4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lnet/klnds/socklnd/socklnd.c
33 * Author: Zach Brown <zab@zabbo.net>
34 * Author: Peter J. Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Eric Barton <eric@bartonsoftware.com>
39 #include <linux/inetdevice.h>
41 #include <linux/sunrpc/addr.h>
43 static const struct lnet_lnd the_ksocklnd;
44 struct ksock_nal_data ksocknal_data;
46 static struct ksock_interface *
47 ksocknal_ip2iface(struct lnet_ni *ni, struct sockaddr *addr)
49 struct ksock_net *net = ni->ni_data;
50 struct ksock_interface *iface;
52 iface = &net->ksnn_interface;
54 if (rpc_cmp_addr((struct sockaddr *)&iface->ksni_addr, addr))
60 static struct ksock_interface *
61 ksocknal_index2iface(struct lnet_ni *ni, int index)
63 struct ksock_net *net = ni->ni_data;
64 struct ksock_interface *iface;
66 iface = &net->ksnn_interface;
68 if (iface->ksni_index == index)
74 static int ksocknal_ip2index(struct sockaddr *addr, struct lnet_ni *ni)
76 struct net_device *dev;
78 DECLARE_CONST_IN_IFADDR(ifa);
80 if (addr->sa_family != AF_INET)
81 /* No IPv6 support yet */
85 for_each_netdev(ni->ni_net_ns, dev) {
86 int flags = dev_get_flags(dev);
87 struct in_device *in_dev;
89 if (flags & IFF_LOOPBACK) /* skip the loopback IF */
92 if (!(flags & IFF_UP))
95 in_dev = __in_dev_get_rcu(dev);
99 in_dev_for_each_ifa_rcu(ifa, in_dev) {
100 if (ifa->ifa_local ==
101 ((struct sockaddr_in *)addr)->sin_addr.s_addr)
113 static struct ksock_conn_cb *
114 ksocknal_create_conn_cb(struct sockaddr *addr)
116 struct ksock_conn_cb *conn_cb;
118 LIBCFS_ALLOC(conn_cb, sizeof(*conn_cb));
122 refcount_set(&conn_cb->ksnr_refcount, 1);
123 conn_cb->ksnr_peer = NULL;
124 conn_cb->ksnr_retry_interval = 0; /* OK to connect at any time */
125 rpc_copy_addr((struct sockaddr *)&conn_cb->ksnr_addr, addr);
126 rpc_set_port((struct sockaddr *)&conn_cb->ksnr_addr,
128 conn_cb->ksnr_myiface = -1;
129 conn_cb->ksnr_scheduled = 0;
130 conn_cb->ksnr_connecting = 0;
131 conn_cb->ksnr_connected = 0;
132 conn_cb->ksnr_deleted = 0;
133 conn_cb->ksnr_conn_count = 0;
134 conn_cb->ksnr_ctrl_conn_count = 0;
135 conn_cb->ksnr_blki_conn_count = 0;
136 conn_cb->ksnr_blko_conn_count = 0;
142 ksocknal_destroy_conn_cb(struct ksock_conn_cb *conn_cb)
144 LASSERT(refcount_read(&conn_cb->ksnr_refcount) == 0);
146 if (conn_cb->ksnr_peer)
147 ksocknal_peer_decref(conn_cb->ksnr_peer);
149 LIBCFS_FREE(conn_cb, sizeof(*conn_cb));
152 static struct ksock_peer_ni *
153 ksocknal_create_peer(struct lnet_ni *ni, struct lnet_process_id id)
155 int cpt = lnet_cpt_of_nid(id.nid, ni);
156 struct ksock_net *net = ni->ni_data;
157 struct ksock_peer_ni *peer_ni;
159 LASSERT(id.nid != LNET_NID_ANY);
160 LASSERT(id.pid != LNET_PID_ANY);
161 LASSERT(!in_interrupt());
163 if (!atomic_inc_unless_negative(&net->ksnn_npeers)) {
164 CERROR("Can't create peer_ni: network shutdown\n");
165 return ERR_PTR(-ESHUTDOWN);
168 LIBCFS_CPT_ALLOC(peer_ni, lnet_cpt_table(), cpt, sizeof(*peer_ni));
170 atomic_dec(&net->ksnn_npeers);
171 return ERR_PTR(-ENOMEM);
174 peer_ni->ksnp_ni = ni;
175 peer_ni->ksnp_id = id;
176 refcount_set(&peer_ni->ksnp_refcount, 1); /* 1 ref for caller */
177 peer_ni->ksnp_closing = 0;
178 peer_ni->ksnp_accepting = 0;
179 peer_ni->ksnp_proto = NULL;
180 peer_ni->ksnp_last_alive = 0;
181 peer_ni->ksnp_zc_next_cookie = SOCKNAL_KEEPALIVE_PING + 1;
182 peer_ni->ksnp_conn_cb = NULL;
184 INIT_LIST_HEAD(&peer_ni->ksnp_conns);
185 INIT_LIST_HEAD(&peer_ni->ksnp_tx_queue);
186 INIT_LIST_HEAD(&peer_ni->ksnp_zc_req_list);
187 spin_lock_init(&peer_ni->ksnp_lock);
193 ksocknal_destroy_peer(struct ksock_peer_ni *peer_ni)
195 struct ksock_net *net = peer_ni->ksnp_ni->ni_data;
197 CDEBUG (D_NET, "peer_ni %s %p deleted\n",
198 libcfs_id2str(peer_ni->ksnp_id), peer_ni);
200 LASSERT(refcount_read(&peer_ni->ksnp_refcount) == 0);
201 LASSERT(peer_ni->ksnp_accepting == 0);
202 LASSERT(list_empty(&peer_ni->ksnp_conns));
203 LASSERT(peer_ni->ksnp_conn_cb == NULL);
204 LASSERT(list_empty(&peer_ni->ksnp_tx_queue));
205 LASSERT(list_empty(&peer_ni->ksnp_zc_req_list));
207 LIBCFS_FREE(peer_ni, sizeof(*peer_ni));
209 /* NB a peer_ni's connections and conn_cb keep a reference on their
210 * peer_ni until they are destroyed, so we can be assured that _all_
211 * state to do with this peer_ni has been cleaned up when its refcount
214 if (atomic_dec_and_test(&net->ksnn_npeers))
215 wake_up_var(&net->ksnn_npeers);
218 struct ksock_peer_ni *
219 ksocknal_find_peer_locked(struct lnet_ni *ni, struct lnet_process_id id)
221 struct ksock_peer_ni *peer_ni;
223 hash_for_each_possible(ksocknal_data.ksnd_peers, peer_ni,
225 LASSERT(!peer_ni->ksnp_closing);
227 if (peer_ni->ksnp_ni != ni)
230 if (peer_ni->ksnp_id.nid != id.nid ||
231 peer_ni->ksnp_id.pid != id.pid)
234 CDEBUG(D_NET, "got peer_ni [%p] -> %s (%d)\n",
235 peer_ni, libcfs_id2str(id),
236 refcount_read(&peer_ni->ksnp_refcount));
242 struct ksock_peer_ni *
243 ksocknal_find_peer(struct lnet_ni *ni, struct lnet_process_id id)
245 struct ksock_peer_ni *peer_ni;
247 read_lock(&ksocknal_data.ksnd_global_lock);
248 peer_ni = ksocknal_find_peer_locked(ni, id);
249 if (peer_ni != NULL) /* +1 ref for caller? */
250 ksocknal_peer_addref(peer_ni);
251 read_unlock(&ksocknal_data.ksnd_global_lock);
257 ksocknal_unlink_peer_locked(struct ksock_peer_ni *peer_ni)
260 struct ksock_interface *iface;
262 for (i = 0; i < peer_ni->ksnp_n_passive_ips; i++) {
263 struct sockaddr_in sa = { .sin_family = AF_INET };
264 LASSERT(i < LNET_INTERFACES_NUM);
265 sa.sin_addr.s_addr = htonl(peer_ni->ksnp_passive_ips[i]);
267 iface = ksocknal_ip2iface(peer_ni->ksnp_ni,
268 (struct sockaddr *)&sa);
270 * All IPs in peer_ni->ksnp_passive_ips[] come from the
271 * interface list, therefore the call must succeed.
