_match_peerip(iface, peerips, n_peerips);
823 xor = (ip ^ peerips[k]);
824 this_netmatch = ((xor & iface->ksni_netmask) == 0) ? 1 : 0;
826 if (!(best_iface == NULL ||
827 best_netmatch < this_netmatch ||
828 (best_netmatch == this_netmatch &&
829 best_npeers > iface->ksni_npeers)))
833 best_netmatch = this_netmatch;
834 best_npeers = iface->ksni_npeers;
837 best_iface->ksni_npeers++;
838 ip = best_iface->ksni_ipaddr;
839 peer->ksnp_passive_ips[i] = ip;
840 peer->ksnp_n_passive_ips = i+1;
843 LASSERT (best_iface != NULL);
845 /* mark the best matching peer IP used */
846 j = ksocknal_match_peerip(best_iface, peerips, n_peerips);
850 /* Overwrite input peer IP addresses */
851 memcpy(peerips, peer->ksnp_passive_ips, n_ips * sizeof(*peerips));
853 cfs_write_unlock_bh (global_lock);
859 ksocknal_create_routes(ksock_peer_t *peer, int port,
860 __u32 *peer_ipaddrs, int npeer_ipaddrs)
862 ksock_route_t *newroute = NULL;
863 cfs_rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
864 lnet_ni_t *ni = peer->ksnp_ni;
865 ksock_net_t *net = ni->ni_data;
866 struct list_head *rtmp;
867 ksock_route_t *route;
868 ksock_interface_t *iface;
869 ksock_interface_t *best_iface;
876 /* CAVEAT EMPTOR: We do all our interface matching with an
877 * exclusive hold of global lock at IRQ priority. We're only
878 * expecting to be dealing with small numbers of interfaces, so the
879 * O(n**3)-ness here shouldn't matter */
881 cfs_write_lock_bh (global_lock);
883 if (net->ksnn_ninterfaces < 2) {
884 /* Only create additional connections
885 * if I have > 1 interface */
886 cfs_write_unlock_bh (global_lock);
890 LASSERT (npeer_ipaddrs <= LNET_MAX_INTERFACES);
892 for (i = 0; i < npeer_ipaddrs; i++) {
893 if (newroute != NULL) {
894 newroute->ksnr_ipaddr = peer_ipaddrs[i];
896 cfs_write_unlock_bh (global_lock);
898 newroute = ksocknal_create_route(peer_ipaddrs[i], port);
899 if (newroute == NULL)
902 cfs_write_lock_bh (global_lock);
905 if (peer->ksnp_closing) {
906 /* peer got closed under me */
910 /* Already got a route? */
912 list_for_each(rtmp, &peer->ksnp_routes) {
913 route = list_entry(rtmp, ksock_route_t, ksnr_list);
915 if (route->ksnr_ipaddr == newroute->ksnr_ipaddr)
927 LASSERT (net->ksnn_ninterfaces <= LNET_MAX_INTERFACES);
929 /* Select interface to connect from */
930 for (j = 0; j < net->ksnn_ninterfaces; j++) {
931 iface = &net->ksnn_interfaces[j];
933 /* Using this interface already? */
934 list_for_each(rtmp, &peer->ksnp_routes) {
935 route = list_entry(rtmp, ksock_route_t, ksnr_list);
937 if (route->ksnr_myipaddr == iface->ksni_ipaddr)
945 this_netmatch = (((iface->ksni_ipaddr ^
946 newroute->ksnr_ipaddr) &
947 iface->ksni_netmask) == 0) ? 1 : 0;
949 if (!(best_iface == NULL ||
950 best_netmatch < this_netmatch ||
951 (best_netmatch == this_netmatch &&
952 best_nroutes > iface->ksni_nroutes)))
956 best_netmatch = this_netmatch;
957 best_nroutes = iface->ksni_nroutes;
960 if (best_iface == NULL)
963 newroute->ksnr_myipaddr = best_iface->ksni_ipaddr;
964 best_iface->ksni_nroutes++;
966 ksocknal_add_route_locked(peer, newroute);
970 cfs_write_unlock_bh (global_lock);
971 if (newroute != NULL)
972 ksocknal_route_decref(newroute);
976 ksocknal_accept (lnet_ni_t *ni, cfs_socket_t *sock)
983 rc = libcfs_sock_getaddr(sock, 1, &peer_ip, &peer_port);
984 LASSERT (rc == 0); /* we succeeded before */
986 LIBCFS_ALLOC(cr, sizeof(*cr));
988 LCONSOLE_ERROR_MSG(0x12f, "Dropping connection request from "
989 "%u.%u.%u.%u: memory exhausted\n",
996 cr->ksncr_sock = sock;
998 cfs_spin_lock_bh (&ksocknal_data.ksnd_connd_lock);
1000 list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
1001 cfs_waitq_signal(&ksocknal_data.ksnd_connd_waitq);
1003 cfs_spin_unlock_bh (&ksocknal_data.ksnd_connd_lock);
1008 ksocknal_connecting (ksock_peer_t *peer, __u32 ipaddr)
1010 ksock_route_t *route;
1012 cfs_list_for_each_entry_typed (route, &peer->ksnp_routes,
1013 ksock_route_t, ksnr_list) {
1015 if (route->ksnr_ipaddr == ipaddr)
1016 return route->ksnr_connecting;
1022 ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
1023 cfs_socket_t *sock, int type)
1025 cfs_rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
1026 CFS_LIST_HEAD (zombies);
1027 lnet_process_id_t peerid;
1028 struct list_head *tmp;
1031 ksock_conn_t *conn2;
1032 ksock_peer_t *peer = NULL;
1033 ksock_peer_t *peer2;
1034 ksock_sched_t *sched;
1035 ksock_hello_msg_t *hello;
1043 active = (route != NULL);
1045 LASSERT (active == (type != SOCKLND_CONN_NONE));
1047 irq = ksocknal_lib_sock_irq (sock);
1049 LIBCFS_ALLOC(conn, sizeof(*conn));
1055 memset (conn, 0, sizeof (*conn));
1057 conn->ksnc_peer = NULL;
1058 conn->ksnc_route = NULL;
1059 conn->ksnc_sock = sock;
1060 /* 2 ref, 1 for conn, another extra ref prevents socket
1061 * being closed before establishment of connection */
1062 cfs_atomic_set (&conn->ksnc_sock_refcount, 2);
1063 conn->ksnc_type = type;
1064 ksocknal_lib_save_callback(sock, conn);
1065 cfs_atomic_set (&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
1067 conn->ksnc_rx_ready = 0;
1068 conn->ksnc_rx_scheduled = 0;
1070 CFS_INIT_LIST_HEAD (&conn->ksnc_tx_queue);
1071 conn->ksnc_tx_ready = 0;
1072 conn->ksnc_tx_scheduled = 0;
1073 conn->ksnc_tx_carrier = NULL;
1074 cfs_atomic_set (&conn->ksnc_tx_nob, 0);
1076 LIBCFS_ALLOC(hello, offsetof(ksock_hello_msg_t,
1077 kshm_ips[LNET_MAX_INTERFACES]));
1078 if (hello == NULL) {
1083 /* stash conn's local and remote addrs */
1084 rc = ksocknal_lib_get_conn_addrs (conn);
1088 /* Find out/confirm peer's NID and connection type and get the
1089 * vector of interfaces she's willing to let me connect to.
