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);
747 CERROR("Can't determine new connection's address\n");
751 LIBCFS_ALLOC(cr, sizeof(*cr));
753 LCONSOLE_ERROR_MSG(0x12f,
754 "Dropping connection request from %pIS: memory exhausted\n",
761 cr->ksncr_sock = sock;
763 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
765 list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
766 wake_up(&ksocknal_data.ksnd_connd_waitq);
768 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
773 ksocknal_connecting(struct ksock_conn_cb *conn_cb, struct sockaddr *sa)
776 rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr, sa))
777 return conn_cb->ksnr_connecting;
782 ksocknal_create_conn(struct lnet_ni *ni, struct ksock_conn_cb *conn_cb,
783 struct socket *sock, int type)
785 rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
787 struct lnet_process_id peerid;
788 struct list_head *tmp;
790 struct ksock_conn *conn;
791 struct ksock_conn *conn2;
792 struct ksock_peer_ni *peer_ni = NULL;
793 struct ksock_peer_ni *peer2;
794 struct ksock_sched *sched;
795 struct ksock_hello_msg *hello;
798 struct ksock_tx *txtmp;
805 active = (conn_cb != NULL);
807 LASSERT(active == (type != SOCKLND_CONN_NONE));
809 LIBCFS_ALLOC(conn, sizeof(*conn));
815 conn->ksnc_peer = NULL;
816 conn->ksnc_conn_cb = NULL;
817 conn->ksnc_sock = sock;
818 /* 2 ref, 1 for conn, another extra ref prevents socket
819 * being closed before establishment of connection */
820 refcount_set(&conn->ksnc_sock_refcount, 2);
821 conn->ksnc_type = type;
822 ksocknal_lib_save_callback(sock, conn);
823 refcount_set(&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
825 conn->ksnc_rx_ready = 0;
826 conn->ksnc_rx_scheduled = 0;
828 INIT_LIST_HEAD(&conn->ksnc_tx_queue);
829 conn->ksnc_tx_ready = 0;
830 conn->ksnc_tx_scheduled = 0;
831 conn->ksnc_tx_carrier = NULL;
832 atomic_set (&conn->ksnc_tx_nob, 0);
834 LIBCFS_ALLOC(hello, offsetof(struct ksock_hello_msg,
835 kshm_ips[LNET_INTERFACES_NUM]));
841 /* stash conn's local and remote addrs */
842 rc = ksocknal_lib_get_conn_addrs(conn);
846 /* Find out/confirm peer_ni's NID and connection type and get the
847 * vector of interfaces she's willing to let me connect to.
848 * Passive connections use the listener timeout since the peer_ni sends
853 peer_ni = conn_cb->ksnr_peer;
854 LASSERT(ni == peer_ni->ksnp_ni);
856 /* Active connection sends HELLO eagerly */
857 hello->kshm_nips = 0;
858 peerid = peer_ni->ksnp_id;
860 write_lock_bh(global_lock);
861 conn->ksnc_proto = peer_ni->ksnp_proto;
862 write_unlock_bh(global_lock);
864 if (conn->ksnc_proto == NULL) {
865 conn->ksnc_proto = &ksocknal_protocol_v3x;
866 #if SOCKNAL_VERSION_DEBUG
867 if (*ksocknal_tunables.ksnd_protocol == 2)
868 conn->ksnc_proto = &ksocknal_protocol_v2x;
869 else if (*ksocknal_tunables.ksnd_protocol == 1)
870 conn->ksnc_proto = &ksocknal_protocol_v1x;
874 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
878 peerid.nid = LNET_NID_ANY;
879 peerid.pid = LNET_PID_ANY;
881 /* Passive, get protocol from peer_ni */
882 conn->ksnc_proto = NULL;
885 rc = ksocknal_recv_hello(ni, conn, hello, &peerid, &incarnation);
889 LASSERT(rc == 0 || active);
890 LASSERT(conn->ksnc_proto != NULL);
891 LASSERT(peerid.nid != LNET_NID_ANY);
893 cpt = lnet_cpt_of_nid(peerid.nid, ni);
896 ksocknal_peer_addref(peer_ni);
897 write_lock_bh(global_lock);
899 peer_ni = ksocknal_create_peer(ni, peerid);
900 if (IS_ERR(peer_ni)) {
901 rc = PTR_ERR(peer_ni);
905 write_lock_bh(global_lock);
907 /* called with a ref on ni, so shutdown can't have started */
908 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers) >= 0);
910 peer2 = ksocknal_find_peer_locked(ni, peerid);
912 /* NB this puts an "empty" peer_ni in the peer_ni
913 * table (which takes my ref) */
914 hash_add(ksocknal_data.ksnd_peers,
915 &peer_ni->ksnp_list, peerid.nid);
917 ksocknal_peer_decref(peer_ni);
922 ksocknal_peer_addref(peer_ni);
923 peer_ni->ksnp_accepting++;
925 /* Am I already connecting to this guy? Resolve in
926 * favour of higher NID...
928 if (peerid.nid < ni->ni_nid &&
929 ksocknal_connecting(peer_ni->ksnp_conn_cb,
930 ((struct sockaddr *) &conn->ksnc_peeraddr))) {
932 warn = "connection race resolution";
937 if (peer_ni->ksnp_closing ||
938 (active && conn_cb->ksnr_deleted)) {
939 /* peer_ni/conn_cb got closed under me */
941 warn = "peer_ni/conn_cb removed";
945 if (peer_ni->ksnp_proto == NULL) {
946 /* Never connected before.
947 * NB recv_hello may have returned EPROTO to signal my peer_ni
948 * wants a different protocol than the one I asked for.
950 LASSERT(list_empty(&peer_ni->ksnp_conns));
952 peer_ni->ksnp_proto = conn->ksnc_proto;
953 peer_ni->ksnp_incarnation = incarnation;
956 if (peer_ni->ksnp_proto != conn->ksnc_proto ||
957 peer_ni->ksnp_incarnation != incarnation) {
958 /* peer_ni rebooted or I've got the wrong protocol version */
959 ksocknal_close_peer_conns_locked(peer_ni, NULL, 0);
961 peer_ni->ksnp_proto = NULL;
963 warn = peer_ni->ksnp_incarnation != incarnation ?