273 LASSERT(iface != NULL);
275 CDEBUG(D_NET, "peer_ni=%p iface=%p ksni_nroutes=%d\n",
276 peer_ni, iface, iface->ksni_nroutes);
277 iface->ksni_npeers--;
280 LASSERT(list_empty(&peer_ni->ksnp_conns));
281 LASSERT(peer_ni->ksnp_conn_cb == NULL);
282 LASSERT(!peer_ni->ksnp_closing);
283 peer_ni->ksnp_closing = 1;
284 hlist_del(&peer_ni->ksnp_list);
285 /* lose peerlist's ref */
286 ksocknal_peer_decref(peer_ni);
290 ksocknal_get_peer_info(struct lnet_ni *ni, int index,
291 struct lnet_process_id *id, __u32 *myip, __u32 *peer_ip,
292 int *port, int *conn_count, int *share_count)
294 struct ksock_peer_ni *peer_ni;
295 struct ksock_conn_cb *conn_cb;
300 read_lock(&ksocknal_data.ksnd_global_lock);
302 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
304 if (peer_ni->ksnp_ni != ni)
307 if (peer_ni->ksnp_n_passive_ips == 0 &&
308 peer_ni->ksnp_conn_cb == NULL) {
312 *id = peer_ni->ksnp_id;
322 for (j = 0; j < peer_ni->ksnp_n_passive_ips; j++) {
326 *id = peer_ni->ksnp_id;
327 *myip = peer_ni->ksnp_passive_ips[j];
336 if (peer_ni->ksnp_conn_cb) {
340 conn_cb = peer_ni->ksnp_conn_cb;
342 *id = peer_ni->ksnp_id;
343 if (conn_cb->ksnr_addr.ss_family == AF_INET) {
344 struct sockaddr_in *sa =
345 (void *)&conn_cb->ksnr_addr;
347 rc = choose_ipv4_src(myip,
348 conn_cb->ksnr_myiface,
349 ntohl(sa->sin_addr.s_addr),
351 *peer_ip = ntohl(sa->sin_addr.s_addr);
352 *port = ntohs(sa->sin_port);
355 *peer_ip = 0xFFFFFFFF;
359 *conn_count = conn_cb->ksnr_conn_count;
365 read_unlock(&ksocknal_data.ksnd_global_lock);
370 ksocknal_get_conn_count_by_type(struct ksock_conn_cb *conn_cb,
373 unsigned int count = 0;
376 case SOCKLND_CONN_CONTROL:
377 count = conn_cb->ksnr_ctrl_conn_count;
379 case SOCKLND_CONN_BULK_IN:
380 count = conn_cb->ksnr_blki_conn_count;
382 case SOCKLND_CONN_BULK_OUT:
383 count = conn_cb->ksnr_blko_conn_count;
385 case SOCKLND_CONN_ANY:
386 count = conn_cb->ksnr_conn_count;
397 ksocknal_incr_conn_count(struct ksock_conn_cb *conn_cb,
400 conn_cb->ksnr_conn_count++;
402 /* check if all connections of the given type got created */
404 case SOCKLND_CONN_CONTROL:
405 conn_cb->ksnr_ctrl_conn_count++;
406 /* there's a single control connection per peer */
407 conn_cb->ksnr_connected |= BIT(type);
409 case SOCKLND_CONN_BULK_IN:
410 conn_cb->ksnr_blki_conn_count++;
411 if (conn_cb->ksnr_blki_conn_count >=
412 *ksocknal_tunables.ksnd_conns_per_peer)
413 conn_cb->ksnr_connected |= BIT(type);
415 case SOCKLND_CONN_BULK_OUT:
416 conn_cb->ksnr_blko_conn_count++;
417 if (conn_cb->ksnr_blko_conn_count >=
418 *ksocknal_tunables.ksnd_conns_per_peer)
419 conn_cb->ksnr_connected |= BIT(type);
421 case SOCKLND_CONN_ANY:
422 if (conn_cb->ksnr_conn_count >=
423 *ksocknal_tunables.ksnd_conns_per_peer)
424 conn_cb->ksnr_connected |= BIT(type);
432 CDEBUG(D_NET, "Add conn type %d, ksnr_connected %x conns_per_peer %d\n",
433 type, conn_cb->ksnr_connected, *ksocknal_tunables.ksnd_conns_per_peer);
437 ksocknal_associate_cb_conn_locked(struct ksock_conn_cb *conn_cb,
438 struct ksock_conn *conn)
440 struct ksock_peer_ni *peer_ni = conn_cb->ksnr_peer;
441 int type = conn->ksnc_type;
442 struct ksock_interface *iface;
445 conn_iface = ksocknal_ip2index((struct sockaddr *)&conn->ksnc_myaddr,
447 conn->ksnc_conn_cb = conn_cb;
448 ksocknal_conn_cb_addref(conn_cb);
450 if (conn_cb->ksnr_myiface != conn_iface) {
451 if (conn_cb->ksnr_myiface < 0) {
452 /* route wasn't bound locally yet (the initial route) */
453 CDEBUG(D_NET, "Binding %s %pIS to interface %d\n",
454 libcfs_id2str(peer_ni->ksnp_id),
459 "Rebinding %s %pIS from interface %d to %d\n",
460 libcfs_id2str(peer_ni->ksnp_id),
462 conn_cb->ksnr_myiface,
465 iface = ksocknal_index2iface(peer_ni->ksnp_ni,
466 conn_cb->ksnr_myiface);
468 iface->ksni_nroutes--;
470 conn_cb->ksnr_myiface = conn_iface;
471 iface = ksocknal_index2iface(peer_ni->ksnp_ni,
472 conn_cb->ksnr_myiface);
474 iface->ksni_nroutes++;
477 ksocknal_incr_conn_count(conn_cb, type);
479 /* Successful connection => further attempts can
480 * proceed immediately
482 conn_cb->ksnr_retry_interval = 0;
486 ksocknal_add_conn_cb_locked(struct ksock_peer_ni *peer_ni,
487 struct ksock_conn_cb *conn_cb)
489 struct list_head *tmp;
490 struct ksock_conn *conn;
491 struct ksock_net *net = peer_ni->ksnp_ni->ni_data;
493 LASSERT(!peer_ni->ksnp_closing);
494 LASSERT(!conn_cb->ksnr_peer);
495 LASSERT(!conn_cb->ksnr_scheduled);
496 LASSERT(!conn_cb->ksnr_connecting);
497 LASSERT(conn_cb->ksnr_connected == 0);
499 conn_cb->ksnr_peer = peer_ni;
500 ksocknal_peer_addref(peer_ni);
502 /* set the conn_cb's interface to the current net's interface */
503 conn_cb->ksnr_myiface = net->ksnn_interface.ksni_index;
504 net->ksnn_interface.ksni_nroutes++;
506 /* peer_ni's route list takes over my ref on 'route' */
507 peer_ni->ksnp_conn_cb = conn_cb;
509 list_for_each(tmp, &peer_ni->ksnp_conns) {
510 conn = list_entry(tmp, struct ksock_conn, ksnc_list);
512 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
513 (struct sockaddr *)&conn_cb->ksnr_addr))
516 ksocknal_associate_cb_conn_locked(conn_cb, conn);
517 /* keep going (typed conns) */
522 ksocknal_del_conn_cb_locked(struct ksock_conn_cb *conn_cb)
524 struct ksock_peer_ni *peer_ni = conn_cb->ksnr_peer;
525 struct ksock_interface *iface;
526 struct ksock_conn *conn;
527 struct ksock_conn *cnxt;
529 LASSERT(!conn_cb->ksnr_deleted);
531 /* Close associated conns */
532 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
533 if (conn->ksnc_conn_cb != conn_cb)
536 ksocknal_close_conn_locked(conn, 0);
539 if (conn_cb->ksnr_myiface >= 0) {
540 iface = ksocknal_index2iface(peer_ni->ksnp_ni,
541 conn_cb->ksnr_myiface);
543 iface->ksni_nroutes--;
546 conn_cb->ksnr_deleted = 1;
547 ksocknal_conn_cb_decref(conn_cb); /* drop peer_ni's ref */
548 peer_ni->ksnp_conn_cb = NULL;
550 if (list_empty(&peer_ni->ksnp_conns)) {
551 /* I've just removed the last route to a peer_ni with no active
554 ksocknal_unlink_peer_locked(peer_ni);
559 ksocknal_add_peer(struct lnet_ni *ni, struct lnet_process_id id,
560 struct sockaddr *addr)
562 struct ksock_peer_ni *peer_ni;
563 struct ksock_peer_ni *peer2;
564 struct ksock_conn_cb *conn_cb;
566 if (id.nid == LNET_NID_ANY ||
567 id.pid == LNET_PID_ANY)
570 /* Have a brand new peer_ni ready... */
571 peer_ni = ksocknal_create_peer(ni, id);
573 return PTR_ERR(peer_ni);
575 conn_cb = ksocknal_create_conn_cb(addr);
577 ksocknal_peer_decref(peer_ni);
581 write_lock_bh(&ksocknal_data.ksnd_global_lock);
583 /* always called with a ref on ni, so shutdown can't have started */
584 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers)
587 peer2 = ksocknal_find_peer_locked(ni, id);
589 ksocknal_peer_decref(peer_ni);
592 /* peer_ni table takes my ref on peer_ni */
593 hash_add(ksocknal_data.ksnd_peers, &peer_ni->ksnp_list, id.nid);
596 ksocknal_add_conn_cb_locked(peer_ni, conn_cb);
598 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
604 ksocknal_del_peer_locked(struct ksock_peer_ni *peer_ni, __u32 ip)
606 struct ksock_conn *conn;
607 struct ksock_conn *cnxt;
608 struct ksock_conn_cb *conn_cb;