1090 * Passive connections use the listener timeout since the peer sends
1094 peer = route->ksnr_peer;
1095 LASSERT(ni == peer->ksnp_ni);
1097 /* Active connection sends HELLO eagerly */
1098 hello->kshm_nips = ksocknal_local_ipvec(ni, hello->kshm_ips);
1099 peerid = peer->ksnp_id;
1101 cfs_write_lock_bh(global_lock);
1102 conn->ksnc_proto = peer->ksnp_proto;
1103 cfs_write_unlock_bh(global_lock);
1105 if (conn->ksnc_proto == NULL) {
1106 conn->ksnc_proto = &ksocknal_protocol_v3x;
1107 #if SOCKNAL_VERSION_DEBUG
1108 if (*ksocknal_tunables.ksnd_protocol == 2)
1109 conn->ksnc_proto = &ksocknal_protocol_v2x;
1110 else if (*ksocknal_tunables.ksnd_protocol == 1)
1111 conn->ksnc_proto = &ksocknal_protocol_v1x;
1115 rc = ksocknal_send_hello (ni, conn, peerid.nid, hello);
1119 peerid.nid = LNET_NID_ANY;
1120 peerid.pid = LNET_PID_ANY;
1122 /* Passive, get protocol from peer */
1123 conn->ksnc_proto = NULL;
1126 rc = ksocknal_recv_hello (ni, conn, hello, &peerid, &incarnation);
1130 LASSERT (rc == 0 || active);
1131 LASSERT (conn->ksnc_proto != NULL);
1132 LASSERT (peerid.nid != LNET_NID_ANY);
1135 ksocknal_peer_addref(peer);
1136 cfs_write_lock_bh (global_lock);
1138 rc = ksocknal_create_peer(&peer, ni, peerid);
1142 cfs_write_lock_bh (global_lock);
1144 /* called with a ref on ni, so shutdown can't have started */
1145 LASSERT (((ksock_net_t *) ni->ni_data)->ksnn_shutdown == 0);
1147 peer2 = ksocknal_find_peer_locked(ni, peerid);
1148 if (peer2 == NULL) {
1149 /* NB this puts an "empty" peer in the peer
1150 * table (which takes my ref) */
1151 list_add_tail(&peer->ksnp_list,
1152 ksocknal_nid2peerlist(peerid.nid));
1154 ksocknal_peer_decref(peer);
1159 ksocknal_peer_addref(peer);
1160 peer->ksnp_accepting++;
1162 /* Am I already connecting to this guy? Resolve in
1163 * favour of higher NID... */
1164 if (peerid.nid < ni->ni_nid &&
1165 ksocknal_connecting(peer, conn->ksnc_ipaddr)) {
1167 warn = "connection race resolution";
1172 if (peer->ksnp_closing ||
1173 (active && route->ksnr_deleted)) {
1174 /* peer/route got closed under me */
1176 warn = "peer/route removed";
1180 if (peer->ksnp_proto == NULL) {
1181 /* Never connected before.
1182 * NB recv_hello may have returned EPROTO to signal my peer
1183 * wants a different protocol than the one I asked for.
1185 LASSERT (list_empty(&peer->ksnp_conns));
1187 peer->ksnp_proto = conn->ksnc_proto;
1188 peer->ksnp_incarnation = incarnation;
1191 if (peer->ksnp_proto != conn->ksnc_proto ||
1192 peer->ksnp_incarnation != incarnation) {
1193 /* Peer rebooted or I've got the wrong protocol version */
1194 ksocknal_close_peer_conns_locked(peer, 0, 0);
1196 peer->ksnp_proto = NULL;
1198 warn = peer->ksnp_incarnation != incarnation ?
1200 "wrong proto version";
1210 warn = "lost conn race";
1213 warn = "retry with different protocol version";
1217 /* Refuse to duplicate an existing connection, unless this is a
1218 * loopback connection */
1219 if (conn->ksnc_ipaddr != conn->ksnc_myipaddr) {
1220 list_for_each(tmp, &peer->ksnp_conns) {
1221 conn2 = list_entry(tmp, ksock_conn_t, ksnc_list);
1223 if (conn2->ksnc_ipaddr != conn->ksnc_ipaddr ||
1224 conn2->ksnc_myipaddr != conn->ksnc_myipaddr ||
1225 conn2->ksnc_type != conn->ksnc_type)
1228 /* Reply on a passive connection attempt so the peer
1229 * realises we're connected. */
1239 /* If the connection created by this route didn't bind to the IP
1240 * address the route connected to, the connection/route matching
1241 * code below probably isn't going to work. */
1243 route->ksnr_ipaddr != conn->ksnc_ipaddr) {
1244 CERROR("Route %s %u.%u.%u.%u connected to %u.%u.%u.%u\n",
1245 libcfs_id2str(peer->ksnp_id),
1246 HIPQUAD(route->ksnr_ipaddr),
1247 HIPQUAD(conn->ksnc_ipaddr));
1250 /* Search for a route corresponding to the new connection and
1251 * create an association. This allows incoming connections created
1252 * by routes in my peer to match my own route entries so I don't
1253 * continually create duplicate routes. */
1254 list_for_each (tmp, &peer->ksnp_routes) {
1255 route = list_entry(tmp, ksock_route_t, ksnr_list);
1257 if (route->ksnr_ipaddr != conn->ksnc_ipaddr)
1260 ksocknal_associate_route_conn_locked(route, conn);
1264 conn->ksnc_peer = peer; /* conn takes my ref on peer */
1265 peer->ksnp_last_alive = cfs_time_current();
1266 peer->ksnp_send_keepalive = 0;
1267 peer->ksnp_error = 0;
1269 sched = ksocknal_choose_scheduler_locked (irq);
1270 sched->kss_nconns++;
1271 conn->ksnc_scheduler = sched;
1273 conn->ksnc_tx_last_post = cfs_time_current();
1274 /* Set the deadline for the outgoing HELLO to drain */
1275 conn->ksnc_tx_bufnob = libcfs_sock_wmem_queued(sock);
1276 conn->ksnc_tx_deadline = cfs_time_shift(*ksocknal_tunables.ksnd_timeout);
1277 cfs_mb(); /* order with adding to peer's conn list */
1279 list_add (&conn->ksnc_list, &peer->ksnp_conns);
1280 ksocknal_conn_addref(conn);
1282 ksocknal_new_packet(conn, 0);
1284 conn->ksnc_zc_capable = ksocknal_lib_zc_capable(conn);
1286 /* Take packets blocking for this connection. */
1287 list_for_each_entry_safe(tx, txtmp, &peer->ksnp_tx_queue, tx_list) {
1288 if (conn->ksnc_proto->pro_match_tx(conn, tx, tx->tx_nonblk) == SOCKNAL_MATCH_NO)
1291 list_del (&tx->tx_list);
1292 ksocknal_queue_tx_locked (tx, conn);
1295 cfs_write_unlock_bh (global_lock);
1297 /* We've now got a new connection. Any errors from here on are just
1298 * like "normal" comms errors and we close the connection normally.
