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/ethtool.h>
40 #include <linux/inetdevice.h>
42 #include <linux/sunrpc/addr.h>
44 static const struct lnet_lnd the_ksocklnd;
45 struct ksock_nal_data ksocknal_data;
47 static struct ksock_interface *
48 ksocknal_ip2iface(struct lnet_ni *ni, struct sockaddr *addr)
50 struct ksock_net *net = ni->ni_data;
51 struct ksock_interface *iface;
53 iface = &net->ksnn_interface;
55 if (rpc_cmp_addr((struct sockaddr *)&iface->ksni_addr, addr))
61 static struct ksock_interface *
62 ksocknal_index2iface(struct lnet_ni *ni, int index)
64 struct ksock_net *net = ni->ni_data;
65 struct ksock_interface *iface;
67 iface = &net->ksnn_interface;
69 if (iface->ksni_index == index)
75 static int ksocknal_ip2index(struct sockaddr *addr, struct lnet_ni *ni)
77 struct net_device *dev;
79 DECLARE_CONST_IN_IFADDR(ifa);
81 if (addr->sa_family != AF_INET)
82 /* No IPv6 support yet */
86 for_each_netdev(ni->ni_net_ns, dev) {
87 int flags = dev_get_flags(dev);
88 struct in_device *in_dev;
90 if (flags & IFF_LOOPBACK) /* skip the loopback IF */
93 if (!(flags & IFF_UP))
96 in_dev = __in_dev_get_rcu(dev);
100 in_dev_for_each_ifa_rcu(ifa, in_dev) {
101 if (ifa->ifa_local ==
102 ((struct sockaddr_in *)addr)->sin_addr.s_addr)
114 static struct ksock_conn_cb *
115 ksocknal_create_conn_cb(struct sockaddr *addr)
117 struct ksock_conn_cb *conn_cb;
119 LIBCFS_ALLOC(conn_cb, sizeof(*conn_cb));
123 refcount_set(&conn_cb->ksnr_refcount, 1);
124 conn_cb->ksnr_peer = NULL;
125 conn_cb->ksnr_retry_interval = 0; /* OK to connect at any time */
126 rpc_copy_addr((struct sockaddr *)&conn_cb->ksnr_addr, addr);
127 rpc_set_port((struct sockaddr *)&conn_cb->ksnr_addr,
129 conn_cb->ksnr_myiface = -1;
130 conn_cb->ksnr_scheduled = 0;
131 conn_cb->ksnr_connecting = 0;
132 conn_cb->ksnr_connected = 0;
133 conn_cb->ksnr_deleted = 0;
134 conn_cb->ksnr_conn_count = 0;
135 conn_cb->ksnr_ctrl_conn_count = 0;
136 conn_cb->ksnr_blki_conn_count = 0;
137 conn_cb->ksnr_blko_conn_count = 0;
138 conn_cb->ksnr_max_conns = 0;
144 ksocknal_destroy_conn_cb(struct ksock_conn_cb *conn_cb)
146 LASSERT(refcount_read(&conn_cb->ksnr_refcount) == 0);
148 if (conn_cb->ksnr_peer)
149 ksocknal_peer_decref(conn_cb->ksnr_peer);
151 LIBCFS_FREE(conn_cb, sizeof(*conn_cb));
154 static struct ksock_peer_ni *
155 ksocknal_create_peer(struct lnet_ni *ni, struct lnet_process_id id)
157 int cpt = lnet_cpt_of_nid(id.nid, ni);
158 struct ksock_net *net = ni->ni_data;
159 struct ksock_peer_ni *peer_ni;
161 LASSERT(id.nid != LNET_NID_ANY);
162 LASSERT(id.pid != LNET_PID_ANY);
163 LASSERT(!in_interrupt());
165 if (!atomic_inc_unless_negative(&net->ksnn_npeers)) {
166 CERROR("Can't create peer_ni: network shutdown\n");
167 return ERR_PTR(-ESHUTDOWN);
170 LIBCFS_CPT_ALLOC(peer_ni, lnet_cpt_table(), cpt, sizeof(*peer_ni));
172 atomic_dec(&net->ksnn_npeers);
173 return ERR_PTR(-ENOMEM);
176 peer_ni->ksnp_ni = ni;
177 peer_ni->ksnp_id = id;
178 refcount_set(&peer_ni->ksnp_refcount, 1); /* 1 ref for caller */
179 peer_ni->ksnp_closing = 0;
180 peer_ni->ksnp_accepting = 0;
181 peer_ni->ksnp_proto = NULL;
182 peer_ni->ksnp_last_alive = 0;
183 peer_ni->ksnp_zc_next_cookie = SOCKNAL_KEEPALIVE_PING + 1;
184 peer_ni->ksnp_conn_cb = NULL;
186 INIT_LIST_HEAD(&peer_ni->ksnp_conns);
187 INIT_LIST_HEAD(&peer_ni->ksnp_tx_queue);
188 INIT_LIST_HEAD(&peer_ni->ksnp_zc_req_list);
189 spin_lock_init(&peer_ni->ksnp_lock);
195 ksocknal_destroy_peer(struct ksock_peer_ni *peer_ni)
197 struct ksock_net *net = peer_ni->ksnp_ni->ni_data;
199 CDEBUG (D_NET, "peer_ni %s %p deleted\n",
200 libcfs_id2str(peer_ni->ksnp_id), peer_ni);
202 LASSERT(refcount_read(&peer_ni->ksnp_refcount) == 0);
203 LASSERT(peer_ni->ksnp_accepting == 0);
204 LASSERT(list_empty(&peer_ni->ksnp_conns));
205 LASSERT(peer_ni->ksnp_conn_cb == NULL);
206 LASSERT(list_empty(&peer_ni->ksnp_tx_queue));
207 LASSERT(list_empty(&peer_ni->ksnp_zc_req_list));
209 LIBCFS_FREE(peer_ni, sizeof(*peer_ni));
211 /* NB a peer_ni's connections and conn_cb keep a reference on their
212 * peer_ni until they are destroyed, so we can be assured that _all_
213 * state to do with this peer_ni has been cleaned up when its refcount
216 if (atomic_dec_and_test(&net->ksnn_npeers))
217 wake_up_var(&net->ksnn_npeers);
220 struct ksock_peer_ni *
221 ksocknal_find_peer_locked(struct lnet_ni *ni, struct lnet_process_id id)
223 struct ksock_peer_ni *peer_ni;
225 hash_for_each_possible(ksocknal_data.ksnd_peers, peer_ni,
227 LASSERT(!peer_ni->ksnp_closing);
229 if (peer_ni->ksnp_ni != ni)
232 if (peer_ni->ksnp_id.nid != id.nid ||
233 peer_ni->ksnp_id.pid != id.pid)
236 CDEBUG(D_NET, "got peer_ni [%p] -> %s (%d)\n",
237 peer_ni, libcfs_id2str(id),
238 refcount_read(&peer_ni->ksnp_refcount));
244 struct ksock_peer_ni *
245 ksocknal_find_peer(struct lnet_ni *ni, struct lnet_process_id id)
247 struct ksock_peer_ni *peer_ni;
249 read_lock(&ksocknal_data.ksnd_global_lock);
250 peer_ni = ksocknal_find_peer_locked(ni, id);
251 if (peer_ni != NULL) /* +1 ref for caller? */
252 ksocknal_peer_addref(peer_ni);
253 read_unlock(&ksocknal_data.ksnd_global_lock);
259 ksocknal_unlink_peer_locked(struct ksock_peer_ni *peer_ni)
262 struct ksock_interface *iface;
264 for (i = 0; i < peer_ni->ksnp_n_passive_ips; i++) {
265 struct sockaddr_in sa = { .sin_family = AF_INET };
266 LASSERT(i < LNET_INTERFACES_NUM);
267 sa.sin_addr.s_addr = htonl(peer_ni->ksnp_passive_ips[i]);
269 iface = ksocknal_ip2iface(peer_ni->ksnp_ni,
270 (struct sockaddr *)&sa);
272 * All IPs in peer_ni->ksnp_passive_ips[] come from the
273 * interface list, therefore the call must succeed.
