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 ksock_conn *conn;
490 struct ksock_net *net = peer_ni->ksnp_ni->ni_data;
492 LASSERT(!peer_ni->ksnp_closing);
493 LASSERT(!conn_cb->ksnr_peer);
494 LASSERT(!conn_cb->ksnr_scheduled);
495 LASSERT(!conn_cb->ksnr_connecting);
496 LASSERT(conn_cb->ksnr_connected == 0);
498 conn_cb->ksnr_peer = peer_ni;
499 ksocknal_peer_addref(peer_ni);
501 /* set the conn_cb's interface to the current net's interface */
502 conn_cb->ksnr_myiface = net->ksnn_interface.ksni_index;
503 net->ksnn_interface.ksni_nroutes++;
505 /* peer_ni's route list takes over my ref on 'route' */
506 peer_ni->ksnp_conn_cb = conn_cb;
508 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
509 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
510 (struct sockaddr *)&conn_cb->ksnr_addr))
513 ksocknal_associate_cb_conn_locked(conn_cb, conn);
514 /* keep going (typed conns) */
519 ksocknal_del_conn_cb_locked(struct ksock_conn_cb *conn_cb)
521 struct ksock_peer_ni *peer_ni = conn_cb->ksnr_peer;
522 struct ksock_interface *iface;
523 struct ksock_conn *conn;
524 struct ksock_conn *cnxt;
526 LASSERT(!conn_cb->ksnr_deleted);
528 /* Close associated conns */
529 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
530 if (conn->ksnc_conn_cb != conn_cb)
533 ksocknal_close_conn_locked(conn, 0);
536 if (conn_cb->ksnr_myiface >= 0) {
537 iface = ksocknal_index2iface(peer_ni->ksnp_ni,
538 conn_cb->ksnr_myiface);
540 iface->ksni_nroutes--;
543 conn_cb->ksnr_deleted = 1;
544 ksocknal_conn_cb_decref(conn_cb); /* drop peer_ni's ref */
545 peer_ni->ksnp_conn_cb = NULL;
547 if (list_empty(&peer_ni->ksnp_conns)) {
548 /* I've just removed the last route to a peer_ni with no active
551 ksocknal_unlink_peer_locked(peer_ni);
556 ksocknal_add_peer(struct lnet_ni *ni, struct lnet_process_id id,
557 struct sockaddr *addr)
559 struct ksock_peer_ni *peer_ni;
560 struct ksock_peer_ni *peer2;
561 struct ksock_conn_cb *conn_cb;
563 if (id.nid == LNET_NID_ANY ||
564 id.pid == LNET_PID_ANY)
567 /* Have a brand new peer_ni ready... */
568 peer_ni = ksocknal_create_peer(ni, id);
570 return PTR_ERR(peer_ni);
572 conn_cb = ksocknal_create_conn_cb(addr);
574 ksocknal_peer_decref(peer_ni);
578 write_lock_bh(&ksocknal_data.ksnd_global_lock);
580 /* always called with a ref on ni, so shutdown can't have started */
581 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers)
584 peer2 = ksocknal_find_peer_locked(ni, id);
586 ksocknal_peer_decref(peer_ni);
589 /* peer_ni table takes my ref on peer_ni */
590 hash_add(ksocknal_data.ksnd_peers, &peer_ni->ksnp_list, id.nid);
593 ksocknal_add_conn_cb_locked(peer_ni, conn_cb);
595 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
601 ksocknal_del_peer_locked(struct ksock_peer_ni *peer_ni, __u32 ip)
603 struct ksock_conn *conn;
604 struct ksock_conn *cnxt;
605 struct ksock_conn_cb *conn_cb;
607 LASSERT(!peer_ni->ksnp_closing);
609 /* Extra ref prevents peer_ni disappearing until I'm done with it */
610 ksocknal_peer_addref(peer_ni);
611 conn_cb = peer_ni->ksnp_conn_cb;
613 ksocknal_del_conn_cb_locked(conn_cb);
615 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns,
617 ksocknal_close_conn_locked(conn, 0);
619 ksocknal_peer_decref(peer_ni);
620 /* NB peer_ni unlinks itself when last conn/conn_cb is removed */
624 ksocknal_del_peer(struct lnet_ni *ni, struct lnet_process_id id, __u32 ip)
627 struct hlist_node *pnxt;
628 struct ksock_peer_ni *peer_ni;
634 write_lock_bh(&ksocknal_data.ksnd_global_lock);
636 if (id.nid != LNET_NID_ANY) {
637 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
641 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
644 for (i = lo; i <= hi; i++) {
645 hlist_for_each_entry_safe(peer_ni, pnxt,
646 &ksocknal_data.ksnd_peers[i],
648 if (peer_ni->ksnp_ni != ni)
651 if (!((id.nid == LNET_NID_ANY ||
652 peer_ni->ksnp_id.nid == id.nid) &&
653 (id.pid == LNET_PID_ANY ||
654 peer_ni->ksnp_id.pid == id.pid)))
657 ksocknal_peer_addref(peer_ni); /* a ref for me... */
659 ksocknal_del_peer_locked(peer_ni, ip);
661 if (peer_ni->ksnp_closing &&
662 !list_empty(&peer_ni->ksnp_tx_queue)) {
663 LASSERT(list_empty(&peer_ni->ksnp_conns));
664 LASSERT(peer_ni->ksnp_conn_cb == NULL);
666 list_splice_init(&peer_ni->ksnp_tx_queue,
670 ksocknal_peer_decref(peer_ni); /* ...till here */
672 rc = 0; /* matched! */
676 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
678 ksocknal_txlist_done(ni, &zombies, -ENETDOWN);
683 static struct ksock_conn *
684 ksocknal_get_conn_by_idx(struct lnet_ni *ni, int index)
686 struct ksock_peer_ni *peer_ni;
687 struct ksock_conn *conn;
690 read_lock(&ksocknal_data.ksnd_global_lock);
692 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
693 LASSERT(!peer_ni->ksnp_closing);
695 if (peer_ni->ksnp_ni != ni)
698 list_for_each_entry(conn, &peer_ni->ksnp_conns,
703 ksocknal_conn_addref(conn);
704 read_unlock(&ksocknal_data.ksnd_global_lock);
709 read_unlock(&ksocknal_data.ksnd_global_lock);
713 static struct ksock_sched *
714 ksocknal_choose_scheduler_locked(unsigned int cpt)
716 struct ksock_sched *sched = ksocknal_data.ksnd_schedulers[cpt];
719 if (sched->kss_nthreads == 0) {
720 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
721 if (sched->kss_nthreads > 0) {
722 CDEBUG(D_NET, "scheduler[%d] has no threads. selected scheduler[%d]\n",
723 cpt, sched->kss_cpt);
734 ksocknal_accept(struct lnet_ni *ni, struct socket *sock)
736 struct ksock_connreq *cr;
738 struct sockaddr_storage peer;
740 rc = lnet_sock_getaddr(sock, true, &peer);
742 CERROR("Can't determine new connection's address\n");
746 LIBCFS_ALLOC(cr, sizeof(*cr));
748 LCONSOLE_ERROR_MSG(0x12f,
749 "Dropping connection request from %pIS: memory exhausted\n",