1299 * NB (a) we still have to send the reply HELLO for passive
1301 * (b) normal I/O on the conn is blocked until I setup and call the
1305 ksocknal_lib_bind_irq (irq);
1307 CDEBUG(D_NET, "New conn %s p %d.x %u.%u.%u.%u -> %u.%u.%u.%u/%d"
1308 " incarnation:"LPD64" sched[%d]/%d\n",
1309 libcfs_id2str(peerid), conn->ksnc_proto->pro_version,
1310 HIPQUAD(conn->ksnc_myipaddr), HIPQUAD(conn->ksnc_ipaddr),
1311 conn->ksnc_port, incarnation,
1312 (int)(conn->ksnc_scheduler - ksocknal_data.ksnd_schedulers), irq);
1315 /* additional routes after interface exchange? */
1316 ksocknal_create_routes(peer, conn->ksnc_port,
1317 hello->kshm_ips, hello->kshm_nips);
1319 hello->kshm_nips = ksocknal_select_ips(peer, hello->kshm_ips,
1321 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
1324 LIBCFS_FREE(hello, offsetof(ksock_hello_msg_t,
1325 kshm_ips[LNET_MAX_INTERFACES]));
1327 /* setup the socket AFTER I've received hello (it disables
1328 * SO_LINGER). I might call back to the acceptor who may want
1329 * to send a protocol version response and then close the
1330 * socket; this ensures the socket only tears down after the
1331 * response has been sent. */
1333 rc = ksocknal_lib_setup_sock(sock);
1335 cfs_write_lock_bh(global_lock);
1337 /* NB my callbacks block while I hold ksnd_global_lock */
1338 ksocknal_lib_set_callback(sock, conn);
1341 peer->ksnp_accepting--;
1343 cfs_write_unlock_bh(global_lock);
1346 cfs_write_lock_bh(global_lock);
1347 ksocknal_close_conn_locked(conn, rc);
1348 cfs_write_unlock_bh(global_lock);
1349 } else if (ksocknal_connsock_addref(conn) == 0) {
1350 /* Allow I/O to proceed. */
1351 ksocknal_read_callback(conn);
1352 ksocknal_write_callback(conn);
1353 ksocknal_connsock_decref(conn);
1356 ksocknal_connsock_decref(conn);
1357 ksocknal_conn_decref(conn);
1361 if (!peer->ksnp_closing &&
1362 list_empty (&peer->ksnp_conns) &&
1363 list_empty (&peer->ksnp_routes)) {
1364 list_add(&zombies, &peer->ksnp_tx_queue);
1365 list_del_init(&peer->ksnp_tx_queue);
1366 ksocknal_unlink_peer_locked(peer);
1369 cfs_write_unlock_bh (global_lock);
1373 CERROR("Not creating conn %s type %d: %s\n",
1374 libcfs_id2str(peerid), conn->ksnc_type, warn);
1376 CDEBUG(D_NET, "Not creating conn %s type %d: %s\n",
1377 libcfs_id2str(peerid), conn->ksnc_type, warn);
1382 /* Request retry by replying with CONN_NONE
1383 * ksnc_proto has been set already */
1384 conn->ksnc_type = SOCKLND_CONN_NONE;
1385 hello->kshm_nips = 0;
1386 ksocknal_send_hello(ni, conn, peerid.nid, hello);
1389 cfs_write_lock_bh(global_lock);
1390 peer->ksnp_accepting--;
1391 cfs_write_unlock_bh(global_lock);
1394 ksocknal_txlist_done(ni, &zombies, 1);
1395 ksocknal_peer_decref(peer);
1399 LIBCFS_FREE(hello, offsetof(ksock_hello_msg_t,
1400 kshm_ips[LNET_MAX_INTERFACES]));
1402 LIBCFS_FREE (conn, sizeof(*conn));
1405 libcfs_sock_release(sock);
1410 ksocknal_close_conn_locked (ksock_conn_t *conn, int error)
1412 /* This just does the immmediate housekeeping, and queues the
1413 * connection for the reaper to terminate.
1414 * Caller holds ksnd_global_lock exclusively in irq context */
1415 ksock_peer_t *peer = conn->ksnc_peer;
1416 ksock_route_t *route;
1417 ksock_conn_t *conn2;
1418 struct list_head *tmp;
1420 LASSERT (peer->ksnp_error == 0);
1421 LASSERT (!conn->ksnc_closing);
1422 conn->ksnc_closing = 1;
1424 /* ksnd_deathrow_conns takes over peer's ref */
1425 list_del (&conn->ksnc_list);
1427 route = conn->ksnc_route;
1428 if (route != NULL) {
1429 /* dissociate conn from route... */
1430 LASSERT (!route->ksnr_deleted);
1431 LASSERT ((route->ksnr_connected & (1 << conn->ksnc_type)) != 0);
1434 list_for_each(tmp, &peer->ksnp_conns) {
1435 conn2 = list_entry(tmp, ksock_conn_t, ksnc_list);
1437 if (conn2->ksnc_route == route &&
1438 conn2->ksnc_type == conn->ksnc_type)
1444 route->ksnr_connected &= ~(1 << conn->ksnc_type);
1446 conn->ksnc_route = NULL;
1448 #if 0 /* irrelevent with only eager routes */
1449 list_del (&route->ksnr_list); /* make route least favourite */
1450 list_add_tail (&route->ksnr_list, &peer->ksnp_routes);
1452 ksocknal_route_decref(route); /* drop conn's ref on route */
1455 if (list_empty (&peer->ksnp_conns)) {
1456 /* No more connections to this peer */
1458 if (!list_empty(&peer->ksnp_tx_queue)) {
1461 LASSERT (conn->ksnc_proto == &ksocknal_protocol_v3x);
1463 /* throw them to the last connection...,
1464 * these TXs will be send to /dev/null by scheduler */
1465 list_for_each_entry(tx, &peer->ksnp_tx_queue, tx_list)
1466 ksocknal_tx_prep(conn, tx);
1468 spin_lock_bh(&conn->ksnc_scheduler->kss_lock);
1469 list_splice_init(&peer->ksnp_tx_queue, &conn->ksnc_tx_queue);
1470 spin_unlock_bh(&conn->ksnc_scheduler->kss_lock);
1473 peer->ksnp_proto = NULL; /* renegotiate protocol version */
1474 peer->ksnp_error = error; /* stash last conn close reason */
1476 if (list_empty (&peer->ksnp_routes)) {
1477 /* I've just closed last conn belonging to a
1478 * peer with no routes to it */
1479 ksocknal_unlink_peer_locked (peer);
1483 cfs_spin_lock_bh (&ksocknal_data.ksnd_reaper_lock);
1485 list_add_tail (&conn->ksnc_list, &ksocknal_data.ksnd_deathrow_conns);
1486 cfs_waitq_signal (&ksocknal_data.ksnd_reaper_waitq);
1488 cfs_spin_unlock_bh (&ksocknal_data.ksnd_reaper_lock);