275 LASSERT(iface != NULL);
277 CDEBUG(D_NET, "peer_ni=%p iface=%p ksni_nroutes=%d\n",
278 peer_ni, iface, iface->ksni_nroutes);
279 iface->ksni_npeers--;
282 LASSERT(list_empty(&peer_ni->ksnp_conns));
283 LASSERT(peer_ni->ksnp_conn_cb == NULL);
284 LASSERT(!peer_ni->ksnp_closing);
285 peer_ni->ksnp_closing = 1;
286 hlist_del(&peer_ni->ksnp_list);
287 /* lose peerlist's ref */
288 ksocknal_peer_decref(peer_ni);
292 ksocknal_get_peer_info(struct lnet_ni *ni, int index,
293 struct lnet_process_id *id, __u32 *myip, __u32 *peer_ip,
294 int *port, int *conn_count, int *share_count)
296 struct ksock_peer_ni *peer_ni;
297 struct ksock_conn_cb *conn_cb;
302 read_lock(&ksocknal_data.ksnd_global_lock);
304 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
306 if (peer_ni->ksnp_ni != ni)
309 if (peer_ni->ksnp_n_passive_ips == 0 &&
310 peer_ni->ksnp_conn_cb == NULL) {
314 *id = peer_ni->ksnp_id;
324 for (j = 0; j < peer_ni->ksnp_n_passive_ips; j++) {
328 *id = peer_ni->ksnp_id;
329 *myip = peer_ni->ksnp_passive_ips[j];
338 if (peer_ni->ksnp_conn_cb) {
342 conn_cb = peer_ni->ksnp_conn_cb;
344 *id = peer_ni->ksnp_id;
345 if (conn_cb->ksnr_addr.ss_family == AF_INET) {
346 struct sockaddr_in *sa =
347 (void *)&conn_cb->ksnr_addr;
349 rc = choose_ipv4_src(myip,
350 conn_cb->ksnr_myiface,
351 ntohl(sa->sin_addr.s_addr),
353 *peer_ip = ntohl(sa->sin_addr.s_addr);
354 *port = ntohs(sa->sin_port);
357 *peer_ip = 0xFFFFFFFF;
361 *conn_count = conn_cb->ksnr_conn_count;
367 read_unlock(&ksocknal_data.ksnd_global_lock);
372 ksocknal_get_conn_count_by_type(struct ksock_conn_cb *conn_cb,
375 unsigned int count = 0;
378 case SOCKLND_CONN_CONTROL:
379 count = conn_cb->ksnr_ctrl_conn_count;
381 case SOCKLND_CONN_BULK_IN:
382 count = conn_cb->ksnr_blki_conn_count;
384 case SOCKLND_CONN_BULK_OUT:
385 count = conn_cb->ksnr_blko_conn_count;
387 case SOCKLND_CONN_ANY:
388 count = conn_cb->ksnr_conn_count;
399 ksocknal_get_conns_per_peer(struct ksock_peer_ni *peer_ni)
401 struct lnet_ni *ni = peer_ni->ksnp_ni;
402 struct lnet_ioctl_config_socklnd_tunables *tunables;
406 tunables = &ni->ni_lnd_tunables.lnd_tun_u.lnd_sock;
408 return tunables->lnd_conns_per_peer;
412 ksocknal_incr_conn_count(struct ksock_conn_cb *conn_cb,
415 conn_cb->ksnr_conn_count++;
417 /* check if all connections of the given type got created */
419 case SOCKLND_CONN_CONTROL:
420 conn_cb->ksnr_ctrl_conn_count++;
421 /* there's a single control connection per peer */
422 conn_cb->ksnr_connected |= BIT(type);
424 case SOCKLND_CONN_BULK_IN:
425 conn_cb->ksnr_blki_conn_count++;
426 if (conn_cb->ksnr_blki_conn_count >= conn_cb->ksnr_max_conns)
427 conn_cb->ksnr_connected |= BIT(type);
429 case SOCKLND_CONN_BULK_OUT:
430 conn_cb->ksnr_blko_conn_count++;
431 if (conn_cb->ksnr_blko_conn_count >= conn_cb->ksnr_max_conns)
432 conn_cb->ksnr_connected |= BIT(type);
434 case SOCKLND_CONN_ANY:
435 if (conn_cb->ksnr_conn_count >= conn_cb->ksnr_max_conns)
436 conn_cb->ksnr_connected |= BIT(type);
443 CDEBUG(D_NET, "Add conn type %d, ksnr_connected %x ksnr_max_conns %d\n",
444 type, conn_cb->ksnr_connected, conn_cb->ksnr_max_conns);
448 ksocknal_associate_cb_conn_locked(struct ksock_conn_cb *conn_cb,
449 struct ksock_conn *conn)
451 struct ksock_peer_ni *peer_ni = conn_cb->ksnr_peer;
452 int type = conn->ksnc_type;
453 struct ksock_interface *iface;
456 conn_iface = ksocknal_ip2index((struct sockaddr *)&conn->ksnc_myaddr,
458 conn->ksnc_conn_cb = conn_cb;
459 ksocknal_conn_cb_addref(conn_cb);
461 if (conn_cb->ksnr_myiface != conn_iface) {
462 if (conn_cb->ksnr_myiface < 0) {
463 /* route wasn't bound locally yet (the initial route) */
464 CDEBUG(D_NET, "Binding %s %pIS to interface %d\n",
465 libcfs_id2str(peer_ni->ksnp_id),
470 "Rebinding %s %pIS from interface %d to %d\n",
471 libcfs_id2str(peer_ni->ksnp_id),
473 conn_cb->ksnr_myiface,
476 iface = ksocknal_index2iface(peer_ni->ksnp_ni,
477 conn_cb->ksnr_myiface);
479 iface->ksni_nroutes--;
481 conn_cb->ksnr_myiface = conn_iface;
482 iface = ksocknal_index2iface(peer_ni->ksnp_ni,
483 conn_cb->ksnr_myiface);
485 iface->ksni_nroutes++;
488 ksocknal_incr_conn_count(conn_cb, type);
490 /* Successful connection => further attempts can
491 * proceed immediately
493 conn_cb->ksnr_retry_interval = 0;
497 ksocknal_add_conn_cb_locked(struct ksock_peer_ni *peer_ni,
498 struct ksock_conn_cb *conn_cb)
500 struct ksock_conn *conn;
501 struct ksock_net *net = peer_ni->ksnp_ni->ni_data;
503 LASSERT(!peer_ni->ksnp_closing);
504 LASSERT(!conn_cb->ksnr_peer);
505 LASSERT(!conn_cb->ksnr_scheduled);
506 LASSERT(!conn_cb->ksnr_connecting);
507 LASSERT(conn_cb->ksnr_connected == 0);
509 conn_cb->ksnr_peer = peer_ni;
510 ksocknal_peer_addref(peer_ni);
512 /* set the conn_cb's interface to the current net's interface */
513 conn_cb->ksnr_myiface = net->ksnn_interface.ksni_index;
514 net->ksnn_interface.ksni_nroutes++;
516 /* peer_ni's route list takes over my ref on 'route' */
517 peer_ni->ksnp_conn_cb = conn_cb;
519 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
520 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
521 (struct sockaddr *)&conn_cb->ksnr_addr))
524 ksocknal_associate_cb_conn_locked(conn_cb, conn);
525 /* keep going (typed conns) */
530 ksocknal_del_conn_cb_locked(struct ksock_conn_cb *conn_cb)
532 struct ksock_peer_ni *peer_ni = conn_cb->ksnr_peer;
533 struct ksock_interface *iface;
534 struct ksock_conn *conn;
535 struct ksock_conn *cnxt;
537 LASSERT(!conn_cb->ksnr_deleted);
539 /* Close associated conns */
540 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
541 if (conn->ksnc_conn_cb != conn_cb)
544 ksocknal_close_conn_locked(conn, 0);
547 if (conn_cb->ksnr_myiface >= 0) {
548 iface = ksocknal_index2iface(peer_ni->ksnp_ni,
549 conn_cb->ksnr_myiface);
551 iface->ksni_nroutes--;
554 conn_cb->ksnr_deleted = 1;
555 ksocknal_conn_cb_decref(conn_cb); /* drop peer_ni's ref */
556 peer_ni->ksnp_conn_cb = NULL;
558 if (list_empty(&peer_ni->ksnp_conns)) {
559 /* I've just removed the last route to a peer_ni with no active
562 ksocknal_unlink_peer_locked(peer_ni);
567 ksocknal_add_peer(struct lnet_ni *ni, struct lnet_process_id id,
568 struct sockaddr *addr)
570 struct ksock_peer_ni *peer_ni;
571 struct ksock_peer_ni *peer2;
572 struct ksock_conn_cb *conn_cb;
574 if (id.nid == LNET_NID_ANY ||
575 id.pid == LNET_PID_ANY)
578 /* Have a brand new peer_ni ready... */
579 peer_ni = ksocknal_create_peer(ni, id);
581 return PTR_ERR(peer_ni);
583 conn_cb = ksocknal_create_conn_cb(addr);
585 ksocknal_peer_decref(peer_ni);
589 write_lock_bh(&ksocknal_data.ksnd_global_lock);
591 /* always called with a ref on ni, so shutdown can't have started */
592 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers)
595 peer2 = ksocknal_find_peer_locked(ni, id);
597 ksocknal_peer_decref(peer_ni);
600 /* peer_ni table takes my ref on peer_ni */
601 hash_add(ksocknal_data.ksnd_peers, &peer_ni->ksnp_list, id.nid);
604 ksocknal_add_conn_cb_locked(peer_ni, conn_cb);
606 /* Remember conns_per_peer setting at the time
607 * of connection initiation. It will define the
608 * max number of conns per type for this conn_cb
611 conn_cb->ksnr_max_conns = ksocknal_get_conns_per_peer(peer_ni);