756 cr->ksncr_sock = sock;
758 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
760 list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
761 wake_up(&ksocknal_data.ksnd_connd_waitq);
763 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
768 ksocknal_connecting(struct ksock_conn_cb *conn_cb, struct sockaddr *sa)
771 rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr, sa))
772 return conn_cb->ksnr_connecting;
777 ksocknal_create_conn(struct lnet_ni *ni, struct ksock_conn_cb *conn_cb,
778 struct socket *sock, int type)
780 rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
782 struct lnet_process_id peerid;
784 struct ksock_conn *conn;
785 struct ksock_conn *conn2;
786 struct ksock_peer_ni *peer_ni = NULL;
787 struct ksock_peer_ni *peer2;
788 struct ksock_sched *sched;
789 struct ksock_hello_msg *hello;
792 struct ksock_tx *txtmp;
799 active = (conn_cb != NULL);
801 LASSERT(active == (type != SOCKLND_CONN_NONE));
803 LIBCFS_ALLOC(conn, sizeof(*conn));
809 conn->ksnc_peer = NULL;
810 conn->ksnc_conn_cb = NULL;
811 conn->ksnc_sock = sock;
812 /* 2 ref, 1 for conn, another extra ref prevents socket
813 * being closed before establishment of connection */
814 refcount_set(&conn->ksnc_sock_refcount, 2);
815 conn->ksnc_type = type;
816 ksocknal_lib_save_callback(sock, conn);
817 refcount_set(&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
819 conn->ksnc_rx_ready = 0;
820 conn->ksnc_rx_scheduled = 0;
822 INIT_LIST_HEAD(&conn->ksnc_tx_queue);
823 conn->ksnc_tx_ready = 0;
824 conn->ksnc_tx_scheduled = 0;
825 conn->ksnc_tx_carrier = NULL;
826 atomic_set (&conn->ksnc_tx_nob, 0);
828 LIBCFS_ALLOC(hello, offsetof(struct ksock_hello_msg,
829 kshm_ips[LNET_INTERFACES_NUM]));
835 /* stash conn's local and remote addrs */
836 rc = ksocknal_lib_get_conn_addrs(conn);
840 /* Find out/confirm peer_ni's NID and connection type and get the
841 * vector of interfaces she's willing to let me connect to.
842 * Passive connections use the listener timeout since the peer_ni sends
847 peer_ni = conn_cb->ksnr_peer;
848 LASSERT(ni == peer_ni->ksnp_ni);
850 /* Active connection sends HELLO eagerly */
851 hello->kshm_nips = 0;
852 peerid = peer_ni->ksnp_id;
854 write_lock_bh(global_lock);
855 conn->ksnc_proto = peer_ni->ksnp_proto;
856 write_unlock_bh(global_lock);
858 if (conn->ksnc_proto == NULL) {
859 conn->ksnc_proto = &ksocknal_protocol_v3x;
860 #if SOCKNAL_VERSION_DEBUG
861 if (*ksocknal_tunables.ksnd_protocol == 2)
862 conn->ksnc_proto = &ksocknal_protocol_v2x;
863 else if (*ksocknal_tunables.ksnd_protocol == 1)
864 conn->ksnc_proto = &ksocknal_protocol_v1x;
868 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
872 peerid.nid = LNET_NID_ANY;
873 peerid.pid = LNET_PID_ANY;
875 /* Passive, get protocol from peer_ni */
876 conn->ksnc_proto = NULL;
879 rc = ksocknal_recv_hello(ni, conn, hello, &peerid, &incarnation);
883 LASSERT(rc == 0 || active);
884 LASSERT(conn->ksnc_proto != NULL);
885 LASSERT(peerid.nid != LNET_NID_ANY);
887 cpt = lnet_cpt_of_nid(peerid.nid, ni);
890 ksocknal_peer_addref(peer_ni);
891 write_lock_bh(global_lock);
893 peer_ni = ksocknal_create_peer(ni, peerid);
894 if (IS_ERR(peer_ni)) {
895 rc = PTR_ERR(peer_ni);
899 write_lock_bh(global_lock);
901 /* called with a ref on ni, so shutdown can't have started */
902 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers) >= 0);
904 peer2 = ksocknal_find_peer_locked(ni, peerid);
906 /* NB this puts an "empty" peer_ni in the peer_ni
907 * table (which takes my ref) */
908 hash_add(ksocknal_data.ksnd_peers,
909 &peer_ni->ksnp_list, peerid.nid);
911 ksocknal_peer_decref(peer_ni);
916 ksocknal_peer_addref(peer_ni);
917 peer_ni->ksnp_accepting++;
919 /* Am I already connecting to this guy? Resolve in
920 * favour of higher NID...
922 if (peerid.nid < ni->ni_nid &&
923 ksocknal_connecting(peer_ni->ksnp_conn_cb,
924 ((struct sockaddr *) &conn->ksnc_peeraddr))) {
926 warn = "connection race resolution";
931 if (peer_ni->ksnp_closing ||
932 (active && conn_cb->ksnr_deleted)) {
933 /* peer_ni/conn_cb got closed under me */
935 warn = "peer_ni/conn_cb removed";
939 if (peer_ni->ksnp_proto == NULL) {
940 /* Never connected before.
941 * NB recv_hello may have returned EPROTO to signal my peer_ni
942 * wants a different protocol than the one I asked for.
944 LASSERT(list_empty(&peer_ni->ksnp_conns));
946 peer_ni->ksnp_proto = conn->ksnc_proto;
947 peer_ni->ksnp_incarnation = incarnation;
950 if (peer_ni->ksnp_proto != conn->ksnc_proto ||
951 peer_ni->ksnp_incarnation != incarnation) {
952 /* peer_ni rebooted or I've got the wrong protocol version */
953 ksocknal_close_peer_conns_locked(peer_ni, NULL, 0);
955 peer_ni->ksnp_proto = NULL;
957 warn = peer_ni->ksnp_incarnation != incarnation ?
959 "wrong proto version";
969 warn = "lost conn race";
972 warn = "retry with different protocol version";
976 /* Refuse to duplicate an existing connection, unless this is a
977 * loopback connection */
978 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
979 (struct sockaddr *)&conn->ksnc_myaddr)) {
980 list_for_each_entry(conn2, &peer_ni->ksnp_conns, ksnc_list) {
982 (struct sockaddr *)&conn2->ksnc_peeraddr,
983 (struct sockaddr *)&conn->ksnc_peeraddr) ||
985 (struct sockaddr *)&conn2->ksnc_myaddr,
986 (struct sockaddr *)&conn->ksnc_myaddr) ||
987 conn2->ksnc_type != conn->ksnc_type)
991 if (num_dup < *ksocknal_tunables.ksnd_conns_per_peer)
994 /* Reply on a passive connection attempt so the peer_ni
995 * realises we're connected.
1005 /* If the connection created by this route didn't bind to the IP
1006 * address the route connected to, the connection/route matching
1007 * code below probably isn't going to work.