1492 ksocknal_peer_failed (ksock_peer_t *peer)
1494 time_t last_alive = 0;
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. */
1501 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
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) {
1508 last_alive = (time_t) (cfs_time_current_sec() -
1509 cfs_duration_sec(cfs_time_current() -
1510 peer->ksnp_last_alive));
1513 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
1516 lnet_notify (peer->ksnp_ni, peer->ksnp_id.nid, 0,
1521 ksocknal_finalize_zcreq(ksock_conn_t *conn)
1523 ksock_peer_t *peer = conn->ksnc_peer;
1526 CFS_LIST_HEAD (zlist);
1528 /* NB safe to finalize TXs because closing of socket will
1529 * abort all buffered data */
1530 LASSERT (conn->ksnc_sock == NULL);
1532 cfs_spin_lock(&peer->ksnp_lock);
1534 cfs_list_for_each_entry_safe_typed(tx, tmp, &peer->ksnp_zc_req_list,
1535 ksock_tx_t, tx_zc_list) {
1536 if (tx->tx_conn != conn)
1539 LASSERT (tx->tx_msg.ksm_zc_cookies[0] != 0);
1541 tx->tx_msg.ksm_zc_cookies[0] = 0;
1542 list_del(&tx->tx_zc_list);
1543 list_add(&tx->tx_zc_list, &zlist);
1546 cfs_spin_unlock(&peer->ksnp_lock);
1548 while (!list_empty(&zlist)) {
1549 tx = list_entry(zlist.next, ksock_tx_t, tx_zc_list);
1551 list_del(&tx->tx_zc_list);
1552 ksocknal_tx_decref(tx);
1557 ksocknal_terminate_conn (ksock_conn_t *conn)
1559 /* This gets called by the reaper (guaranteed thread context) to
1560 * disengage the socket from its callbacks and close it.
1561 * ksnc_refcount will eventually hit zero, and then the reaper will
1563 ksock_peer_t *peer = conn->ksnc_peer;
1564 ksock_sched_t *sched = conn->ksnc_scheduler;
1567 LASSERT(conn->ksnc_closing);
1569 /* wake up the scheduler to "send" all remaining packets to /dev/null */
1570 cfs_spin_lock_bh (&sched->kss_lock);
1572 /* a closing conn is always ready to tx */
1573 conn->ksnc_tx_ready = 1;
1575 if (!conn->ksnc_tx_scheduled &&
1576 !list_empty(&conn->ksnc_tx_queue)){
1577 list_add_tail (&conn->ksnc_tx_list,
1578 &sched->kss_tx_conns);
1579 conn->ksnc_tx_scheduled = 1;
1580 /* extra ref for scheduler */
1581 ksocknal_conn_addref(conn);
1583 cfs_waitq_signal (&sched->kss_waitq);
1586 cfs_spin_unlock_bh (&sched->kss_lock);
1588 /* serialise with callbacks */
1589 cfs_write_lock_bh (&ksocknal_data.ksnd_global_lock);
1591 ksocknal_lib_reset_callback(conn->ksnc_sock, conn);
1593 /* OK, so this conn may not be completely disengaged from its
1594 * scheduler yet, but it _has_ committed to terminate... */
1595 conn->ksnc_scheduler->kss_nconns--;
1597 if (peer->ksnp_error != 0) {
1598 /* peer's last conn closed in error */
1599 LASSERT (list_empty (&peer->ksnp_conns));
1601 peer->ksnp_error = 0; /* avoid multiple notifications */
1604 cfs_write_unlock_bh (&ksocknal_data.ksnd_global_lock);
1607 ksocknal_peer_failed(peer);
1609 /* The socket is closed on the final put; either here, or in
1610 * ksocknal_{send,recv}msg(). Since we set up the linger2 option
1611 * when the connection was established, this will close the socket
1612 * immediately, aborting anything buffered in it. Any hung
1613 * zero-copy transmits will therefore complete in finite time. */
1614 ksocknal_connsock_decref(conn);
1618 ksocknal_queue_zombie_conn (ksock_conn_t *conn)
1620 /* Queue the conn for the reaper to destroy */
1622 LASSERT (cfs_atomic_read(&conn->ksnc_conn_refcount) == 0);
1623 cfs_spin_lock_bh (&ksocknal_data.ksnd_reaper_lock);
1625 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
1626 cfs_waitq_signal(&ksocknal_data.ksnd_reaper_waitq);
1628 cfs_spin_unlock_bh (&ksocknal_data.ksnd_reaper_lock);
1632 ksocknal_destroy_conn (ksock_conn_t *conn)
1634 /* Final coup-de-grace of the reaper */
1635 CDEBUG (D_NET, "connection %p\n", conn);
1637 LASSERT (cfs_atomic_read (&conn->ksnc_conn_refcount) == 0);
1638 LASSERT (cfs_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));
1645 /* complete current receive if any */
1646 switch (conn->ksnc_rx_state) {
1647 case SOCKNAL_RX_LNET_PAYLOAD:
1648 CERROR("Completing partial receive from %s"
1649 ", ip %d.%d.%d.%d:%d, with error\n",
1650 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1651 HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port);
1652 lnet_finalize (conn->ksnc_peer->ksnp_ni,
1653 conn->ksnc_cookie, -EIO);
1655 case SOCKNAL_RX_LNET_HEADER:
1656 if (conn->ksnc_rx_started)
1657 CERROR("Incomplete receive of lnet header from %s"
1658 ", ip %d.%d.%d.%d:%d, with error, protocol: %d.x.\n",
1659 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1660 HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port,
1661 conn->ksnc_proto->pro_version);
1663 case SOCKNAL_RX_KSM_HEADER:
1664 if (conn->ksnc_rx_started)
1665 CERROR("Incomplete receive of ksock message from %s"
1666 ", ip %d.%d.%d.%d:%d, with error, protocol: %d.x.\n",
1667 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1668 HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port,
1669 conn->ksnc_proto->pro_version);
1671 case SOCKNAL_RX_SLOP:
1672 if (conn->ksnc_rx_started)
1673 CERROR("Incomplete receive of slops from %s"
1674 ", ip %d.%d.%d.%d:%d, with error\n",
1675 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1676 HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port);