965 "wrong proto version";
975 warn = "lost conn race";
978 warn = "retry with different protocol version";
982 /* Refuse to duplicate an existing connection, unless this is a
983 * loopback connection */
984 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
985 (struct sockaddr *)&conn->ksnc_myaddr)) {
986 list_for_each(tmp, &peer_ni->ksnp_conns) {
987 conn2 = list_entry(tmp, struct ksock_conn, ksnc_list);
990 (struct sockaddr *)&conn2->ksnc_peeraddr,
991 (struct sockaddr *)&conn->ksnc_peeraddr) ||
993 (struct sockaddr *)&conn2->ksnc_myaddr,
994 (struct sockaddr *)&conn->ksnc_myaddr) ||
995 conn2->ksnc_type != conn->ksnc_type)
999 if (num_dup < *ksocknal_tunables.ksnd_conns_per_peer)
1002 /* Reply on a passive connection attempt so the peer_ni
1003 * realises we're connected.
1013 /* If the connection created by this route didn't bind to the IP
1014 * address the route connected to, the connection/route matching
1015 * code below probably isn't going to work.
1018 !rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr,
1019 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1020 CERROR("Route %s %pIS connected to %pIS\n",
1021 libcfs_id2str(peer_ni->ksnp_id),
1022 &conn_cb->ksnr_addr,
1023 &conn->ksnc_peeraddr);
1026 /* Search for a conn_cb corresponding to the new connection and
1027 * create an association. This allows incoming connections created
1028 * by conn_cbs in my peer_ni to match my own conn_cb entries so I don't
1029 * continually create duplicate conn_cbs.
1031 conn_cb = peer_ni->ksnp_conn_cb;
1033 if (conn_cb && rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
1034 (struct sockaddr *)&conn_cb->ksnr_addr))
1035 ksocknal_associate_cb_conn_locked(conn_cb, conn);
1037 conn->ksnc_peer = peer_ni; /* conn takes my ref on peer_ni */
1038 peer_ni->ksnp_last_alive = ktime_get_seconds();
1039 peer_ni->ksnp_send_keepalive = 0;
1040 peer_ni->ksnp_error = 0;
1042 sched = ksocknal_choose_scheduler_locked(cpt);
1044 CERROR("no schedulers available. node is unhealthy\n");
1048 * The cpt might have changed if we ended up selecting a non cpt
1049 * native scheduler. So use the scheduler's cpt instead.
1051 cpt = sched->kss_cpt;
1052 sched->kss_nconns++;
1053 conn->ksnc_scheduler = sched;
1055 conn->ksnc_tx_last_post = ktime_get_seconds();
1056 /* Set the deadline for the outgoing HELLO to drain */
1057 conn->ksnc_tx_bufnob = sock->sk->sk_wmem_queued;
1058 conn->ksnc_tx_deadline = ktime_get_seconds() +
1060 smp_mb(); /* order with adding to peer_ni's conn list */
1062 list_add(&conn->ksnc_list, &peer_ni->ksnp_conns);
1063 ksocknal_conn_addref(conn);
1065 ksocknal_new_packet(conn, 0);
1067 conn->ksnc_zc_capable = ksocknal_lib_zc_capable(conn);
1069 /* Take packets blocking for this connection. */
1070 list_for_each_entry_safe(tx, txtmp, &peer_ni->ksnp_tx_queue, tx_list) {
1071 if (conn->ksnc_proto->pro_match_tx(conn, tx, tx->tx_nonblk) ==
1075 list_del(&tx->tx_list);
1076 ksocknal_queue_tx_locked(tx, conn);
1079 write_unlock_bh(global_lock);
1081 /* We've now got a new connection. Any errors from here on are just
1082 * like "normal" comms errors and we close the connection normally.
1083 * NB (a) we still have to send the reply HELLO for passive
1085 * (b) normal I/O on the conn is blocked until I setup and call the
1089 CDEBUG(D_NET, "New conn %s p %d.x %pIS -> %pISp"
1090 " incarnation:%lld sched[%d]\n",
1091 libcfs_id2str(peerid), conn->ksnc_proto->pro_version,
1092 &conn->ksnc_myaddr, &conn->ksnc_peeraddr,
1096 hello->kshm_nips = 0;
1097 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
1100 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1101 kshm_ips[LNET_INTERFACES_NUM]));
1103 /* setup the socket AFTER I've received hello (it disables
1104 * SO_LINGER). I might call back to the acceptor who may want
1105 * to send a protocol version response and then close the
1106 * socket; this ensures the socket only tears down after the
1107 * response has been sent.
1110 rc = ksocknal_lib_setup_sock(sock);
1112 write_lock_bh(global_lock);
1114 /* NB my callbacks block while I hold ksnd_global_lock */
1115 ksocknal_lib_set_callback(sock, conn);
1118 peer_ni->ksnp_accepting--;
1120 write_unlock_bh(global_lock);
1123 write_lock_bh(global_lock);
1124 if (!conn->ksnc_closing) {
1125 /* could be closed by another thread */
1126 ksocknal_close_conn_locked(conn, rc);
1128 write_unlock_bh(global_lock);
1129 } else if (ksocknal_connsock_addref(conn) == 0) {
1130 /* Allow I/O to proceed. */
1131 ksocknal_read_callback(conn);
1132 ksocknal_write_callback(conn);
1133 ksocknal_connsock_decref(conn);
1136 ksocknal_connsock_decref(conn);
1137 ksocknal_conn_decref(conn);
1142 if (!peer_ni->ksnp_closing &&
1143 list_empty(&peer_ni->ksnp_conns) &&
1144 peer_ni->ksnp_conn_cb == NULL) {
1145 list_splice_init(&peer_ni->ksnp_tx_queue, &zombies);
1146 ksocknal_unlink_peer_locked(peer_ni);
1149 write_unlock_bh(global_lock);
1153 CERROR("Not creating conn %s type %d: %s\n",
1154 libcfs_id2str(peerid), conn->ksnc_type, warn);
1156 CDEBUG(D_NET, "Not creating conn %s type %d: %s\n",
1157 libcfs_id2str(peerid), conn->ksnc_type, warn);
1162 /* Request retry by replying with CONN_NONE
1163 * ksnc_proto has been set already
1165 conn->ksnc_type = SOCKLND_CONN_NONE;
1166 hello->kshm_nips = 0;
1167 ksocknal_send_hello(ni, conn, peerid.nid, hello);
1170 write_lock_bh(global_lock);
1171 peer_ni->ksnp_accepting--;
1172 write_unlock_bh(global_lock);
1176 * If we get here without an error code, just use -EALREADY.