1010 !rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr,
1011 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1012 CERROR("Route %s %pIS connected to %pIS\n",
1013 libcfs_id2str(peer_ni->ksnp_id),
1014 &conn_cb->ksnr_addr,
1015 &conn->ksnc_peeraddr);
1018 /* Search for a conn_cb corresponding to the new connection and
1019 * create an association. This allows incoming connections created
1020 * by conn_cbs in my peer_ni to match my own conn_cb entries so I don't
1021 * continually create duplicate conn_cbs.
1023 conn_cb = peer_ni->ksnp_conn_cb;
1025 if (conn_cb && rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
1026 (struct sockaddr *)&conn_cb->ksnr_addr))
1027 ksocknal_associate_cb_conn_locked(conn_cb, conn);
1029 conn->ksnc_peer = peer_ni; /* conn takes my ref on peer_ni */
1030 peer_ni->ksnp_last_alive = ktime_get_seconds();
1031 peer_ni->ksnp_send_keepalive = 0;
1032 peer_ni->ksnp_error = 0;
1034 sched = ksocknal_choose_scheduler_locked(cpt);
1036 CERROR("no schedulers available. node is unhealthy\n");
1040 * The cpt might have changed if we ended up selecting a non cpt
1041 * native scheduler. So use the scheduler's cpt instead.
1043 cpt = sched->kss_cpt;
1044 sched->kss_nconns++;
1045 conn->ksnc_scheduler = sched;
1047 conn->ksnc_tx_last_post = ktime_get_seconds();
1048 /* Set the deadline for the outgoing HELLO to drain */
1049 conn->ksnc_tx_bufnob = sock->sk->sk_wmem_queued;
1050 conn->ksnc_tx_deadline = ktime_get_seconds() +
1052 smp_mb(); /* order with adding to peer_ni's conn list */
1054 list_add(&conn->ksnc_list, &peer_ni->ksnp_conns);
1055 ksocknal_conn_addref(conn);
1057 ksocknal_new_packet(conn, 0);
1059 conn->ksnc_zc_capable = ksocknal_lib_zc_capable(conn);
1061 /* Take packets blocking for this connection. */
1062 list_for_each_entry_safe(tx, txtmp, &peer_ni->ksnp_tx_queue, tx_list) {
1063 if (conn->ksnc_proto->pro_match_tx(conn, tx, tx->tx_nonblk) ==
1067 list_del(&tx->tx_list);
1068 ksocknal_queue_tx_locked(tx, conn);
1071 write_unlock_bh(global_lock);
1073 /* We've now got a new connection. Any errors from here on are just
1074 * like "normal" comms errors and we close the connection normally.
1075 * NB (a) we still have to send the reply HELLO for passive
1077 * (b) normal I/O on the conn is blocked until I setup and call the
1081 CDEBUG(D_NET, "New conn %s p %d.x %pIS -> %pISp"
1082 " incarnation:%lld sched[%d]\n",
1083 libcfs_id2str(peerid), conn->ksnc_proto->pro_version,
1084 &conn->ksnc_myaddr, &conn->ksnc_peeraddr,
1088 hello->kshm_nips = 0;
1089 rc = ksocknal_send_hello(ni, conn, peerid.nid, hello);
1092 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1093 kshm_ips[LNET_INTERFACES_NUM]));
1095 /* setup the socket AFTER I've received hello (it disables
1096 * SO_LINGER). I might call back to the acceptor who may want
1097 * to send a protocol version response and then close the
1098 * socket; this ensures the socket only tears down after the
1099 * response has been sent.
1102 rc = ksocknal_lib_setup_sock(sock);
1104 write_lock_bh(global_lock);
1106 /* NB my callbacks block while I hold ksnd_global_lock */
1107 ksocknal_lib_set_callback(sock, conn);
1110 peer_ni->ksnp_accepting--;
1112 write_unlock_bh(global_lock);
1115 write_lock_bh(global_lock);
1116 if (!conn->ksnc_closing) {
1117 /* could be closed by another thread */
1118 ksocknal_close_conn_locked(conn, rc);
1120 write_unlock_bh(global_lock);
1121 } else if (ksocknal_connsock_addref(conn) == 0) {
1122 /* Allow I/O to proceed. */
1123 ksocknal_read_callback(conn);
1124 ksocknal_write_callback(conn);
1125 ksocknal_connsock_decref(conn);
1128 ksocknal_connsock_decref(conn);
1129 ksocknal_conn_decref(conn);
1134 if (!peer_ni->ksnp_closing &&
1135 list_empty(&peer_ni->ksnp_conns) &&
1136 peer_ni->ksnp_conn_cb == NULL) {
1137 list_splice_init(&peer_ni->ksnp_tx_queue, &zombies);
1138 ksocknal_unlink_peer_locked(peer_ni);
1141 write_unlock_bh(global_lock);
1145 CERROR("Not creating conn %s type %d: %s\n",
1146 libcfs_id2str(peerid), conn->ksnc_type, warn);
1148 CDEBUG(D_NET, "Not creating conn %s type %d: %s\n",
1149 libcfs_id2str(peerid), conn->ksnc_type, warn);
1154 /* Request retry by replying with CONN_NONE
1155 * ksnc_proto has been set already
1157 conn->ksnc_type = SOCKLND_CONN_NONE;
1158 hello->kshm_nips = 0;
1159 ksocknal_send_hello(ni, conn, peerid.nid, hello);
1162 write_lock_bh(global_lock);
1163 peer_ni->ksnp_accepting--;
1164 write_unlock_bh(global_lock);
1168 * If we get here without an error code, just use -EALREADY.
1169 * Depending on how we got here, the error may be positive
1170 * or negative. Normalize the value for ksocknal_txlist_done().
1172 rc2 = (rc == 0 ? -EALREADY : (rc > 0 ? -rc : rc));
1173 ksocknal_txlist_done(ni, &zombies, rc2);
1174 ksocknal_peer_decref(peer_ni);
1178 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1179 kshm_ips[LNET_INTERFACES_NUM]));
1181 LIBCFS_FREE(conn, sizeof(*conn));
1190 ksocknal_close_conn_locked(struct ksock_conn *conn, int error)
1192 /* This just does the immmediate housekeeping, and queues the
1193 * connection for the reaper to terminate.