613 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
619 ksocknal_del_peer_locked(struct ksock_peer_ni *peer_ni, __u32 ip)
621 struct ksock_conn *conn;
622 struct ksock_conn *cnxt;
623 struct ksock_conn_cb *conn_cb;
625 LASSERT(!peer_ni->ksnp_closing);
627 /* Extra ref prevents peer_ni disappearing until I'm done with it */
628 ksocknal_peer_addref(peer_ni);
629 conn_cb = peer_ni->ksnp_conn_cb;
631 ksocknal_del_conn_cb_locked(conn_cb);
633 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns,
635 ksocknal_close_conn_locked(conn, 0);
637 ksocknal_peer_decref(peer_ni);
638 /* NB peer_ni unlinks itself when last conn/conn_cb is removed */
642 ksocknal_del_peer(struct lnet_ni *ni, struct lnet_process_id id, __u32 ip)
645 struct hlist_node *pnxt;
646 struct ksock_peer_ni *peer_ni;
652 write_lock_bh(&ksocknal_data.ksnd_global_lock);
654 if (id.nid != LNET_NID_ANY) {
655 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
659 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
662 for (i = lo; i <= hi; i++) {
663 hlist_for_each_entry_safe(peer_ni, pnxt,
664 &ksocknal_data.ksnd_peers[i],
666 if (peer_ni->ksnp_ni != ni)
669 if (!((id.nid == LNET_NID_ANY ||
670 peer_ni->ksnp_id.nid == id.nid) &&
671 (id.pid == LNET_PID_ANY ||
672 peer_ni->ksnp_id.pid == id.pid)))
675 ksocknal_peer_addref(peer_ni); /* a ref for me... */
677 ksocknal_del_peer_locked(peer_ni, ip);
679 if (peer_ni->ksnp_closing &&
680 !list_empty(&peer_ni->ksnp_tx_queue)) {
681 LASSERT(list_empty(&peer_ni->ksnp_conns));
682 LASSERT(peer_ni->ksnp_conn_cb == NULL);
684 list_splice_init(&peer_ni->ksnp_tx_queue,
688 ksocknal_peer_decref(peer_ni); /* ...till here */
690 rc = 0; /* matched! */
694 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
696 ksocknal_txlist_done(ni, &zombies, -ENETDOWN);
701 static struct ksock_conn *
702 ksocknal_get_conn_by_idx(struct lnet_ni *ni, int index)
704 struct ksock_peer_ni *peer_ni;
705 struct ksock_conn *conn;
708 read_lock(&ksocknal_data.ksnd_global_lock);
710 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
711 LASSERT(!peer_ni->ksnp_closing);
713 if (peer_ni->ksnp_ni != ni)
716 list_for_each_entry(conn, &peer_ni->ksnp_conns,
721 ksocknal_conn_addref(conn);
722 read_unlock(&ksocknal_data.ksnd_global_lock);
727 read_unlock(&ksocknal_data.ksnd_global_lock);
731 static struct ksock_sched *
732 ksocknal_choose_scheduler_locked(unsigned int cpt)
734 struct ksock_sched *sched = ksocknal_data.ksnd_schedulers[cpt];
737 if (sched->kss_nthreads == 0) {
738 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
739 if (sched->kss_nthreads > 0) {
740 CDEBUG(D_NET, "scheduler[%d] has no threads. selected scheduler[%d]\n",
741 cpt, sched->kss_cpt);
752 ksocknal_accept(struct lnet_ni *ni, struct socket *sock)
754 struct ksock_connreq *cr;
756 struct sockaddr_storage peer;
758 rc = lnet_sock_getaddr(sock, true, &peer);
760 CERROR("Can't determine new connection's address\n");
764 LIBCFS_ALLOC(cr, sizeof(*cr));
766 LCONSOLE_ERROR_MSG(0x12f,
767 "Dropping connection request from %pIS: memory exhausted\n",
774 cr->ksncr_sock = sock;
776 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
778 list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
779 wake_up(&ksocknal_data.ksnd_connd_waitq);
781 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
786 ksocknal_connecting(struct ksock_conn_cb *conn_cb, struct sockaddr *sa)
789 rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr, sa))
790 return conn_cb->ksnr_connecting;
795 ksocknal_create_conn(struct lnet_ni *ni, struct ksock_conn_cb *conn_cb,
796 struct socket *sock, int type)
798 rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
800 struct lnet_process_id peerid;
802 struct ksock_conn *conn;
803 struct ksock_conn *conn2;
804 struct ksock_peer_ni *peer_ni = NULL;
805 struct ksock_peer_ni *peer2;
806 struct ksock_sched *sched;
807 struct ksock_hello_msg *hello;
810 struct ksock_tx *txtmp;
817 active = (conn_cb != NULL);
819 LASSERT(active == (type != SOCKLND_CONN_NONE));
821 LIBCFS_ALLOC(conn, sizeof(*conn));
827 conn->ksnc_peer = NULL;
828 conn->ksnc_conn_cb = NULL;
829 conn->ksnc_sock = sock;
830 /* 2 ref, 1 for conn, another extra ref prevents socket
831 * being closed before establishment of connection */
832 refcount_set(&conn->ksnc_sock_refcount, 2);
833 conn->ksnc_type = type;
834 ksocknal_lib_save_callback(sock, conn);
835 refcount_set(&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
837 conn->ksnc_rx_ready = 0;
838 conn->ksnc_rx_scheduled = 0;
840 INIT_LIST_HEAD(&conn->ksnc_tx_queue);
841 conn->ksnc_tx_ready = 0;
842 conn->ksnc_tx_scheduled = 0;
843 conn->ksnc_tx_carrier = NULL;
844 atomic_set (&conn->ksnc_tx_nob, 0);
846 LIBCFS_ALLOC(hello, offsetof(struct ksock_hello_msg,
847 kshm_ips[LNET_INTERFACES_NUM]));
853 /* stash conn's local and remote addrs */
854 rc = ksocknal_lib_get_conn_addrs(conn);
858 /* Find out/confirm peer_ni's NID and connection type and get the
859 * vector of interfaces she's willing to let me connect to.
860 * Passive connections use the listener timeout since the peer_ni sends
865 peer_ni = conn_cb->ksnr_peer;
866 LASSERT(ni == peer_ni->ksnp_ni);
868 /* Active connection sends HELLO eagerly */
869 hello->kshm_nips = 0;
870 peerid = peer_ni->ksnp_id;
872 write_lock_bh(global_lock);
873 conn->ksnc_proto = peer_ni->ksnp_proto;
874 write_unlock_bh(global_lock);
876 if (conn->ksnc_proto == NULL) {
877 conn->ksnc_proto = &ksocknal_protocol_v3x;
878 #if SOCKNAL_VERSION_DEBUG
879 if (*ksocknal_tunables.ksnd_protocol == 2)
880 conn->ksnc_proto = &ksocknal_protocol_v2x;
881 else if (*ksocknal_tunables.ksnd_protocol == 1)
882 conn->ksnc_proto = &ksocknal_protocol_v1x;
886 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
890 peerid.nid = LNET_NID_ANY;
891 peerid.pid = LNET_PID_ANY;
893 /* Passive, get protocol from peer_ni */
894 conn->ksnc_proto = NULL;
897 rc = ksocknal_recv_hello(ni, conn, hello, &peerid, &incarnation);
901 LASSERT(rc == 0 || active);
902 LASSERT(conn->ksnc_proto != NULL);
903 LASSERT(peerid.nid != LNET_NID_ANY);
905 cpt = lnet_cpt_of_nid(peerid.nid, ni);
908 ksocknal_peer_addref(peer_ni);
909 write_lock_bh(global_lock);
911 peer_ni = ksocknal_create_peer(ni, peerid);
912 if (IS_ERR(peer_ni)) {
913 rc = PTR_ERR(peer_ni);
917 write_lock_bh(global_lock);
919 /* called with a ref on ni, so shutdown can't have started */
920 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers) >= 0);
922 peer2 = ksocknal_find_peer_locked(ni, peerid);
924 /* NB this puts an "empty" peer_ni in the peer_ni
925 * table (which takes my ref) */
926 hash_add(ksocknal_data.ksnd_peers,
927 &peer_ni->ksnp_list, peerid.nid);
929 ksocknal_peer_decref(peer_ni);
934 ksocknal_peer_addref(peer_ni);
935 peer_ni->ksnp_accepting++;
937 /* Am I already connecting to this guy? Resolve in
938 * favour of higher NID...
940 if (peerid.nid < ni->ni_nid &&
941 ksocknal_connecting(peer_ni->ksnp_conn_cb,
942 ((struct sockaddr *) &conn->ksnc_peeraddr))) {
944 warn = "connection race resolution";
949 if (peer_ni->ksnp_closing ||
950 (active && conn_cb->ksnr_deleted)) {
951 /* peer_ni/conn_cb got closed under me */
953 warn = "peer_ni/conn_cb removed";
957 if (peer_ni->ksnp_proto == NULL) {
958 /* Never connected before.