610 LASSERT(!peer_ni->ksnp_closing);
612 /* Extra ref prevents peer_ni disappearing until I'm done with it */
613 ksocknal_peer_addref(peer_ni);
614 conn_cb = peer_ni->ksnp_conn_cb;
616 ksocknal_del_conn_cb_locked(conn_cb);
618 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns,
620 ksocknal_close_conn_locked(conn, 0);
622 ksocknal_peer_decref(peer_ni);
623 /* NB peer_ni unlinks itself when last conn/conn_cb is removed */
627 ksocknal_del_peer(struct lnet_ni *ni, struct lnet_process_id id, __u32 ip)
630 struct hlist_node *pnxt;
631 struct ksock_peer_ni *peer_ni;
637 write_lock_bh(&ksocknal_data.ksnd_global_lock);
639 if (id.nid != LNET_NID_ANY) {
640 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
644 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
647 for (i = lo; i <= hi; i++) {
648 hlist_for_each_entry_safe(peer_ni, pnxt,
649 &ksocknal_data.ksnd_peers[i],
651 if (peer_ni->ksnp_ni != ni)
654 if (!((id.nid == LNET_NID_ANY ||
655 peer_ni->ksnp_id.nid == id.nid) &&
656 (id.pid == LNET_PID_ANY ||
657 peer_ni->ksnp_id.pid == id.pid)))
660 ksocknal_peer_addref(peer_ni); /* a ref for me... */
662 ksocknal_del_peer_locked(peer_ni, ip);
664 if (peer_ni->ksnp_closing &&
665 !list_empty(&peer_ni->ksnp_tx_queue)) {
666 LASSERT(list_empty(&peer_ni->ksnp_conns));
667 LASSERT(peer_ni->ksnp_conn_cb == NULL);
669 list_splice_init(&peer_ni->ksnp_tx_queue,
673 ksocknal_peer_decref(peer_ni); /* ...till here */
675 rc = 0; /* matched! */
679 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
681 ksocknal_txlist_done(ni, &zombies, -ENETDOWN);
686 static struct ksock_conn *
687 ksocknal_get_conn_by_idx(struct lnet_ni *ni, int index)
689 struct ksock_peer_ni *peer_ni;
690 struct ksock_conn *conn;
691 struct list_head *ctmp;
694 read_lock(&ksocknal_data.ksnd_global_lock);
696 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
697 LASSERT(!peer_ni->ksnp_closing);
699 if (peer_ni->ksnp_ni != ni)
702 list_for_each(ctmp, &peer_ni->ksnp_conns) {
706 conn = list_entry(ctmp, struct ksock_conn,
708 ksocknal_conn_addref(conn);
709 read_unlock(&ksocknal_data.ksnd_global_lock);
714 read_unlock(&ksocknal_data.ksnd_global_lock);
718 static struct ksock_sched *
719 ksocknal_choose_scheduler_locked(unsigned int cpt)
721 struct ksock_sched *sched = ksocknal_data.ksnd_schedulers[cpt];
724 if (sched->kss_nthreads == 0) {
725 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
726 if (sched->kss_nthreads > 0) {
727 CDEBUG(D_NET, "scheduler[%d] has no threads. selected scheduler[%d]\n",
728 cpt, sched->kss_cpt);
739 ksocknal_accept(struct lnet_ni *ni, struct socket *sock)
741 struct ksock_connreq *cr;
743 struct sockaddr_storage peer;
745 rc = lnet_sock_getaddr(sock, true, &peer);
746 LASSERT(rc == 0); /* we succeeded before */
748 LIBCFS_ALLOC(cr, sizeof(*cr));
750 LCONSOLE_ERROR_MSG(0x12f,
751 "Dropping connection request from %pIS: memory exhausted\n",
758 cr->ksncr_sock = sock;
760 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
762 list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
763 wake_up(&ksocknal_data.ksnd_connd_waitq);
765 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
770 ksocknal_connecting(struct ksock_conn_cb *conn_cb, struct sockaddr *sa)
773 rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr, sa))
774 return conn_cb->ksnr_connecting;
779 ksocknal_create_conn(struct lnet_ni *ni, struct ksock_conn_cb *conn_cb,
780 struct socket *sock, int type)
782 rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
784 struct lnet_process_id peerid;
785 struct list_head *tmp;
787 struct ksock_conn *conn;
788 struct ksock_conn *conn2;
789 struct ksock_peer_ni *peer_ni = NULL;
790 struct ksock_peer_ni *peer2;
791 struct ksock_sched *sched;
792 struct ksock_hello_msg *hello;
795 struct ksock_tx *txtmp;
802 active = (conn_cb != NULL);
804 LASSERT(active == (type != SOCKLND_CONN_NONE));
806 LIBCFS_ALLOC(conn, sizeof(*conn));
812 conn->ksnc_peer = NULL;
813 conn->ksnc_conn_cb = NULL;
814 conn->ksnc_sock = sock;
815 /* 2 ref, 1 for conn, another extra ref prevents socket
816 * being closed before establishment of connection */
817 refcount_set(&conn->ksnc_sock_refcount, 2);
818 conn->ksnc_type = type;
819 ksocknal_lib_save_callback(sock, conn);
820 refcount_set(&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
822 conn->ksnc_rx_ready = 0;
823 conn->ksnc_rx_scheduled = 0;
825 INIT_LIST_HEAD(&conn->ksnc_tx_queue);
826 conn->ksnc_tx_ready = 0;
827 conn->ksnc_tx_scheduled = 0;
828 conn->ksnc_tx_carrier = NULL;
829 atomic_set (&conn->ksnc_tx_nob, 0);
831 LIBCFS_ALLOC(hello, offsetof(struct ksock_hello_msg,
832 kshm_ips[LNET_INTERFACES_NUM]));
838 /* stash conn's local and remote addrs */
839 rc = ksocknal_lib_get_conn_addrs(conn);
843 /* Find out/confirm peer_ni's NID and connection type and get the
844 * vector of interfaces she's willing to let me connect to.
845 * Passive connections use the listener timeout since the peer_ni sends
850 peer_ni = conn_cb->ksnr_peer;
851 LASSERT(ni == peer_ni->ksnp_ni);
853 /* Active connection sends HELLO eagerly */
854 hello->kshm_nips = 0;
855 peerid = peer_ni->ksnp_id;
857 write_lock_bh(global_lock);
858 conn->ksnc_proto = peer_ni->ksnp_proto;
859 write_unlock_bh(global_lock);
861 if (conn->ksnc_proto == NULL) {
862 conn->ksnc_proto = &ksocknal_protocol_v3x;
863 #if SOCKNAL_VERSION_DEBUG
864 if (*ksocknal_tunables.ksnd_protocol == 2)
865 conn->ksnc_proto = &ksocknal_protocol_v2x;
866 else if (*ksocknal_tunables.ksnd_protocol == 1)
867 conn->ksnc_proto = &ksocknal_protocol_v1x;
871 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
875 peerid.nid = LNET_NID_ANY;
876 peerid.pid = LNET_PID_ANY;
878 /* Passive, get protocol from peer_ni */
879 conn->ksnc_proto = NULL;
882 rc = ksocknal_recv_hello(ni, conn, hello, &peerid, &incarnation);
886 LASSERT(rc == 0 || active);
887 LASSERT(conn->ksnc_proto != NULL);
888 LASSERT(peerid.nid != LNET_NID_ANY);
890 cpt = lnet_cpt_of_nid(peerid.nid, ni);
893 ksocknal_peer_addref(peer_ni);
894 write_lock_bh(global_lock);
896 peer_ni = ksocknal_create_peer(ni, peerid);
897 if (IS_ERR(peer_ni)) {
898 rc = PTR_ERR(peer_ni);
902 write_lock_bh(global_lock);
904 /* called with a ref on ni, so shutdown can't have started */
905 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers) >= 0);
907 peer2 = ksocknal_find_peer_locked(ni, peerid);
909 /* NB this puts an "empty" peer_ni in the peer_ni
910 * table (which takes my ref) */
911 hash_add(ksocknal_data.ksnd_peers,
912 &peer_ni->ksnp_list, peerid.nid);
914 ksocknal_peer_decref(peer_ni);
919 ksocknal_peer_addref(peer_ni);
920 peer_ni->ksnp_accepting++;
922 /* Am I already connecting to this guy? Resolve in
923 * favour of higher NID...
925 if (peerid.nid < ni->ni_nid &&
926 ksocknal_connecting(peer_ni->ksnp_conn_cb,
927 ((struct sockaddr *) &conn->ksnc_peeraddr))) {
929 warn = "connection race resolution";
934 if (peer_ni->ksnp_closing ||
935 (active && conn_cb->ksnr_deleted)) {
936 /* peer_ni/conn_cb got closed under me */
938 warn = "peer_ni/conn_cb removed";
942 if (peer_ni->ksnp_proto == NULL) {
943 /* Never connected before.