1177 * Depending on how we got here, the error may be positive
1178 * or negative. Normalize the value for ksocknal_txlist_done().
1180 rc2 = (rc == 0 ? -EALREADY : (rc > 0 ? -rc : rc));
1181 ksocknal_txlist_done(ni, &zombies, rc2);
1182 ksocknal_peer_decref(peer_ni);
1186 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1187 kshm_ips[LNET_INTERFACES_NUM]));
1189 LIBCFS_FREE(conn, sizeof(*conn));
1198 ksocknal_close_conn_locked(struct ksock_conn *conn, int error)
1200 /* This just does the immmediate housekeeping, and queues the
1201 * connection for the reaper to terminate.
1202 * Caller holds ksnd_global_lock exclusively in irq context */
1203 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1204 struct ksock_conn_cb *conn_cb;
1205 struct ksock_conn *conn2;
1206 struct list_head *tmp;
1208 LASSERT(peer_ni->ksnp_error == 0);
1209 LASSERT(!conn->ksnc_closing);
1210 conn->ksnc_closing = 1;
1212 /* ksnd_deathrow_conns takes over peer_ni's ref */
1213 list_del(&conn->ksnc_list);
1215 conn_cb = conn->ksnc_conn_cb;
1216 if (conn_cb != NULL) {
1217 /* dissociate conn from cb... */
1218 LASSERT(!conn_cb->ksnr_deleted);
1220 /* connected bit is set only if all connections
1221 * of the given type got created
1223 if (ksocknal_get_conn_count_by_type(conn_cb, conn->ksnc_type) ==
1224 *ksocknal_tunables.ksnd_conns_per_peer)
1225 LASSERT((conn_cb->ksnr_connected &
1226 BIT(conn->ksnc_type)) != 0);
1229 list_for_each(tmp, &peer_ni->ksnp_conns) {
1230 conn2 = list_entry(tmp, struct ksock_conn, ksnc_list);
1232 if (conn2->ksnc_conn_cb == conn_cb &&
1233 conn2->ksnc_type == conn->ksnc_type)
1239 conn_cb->ksnr_connected &= ~BIT(conn->ksnc_type);
1241 conn->ksnc_conn_cb = NULL;
1243 /* drop conn's ref on conn_cb */
1244 ksocknal_conn_cb_decref(conn_cb);
1247 if (list_empty(&peer_ni->ksnp_conns)) {
1248 /* No more connections to this peer_ni */
1250 if (!list_empty(&peer_ni->ksnp_tx_queue)) {
1251 struct ksock_tx *tx;
1253 LASSERT(conn->ksnc_proto == &ksocknal_protocol_v3x);
1255 /* throw them to the last connection...,
1256 * these TXs will be send to /dev/null by scheduler */
1257 list_for_each_entry(tx, &peer_ni->ksnp_tx_queue,
1259 ksocknal_tx_prep(conn, tx);
1261 spin_lock_bh(&conn->ksnc_scheduler->kss_lock);
1262 list_splice_init(&peer_ni->ksnp_tx_queue,
1263 &conn->ksnc_tx_queue);
1264 spin_unlock_bh(&conn->ksnc_scheduler->kss_lock);
1267 /* renegotiate protocol version */
1268 peer_ni->ksnp_proto = NULL;
1269 /* stash last conn close reason */
1270 peer_ni->ksnp_error = error;
1272 if (peer_ni->ksnp_conn_cb == NULL) {
1273 /* I've just closed last conn belonging to a
1274 * peer_ni with no connections to it
1276 ksocknal_unlink_peer_locked(peer_ni);
1280 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1282 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_deathrow_conns);
1283 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1285 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1289 ksocknal_peer_failed(struct ksock_peer_ni *peer_ni)
1291 bool notify = false;
1292 time64_t last_alive = 0;
1294 /* There has been a connection failure or comms error; but I'll only
1295 * tell LNET I think the peer_ni is dead if it's to another kernel and
1296 * there are no connections or connection attempts in existence. */
1298 read_lock(&ksocknal_data.ksnd_global_lock);
1300 if ((peer_ni->ksnp_id.pid & LNET_PID_USERFLAG) == 0 &&
1301 list_empty(&peer_ni->ksnp_conns) &&
1302 peer_ni->ksnp_accepting == 0 &&
1303 !ksocknal_find_connecting_conn_cb_locked(peer_ni)) {
1305 last_alive = peer_ni->ksnp_last_alive;
1308 read_unlock(&ksocknal_data.ksnd_global_lock);
1311 lnet_notify(peer_ni->ksnp_ni, peer_ni->ksnp_id.nid,
1312 false, false, last_alive);
1316 ksocknal_finalize_zcreq(struct ksock_conn *conn)
1318 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1319 struct ksock_tx *tx;
1320 struct ksock_tx *tmp;
1323 /* NB safe to finalize TXs because closing of socket will
1324 * abort all buffered data */
1325 LASSERT(conn->ksnc_sock == NULL);
1327 spin_lock(&peer_ni->ksnp_lock);
1329 list_for_each_entry_safe(tx, tmp, &peer_ni->ksnp_zc_req_list, tx_zc_list) {
1330 if (tx->tx_conn != conn)
1333 LASSERT(tx->tx_msg.ksm_zc_cookies[0] != 0);
1335 tx->tx_msg.ksm_zc_cookies[0] = 0;
1336 tx->tx_zc_aborted = 1; /* mark it as not-acked */
1337 list_move(&tx->tx_zc_list, &zlist);
1340 spin_unlock(&peer_ni->ksnp_lock);
1342 while ((tx = list_first_entry_or_null(&zlist, struct ksock_tx,
1343 tx_zc_list)) != NULL) {
1344 list_del(&tx->tx_zc_list);
1345 ksocknal_tx_decref(tx);
1350 ksocknal_terminate_conn(struct ksock_conn *conn)
1352 /* This gets called by the reaper (guaranteed thread context) to
1353 * disengage the socket from its callbacks and close it.