1683 ksocknal_peer_decref(conn->ksnc_peer);
1685 LIBCFS_FREE (conn, sizeof (*conn));
1689 ksocknal_close_peer_conns_locked (ksock_peer_t *peer, __u32 ipaddr, int why)
1692 struct list_head *ctmp;
1693 struct list_head *cnxt;
1696 list_for_each_safe (ctmp, cnxt, &peer->ksnp_conns) {
1697 conn = list_entry (ctmp, ksock_conn_t, ksnc_list);
1700 conn->ksnc_ipaddr == ipaddr) {
1702 ksocknal_close_conn_locked (conn, why);
1710 ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why)
1712 ksock_peer_t *peer = conn->ksnc_peer;
1713 __u32 ipaddr = conn->ksnc_ipaddr;
1716 cfs_write_lock_bh (&ksocknal_data.ksnd_global_lock);
1718 count = ksocknal_close_peer_conns_locked (peer, ipaddr, why);
1720 cfs_write_unlock_bh (&ksocknal_data.ksnd_global_lock);
1726 ksocknal_close_matching_conns (lnet_process_id_t id, __u32 ipaddr)
1729 struct list_head *ptmp;
1730 struct list_head *pnxt;
1736 cfs_write_lock_bh (&ksocknal_data.ksnd_global_lock);
1738 if (id.nid != LNET_NID_ANY)
1739 lo = hi = (int)(ksocknal_nid2peerlist(id.nid) - ksocknal_data.ksnd_peers);
1742 hi = ksocknal_data.ksnd_peer_hash_size - 1;
1745 for (i = lo; i <= hi; i++) {
1746 list_for_each_safe (ptmp, pnxt, &ksocknal_data.ksnd_peers[i]) {
1748 peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
1750 if (!((id.nid == LNET_NID_ANY || id.nid == peer->ksnp_id.nid) &&
1751 (id.pid == LNET_PID_ANY || id.pid == peer->ksnp_id.pid)))
1754 count += ksocknal_close_peer_conns_locked (peer, ipaddr, 0);
1758 cfs_write_unlock_bh (&ksocknal_data.ksnd_global_lock);
1760 /* wildcards always succeed */
1761 if (id.nid == LNET_NID_ANY || id.pid == LNET_PID_ANY || ipaddr == 0)
1764 return (count == 0 ? -ENOENT : 0);
1768 ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive)
1770 /* The router is telling me she's been notified of a change in
1771 * gateway state.... */
1772 lnet_process_id_t id = {0};
1775 id.pid = LNET_PID_ANY;
1777 CDEBUG (D_NET, "gw %s %s\n", libcfs_nid2str(gw_nid),
1778 alive ? "up" : "down");
1781 /* If the gateway crashed, close all open connections... */
1782 ksocknal_close_matching_conns (id, 0);
1786 /* ...otherwise do nothing. We can only establish new connections
1787 * if we have autroutes, and these connect on demand. */
1791 ksocknal_push_peer (ksock_peer_t *peer)
1795 struct list_head *tmp;
1798 for (index = 0; ; index++) {
1799 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
1804 list_for_each (tmp, &peer->ksnp_conns) {
1806 conn = list_entry (tmp, ksock_conn_t, ksnc_list);
1807 ksocknal_conn_addref(conn);
1812 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
1817 ksocknal_lib_push_conn (conn);
1818 ksocknal_conn_decref(conn);
1823 ksocknal_push (lnet_ni_t *ni, lnet_process_id_t id)
1826 struct list_head *tmp;
1832 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
1833 for (j = 0; ; j++) {
1834 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
1839 list_for_each (tmp, &ksocknal_data.ksnd_peers[i]) {
1840 peer = list_entry(tmp, ksock_peer_t,
1843 if (!((id.nid == LNET_NID_ANY ||
1844 id.nid == peer->ksnp_id.nid) &&
1845 (id.pid == LNET_PID_ANY ||
1846 id.pid == peer->ksnp_id.pid))) {
1852 ksocknal_peer_addref(peer);
1857 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
1861 ksocknal_push_peer (peer);
1862 ksocknal_peer_decref(peer);
1872 ksocknal_add_interface(lnet_ni_t *ni, __u32 ipaddress, __u32 netmask)
1874 ksock_net_t *net = ni->ni_data;
1875 ksock_interface_t *iface;
1879 struct list_head *ptmp;
1881 struct list_head *rtmp;
1882 ksock_route_t *route;
1884 if (ipaddress == 0 ||
1888 cfs_write_lock_bh (&ksocknal_data.ksnd_global_lock);
1890 iface = ksocknal_ip2iface(ni, ipaddress);
1891 if (iface != NULL) {
1892 /* silently ignore dups */
1894 } else if (net->ksnn_ninterfaces == LNET_MAX_INTERFACES) {
1897 iface = &net->ksnn_interfaces[net->ksnn_ninterfaces++];
1899 iface->ksni_ipaddr = ipaddress;
1900 iface->ksni_netmask = netmask;
1901 iface->ksni_nroutes = 0;
1902 iface->ksni_npeers = 0;
1904 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
1905 list_for_each(ptmp, &ksocknal_data.ksnd_peers[i]) {
1906 peer = list_entry(ptmp, ksock_peer_t, ksnp_list);
1908 for (j = 0; j < peer->ksnp_n_passive_ips; j++)
1909 if (peer->ksnp_passive_ips[j] == ipaddress)
1910 iface->ksni_npeers++;
1912 list_for_each(rtmp, &peer->ksnp_routes) {
1913 route = list_entry(rtmp, ksock_route_t, ksnr_list);
1915 if (route->ksnr_myipaddr == ipaddress)
1916 iface->ksni_nroutes++;
1922 /* NB only new connections will pay attention to the new interface! */
1925 cfs_write_unlock_bh (&ksocknal_data.ksnd_global_lock);
1931 ksocknal_peer_del_interface_locked(ksock_peer_t *peer, __u32 ipaddr)
1933 struct list_head *tmp;
1934 struct list_head *nxt;
1935 ksock_route_t *route;
1940 for (i = 0; i < peer->ksnp_n_passive_ips; i++)
1941 if (peer->ksnp_passive_ips[i] == ipaddr) {
1942 for (j = i+1; j < peer->ksnp_n_passive_ips; j++)
1943 peer->ksnp_passive_ips[j-1] =
1944 peer->ksnp_passive_ips[j];
1945 peer->ksnp_n_passive_ips--;
1949 list_for_each_safe(tmp, nxt, &peer->ksnp_routes) {
1950 route = list_entry (tmp, ksock_route_t, ksnr_list);
1952 if (route->ksnr_myipaddr != ipaddr)
1955 if (route->ksnr_share_count != 0) {
1956 /* Manually created; keep, but unbind */