959 * NB recv_hello may have returned EPROTO to signal my peer_ni
960 * wants a different protocol than the one I asked for.
962 LASSERT(list_empty(&peer_ni->ksnp_conns));
964 peer_ni->ksnp_proto = conn->ksnc_proto;
965 peer_ni->ksnp_incarnation = incarnation;
968 if (peer_ni->ksnp_proto != conn->ksnc_proto ||
969 peer_ni->ksnp_incarnation != incarnation) {
970 /* peer_ni rebooted or I've got the wrong protocol version */
971 ksocknal_close_peer_conns_locked(peer_ni, NULL, 0);
973 peer_ni->ksnp_proto = NULL;
975 warn = peer_ni->ksnp_incarnation != incarnation ?
977 "wrong proto version";
987 warn = "lost conn race";
990 warn = "retry with different protocol version";
994 /* Refuse to duplicate an existing connection, unless this is a
995 * loopback connection */
996 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
997 (struct sockaddr *)&conn->ksnc_myaddr)) {
998 list_for_each_entry(conn2, &peer_ni->ksnp_conns, ksnc_list) {
1000 (struct sockaddr *)&conn2->ksnc_peeraddr,
1001 (struct sockaddr *)&conn->ksnc_peeraddr) ||
1003 (struct sockaddr *)&conn2->ksnc_myaddr,
1004 (struct sockaddr *)&conn->ksnc_myaddr) ||
1005 conn2->ksnc_type != conn->ksnc_type)
1009 /* If max conns per type is not registered in conn_cb
1010 * as ksnr_max_conns, use ni's conns_per_peer
1012 if ((peer_ni->ksnp_conn_cb &&
1013 num_dup < peer_ni->ksnp_conn_cb->ksnr_max_conns) ||
1014 (!peer_ni->ksnp_conn_cb &&
1015 num_dup < ksocknal_get_conns_per_peer(peer_ni)))
1018 /* Reply on a passive connection attempt so the peer_ni
1019 * realises we're connected.
1029 /* If the connection created by this route didn't bind to the IP
1030 * address the route connected to, the connection/route matching
1031 * code below probably isn't going to work.
1034 !rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr,
1035 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1036 CERROR("Route %s %pIS connected to %pIS\n",
1037 libcfs_id2str(peer_ni->ksnp_id),
1038 &conn_cb->ksnr_addr,
1039 &conn->ksnc_peeraddr);
1042 /* Search for a conn_cb corresponding to the new connection and
1043 * create an association. This allows incoming connections created
1044 * by conn_cbs in my peer_ni to match my own conn_cb entries so I don't
1045 * continually create duplicate conn_cbs.
1047 conn_cb = peer_ni->ksnp_conn_cb;
1049 if (conn_cb && rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
1050 (struct sockaddr *)&conn_cb->ksnr_addr))
1051 ksocknal_associate_cb_conn_locked(conn_cb, conn);
1053 conn->ksnc_peer = peer_ni; /* conn takes my ref on peer_ni */
1054 peer_ni->ksnp_last_alive = ktime_get_seconds();
1055 peer_ni->ksnp_send_keepalive = 0;
1056 peer_ni->ksnp_error = 0;
1058 sched = ksocknal_choose_scheduler_locked(cpt);
1060 CERROR("no schedulers available. node is unhealthy\n");
1064 * The cpt might have changed if we ended up selecting a non cpt
1065 * native scheduler. So use the scheduler's cpt instead.
1067 cpt = sched->kss_cpt;
1068 sched->kss_nconns++;
1069 conn->ksnc_scheduler = sched;
1071 conn->ksnc_tx_last_post = ktime_get_seconds();
1072 /* Set the deadline for the outgoing HELLO to drain */
1073 conn->ksnc_tx_bufnob = sock->sk->sk_wmem_queued;
1074 conn->ksnc_tx_deadline = ktime_get_seconds() +
1076 smp_mb(); /* order with adding to peer_ni's conn list */
1078 list_add(&conn->ksnc_list, &peer_ni->ksnp_conns);
1079 ksocknal_conn_addref(conn);
1081 ksocknal_new_packet(conn, 0);
1083 conn->ksnc_zc_capable = ksocknal_lib_zc_capable(conn);
1085 /* Take packets blocking for this connection. */
1086 list_for_each_entry_safe(tx, txtmp, &peer_ni->ksnp_tx_queue, tx_list) {
1087 if (conn->ksnc_proto->pro_match_tx(conn, tx, tx->tx_nonblk) ==
1091 list_del(&tx->tx_list);
1092 ksocknal_queue_tx_locked(tx, conn);
1095 write_unlock_bh(global_lock);
1097 /* We've now got a new connection. Any errors from here on are just
1098 * like "normal" comms errors and we close the connection normally.
1099 * NB (a) we still have to send the reply HELLO for passive
1101 * (b) normal I/O on the conn is blocked until I setup and call the
1105 CDEBUG(D_NET, "New conn %s p %d.x %pIS -> %pISp"
1106 " incarnation:%lld sched[%d]\n",
1107 libcfs_id2str(peerid), conn->ksnc_proto->pro_version,
1108 &conn->ksnc_myaddr, &conn->ksnc_peeraddr,
1112 hello->kshm_nips = 0;
1113 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
1116 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1117 kshm_ips[LNET_INTERFACES_NUM]));
1119 /* setup the socket AFTER I've received hello (it disables
1120 * SO_LINGER). I might call back to the acceptor who may want
1121 * to send a protocol version response and then close the
1122 * socket; this ensures the socket only tears down after the
1123 * response has been sent.
1126 rc = ksocknal_lib_setup_sock(sock);
1128 write_lock_bh(global_lock);
1130 /* NB my callbacks block while I hold ksnd_global_lock */
1131 ksocknal_lib_set_callback(sock, conn);
1134 peer_ni->ksnp_accepting--;
1136 write_unlock_bh(global_lock);
1139 write_lock_bh(global_lock);
1140 if (!conn->ksnc_closing) {
1141 /* could be closed by another thread */
1142 ksocknal_close_conn_locked(conn, rc);
1144 write_unlock_bh(global_lock);
1145 } else if (ksocknal_connsock_addref(conn) == 0) {
1146 /* Allow I/O to proceed. */
1147 ksocknal_read_callback(conn);
1148 ksocknal_write_callback(conn);
1149 ksocknal_connsock_decref(conn);
1152 ksocknal_connsock_decref(conn);
1153 ksocknal_conn_decref(conn);
1158 if (!peer_ni->ksnp_closing &&
1159 list_empty(&peer_ni->ksnp_conns) &&
1160 peer_ni->ksnp_conn_cb == NULL) {
1161 list_splice_init(&peer_ni->ksnp_tx_queue, &zombies);
1162 ksocknal_unlink_peer_locked(peer_ni);
1165 write_unlock_bh(global_lock);
1169 CERROR("Not creating conn %s type %d: %s\n",
1170 libcfs_id2str(peerid), conn->ksnc_type, warn);
1172 CDEBUG(D_NET, "Not creating conn %s type %d: %s\n",
1173 libcfs_id2str(peerid), conn->ksnc_type, warn);
1178 /* Request retry by replying with CONN_NONE
1179 * ksnc_proto has been set already
1181 conn->ksnc_type = SOCKLND_CONN_NONE;
1182 hello->kshm_nips = 0;
1183 ksocknal_send_hello(ni, conn, peerid.nid, hello);
1186 write_lock_bh(global_lock);
1187 peer_ni->ksnp_accepting--;
1188 write_unlock_bh(global_lock);
1192 * If we get here without an error code, just use -EALREADY.
1193 * Depending on how we got here, the error may be positive
1194 * or negative. Normalize the value for ksocknal_txlist_done().
1196 rc2 = (rc == 0 ? -EALREADY : (rc > 0 ? -rc : rc));
1197 ksocknal_txlist_done(ni, &zombies, rc2);
1198 ksocknal_peer_decref(peer_ni);
1202 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1203 kshm_ips[LNET_INTERFACES_NUM]));
1205 LIBCFS_FREE(conn, sizeof(*conn));
1214 ksocknal_close_conn_locked(struct ksock_conn *conn, int error)
1216 /* This just does the immmediate housekeeping, and queues the
1217 * connection for the reaper to terminate.