1354 * ksnc_refcount will eventually hit zero, and then the reaper will
1357 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1358 struct ksock_sched *sched = conn->ksnc_scheduler;
1359 bool failed = false;
1361 LASSERT(conn->ksnc_closing);
1363 /* wake up the scheduler to "send" all remaining packets to /dev/null */
1364 spin_lock_bh(&sched->kss_lock);
1366 /* a closing conn is always ready to tx */
1367 conn->ksnc_tx_ready = 1;
1369 if (!conn->ksnc_tx_scheduled &&
1370 !list_empty(&conn->ksnc_tx_queue)) {
1371 list_add_tail(&conn->ksnc_tx_list,
1372 &sched->kss_tx_conns);
1373 conn->ksnc_tx_scheduled = 1;
1374 /* extra ref for scheduler */
1375 ksocknal_conn_addref(conn);
1377 wake_up(&sched->kss_waitq);
1380 spin_unlock_bh(&sched->kss_lock);
1382 /* serialise with callbacks */
1383 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1385 ksocknal_lib_reset_callback(conn->ksnc_sock, conn);
1387 /* OK, so this conn may not be completely disengaged from its
1388 * scheduler yet, but it _has_ committed to terminate...
1390 conn->ksnc_scheduler->kss_nconns--;
1392 if (peer_ni->ksnp_error != 0) {
1393 /* peer_ni's last conn closed in error */
1394 LASSERT(list_empty(&peer_ni->ksnp_conns));
1396 peer_ni->ksnp_error = 0; /* avoid multiple notifications */
1399 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1402 ksocknal_peer_failed(peer_ni);
1404 /* The socket is closed on the final put; either here, or in
1405 * ksocknal_{send,recv}msg(). Since we set up the linger2 option
1406 * when the connection was established, this will close the socket
1407 * immediately, aborting anything buffered in it. Any hung
1408 * zero-copy transmits will therefore complete in finite time.
1410 ksocknal_connsock_decref(conn);
1414 ksocknal_queue_zombie_conn(struct ksock_conn *conn)
1416 /* Queue the conn for the reaper to destroy */
1417 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1418 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1420 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
1421 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1423 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1427 ksocknal_destroy_conn(struct ksock_conn *conn)
1431 /* Final coup-de-grace of the reaper */
1432 CDEBUG(D_NET, "connection %p\n", conn);
1434 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1435 LASSERT(refcount_read(&conn->ksnc_sock_refcount) == 0);
1436 LASSERT(conn->ksnc_sock == NULL);
1437 LASSERT(conn->ksnc_conn_cb == NULL);
1438 LASSERT(!conn->ksnc_tx_scheduled);
1439 LASSERT(!conn->ksnc_rx_scheduled);
1440 LASSERT(list_empty(&conn->ksnc_tx_queue));
1442 /* complete current receive if any */
1443 switch (conn->ksnc_rx_state) {
1444 case SOCKNAL_RX_LNET_PAYLOAD:
1445 last_rcv = conn->ksnc_rx_deadline -
1447 CERROR("Completing partial receive from %s[%d], ip %pISp, with error, wanted: %d, left: %d, last alive is %lld secs ago\n",
1448 libcfs_id2str(conn->ksnc_peer->ksnp_id), conn->ksnc_type,
1449 &conn->ksnc_peeraddr,
1450 conn->ksnc_rx_nob_wanted, conn->ksnc_rx_nob_left,
1451 ktime_get_seconds() - last_rcv);
1452 if (conn->ksnc_lnet_msg)
1453 conn->ksnc_lnet_msg->msg_health_status =
1454 LNET_MSG_STATUS_REMOTE_ERROR;
1455 lnet_finalize(conn->ksnc_lnet_msg, -EIO);
1457 case SOCKNAL_RX_LNET_HEADER:
1458 if (conn->ksnc_rx_started)
1459 CERROR("Incomplete receive of lnet header from %s, ip %pISp, with error, protocol: %d.x.\n",
1460 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1461 &conn->ksnc_peeraddr,
1462 conn->ksnc_proto->pro_version);
1464 case SOCKNAL_RX_KSM_HEADER:
1465 if (conn->ksnc_rx_started)
1466 CERROR("Incomplete receive of ksock message from %s, ip %pISp, with error, protocol: %d.x.\n",
1467 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1468 &conn->ksnc_peeraddr,
1469 conn->ksnc_proto->pro_version);
1471 case SOCKNAL_RX_SLOP:
1472 if (conn->ksnc_rx_started)
1473 CERROR("Incomplete receive of slops from %s, ip %pISp, with error\n",
1474 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1475 &conn->ksnc_peeraddr);
1482 ksocknal_peer_decref(conn->ksnc_peer);
1484 LIBCFS_FREE (conn, sizeof (*conn));
1488 ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
1489 struct sockaddr *addr, int why)
1491 struct ksock_conn *conn;
1492 struct ksock_conn *cnxt;
1495 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
1498 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1500 ksocknal_close_conn_locked(conn, why);
1508 ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why)
1510 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1513 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1515 count = ksocknal_close_peer_conns_locked(
1516 peer_ni, (struct sockaddr *)&conn->ksnc_peeraddr, why);
1518 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1524 ksocknal_close_matching_conns(struct lnet_process_id id, __u32 ipaddr)