944 * NB recv_hello may have returned EPROTO to signal my peer_ni
945 * wants a different protocol than the one I asked for.
947 LASSERT(list_empty(&peer_ni->ksnp_conns));
949 peer_ni->ksnp_proto = conn->ksnc_proto;
950 peer_ni->ksnp_incarnation = incarnation;
953 if (peer_ni->ksnp_proto != conn->ksnc_proto ||
954 peer_ni->ksnp_incarnation != incarnation) {
955 /* peer_ni rebooted or I've got the wrong protocol version */
956 ksocknal_close_peer_conns_locked(peer_ni, NULL, 0);
958 peer_ni->ksnp_proto = NULL;
960 warn = peer_ni->ksnp_incarnation != incarnation ?
962 "wrong proto version";
972 warn = "lost conn race";
975 warn = "retry with different protocol version";
979 /* Refuse to duplicate an existing connection, unless this is a
980 * loopback connection */
981 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
982 (struct sockaddr *)&conn->ksnc_myaddr)) {
983 list_for_each(tmp, &peer_ni->ksnp_conns) {
984 conn2 = list_entry(tmp, struct ksock_conn, ksnc_list);
987 (struct sockaddr *)&conn2->ksnc_peeraddr,
988 (struct sockaddr *)&conn->ksnc_peeraddr) ||
990 (struct sockaddr *)&conn2->ksnc_myaddr,
991 (struct sockaddr *)&conn->ksnc_myaddr) ||
992 conn2->ksnc_type != conn->ksnc_type)
996 if (num_dup < *ksocknal_tunables.ksnd_conns_per_peer)
999 /* Reply on a passive connection attempt so the peer_ni
1000 * realises we're connected.
1010 /* If the connection created by this route didn't bind to the IP
1011 * address the route connected to, the connection/route matching
1012 * code below probably isn't going to work.
1015 !rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr,
1016 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1017 CERROR("Route %s %pIS connected to %pIS\n",
1018 libcfs_id2str(peer_ni->ksnp_id),
1019 &conn_cb->ksnr_addr,
1020 &conn->ksnc_peeraddr);
1023 /* Search for a conn_cb corresponding to the new connection and
1024 * create an association. This allows incoming connections created
1025 * by conn_cbs in my peer_ni to match my own conn_cb entries so I don't
1026 * continually create duplicate conn_cbs.
1028 conn_cb = peer_ni->ksnp_conn_cb;
1030 if (conn_cb && rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
1031 (struct sockaddr *)&conn_cb->ksnr_addr))
1032 ksocknal_associate_cb_conn_locked(conn_cb, conn);
1034 conn->ksnc_peer = peer_ni; /* conn takes my ref on peer_ni */
1035 peer_ni->ksnp_last_alive = ktime_get_seconds();
1036 peer_ni->ksnp_send_keepalive = 0;
1037 peer_ni->ksnp_error = 0;
1039 sched = ksocknal_choose_scheduler_locked(cpt);
1041 CERROR("no schedulers available. node is unhealthy\n");
1045 * The cpt might have changed if we ended up selecting a non cpt
1046 * native scheduler. So use the scheduler's cpt instead.
1048 cpt = sched->kss_cpt;
1049 sched->kss_nconns++;
1050 conn->ksnc_scheduler = sched;
1052 conn->ksnc_tx_last_post = ktime_get_seconds();
1053 /* Set the deadline for the outgoing HELLO to drain */
1054 conn->ksnc_tx_bufnob = sock->sk->sk_wmem_queued;
1055 conn->ksnc_tx_deadline = ktime_get_seconds() +
1057 smp_mb(); /* order with adding to peer_ni's conn list */
1059 list_add(&conn->ksnc_list, &peer_ni->ksnp_conns);
1060 ksocknal_conn_addref(conn);
1062 ksocknal_new_packet(conn, 0);
1064 conn->ksnc_zc_capable = ksocknal_lib_zc_capable(conn);
1066 /* Take packets blocking for this connection. */
1067 list_for_each_entry_safe(tx, txtmp, &peer_ni->ksnp_tx_queue, tx_list) {
1068 if (conn->ksnc_proto->pro_match_tx(conn, tx, tx->tx_nonblk) ==
1072 list_del(&tx->tx_list);
1073 ksocknal_queue_tx_locked(tx, conn);
1076 write_unlock_bh(global_lock);
1078 /* We've now got a new connection. Any errors from here on are just
1079 * like "normal" comms errors and we close the connection normally.
1080 * NB (a) we still have to send the reply HELLO for passive
1082 * (b) normal I/O on the conn is blocked until I setup and call the
1086 CDEBUG(D_NET, "New conn %s p %d.x %pIS -> %pISp"
1087 " incarnation:%lld sched[%d]\n",
1088 libcfs_id2str(peerid), conn->ksnc_proto->pro_version,
1089 &conn->ksnc_myaddr, &conn->ksnc_peeraddr,
1093 hello->kshm_nips = 0;
1094 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
1097 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1098 kshm_ips[LNET_INTERFACES_NUM]));
1100 /* setup the socket AFTER I've received hello (it disables
1101 * SO_LINGER). I might call back to the acceptor who may want
1102 * to send a protocol version response and then close the
1103 * socket; this ensures the socket only tears down after the
1104 * response has been sent.
1107 rc = ksocknal_lib_setup_sock(sock);
1109 write_lock_bh(global_lock);
1111 /* NB my callbacks block while I hold ksnd_global_lock */
1112 ksocknal_lib_set_callback(sock, conn);
1115 peer_ni->ksnp_accepting--;
1117 write_unlock_bh(global_lock);
1120 write_lock_bh(global_lock);
1121 if (!conn->ksnc_closing) {
1122 /* could be closed by another thread */
1123 ksocknal_close_conn_locked(conn, rc);
1125 write_unlock_bh(global_lock);
1126 } else if (ksocknal_connsock_addref(conn) == 0) {
1127 /* Allow I/O to proceed. */
1128 ksocknal_read_callback(conn);
1129 ksocknal_write_callback(conn);
1130 ksocknal_connsock_decref(conn);
1133 ksocknal_connsock_decref(conn);
1134 ksocknal_conn_decref(conn);
1139 if (!peer_ni->ksnp_closing &&
1140 list_empty(&peer_ni->ksnp_conns) &&
1141 peer_ni->ksnp_conn_cb == NULL) {
1142 list_splice_init(&peer_ni->ksnp_tx_queue, &zombies);
1143 ksocknal_unlink_peer_locked(peer_ni);
1146 write_unlock_bh(global_lock);
1150 CERROR("Not creating conn %s type %d: %s\n",
1151 libcfs_id2str(peerid), conn->ksnc_type, warn);
1153 CDEBUG(D_NET, "Not creating conn %s type %d: %s\n",
1154 libcfs_id2str(peerid), conn->ksnc_type, warn);
1159 /* Request retry by replying with CONN_NONE
1160 * ksnc_proto has been set already
1162 conn->ksnc_type = SOCKLND_CONN_NONE;
1163 hello->kshm_nips = 0;
1164 ksocknal_send_hello(ni, conn, peerid.nid, hello);
1167 write_lock_bh(global_lock);
1168 peer_ni->ksnp_accepting--;
1169 write_unlock_bh(global_lock);
1173 * If we get here without an error code, just use -EALREADY.
1174 * Depending on how we got here, the error may be positive
1175 * or negative. Normalize the value for ksocknal_txlist_done().
1177 rc2 = (rc == 0 ? -EALREADY : (rc > 0 ? -rc : rc));
1178 ksocknal_txlist_done(ni, &zombies, rc2);
1179 ksocknal_peer_decref(peer_ni);
1183 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1184 kshm_ips[LNET_INTERFACES_NUM]));
1186 LIBCFS_FREE(conn, sizeof(*conn));
1195 ksocknal_close_conn_locked(struct ksock_conn *conn, int error)
1197 /* This just does the immmediate housekeeping, and queues the
1198 * connection for the reaper to terminate.