1218 * Caller holds ksnd_global_lock exclusively in irq context */
1219 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1220 struct ksock_conn_cb *conn_cb;
1221 struct ksock_conn *conn2;
1223 LASSERT(peer_ni->ksnp_error == 0);
1224 LASSERT(!conn->ksnc_closing);
1225 conn->ksnc_closing = 1;
1227 /* ksnd_deathrow_conns takes over peer_ni's ref */
1228 list_del(&conn->ksnc_list);
1230 conn_cb = conn->ksnc_conn_cb;
1231 if (conn_cb != NULL) {
1232 /* dissociate conn from cb... */
1233 LASSERT(!conn_cb->ksnr_deleted);
1235 /* connected bit is set only if all connections
1236 * of the given type got created
1238 if (ksocknal_get_conn_count_by_type(conn_cb, conn->ksnc_type) ==
1239 conn_cb->ksnr_max_conns)
1240 LASSERT((conn_cb->ksnr_connected &
1241 BIT(conn->ksnc_type)) != 0);
1243 list_for_each_entry(conn2, &peer_ni->ksnp_conns, ksnc_list) {
1244 if (conn2->ksnc_conn_cb == conn_cb &&
1245 conn2->ksnc_type == conn->ksnc_type)
1248 conn_cb->ksnr_connected &= ~BIT(conn->ksnc_type);
1250 conn->ksnc_conn_cb = NULL;
1252 /* drop conn's ref on conn_cb */
1253 ksocknal_conn_cb_decref(conn_cb);
1256 if (list_empty(&peer_ni->ksnp_conns)) {
1257 /* No more connections to this peer_ni */
1259 if (!list_empty(&peer_ni->ksnp_tx_queue)) {
1260 struct ksock_tx *tx;
1262 LASSERT(conn->ksnc_proto == &ksocknal_protocol_v3x);
1264 /* throw them to the last connection...,
1265 * these TXs will be send to /dev/null by scheduler */
1266 list_for_each_entry(tx, &peer_ni->ksnp_tx_queue,
1268 ksocknal_tx_prep(conn, tx);
1270 spin_lock_bh(&conn->ksnc_scheduler->kss_lock);
1271 list_splice_init(&peer_ni->ksnp_tx_queue,
1272 &conn->ksnc_tx_queue);
1273 spin_unlock_bh(&conn->ksnc_scheduler->kss_lock);
1276 /* renegotiate protocol version */
1277 peer_ni->ksnp_proto = NULL;
1278 /* stash last conn close reason */
1279 peer_ni->ksnp_error = error;
1281 if (peer_ni->ksnp_conn_cb == NULL) {
1282 /* I've just closed last conn belonging to a
1283 * peer_ni with no connections to it
1285 ksocknal_unlink_peer_locked(peer_ni);
1289 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1291 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_deathrow_conns);
1292 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1294 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1298 ksocknal_peer_failed(struct ksock_peer_ni *peer_ni)
1300 bool notify = false;
1301 time64_t last_alive = 0;
1303 /* There has been a connection failure or comms error; but I'll only
1304 * tell LNET I think the peer_ni is dead if it's to another kernel and
1305 * there are no connections or connection attempts in existence. */
1307 read_lock(&ksocknal_data.ksnd_global_lock);
1309 if ((peer_ni->ksnp_id.pid & LNET_PID_USERFLAG) == 0 &&
1310 list_empty(&peer_ni->ksnp_conns) &&
1311 peer_ni->ksnp_accepting == 0 &&
1312 !ksocknal_find_connecting_conn_cb_locked(peer_ni)) {
1314 last_alive = peer_ni->ksnp_last_alive;
1317 read_unlock(&ksocknal_data.ksnd_global_lock);
1320 lnet_notify(peer_ni->ksnp_ni, peer_ni->ksnp_id.nid,
1321 false, false, last_alive);
1325 ksocknal_finalize_zcreq(struct ksock_conn *conn)
1327 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1328 struct ksock_tx *tx;
1329 struct ksock_tx *tmp;
1332 /* NB safe to finalize TXs because closing of socket will
1333 * abort all buffered data */
1334 LASSERT(conn->ksnc_sock == NULL);
1336 spin_lock(&peer_ni->ksnp_lock);
1338 list_for_each_entry_safe(tx, tmp, &peer_ni->ksnp_zc_req_list,
1340 if (tx->tx_conn != conn)
1343 LASSERT(tx->tx_msg.ksm_zc_cookies[0] != 0);
1345 tx->tx_msg.ksm_zc_cookies[0] = 0;
1346 tx->tx_zc_aborted = 1; /* mark it as not-acked */
1347 list_move(&tx->tx_zc_list, &zlist);
1350 spin_unlock(&peer_ni->ksnp_lock);
1352 while ((tx = list_first_entry_or_null(&zlist, struct ksock_tx,
1353 tx_zc_list)) != NULL) {
1354 list_del(&tx->tx_zc_list);
1355 ksocknal_tx_decref(tx);
1360 ksocknal_terminate_conn(struct ksock_conn *conn)
1362 /* This gets called by the reaper (guaranteed thread context) to
1363 * disengage the socket from its callbacks and close it.
1364 * ksnc_refcount will eventually hit zero, and then the reaper will
1367 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1368 struct ksock_sched *sched = conn->ksnc_scheduler;
1369 bool failed = false;
1371 LASSERT(conn->ksnc_closing);
1373 /* wake up the scheduler to "send" all remaining packets to /dev/null */
1374 spin_lock_bh(&sched->kss_lock);
1376 /* a closing conn is always ready to tx */
1377 conn->ksnc_tx_ready = 1;
1379 if (!conn->ksnc_tx_scheduled &&
1380 !list_empty(&conn->ksnc_tx_queue)) {
1381 list_add_tail(&conn->ksnc_tx_list,
1382 &sched->kss_tx_conns);
1383 conn->ksnc_tx_scheduled = 1;
1384 /* extra ref for scheduler */
1385 ksocknal_conn_addref(conn);
1387 wake_up(&sched->kss_waitq);
1390 spin_unlock_bh(&sched->kss_lock);
1392 /* serialise with callbacks */
1393 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1395 ksocknal_lib_reset_callback(conn->ksnc_sock, conn);
1397 /* OK, so this conn may not be completely disengaged from its
1398 * scheduler yet, but it _has_ committed to terminate...
1400 conn->ksnc_scheduler->kss_nconns--;
1402 if (peer_ni->ksnp_error != 0) {
1403 /* peer_ni's last conn closed in error */
1404 LASSERT(list_empty(&peer_ni->ksnp_conns));
1406 peer_ni->ksnp_error = 0; /* avoid multiple notifications */
1409 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1412 ksocknal_peer_failed(peer_ni);
1414 /* The socket is closed on the final put; either here, or in
1415 * ksocknal_{send,recv}msg(). Since we set up the linger2 option
1416 * when the connection was established, this will close the socket
1417 * immediately, aborting anything buffered in it. Any hung
1418 * zero-copy transmits will therefore complete in finite time.