1526 struct ksock_peer_ni *peer_ni;
1527 struct hlist_node *pnxt;
1532 struct sockaddr_in sa = {.sin_family = AF_INET};
1534 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1536 if (id.nid != LNET_NID_ANY) {
1537 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
1541 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1544 sa.sin_addr.s_addr = htonl(ipaddr);
1545 for (i = lo; i <= hi; i++) {
1546 hlist_for_each_entry_safe(peer_ni, pnxt,
1547 &ksocknal_data.ksnd_peers[i],
1550 if (!((id.nid == LNET_NID_ANY ||
1551 id.nid == peer_ni->ksnp_id.nid) &&
1552 (id.pid == LNET_PID_ANY ||
1553 id.pid == peer_ni->ksnp_id.pid)))
1556 count += ksocknal_close_peer_conns_locked(
1558 ipaddr ? (struct sockaddr *)&sa : NULL, 0);
1562 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1564 /* wildcards always succeed */
1565 if (id.nid == LNET_NID_ANY || id.pid == LNET_PID_ANY || ipaddr == 0)
1568 return (count == 0 ? -ENOENT : 0);
1572 ksocknal_notify_gw_down(lnet_nid_t gw_nid)
1574 /* The router is telling me she's been notified of a change in
1577 struct lnet_process_id id = {
1579 .pid = LNET_PID_ANY,
1582 CDEBUG(D_NET, "gw %s down\n", libcfs_nid2str(gw_nid));
1584 /* If the gateway crashed, close all open connections... */
1585 ksocknal_close_matching_conns(id, 0);
1588 /* We can only establish new connections
1589 * if we have autroutes, and these connect on demand. */
1593 ksocknal_push_peer(struct ksock_peer_ni *peer_ni)
1597 struct list_head *tmp;
1598 struct ksock_conn *conn;
1600 for (index = 0; ; index++) {
1601 read_lock(&ksocknal_data.ksnd_global_lock);
1606 list_for_each(tmp, &peer_ni->ksnp_conns) {
1608 conn = list_entry(tmp, struct ksock_conn,
1610 ksocknal_conn_addref(conn);
1615 read_unlock(&ksocknal_data.ksnd_global_lock);
1620 ksocknal_lib_push_conn (conn);
1621 ksocknal_conn_decref(conn);
1626 ksocknal_push(struct lnet_ni *ni, struct lnet_process_id id)
1633 if (id.nid != LNET_NID_ANY) {
1634 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
1638 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1641 for (bkt = lo; bkt <= hi; bkt++) {
1642 int peer_off; /* searching offset in peer_ni hash table */
1644 for (peer_off = 0; ; peer_off++) {
1645 struct ksock_peer_ni *peer_ni;
1648 read_lock(&ksocknal_data.ksnd_global_lock);
1649 hlist_for_each_entry(peer_ni,
1650 &ksocknal_data.ksnd_peers[bkt],
1652 if (!((id.nid == LNET_NID_ANY ||
1653 id.nid == peer_ni->ksnp_id.nid) &&
1654 (id.pid == LNET_PID_ANY ||
1655 id.pid == peer_ni->ksnp_id.pid)))
1658 if (i++ == peer_off) {
1659 ksocknal_peer_addref(peer_ni);
1663 read_unlock(&ksocknal_data.ksnd_global_lock);
1665 if (i <= peer_off) /* no match */
1669 ksocknal_push_peer(peer_ni);
1670 ksocknal_peer_decref(peer_ni);
1677 ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg)
1679 struct lnet_process_id id = {0};
1680 struct libcfs_ioctl_data *data = arg;
1684 case IOC_LIBCFS_GET_INTERFACE: {
1685 struct ksock_net *net = ni->ni_data;
1686 struct ksock_interface *iface;
1687 struct sockaddr_in *sa;
1689 read_lock(&ksocknal_data.ksnd_global_lock);
1691 if (data->ioc_count >= 1) {
1695 iface = &net->ksnn_interface;
1697 sa = (void *)&iface->ksni_addr;
1698 if (sa->sin_family == AF_INET)
1699 data->ioc_u32[0] = ntohl(sa->sin_addr.s_addr);
1701 data->ioc_u32[0] = 0xFFFFFFFF;
1702 data->ioc_u32[1] = iface->ksni_netmask;
1703 data->ioc_u32[2] = iface->ksni_npeers;
1704 data->ioc_u32[3] = iface->ksni_nroutes;
1707 read_unlock(&ksocknal_data.ksnd_global_lock);
1711 case IOC_LIBCFS_GET_PEER: {
1716 int share_count = 0;
1718 rc = ksocknal_get_peer_info(ni, data->ioc_count,
1719 &id, &myip, &ip, &port,
1720 &conn_count, &share_count);
1724 data->ioc_nid = id.nid;
1725 data->ioc_count = share_count;
1726 data->ioc_u32[0] = ip;
1727 data->ioc_u32[1] = port;
1728 data->ioc_u32[2] = myip;
1729 data->ioc_u32[3] = conn_count;
1730 data->ioc_u32[4] = id.pid;
1734 case IOC_LIBCFS_ADD_PEER: {
1735 struct sockaddr_in sa = {.sin_family = AF_INET};
1737 id.nid = data->ioc_nid;
1738 id.pid = LNET_PID_LUSTRE;
1739 sa.sin_addr.s_addr = htonl(data->ioc_u32[0]);
1740 sa.sin_port = htons(data->ioc_u32[1]);
1741 return ksocknal_add_peer(ni, id, (struct sockaddr *)&sa);
1743 case IOC_LIBCFS_DEL_PEER:
1744 id.nid = data->ioc_nid;
1745 id.pid = LNET_PID_ANY;
1746 return ksocknal_del_peer (ni, id,
1747 data->ioc_u32[0]); /* IP */
1749 case IOC_LIBCFS_GET_CONN: {
1753 struct ksock_conn *conn = ksocknal_get_conn_by_idx(ni, data->ioc_count);
1754 struct sockaddr_in *psa = (void *)&conn->ksnc_peeraddr;
1755 struct sockaddr_in *mysa = (void *)&conn->ksnc_myaddr;
1760 ksocknal_lib_get_conn_tunables(conn, &txmem, &rxmem, &nagle);
1762 data->ioc_count = txmem;
1763 data->ioc_nid = conn->ksnc_peer->ksnp_id.nid;
1764 data->ioc_flags = nagle;
1765 if (psa->sin_family == AF_INET)
1766 data->ioc_u32[0] = ntohl(psa->sin_addr.s_addr);
