4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lnet/klnds/socklnd/socklnd.c
33 * Author: Zach Brown <zab@zabbo.net>
34 * Author: Peter J. Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Eric Barton <eric@bartonsoftware.com>
39 #include <linux/ethtool.h>
40 #include <linux/inetdevice.h>
41 #include <linux/kernel.h>
42 #include <linux/sunrpc/addr.h>
43 #include <net/addrconf.h>
46 static const struct lnet_lnd the_ksocklnd;
47 struct ksock_nal_data ksocknal_data;
49 static int ksocknal_ip2index(struct sockaddr *addr, struct lnet_ni *ni,
52 struct net_device *dev;
54 DECLARE_CONST_IN_IFADDR(ifa);
58 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
62 for_each_netdev(ni->ni_net_ns, dev) {
63 int flags = dev_get_flags(dev);
64 struct in_device *in_dev;
66 if (flags & IFF_LOOPBACK) /* skip the loopback IF */
69 if (!(flags & IFF_UP))
72 switch (addr->sa_family) {
74 in_dev = __in_dev_get_rcu(dev);
78 in_dev_for_each_ifa_rcu(ifa, in_dev) {
80 ((struct sockaddr_in *)addr)->sin_addr.s_addr)
85 #if IS_ENABLED(CONFIG_IPV6)
87 struct inet6_dev *in6_dev;
88 const struct inet6_ifaddr *ifa6;
89 struct sockaddr_in6 *addr6 = (struct sockaddr_in6*)addr;
91 in6_dev = __in6_dev_get(dev);
95 list_for_each_entry_rcu(ifa6, &in6_dev->addr_list, if_list) {
96 if (ipv6_addr_cmp(&ifa6->addr,
97 &addr6->sin6_addr) == 0)
102 #endif /* IS_ENABLED(CONFIG_IPV6) */
113 ((dev->reg_state == NETREG_UNREGISTERING) ||
114 ((dev->operstate != IF_OPER_UP) &&
115 (dev->operstate != IF_OPER_UNKNOWN))) ||
116 (lnet_get_link_status(dev) == 0))
122 static struct ksock_conn_cb *
123 ksocknal_create_conn_cb(struct sockaddr *addr)
125 struct ksock_conn_cb *conn_cb;
127 LIBCFS_ALLOC(conn_cb, sizeof(*conn_cb));
131 refcount_set(&conn_cb->ksnr_refcount, 1);
132 conn_cb->ksnr_peer = NULL;
133 conn_cb->ksnr_retry_interval = 0; /* OK to connect at any time */
134 rpc_copy_addr((struct sockaddr *)&conn_cb->ksnr_addr, addr);
135 rpc_set_port((struct sockaddr *)&conn_cb->ksnr_addr,
137 conn_cb->ksnr_scheduled = 0;
138 conn_cb->ksnr_connecting = 0;
139 conn_cb->ksnr_connected = 0;
140 conn_cb->ksnr_deleted = 0;
141 conn_cb->ksnr_conn_count = 0;
142 conn_cb->ksnr_ctrl_conn_count = 0;
143 conn_cb->ksnr_blki_conn_count = 0;
144 conn_cb->ksnr_blko_conn_count = 0;
145 conn_cb->ksnr_max_conns = 0;
146 conn_cb->ksnr_busy_retry_count = 0;
152 ksocknal_destroy_conn_cb(struct ksock_conn_cb *conn_cb)
154 LASSERT(refcount_read(&conn_cb->ksnr_refcount) == 0);
156 if (conn_cb->ksnr_peer)
157 ksocknal_peer_decref(conn_cb->ksnr_peer);
159 LIBCFS_FREE(conn_cb, sizeof(*conn_cb));
162 static struct ksock_peer_ni *
163 ksocknal_create_peer(struct lnet_ni *ni, struct lnet_processid *id)
165 int cpt = lnet_nid2cpt(&id->nid, ni);
166 struct ksock_net *net = ni->ni_data;
167 struct ksock_peer_ni *peer_ni;
169 LASSERT(!LNET_NID_IS_ANY(&id->nid));
170 LASSERT(id->pid != LNET_PID_ANY);
171 LASSERT(!in_interrupt());
173 if (!atomic_inc_unless_negative(&net->ksnn_npeers)) {
174 CERROR("Can't create peer_ni: network shutdown\n");
175 return ERR_PTR(-ESHUTDOWN);
178 LIBCFS_CPT_ALLOC(peer_ni, lnet_cpt_table(), cpt, sizeof(*peer_ni));
180 atomic_dec(&net->ksnn_npeers);
181 return ERR_PTR(-ENOMEM);
184 peer_ni->ksnp_ni = ni;
185 peer_ni->ksnp_id = *id;
186 refcount_set(&peer_ni->ksnp_refcount, 1); /* 1 ref for caller */
187 peer_ni->ksnp_closing = 0;
188 peer_ni->ksnp_accepting = 0;
189 peer_ni->ksnp_proto = NULL;
190 peer_ni->ksnp_last_alive = 0;
191 peer_ni->ksnp_zc_next_cookie = SOCKNAL_KEEPALIVE_PING + 1;
192 peer_ni->ksnp_conn_cb = NULL;
194 INIT_LIST_HEAD(&peer_ni->ksnp_conns);
195 INIT_LIST_HEAD(&peer_ni->ksnp_tx_queue);
196 INIT_LIST_HEAD(&peer_ni->ksnp_zc_req_list);
197 spin_lock_init(&peer_ni->ksnp_lock);
203 ksocknal_destroy_peer(struct ksock_peer_ni *peer_ni)
205 struct ksock_net *net = peer_ni->ksnp_ni->ni_data;
207 CDEBUG (D_NET, "peer_ni %s %p deleted\n",
208 libcfs_idstr(&peer_ni->ksnp_id), peer_ni);
210 LASSERT(refcount_read(&peer_ni->ksnp_refcount) == 0);
211 LASSERT(peer_ni->ksnp_accepting == 0);
212 LASSERT(list_empty(&peer_ni->ksnp_conns));
213 LASSERT(peer_ni->ksnp_conn_cb == NULL);
214 LASSERT(list_empty(&peer_ni->ksnp_tx_queue));
215 LASSERT(list_empty(&peer_ni->ksnp_zc_req_list));
217 LIBCFS_FREE(peer_ni, sizeof(*peer_ni));
219 /* NB a peer_ni's connections and conn_cb keep a reference on their
220 * peer_ni until they are destroyed, so we can be assured that _all_
221 * state to do with this peer_ni has been cleaned up when its refcount
224 if (atomic_dec_and_test(&net->ksnn_npeers))
225 wake_up_var(&net->ksnn_npeers);
228 struct ksock_peer_ni *
229 ksocknal_find_peer_locked(struct lnet_ni *ni, struct lnet_processid *id)
231 struct ksock_peer_ni *peer_ni;
232 unsigned long hash = nidhash(&id->nid);
234 hash_for_each_possible(ksocknal_data.ksnd_peers, peer_ni,
236 LASSERT(!peer_ni->ksnp_closing);
238 if (peer_ni->ksnp_ni != ni)
241 if (!nid_same(&peer_ni->ksnp_id.nid, &id->nid) ||
242 peer_ni->ksnp_id.pid != id->pid)
245 CDEBUG(D_NET, "got peer_ni [%p] -> %s (%d)\n",
246 peer_ni, libcfs_idstr(id),
247 refcount_read(&peer_ni->ksnp_refcount));
253 struct ksock_peer_ni *
254 ksocknal_find_peer(struct lnet_ni *ni, struct lnet_processid *id)
256 struct ksock_peer_ni *peer_ni;
258 read_lock(&ksocknal_data.ksnd_global_lock);
259 peer_ni = ksocknal_find_peer_locked(ni, id);
260 if (peer_ni != NULL) /* +1 ref for caller? */
261 ksocknal_peer_addref(peer_ni);
262 read_unlock(&ksocknal_data.ksnd_global_lock);
268 ksocknal_unlink_peer_locked(struct ksock_peer_ni *peer_ni)
270 LASSERT(list_empty(&peer_ni->ksnp_conns));
271 LASSERT(peer_ni->ksnp_conn_cb == NULL);
272 LASSERT(!peer_ni->ksnp_closing);
273 peer_ni->ksnp_closing = 1;
274 hlist_del(&peer_ni->ksnp_list);
275 /* lose peerlist's ref */
276 ksocknal_peer_decref(peer_ni);
281 ksocknal_dump_peer_debug_info(struct ksock_peer_ni *peer_ni)
283 struct ksock_conn *conn;
284 struct list_head *ctmp;
285 struct list_head *txtmp;
289 list_for_each(ctmp, &peer_ni->ksnp_conns) {
290 conn = list_entry(ctmp, struct ksock_conn, ksnc_list);
292 if (!list_empty(&conn->ksnc_tx_queue))
293 list_for_each(txtmp, &conn->ksnc_tx_queue) txcount++;
295 CDEBUG(D_CONSOLE, "Conn %d [type, closing, crefcnt, srefcnt]: %d, %d, %d, %d\n",
299 refcount_read(&conn->ksnc_conn_refcount),
300 refcount_read(&conn->ksnc_sock_refcount));
301 CDEBUG(D_CONSOLE, "Conn %d rx [scheduled, ready, state]: %d, %d, %d\n",
303 conn->ksnc_rx_scheduled,
305 conn->ksnc_rx_state);
306 CDEBUG(D_CONSOLE, "Conn %d tx [txqcnt, scheduled, last_post, ready, deadline]: %d, %d, %lld, %d, %lld\n",
309 conn->ksnc_tx_scheduled,
310 conn->ksnc_tx_last_post,
312 conn->ksnc_rx_deadline);
314 if (conn->ksnc_scheduler)
315 CDEBUG(D_CONSOLE, "Conn %d sched [nconns, cpt]: %d, %d\n",
317 conn->ksnc_scheduler->kss_nconns,
318 conn->ksnc_scheduler->kss_cpt);
326 ksocknal_get_peer_info(struct lnet_ni *ni, int index,
327 struct lnet_processid *id, __u32 *myip, __u32 *peer_ip,
328 int *port, int *conn_count, int *share_count)
330 struct ksock_peer_ni *peer_ni;
331 struct ksock_conn_cb *conn_cb;
334 struct ksock_net *net;
336 read_lock(&ksocknal_data.ksnd_global_lock);
338 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
340 if (peer_ni->ksnp_ni != ni)
345 *id = peer_ni->ksnp_id;
346 conn_cb = peer_ni->ksnp_conn_cb;
347 if (conn_cb == NULL) {
355 ksocknal_dump_peer_debug_info(peer_ni);
357 if (conn_cb->ksnr_addr.ss_family == AF_INET) {
358 struct sockaddr_in *sa =
359 (void *)&conn_cb->ksnr_addr;
361 rc = choose_ipv4_src(myip,
362 net->ksnn_interface.ksni_index,
363 ntohl(sa->sin_addr.s_addr),
365 *peer_ip = ntohl(sa->sin_addr.s_addr);
366 *port = ntohs(sa->sin_port);
370 *peer_ip = 0xFFFFFFFF;
374 *conn_count = conn_cb->ksnr_conn_count;
379 read_unlock(&ksocknal_data.ksnd_global_lock);