1420 ksocknal_connsock_decref(conn);
1424 ksocknal_queue_zombie_conn(struct ksock_conn *conn)
1426 /* Queue the conn for the reaper to destroy */
1427 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1428 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1430 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
1431 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1433 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1437 ksocknal_destroy_conn(struct ksock_conn *conn)
1441 /* Final coup-de-grace of the reaper */
1442 CDEBUG(D_NET, "connection %p\n", conn);
1444 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1445 LASSERT(refcount_read(&conn->ksnc_sock_refcount) == 0);
1446 LASSERT(conn->ksnc_sock == NULL);
1447 LASSERT(conn->ksnc_conn_cb == NULL);
1448 LASSERT(!conn->ksnc_tx_scheduled);
1449 LASSERT(!conn->ksnc_rx_scheduled);
1450 LASSERT(list_empty(&conn->ksnc_tx_queue));
1452 /* complete current receive if any */
1453 switch (conn->ksnc_rx_state) {
1454 case SOCKNAL_RX_LNET_PAYLOAD:
1455 last_rcv = conn->ksnc_rx_deadline -
1457 CERROR("Completing partial receive from %s[%d], ip %pISp, with error, wanted: %d, left: %d, last alive is %lld secs ago\n",
1458 libcfs_id2str(conn->ksnc_peer->ksnp_id), conn->ksnc_type,
1459 &conn->ksnc_peeraddr,
1460 conn->ksnc_rx_nob_wanted, conn->ksnc_rx_nob_left,
1461 ktime_get_seconds() - last_rcv);
1462 if (conn->ksnc_lnet_msg)
1463 conn->ksnc_lnet_msg->msg_health_status =
1464 LNET_MSG_STATUS_REMOTE_ERROR;
1465 lnet_finalize(conn->ksnc_lnet_msg, -EIO);
1467 case SOCKNAL_RX_LNET_HEADER:
1468 if (conn->ksnc_rx_started)
1469 CERROR("Incomplete receive of lnet header from %s, ip %pISp, with error, protocol: %d.x.\n",
1470 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1471 &conn->ksnc_peeraddr,
1472 conn->ksnc_proto->pro_version);
1474 case SOCKNAL_RX_KSM_HEADER:
1475 if (conn->ksnc_rx_started)
1476 CERROR("Incomplete receive of ksock message from %s, ip %pISp, with error, protocol: %d.x.\n",
1477 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1478 &conn->ksnc_peeraddr,
1479 conn->ksnc_proto->pro_version);
1481 case SOCKNAL_RX_SLOP:
1482 if (conn->ksnc_rx_started)
1483 CERROR("Incomplete receive of slops from %s, ip %pISp, with error\n",
1484 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1485 &conn->ksnc_peeraddr);
1492 ksocknal_peer_decref(conn->ksnc_peer);
1494 LIBCFS_FREE (conn, sizeof (*conn));
1498 ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
1499 struct sockaddr *addr, int why)
1501 struct ksock_conn *conn;
1502 struct ksock_conn *cnxt;
1505 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
1508 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1510 ksocknal_close_conn_locked(conn, why);
1518 ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why)
1520 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1523 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1525 count = ksocknal_close_peer_conns_locked(
1526 peer_ni, (struct sockaddr *)&conn->ksnc_peeraddr, why);
1528 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1534 ksocknal_close_matching_conns(struct lnet_process_id id, __u32 ipaddr)
1536 struct ksock_peer_ni *peer_ni;
1537 struct hlist_node *pnxt;
1542 struct sockaddr_in sa = {.sin_family = AF_INET};
1544 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1546 if (id.nid != LNET_NID_ANY) {
1547 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
1551 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1554 sa.sin_addr.s_addr = htonl(ipaddr);
1555 for (i = lo; i <= hi; i++) {
1556 hlist_for_each_entry_safe(peer_ni, pnxt,
1557 &ksocknal_data.ksnd_peers[i],
1560 if (!((id.nid == LNET_NID_ANY ||
1561 id.nid == peer_ni->ksnp_id.nid) &&
1562 (id.pid == LNET_PID_ANY ||
1563 id.pid == peer_ni->ksnp_id.pid)))
1566 count += ksocknal_close_peer_conns_locked(
1568 ipaddr ? (struct sockaddr *)&sa : NULL, 0);
1572 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1574 /* wildcards always succeed */
1575 if (id.nid == LNET_NID_ANY || id.pid == LNET_PID_ANY || ipaddr == 0)
1578 return (count == 0 ? -ENOENT : 0);
1582 ksocknal_notify_gw_down(lnet_nid_t gw_nid)
1584 /* The router is telling me she's been notified of a change in
1587 struct lnet_process_id id = {
1589 .pid = LNET_PID_ANY,
1592 CDEBUG(D_NET, "gw %s down\n", libcfs_nid2str(gw_nid));
1594 /* If the gateway crashed, close all open connections... */
1595 ksocknal_close_matching_conns(id, 0);
1598 /* We can only establish new connections
1599 * if we have autroutes, and these connect on demand. */
1603 ksocknal_push_peer(struct ksock_peer_ni *peer_ni)
1607 struct ksock_conn *conn;
1609 for (index = 0; ; index++) {
1610 read_lock(&ksocknal_data.ksnd_global_lock);
1615 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
1617 ksocknal_conn_addref(conn);
1622 read_unlock(&ksocknal_data.ksnd_global_lock);
1627 ksocknal_lib_push_conn (conn);
1628 ksocknal_conn_decref(conn);
1633 ksocknal_push(struct lnet_ni *ni, struct lnet_process_id id)
1640 if (id.nid != LNET_NID_ANY) {
1641 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
1645 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1648 for (bkt = lo; bkt <= hi; bkt++) {
1649 int peer_off; /* searching offset in peer_ni hash table */
1651 for (peer_off = 0; ; peer_off++) {
1652 struct ksock_peer_ni *peer_ni;
1655 read_lock(&ksocknal_data.ksnd_global_lock);
1656 hlist_for_each_entry(peer_ni,
1657 &ksocknal_data.ksnd_peers[bkt],
1659 if (!((id.nid == LNET_NID_ANY ||
1660 id.nid == peer_ni->ksnp_id.nid) &&
1661 (id.pid == LNET_PID_ANY ||
1662 id.pid == peer_ni->ksnp_id.pid)))
1665 if (i++ == peer_off) {
1666 ksocknal_peer_addref(peer_ni);
1670 read_unlock(&ksocknal_data.ksnd_global_lock);
1672 if (i <= peer_off) /* no match */
1676 ksocknal_push_peer(peer_ni);
1677 ksocknal_peer_decref(peer_ni);
1684 ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg)
1686 struct lnet_process_id id = {0};
1687 struct libcfs_ioctl_data *data = arg;
1691 case IOC_LIBCFS_GET_INTERFACE: {
1692 struct ksock_net *net = ni->ni_data;
1693 struct ksock_interface *iface;
1694 struct sockaddr_in *sa;
1696 read_lock(&ksocknal_data.ksnd_global_lock);
1698 if (data->ioc_count >= 1) {
1702 iface = &net->ksnn_interface;
1704 sa = (void *)&iface->ksni_addr;
1705 if (sa->sin_family == AF_INET)
1706 data->ioc_u32[0] = ntohl(sa->sin_addr.s_addr);
1708 data->ioc_u32[0] = 0xFFFFFFFF;
1709 data->ioc_u32[1] = iface->ksni_netmask;
1710 data->ioc_u32[2] = iface->ksni_npeers;
1711 data->ioc_u32[3] = iface->ksni_nroutes;
1714 read_unlock(&ksocknal_data.ksnd_global_lock);
1718 case IOC_LIBCFS_GET_PEER: {
1723 int share_count = 0;
1725 rc = ksocknal_get_peer_info(ni, data->ioc_count,
1726 &id, &myip, &ip, &port,
1727 &conn_count, &share_count);
1731 data->ioc_nid = id.nid;
1732 data->ioc_count = share_count;
1733 data->ioc_u32[0] = ip;
1734 data->ioc_u32[1] = port;
1735 data->ioc_u32[2] = myip;
1736 data->ioc_u32[3] = conn_count;
1737 data->ioc_u32[4] = id.pid;
1741 case IOC_LIBCFS_ADD_PEER: {
1742 struct sockaddr_in sa = {.sin_family = AF_INET};
1744 id.nid = data->ioc_nid;
1745 id.pid = LNET_PID_LUSTRE;
1746 sa.sin_addr.s_addr = htonl(data->ioc_u32[0]);
1747 sa.sin_port = htons(data->ioc_u32[1]);
1748 return ksocknal_add_peer(ni, id, (struct sockaddr *)&sa);
1750 case IOC_LIBCFS_DEL_PEER:
1751 id.nid = data->ioc_nid;
1752 id.pid = LNET_PID_ANY;
1753 return ksocknal_del_peer (ni, id,
1754 data->ioc_u32[0]); /* IP */
1756 case IOC_LIBCFS_GET_CONN: {
1760 struct ksock_conn *conn = ksocknal_get_conn_by_idx(ni, data->ioc_count);
1761 struct sockaddr_in *psa = (void *)&conn->ksnc_peeraddr;
1762 struct sockaddr_in *mysa = (void *)&conn->ksnc_myaddr;
1767 ksocknal_lib_get_conn_tunables(conn, &txmem, &rxmem, &nagle);