1194 * Caller holds ksnd_global_lock exclusively in irq context */
1195 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1196 struct ksock_conn_cb *conn_cb;
1197 struct ksock_conn *conn2;
1199 LASSERT(peer_ni->ksnp_error == 0);
1200 LASSERT(!conn->ksnc_closing);
1201 conn->ksnc_closing = 1;
1203 /* ksnd_deathrow_conns takes over peer_ni's ref */
1204 list_del(&conn->ksnc_list);
1206 conn_cb = conn->ksnc_conn_cb;
1207 if (conn_cb != NULL) {
1208 /* dissociate conn from cb... */
1209 LASSERT(!conn_cb->ksnr_deleted);
1211 /* connected bit is set only if all connections
1212 * of the given type got created
1214 if (ksocknal_get_conn_count_by_type(conn_cb, conn->ksnc_type) ==
1215 *ksocknal_tunables.ksnd_conns_per_peer)
1216 LASSERT((conn_cb->ksnr_connected &
1217 BIT(conn->ksnc_type)) != 0);
1219 list_for_each_entry(conn2, &peer_ni->ksnp_conns, ksnc_list) {
1220 if (conn2->ksnc_conn_cb == conn_cb &&
1221 conn2->ksnc_type == conn->ksnc_type)
1224 conn_cb->ksnr_connected &= ~BIT(conn->ksnc_type);
1226 conn->ksnc_conn_cb = NULL;
1228 /* drop conn's ref on conn_cb */
1229 ksocknal_conn_cb_decref(conn_cb);
1232 if (list_empty(&peer_ni->ksnp_conns)) {
1233 /* No more connections to this peer_ni */
1235 if (!list_empty(&peer_ni->ksnp_tx_queue)) {
1236 struct ksock_tx *tx;
1238 LASSERT(conn->ksnc_proto == &ksocknal_protocol_v3x);
1240 /* throw them to the last connection...,
1241 * these TXs will be send to /dev/null by scheduler */
1242 list_for_each_entry(tx, &peer_ni->ksnp_tx_queue,
1244 ksocknal_tx_prep(conn, tx);
1246 spin_lock_bh(&conn->ksnc_scheduler->kss_lock);
1247 list_splice_init(&peer_ni->ksnp_tx_queue,
1248 &conn->ksnc_tx_queue);
1249 spin_unlock_bh(&conn->ksnc_scheduler->kss_lock);
1252 /* renegotiate protocol version */
1253 peer_ni->ksnp_proto = NULL;
1254 /* stash last conn close reason */
1255 peer_ni->ksnp_error = error;
1257 if (peer_ni->ksnp_conn_cb == NULL) {
1258 /* I've just closed last conn belonging to a
1259 * peer_ni with no connections to it
1261 ksocknal_unlink_peer_locked(peer_ni);
1265 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1267 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_deathrow_conns);
1268 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1270 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1274 ksocknal_peer_failed(struct ksock_peer_ni *peer_ni)
1276 bool notify = false;
1277 time64_t last_alive = 0;
1279 /* There has been a connection failure or comms error; but I'll only
1280 * tell LNET I think the peer_ni is dead if it's to another kernel and
1281 * there are no connections or connection attempts in existence. */
1283 read_lock(&ksocknal_data.ksnd_global_lock);
1285 if ((peer_ni->ksnp_id.pid & LNET_PID_USERFLAG) == 0 &&
1286 list_empty(&peer_ni->ksnp_conns) &&
1287 peer_ni->ksnp_accepting == 0 &&
1288 !ksocknal_find_connecting_conn_cb_locked(peer_ni)) {
1290 last_alive = peer_ni->ksnp_last_alive;
1293 read_unlock(&ksocknal_data.ksnd_global_lock);
1296 lnet_notify(peer_ni->ksnp_ni, peer_ni->ksnp_id.nid,
1297 false, false, last_alive);
1301 ksocknal_finalize_zcreq(struct ksock_conn *conn)
1303 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1304 struct ksock_tx *tx;
1305 struct ksock_tx *tmp;
1308 /* NB safe to finalize TXs because closing of socket will
1309 * abort all buffered data */
1310 LASSERT(conn->ksnc_sock == NULL);
1312 spin_lock(&peer_ni->ksnp_lock);
1314 list_for_each_entry_safe(tx, tmp, &peer_ni->ksnp_zc_req_list,
1316 if (tx->tx_conn != conn)
1319 LASSERT(tx->tx_msg.ksm_zc_cookies[0] != 0);
1321 tx->tx_msg.ksm_zc_cookies[0] = 0;
1322 tx->tx_zc_aborted = 1; /* mark it as not-acked */
1323 list_move(&tx->tx_zc_list, &zlist);
1326 spin_unlock(&peer_ni->ksnp_lock);
1328 while ((tx = list_first_entry_or_null(&zlist, struct ksock_tx,
1329 tx_zc_list)) != NULL) {
1330 list_del(&tx->tx_zc_list);
1331 ksocknal_tx_decref(tx);
1336 ksocknal_terminate_conn(struct ksock_conn *conn)
1338 /* This gets called by the reaper (guaranteed thread context) to
1339 * disengage the socket from its callbacks and close it.
1340 * ksnc_refcount will eventually hit zero, and then the reaper will
1343 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1344 struct ksock_sched *sched = conn->ksnc_scheduler;
1345 bool failed = false;
1347 LASSERT(conn->ksnc_closing);
1349 /* wake up the scheduler to "send" all remaining packets to /dev/null */
1350 spin_lock_bh(&sched->kss_lock);
1352 /* a closing conn is always ready to tx */
1353 conn->ksnc_tx_ready = 1;
1355 if (!conn->ksnc_tx_scheduled &&
1356 !list_empty(&conn->ksnc_tx_queue)) {
1357 list_add_tail(&conn->ksnc_tx_list,
1358 &sched->kss_tx_conns);
1359 conn->ksnc_tx_scheduled = 1;
1360 /* extra ref for scheduler */
1361 ksocknal_conn_addref(conn);
1363 wake_up(&sched->kss_waitq);
1366 spin_unlock_bh(&sched->kss_lock);
1368 /* serialise with callbacks */
1369 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1371 ksocknal_lib_reset_callback(conn->ksnc_sock, conn);
1373 /* OK, so this conn may not be completely disengaged from its
1374 * scheduler yet, but it _has_ committed to terminate...
1376 conn->ksnc_scheduler->kss_nconns--;
1378 if (peer_ni->ksnp_error != 0) {
1379 /* peer_ni's last conn closed in error */
1380 LASSERT(list_empty(&peer_ni->ksnp_conns));
1382 peer_ni->ksnp_error = 0; /* avoid multiple notifications */
1385 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1388 ksocknal_peer_failed(peer_ni);
1390 /* The socket is closed on the final put; either here, or in
1391 * ksocknal_{send,recv}msg(). Since we set up the linger2 option
1392 * when the connection was established, this will close the socket
1393 * immediately, aborting anything buffered in it. Any hung
1394 * zero-copy transmits will therefore complete in finite time.