1768 data->ioc_u32[0] = 0xFFFFFFFF;
1769 data->ioc_u32[1] = rpc_get_port((struct sockaddr *)
1770 &conn->ksnc_peeraddr);
1771 if (mysa->sin_family == AF_INET)
1772 data->ioc_u32[2] = ntohl(mysa->sin_addr.s_addr);
1774 data->ioc_u32[2] = 0xFFFFFFFF;
1775 data->ioc_u32[3] = conn->ksnc_type;
1776 data->ioc_u32[4] = conn->ksnc_scheduler->kss_cpt;
1777 data->ioc_u32[5] = rxmem;
1778 data->ioc_u32[6] = conn->ksnc_peer->ksnp_id.pid;
1779 ksocknal_conn_decref(conn);
1783 case IOC_LIBCFS_CLOSE_CONNECTION:
1784 id.nid = data->ioc_nid;
1785 id.pid = LNET_PID_ANY;
1786 return ksocknal_close_matching_conns (id,
1789 case IOC_LIBCFS_REGISTER_MYNID:
1790 /* Ignore if this is a noop */
1791 if (data->ioc_nid == ni->ni_nid)
1794 CERROR("obsolete IOC_LIBCFS_REGISTER_MYNID: %s(%s)\n",
1795 libcfs_nid2str(data->ioc_nid),
1796 libcfs_nid2str(ni->ni_nid));
1799 case IOC_LIBCFS_PUSH_CONNECTION:
1800 id.nid = data->ioc_nid;
1801 id.pid = LNET_PID_ANY;
1802 return ksocknal_push(ni, id);
1811 ksocknal_free_buffers (void)
1813 LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
1815 if (ksocknal_data.ksnd_schedulers != NULL)
1816 cfs_percpt_free(ksocknal_data.ksnd_schedulers);
1818 spin_lock(&ksocknal_data.ksnd_tx_lock);
1820 if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
1822 struct ksock_tx *tx;
1824 list_splice_init(&ksocknal_data.ksnd_idle_noop_txs, &zlist);
1825 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1827 while ((tx = list_first_entry_or_null(&zlist, struct ksock_tx,
1828 tx_list)) != NULL) {
1829 list_del(&tx->tx_list);
1830 LIBCFS_FREE(tx, tx->tx_desc_size);
1833 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1838 ksocknal_base_shutdown(void)
1840 struct ksock_sched *sched;
1841 struct ksock_peer_ni *peer_ni;
1844 CDEBUG(D_MALLOC, "before NAL cleanup: kmem %lld\n",
1845 libcfs_kmem_read());
1846 LASSERT (ksocknal_data.ksnd_nnets == 0);
1848 switch (ksocknal_data.ksnd_init) {
1853 case SOCKNAL_INIT_ALL:
1854 case SOCKNAL_INIT_DATA:
1855 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list)
1858 LASSERT(list_empty(&ksocknal_data.ksnd_nets));
1859 LASSERT(list_empty(&ksocknal_data.ksnd_enomem_conns));
1860 LASSERT(list_empty(&ksocknal_data.ksnd_zombie_conns));
1861 LASSERT(list_empty(&ksocknal_data.ksnd_connd_connreqs));
1862 LASSERT(list_empty(&ksocknal_data.ksnd_connd_routes));
1864 if (ksocknal_data.ksnd_schedulers != NULL) {
1865 cfs_percpt_for_each(sched, i,
1866 ksocknal_data.ksnd_schedulers) {
1868 LASSERT(list_empty(&sched->kss_tx_conns));
1869 LASSERT(list_empty(&sched->kss_rx_conns));
1870 LASSERT(list_empty(&sched->kss_zombie_noop_txs));
1871 LASSERT(sched->kss_nconns == 0);
1875 /* flag threads to terminate; wake and wait for them to die */
1876 ksocknal_data.ksnd_shuttingdown = 1;
1877 wake_up_all(&ksocknal_data.ksnd_connd_waitq);
1878 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1880 if (ksocknal_data.ksnd_schedulers != NULL) {
1881 cfs_percpt_for_each(sched, i,
1882 ksocknal_data.ksnd_schedulers)
1883 wake_up_all(&sched->kss_waitq);
1886 wait_var_event_warning(&ksocknal_data.ksnd_nthreads,
1887 atomic_read(&ksocknal_data.ksnd_nthreads) == 0,
1888 "waiting for %d threads to terminate\n",
1889 atomic_read(&ksocknal_data.ksnd_nthreads));
1891 ksocknal_free_buffers();
1893 ksocknal_data.ksnd_init = SOCKNAL_INIT_NOTHING;
1897 CDEBUG(D_MALLOC, "after NAL cleanup: kmem %lld\n",
1898 libcfs_kmem_read());
1900 module_put(THIS_MODULE);
1904 ksocknal_base_startup(void)
1906 struct ksock_sched *sched;
1910 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
1911 LASSERT(ksocknal_data.ksnd_nnets == 0);
1913 memset(&ksocknal_data, 0, sizeof(ksocknal_data)); /* zero pointers */
1915 hash_init(ksocknal_data.ksnd_peers);
1917 rwlock_init(&ksocknal_data.ksnd_global_lock);
1918 INIT_LIST_HEAD(&ksocknal_data.ksnd_nets);
1920 spin_lock_init(&ksocknal_data.ksnd_reaper_lock);
1921 INIT_LIST_HEAD(&ksocknal_data.ksnd_enomem_conns);
1922 INIT_LIST_HEAD(&ksocknal_data.ksnd_zombie_conns);
1923 INIT_LIST_HEAD(&ksocknal_data.ksnd_deathrow_conns);
1924 init_waitqueue_head(&ksocknal_data.ksnd_reaper_waitq);
1926 spin_lock_init(&ksocknal_data.ksnd_connd_lock);
1927 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_connreqs);
1928 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_routes);
1929 init_waitqueue_head(&ksocknal_data.ksnd_connd_waitq);
1931 spin_lock_init(&ksocknal_data.ksnd_tx_lock);
1932 INIT_LIST_HEAD(&ksocknal_data.ksnd_idle_noop_txs);
1934 /* NB memset above zeros whole of ksocknal_data */
1936 /* flag lists/ptrs/locks initialised */
1937 ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
1938 if (!try_module_get(THIS_MODULE))
1941 /* Create a scheduler block per available CPT */
1942 ksocknal_data.ksnd_schedulers = cfs_percpt_alloc(lnet_cpt_table(),
1944 if (ksocknal_data.ksnd_schedulers == NULL)