1769 data->ioc_count = txmem;
1770 data->ioc_nid = conn->ksnc_peer->ksnp_id.nid;
1771 data->ioc_flags = nagle;
1772 if (psa->sin_family == AF_INET)
1773 data->ioc_u32[0] = ntohl(psa->sin_addr.s_addr);
1775 data->ioc_u32[0] = 0xFFFFFFFF;
1776 data->ioc_u32[1] = rpc_get_port((struct sockaddr *)
1777 &conn->ksnc_peeraddr);
1778 if (mysa->sin_family == AF_INET)
1779 data->ioc_u32[2] = ntohl(mysa->sin_addr.s_addr);
1781 data->ioc_u32[2] = 0xFFFFFFFF;
1782 data->ioc_u32[3] = conn->ksnc_type;
1783 data->ioc_u32[4] = conn->ksnc_scheduler->kss_cpt;
1784 data->ioc_u32[5] = rxmem;
1785 data->ioc_u32[6] = conn->ksnc_peer->ksnp_id.pid;
1786 ksocknal_conn_decref(conn);
1790 case IOC_LIBCFS_CLOSE_CONNECTION:
1791 id.nid = data->ioc_nid;
1792 id.pid = LNET_PID_ANY;
1793 return ksocknal_close_matching_conns (id,
1796 case IOC_LIBCFS_REGISTER_MYNID:
1797 /* Ignore if this is a noop */
1798 if (data->ioc_nid == ni->ni_nid)
1801 CERROR("obsolete IOC_LIBCFS_REGISTER_MYNID: %s(%s)\n",
1802 libcfs_nid2str(data->ioc_nid),
1803 libcfs_nid2str(ni->ni_nid));
1806 case IOC_LIBCFS_PUSH_CONNECTION:
1807 id.nid = data->ioc_nid;
1808 id.pid = LNET_PID_ANY;
1809 return ksocknal_push(ni, id);
1818 ksocknal_free_buffers (void)
1820 LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
1822 if (ksocknal_data.ksnd_schedulers != NULL)
1823 cfs_percpt_free(ksocknal_data.ksnd_schedulers);
1825 spin_lock(&ksocknal_data.ksnd_tx_lock);
1827 if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
1829 struct ksock_tx *tx;
1831 list_splice_init(&ksocknal_data.ksnd_idle_noop_txs, &zlist);
1832 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1834 while ((tx = list_first_entry_or_null(&zlist, struct ksock_tx,
1835 tx_list)) != NULL) {
1836 list_del(&tx->tx_list);
1837 LIBCFS_FREE(tx, tx->tx_desc_size);
1840 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1844 static int ksocknal_get_link_status(struct net_device *dev)
1850 if (!netif_running(dev))
1852 /* Some devices may not be providing link settings */
1853 else if (dev->ethtool_ops->get_link)
1854 ret = dev->ethtool_ops->get_link(dev);
1860 ksocknal_handle_link_state_change(struct net_device *dev,
1861 unsigned char operstate)
1864 struct ksock_net *net;
1865 struct ksock_net *cnxt;
1867 unsigned char link_down = !(operstate == IF_OPER_UP);
1869 ifindex = dev->ifindex;
1871 if (!ksocknal_data.ksnd_nnets)
1874 list_for_each_entry_safe(net, cnxt, &ksocknal_data.ksnd_nets,
1876 if (net->ksnn_interface.ksni_index != ifindex)
1880 atomic_set(&ni->ni_fatal_error_on, link_down);
1882 atomic_set(&ni->ni_fatal_error_on,
1883 (ksocknal_get_link_status(dev) == 0));
1890 /************************************
1891 * Net device notifier event handler
1892 ************************************/
1893 static int ksocknal_device_event(struct notifier_block *unused,
1894 unsigned long event, void *ptr)
1896 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1897 unsigned char operstate;
1899 operstate = dev->operstate;
1905 ksocknal_handle_link_state_change(dev, operstate);
1912 static struct notifier_block ksocknal_notifier_block = {
1913 .notifier_call = ksocknal_device_event,
1917 ksocknal_base_shutdown(void)
1919 struct ksock_sched *sched;
1920 struct ksock_peer_ni *peer_ni;
1923 CDEBUG(D_MALLOC, "before NAL cleanup: kmem %lld\n",
1924 libcfs_kmem_read());
1925 LASSERT (ksocknal_data.ksnd_nnets == 0);
1927 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL)
1928 unregister_netdevice_notifier(&ksocknal_notifier_block);
1930 switch (ksocknal_data.ksnd_init) {
1935 case SOCKNAL_INIT_ALL:
1936 case SOCKNAL_INIT_DATA:
1937 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list)
1940 LASSERT(list_empty(&ksocknal_data.ksnd_nets));
1941 LASSERT(list_empty(&ksocknal_data.ksnd_enomem_conns));
1942 LASSERT(list_empty(&ksocknal_data.ksnd_zombie_conns));
1943 LASSERT(list_empty(&ksocknal_data.ksnd_connd_connreqs));
1944 LASSERT(list_empty(&ksocknal_data.ksnd_connd_routes));
1946 if (ksocknal_data.ksnd_schedulers != NULL) {
1947 cfs_percpt_for_each(sched, i,
1948 ksocknal_data.ksnd_schedulers) {
1950 LASSERT(list_empty(&sched->kss_tx_conns));
1951 LASSERT(list_empty(&sched->kss_rx_conns));
1952 LASSERT(list_empty(&sched->kss_zombie_noop_txs));
1953 LASSERT(sched->kss_nconns == 0);
1957 /* flag threads to terminate; wake and wait for them to die */
1958 ksocknal_data.ksnd_shuttingdown = 1;
1959 wake_up_all(&ksocknal_data.ksnd_connd_waitq);
1960 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1962 if (ksocknal_data.ksnd_schedulers != NULL) {
1963 cfs_percpt_for_each(sched, i,
1964 ksocknal_data.ksnd_schedulers)
1965 wake_up_all(&sched->kss_waitq);
1968 wait_var_event_warning(&ksocknal_data.ksnd_nthreads,
1969 atomic_read(&ksocknal_data.ksnd_nthreads) == 0,
1970 "waiting for %d threads to terminate\n",
1971 atomic_read(&ksocknal_data.ksnd_nthreads));
1973 ksocknal_free_buffers();
1975 ksocknal_data.ksnd_init = SOCKNAL_INIT_NOTHING;
1979 CDEBUG(D_MALLOC, "after NAL cleanup: kmem %lld\n",
1980 libcfs_kmem_read());
1982 module_put(THIS_MODULE);
1986 ksocknal_base_startup(void)
1988 struct ksock_sched *sched;
1992 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
1993 LASSERT(ksocknal_data.ksnd_nnets == 0);
1995 memset(&ksocknal_data, 0, sizeof(ksocknal_data)); /* zero pointers */
1997 hash_init(ksocknal_data.ksnd_peers);
1999 rwlock_init(&ksocknal_data.ksnd_global_lock);
2000 INIT_LIST_HEAD(&ksocknal_data.ksnd_nets);
2002 spin_lock_init(&ksocknal_data.ksnd_reaper_lock);
2003 INIT_LIST_HEAD(&ksocknal_data.ksnd_enomem_conns);
2004 INIT_LIST_HEAD(&ksocknal_data.ksnd_zombie_conns);
2005 INIT_LIST_HEAD(&ksocknal_data.ksnd_deathrow_conns);
2006 init_waitqueue_head(&ksocknal_data.ksnd_reaper_waitq);
2008 spin_lock_init(&ksocknal_data.ksnd_connd_lock);
2009 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_connreqs);
2010 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_routes);
2011 init_waitqueue_head(&ksocknal_data.ksnd_connd_waitq);
2013 spin_lock_init(&ksocknal_data.ksnd_tx_lock);
2014 INIT_LIST_HEAD(&ksocknal_data.ksnd_idle_noop_txs);
2016 /* NB memset above zeros whole of ksocknal_data */
2018 /* flag lists/ptrs/locks initialised */
2019 ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
2020 if (!try_module_get(THIS_MODULE))
2023 /* Create a scheduler block per available CPT */
2024 ksocknal_data.ksnd_schedulers = cfs_percpt_alloc(lnet_cpt_table(),
2026 if (ksocknal_data.ksnd_schedulers == NULL)
2029 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
2033 * make sure not to allocate more threads than there are
2034 * cores/CPUs in teh CPT
2036 nthrs = cfs_cpt_weight(lnet_cpt_table(), i);
2037 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2038 nthrs = min(nthrs, *ksocknal_tunables.ksnd_nscheds);
2041 * max to half of CPUs, assume another half should be
2042 * reserved for upper layer modules
2044 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2047 sched->kss_nthreads_max = nthrs;
2050 spin_lock_init(&sched->kss_lock);
2051 INIT_LIST_HEAD(&sched->kss_rx_conns);
2052 INIT_LIST_HEAD(&sched->kss_tx_conns);
2053 INIT_LIST_HEAD(&sched->kss_zombie_noop_txs);
2054 init_waitqueue_head(&sched->kss_waitq);
2057 ksocknal_data.ksnd_connd_starting = 0;
2058 ksocknal_data.ksnd_connd_failed_stamp = 0;
2059 ksocknal_data.ksnd_connd_starting_stamp = ktime_get_real_seconds();
2060 /* must have at least 2 connds to remain responsive to accepts while
2062 if (*ksocknal_tunables.ksnd_nconnds < SOCKNAL_CONND_RESV + 1)
2063 *ksocknal_tunables.ksnd_nconnds = SOCKNAL_CONND_RESV + 1;
2065 if (*ksocknal_tunables.ksnd_nconnds_max <
2066 *ksocknal_tunables.ksnd_nconnds) {
2067 ksocknal_tunables.ksnd_nconnds_max =