384 ksocknal_get_conns_per_peer(struct ksock_peer_ni *peer_ni)
386 struct lnet_ni *ni = peer_ni->ksnp_ni;
387 struct lnet_ioctl_config_socklnd_tunables *tunables;
391 tunables = &ni->ni_lnd_tunables.lnd_tun_u.lnd_sock;
393 return tunables->lnd_conns_per_peer;
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 * two in case of loopback
409 conn_cb->ksnr_connected |= BIT(type);
411 case SOCKLND_CONN_BULK_IN:
412 conn_cb->ksnr_blki_conn_count++;
413 if (conn_cb->ksnr_blki_conn_count >= conn_cb->ksnr_max_conns)
414 conn_cb->ksnr_connected |= BIT(type);
416 case SOCKLND_CONN_BULK_OUT:
417 conn_cb->ksnr_blko_conn_count++;
418 if (conn_cb->ksnr_blko_conn_count >= conn_cb->ksnr_max_conns)
419 conn_cb->ksnr_connected |= BIT(type);
421 case SOCKLND_CONN_ANY:
422 if (conn_cb->ksnr_conn_count >= conn_cb->ksnr_max_conns)
423 conn_cb->ksnr_connected |= BIT(type);
430 CDEBUG(D_NET, "Add conn type %d, ksnr_connected %x ksnr_max_conns %d\n",
431 type, conn_cb->ksnr_connected, conn_cb->ksnr_max_conns);
436 ksocknal_decr_conn_count(struct ksock_conn_cb *conn_cb,
439 conn_cb->ksnr_conn_count--;
441 /* check if all connections of the given type got created */
443 case SOCKLND_CONN_CONTROL:
444 conn_cb->ksnr_ctrl_conn_count--;
445 /* there's a single control connection per peer,
446 * two in case of loopback
448 if (conn_cb->ksnr_ctrl_conn_count == 0)
449 conn_cb->ksnr_connected &= ~BIT(type);
451 case SOCKLND_CONN_BULK_IN:
452 conn_cb->ksnr_blki_conn_count--;
453 if (conn_cb->ksnr_blki_conn_count == 0)
454 conn_cb->ksnr_connected &= ~BIT(type);
456 case SOCKLND_CONN_BULK_OUT:
457 conn_cb->ksnr_blko_conn_count--;
458 if (conn_cb->ksnr_blko_conn_count == 0)
459 conn_cb->ksnr_connected &= ~BIT(type);
461 case SOCKLND_CONN_ANY:
462 if (conn_cb->ksnr_conn_count == 0)
463 conn_cb->ksnr_connected &= ~BIT(type);
470 CDEBUG(D_NET, "Del conn type %d, ksnr_connected %x ksnr_max_conns %d\n",
471 type, conn_cb->ksnr_connected, conn_cb->ksnr_max_conns);
475 ksocknal_associate_cb_conn_locked(struct ksock_conn_cb *conn_cb,
476 struct ksock_conn *conn)
478 int type = conn->ksnc_type;
480 conn->ksnc_conn_cb = conn_cb;
481 ksocknal_conn_cb_addref(conn_cb);
482 ksocknal_incr_conn_count(conn_cb, type);
484 /* Successful connection => further attempts can
485 * proceed immediately
487 conn_cb->ksnr_retry_interval = 0;
491 ksocknal_add_conn_cb_locked(struct ksock_peer_ni *peer_ni,
492 struct ksock_conn_cb *conn_cb)
494 struct ksock_conn *conn;
495 struct ksock_net *net = peer_ni->ksnp_ni->ni_data;
497 LASSERT(!peer_ni->ksnp_closing);
498 LASSERT(!conn_cb->ksnr_peer);
499 LASSERT(!conn_cb->ksnr_scheduled);
500 LASSERT(!conn_cb->ksnr_connecting);
501 LASSERT(conn_cb->ksnr_connected == 0);
503 conn_cb->ksnr_peer = peer_ni;
504 ksocknal_peer_addref(peer_ni);
506 /* peer_ni's route list takes over my ref on 'route' */
507 peer_ni->ksnp_conn_cb = conn_cb;
508 net->ksnn_interface.ksni_nroutes++;
510 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
511 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
512 (struct sockaddr *)&conn_cb->ksnr_addr))
514 CDEBUG(D_NET, "call ksocknal_associate_cb_conn_locked\n");
515 ksocknal_associate_cb_conn_locked(conn_cb, conn);
516 /* keep going (typed conns) */
521 ksocknal_del_conn_cb_locked(struct ksock_conn_cb *conn_cb)
523 struct ksock_peer_ni *peer_ni = conn_cb->ksnr_peer;
524 struct ksock_conn *conn;
525 struct ksock_conn *cnxt;
526 struct ksock_net *net;
528 LASSERT(!conn_cb->ksnr_deleted);
530 /* Close associated conns */
531 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
532 if (conn->ksnc_conn_cb != conn_cb)
535 ksocknal_close_conn_locked(conn, 0);
538 net = (struct ksock_net *)(peer_ni->ksnp_ni->ni_data);
539 net->ksnn_interface.ksni_nroutes--;
540 LASSERT(net->ksnn_interface.ksni_nroutes >= 0);
542 conn_cb->ksnr_deleted = 1;
543 ksocknal_conn_cb_decref(conn_cb); /* drop peer_ni's ref */
544 peer_ni->ksnp_conn_cb = NULL;
546 if (list_empty(&peer_ni->ksnp_conns)) {
547 /* I've just removed the last route to a peer_ni with no active
550 ksocknal_unlink_peer_locked(peer_ni);
555 ksocknal_get_conn_count_by_type(struct ksock_conn_cb *conn_cb,
558 unsigned int count = 0;
561 case SOCKLND_CONN_CONTROL:
562 count = conn_cb->ksnr_ctrl_conn_count;
564 case SOCKLND_CONN_BULK_IN:
565 count = conn_cb->ksnr_blki_conn_count;
567 case SOCKLND_CONN_BULK_OUT:
568 count = conn_cb->ksnr_blko_conn_count;
570 case SOCKLND_CONN_ANY:
571 count = conn_cb->ksnr_conn_count;
582 ksocknal_add_peer(struct lnet_ni *ni, struct lnet_processid *id,
583 struct sockaddr *addr)
585 struct ksock_peer_ni *peer_ni;
586 struct ksock_peer_ni *peer2;
587 struct ksock_conn_cb *conn_cb;
589 if (LNET_NID_IS_ANY(&id->nid) ||
590 id->pid == LNET_PID_ANY)
593 /* Have a brand new peer_ni ready... */
594 peer_ni = ksocknal_create_peer(ni, id);
596 return PTR_ERR(peer_ni);
598 conn_cb = ksocknal_create_conn_cb(addr);
600 ksocknal_peer_decref(peer_ni);
604 write_lock_bh(&ksocknal_data.ksnd_global_lock);
606 /* always called with a ref on ni, so shutdown can't have started */
607 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers)
610 peer2 = ksocknal_find_peer_locked(ni, id);
612 ksocknal_peer_decref(peer_ni);
615 /* peer_ni table takes my ref on peer_ni */
616 hash_add(ksocknal_data.ksnd_peers, &peer_ni->ksnp_list,
620 if (peer_ni->ksnp_conn_cb) {
621 ksocknal_conn_cb_decref(conn_cb);
623 /* Remember conns_per_peer setting at the time
624 * of connection initiation. It will define the
625 * max number of conns per type for this conn_cb
628 conn_cb->ksnr_max_conns = ksocknal_get_conns_per_peer(peer_ni);
629 ksocknal_add_conn_cb_locked(peer_ni, conn_cb);
632 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
638 ksocknal_del_peer_locked(struct ksock_peer_ni *peer_ni)
640 struct ksock_conn *conn;
641 struct ksock_conn *cnxt;
642 struct ksock_conn_cb *conn_cb;
644 LASSERT(!peer_ni->ksnp_closing);
646 /* Extra ref prevents peer_ni disappearing until I'm done with it */
647 ksocknal_peer_addref(peer_ni);
648 conn_cb = peer_ni->ksnp_conn_cb;
650 ksocknal_del_conn_cb_locked(conn_cb);
652 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns,
654 ksocknal_close_conn_locked(conn, 0);
656 ksocknal_peer_decref(peer_ni);
657 /* NB peer_ni unlinks itself when last conn/conn_cb is removed */
661 ksocknal_del_peer(struct lnet_ni *ni, struct lnet_processid *id)
664 struct hlist_node *pnxt;
665 struct ksock_peer_ni *peer_ni;
671 write_lock_bh(&ksocknal_data.ksnd_global_lock);
673 if (id && !LNET_NID_IS_ANY(&id->nid)) {
674 lo = hash_min(nidhash(&id->nid),
675 HASH_BITS(ksocknal_data.ksnd_peers));
679 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
682 for (i = lo; i <= hi; i++) {
683 hlist_for_each_entry_safe(peer_ni, pnxt,
684 &ksocknal_data.ksnd_peers[i],
686 if (peer_ni->ksnp_ni != ni)
689 if (!((!id || LNET_NID_IS_ANY(&id->nid) ||
690 nid_same(&peer_ni->ksnp_id.nid, &id->nid)) &&
691 (!id || id->pid == LNET_PID_ANY ||
692 peer_ni->ksnp_id.pid == id->pid)))
695 ksocknal_peer_addref(peer_ni); /* a ref for me... */
697 ksocknal_del_peer_locked(peer_ni);
699 if (peer_ni->ksnp_closing &&
700 !list_empty(&peer_ni->ksnp_tx_queue)) {
701 LASSERT(list_empty(&peer_ni->ksnp_conns));
702 LASSERT(peer_ni->ksnp_conn_cb == NULL);
704 list_splice_init(&peer_ni->ksnp_tx_queue,
708 ksocknal_peer_decref(peer_ni); /* ...till here */
710 rc = 0; /* matched! */
714 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
716 ksocknal_txlist_done(ni, &zombies, -ENETDOWN);
721 static struct ksock_conn *
722 ksocknal_get_conn_by_idx(struct lnet_ni *ni, int index)
724 struct ksock_peer_ni *peer_ni;
725 struct ksock_conn *conn;
728 read_lock(&ksocknal_data.ksnd_global_lock);
730 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
731 LASSERT(!peer_ni->ksnp_closing);
733 if (peer_ni->ksnp_ni != ni)