1396 ksocknal_connsock_decref(conn);
1400 ksocknal_queue_zombie_conn(struct ksock_conn *conn)
1402 /* Queue the conn for the reaper to destroy */
1403 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1404 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1406 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
1407 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1409 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1413 ksocknal_destroy_conn(struct ksock_conn *conn)
1417 /* Final coup-de-grace of the reaper */
1418 CDEBUG(D_NET, "connection %p\n", conn);
1420 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1421 LASSERT(refcount_read(&conn->ksnc_sock_refcount) == 0);
1422 LASSERT(conn->ksnc_sock == NULL);
1423 LASSERT(conn->ksnc_conn_cb == NULL);
1424 LASSERT(!conn->ksnc_tx_scheduled);
1425 LASSERT(!conn->ksnc_rx_scheduled);
1426 LASSERT(list_empty(&conn->ksnc_tx_queue));
1428 /* complete current receive if any */
1429 switch (conn->ksnc_rx_state) {
1430 case SOCKNAL_RX_LNET_PAYLOAD:
1431 last_rcv = conn->ksnc_rx_deadline -
1433 CERROR("Completing partial receive from %s[%d], ip %pISp, with error, wanted: %d, left: %d, last alive is %lld secs ago\n",
1434 libcfs_id2str(conn->ksnc_peer->ksnp_id), conn->ksnc_type,
1435 &conn->ksnc_peeraddr,
1436 conn->ksnc_rx_nob_wanted, conn->ksnc_rx_nob_left,
1437 ktime_get_seconds() - last_rcv);
1438 if (conn->ksnc_lnet_msg)
1439 conn->ksnc_lnet_msg->msg_health_status =
1440 LNET_MSG_STATUS_REMOTE_ERROR;
1441 lnet_finalize(conn->ksnc_lnet_msg, -EIO);
1443 case SOCKNAL_RX_LNET_HEADER:
1444 if (conn->ksnc_rx_started)
1445 CERROR("Incomplete receive of lnet header from %s, ip %pISp, with error, protocol: %d.x.\n",
1446 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1447 &conn->ksnc_peeraddr,
1448 conn->ksnc_proto->pro_version);
1450 case SOCKNAL_RX_KSM_HEADER:
1451 if (conn->ksnc_rx_started)
1452 CERROR("Incomplete receive of ksock message from %s, ip %pISp, with error, protocol: %d.x.\n",
1453 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1454 &conn->ksnc_peeraddr,
1455 conn->ksnc_proto->pro_version);
1457 case SOCKNAL_RX_SLOP:
1458 if (conn->ksnc_rx_started)
1459 CERROR("Incomplete receive of slops from %s, ip %pISp, with error\n",
1460 libcfs_id2str(conn->ksnc_peer->ksnp_id),
1461 &conn->ksnc_peeraddr);
1468 ksocknal_peer_decref(conn->ksnc_peer);
1470 LIBCFS_FREE (conn, sizeof (*conn));
1474 ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
1475 struct sockaddr *addr, int why)
1477 struct ksock_conn *conn;
1478 struct ksock_conn *cnxt;
1481 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
1484 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1486 ksocknal_close_conn_locked(conn, why);
1494 ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why)
1496 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1499 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1501 count = ksocknal_close_peer_conns_locked(
1502 peer_ni, (struct sockaddr *)&conn->ksnc_peeraddr, why);
1504 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1510 ksocknal_close_matching_conns(struct lnet_process_id id, __u32 ipaddr)
1512 struct ksock_peer_ni *peer_ni;
1513 struct hlist_node *pnxt;
1518 struct sockaddr_in sa = {.sin_family = AF_INET};
1520 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1522 if (id.nid != LNET_NID_ANY) {
1523 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
1527 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1530 sa.sin_addr.s_addr = htonl(ipaddr);
1531 for (i = lo; i <= hi; i++) {
1532 hlist_for_each_entry_safe(peer_ni, pnxt,
1533 &ksocknal_data.ksnd_peers[i],
1536 if (!((id.nid == LNET_NID_ANY ||
1537 id.nid == peer_ni->ksnp_id.nid) &&
1538 (id.pid == LNET_PID_ANY ||
1539 id.pid == peer_ni->ksnp_id.pid)))
1542 count += ksocknal_close_peer_conns_locked(
1544 ipaddr ? (struct sockaddr *)&sa : NULL, 0);
1548 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1550 /* wildcards always succeed */
1551 if (id.nid == LNET_NID_ANY || id.pid == LNET_PID_ANY || ipaddr == 0)
1554 return (count == 0 ? -ENOENT : 0);
1558 ksocknal_notify_gw_down(lnet_nid_t gw_nid)
1560 /* The router is telling me she's been notified of a change in
1563 struct lnet_process_id id = {
1565 .pid = LNET_PID_ANY,
1568 CDEBUG(D_NET, "gw %s down\n", libcfs_nid2str(gw_nid));
1570 /* If the gateway crashed, close all open connections... */
1571 ksocknal_close_matching_conns(id, 0);
1574 /* We can only establish new connections
1575 * if we have autroutes, and these connect on demand. */
1579 ksocknal_push_peer(struct ksock_peer_ni *peer_ni)
1583 struct ksock_conn *conn;
1585 for (index = 0; ; index++) {
1586 read_lock(&ksocknal_data.ksnd_global_lock);
1591 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
1593 ksocknal_conn_addref(conn);
1598 read_unlock(&ksocknal_data.ksnd_global_lock);
1603 ksocknal_lib_push_conn (conn);
1604 ksocknal_conn_decref(conn);
1609 ksocknal_push(struct lnet_ni *ni, struct lnet_process_id id)
1616 if (id.nid != LNET_NID_ANY) {
1617 lo = hash_min(id.nid, HASH_BITS(ksocknal_data.ksnd_peers));
1621 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1624 for (bkt = lo; bkt <= hi; bkt++) {
1625 int peer_off; /* searching offset in peer_ni hash table */
1627 for (peer_off = 0; ; peer_off++) {
1628 struct ksock_peer_ni *peer_ni;
1631 read_lock(&ksocknal_data.ksnd_global_lock);
1632 hlist_for_each_entry(peer_ni,
1633 &ksocknal_data.ksnd_peers[bkt],
1635 if (!((id.nid == LNET_NID_ANY ||
1636 id.nid == peer_ni->ksnp_id.nid) &&
1637 (id.pid == LNET_PID_ANY ||
1638 id.pid == peer_ni->ksnp_id.pid)))
1641 if (i++ == peer_off) {
1642 ksocknal_peer_addref(peer_ni);