1199 * Caller holds ksnd_global_lock exclusively in irq context */
1200 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1201 struct ksock_conn_cb *conn_cb;
1202 struct ksock_conn *conn2;
1203 struct list_head *tmp;
1205 LASSERT(peer_ni->ksnp_error == 0);
1206 LASSERT(!conn->ksnc_closing);
1207 conn->ksnc_closing = 1;
1209 /* ksnd_deathrow_conns takes over peer_ni's ref */
1210 list_del(&conn->ksnc_list);
1212 conn_cb = conn->ksnc_conn_cb;
1213 if (conn_cb != NULL) {
1214 /* dissociate conn from cb... */
1215 LASSERT(!conn_cb->ksnr_deleted);
1217 /* connected bit is set only if all connections
1218 * of the given type got created
1220 if (ksocknal_get_conn_count_by_type(conn_cb, conn->ksnc_type) ==
1221 *ksocknal_tunables.ksnd_conns_per_peer)
1222 LASSERT((conn_cb->ksnr_connected &
1223 BIT(conn->ksnc_type)) != 0);
1226 list_for_each(tmp, &peer_ni->ksnp_conns) {
1227 conn2 = list_entry(tmp, struct ksock_conn, ksnc_list);
1229 if (conn2->ksnc_conn_cb == conn_cb &&
1230 conn2->ksnc_type == conn->ksnc_type)
1236 conn_cb->ksnr_connected &= ~BIT(conn->ksnc_type);
1238 conn->ksnc_conn_cb = NULL;
1240 /* drop conn's ref on conn_cb */
1241 ksocknal_conn_cb_decref(conn_cb);
1244 if (list_empty(&peer_ni->ksnp_conns)) {
1245 /* No more connections to this peer_ni */
1247 if (!list_empty(&peer_ni->ksnp_tx_queue)) {
1248 struct ksock_tx *tx;
1250 LASSERT(conn->ksnc_proto == &ksocknal_protocol_v3x);
1252 /* throw them to the last connection...,
1253 * these TXs will be send to /dev/null by scheduler */
1254 list_for_each_entry(tx, &peer_ni->ksnp_tx_queue,
1256 ksocknal_tx_prep(conn, tx);
1258 spin_lock_bh(&conn->ksnc_scheduler->kss_lock);
1259 list_splice_init(&peer_ni->ksnp_tx_queue,
1260 &conn->ksnc_tx_queue);
1261 spin_unlock_bh(&conn->ksnc_scheduler->kss_lock);
1264 /* renegotiate protocol version */
1265 peer_ni->ksnp_proto = NULL;
1266 /* stash last conn close reason */
1267 peer_ni->ksnp_error = error;
1269 if (peer_ni->ksnp_conn_cb == NULL) {
1270 /* I've just closed last conn belonging to a
1271 * peer_ni with no connections to it
1273 ksocknal_unlink_peer_locked(peer_ni);
1277 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1279 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_deathrow_conns);
1280 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1282 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1286 ksocknal_peer_failed(struct ksock_peer_ni *peer_ni)
1288 bool notify = false;
1289 time64_t last_alive = 0;
1291 /* There has been a connection failure or comms error; but I'll only
1292 * tell LNET I think the peer_ni is dead if it's to another kernel and
1293 * there are no connections or connection attempts in existence. */
1295 read_lock(&ksocknal_data.ksnd_global_lock);
1297 if ((peer_ni->ksnp_id.pid & LNET_PID_USERFLAG) == 0 &&
1298 list_empty(&peer_ni->ksnp_conns) &&
1299 peer_ni->ksnp_accepting == 0 &&
1300 !ksocknal_find_connecting_conn_cb_locked(peer_ni)) {
1302 last_alive = peer_ni->ksnp_last_alive;
1305 read_unlock(&ksocknal_data.ksnd_global_lock);
1308 lnet_notify(peer_ni->ksnp_ni, peer_ni->ksnp_id.nid,
1309 false, false, last_alive);
1313 ksocknal_finalize_zcreq(struct ksock_conn *conn)
1315 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1316 struct ksock_tx *tx;
1317 struct ksock_tx *tmp;
1320 /* NB safe to finalize TXs because closing of socket will
1321 * abort all buffered data */
1322 LASSERT(conn->ksnc_sock == NULL);
1324 spin_lock(&peer_ni->ksnp_lock);
1326 list_for_each_entry_safe(tx, tmp, &peer_ni->ksnp_zc_req_list, tx_zc_list) {
1327 if (tx->tx_conn != conn)
1330 LASSERT(tx->tx_msg.ksm_zc_cookies[0] != 0);
1332 tx->tx_msg.ksm_zc_cookies[0] = 0;
1333 tx->tx_zc_aborted = 1; /* mark it as not-acked */
1334 list_move(&tx->tx_zc_list, &zlist);
1337 spin_unlock(&peer_ni->ksnp_lock);
1339 while (!list_empty(&zlist)) {
1340 tx = list_entry(zlist.next, struct ksock_tx, tx_zc_list);
1342 list_del(&tx->tx_zc_list);
1343 ksocknal_tx_decref(tx);
1348 ksocknal_terminate_conn(struct ksock_conn *conn)
1350 /* This gets called by the reaper (guaranteed thread context) to
1351 * disengage the socket from its callbacks and close it.
1352 * ksnc_refcount will eventually hit zero, and then the reaper will
1355 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1356 struct ksock_sched *sched = conn->ksnc_scheduler;
1357 bool failed = false;
1359 LASSERT(conn->ksnc_closing);
1361 /* wake up the scheduler to "send" all remaining packets to /dev/null */
1362 spin_lock_bh(&sched->kss_lock);
1364 /* a closing conn is always ready to tx */
1365 conn->ksnc_tx_ready = 1;
1367 if (!conn->ksnc_tx_scheduled &&
1368 !list_empty(&conn->ksnc_tx_queue)) {
1369 list_add_tail(&conn->ksnc_tx_list,
1370 &sched->kss_tx_conns);
1371 conn->ksnc_tx_scheduled = 1;
1372 /* extra ref for scheduler */
1373 ksocknal_conn_addref(conn);
1375 wake_up(&sched->kss_waitq);
1378 spin_unlock_bh(&sched->kss_lock);
1380 /* serialise with callbacks */
1381 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1383 ksocknal_lib_reset_callback(conn->ksnc_sock, conn);
1385 /* OK, so this conn may not be completely disengaged from its
1386 * scheduler yet, but it _has_ committed to terminate...
1388 conn->ksnc_scheduler->kss_nconns--;
1390 if (peer_ni->ksnp_error != 0) {
1391 /* peer_ni's last conn closed in error */
1392 LASSERT(list_empty(&peer_ni->ksnp_conns));
1394 peer_ni->ksnp_error = 0; /* avoid multiple notifications */
1397 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1400 ksocknal_peer_failed(peer_ni);
1402 /* The socket is closed on the final put; either here, or in
1403 * ksocknal_{send,recv}msg(). Since we set up the linger2 option
1404 * when the connection was established, this will close the socket
1405 * immediately, aborting anything buffered in it. Any hung
1406 * zero-copy transmits will therefore complete in finite time.