736 list_for_each_entry(conn, &peer_ni->ksnp_conns,
741 ksocknal_conn_addref(conn);
742 read_unlock(&ksocknal_data.ksnd_global_lock);
747 read_unlock(&ksocknal_data.ksnd_global_lock);
751 static struct ksock_sched *
752 ksocknal_choose_scheduler_locked(unsigned int cpt)
754 struct ksock_sched *sched = ksocknal_data.ksnd_schedulers[cpt];
757 if (sched->kss_nthreads == 0) {
758 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
759 if (sched->kss_nthreads > 0) {
760 CDEBUG(D_NET, "scheduler[%d] has no threads. selected scheduler[%d]\n",
761 cpt, sched->kss_cpt);
772 ksocknal_accept(struct lnet_ni *ni, struct socket *sock)
774 struct ksock_connreq *cr;
776 struct sockaddr_storage peer;
778 rc = lnet_sock_getaddr(sock, true, &peer);
780 CERROR("Can't determine new connection's address\n");
784 LIBCFS_ALLOC(cr, sizeof(*cr));
786 LCONSOLE_ERROR_MSG(0x12f,
787 "Dropping connection request from %pISc: memory exhausted\n",
794 cr->ksncr_sock = sock;
796 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
798 list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
799 wake_up(&ksocknal_data.ksnd_connd_waitq);
801 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
805 static const struct ln_key_list ksocknal_tunables_keys = {
806 .lkl_maxattr = LNET_NET_SOCKLND_TUNABLES_ATTR_MAX,
808 [LNET_NET_SOCKLND_TUNABLES_ATTR_CONNS_PER_PEER] = {
809 .lkp_value = "conns_per_peer",
810 .lkp_data_type = NLA_U16
812 [LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TIMEOUT] = {
813 .lkp_value = "timeout",
814 .lkp_data_type = NLA_U32
816 [LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TOS] = {
818 .lkp_data_type = NLA_S16,
824 ksocknal_nl_get(int cmd, struct sk_buff *msg, int type, void *data)
826 struct lnet_lnd_tunables *tun;
827 struct lnet_ni *ni = data;
832 if (cmd != LNET_CMD_NETS || type != LNET_NET_LOCAL_NI_ATTR_LND_TUNABLES)
835 tun = &ni->ni_lnd_tunables;
836 nla_put_u16(msg, LNET_NET_SOCKLND_TUNABLES_ATTR_CONNS_PER_PEER,
837 tun->lnd_tun_u.lnd_sock.lnd_conns_per_peer);
838 nla_put_u32(msg, LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TIMEOUT,
840 nla_put_s16(msg, LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TOS,
841 tun->lnd_tun_u.lnd_sock.lnd_tos);
847 ksocknal_nl_set_default(int cmd, int type, void *data)
849 struct lnet_lnd_tunables *tunables = data;
850 struct lnet_ioctl_config_socklnd_tunables *lt;
851 struct lnet_ioctl_config_socklnd_tunables *df;
853 lt = &tunables->lnd_tun_u.lnd_sock;
854 df = &ksock_default_tunables;
856 case LNET_NET_SOCKLND_TUNABLES_ATTR_CONNS_PER_PEER:
857 lt->lnd_conns_per_peer = df->lnd_conns_per_peer;
859 case LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TIMEOUT:
860 lt->lnd_timeout = df->lnd_timeout;
868 ksocknal_nl_set(int cmd, struct nlattr *attr, int type, void *data)
870 struct lnet_lnd_tunables *tunables = data;
874 if (cmd != LNET_CMD_NETS)
878 ksocknal_nl_set_default(cmd, type, data);
882 if (nla_type(attr) != LN_SCALAR_ATTR_INT_VALUE)
886 case LNET_NET_SOCKLND_TUNABLES_ATTR_CONNS_PER_PEER:
887 /* value values are 1 to 127. Zero mean calculate the value */
888 num = nla_get_s64(attr);
889 if (num > -1 && num < 128)
890 tunables->lnd_tun_u.lnd_sock.lnd_conns_per_peer = num;
894 case LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TIMEOUT:
895 num = nla_get_s64(attr);
896 tunables->lnd_tun_u.lnd_sock.lnd_timeout = num;
898 case LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TOS:
899 num = nla_get_s64(attr);
900 clamp_t(s64, num, -1, 0xff);
901 tunables->lnd_tun_u.lnd_sock.lnd_tos = num;
911 ksocknal_connecting(struct ksock_conn_cb *conn_cb, struct sockaddr *sa)
914 rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr, sa))
915 return conn_cb->ksnr_connecting;
920 ksocknal_create_conn(struct lnet_ni *ni, struct ksock_conn_cb *conn_cb,
921 struct socket *sock, int type)
923 rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
925 struct lnet_processid peerid;
927 struct ksock_conn *conn;
928 struct ksock_conn *conn2;
929 struct ksock_peer_ni *peer_ni = NULL;
930 struct ksock_peer_ni *peer2;
931 struct ksock_sched *sched;
932 struct ksock_hello_msg *hello;
935 struct ksock_tx *txtmp;
942 active = (conn_cb != NULL);
944 LASSERT(active == (type != SOCKLND_CONN_NONE));
946 LIBCFS_ALLOC(conn, sizeof(*conn));
952 conn->ksnc_peer = NULL;
953 conn->ksnc_conn_cb = NULL;
954 conn->ksnc_sock = sock;
955 /* 2 ref, 1 for conn, another extra ref prevents socket
956 * being closed before establishment of connection */
957 refcount_set(&conn->ksnc_sock_refcount, 2);
958 conn->ksnc_type = type;
959 ksocknal_lib_save_callback(sock, conn);
960 refcount_set(&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
962 conn->ksnc_rx_ready = 0;
963 conn->ksnc_rx_scheduled = 0;
965 INIT_LIST_HEAD(&conn->ksnc_tx_queue);
966 conn->ksnc_tx_ready = 0;
967 conn->ksnc_tx_scheduled = 0;
968 conn->ksnc_tx_carrier = NULL;
969 atomic_set (&conn->ksnc_tx_nob, 0);
971 LIBCFS_ALLOC(hello, offsetof(struct ksock_hello_msg,
972 kshm_ips[LNET_INTERFACES_NUM]));
978 /* stash conn's local and remote addrs */
979 rc = ksocknal_lib_get_conn_addrs(conn);
983 /* Find out/confirm peer_ni's NID and connection type and get the
984 * vector of interfaces she's willing to let me connect to.
985 * Passive connections use the listener timeout since the peer_ni sends
989 struct sockaddr_in *psa = (void *)&conn->ksnc_peeraddr;
991 peer_ni = conn_cb->ksnr_peer;
992 LASSERT(ni == peer_ni->ksnp_ni);
994 /* Active connection sends HELLO eagerly */
995 hello->kshm_nips = 0;
996 peerid = peer_ni->ksnp_id;
998 write_lock_bh(global_lock);
999 conn->ksnc_proto = peer_ni->ksnp_proto;
1000 write_unlock_bh(global_lock);
1002 if (conn->ksnc_proto == NULL) {
1003 if (psa->sin_family == AF_INET6)
1004 conn->ksnc_proto = &ksocknal_protocol_v4x;
1005 else if (psa->sin_family == AF_INET)
1006 conn->ksnc_proto = &ksocknal_protocol_v3x;
1007 #if SOCKNAL_VERSION_DEBUG
1008 if (*ksocknal_tunables.ksnd_protocol == 2)
1009 conn->ksnc_proto = &ksocknal_protocol_v2x;
1010 else if (*ksocknal_tunables.ksnd_protocol == 1)
1011 conn->ksnc_proto = &ksocknal_protocol_v1x;
1014 if (!conn->ksnc_proto) {
1019 rc = ksocknal_send_hello(ni, conn, &peerid.nid, hello);
1023 peerid.nid = LNET_ANY_NID;
1024 peerid.pid = LNET_PID_ANY;
1026 /* Passive, get protocol from peer_ni */
1027 conn->ksnc_proto = NULL;
1030 rc = ksocknal_recv_hello(ni, conn, hello, &peerid, &incarnation);
1034 LASSERT(rc == 0 || active);
1035 LASSERT(conn->ksnc_proto != NULL);
1036 LASSERT(!LNET_NID_IS_ANY(&peerid.nid));
1038 cpt = lnet_nid2cpt(&peerid.nid, ni);
1041 ksocknal_peer_addref(peer_ni);
1042 write_lock_bh(global_lock);
1044 peer_ni = ksocknal_create_peer(ni, &peerid);
1045 if (IS_ERR(peer_ni)) {
1046 rc = PTR_ERR(peer_ni);
1050 write_lock_bh(global_lock);
1052 /* called with a ref on ni, so shutdown can't have started */
1053 LASSERT(atomic_read(&((struct ksock_net *)ni->ni_data)->ksnn_npeers) >= 0);
1055 peer2 = ksocknal_find_peer_locked(ni, &peerid);
1056 if (peer2 == NULL) {
1057 /* NB this puts an "empty" peer_ni in the peer_ni
1058 * table (which takes my ref) */
1059 hash_add(ksocknal_data.ksnd_peers,
1060 &peer_ni->ksnp_list, nidhash(&peerid.nid));
1062 ksocknal_peer_decref(peer_ni);
1067 ksocknal_peer_addref(peer_ni);
1068 peer_ni->ksnp_accepting++;
1070 /* Am I already connecting to this guy? Resolve in
1071 * favour of higher NID...
1073 if (memcmp(&peerid.nid, &ni->ni_nid, sizeof(peerid.nid)) < 0 &&
1074 ksocknal_connecting(peer_ni->ksnp_conn_cb,
1075 ((struct sockaddr *) &conn->ksnc_peeraddr))) {
1077 warn = "connection race resolution";
1082 if (peer_ni->ksnp_closing ||
1083 (active && conn_cb->ksnr_deleted)) {
1084 /* peer_ni/conn_cb got closed under me */
1086 warn = "peer_ni/conn_cb removed";
1090 if (peer_ni->ksnp_proto == NULL) {
1091 /* Never connected before.