1646 read_unlock(&ksocknal_data.ksnd_global_lock);
1648 if (i <= peer_off) /* no match */
1652 ksocknal_push_peer(peer_ni);
1653 ksocknal_peer_decref(peer_ni);
1660 ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg)
1662 struct lnet_process_id id = {0};
1663 struct libcfs_ioctl_data *data = arg;
1667 case IOC_LIBCFS_GET_INTERFACE: {
1668 struct ksock_net *net = ni->ni_data;
1669 struct ksock_interface *iface;
1670 struct sockaddr_in *sa;
1672 read_lock(&ksocknal_data.ksnd_global_lock);
1674 if (data->ioc_count >= 1) {
1678 iface = &net->ksnn_interface;
1680 sa = (void *)&iface->ksni_addr;
1681 if (sa->sin_family == AF_INET)
1682 data->ioc_u32[0] = ntohl(sa->sin_addr.s_addr);
1684 data->ioc_u32[0] = 0xFFFFFFFF;
1685 data->ioc_u32[1] = iface->ksni_netmask;
1686 data->ioc_u32[2] = iface->ksni_npeers;
1687 data->ioc_u32[3] = iface->ksni_nroutes;
1690 read_unlock(&ksocknal_data.ksnd_global_lock);
1694 case IOC_LIBCFS_GET_PEER: {
1699 int share_count = 0;
1701 rc = ksocknal_get_peer_info(ni, data->ioc_count,
1702 &id, &myip, &ip, &port,
1703 &conn_count, &share_count);
1707 data->ioc_nid = id.nid;
1708 data->ioc_count = share_count;
1709 data->ioc_u32[0] = ip;
1710 data->ioc_u32[1] = port;
1711 data->ioc_u32[2] = myip;
1712 data->ioc_u32[3] = conn_count;
1713 data->ioc_u32[4] = id.pid;
1717 case IOC_LIBCFS_ADD_PEER: {
1718 struct sockaddr_in sa = {.sin_family = AF_INET};
1720 id.nid = data->ioc_nid;
1721 id.pid = LNET_PID_LUSTRE;
1722 sa.sin_addr.s_addr = htonl(data->ioc_u32[0]);
1723 sa.sin_port = htons(data->ioc_u32[1]);
1724 return ksocknal_add_peer(ni, id, (struct sockaddr *)&sa);
1726 case IOC_LIBCFS_DEL_PEER:
1727 id.nid = data->ioc_nid;
1728 id.pid = LNET_PID_ANY;
1729 return ksocknal_del_peer (ni, id,
1730 data->ioc_u32[0]); /* IP */
1732 case IOC_LIBCFS_GET_CONN: {
1736 struct ksock_conn *conn = ksocknal_get_conn_by_idx(ni, data->ioc_count);
1737 struct sockaddr_in *psa = (void *)&conn->ksnc_peeraddr;
1738 struct sockaddr_in *mysa = (void *)&conn->ksnc_myaddr;
1743 ksocknal_lib_get_conn_tunables(conn, &txmem, &rxmem, &nagle);
1745 data->ioc_count = txmem;
1746 data->ioc_nid = conn->ksnc_peer->ksnp_id.nid;
1747 data->ioc_flags = nagle;
1748 if (psa->sin_family == AF_INET)
1749 data->ioc_u32[0] = ntohl(psa->sin_addr.s_addr);
1751 data->ioc_u32[0] = 0xFFFFFFFF;
1752 data->ioc_u32[1] = rpc_get_port((struct sockaddr *)
1753 &conn->ksnc_peeraddr);
1754 if (mysa->sin_family == AF_INET)
1755 data->ioc_u32[2] = ntohl(mysa->sin_addr.s_addr);
1757 data->ioc_u32[2] = 0xFFFFFFFF;
1758 data->ioc_u32[3] = conn->ksnc_type;
1759 data->ioc_u32[4] = conn->ksnc_scheduler->kss_cpt;
1760 data->ioc_u32[5] = rxmem;
1761 data->ioc_u32[6] = conn->ksnc_peer->ksnp_id.pid;
1762 ksocknal_conn_decref(conn);
1766 case IOC_LIBCFS_CLOSE_CONNECTION:
1767 id.nid = data->ioc_nid;
1768 id.pid = LNET_PID_ANY;
1769 return ksocknal_close_matching_conns (id,
1772 case IOC_LIBCFS_REGISTER_MYNID:
1773 /* Ignore if this is a noop */
1774 if (data->ioc_nid == ni->ni_nid)
1777 CERROR("obsolete IOC_LIBCFS_REGISTER_MYNID: %s(%s)\n",
1778 libcfs_nid2str(data->ioc_nid),
1779 libcfs_nid2str(ni->ni_nid));
1782 case IOC_LIBCFS_PUSH_CONNECTION:
1783 id.nid = data->ioc_nid;
1784 id.pid = LNET_PID_ANY;
1785 return ksocknal_push(ni, id);
1794 ksocknal_free_buffers (void)
1796 LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
1798 if (ksocknal_data.ksnd_schedulers != NULL)
1799 cfs_percpt_free(ksocknal_data.ksnd_schedulers);
1801 spin_lock(&ksocknal_data.ksnd_tx_lock);
1803 if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
1805 struct ksock_tx *tx;
1807 list_splice_init(&ksocknal_data.ksnd_idle_noop_txs, &zlist);
1808 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1810 while ((tx = list_first_entry_or_null(&zlist, struct ksock_tx,
1811 tx_list)) != NULL) {
1812 list_del(&tx->tx_list);
1813 LIBCFS_FREE(tx, tx->tx_desc_size);
1816 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1820 static int ksocknal_get_link_status(struct net_device *dev)
1826 if (!netif_running(dev))
1828 /* Some devices may not be providing link settings */
1829 else if (dev->ethtool_ops->get_link)
1830 ret = dev->ethtool_ops->get_link(dev);
1836 ksocknal_handle_link_state_change(struct net_device *dev,
1837 unsigned char operstate)
1840 struct ksock_net *net;
1841 struct ksock_net *cnxt;
1843 unsigned char link_down = !(operstate == IF_OPER_UP);
1845 ifindex = dev->ifindex;
1847 if (!ksocknal_data.ksnd_nnets)
1850 list_for_each_entry_safe(net, cnxt, &ksocknal_data.ksnd_nets,
1852 if (net->ksnn_interface.ksni_index != ifindex)
1856 atomic_set(&ni->ni_fatal_error_on, link_down);
1858 atomic_set(&ni->ni_fatal_error_on,
1859 (ksocknal_get_link_status(dev) == 0));
1866 /************************************
1867 * Net device notifier event handler
1868 ************************************/
1869 static int ksocknal_device_event(struct notifier_block *unused,
1870 unsigned long event, void *ptr)
1872 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1873 unsigned char operstate;
1875 operstate = dev->operstate;
1881 ksocknal_handle_link_state_change(dev, operstate);
1888 static struct notifier_block ksocknal_notifier_block = {
1889 .notifier_call = ksocknal_device_event,
1893 ksocknal_base_shutdown(void)
1895 struct ksock_sched *sched;
1896 struct ksock_peer_ni *peer_ni;
1899 CDEBUG(D_MALLOC, "before NAL cleanup: kmem %lld\n",
1900 libcfs_kmem_read());
1901 LASSERT (ksocknal_data.ksnd_nnets == 0);
1903 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL)
1904 unregister_netdevice_notifier(&ksocknal_notifier_block);
1906 switch (ksocknal_data.ksnd_init) {
1911 case SOCKNAL_INIT_ALL:
1912 case SOCKNAL_INIT_DATA:
1913 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list)
1916 LASSERT(list_empty(&ksocknal_data.ksnd_nets));
1917 LASSERT(list_empty(&ksocknal_data.ksnd_enomem_conns));
1918 LASSERT(list_empty(&ksocknal_data.ksnd_zombie_conns));
1919 LASSERT(list_empty(&ksocknal_data.ksnd_connd_connreqs));
1920 LASSERT(list_empty(&ksocknal_data.ksnd_connd_routes));
1922 if (ksocknal_data.ksnd_schedulers != NULL) {
1923 cfs_percpt_for_each(sched, i,
1924 ksocknal_data.ksnd_schedulers) {
1926 LASSERT(list_empty(&sched->kss_tx_conns));
1927 LASSERT(list_empty(&sched->kss_rx_conns));
1928 LASSERT(list_empty(&sched->kss_zombie_noop_txs));
1929 LASSERT(sched->kss_nconns == 0);
1933 /* flag threads to terminate; wake and wait for them to die */
1934 ksocknal_data.ksnd_shuttingdown = 1;
1935 wake_up_all(&ksocknal_data.ksnd_connd_waitq);
1936 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1938 if (ksocknal_data.ksnd_schedulers != NULL) {
1939 cfs_percpt_for_each(sched, i,
1940 ksocknal_data.ksnd_schedulers)
1941 wake_up_all(&sched->kss_waitq);
1944 wait_var_event_warning(&ksocknal_data.ksnd_nthreads,
1945 atomic_read(&ksocknal_data.ksnd_nthreads) == 0,
1946 "waiting for %d threads to terminate\n",
1947 atomic_read(&ksocknal_data.ksnd_nthreads));
1949 ksocknal_free_buffers();
1951 ksocknal_data.ksnd_init = SOCKNAL_INIT_NOTHING;
1955 CDEBUG(D_MALLOC, "after NAL cleanup: kmem %lld\n",
1956 libcfs_kmem_read());
1958 module_put(THIS_MODULE);
1962 ksocknal_base_startup(void)
1964 struct ksock_sched *sched;
1968 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
1969 LASSERT(ksocknal_data.ksnd_nnets == 0);
1971 memset(&ksocknal_data, 0, sizeof(ksocknal_data)); /* zero pointers */
1973 hash_init(ksocknal_data.ksnd_peers);
1975 rwlock_init(&ksocknal_data.ksnd_global_lock);
1976 INIT_LIST_HEAD(&ksocknal_data.ksnd_nets);
1978 spin_lock_init(&ksocknal_data.ksnd_reaper_lock);
1979 INIT_LIST_HEAD(&ksocknal_data.ksnd_enomem_conns);
1980 INIT_LIST_HEAD(&ksocknal_data.ksnd_zombie_conns);
1981 INIT_LIST_HEAD(&ksocknal_data.ksnd_deathrow_conns);
1982 init_waitqueue_head(&ksocknal_data.ksnd_reaper_waitq);
1984 spin_lock_init(&ksocknal_data.ksnd_connd_lock);
1985 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_connreqs);
1986 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_routes);
1987 init_waitqueue_head(&ksocknal_data.ksnd_connd_waitq);
1989 spin_lock_init(&ksocknal_data.ksnd_tx_lock);
1990 INIT_LIST_HEAD(&ksocknal_data.ksnd_idle_noop_txs);
1992 /* NB memset above zeros whole of ksocknal_data */
1994 /* flag lists/ptrs/locks initialised */
1995 ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
1996 if (!try_module_get(THIS_MODULE))
1999 /* Create a scheduler block per available CPT */
2000 ksocknal_data.ksnd_schedulers = cfs_percpt_alloc(lnet_cpt_table(),
2002 if (ksocknal_data.ksnd_schedulers == NULL)
2005 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
2009 * make sure not to allocate more threads than there are
2010 * cores/CPUs in teh CPT
2012 nthrs = cfs_cpt_weight(lnet_cpt_table(), i);
2013 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2014 nthrs = min(nthrs, *ksocknal_tunables.ksnd_nscheds);
2017 * max to half of CPUs, assume another half should be
2018 * reserved for upper layer modules
2020 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2023 sched->kss_nthreads_max = nthrs;
2026 spin_lock_init(&sched->kss_lock);
2027 INIT_LIST_HEAD(&sched->kss_rx_conns);
2028 INIT_LIST_HEAD(&sched->kss_tx_conns);
2029 INIT_LIST_HEAD(&sched->kss_zombie_noop_txs);
2030 init_waitqueue_head(&sched->kss_waitq);
2033 ksocknal_data.ksnd_connd_starting = 0;
2034 ksocknal_data.ksnd_connd_failed_stamp = 0;
2035 ksocknal_data.ksnd_connd_starting_stamp = ktime_get_real_seconds();
2036 /* must have at least 2 connds to remain responsive to accepts while
2038 if (*ksocknal_tunables.ksnd_nconnds < SOCKNAL_CONND_RESV + 1)
2039 *ksocknal_tunables.ksnd_nconnds = SOCKNAL_CONND_RESV + 1;
2041 if (*ksocknal_tunables.ksnd_nconnds_max <
2042 *ksocknal_tunables.ksnd_nconnds) {
2043 ksocknal_tunables.ksnd_nconnds_max =
2044 ksocknal_tunables.ksnd_nconnds;
2047 for (i = 0; i < *ksocknal_tunables.ksnd_nconnds; i++) {
2049 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2050 ksocknal_data.ksnd_connd_starting++;
2051 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2054 snprintf(name, sizeof(name), "socknal_cd%02d", i);
2055 rc = ksocknal_thread_start(ksocknal_connd,
2056 (void *)((uintptr_t)i), name);
2058 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2059 ksocknal_data.ksnd_connd_starting--;
2060 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2061 CERROR("Can't spawn socknal connd: %d\n", rc);
2066 rc = ksocknal_thread_start(ksocknal_reaper, NULL, "socknal_reaper");
2068 CERROR ("Can't spawn socknal reaper: %d\n", rc);
2072 register_netdevice_notifier(&ksocknal_notifier_block);
2074 /* flag everything initialised */
2075 ksocknal_data.ksnd_init = SOCKNAL_INIT_ALL;
2080 ksocknal_base_shutdown();
2085 ksocknal_debug_peerhash(struct lnet_ni *ni)
2087 struct ksock_peer_ni *peer_ni;
2090 read_lock(&ksocknal_data.ksnd_global_lock);
2092 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
2093 struct ksock_conn_cb *conn_cb;
2094 struct ksock_conn *conn;
2096 if (peer_ni->ksnp_ni != ni)
2099 CWARN("Active peer_ni on shutdown: %s, ref %d, "
2100 "closing %d, accepting %d, err %d, zcookie %llu, "
2101 "txq %d, zc_req %d\n", libcfs_id2str(peer_ni->ksnp_id),
2102 refcount_read(&peer_ni->ksnp_refcount),
2103 peer_ni->ksnp_closing,
2104 peer_ni->ksnp_accepting, peer_ni->ksnp_error,
2105 peer_ni->ksnp_zc_next_cookie,
2106 !list_empty(&peer_ni->ksnp_tx_queue),
2107 !list_empty(&peer_ni->ksnp_zc_req_list));