2068 ksocknal_tunables.ksnd_nconnds;
2071 for (i = 0; i < *ksocknal_tunables.ksnd_nconnds; i++) {
2072 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2073 ksocknal_data.ksnd_connd_starting++;
2074 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2076 rc = ksocknal_thread_start(ksocknal_connd,
2077 (void *)((uintptr_t)i),
2078 "socknal_cd%02d", i);
2080 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2081 ksocknal_data.ksnd_connd_starting--;
2082 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2083 CERROR("Can't spawn socknal connd: %d\n", rc);
2088 rc = ksocknal_thread_start(ksocknal_reaper, NULL, "socknal_reaper");
2090 CERROR ("Can't spawn socknal reaper: %d\n", rc);
2094 register_netdevice_notifier(&ksocknal_notifier_block);
2096 /* flag everything initialised */
2097 ksocknal_data.ksnd_init = SOCKNAL_INIT_ALL;
2102 ksocknal_base_shutdown();
2107 ksocknal_debug_peerhash(struct lnet_ni *ni)
2109 struct ksock_peer_ni *peer_ni;
2112 read_lock(&ksocknal_data.ksnd_global_lock);
2114 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
2115 struct ksock_conn_cb *conn_cb;
2116 struct ksock_conn *conn;
2118 if (peer_ni->ksnp_ni != ni)
2121 CWARN("Active peer_ni on shutdown: %s, ref %d, "
2122 "closing %d, accepting %d, err %d, zcookie %llu, "
2123 "txq %d, zc_req %d\n", libcfs_id2str(peer_ni->ksnp_id),
2124 refcount_read(&peer_ni->ksnp_refcount),
2125 peer_ni->ksnp_closing,
2126 peer_ni->ksnp_accepting, peer_ni->ksnp_error,
2127 peer_ni->ksnp_zc_next_cookie,
2128 !list_empty(&peer_ni->ksnp_tx_queue),
2129 !list_empty(&peer_ni->ksnp_zc_req_list));
2131 conn_cb = peer_ni->ksnp_conn_cb;
2133 CWARN("ConnCB: ref %d, schd %d, conn %d, cnted %d, del %d\n",
2134 refcount_read(&conn_cb->ksnr_refcount),
2135 conn_cb->ksnr_scheduled, conn_cb->ksnr_connecting,
2136 conn_cb->ksnr_connected, conn_cb->ksnr_deleted);
2139 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
2140 CWARN("Conn: ref %d, sref %d, t %d, c %d\n",
2141 refcount_read(&conn->ksnc_conn_refcount),
2142 refcount_read(&conn->ksnc_sock_refcount),
2143 conn->ksnc_type, conn->ksnc_closing);
2148 read_unlock(&ksocknal_data.ksnd_global_lock);
2153 ksocknal_shutdown(struct lnet_ni *ni)
2155 struct ksock_net *net = ni->ni_data;
2156 struct lnet_process_id anyid = {
2157 .nid = LNET_NID_ANY,
2158 .pid = LNET_PID_ANY,
2161 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL);
2162 LASSERT(ksocknal_data.ksnd_nnets > 0);
2164 /* prevent new peers */
2165 atomic_add(SOCKNAL_SHUTDOWN_BIAS, &net->ksnn_npeers);
2167 /* Delete all peers */
2168 ksocknal_del_peer(ni, anyid, 0);
2170 /* Wait for all peer_ni state to clean up */
2171 wait_var_event_warning(&net->ksnn_npeers,
2172 atomic_read(&net->ksnn_npeers) ==
2173 SOCKNAL_SHUTDOWN_BIAS,
2174 "waiting for %d peers to disconnect\n",
2175 ksocknal_debug_peerhash(ni) +
2176 atomic_read(&net->ksnn_npeers) -
2177 SOCKNAL_SHUTDOWN_BIAS);
2179 LASSERT(net->ksnn_interface.ksni_npeers == 0);
2180 LASSERT(net->ksnn_interface.ksni_nroutes == 0);
2182 list_del(&net->ksnn_list);
2183 LIBCFS_FREE(net, sizeof(*net));
2185 ksocknal_data.ksnd_nnets--;
2186 if (ksocknal_data.ksnd_nnets == 0)
2187 ksocknal_base_shutdown();
2191 ksocknal_search_new_ipif(struct ksock_net *net)
2194 char *ifnam = &net->ksnn_interface.ksni_name[0];
2195 char *colon = strchr(ifnam, ':');
2197 struct ksock_net *tmp;
2202 list_for_each_entry(tmp, &ksocknal_data.ksnd_nets, ksnn_list) {
2203 char *ifnam2 = &tmp->ksnn_interface.ksni_name[0];
2204 char *colon2 = strchr(ifnam2, ':');
2209 found = strcmp(ifnam, ifnam2) == 0;
2222 ksocknal_start_schedulers(struct ksock_sched *sched)
2228 if (sched->kss_nthreads == 0) {
2229 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2230 nthrs = sched->kss_nthreads_max;
2232 nthrs = cfs_cpt_weight(lnet_cpt_table(),
2234 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2235 nthrs = min(SOCKNAL_NSCHEDS_HIGH, nthrs);
2237 nthrs = min(nthrs, sched->kss_nthreads_max);
2239 LASSERT(sched->kss_nthreads <= sched->kss_nthreads_max);
2240 /* increase two threads if there is new interface */
2241 nthrs = min(2, sched->kss_nthreads_max - sched->kss_nthreads);
2244 for (i = 0; i < nthrs; i++) {
2247 id = KSOCK_THREAD_ID(sched->kss_cpt, sched->kss_nthreads + i);
2248 rc = ksocknal_thread_start(ksocknal_scheduler, (void *)id,
2249 "socknal_sd%02d_%02d",
2251 (int)KSOCK_THREAD_SID(id));
2255 CERROR("Can't spawn thread %d for scheduler[%d]: %d\n",
2256 sched->kss_cpt, (int) KSOCK_THREAD_SID(id), rc);
2260 sched->kss_nthreads += i;
2265 ksocknal_net_start_threads(struct ksock_net *net, __u32 *cpts, int ncpts)
2267 int newif = ksocknal_search_new_ipif(net);
2271 if (ncpts > 0 && ncpts > cfs_cpt_number(lnet_cpt_table()))
2274 for (i = 0; i < ncpts; i++) {
2275 struct ksock_sched *sched;
2276 int cpt = (cpts == NULL) ? i : cpts[i];
2278 LASSERT(cpt < cfs_cpt_number(lnet_cpt_table()));
2279 sched = ksocknal_data.ksnd_schedulers[cpt];
2281 if (!newif && sched->kss_nthreads > 0)
2284 rc = ksocknal_start_schedulers(sched);
2292 ksocknal_startup(struct lnet_ni *ni)
2294 struct ksock_net *net;
2295 struct ksock_interface *ksi = NULL;
2296 struct lnet_inetdev *ifaces = NULL;
2297 struct sockaddr_in *sa;
2301 LASSERT (ni->ni_net->net_lnd == &the_ksocklnd);
2302 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING) {
2303 rc = ksocknal_base_startup();
2307 LIBCFS_ALLOC(net, sizeof(*net));
2310 net->ksnn_incarnation = ktime_get_real_ns();
2313 ksocknal_tunables_setup(ni);
2315 rc = lnet_inet_enumerate(&ifaces, ni->ni_net_ns);
2319 ksi = &net->ksnn_interface;
2321 /* Use the first discovered interface or look in the list */
2322 if (ni->ni_interface) {
2323 for (i = 0; i < rc; i++)
2324 if (strcmp(ifaces[i].li_name, ni->ni_interface) == 0)
2327 /* ni_interfaces doesn't contain the interface we want */
2329 CERROR("ksocklnd: failed to find interface %s\n",
2335 ni->ni_dev_cpt = ifaces[i].li_cpt;
2336 sa = (void *)&ksi->ksni_addr;
2337 memset(sa, 0, sizeof(*sa));
2338 sa->sin_family = AF_INET;
2339 sa->sin_addr.s_addr = htonl(ifaces[i].li_ipaddr);
2340 ksi->ksni_index = ksocknal_ip2index((struct sockaddr *)sa, ni);
2341 ksi->ksni_netmask = ifaces[i].li_netmask;
2342 strlcpy(ksi->ksni_name, ifaces[i].li_name, sizeof(ksi->ksni_name));
2344 /* call it before add it to ksocknal_data.ksnd_nets */
2345 rc = ksocknal_net_start_threads(net, ni->ni_cpts, ni->ni_ncpts);
2350 LASSERT(ksi->ksni_addr.ss_family == AF_INET);
2351 ni->ni_nid = LNET_MKNID(
2352 LNET_NIDNET(ni->ni_nid),
2353 ntohl(((struct sockaddr_in *)
2354 &ksi->ksni_addr)->sin_addr.s_addr));
2355 list_add(&net->ksnn_list, &ksocknal_data.ksnd_nets);
2357 ksocknal_data.ksnd_nnets++;
2362 LIBCFS_FREE(net, sizeof(*net));
2364 if (ksocknal_data.ksnd_nnets == 0)
2365 ksocknal_base_shutdown();
2370 static void __exit ksocklnd_exit(void)
2372 lnet_unregister_lnd(&the_ksocklnd);
2375 static const struct lnet_lnd the_ksocklnd = {
2376 .lnd_type = SOCKLND,
2377 .lnd_startup = ksocknal_startup,
2378 .lnd_shutdown = ksocknal_shutdown,
2379 .lnd_ctl = ksocknal_ctl,
2380 .lnd_send = ksocknal_send,
2381 .lnd_recv = ksocknal_recv,
2382 .lnd_notify_peer_down = ksocknal_notify_gw_down,
2383 .lnd_accept = ksocknal_accept,
2386 static int __init ksocklnd_init(void)
2390 /* check ksnr_connected/connecting field large enough */
2391 BUILD_BUG_ON(SOCKLND_CONN_NTYPES > 4);
2392 BUILD_BUG_ON(SOCKLND_CONN_ACK != SOCKLND_CONN_BULK_IN);
2394 rc = ksocknal_tunables_init();
2398 lnet_register_lnd(&the_ksocklnd);
2403 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2404 MODULE_DESCRIPTION("TCP Socket LNet Network Driver");
2405 MODULE_VERSION("2.8.0");
2406 MODULE_LICENSE("GPL");
2408 module_init(ksocklnd_init);
2409 module_exit(ksocklnd_exit);