1408 ksocknal_connsock_decref(conn);
1412 ksocknal_queue_zombie_conn(struct ksock_conn *conn)
1414 /* Queue the conn for the reaper to destroy */
1415 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1416 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1418 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
1419 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1421 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1425 ksocknal_destroy_conn(struct ksock_conn *conn)
1429 /* Final coup-de-grace of the reaper */
1430 CDEBUG(D_NET, "connection %p\n", conn);
1432 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1433 LASSERT(refcount_read(&conn->ksnc_sock_refcount) == 0);
1434 LASSERT(conn->ksnc_sock == NULL);
1435 LASSERT(conn->ksnc_conn_cb == NULL);
1436 LASSERT(!conn->ksnc_tx_scheduled);
1437 LASSERT(!conn->ksnc_rx_scheduled);
1438 LASSERT(list_empty(&conn->ksnc_tx_queue));
1440 /* complete current receive if any */
1441 switch (conn->ksnc_rx_state) {
1442 case SOCKNAL_RX_LNET_PAYLOAD:
1443 last_rcv = conn->ksnc_rx_deadline -
1445 CERROR("Completing partial receive from %s[%d], ip %pISp, with error, wanted: %d, left: %d, last alive is %lld secs ago\n",
1446 libcfs_id2str(conn->ksnc_peer->ksnp_id), conn->ksnc_type,
1447 &conn->ksnc_peeraddr,
1448 conn->ksnc_rx_nob_wanted, conn->ksnc_rx_nob_left,
1449 ktime_get_seconds() - last_rcv);
1450 if (conn->ksnc_lnet_msg)
1451 conn->ksnc_lnet_msg->msg_health_status =
1452 LNET_MSG_STATUS_REMOTE_ERROR;
1453 lnet_finalize(conn->ksnc_lnet_msg, -EIO);
1455 case SOCKNAL_RX_LNET_HEADER:
1456 if (conn->ksnc_rx_started)
1457 CERROR("Incomplete receive of lnet header from %s, ip %pISp, with error, protocol: %d.x.\n",
1458 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1459 &conn->ksnc_peeraddr,
1460 conn->ksnc_proto->pro_version);
1462 case SOCKNAL_RX_KSM_HEADER:
1463 if (conn->ksnc_rx_started)
1464 CERROR("Incomplete receive of ksock message from %s, ip %pISp, with error, protocol: %d.x.\n",
1465 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1466 &conn->ksnc_peeraddr,
1467 conn->ksnc_proto->pro_version);
1469 case SOCKNAL_RX_SLOP:
1470 if (conn->ksnc_rx_started)
1471 CERROR("Incomplete receive of slops from %s, ip %pISp, with error\n",
1472 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1473 &conn->ksnc_peeraddr);
1480 ksocknal_peer_decref(conn->ksnc_peer);
1482 LIBCFS_FREE (conn, sizeof (*conn));
1486 ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
1487 struct sockaddr *addr, int why)
1489 struct ksock_conn *conn;
1490 struct ksock_conn *cnxt;
1493 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
1496 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1498 ksocknal_close_conn_locked(conn, why);
1506 ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why)
1508 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1511 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1513 count = ksocknal_close_peer_conns_locked(
1514 peer_ni, (struct sockaddr *)&conn->ksnc_peeraddr, why);
1516 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1522 ksocknal_close_matching_conns(struct lnet_process_id id, __u32 ipaddr)
1524 struct ksock_peer_ni *peer_ni;
1525 struct hlist_node *pnxt;
1530 struct sockaddr_in sa = {.sin_family = AF_INET};
1532 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1534 if (id.nid != LNET_NID_ANY) {
1535 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
1539 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1542 sa.sin_addr.s_addr = htonl(ipaddr);
1543 for (i = lo; i <= hi; i++) {
1544 hlist_for_each_entry_safe(peer_ni, pnxt,
1545 &ksocknal_data.ksnd_peers[i],
1548 if (!((id.nid == LNET_NID_ANY ||
1549 id.nid == peer_ni->ksnp_id.nid) &&
1550 (id.pid == LNET_PID_ANY ||
1551 id.pid == peer_ni->ksnp_id.pid)))
1554 count += ksocknal_close_peer_conns_locked(
1556 ipaddr ? (struct sockaddr *)&sa : NULL, 0);
1560 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1562 /* wildcards always succeed */
1563 if (id.nid == LNET_NID_ANY || id.pid == LNET_PID_ANY || ipaddr == 0)
1566 return (count == 0 ? -ENOENT : 0);
1570 ksocknal_notify_gw_down(lnet_nid_t gw_nid)
1572 /* The router is telling me she's been notified of a change in
1575 struct lnet_process_id id = {
1577 .pid = LNET_PID_ANY,
1580 CDEBUG(D_NET, "gw %s down\n", libcfs_nid2str(gw_nid));
1582 /* If the gateway crashed, close all open connections... */
1583 ksocknal_close_matching_conns(id, 0);
1586 /* We can only establish new connections
1587 * if we have autroutes, and these connect on demand. */
1591 ksocknal_push_peer(struct ksock_peer_ni *peer_ni)
1595 struct list_head *tmp;
1596 struct ksock_conn *conn;
1598 for (index = 0; ; index++) {
1599 read_lock(&ksocknal_data.ksnd_global_lock);
1604 list_for_each(tmp, &peer_ni->ksnp_conns) {
1606 conn = list_entry(tmp, struct ksock_conn,
1608 ksocknal_conn_addref(conn);
1613 read_unlock(&ksocknal_data.ksnd_global_lock);
1618 ksocknal_lib_push_conn (conn);
1619 ksocknal_conn_decref(conn);
1624 ksocknal_push(struct lnet_ni *ni, struct lnet_process_id id)
1631 if (id.nid != LNET_NID_ANY) {
1632 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
1636 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1639 for (bkt = lo; bkt <= hi; bkt++) {
1640 int peer_off; /* searching offset in peer_ni hash table */
1642 for (peer_off = 0; ; peer_off++) {
1643 struct ksock_peer_ni *peer_ni;
1646 read_lock(&ksocknal_data.ksnd_global_lock);
1647 hlist_for_each_entry(peer_ni,
1648 &ksocknal_data.ksnd_peers[bkt],
1650 if (!((id.nid == LNET_NID_ANY ||
1651 id.nid == peer_ni->ksnp_id.nid) &&
1652 (id.pid == LNET_PID_ANY ||
1653 id.pid == peer_ni->ksnp_id.pid)))
1656 if (i++ == peer_off) {
1657 ksocknal_peer_addref(peer_ni);
1661 read_unlock(&ksocknal_data.ksnd_global_lock);
1663 if (i <= peer_off) /* no match */
1667 ksocknal_push_peer(peer_ni);
1668 ksocknal_peer_decref(peer_ni);
1675 ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg)
1677 struct lnet_process_id id = {0};
1678 struct libcfs_ioctl_data *data = arg;
1682 case IOC_LIBCFS_GET_INTERFACE: {
1683 struct ksock_net *net = ni->ni_data;
1684 struct ksock_interface *iface;
1685 struct sockaddr_in *sa;
1687 read_lock(&ksocknal_data.ksnd_global_lock);
1689 if (data->ioc_count >= 1) {
1693 iface = &net->ksnn_interface;
1695 sa = (void *)&iface->ksni_addr;
1696 if (sa->sin_family == AF_INET)
1697 data->ioc_u32[0] = ntohl(sa->sin_addr.s_addr);
1699 data->ioc_u32[0] = 0xFFFFFFFF;
1700 data->ioc_u32[1] = iface->ksni_netmask;
1701 data->ioc_u32[2] = iface->ksni_npeers;
1702 data->ioc_u32[3] = iface->ksni_nroutes;
1705 read_unlock(&ksocknal_data.ksnd_global_lock);
1709 case IOC_LIBCFS_GET_PEER: {
1714 int share_count = 0;
1716 rc = ksocknal_get_peer_info(ni, data->ioc_count,
1717 &id, &myip, &ip, &port,
1718 &conn_count, &share_count);
1722 data->ioc_nid = id.nid;
1723 data->ioc_count = share_count;
1724 data->ioc_u32[0] = ip;
1725 data->ioc_u32[1] = port;
1726 data->ioc_u32[2] = myip;
1727 data->ioc_u32[3] = conn_count;
1728 data->ioc_u32[4] = id.pid;
1732 case IOC_LIBCFS_ADD_PEER: {
1733 struct sockaddr_in sa = {.sin_family = AF_INET};
1735 id.nid = data->ioc_nid;
1736 id.pid = LNET_PID_LUSTRE;
1737 sa.sin_addr.s_addr = htonl(data->ioc_u32[0]);
1738 sa.sin_port = htons(data->ioc_u32[1]);
1739 return ksocknal_add_peer(ni, id, (struct sockaddr *)&sa);
1741 case IOC_LIBCFS_DEL_PEER:
1742 id.nid = data->ioc_nid;
1743 id.pid = LNET_PID_ANY;
1744 return ksocknal_del_peer (ni, id,
1745 data->ioc_u32[0]); /* IP */