1092 * NB recv_hello may have returned EPROTO to signal my peer_ni
1093 * wants a different protocol than the one I asked for.
1095 LASSERT(list_empty(&peer_ni->ksnp_conns));
1097 peer_ni->ksnp_proto = conn->ksnc_proto;
1098 peer_ni->ksnp_incarnation = incarnation;
1101 if (peer_ni->ksnp_proto != conn->ksnc_proto ||
1102 peer_ni->ksnp_incarnation != incarnation) {
1103 /* peer_ni rebooted or I've got the wrong protocol version */
1104 ksocknal_close_peer_conns_locked(peer_ni, NULL, 0);
1106 peer_ni->ksnp_proto = NULL;
1108 warn = peer_ni->ksnp_incarnation != incarnation ?
1109 "peer_ni rebooted" :
1110 "wrong proto version";
1120 warn = "lost conn race";
1123 warn = "retry with different protocol version";
1127 /* Refuse to duplicate an existing connection, unless this is a
1128 * loopback connection */
1129 if (!rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
1130 (struct sockaddr *)&conn->ksnc_myaddr)) {
1131 list_for_each_entry(conn2, &peer_ni->ksnp_conns, ksnc_list) {
1133 (struct sockaddr *)&conn2->ksnc_peeraddr,
1134 (struct sockaddr *)&conn->ksnc_peeraddr) ||
1136 (struct sockaddr *)&conn2->ksnc_myaddr,
1137 (struct sockaddr *)&conn->ksnc_myaddr) ||
1138 conn2->ksnc_type != conn->ksnc_type)
1142 /* If max conns per type is not registered in conn_cb
1143 * as ksnr_max_conns, use ni's conns_per_peer
1145 if ((peer_ni->ksnp_conn_cb &&
1146 num_dup < peer_ni->ksnp_conn_cb->ksnr_max_conns) ||
1147 (!peer_ni->ksnp_conn_cb &&
1148 num_dup < ksocknal_get_conns_per_peer(peer_ni)))
1151 /* Reply on a passive connection attempt so the peer_ni
1152 * realises we're connected.
1162 /* If the connection created by this route didn't bind to the IP
1163 * address the route connected to, the connection/route matching
1164 * code below probably isn't going to work.
1167 !rpc_cmp_addr((struct sockaddr *)&conn_cb->ksnr_addr,
1168 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1169 CERROR("Route %s %pISc connected to %pISc\n",
1170 libcfs_idstr(&peer_ni->ksnp_id),
1171 &conn_cb->ksnr_addr,
1172 &conn->ksnc_peeraddr);
1175 /* Search for a conn_cb corresponding to the new connection and
1176 * create an association. This allows incoming connections created
1177 * by conn_cbs in my peer_ni to match my own conn_cb entries so I don't
1178 * continually create duplicate conn_cbs.
1180 conn_cb = peer_ni->ksnp_conn_cb;
1182 if (conn_cb && rpc_cmp_addr((struct sockaddr *)&conn->ksnc_peeraddr,
1183 (struct sockaddr *)&conn_cb->ksnr_addr))
1184 ksocknal_associate_cb_conn_locked(conn_cb, conn);
1186 conn->ksnc_peer = peer_ni; /* conn takes my ref on peer_ni */
1187 peer_ni->ksnp_last_alive = ktime_get_seconds();
1188 peer_ni->ksnp_send_keepalive = 0;
1189 peer_ni->ksnp_error = 0;
1191 sched = ksocknal_choose_scheduler_locked(cpt);
1193 CERROR("no schedulers available. node is unhealthy\n");
1197 * The cpt might have changed if we ended up selecting a non cpt
1198 * native scheduler. So use the scheduler's cpt instead.
1200 cpt = sched->kss_cpt;
1201 sched->kss_nconns++;
1202 conn->ksnc_scheduler = sched;
1204 conn->ksnc_tx_last_post = ktime_get_seconds();
1205 /* Set the deadline for the outgoing HELLO to drain */
1206 conn->ksnc_tx_bufnob = sock->sk->sk_wmem_queued;
1207 conn->ksnc_tx_deadline = ktime_get_seconds() +
1209 smp_mb(); /* order with adding to peer_ni's conn list */
1211 list_add(&conn->ksnc_list, &peer_ni->ksnp_conns);
1212 ksocknal_conn_addref(conn);
1214 ksocknal_new_packet(conn, 0);
1216 conn->ksnc_zc_capable = ksocknal_lib_zc_capable(conn);
1218 /* Take packets blocking for this connection. */
1219 list_for_each_entry_safe(tx, txtmp, &peer_ni->ksnp_tx_queue, tx_list) {
1220 if (conn->ksnc_proto->pro_match_tx(conn, tx, tx->tx_nonblk) ==
1224 list_del(&tx->tx_list);
1225 ksocknal_queue_tx_locked(tx, conn);
1228 write_unlock_bh(global_lock);
1229 /* We've now got a new connection. Any errors from here on are just
1230 * like "normal" comms errors and we close the connection normally.
1231 * NB (a) we still have to send the reply HELLO for passive
1233 * (b) normal I/O on the conn is blocked until I setup and call the
1237 CDEBUG(D_NET, "New conn %s p %d.x %pISc -> %pIScp"
1238 " incarnation:%lld sched[%d]\n",
1239 libcfs_idstr(&peerid), conn->ksnc_proto->pro_version,
1240 &conn->ksnc_myaddr, &conn->ksnc_peeraddr,
1244 hello->kshm_nips = 0;
1245 rc = ksocknal_send_hello(ni, conn, &peerid.nid, hello);
1248 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1249 kshm_ips[LNET_INTERFACES_NUM]));
1251 /* setup the socket AFTER I've received hello (it disables
1252 * SO_LINGER). I might call back to the acceptor who may want
1253 * to send a protocol version response and then close the
1254 * socket; this ensures the socket only tears down after the
1255 * response has been sent.
1258 rc = ksocknal_lib_setup_sock(sock, ni);
1260 write_lock_bh(global_lock);
1262 /* NB my callbacks block while I hold ksnd_global_lock */
1263 ksocknal_lib_set_callback(sock, conn);
1266 peer_ni->ksnp_accepting--;
1268 write_unlock_bh(global_lock);
1271 write_lock_bh(global_lock);
1272 if (!conn->ksnc_closing) {
1273 /* could be closed by another thread */
1274 ksocknal_close_conn_locked(conn, rc);
1276 write_unlock_bh(global_lock);
1277 } else if (ksocknal_connsock_addref(conn) == 0) {
1278 /* Allow I/O to proceed. */
1279 ksocknal_read_callback(conn);
1280 ksocknal_write_callback(conn);
1281 ksocknal_connsock_decref(conn);
1284 ksocknal_connsock_decref(conn);
1285 ksocknal_conn_decref(conn);
1290 if (!peer_ni->ksnp_closing &&
1291 list_empty(&peer_ni->ksnp_conns) &&
1292 peer_ni->ksnp_conn_cb == NULL) {
1293 list_splice_init(&peer_ni->ksnp_tx_queue, &zombies);
1294 ksocknal_unlink_peer_locked(peer_ni);
1297 write_unlock_bh(global_lock);
1301 CERROR("Not creating conn %s type %d: %s\n",
1302 libcfs_idstr(&peerid), conn->ksnc_type, warn);
1304 CDEBUG(D_NET, "Not creating conn %s type %d: %s\n",
1305 libcfs_idstr(&peerid), conn->ksnc_type, warn);
1310 /* Request retry by replying with CONN_NONE
1311 * ksnc_proto has been set already
1313 conn->ksnc_type = SOCKLND_CONN_NONE;
1314 hello->kshm_nips = 0;
1315 ksocknal_send_hello(ni, conn, &peerid.nid, hello);
1318 write_lock_bh(global_lock);
1319 peer_ni->ksnp_accepting--;
1320 write_unlock_bh(global_lock);
1324 * If we get here without an error code, just use -EALREADY.
1325 * Depending on how we got here, the error may be positive
1326 * or negative. Normalize the value for ksocknal_txlist_done().
1328 rc2 = (rc == 0 ? -EALREADY : (rc > 0 ? -rc : rc));
1329 ksocknal_txlist_done(ni, &zombies, rc2);
1330 ksocknal_peer_decref(peer_ni);
1334 LIBCFS_FREE(hello, offsetof(struct ksock_hello_msg,
1335 kshm_ips[LNET_INTERFACES_NUM]));
1337 LIBCFS_FREE(conn, sizeof(*conn));
1346 ksocknal_close_conn_locked(struct ksock_conn *conn, int error)
1348 /* This just does the immmediate housekeeping, and queues the
1349 * connection for the reaper to terminate.