2109 conn_cb = peer_ni->ksnp_conn_cb;
2111 CWARN("ConnCB: ref %d, schd %d, conn %d, cnted %d, del %d\n",
2112 refcount_read(&conn_cb->ksnr_refcount),
2113 conn_cb->ksnr_scheduled, conn_cb->ksnr_connecting,
2114 conn_cb->ksnr_connected, conn_cb->ksnr_deleted);
2117 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
2118 CWARN("Conn: ref %d, sref %d, t %d, c %d\n",
2119 refcount_read(&conn->ksnc_conn_refcount),
2120 refcount_read(&conn->ksnc_sock_refcount),
2121 conn->ksnc_type, conn->ksnc_closing);
2126 read_unlock(&ksocknal_data.ksnd_global_lock);
2131 ksocknal_shutdown(struct lnet_ni *ni)
2133 struct ksock_net *net = ni->ni_data;
2134 struct lnet_process_id anyid = {
2135 .nid = LNET_NID_ANY,
2136 .pid = LNET_PID_ANY,
2139 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL);
2140 LASSERT(ksocknal_data.ksnd_nnets > 0);
2142 /* prevent new peers */
2143 atomic_add(SOCKNAL_SHUTDOWN_BIAS, &net->ksnn_npeers);
2145 /* Delete all peers */
2146 ksocknal_del_peer(ni, anyid, 0);
2148 /* Wait for all peer_ni state to clean up */
2149 wait_var_event_warning(&net->ksnn_npeers,
2150 atomic_read(&net->ksnn_npeers) ==
2151 SOCKNAL_SHUTDOWN_BIAS,
2152 "waiting for %d peers to disconnect\n",
2153 ksocknal_debug_peerhash(ni) +
2154 atomic_read(&net->ksnn_npeers) -
2155 SOCKNAL_SHUTDOWN_BIAS);
2157 LASSERT(net->ksnn_interface.ksni_npeers == 0);
2158 LASSERT(net->ksnn_interface.ksni_nroutes == 0);
2160 list_del(&net->ksnn_list);
2161 LIBCFS_FREE(net, sizeof(*net));
2163 ksocknal_data.ksnd_nnets--;
2164 if (ksocknal_data.ksnd_nnets == 0)
2165 ksocknal_base_shutdown();
2169 ksocknal_search_new_ipif(struct ksock_net *net)
2172 char *ifnam = &net->ksnn_interface.ksni_name[0];
2173 char *colon = strchr(ifnam, ':');
2175 struct ksock_net *tmp;
2180 list_for_each_entry(tmp, &ksocknal_data.ksnd_nets, ksnn_list) {
2181 char *ifnam2 = &tmp->ksnn_interface.ksni_name[0];
2182 char *colon2 = strchr(ifnam2, ':');
2187 found = strcmp(ifnam, ifnam2) == 0;
2200 ksocknal_start_schedulers(struct ksock_sched *sched)
2206 if (sched->kss_nthreads == 0) {
2207 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2208 nthrs = sched->kss_nthreads_max;
2210 nthrs = cfs_cpt_weight(lnet_cpt_table(),
2212 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2213 nthrs = min(SOCKNAL_NSCHEDS_HIGH, nthrs);
2215 nthrs = min(nthrs, sched->kss_nthreads_max);
2217 LASSERT(sched->kss_nthreads <= sched->kss_nthreads_max);
2218 /* increase two threads if there is new interface */
2219 nthrs = min(2, sched->kss_nthreads_max - sched->kss_nthreads);
2222 for (i = 0; i < nthrs; i++) {
2226 id = KSOCK_THREAD_ID(sched->kss_cpt, sched->kss_nthreads + i);
2227 snprintf(name, sizeof(name), "socknal_sd%02d_%02d",
2228 sched->kss_cpt, (int)KSOCK_THREAD_SID(id));
2230 rc = ksocknal_thread_start(ksocknal_scheduler,
2235 CERROR("Can't spawn thread %d for scheduler[%d]: %d\n",
2236 sched->kss_cpt, (int) KSOCK_THREAD_SID(id), rc);
2240 sched->kss_nthreads += i;
2245 ksocknal_net_start_threads(struct ksock_net *net, __u32 *cpts, int ncpts)
2247 int newif = ksocknal_search_new_ipif(net);
2251 if (ncpts > 0 && ncpts > cfs_cpt_number(lnet_cpt_table()))
2254 for (i = 0; i < ncpts; i++) {
2255 struct ksock_sched *sched;
2256 int cpt = (cpts == NULL) ? i : cpts[i];
2258 LASSERT(cpt < cfs_cpt_number(lnet_cpt_table()));
2259 sched = ksocknal_data.ksnd_schedulers[cpt];
2261 if (!newif && sched->kss_nthreads > 0)
2264 rc = ksocknal_start_schedulers(sched);
2272 ksocknal_startup(struct lnet_ni *ni)
2274 struct ksock_net *net;
2275 struct lnet_ioctl_config_lnd_cmn_tunables *net_tunables;
2276 struct ksock_interface *ksi = NULL;
2277 struct lnet_inetdev *ifaces = NULL;
2278 struct sockaddr_in *sa;
2282 LASSERT (ni->ni_net->net_lnd == &the_ksocklnd);
2283 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING) {
2284 rc = ksocknal_base_startup();
2288 LIBCFS_ALLOC(net, sizeof(*net));
2291 net->ksnn_incarnation = ktime_get_real_ns();
2293 net_tunables = &ni->ni_net->net_tunables;
2294 if (net_tunables->lct_peer_timeout == -1)
2295 net_tunables->lct_peer_timeout =
2296 *ksocknal_tunables.ksnd_peertimeout;
2298 if (net_tunables->lct_max_tx_credits == -1)
2299 net_tunables->lct_max_tx_credits =
2300 *ksocknal_tunables.ksnd_credits;
2302 if (net_tunables->lct_peer_tx_credits == -1)
2303 net_tunables->lct_peer_tx_credits =
2304 *ksocknal_tunables.ksnd_peertxcredits;
2306 if (net_tunables->lct_peer_tx_credits >
2307 net_tunables->lct_max_tx_credits)
2308 net_tunables->lct_peer_tx_credits =
2309 net_tunables->lct_max_tx_credits;
2311 if (net_tunables->lct_peer_rtr_credits == -1)
2312 net_tunables->lct_peer_rtr_credits =
2313 *ksocknal_tunables.ksnd_peerrtrcredits;
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();
2371 static void __exit ksocklnd_exit(void)
2373 lnet_unregister_lnd(&the_ksocklnd);
2376 static const struct lnet_lnd the_ksocklnd = {
2377 .lnd_type = SOCKLND,
2378 .lnd_startup = ksocknal_startup,
2379 .lnd_shutdown = ksocknal_shutdown,
2380 .lnd_ctl = ksocknal_ctl,
2381 .lnd_send = ksocknal_send,
2382 .lnd_recv = ksocknal_recv,
2383 .lnd_notify_peer_down = ksocknal_notify_gw_down,
2384 .lnd_accept = ksocknal_accept,
2387 static int __init ksocklnd_init(void)
2391 /* check ksnr_connected/connecting field large enough */
2392 BUILD_BUG_ON(SOCKLND_CONN_NTYPES > 4);
2393 BUILD_BUG_ON(SOCKLND_CONN_ACK != SOCKLND_CONN_BULK_IN);
2395 rc = ksocknal_tunables_init();
2399 lnet_register_lnd(&the_ksocklnd);
2404 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2405 MODULE_DESCRIPTION("TCP Socket LNet Network Driver");
2406 MODULE_VERSION("2.8.0");
2407 MODULE_LICENSE("GPL");
2409 module_init(ksocklnd_init);
2410 module_exit(ksocklnd_exit);