1747 case IOC_LIBCFS_GET_CONN: {
1751 struct ksock_conn *conn = ksocknal_get_conn_by_idx(ni, data->ioc_count);
1752 struct sockaddr_in *psa = (void *)&conn->ksnc_peeraddr;
1753 struct sockaddr_in *mysa = (void *)&conn->ksnc_myaddr;
1758 ksocknal_lib_get_conn_tunables(conn, &txmem, &rxmem, &nagle);
1760 data->ioc_count = txmem;
1761 data->ioc_nid = conn->ksnc_peer->ksnp_id.nid;
1762 data->ioc_flags = nagle;
1763 if (psa->sin_family == AF_INET)
1764 data->ioc_u32[0] = ntohl(psa->sin_addr.s_addr);
1766 data->ioc_u32[0] = 0xFFFFFFFF;
1767 data->ioc_u32[1] = rpc_get_port((struct sockaddr *)
1768 &conn->ksnc_peeraddr);
1769 if (mysa->sin_family == AF_INET)
1770 data->ioc_u32[2] = ntohl(mysa->sin_addr.s_addr);
1772 data->ioc_u32[2] = 0xFFFFFFFF;
1773 data->ioc_u32[3] = conn->ksnc_type;
1774 data->ioc_u32[4] = conn->ksnc_scheduler->kss_cpt;
1775 data->ioc_u32[5] = rxmem;
1776 data->ioc_u32[6] = conn->ksnc_peer->ksnp_id.pid;
1777 ksocknal_conn_decref(conn);
1781 case IOC_LIBCFS_CLOSE_CONNECTION:
1782 id.nid = data->ioc_nid;
1783 id.pid = LNET_PID_ANY;
1784 return ksocknal_close_matching_conns (id,
1787 case IOC_LIBCFS_REGISTER_MYNID:
1788 /* Ignore if this is a noop */
1789 if (data->ioc_nid == ni->ni_nid)
1792 CERROR("obsolete IOC_LIBCFS_REGISTER_MYNID: %s(%s)\n",
1793 libcfs_nid2str(data->ioc_nid),
1794 libcfs_nid2str(ni->ni_nid));
1797 case IOC_LIBCFS_PUSH_CONNECTION:
1798 id.nid = data->ioc_nid;
1799 id.pid = LNET_PID_ANY;
1800 return ksocknal_push(ni, id);
1809 ksocknal_free_buffers (void)
1811 LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
1813 if (ksocknal_data.ksnd_schedulers != NULL)
1814 cfs_percpt_free(ksocknal_data.ksnd_schedulers);
1816 spin_lock(&ksocknal_data.ksnd_tx_lock);
1818 if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
1820 struct ksock_tx *tx;
1822 list_splice_init(&ksocknal_data.ksnd_idle_noop_txs, &zlist);
1823 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1825 while (!list_empty(&zlist)) {
1826 tx = list_entry(zlist.next, struct ksock_tx, tx_list);
1827 list_del(&tx->tx_list);
1828 LIBCFS_FREE(tx, tx->tx_desc_size);
1831 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1836 ksocknal_base_shutdown(void)
1838 struct ksock_sched *sched;
1839 struct ksock_peer_ni *peer_ni;
1842 CDEBUG(D_MALLOC, "before NAL cleanup: kmem %lld\n",
1843 libcfs_kmem_read());
1844 LASSERT (ksocknal_data.ksnd_nnets == 0);
1846 switch (ksocknal_data.ksnd_init) {
1851 case SOCKNAL_INIT_ALL:
1852 case SOCKNAL_INIT_DATA:
1853 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list)
1856 LASSERT(list_empty(&ksocknal_data.ksnd_nets));
1857 LASSERT(list_empty(&ksocknal_data.ksnd_enomem_conns));
1858 LASSERT(list_empty(&ksocknal_data.ksnd_zombie_conns));
1859 LASSERT(list_empty(&ksocknal_data.ksnd_connd_connreqs));
1860 LASSERT(list_empty(&ksocknal_data.ksnd_connd_routes));
1862 if (ksocknal_data.ksnd_schedulers != NULL) {
1863 cfs_percpt_for_each(sched, i,
1864 ksocknal_data.ksnd_schedulers) {
1866 LASSERT(list_empty(&sched->kss_tx_conns));
1867 LASSERT(list_empty(&sched->kss_rx_conns));
1868 LASSERT(list_empty(&sched->kss_zombie_noop_txs));
1869 LASSERT(sched->kss_nconns == 0);
1873 /* flag threads to terminate; wake and wait for them to die */
1874 ksocknal_data.ksnd_shuttingdown = 1;
1875 wake_up_all(&ksocknal_data.ksnd_connd_waitq);
1876 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1878 if (ksocknal_data.ksnd_schedulers != NULL) {
1879 cfs_percpt_for_each(sched, i,
1880 ksocknal_data.ksnd_schedulers)
1881 wake_up_all(&sched->kss_waitq);
1884 wait_var_event_warning(&ksocknal_data.ksnd_nthreads,
1885 atomic_read(&ksocknal_data.ksnd_nthreads) == 0,
1886 "waiting for %d threads to terminate\n",
1887 atomic_read(&ksocknal_data.ksnd_nthreads));
1889 ksocknal_free_buffers();
1891 ksocknal_data.ksnd_init = SOCKNAL_INIT_NOTHING;
1895 CDEBUG(D_MALLOC, "after NAL cleanup: kmem %lld\n",
1896 libcfs_kmem_read());
1898 module_put(THIS_MODULE);
1902 ksocknal_base_startup(void)
1904 struct ksock_sched *sched;
1908 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
1909 LASSERT(ksocknal_data.ksnd_nnets == 0);
1911 memset(&ksocknal_data, 0, sizeof(ksocknal_data)); /* zero pointers */
1913 hash_init(ksocknal_data.ksnd_peers);
1915 rwlock_init(&ksocknal_data.ksnd_global_lock);
1916 INIT_LIST_HEAD(&ksocknal_data.ksnd_nets);
1918 spin_lock_init(&ksocknal_data.ksnd_reaper_lock);
1919 INIT_LIST_HEAD(&ksocknal_data.ksnd_enomem_conns);
1920 INIT_LIST_HEAD(&ksocknal_data.ksnd_zombie_conns);
1921 INIT_LIST_HEAD(&ksocknal_data.ksnd_deathrow_conns);
1922 init_waitqueue_head(&ksocknal_data.ksnd_reaper_waitq);
1924 spin_lock_init(&ksocknal_data.ksnd_connd_lock);
1925 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_connreqs);
1926 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_routes);
1927 init_waitqueue_head(&ksocknal_data.ksnd_connd_waitq);
1929 spin_lock_init(&ksocknal_data.ksnd_tx_lock);
1930 INIT_LIST_HEAD(&ksocknal_data.ksnd_idle_noop_txs);
1932 /* NB memset above zeros whole of ksocknal_data */
1934 /* flag lists/ptrs/locks initialised */
1935 ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
1936 if (!try_module_get(THIS_MODULE))
1939 /* Create a scheduler block per available CPT */
1940 ksocknal_data.ksnd_schedulers = cfs_percpt_alloc(lnet_cpt_table(),
1942 if (ksocknal_data.ksnd_schedulers == NULL)
1945 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
1949 * make sure not to allocate more threads than there are
1950 * cores/CPUs in teh CPT
1952 nthrs = cfs_cpt_weight(lnet_cpt_table(), i);
1953 if (*ksocknal_tunables.ksnd_nscheds > 0) {
1954 nthrs = min(nthrs, *ksocknal_tunables.ksnd_nscheds);
1957 * max to half of CPUs, assume another half should be
1958 * reserved for upper layer modules
1960 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
1963 sched->kss_nthreads_max = nthrs;
1966 spin_lock_init(&sched->kss_lock);
1967 INIT_LIST_HEAD(&sched->kss_rx_conns);
1968 INIT_LIST_HEAD(&sched->kss_tx_conns);
1969 INIT_LIST_HEAD(&sched->kss_zombie_noop_txs);
1970 init_waitqueue_head(&sched->kss_waitq);
1973 ksocknal_data.ksnd_connd_starting = 0;
1974 ksocknal_data.ksnd_connd_failed_stamp = 0;
1975 ksocknal_data.ksnd_connd_starting_stamp = ktime_get_real_seconds();
1976 /* must have at least 2 connds to remain responsive to accepts while
1978 if (*ksocknal_tunables.ksnd_nconnds < SOCKNAL_CONND_RESV + 1)
1979 *ksocknal_tunables.ksnd_nconnds = SOCKNAL_CONND_RESV + 1;
1981 if (*ksocknal_tunables.ksnd_nconnds_max <
1982 *ksocknal_tunables.ksnd_nconnds) {
1983 ksocknal_tunables.ksnd_nconnds_max =
1984 ksocknal_tunables.ksnd_nconnds;
1987 for (i = 0; i < *ksocknal_tunables.ksnd_nconnds; i++) {
1989 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
1990 ksocknal_data.ksnd_connd_starting++;
1991 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
1994 snprintf(name, sizeof(name), "socknal_cd%02d", i);
1995 rc = ksocknal_thread_start(ksocknal_connd,
1996 (void *)((uintptr_t)i), name);
1998 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
1999 ksocknal_data.ksnd_connd_starting--;
2000 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2001 CERROR("Can't spawn socknal connd: %d\n", rc);
2006 rc = ksocknal_thread_start(ksocknal_reaper, NULL, "socknal_reaper");
2008 CERROR ("Can't spawn socknal reaper: %d\n", rc);
2012 /* flag everything initialised */
2013 ksocknal_data.ksnd_init = SOCKNAL_INIT_ALL;
2018 ksocknal_base_shutdown();
2023 ksocknal_debug_peerhash(struct lnet_ni *ni)
2025 struct ksock_peer_ni *peer_ni;
2028 read_lock(&ksocknal_data.ksnd_global_lock);