1957 route->ksnr_myipaddr = 0;
1959 ksocknal_del_route_locked(route);
1963 list_for_each_safe(tmp, nxt, &peer->ksnp_conns) {
1964 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
1966 if (conn->ksnc_myipaddr == ipaddr)
1967 ksocknal_close_conn_locked (conn, 0);
1972 ksocknal_del_interface(lnet_ni_t *ni, __u32 ipaddress)
1974 ksock_net_t *net = ni->ni_data;
1976 struct list_head *tmp;
1977 struct list_head *nxt;
1983 cfs_write_lock_bh (&ksocknal_data.ksnd_global_lock);
1985 for (i = 0; i < net->ksnn_ninterfaces; i++) {
1986 this_ip = net->ksnn_interfaces[i].ksni_ipaddr;
1988 if (!(ipaddress == 0 ||
1989 ipaddress == this_ip))
1994 for (j = i+1; j < net->ksnn_ninterfaces; j++)
1995 net->ksnn_interfaces[j-1] =
1996 net->ksnn_interfaces[j];
1998 net->ksnn_ninterfaces--;
2000 for (j = 0; j < ksocknal_data.ksnd_peer_hash_size; j++) {
2001 list_for_each_safe(tmp, nxt, &ksocknal_data.ksnd_peers[j]) {
2002 peer = list_entry(tmp, ksock_peer_t, ksnp_list);
2004 if (peer->ksnp_ni != ni)
2007 ksocknal_peer_del_interface_locked(peer, this_ip);
2012 cfs_write_unlock_bh (&ksocknal_data.ksnd_global_lock);
2018 ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
2020 lnet_process_id_t id = {0};
2021 struct libcfs_ioctl_data *data = arg;
2025 case IOC_LIBCFS_GET_INTERFACE: {
2026 ksock_net_t *net = ni->ni_data;
2027 ksock_interface_t *iface;
2029 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
2031 if (data->ioc_count >= (__u32)net->ksnn_ninterfaces) {
2035 iface = &net->ksnn_interfaces[data->ioc_count];
2037 data->ioc_u32[0] = iface->ksni_ipaddr;
2038 data->ioc_u32[1] = iface->ksni_netmask;
2039 data->ioc_u32[2] = iface->ksni_npeers;
2040 data->ioc_u32[3] = iface->ksni_nroutes;
2043 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
2047 case IOC_LIBCFS_ADD_INTERFACE:
2048 return ksocknal_add_interface(ni,
2049 data->ioc_u32[0], /* IP address */
2050 data->ioc_u32[1]); /* net mask */
2052 case IOC_LIBCFS_DEL_INTERFACE:
2053 return ksocknal_del_interface(ni,
2054 data->ioc_u32[0]); /* IP address */
2056 case IOC_LIBCFS_GET_PEER: {
2061 int share_count = 0;
2063 rc = ksocknal_get_peer_info(ni, data->ioc_count,
2064 &id, &myip, &ip, &port,
2065 &conn_count, &share_count);
2069 data->ioc_nid = id.nid;
2070 data->ioc_count = share_count;
2071 data->ioc_u32[0] = ip;
2072 data->ioc_u32[1] = port;
2073 data->ioc_u32[2] = myip;
2074 data->ioc_u32[3] = conn_count;
2075 data->ioc_u32[4] = id.pid;
2079 case IOC_LIBCFS_ADD_PEER:
2080 id.nid = data->ioc_nid;
2081 id.pid = LUSTRE_SRV_LNET_PID;
2082 return ksocknal_add_peer (ni, id,
2083 data->ioc_u32[0], /* IP */
2084 data->ioc_u32[1]); /* port */
2086 case IOC_LIBCFS_DEL_PEER:
2087 id.nid = data->ioc_nid;
2088 id.pid = LNET_PID_ANY;
2089 return ksocknal_del_peer (ni, id,
2090 data->ioc_u32[0]); /* IP */
2092 case IOC_LIBCFS_GET_CONN: {
2096 ksock_conn_t *conn = ksocknal_get_conn_by_idx (ni, data->ioc_count);
2101 ksocknal_lib_get_conn_tunables(conn, &txmem, &rxmem, &nagle);
2103 data->ioc_count = txmem;
2104 data->ioc_nid = conn->ksnc_peer->ksnp_id.nid;
2105 data->ioc_flags = nagle;
2106 data->ioc_u32[0] = conn->ksnc_ipaddr;
2107 data->ioc_u32[1] = conn->ksnc_port;
2108 data->ioc_u32[2] = conn->ksnc_myipaddr;
2109 data->ioc_u32[3] = conn->ksnc_type;
2110 data->ioc_u32[4] = (__u32)(conn->ksnc_scheduler -
2111 ksocknal_data.ksnd_schedulers);
2112 data->ioc_u32[5] = rxmem;
2113 data->ioc_u32[6] = conn->ksnc_peer->ksnp_id.pid;
2114 ksocknal_conn_decref(conn);
2118 case IOC_LIBCFS_CLOSE_CONNECTION:
2119 id.nid = data->ioc_nid;
2120 id.pid = LNET_PID_ANY;
2121 return ksocknal_close_matching_conns (id,
2124 case IOC_LIBCFS_REGISTER_MYNID:
2125 /* Ignore if this is a noop */
2126 if (data->ioc_nid == ni->ni_nid)
2129 CERROR("obsolete IOC_LIBCFS_REGISTER_MYNID: %s(%s)\n",
2130 libcfs_nid2str(data->ioc_nid),
2131 libcfs_nid2str(ni->ni_nid));
2134 case IOC_LIBCFS_PUSH_CONNECTION:
2135 id.nid = data->ioc_nid;
2136 id.pid = LNET_PID_ANY;
2137 return ksocknal_push(ni, id);
2146 ksocknal_free_buffers (void)
2148 LASSERT (cfs_atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
2150 if (ksocknal_data.ksnd_schedulers != NULL)
2151 LIBCFS_FREE (ksocknal_data.ksnd_schedulers,
2152 sizeof (ksock_sched_t) * ksocknal_data.ksnd_nschedulers);
2154 LIBCFS_FREE (ksocknal_data.ksnd_peers,
2155 sizeof (struct list_head) *
2156 ksocknal_data.ksnd_peer_hash_size);
2158 cfs_spin_lock(&ksocknal_data.ksnd_tx_lock);
2160 if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
2161 struct list_head zlist;
2164 list_add(&zlist, &ksocknal_data.ksnd_idle_noop_txs);
2165 list_del_init(&ksocknal_data.ksnd_idle_noop_txs);
2166 cfs_spin_unlock(&ksocknal_data.ksnd_tx_lock);
2168 while(!list_empty(&zlist)) {
2169 tx = list_entry(zlist.next, ksock_tx_t, tx_list);
2170 list_del(&tx->tx_list);
2171 LIBCFS_FREE(tx, tx->tx_desc_size);
2174 cfs_spin_unlock(&ksocknal_data.ksnd_tx_lock);
2179 ksocknal_base_shutdown (void)
2181 ksock_sched_t *sched;
2184 CDEBUG(D_MALLOC, "before NAL cleanup: kmem %d\n",
2185 cfs_atomic_read (&libcfs_kmemory));
2186 LASSERT (ksocknal_data.ksnd_nnets == 0);
2188 switch (ksocknal_data.ksnd_init) {
2192 case SOCKNAL_INIT_ALL:
2193 case SOCKNAL_INIT_DATA:
2194 LASSERT (ksocknal_data.ksnd_peers != NULL);