1947 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
1951 * make sure not to allocate more threads than there are
1952 * cores/CPUs in teh CPT
1954 nthrs = cfs_cpt_weight(lnet_cpt_table(), i);
1955 if (*ksocknal_tunables.ksnd_nscheds > 0) {
1956 nthrs = min(nthrs, *ksocknal_tunables.ksnd_nscheds);
1959 * max to half of CPUs, assume another half should be
1960 * reserved for upper layer modules
1962 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
1965 sched->kss_nthreads_max = nthrs;
1968 spin_lock_init(&sched->kss_lock);
1969 INIT_LIST_HEAD(&sched->kss_rx_conns);
1970 INIT_LIST_HEAD(&sched->kss_tx_conns);
1971 INIT_LIST_HEAD(&sched->kss_zombie_noop_txs);
1972 init_waitqueue_head(&sched->kss_waitq);
1975 ksocknal_data.ksnd_connd_starting = 0;
1976 ksocknal_data.ksnd_connd_failed_stamp = 0;
1977 ksocknal_data.ksnd_connd_starting_stamp = ktime_get_real_seconds();
1978 /* must have at least 2 connds to remain responsive to accepts while
1980 if (*ksocknal_tunables.ksnd_nconnds < SOCKNAL_CONND_RESV + 1)
1981 *ksocknal_tunables.ksnd_nconnds = SOCKNAL_CONND_RESV + 1;
1983 if (*ksocknal_tunables.ksnd_nconnds_max <
1984 *ksocknal_tunables.ksnd_nconnds) {
1985 ksocknal_tunables.ksnd_nconnds_max =
1986 ksocknal_tunables.ksnd_nconnds;
1989 for (i = 0; i < *ksocknal_tunables.ksnd_nconnds; i++) {
1991 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
1992 ksocknal_data.ksnd_connd_starting++;
1993 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
1996 snprintf(name, sizeof(name), "socknal_cd%02d", i);
1997 rc = ksocknal_thread_start(ksocknal_connd,
1998 (void *)((uintptr_t)i), name);
2000 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2001 ksocknal_data.ksnd_connd_starting--;
2002 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2003 CERROR("Can't spawn socknal connd: %d\n", rc);
2008 rc = ksocknal_thread_start(ksocknal_reaper, NULL, "socknal_reaper");
2010 CERROR ("Can't spawn socknal reaper: %d\n", rc);
2014 /* flag everything initialised */
2015 ksocknal_data.ksnd_init = SOCKNAL_INIT_ALL;
2020 ksocknal_base_shutdown();
2025 ksocknal_debug_peerhash(struct lnet_ni *ni)
2027 struct ksock_peer_ni *peer_ni;
2030 read_lock(&ksocknal_data.ksnd_global_lock);
2032 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
2033 struct ksock_conn_cb *conn_cb;
2034 struct ksock_conn *conn;
2036 if (peer_ni->ksnp_ni != ni)
2039 CWARN("Active peer_ni on shutdown: %s, ref %d, "
2040 "closing %d, accepting %d, err %d, zcookie %llu, "
2041 "txq %d, zc_req %d\n", libcfs_id2str(peer_ni->ksnp_id),
2042 refcount_read(&peer_ni->ksnp_refcount),
2043 peer_ni->ksnp_closing,
2044 peer_ni->ksnp_accepting, peer_ni->ksnp_error,
2045 peer_ni->ksnp_zc_next_cookie,
2046 !list_empty(&peer_ni->ksnp_tx_queue),
2047 !list_empty(&peer_ni->ksnp_zc_req_list));
2049 conn_cb = peer_ni->ksnp_conn_cb;
2051 CWARN("ConnCB: ref %d, schd %d, conn %d, cnted %d, del %d\n",
2052 refcount_read(&conn_cb->ksnr_refcount),
2053 conn_cb->ksnr_scheduled, conn_cb->ksnr_connecting,
2054 conn_cb->ksnr_connected, conn_cb->ksnr_deleted);
2057 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
2058 CWARN("Conn: ref %d, sref %d, t %d, c %d\n",
2059 refcount_read(&conn->ksnc_conn_refcount),
2060 refcount_read(&conn->ksnc_sock_refcount),
2061 conn->ksnc_type, conn->ksnc_closing);
2066 read_unlock(&ksocknal_data.ksnd_global_lock);
2071 ksocknal_shutdown(struct lnet_ni *ni)
2073 struct ksock_net *net = ni->ni_data;
2074 struct lnet_process_id anyid = {
2075 .nid = LNET_NID_ANY,
2076 .pid = LNET_PID_ANY,
2079 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL);
2080 LASSERT(ksocknal_data.ksnd_nnets > 0);
2082 /* prevent new peers */
2083 atomic_add(SOCKNAL_SHUTDOWN_BIAS, &net->ksnn_npeers);
2085 /* Delete all peers */
2086 ksocknal_del_peer(ni, anyid, 0);
2088 /* Wait for all peer_ni state to clean up */
2089 wait_var_event_warning(&net->ksnn_npeers,
2090 atomic_read(&net->ksnn_npeers) ==
2091 SOCKNAL_SHUTDOWN_BIAS,
2092 "waiting for %d peers to disconnect\n",
2093 ksocknal_debug_peerhash(ni) +
2094 atomic_read(&net->ksnn_npeers) -
2095 SOCKNAL_SHUTDOWN_BIAS);
2097 LASSERT(net->ksnn_interface.ksni_npeers == 0);
2098 LASSERT(net->ksnn_interface.ksni_nroutes == 0);
2100 list_del(&net->ksnn_list);
2101 LIBCFS_FREE(net, sizeof(*net));
2103 ksocknal_data.ksnd_nnets--;
2104 if (ksocknal_data.ksnd_nnets == 0)
2105 ksocknal_base_shutdown();
2109 ksocknal_search_new_ipif(struct ksock_net *net)
2112 char *ifnam = &net->ksnn_interface.ksni_name[0];
2113 char *colon = strchr(ifnam, ':');
2115 struct ksock_net *tmp;
2120 list_for_each_entry(tmp, &ksocknal_data.ksnd_nets, ksnn_list) {
2121 char *ifnam2 = &tmp->ksnn_interface.ksni_name[0];
2122 char *colon2 = strchr(ifnam2, ':');
2127 found = strcmp(ifnam, ifnam2) == 0;
2140 ksocknal_start_schedulers(struct ksock_sched *sched)
2146 if (sched->kss_nthreads == 0) {