2030 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
2031 struct ksock_conn_cb *conn_cb;
2032 struct ksock_conn *conn;
2034 if (peer_ni->ksnp_ni != ni)
2037 CWARN("Active peer_ni on shutdown: %s, ref %d, "
2038 "closing %d, accepting %d, err %d, zcookie %llu, "
2039 "txq %d, zc_req %d\n", libcfs_id2str(peer_ni->ksnp_id),
2040 refcount_read(&peer_ni->ksnp_refcount),
2041 peer_ni->ksnp_closing,
2042 peer_ni->ksnp_accepting, peer_ni->ksnp_error,
2043 peer_ni->ksnp_zc_next_cookie,
2044 !list_empty(&peer_ni->ksnp_tx_queue),
2045 !list_empty(&peer_ni->ksnp_zc_req_list));
2047 conn_cb = peer_ni->ksnp_conn_cb;
2049 CWARN("ConnCB: ref %d, schd %d, conn %d, cnted %d, del %d\n",
2050 refcount_read(&conn_cb->ksnr_refcount),
2051 conn_cb->ksnr_scheduled, conn_cb->ksnr_connecting,
2052 conn_cb->ksnr_connected, conn_cb->ksnr_deleted);
2055 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
2056 CWARN("Conn: ref %d, sref %d, t %d, c %d\n",
2057 refcount_read(&conn->ksnc_conn_refcount),
2058 refcount_read(&conn->ksnc_sock_refcount),
2059 conn->ksnc_type, conn->ksnc_closing);
2064 read_unlock(&ksocknal_data.ksnd_global_lock);
2069 ksocknal_shutdown(struct lnet_ni *ni)
2071 struct ksock_net *net = ni->ni_data;
2072 struct lnet_process_id anyid = {
2073 .nid = LNET_NID_ANY,
2074 .pid = LNET_PID_ANY,
2077 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL);
2078 LASSERT(ksocknal_data.ksnd_nnets > 0);
2080 /* prevent new peers */
2081 atomic_add(SOCKNAL_SHUTDOWN_BIAS, &net->ksnn_npeers);
2083 /* Delete all peers */
2084 ksocknal_del_peer(ni, anyid, 0);
2086 /* Wait for all peer_ni state to clean up */
2087 wait_var_event_warning(&net->ksnn_npeers,
2088 atomic_read(&net->ksnn_npeers) ==
2089 SOCKNAL_SHUTDOWN_BIAS,
2090 "waiting for %d peers to disconnect\n",
2091 ksocknal_debug_peerhash(ni) +
2092 atomic_read(&net->ksnn_npeers) -
2093 SOCKNAL_SHUTDOWN_BIAS);
2095 LASSERT(net->ksnn_interface.ksni_npeers == 0);
2096 LASSERT(net->ksnn_interface.ksni_nroutes == 0);
2098 list_del(&net->ksnn_list);
2099 LIBCFS_FREE(net, sizeof(*net));
2101 ksocknal_data.ksnd_nnets--;
2102 if (ksocknal_data.ksnd_nnets == 0)
2103 ksocknal_base_shutdown();
2107 ksocknal_search_new_ipif(struct ksock_net *net)
2110 char *ifnam = &net->ksnn_interface.ksni_name[0];
2111 char *colon = strchr(ifnam, ':');
2113 struct ksock_net *tmp;
2118 list_for_each_entry(tmp, &ksocknal_data.ksnd_nets, ksnn_list) {
2119 char *ifnam2 = &tmp->ksnn_interface.ksni_name[0];
2120 char *colon2 = strchr(ifnam2, ':');
2125 found = strcmp(ifnam, ifnam2) == 0;
2138 ksocknal_start_schedulers(struct ksock_sched *sched)
2144 if (sched->kss_nthreads == 0) {
2145 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2146 nthrs = sched->kss_nthreads_max;
2148 nthrs = cfs_cpt_weight(lnet_cpt_table(),
2150 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2151 nthrs = min(SOCKNAL_NSCHEDS_HIGH, nthrs);
2153 nthrs = min(nthrs, sched->kss_nthreads_max);
2155 LASSERT(sched->kss_nthreads <= sched->kss_nthreads_max);
2156 /* increase two threads if there is new interface */
2157 nthrs = min(2, sched->kss_nthreads_max - sched->kss_nthreads);
2160 for (i = 0; i < nthrs; i++) {
2164 id = KSOCK_THREAD_ID(sched->kss_cpt, sched->kss_nthreads + i);
2165 snprintf(name, sizeof(name), "socknal_sd%02d_%02d",
2166 sched->kss_cpt, (int)KSOCK_THREAD_SID(id));
2168 rc = ksocknal_thread_start(ksocknal_scheduler,
2173 CERROR("Can't spawn thread %d for scheduler[%d]: %d\n",
2174 sched->kss_cpt, (int) KSOCK_THREAD_SID(id), rc);
2178 sched->kss_nthreads += i;
2183 ksocknal_net_start_threads(struct ksock_net *net, __u32 *cpts, int ncpts)
2185 int newif = ksocknal_search_new_ipif(net);
2189 if (ncpts > 0 && ncpts > cfs_cpt_number(lnet_cpt_table()))
2192 for (i = 0; i < ncpts; i++) {
2193 struct ksock_sched *sched;
2194 int cpt = (cpts == NULL) ? i : cpts[i];
2196 LASSERT(cpt < cfs_cpt_number(lnet_cpt_table()));
2197 sched = ksocknal_data.ksnd_schedulers[cpt];
2199 if (!newif && sched->kss_nthreads > 0)
2202 rc = ksocknal_start_schedulers(sched);
2210 ksocknal_startup(struct lnet_ni *ni)
2212 struct ksock_net *net;
2213 struct lnet_ioctl_config_lnd_cmn_tunables *net_tunables;
2214 struct ksock_interface *ksi = NULL;
2215 struct lnet_inetdev *ifaces = NULL;
2216 struct sockaddr_in *sa;
2220 LASSERT (ni->ni_net->net_lnd == &the_ksocklnd);
2221 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING) {
2222 rc = ksocknal_base_startup();
2226 LIBCFS_ALLOC(net, sizeof(*net));
2229 net->ksnn_incarnation = ktime_get_real_ns();
2231 net_tunables = &ni->ni_net->net_tunables;
2232 if (net_tunables->lct_peer_timeout == -1)
2233 net_tunables->lct_peer_timeout =
2234 *ksocknal_tunables.ksnd_peertimeout;
2236 if (net_tunables->lct_max_tx_credits == -1)
2237 net_tunables->lct_max_tx_credits =
2238 *ksocknal_tunables.ksnd_credits;
2240 if (net_tunables->lct_peer_tx_credits == -1)
2241 net_tunables->lct_peer_tx_credits =
2242 *ksocknal_tunables.ksnd_peertxcredits;
2244 if (net_tunables->lct_peer_tx_credits >
2245 net_tunables->lct_max_tx_credits)
2246 net_tunables->lct_peer_tx_credits =
2247 net_tunables->lct_max_tx_credits;
2249 if (net_tunables->lct_peer_rtr_credits == -1)
2250 net_tunables->lct_peer_rtr_credits =
2251 *ksocknal_tunables.ksnd_peerrtrcredits;
2253 rc = lnet_inet_enumerate(&ifaces, ni->ni_net_ns);
2257 ksi = &net->ksnn_interface;
2259 /* Use the first discovered interface or look in the list */
2260 if (ni->ni_interface) {
2261 for (i = 0; i < rc; i++)
2262 if (strcmp(ifaces[i].li_name, ni->ni_interface) == 0)
2265 /* ni_interfaces doesn't contain the interface we want */
2267 CERROR("ksocklnd: failed to find interface %s\n",
2273 ni->ni_dev_cpt = ifaces[i].li_cpt;
2274 sa = (void *)&ksi->ksni_addr;
2275 memset(sa, 0, sizeof(*sa));
2276 sa->sin_family = AF_INET;
2277 sa->sin_addr.s_addr = htonl(ifaces[i].li_ipaddr);
2278 ksi->ksni_index = ksocknal_ip2index((struct sockaddr *)sa, ni);
2279 ksi->ksni_netmask = ifaces[i].li_netmask;
2280 strlcpy(ksi->ksni_name, ifaces[i].li_name, sizeof(ksi->ksni_name));
2282 /* call it before add it to ksocknal_data.ksnd_nets */
2283 rc = ksocknal_net_start_threads(net, ni->ni_cpts, ni->ni_ncpts);
2288 LASSERT(ksi->ksni_addr.ss_family == AF_INET);
2289 ni->ni_nid = LNET_MKNID(
2290 LNET_NIDNET(ni->ni_nid),
2291 ntohl(((struct sockaddr_in *)
2292 &ksi->ksni_addr)->sin_addr.s_addr));
2293 list_add(&net->ksnn_list, &ksocknal_data.ksnd_nets);
2294 ksocknal_data.ksnd_nnets++;
2299 LIBCFS_FREE(net, sizeof(*net));
2301 if (ksocknal_data.ksnd_nnets == 0)
2302 ksocknal_base_shutdown();
2308 static void __exit ksocklnd_exit(void)
2310 lnet_unregister_lnd(&the_ksocklnd);
2313 static const struct lnet_lnd the_ksocklnd = {
2314 .lnd_type = SOCKLND,
2315 .lnd_startup = ksocknal_startup,
2316 .lnd_shutdown = ksocknal_shutdown,
2317 .lnd_ctl = ksocknal_ctl,
2318 .lnd_send = ksocknal_send,
2319 .lnd_recv = ksocknal_recv,
2320 .lnd_notify_peer_down = ksocknal_notify_gw_down,
2321 .lnd_accept = ksocknal_accept,
2324 static int __init ksocklnd_init(void)
2328 /* check ksnr_connected/connecting field large enough */
2329 BUILD_BUG_ON(SOCKLND_CONN_NTYPES > 4);
2330 BUILD_BUG_ON(SOCKLND_CONN_ACK != SOCKLND_CONN_BULK_IN);
2332 rc = ksocknal_tunables_init();
2336 lnet_register_lnd(&the_ksocklnd);
2341 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2342 MODULE_DESCRIPTION("TCP Socket LNet Network Driver");
2343 MODULE_VERSION("2.8.0");
2344 MODULE_LICENSE("GPL");
2346 module_init(ksocklnd_init);
2347 module_exit(ksocklnd_exit);