1350 * Caller holds ksnd_global_lock exclusively in irq context */
1351 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1352 struct ksock_conn_cb *conn_cb;
1353 struct ksock_conn *conn2;
1355 int duplicate_count = 0;
1357 LASSERT(peer_ni->ksnp_error == 0);
1358 LASSERT(!conn->ksnc_closing);
1359 conn->ksnc_closing = 1;
1361 /* ksnd_deathrow_conns takes over peer_ni's ref */
1362 list_del(&conn->ksnc_list);
1364 conn_cb = conn->ksnc_conn_cb;
1365 if (conn_cb != NULL) {
1366 /* dissociate conn from cb... */
1367 LASSERT(!conn_cb->ksnr_deleted);
1369 conn_count = ksocknal_get_conn_count_by_type(conn_cb,
1371 /* connected bit is set only if all connections
1372 * of the given type got created
1374 if (conn_count == conn_cb->ksnr_max_conns)
1375 LASSERT((conn_cb->ksnr_connected &
1376 BIT(conn->ksnc_type)) != 0);
1378 if (conn_count == 1) {
1379 list_for_each_entry(conn2, &peer_ni->ksnp_conns,
1381 if (conn2->ksnc_conn_cb == conn_cb &&
1382 conn2->ksnc_type == conn->ksnc_type)
1383 duplicate_count += 1;
1385 if (duplicate_count > 0)
1386 CERROR("Found %d duplicate conns type %d\n",
1390 ksocknal_decr_conn_count(conn_cb, conn->ksnc_type);
1392 conn->ksnc_conn_cb = NULL;
1394 /* drop conn's ref on conn_cb */
1395 ksocknal_conn_cb_decref(conn_cb);
1398 if (list_empty(&peer_ni->ksnp_conns)) {
1399 /* No more connections to this peer_ni */
1401 if (!list_empty(&peer_ni->ksnp_tx_queue)) {
1402 struct ksock_tx *tx;
1404 LASSERT(conn->ksnc_proto == &ksocknal_protocol_v3x);
1406 /* throw them to the last connection...,
1407 * these TXs will be send to /dev/null by scheduler */
1408 list_for_each_entry(tx, &peer_ni->ksnp_tx_queue,
1410 ksocknal_tx_prep(conn, tx);
1412 spin_lock_bh(&conn->ksnc_scheduler->kss_lock);
1413 list_splice_init(&peer_ni->ksnp_tx_queue,
1414 &conn->ksnc_tx_queue);
1415 spin_unlock_bh(&conn->ksnc_scheduler->kss_lock);
1418 /* renegotiate protocol version */
1419 peer_ni->ksnp_proto = NULL;
1420 /* stash last conn close reason */
1421 peer_ni->ksnp_error = error;
1423 if (peer_ni->ksnp_conn_cb == NULL) {
1424 /* I've just closed last conn belonging to a
1425 * peer_ni with no connections to it
1427 ksocknal_unlink_peer_locked(peer_ni);
1431 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1433 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_deathrow_conns);
1434 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1436 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1440 ksocknal_peer_failed(struct ksock_peer_ni *peer_ni)
1442 bool notify = false;
1443 time64_t last_alive = 0;
1445 /* There has been a connection failure or comms error; but I'll only
1446 * tell LNET I think the peer_ni is dead if it's to another kernel and
1447 * there are no connections or connection attempts in existence. */
1449 read_lock(&ksocknal_data.ksnd_global_lock);
1451 if ((peer_ni->ksnp_id.pid & LNET_PID_USERFLAG) == 0 &&
1452 list_empty(&peer_ni->ksnp_conns) &&
1453 peer_ni->ksnp_accepting == 0 &&
1454 !ksocknal_find_connecting_conn_cb_locked(peer_ni)) {
1456 last_alive = peer_ni->ksnp_last_alive;
1459 read_unlock(&ksocknal_data.ksnd_global_lock);
1462 lnet_notify(peer_ni->ksnp_ni,
1463 &peer_ni->ksnp_id.nid,
1464 false, false, last_alive);
1468 ksocknal_finalize_zcreq(struct ksock_conn *conn)
1470 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1471 struct ksock_tx *tx;
1472 struct ksock_tx *tmp;
1475 /* NB safe to finalize TXs because closing of socket will
1476 * abort all buffered data */
1477 LASSERT(conn->ksnc_sock == NULL);
1479 spin_lock(&peer_ni->ksnp_lock);
1481 list_for_each_entry_safe(tx, tmp, &peer_ni->ksnp_zc_req_list,
1483 if (tx->tx_conn != conn)
1486 LASSERT(tx->tx_msg.ksm_zc_cookies[0] != 0);
1488 tx->tx_msg.ksm_zc_cookies[0] = 0;
1489 tx->tx_zc_aborted = 1; /* mark it as not-acked */
1490 list_move(&tx->tx_zc_list, &zlist);
1493 spin_unlock(&peer_ni->ksnp_lock);
1495 while ((tx = list_first_entry_or_null(&zlist, struct ksock_tx,
1496 tx_zc_list)) != NULL) {
1497 list_del(&tx->tx_zc_list);
1498 ksocknal_tx_decref(tx);
1503 ksocknal_terminate_conn(struct ksock_conn *conn)
1505 /* This gets called by the reaper (guaranteed thread context) to
1506 * disengage the socket from its callbacks and close it.
1507 * ksnc_refcount will eventually hit zero, and then the reaper will
1510 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1511 struct ksock_sched *sched = conn->ksnc_scheduler;
1512 bool failed = false;
1514 LASSERT(conn->ksnc_closing);
1516 /* wake up the scheduler to "send" all remaining packets to /dev/null */
1517 spin_lock_bh(&sched->kss_lock);
1519 /* a closing conn is always ready to tx */
1520 conn->ksnc_tx_ready = 1;
1522 if (!conn->ksnc_tx_scheduled &&
1523 !list_empty(&conn->ksnc_tx_queue)) {
1524 list_add_tail(&conn->ksnc_tx_list,
1525 &sched->kss_tx_conns);
1526 conn->ksnc_tx_scheduled = 1;
1527 /* extra ref for scheduler */
1528 ksocknal_conn_addref(conn);
1530 wake_up(&sched->kss_waitq);
1533 spin_unlock_bh(&sched->kss_lock);
1535 /* serialise with callbacks */
1536 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1538 ksocknal_lib_reset_callback(conn->ksnc_sock, conn);
1540 /* OK, so this conn may not be completely disengaged from its
1541 * scheduler yet, but it _has_ committed to terminate...
1543 conn->ksnc_scheduler->kss_nconns--;
1545 if (peer_ni->ksnp_error != 0) {
1546 /* peer_ni's last conn closed in error */
1547 LASSERT(list_empty(&peer_ni->ksnp_conns));
1549 peer_ni->ksnp_error = 0; /* avoid multiple notifications */
1552 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1555 ksocknal_peer_failed(peer_ni);
1557 /* The socket is closed on the final put; either here, or in
1558 * ksocknal_{send,recv}msg(). Since we set up the linger2 option
1559 * when the connection was established, this will close the socket
1560 * immediately, aborting anything buffered in it. Any hung
1561 * zero-copy transmits will therefore complete in finite time.
1563 ksocknal_connsock_decref(conn);
1567 ksocknal_queue_zombie_conn(struct ksock_conn *conn)
1569 /* Queue the conn for the reaper to destroy */
1570 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1571 spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
1573 list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
1574 wake_up(&ksocknal_data.ksnd_reaper_waitq);
1576 spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
1580 ksocknal_destroy_conn(struct ksock_conn *conn)
1584 /* Final coup-de-grace of the reaper */
1585 CDEBUG(D_NET, "connection %p\n", conn);
1587 LASSERT(refcount_read(&conn->ksnc_conn_refcount) == 0);
1588 LASSERT(refcount_read(&conn->ksnc_sock_refcount) == 0);
1589 LASSERT(conn->ksnc_sock == NULL);
1590 LASSERT(conn->ksnc_conn_cb == NULL);
1591 LASSERT(!conn->ksnc_tx_scheduled);
1592 LASSERT(!conn->ksnc_rx_scheduled);
1593 LASSERT(list_empty(&conn->ksnc_tx_queue));
1595 /* complete current receive if any */
1596 switch (conn->ksnc_rx_state) {
1597 case SOCKNAL_RX_LNET_PAYLOAD:
1598 last_rcv = conn->ksnc_rx_deadline -
1600 CERROR("Completing partial receive from %s[%d], ip %pIScp, with error, wanted: %d, left: %d, last alive is %lld secs ago\n",
1601 libcfs_idstr(&conn->ksnc_peer->ksnp_id),
1603 &conn->ksnc_peeraddr,
1604 conn->ksnc_rx_nob_wanted, conn->ksnc_rx_nob_left,
1605 ktime_get_seconds() - last_rcv);
1606 if (conn->ksnc_lnet_msg)
1607 conn->ksnc_lnet_msg->msg_health_status =
1608 LNET_MSG_STATUS_REMOTE_ERROR;
1609 lnet_finalize(conn->ksnc_lnet_msg, -EIO);
1611 case SOCKNAL_RX_LNET_HEADER:
1612 if (conn->ksnc_rx_started)
1613 CERROR("Incomplete receive of lnet header from %s, ip %pIScp, with error, protocol: %d.x.\n",
1614 libcfs_idstr(&conn->ksnc_peer->ksnp_id),
1615 &conn->ksnc_peeraddr,
1616 conn->ksnc_proto->pro_version);
1618 case SOCKNAL_RX_KSM_HEADER:
1619 if (conn->ksnc_rx_started)
1620 CERROR("Incomplete receive of ksock message from %s, ip %pIScp, with error, protocol: %d.x.\n",
1621 libcfs_idstr(&conn->ksnc_peer->ksnp_id),
1622 &conn->ksnc_peeraddr,
1623 conn->ksnc_proto->pro_version);
1625 case SOCKNAL_RX_SLOP:
1626 if (conn->ksnc_rx_started)
1627 CERROR("Incomplete receive of slops from %s, ip %pIScp, with error\n",
1628 libcfs_idstr(&conn->ksnc_peer->ksnp_id),
1629 &conn->ksnc_peeraddr);
1636 ksocknal_peer_decref(conn->ksnc_peer);
1638 LIBCFS_FREE(conn, sizeof(*conn));
1642 ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
1643 struct sockaddr *addr, int why)
1645 struct ksock_conn *conn;
1646 struct ksock_conn *cnxt;
1649 list_for_each_entry_safe(conn, cnxt, &peer_ni->ksnp_conns, ksnc_list) {
1652 (struct sockaddr *)&conn->ksnc_peeraddr)) {
1654 ksocknal_close_conn_locked(conn, why);
1662 ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why)
1664 struct ksock_peer_ni *peer_ni = conn->ksnc_peer;
1667 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1669 count = ksocknal_close_peer_conns_locked(
1670 peer_ni, (struct sockaddr *)&conn->ksnc_peeraddr, why);
1672 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1678 ksocknal_close_matching_conns(struct lnet_processid *id, __u32 ipaddr)
1680 struct ksock_peer_ni *peer_ni;
1681 struct hlist_node *pnxt;
1686 struct sockaddr_in sa = {.sin_family = AF_INET};
1688 write_lock_bh(&ksocknal_data.ksnd_global_lock);
1690 if (!LNET_NID_IS_ANY(&id->nid)) {
1691 lo = hash_min(nidhash(&id->nid),
1692 HASH_BITS(ksocknal_data.ksnd_peers));
1696 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1699 sa.sin_addr.s_addr = htonl(ipaddr);
1700 for (i = lo; i <= hi; i++) {
1701 hlist_for_each_entry_safe(peer_ni, pnxt,
1702 &ksocknal_data.ksnd_peers[i],
1705 if (!((LNET_NID_IS_ANY(&id->nid) ||
1706 nid_same(&id->nid, &peer_ni->ksnp_id.nid)) &&
1707 (id->pid == LNET_PID_ANY ||
1708 id->pid == peer_ni->ksnp_id.pid)))
1711 count += ksocknal_close_peer_conns_locked(
1713 ipaddr ? (struct sockaddr *)&sa : NULL, 0);
1717 write_unlock_bh(&ksocknal_data.ksnd_global_lock);
1719 /* wildcards always succeed */
1720 if (LNET_NID_IS_ANY(&id->nid) || id->pid == LNET_PID_ANY ||
1724 return (count == 0 ? -ENOENT : 0);
1728 ksocknal_notify_gw_down(struct lnet_nid *gw_nid)
1730 /* The router is telling me she's been notified of a change in
1733 struct lnet_processid id = {
1734 .pid = LNET_PID_ANY,
1738 CDEBUG(D_NET, "gw %s down\n", libcfs_nidstr(gw_nid));
1740 /* If the gateway crashed, close all open connections... */
1741 ksocknal_close_matching_conns(&id, 0);
1744 /* We can only establish new connections
1745 * if we have autroutes, and these connect on demand.