2195 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
2196 LASSERT (list_empty (&ksocknal_data.ksnd_peers[i]));
2198 LASSERT (list_empty (&ksocknal_data.ksnd_enomem_conns));
2199 LASSERT (list_empty (&ksocknal_data.ksnd_zombie_conns));
2200 LASSERT (list_empty (&ksocknal_data.ksnd_connd_connreqs));
2201 LASSERT (list_empty (&ksocknal_data.ksnd_connd_routes));
2203 if (ksocknal_data.ksnd_schedulers != NULL)
2204 for (i = 0; i < ksocknal_data.ksnd_nschedulers; i++) {
2205 ksock_sched_t *kss =
2206 &ksocknal_data.ksnd_schedulers[i];
2208 LASSERT (list_empty (&kss->kss_tx_conns));
2209 LASSERT (list_empty (&kss->kss_rx_conns));
2210 LASSERT (list_empty (&kss->kss_zombie_noop_txs));
2211 LASSERT (kss->kss_nconns == 0);
2214 /* flag threads to terminate; wake and wait for them to die */
2215 ksocknal_data.ksnd_shuttingdown = 1;
2216 cfs_waitq_broadcast (&ksocknal_data.ksnd_connd_waitq);
2217 cfs_waitq_broadcast (&ksocknal_data.ksnd_reaper_waitq);
2219 if (ksocknal_data.ksnd_schedulers != NULL)
2220 for (i = 0; i < ksocknal_data.ksnd_nschedulers; i++) {
2221 sched = &ksocknal_data.ksnd_schedulers[i];
2222 cfs_waitq_broadcast(&sched->kss_waitq);
2226 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
2227 while (ksocknal_data.ksnd_nthreads != 0) {
2229 CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
2230 "waiting for %d threads to terminate\n",
2231 ksocknal_data.ksnd_nthreads);
2232 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
2233 cfs_pause(cfs_time_seconds(1));
2234 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
2236 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
2238 ksocknal_free_buffers();
2240 ksocknal_data.ksnd_init = SOCKNAL_INIT_NOTHING;
2244 CDEBUG(D_MALLOC, "after NAL cleanup: kmem %d\n",
2245 cfs_atomic_read (&libcfs_kmemory));
2247 PORTAL_MODULE_UNUSE;
2251 ksocknal_base_startup (void)
2256 LASSERT (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
2257 LASSERT (ksocknal_data.ksnd_nnets == 0);
2259 memset (&ksocknal_data, 0, sizeof (ksocknal_data)); /* zero pointers */
2261 ksocknal_data.ksnd_peer_hash_size = SOCKNAL_PEER_HASH_SIZE;
2262 LIBCFS_ALLOC (ksocknal_data.ksnd_peers,
2263 sizeof (struct list_head) * ksocknal_data.ksnd_peer_hash_size);
2264 if (ksocknal_data.ksnd_peers == NULL)
2267 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++)
2268 CFS_INIT_LIST_HEAD(&ksocknal_data.ksnd_peers[i]);
2270 cfs_rwlock_init(&ksocknal_data.ksnd_global_lock);
2272 cfs_spin_lock_init (&ksocknal_data.ksnd_reaper_lock);
2273 CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_enomem_conns);
2274 CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_zombie_conns);
2275 CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_deathrow_conns);
2276 cfs_waitq_init(&ksocknal_data.ksnd_reaper_waitq);
2278 cfs_spin_lock_init (&ksocknal_data.ksnd_connd_lock);
2279 CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_connd_connreqs);
2280 CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_connd_routes);
2281 cfs_waitq_init(&ksocknal_data.ksnd_connd_waitq);
2283 cfs_spin_lock_init (&ksocknal_data.ksnd_tx_lock);
2284 CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_idle_noop_txs);
2286 /* NB memset above zeros whole of ksocknal_data, including
2287 * ksocknal_data.ksnd_irqinfo[all].ksni_valid */
2289 /* flag lists/ptrs/locks initialised */
2290 ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
2293 ksocknal_data.ksnd_nschedulers = ksocknal_nsched();
2294 LIBCFS_ALLOC(ksocknal_data.ksnd_schedulers,
2295 sizeof(ksock_sched_t) * ksocknal_data.ksnd_nschedulers);
2296 if (ksocknal_data.ksnd_schedulers == NULL)
2299 for (i = 0; i < ksocknal_data.ksnd_nschedulers; i++) {
2300 ksock_sched_t *kss = &ksocknal_data.ksnd_schedulers[i];
2302 cfs_spin_lock_init (&kss->kss_lock);
2303 CFS_INIT_LIST_HEAD (&kss->kss_rx_conns);
2304 CFS_INIT_LIST_HEAD (&kss->kss_tx_conns);
2305 CFS_INIT_LIST_HEAD (&kss->kss_zombie_noop_txs);
2306 cfs_waitq_init (&kss->kss_waitq);
2309 for (i = 0; i < ksocknal_data.ksnd_nschedulers; i++) {
2310 rc = ksocknal_thread_start (ksocknal_scheduler,
2311 &ksocknal_data.ksnd_schedulers[i]);
2313 CERROR("Can't spawn socknal scheduler[%d]: %d\n",
2319 /* must have at least 2 connds to remain responsive to accepts while
2321 if (*ksocknal_tunables.ksnd_nconnds < 2)
2322 *ksocknal_tunables.ksnd_nconnds = 2;
2324 for (i = 0; i < *ksocknal_tunables.ksnd_nconnds; i++) {
2325 rc = ksocknal_thread_start (ksocknal_connd,
2326 (void *)((ulong_ptr_t)i));
2328 CERROR("Can't spawn socknal connd: %d\n", rc);
2333 rc = ksocknal_thread_start (ksocknal_reaper, NULL);
2335 CERROR ("Can't spawn socknal reaper: %d\n", rc);
2339 /* flag everything initialised */
2340 ksocknal_data.ksnd_init = SOCKNAL_INIT_ALL;
2345 ksocknal_base_shutdown();
2350 ksocknal_debug_peerhash (lnet_ni_t *ni)
2352 ksock_peer_t *peer = NULL;
2353 struct list_head *tmp;
2356 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
2358 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
2359 list_for_each (tmp, &ksocknal_data.ksnd_peers[i]) {
2360 peer = list_entry (tmp, ksock_peer_t, ksnp_list);
2362 if (peer->ksnp_ni == ni) break;
2369 ksock_route_t *route;
2372 CWARN ("Active peer on shutdown: %s, ref %d, scnt %d, "