2147 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2148 nthrs = sched->kss_nthreads_max;
2150 nthrs = cfs_cpt_weight(lnet_cpt_table(),
2152 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2153 nthrs = min(SOCKNAL_NSCHEDS_HIGH, nthrs);
2155 nthrs = min(nthrs, sched->kss_nthreads_max);
2157 LASSERT(sched->kss_nthreads <= sched->kss_nthreads_max);
2158 /* increase two threads if there is new interface */
2159 nthrs = min(2, sched->kss_nthreads_max - sched->kss_nthreads);
2162 for (i = 0; i < nthrs; i++) {
2166 id = KSOCK_THREAD_ID(sched->kss_cpt, sched->kss_nthreads + i);
2167 snprintf(name, sizeof(name), "socknal_sd%02d_%02d",
2168 sched->kss_cpt, (int)KSOCK_THREAD_SID(id));
2170 rc = ksocknal_thread_start(ksocknal_scheduler,
2175 CERROR("Can't spawn thread %d for scheduler[%d]: %d\n",
2176 sched->kss_cpt, (int) KSOCK_THREAD_SID(id), rc);
2180 sched->kss_nthreads += i;
2185 ksocknal_net_start_threads(struct ksock_net *net, __u32 *cpts, int ncpts)
2187 int newif = ksocknal_search_new_ipif(net);
2191 if (ncpts > 0 && ncpts > cfs_cpt_number(lnet_cpt_table()))
2194 for (i = 0; i < ncpts; i++) {
2195 struct ksock_sched *sched;
2196 int cpt = (cpts == NULL) ? i : cpts[i];
2198 LASSERT(cpt < cfs_cpt_number(lnet_cpt_table()));
2199 sched = ksocknal_data.ksnd_schedulers[cpt];
2201 if (!newif && sched->kss_nthreads > 0)
2204 rc = ksocknal_start_schedulers(sched);
2212 ksocknal_startup(struct lnet_ni *ni)
2214 struct ksock_net *net;
2215 struct lnet_ioctl_config_lnd_cmn_tunables *net_tunables;
2216 struct ksock_interface *ksi = NULL;
2217 struct lnet_inetdev *ifaces = NULL;
2218 struct sockaddr_in *sa;
2222 LASSERT (ni->ni_net->net_lnd == &the_ksocklnd);
2223 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING) {
2224 rc = ksocknal_base_startup();
2228 LIBCFS_ALLOC(net, sizeof(*net));
2231 net->ksnn_incarnation = ktime_get_real_ns();
2233 net_tunables = &ni->ni_net->net_tunables;
2234 if (net_tunables->lct_peer_timeout == -1)
2235 net_tunables->lct_peer_timeout =
2236 *ksocknal_tunables.ksnd_peertimeout;
2238 if (net_tunables->lct_max_tx_credits == -1)
2239 net_tunables->lct_max_tx_credits =
2240 *ksocknal_tunables.ksnd_credits;
2242 if (net_tunables->lct_peer_tx_credits == -1)
2243 net_tunables->lct_peer_tx_credits =
2244 *ksocknal_tunables.ksnd_peertxcredits;
2246 if (net_tunables->lct_peer_tx_credits >
2247 net_tunables->lct_max_tx_credits)
2248 net_tunables->lct_peer_tx_credits =
2249 net_tunables->lct_max_tx_credits;
2251 if (net_tunables->lct_peer_rtr_credits == -1)
2252 net_tunables->lct_peer_rtr_credits =
2253 *ksocknal_tunables.ksnd_peerrtrcredits;
2255 rc = lnet_inet_enumerate(&ifaces, ni->ni_net_ns);
2259 ksi = &net->ksnn_interface;
2261 /* Use the first discovered interface or look in the list */
2262 if (ni->ni_interface) {
2263 for (i = 0; i < rc; i++)
2264 if (strcmp(ifaces[i].li_name, ni->ni_interface) == 0)
2267 /* ni_interfaces doesn't contain the interface we want */
2269 CERROR("ksocklnd: failed to find interface %s\n",
2275 ni->ni_dev_cpt = ifaces[i].li_cpt;
2276 sa = (void *)&ksi->ksni_addr;
2277 memset(sa, 0, sizeof(*sa));
2278 sa->sin_family = AF_INET;
2279 sa->sin_addr.s_addr = htonl(ifaces[i].li_ipaddr);
2280 ksi->ksni_index = ksocknal_ip2index((struct sockaddr *)sa, ni);
2281 ksi->ksni_netmask = ifaces[i].li_netmask;
2282 strlcpy(ksi->ksni_name, ifaces[i].li_name, sizeof(ksi->ksni_name));
2284 /* call it before add it to ksocknal_data.ksnd_nets */
2285 rc = ksocknal_net_start_threads(net, ni->ni_cpts, ni->ni_ncpts);
2290 LASSERT(ksi->ksni_addr.ss_family == AF_INET);
2291 ni->ni_nid = LNET_MKNID(
2292 LNET_NIDNET(ni->ni_nid),
2293 ntohl(((struct sockaddr_in *)
2294 &ksi->ksni_addr)->sin_addr.s_addr));
2295 list_add(&net->ksnn_list, &ksocknal_data.ksnd_nets);
2296 ksocknal_data.ksnd_nnets++;
2301 LIBCFS_FREE(net, sizeof(*net));
2303 if (ksocknal_data.ksnd_nnets == 0)
2304 ksocknal_base_shutdown();
2310 static void __exit ksocklnd_exit(void)
2312 lnet_unregister_lnd(&the_ksocklnd);
2315 static const struct lnet_lnd the_ksocklnd = {
2316 .lnd_type = SOCKLND,
2317 .lnd_startup = ksocknal_startup,
2318 .lnd_shutdown = ksocknal_shutdown,
2319 .lnd_ctl = ksocknal_ctl,
2320 .lnd_send = ksocknal_send,
2321 .lnd_recv = ksocknal_recv,
2322 .lnd_notify_peer_down = ksocknal_notify_gw_down,
2323 .lnd_accept = ksocknal_accept,
2326 static int __init ksocklnd_init(void)
2330 /* check ksnr_connected/connecting field large enough */
2331 BUILD_BUG_ON(SOCKLND_CONN_NTYPES > 4);
2332 BUILD_BUG_ON(SOCKLND_CONN_ACK != SOCKLND_CONN_BULK_IN);
2334 rc = ksocknal_tunables_init();
2338 lnet_register_lnd(&the_ksocklnd);
2343 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2344 MODULE_DESCRIPTION("TCP Socket LNet Network Driver");
2345 MODULE_VERSION("2.8.0");
2346 MODULE_LICENSE("GPL");
2348 module_init(ksocklnd_init);
2349 module_exit(ksocklnd_exit);