1750 ksocknal_push_peer(struct ksock_peer_ni *peer_ni)
1754 struct ksock_conn *conn;
1756 for (index = 0; ; index++) {
1757 read_lock(&ksocknal_data.ksnd_global_lock);
1762 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
1764 ksocknal_conn_addref(conn);
1769 read_unlock(&ksocknal_data.ksnd_global_lock);
1774 ksocknal_lib_push_conn (conn);
1775 ksocknal_conn_decref(conn);
1780 ksocknal_push(struct lnet_ni *ni, struct lnet_processid *id)
1787 if (!LNET_NID_IS_ANY(&id->nid)) {
1788 lo = hash_min(nidhash(&id->nid),
1789 HASH_BITS(ksocknal_data.ksnd_peers));
1793 hi = HASH_SIZE(ksocknal_data.ksnd_peers) - 1;
1796 for (bkt = lo; bkt <= hi; bkt++) {
1797 int peer_off; /* searching offset in peer_ni hash table */
1799 for (peer_off = 0; ; peer_off++) {
1800 struct ksock_peer_ni *peer_ni;
1803 read_lock(&ksocknal_data.ksnd_global_lock);
1804 hlist_for_each_entry(peer_ni,
1805 &ksocknal_data.ksnd_peers[bkt],
1807 if (!((LNET_NID_IS_ANY(&id->nid) ||
1809 &peer_ni->ksnp_id.nid)) &&
1810 (id->pid == LNET_PID_ANY ||
1811 id->pid == peer_ni->ksnp_id.pid)))
1814 if (i++ == peer_off) {
1815 ksocknal_peer_addref(peer_ni);
1819 read_unlock(&ksocknal_data.ksnd_global_lock);
1821 if (i <= peer_off) /* no match */
1825 ksocknal_push_peer(peer_ni);
1826 ksocknal_peer_decref(peer_ni);
1833 ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg)
1835 struct lnet_processid id = {};
1836 struct libcfs_ioctl_data *data = arg;
1840 case IOC_LIBCFS_GET_INTERFACE: {
1841 struct ksock_net *net = ni->ni_data;
1842 struct ksock_interface *iface;
1843 struct sockaddr_in *sa;
1845 read_lock(&ksocknal_data.ksnd_global_lock);
1847 if (data->ioc_count >= 1) {
1851 iface = &net->ksnn_interface;
1853 sa = (void *)&iface->ksni_addr;
1854 if (sa->sin_family == AF_INET) {
1855 data->ioc_u32[0] = ntohl(sa->sin_addr.s_addr);
1856 data->ioc_u32[1] = iface->ksni_netmask;
1858 data->ioc_u32[0] = 0xFFFFFFFF;
1859 data->ioc_u32[1] = 0;
1861 data->ioc_u32[2] = iface->ksni_npeers;
1862 data->ioc_u32[3] = iface->ksni_nroutes;
1865 read_unlock(&ksocknal_data.ksnd_global_lock);
1869 case IOC_LIBCFS_GET_PEER: {
1874 int share_count = 0;
1876 rc = ksocknal_get_peer_info(ni, data->ioc_count,
1877 &id, &myip, &ip, &port,
1878 &conn_count, &share_count);
1882 if (!nid_is_nid4(&id.nid))
1884 data->ioc_nid = lnet_nid_to_nid4(&id.nid);
1885 data->ioc_count = share_count;
1886 data->ioc_u32[0] = ip;
1887 data->ioc_u32[1] = port;
1888 data->ioc_u32[2] = myip;
1889 data->ioc_u32[3] = conn_count;
1890 data->ioc_u32[4] = id.pid;
1894 case IOC_LIBCFS_ADD_PEER: {
1895 struct sockaddr_in sa = {.sin_family = AF_INET};
1897 id.pid = LNET_PID_LUSTRE;
1898 lnet_nid4_to_nid(data->ioc_nid, &id.nid);
1899 sa.sin_addr.s_addr = htonl(data->ioc_u32[0]);
1900 sa.sin_port = htons(data->ioc_u32[1]);
1901 return ksocknal_add_peer(ni, &id, (struct sockaddr *)&sa);
1903 case IOC_LIBCFS_DEL_PEER:
1904 lnet_nid4_to_nid(data->ioc_nid, &id.nid);
1905 id.pid = LNET_PID_ANY;
1906 return ksocknal_del_peer(ni, &id);
1908 case IOC_LIBCFS_GET_CONN: {
1912 struct ksock_conn *conn = ksocknal_get_conn_by_idx(ni, data->ioc_count);
1913 struct sockaddr_in *psa = (void *)&conn->ksnc_peeraddr;
1914 struct sockaddr_in *mysa = (void *)&conn->ksnc_myaddr;
1919 ksocknal_lib_get_conn_tunables(conn, &txmem, &rxmem, &nagle);
1921 data->ioc_count = txmem;
1922 data->ioc_nid = lnet_nid_to_nid4(&conn->ksnc_peer->ksnp_id.nid);
1923 data->ioc_flags = nagle;
1924 if (psa->sin_family == AF_INET)
1925 data->ioc_u32[0] = ntohl(psa->sin_addr.s_addr);
1927 data->ioc_u32[0] = 0xFFFFFFFF;
1928 data->ioc_u32[1] = rpc_get_port((struct sockaddr *)
1929 &conn->ksnc_peeraddr);
1930 if (mysa->sin_family == AF_INET)
1931 data->ioc_u32[2] = ntohl(mysa->sin_addr.s_addr);
1933 data->ioc_u32[2] = 0xFFFFFFFF;
1934 data->ioc_u32[3] = conn->ksnc_type;
1935 data->ioc_u32[4] = conn->ksnc_scheduler->kss_cpt;
1936 data->ioc_u32[5] = rxmem;
1937 data->ioc_u32[6] = conn->ksnc_peer->ksnp_id.pid;
1938 ksocknal_conn_decref(conn);
1942 case IOC_LIBCFS_CLOSE_CONNECTION:
1943 lnet_nid4_to_nid(data->ioc_nid, &id.nid);
1944 id.pid = LNET_PID_ANY;
1945 return ksocknal_close_matching_conns(&id,
1948 case IOC_LIBCFS_REGISTER_MYNID:
1949 /* Ignore if this is a noop */
1950 if (nid_is_nid4(&ni->ni_nid) &&
1951 data->ioc_nid == lnet_nid_to_nid4(&ni->ni_nid))
1954 CERROR("obsolete IOC_LIBCFS_REGISTER_MYNID: %s(%s)\n",
1955 libcfs_nid2str(data->ioc_nid),
1956 libcfs_nidstr(&ni->ni_nid));
1959 case IOC_LIBCFS_PUSH_CONNECTION:
1960 lnet_nid4_to_nid(data->ioc_nid, &id.nid);
1961 id.pid = LNET_PID_ANY;
1962 return ksocknal_push(ni, &id);
1971 ksocknal_free_buffers (void)
1973 LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
1975 if (ksocknal_data.ksnd_schedulers != NULL)
1976 cfs_percpt_free(ksocknal_data.ksnd_schedulers);
1978 spin_lock(&ksocknal_data.ksnd_tx_lock);
1980 if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
1982 struct ksock_tx *tx;
1984 list_splice_init(&ksocknal_data.ksnd_idle_noop_txs, &zlist);
1985 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1987 while ((tx = list_first_entry_or_null(&zlist, struct ksock_tx,
1988 tx_list)) != NULL) {
1989 list_del(&tx->tx_list);
1990 LIBCFS_FREE(tx, tx->tx_desc_size);
1993 spin_unlock(&ksocknal_data.ksnd_tx_lock);
1998 ksocknal_handle_link_state_change(struct net_device *dev,
1999 unsigned char operstate)
2001 struct lnet_ni *ni = NULL;
2002 struct ksock_net *net;
2003 struct ksock_net *cnxt;
2005 unsigned char link_down;
2006 struct in_device *in_dev;
2007 bool found_ip = false;
2008 struct ksock_interface *ksi = NULL;
2009 struct sockaddr_in *sa;
2010 __u32 ni_state_before;
2011 bool update_ping_buf = false;
2013 DECLARE_CONST_IN_IFADDR(ifa);
2015 link_down = !((operstate == IF_OPER_UP) || (operstate == IF_OPER_UNKNOWN));
2016 ifindex = dev->ifindex;
2018 if (!ksocknal_data.ksnd_nnets)
2021 list_for_each_entry_safe(net, cnxt, &ksocknal_data.ksnd_nets,
2024 ksi = &net->ksnn_interface;
2025 sa = (void *)&ksi->ksni_addr;
2028 if (strcmp(ksi->ksni_name, dev->name))
2031 if (ksi->ksni_index == -1) {
2032 if (dev->reg_state != NETREG_REGISTERED)
2034 /* A registration just happened: save the new index for
2036 ksi->ksni_index = ifindex;
2040 if (ksi->ksni_index != ifindex)
2043 if (dev->reg_state == NETREG_UNREGISTERING) {
2044 /* Device is being unregistered, we need to clear the
2045 * index, it can change when device will be back */
2046 ksi->ksni_index = -1;
2052 in_dev = __in_dev_get_rtnl(dev);
2054 CDEBUG(D_NET, "Interface %s has no IPv4 status.\n",
2056 ni_state_before = lnet_set_link_fatal_state(ni, 1);
2059 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
2060 if (sa->sin_addr.s_addr == ifa->ifa_local)
2066 CDEBUG(D_NET, "Interface %s has no matching ip\n",
2068 ni_state_before = lnet_set_link_fatal_state(ni, 1);
2073 ni_state_before = lnet_set_link_fatal_state(ni, 1);
2075 state = (lnet_get_link_status(dev) == 0);
2076 ni_state_before = lnet_set_link_fatal_state(ni,
2080 if (!update_ping_buf &&
2081 (ni->ni_state == LNET_NI_STATE_ACTIVE) &&
2082 (atomic_read(&ni->ni_fatal_error_on) != ni_state_before))
2083 update_ping_buf = true;
2086 if (update_ping_buf)
2087 lnet_mark_ping_buffer_for_update();
2094 ksocknal_handle_inetaddr_change(struct in_ifaddr *ifa, unsigned long event)
2096 struct lnet_ni *ni = NULL;
2097 struct ksock_net *net;
2098 struct ksock_net *cnxt;
2099 struct net_device *event_netdev = ifa->ifa_dev->dev;
2101 struct ksock_interface *ksi = NULL;