2373 "closing %d, accepting %d, err %d, zcookie "LPU64", "
2374 "txq %d, zc_req %d\n", libcfs_id2str(peer->ksnp_id),
2375 cfs_atomic_read(&peer->ksnp_refcount),
2376 peer->ksnp_sharecount, peer->ksnp_closing,
2377 peer->ksnp_accepting, peer->ksnp_error,
2378 peer->ksnp_zc_next_cookie,
2379 !list_empty(&peer->ksnp_tx_queue),
2380 !list_empty(&peer->ksnp_zc_req_list));
2382 list_for_each (tmp, &peer->ksnp_routes) {
2383 route = list_entry(tmp, ksock_route_t, ksnr_list);
2384 CWARN ("Route: ref %d, schd %d, conn %d, cnted %d, "
2385 "del %d\n", cfs_atomic_read(&route->ksnr_refcount),
2386 route->ksnr_scheduled, route->ksnr_connecting,
2387 route->ksnr_connected, route->ksnr_deleted);
2390 list_for_each (tmp, &peer->ksnp_conns) {
2391 conn = list_entry(tmp, ksock_conn_t, ksnc_list);
2392 CWARN ("Conn: ref %d, sref %d, t %d, c %d\n",
2393 cfs_atomic_read(&conn->ksnc_conn_refcount),
2394 cfs_atomic_read(&conn->ksnc_sock_refcount),
2395 conn->ksnc_type, conn->ksnc_closing);
2399 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
2404 ksocknal_shutdown (lnet_ni_t *ni)
2406 ksock_net_t *net = ni->ni_data;
2408 lnet_process_id_t anyid = {0};
2410 anyid.nid = LNET_NID_ANY;
2411 anyid.pid = LNET_PID_ANY;
2413 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL);
2414 LASSERT(ksocknal_data.ksnd_nnets > 0);
2416 cfs_spin_lock_bh (&net->ksnn_lock);
2417 net->ksnn_shutdown = 1; /* prevent new peers */
2418 cfs_spin_unlock_bh (&net->ksnn_lock);
2420 /* Delete all peers */
2421 ksocknal_del_peer(ni, anyid, 0);
2423 /* Wait for all peer state to clean up */
2425 cfs_spin_lock_bh (&net->ksnn_lock);
2426 while (net->ksnn_npeers != 0) {
2427 cfs_spin_unlock_bh (&net->ksnn_lock);
2430 CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
2431 "waiting for %d peers to disconnect\n",
2433 cfs_pause(cfs_time_seconds(1));
2435 ksocknal_debug_peerhash(ni);
2437 cfs_spin_lock_bh (&net->ksnn_lock);
2439 cfs_spin_unlock_bh (&net->ksnn_lock);
2441 for (i = 0; i < net->ksnn_ninterfaces; i++) {
2442 LASSERT (net->ksnn_interfaces[i].ksni_npeers == 0);
2443 LASSERT (net->ksnn_interfaces[i].ksni_nroutes == 0);
2446 LIBCFS_FREE(net, sizeof(*net));
2448 ksocknal_data.ksnd_nnets--;
2449 if (ksocknal_data.ksnd_nnets == 0)
2450 ksocknal_base_shutdown();
2454 ksocknal_enumerate_interfaces(ksock_net_t *net)
2462 n = libcfs_ipif_enumerate(&names);
2464 CERROR("Can't enumerate interfaces: %d\n", n);
2468 for (i = j = 0; i < n; i++) {
2473 if (!strcmp(names[i], "lo")) /* skip the loopback IF */
2476 rc = libcfs_ipif_query(names[i], &up, &ip, &mask);
2478 CWARN("Can't get interface %s info: %d\n",
2484 CWARN("Ignoring interface %s (down)\n",
2489 if (j == LNET_MAX_INTERFACES) {
2490 CWARN("Ignoring interface %s (too many interfaces)\n",
2495 net->ksnn_interfaces[j].ksni_ipaddr = ip;
2496 net->ksnn_interfaces[j].ksni_netmask = mask;
2500 libcfs_ipif_free_enumeration(names, n);
2503 CERROR("Can't find any usable interfaces\n");
2509 ksocknal_startup (lnet_ni_t *ni)
2515 LASSERT (ni->ni_lnd == &the_ksocklnd);
2517 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING) {
2518 rc = ksocknal_base_startup();
2523 LIBCFS_ALLOC(net, sizeof(*net));
2527 memset(net, 0, sizeof(*net));
2528 cfs_spin_lock_init(&net->ksnn_lock);
2529 net->ksnn_incarnation = ksocknal_lib_new_incarnation();
2531 ni->ni_maxtxcredits = *ksocknal_tunables.ksnd_credits;
2532 ni->ni_peertxcredits = *ksocknal_tunables.ksnd_peercredits;
2534 if (ni->ni_interfaces[0] == NULL) {
2535 rc = ksocknal_enumerate_interfaces(net);
2539 net->ksnn_ninterfaces = 1;
2541 for (i = 0; i < LNET_MAX_INTERFACES; i++) {
2544 if (ni->ni_interfaces[i] == NULL)
2547 rc = libcfs_ipif_query(
2548 ni->ni_interfaces[i], &up,
2549 &net->ksnn_interfaces[i].ksni_ipaddr,
2550 &net->ksnn_interfaces[i].ksni_netmask);
2553 CERROR("Can't get interface %s info: %d\n",
2554 ni->ni_interfaces[i], rc);
2559 CERROR("Interface %s is down\n",
2560 ni->ni_interfaces[i]);
2564 net->ksnn_ninterfaces = i;
2567 ni->ni_nid = LNET_MKNID(LNET_NIDNET(ni->ni_nid),
2568 net->ksnn_interfaces[0].ksni_ipaddr);
2570 ksocknal_data.ksnd_nnets++;
2575 LIBCFS_FREE(net, sizeof(*net));
2577 if (ksocknal_data.ksnd_nnets == 0)
2578 ksocknal_base_shutdown();
2585 ksocknal_module_fini (void)
2587 lnet_unregister_lnd(&the_ksocklnd);
2588 ksocknal_tunables_fini();
2592 ksocknal_module_init (void)
2596 /* check ksnr_connected/connecting field large enough */
2597 CLASSERT (SOCKLND_CONN_NTYPES <= 4);
2598 CLASSERT (SOCKLND_CONN_ACK == SOCKLND_CONN_BULK_IN);
2600 /* initialize the_ksocklnd */
2601 the_ksocklnd.lnd_type = SOCKLND;
2602 the_ksocklnd.lnd_startup = ksocknal_startup;
2603 the_ksocklnd.lnd_shutdown = ksocknal_shutdown;
2604 the_ksocklnd.lnd_ctl = ksocknal_ctl;
2605 the_ksocklnd.lnd_send = ksocknal_send;
2606 the_ksocklnd.lnd_recv = ksocknal_recv;
2607 the_ksocklnd.lnd_notify = ksocknal_notify;
2608 the_ksocklnd.lnd_accept = ksocknal_accept;
2610 rc = ksocknal_tunables_init();
2614 lnet_register_lnd(&the_ksocklnd);
2619 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2620 MODULE_DESCRIPTION("Kernel TCP Socket LND v3.0.0");
2621 MODULE_LICENSE("GPL");
2623 cfs_module(ksocknal, "3.0.0", ksocknal_module_init, ksocknal_module_fini);