2102 struct sockaddr_in *sa;
2103 __u32 ni_state_before;
2104 bool update_ping_buf = false;
2107 if (!ksocknal_data.ksnd_nnets)
2110 ifindex = event_netdev->ifindex;
2112 list_for_each_entry_safe(net, cnxt, &ksocknal_data.ksnd_nets,
2115 ksi = &net->ksnn_interface;
2116 sa = (void *)&ksi->ksni_addr;
2118 if (ksi->ksni_index != ifindex ||
2119 strcmp(ksi->ksni_name, event_netdev->name))
2122 if (sa->sin_addr.s_addr == ifa->ifa_local) {
2124 link_down = (event == NETDEV_DOWN);
2125 ni_state_before = lnet_set_link_fatal_state(ni,
2128 if (!update_ping_buf &&
2129 (ni->ni_state == LNET_NI_STATE_ACTIVE) &&
2130 ((event == NETDEV_DOWN) != ni_state_before))
2131 update_ping_buf = true;
2135 if (update_ping_buf)
2136 lnet_mark_ping_buffer_for_update();
2141 /************************************
2142 * Net device notifier event handler
2143 ************************************/
2144 static int ksocknal_device_event(struct notifier_block *unused,
2145 unsigned long event, void *ptr)
2147 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2148 unsigned char operstate;
2150 operstate = dev->operstate;
2152 CDEBUG(D_NET, "devevent: status=%ld, iface=%s ifindex %d state %u\n",
2153 event, dev->name, dev->ifindex, operstate);
2159 case NETDEV_REGISTER:
2160 case NETDEV_UNREGISTER:
2161 ksocknal_handle_link_state_change(dev, operstate);
2168 /************************************
2169 * Inetaddr notifier event handler
2170 ************************************/
2171 static int ksocknal_inetaddr_event(struct notifier_block *unused,
2172 unsigned long event, void *ptr)
2174 struct in_ifaddr *ifa = ptr;
2176 CDEBUG(D_NET, "addrevent: status %ld ip addr %pI4, netmask %pI4.\n",
2177 event, &ifa->ifa_address, &ifa->ifa_mask);
2183 ksocknal_handle_inetaddr_change(ifa, event);
2190 static struct notifier_block ksocknal_dev_notifier_block = {
2191 .notifier_call = ksocknal_device_event,
2194 static struct notifier_block ksocknal_inetaddr_notifier_block = {
2195 .notifier_call = ksocknal_inetaddr_event,
2199 ksocknal_base_shutdown(void)
2201 struct ksock_sched *sched;
2202 struct ksock_peer_ni *peer_ni;
2205 CDEBUG(D_MALLOC, "before NAL cleanup: kmem %lld\n",
2206 libcfs_kmem_read());
2207 LASSERT (ksocknal_data.ksnd_nnets == 0);
2209 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL) {
2210 unregister_netdevice_notifier(&ksocknal_dev_notifier_block);
2211 unregister_inetaddr_notifier(&ksocknal_inetaddr_notifier_block);
2214 switch (ksocknal_data.ksnd_init) {
2219 case SOCKNAL_INIT_ALL:
2220 case SOCKNAL_INIT_DATA:
2221 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list)
2224 LASSERT(list_empty(&ksocknal_data.ksnd_nets));
2225 LASSERT(list_empty(&ksocknal_data.ksnd_enomem_conns));
2226 LASSERT(list_empty(&ksocknal_data.ksnd_zombie_conns));
2227 LASSERT(list_empty(&ksocknal_data.ksnd_connd_connreqs));
2228 LASSERT(list_empty(&ksocknal_data.ksnd_connd_routes));
2230 if (ksocknal_data.ksnd_schedulers != NULL) {
2231 cfs_percpt_for_each(sched, i,
2232 ksocknal_data.ksnd_schedulers) {
2234 LASSERT(list_empty(&sched->kss_tx_conns));
2235 LASSERT(list_empty(&sched->kss_rx_conns));
2236 LASSERT(list_empty(&sched->kss_zombie_noop_txs));
2237 LASSERT(sched->kss_nconns == 0);
2241 /* flag threads to terminate; wake and wait for them to die */
2242 ksocknal_data.ksnd_shuttingdown = 1;
2243 wake_up_all(&ksocknal_data.ksnd_connd_waitq);
2244 wake_up(&ksocknal_data.ksnd_reaper_waitq);
2246 if (ksocknal_data.ksnd_schedulers != NULL) {
2247 cfs_percpt_for_each(sched, i,
2248 ksocknal_data.ksnd_schedulers)
2249 wake_up_all(&sched->kss_waitq);
2252 wait_var_event_warning(&ksocknal_data.ksnd_nthreads,
2253 atomic_read(&ksocknal_data.ksnd_nthreads) == 0,
2254 "waiting for %d threads to terminate\n",
2255 atomic_read(&ksocknal_data.ksnd_nthreads));
2257 ksocknal_free_buffers();
2259 ksocknal_data.ksnd_init = SOCKNAL_INIT_NOTHING;
2263 CDEBUG(D_MALLOC, "after NAL cleanup: kmem %lld\n",
2264 libcfs_kmem_read());
2266 module_put(THIS_MODULE);
2270 ksocknal_base_startup(void)
2272 struct ksock_sched *sched;
2276 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
2277 LASSERT(ksocknal_data.ksnd_nnets == 0);
2279 memset(&ksocknal_data, 0, sizeof(ksocknal_data)); /* zero pointers */
2281 hash_init(ksocknal_data.ksnd_peers);
2283 rwlock_init(&ksocknal_data.ksnd_global_lock);
2284 INIT_LIST_HEAD(&ksocknal_data.ksnd_nets);
2286 spin_lock_init(&ksocknal_data.ksnd_reaper_lock);
2287 INIT_LIST_HEAD(&ksocknal_data.ksnd_enomem_conns);
2288 INIT_LIST_HEAD(&ksocknal_data.ksnd_zombie_conns);
2289 INIT_LIST_HEAD(&ksocknal_data.ksnd_deathrow_conns);
2290 init_waitqueue_head(&ksocknal_data.ksnd_reaper_waitq);
2292 spin_lock_init(&ksocknal_data.ksnd_connd_lock);
2293 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_connreqs);
2294 INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_routes);
2295 init_waitqueue_head(&ksocknal_data.ksnd_connd_waitq);
2297 spin_lock_init(&ksocknal_data.ksnd_tx_lock);
2298 INIT_LIST_HEAD(&ksocknal_data.ksnd_idle_noop_txs);
2300 /* NB memset above zeros whole of ksocknal_data */
2302 /* flag lists/ptrs/locks initialised */
2303 ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
2304 if (!try_module_get(THIS_MODULE))
2307 /* Create a scheduler block per available CPT */
2308 ksocknal_data.ksnd_schedulers = cfs_percpt_alloc(lnet_cpt_table(),
2310 if (ksocknal_data.ksnd_schedulers == NULL)
2313 cfs_percpt_for_each(sched, i, ksocknal_data.ksnd_schedulers) {
2317 * make sure not to allocate more threads than there are
2318 * cores/CPUs in teh CPT
2320 nthrs = cfs_cpt_weight(lnet_cpt_table(), i);
2321 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2322 nthrs = min(nthrs, *ksocknal_tunables.ksnd_nscheds);
2325 * max to half of CPUs, assume another half should be
2326 * reserved for upper layer modules
2328 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2331 sched->kss_nthreads_max = nthrs;
2334 spin_lock_init(&sched->kss_lock);
2335 INIT_LIST_HEAD(&sched->kss_rx_conns);
2336 INIT_LIST_HEAD(&sched->kss_tx_conns);
2337 INIT_LIST_HEAD(&sched->kss_zombie_noop_txs);
2338 init_waitqueue_head(&sched->kss_waitq);
2341 ksocknal_data.ksnd_connd_starting = 0;
2342 ksocknal_data.ksnd_connd_failed_stamp = 0;
2343 ksocknal_data.ksnd_connd_starting_stamp = ktime_get_real_seconds();
2344 /* must have at least 2 connds to remain responsive to accepts while
2346 if (*ksocknal_tunables.ksnd_nconnds < SOCKNAL_CONND_RESV + 1)
2347 *ksocknal_tunables.ksnd_nconnds = SOCKNAL_CONND_RESV + 1;
2349 if (*ksocknal_tunables.ksnd_nconnds_max <
2350 *ksocknal_tunables.ksnd_nconnds) {
2351 ksocknal_tunables.ksnd_nconnds_max =
2352 ksocknal_tunables.ksnd_nconnds;
2355 for (i = 0; i < *ksocknal_tunables.ksnd_nconnds; i++) {
2356 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2357 ksocknal_data.ksnd_connd_starting++;
2358 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2360 rc = ksocknal_thread_start(ksocknal_connd,
2361 (void *)((uintptr_t)i),
2362 "socknal_cd%02d", i);
2364 spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
2365 ksocknal_data.ksnd_connd_starting--;
2366 spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
2367 CERROR("Can't spawn socknal connd: %d\n", rc);
2372 rc = ksocknal_thread_start(ksocknal_reaper, NULL, "socknal_reaper");
2374 CERROR ("Can't spawn socknal reaper: %d\n", rc);
2378 register_netdevice_notifier(&ksocknal_dev_notifier_block);
2379 register_inetaddr_notifier(&ksocknal_inetaddr_notifier_block);
2381 /* flag everything initialised */
2382 ksocknal_data.ksnd_init = SOCKNAL_INIT_ALL;
2387 ksocknal_base_shutdown();
2392 ksocknal_debug_peerhash(struct lnet_ni *ni)
2394 struct ksock_peer_ni *peer_ni;
2397 read_lock(&ksocknal_data.ksnd_global_lock);
2399 hash_for_each(ksocknal_data.ksnd_peers, i, peer_ni, ksnp_list) {
2400 struct ksock_conn_cb *conn_cb;
2401 struct ksock_conn *conn;
2403 if (peer_ni->ksnp_ni != ni)
2406 CWARN("Active peer_ni on shutdown: %s, ref %d, closing %d, accepting %d, err %d, zcookie %llu, txq %d, zc_req %d\n",
2407 libcfs_idstr(&peer_ni->ksnp_id),
2408 refcount_read(&peer_ni->ksnp_refcount),
2409 peer_ni->ksnp_closing,
2410 peer_ni->ksnp_accepting, peer_ni->ksnp_error,
2411 peer_ni->ksnp_zc_next_cookie,
2412 !list_empty(&peer_ni->ksnp_tx_queue),
2413 !list_empty(&peer_ni->ksnp_zc_req_list));
2415 conn_cb = peer_ni->ksnp_conn_cb;
2417 CWARN("ConnCB: ref %d, schd %d, conn %d, cnted %d, del %d\n",
2418 refcount_read(&conn_cb->ksnr_refcount),
2419 conn_cb->ksnr_scheduled, conn_cb->ksnr_connecting,
2420 conn_cb->ksnr_connected, conn_cb->ksnr_deleted);
2423 list_for_each_entry(conn, &peer_ni->ksnp_conns, ksnc_list) {
2424 CWARN("Conn: ref %d, sref %d, t %d, c %d\n",
2425 refcount_read(&conn->ksnc_conn_refcount),
2426 refcount_read(&conn->ksnc_sock_refcount),
2427 conn->ksnc_type, conn->ksnc_closing);
2432 read_unlock(&ksocknal_data.ksnd_global_lock);
2437 ksocknal_shutdown(struct lnet_ni *ni)
2439 struct ksock_net *net = ni->ni_data;
2441 LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL);
2442 LASSERT(ksocknal_data.ksnd_nnets > 0);
2444 /* prevent new peers */
2445 atomic_add(SOCKNAL_SHUTDOWN_BIAS, &net->ksnn_npeers);
2447 /* Delete all peers */
2448 ksocknal_del_peer(ni, NULL);
2450 /* Wait for all peer_ni state to clean up */
2451 wait_var_event_warning(&net->ksnn_npeers,
2452 atomic_read(&net->ksnn_npeers) ==
2453 SOCKNAL_SHUTDOWN_BIAS,
2454 "waiting for %d peers to disconnect\n",
2455 ksocknal_debug_peerhash(ni) +
2456 atomic_read(&net->ksnn_npeers) -
2457 SOCKNAL_SHUTDOWN_BIAS);
2459 LASSERT(net->ksnn_interface.ksni_npeers == 0);
2460 LASSERT(net->ksnn_interface.ksni_nroutes == 0);
2462 list_del(&net->ksnn_list);
2463 LIBCFS_FREE(net, sizeof(*net));
2465 ksocknal_data.ksnd_nnets--;
2466 if (ksocknal_data.ksnd_nnets == 0)
2467 ksocknal_base_shutdown();
2471 ksocknal_search_new_ipif(struct ksock_net *net)
2474 char *ifnam = &net->ksnn_interface.ksni_name[0];
2475 char *colon = strchr(ifnam, ':');
2477 struct ksock_net *tmp;
2482 list_for_each_entry(tmp, &ksocknal_data.ksnd_nets, ksnn_list) {
2483 char *ifnam2 = &tmp->ksnn_interface.ksni_name[0];
2484 char *colon2 = strchr(ifnam2, ':');
2489 found = strcmp(ifnam, ifnam2) == 0;
2502 ksocknal_start_schedulers(struct ksock_sched *sched)
2508 if (sched->kss_nthreads == 0) {
2509 if (*ksocknal_tunables.ksnd_nscheds > 0) {
2510 nthrs = sched->kss_nthreads_max;
2512 nthrs = cfs_cpt_weight(lnet_cpt_table(),
2514 nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
2515 nthrs = min(SOCKNAL_NSCHEDS_HIGH, nthrs);
2517 nthrs = min(nthrs, sched->kss_nthreads_max);
2519 LASSERT(sched->kss_nthreads <= sched->kss_nthreads_max);
2520 /* increase two threads if there is new interface */
2521 nthrs = min(2, sched->kss_nthreads_max - sched->kss_nthreads);
2524 for (i = 0; i < nthrs; i++) {
2527 id = KSOCK_THREAD_ID(sched->kss_cpt, sched->kss_nthreads + i);
2528 rc = ksocknal_thread_start(ksocknal_scheduler, (void *)id,
2529 "socknal_sd%02d_%02d",
2531 (int)KSOCK_THREAD_SID(id));
2535 CERROR("Can't spawn thread %d for scheduler[%d]: %d\n",
2536 sched->kss_cpt, (int) KSOCK_THREAD_SID(id), rc);
2540 sched->kss_nthreads += i;
2545 ksocknal_net_start_threads(struct ksock_net *net, __u32 *cpts, int ncpts)
2547 int newif = ksocknal_search_new_ipif(net);
2551 if (ncpts > 0 && ncpts > cfs_cpt_number(lnet_cpt_table()))
2554 for (i = 0; i < ncpts; i++) {
2555 struct ksock_sched *sched;
2556 int cpt = (cpts == NULL) ? i : cpts[i];
2558 LASSERT(cpt < cfs_cpt_number(lnet_cpt_table()));
2559 sched = ksocknal_data.ksnd_schedulers[cpt];
2561 if (!newif && sched->kss_nthreads > 0)
2564 rc = ksocknal_start_schedulers(sched);
2572 ksocknal_startup(struct lnet_ni *ni)
2574 struct ksock_net *net;
2575 struct ksock_interface *ksi = NULL;
2576 struct lnet_inetdev *ifaces = NULL;
2580 LASSERT (ni->ni_net->net_lnd == &the_ksocklnd);
2581 if (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING) {
2582 rc = ksocknal_base_startup();
2586 LIBCFS_ALLOC(net, sizeof(*net));
2590 net->ksnn_incarnation = ktime_get_real_ns();
2593 ksocknal_tunables_setup(ni);
2595 rc = lnet_inet_enumerate(&ifaces, ni->ni_net_ns, true);
2599 ksi = &net->ksnn_interface;
2601 /* Interface and/or IP address is specified otherwise default to
2602 * the first Interface
2604 if_idx = lnet_inet_select(ni, ifaces, rc);
2608 if (!ni->ni_interface) {
2609 rc = lnet_ni_add_interface(ni, ifaces[if_idx].li_name);
2611 CWARN("ksocklnd failed to allocate ni_interface\n");
2614 ni->ni_dev_cpt = ifaces[if_idx].li_cpt;
2615 ksi->ksni_index = ifaces[if_idx].li_index;
2616 if (ifaces[if_idx].li_size == sizeof(struct in6_addr)) {
2617 struct sockaddr_in6 *sa;
2618 sa = (void *)&ksi->ksni_addr;
2619 memset(sa, 0, sizeof(*sa));
2620 sa->sin6_family = AF_INET6;
2621 memcpy(&sa->sin6_addr, ifaces[if_idx].li_ipv6addr,
2622 sizeof(struct in6_addr));
2623 ni->ni_nid.nid_size = sizeof(struct in6_addr) - 4;
2624 memcpy(&ni->ni_nid.nid_addr, ifaces[if_idx].li_ipv6addr,
2625 sizeof(struct in6_addr));
2627 struct sockaddr_in *sa;
2628 sa = (void *)&ksi->ksni_addr;
2629 memset(sa, 0, sizeof(*sa));
2630 sa->sin_family = AF_INET;
2631 sa->sin_addr.s_addr = ifaces[if_idx].li_ipaddr;
2632 ksi->ksni_netmask = ifaces[if_idx].li_netmask;
2633 ni->ni_nid.nid_size = 0;
2634 ni->ni_nid.nid_addr[0] = sa->sin_addr.s_addr;
2636 strlcpy(ksi->ksni_name, ifaces[if_idx].li_name, sizeof(ksi->ksni_name));
2638 /* call it before add it to ksocknal_data.ksnd_nets */
2639 rc = ksocknal_net_start_threads(net, ni->ni_cpts, ni->ni_ncpts);
2643 if ((ksocknal_ip2index((struct sockaddr *)&ksi->ksni_addr,
2645 &dev_status) < 0) ||
2647 lnet_set_link_fatal_state(ni, 1);
2649 list_add(&net->ksnn_list, &ksocknal_data.ksnd_nets);
2651 ksocknal_data.ksnd_nnets++;
2656 LIBCFS_FREE(net, sizeof(*net));
2658 if (ksocknal_data.ksnd_nnets == 0)
2659 ksocknal_base_shutdown();
2664 static void __exit ksocklnd_exit(void)
2666 lnet_unregister_lnd(&the_ksocklnd);
2669 static const struct lnet_lnd the_ksocklnd = {
2670 .lnd_type = SOCKLND,
2671 .lnd_startup = ksocknal_startup,
2672 .lnd_shutdown = ksocknal_shutdown,
2673 .lnd_ctl = ksocknal_ctl,
2674 .lnd_send = ksocknal_send,
2675 .lnd_recv = ksocknal_recv,
2676 .lnd_notify_peer_down = ksocknal_notify_gw_down,
2677 .lnd_accept = ksocknal_accept,
2678 .lnd_nl_get = ksocknal_nl_get,
2679 .lnd_nl_set = ksocknal_nl_set,
2680 .lnd_keys = &ksocknal_tunables_keys,
2683 static int __init ksocklnd_init(void)
2687 /* check ksnr_connected/connecting field large enough */
2688 BUILD_BUG_ON(SOCKLND_CONN_NTYPES > 4);
2689 BUILD_BUG_ON(SOCKLND_CONN_ACK != SOCKLND_CONN_BULK_IN);
2691 rc = ksocknal_tunables_init();
2695 rc = libcfs_setup();
2699 lnet_register_lnd(&the_ksocklnd);
2704 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2705 MODULE_DESCRIPTION("TCP Socket LNet Network Driver");
2706 MODULE_VERSION("2.8.0");
2707 MODULE_LICENSE("GPL");
2709 module_init(ksocklnd_init);
2710 module_exit(ksocklnd_exit);