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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LNET
39 #include <linux/completion.h>
41 #include <lnet/lib-lnet.h>
43 static int accept_port = 988;
44 static int accept_backlog = 127;
45 static int accept_timeout = 5;
49 struct socket *pta_sock;
50 struct completion pta_signal;
51 } lnet_acceptor_state = {
56 lnet_acceptor_port(void)
62 lnet_accept_magic(__u32 magic, __u32 constant)
64 return (magic == constant ||
65 magic == __swab32(constant));
68 EXPORT_SYMBOL(lnet_acceptor_port);
70 static char *accept = "secure";
72 module_param(accept, charp, 0444);
73 MODULE_PARM_DESC(accept, "Accept connections (secure|all|none)");
74 module_param(accept_port, int, 0444);
75 MODULE_PARM_DESC(accept_port, "Acceptor's port (same on all nodes)");
76 module_param(accept_backlog, int, 0444);
77 MODULE_PARM_DESC(accept_backlog, "Acceptor's listen backlog");
78 module_param(accept_timeout, int, 0644);
79 MODULE_PARM_DESC(accept_timeout, "Acceptor's timeout (seconds)");
81 static char *accept_type = NULL;
84 lnet_acceptor_get_tunables(void)
86 /* Userland acceptor uses 'accept_type' instead of 'accept', due to
87 * conflict with 'accept(2)', but kernel acceptor still uses 'accept'
88 * for compatibility. Hence the trick. */
94 lnet_acceptor_timeout(void)
96 return accept_timeout;
98 EXPORT_SYMBOL(lnet_acceptor_timeout);
101 lnet_connect_console_error (int rc, lnet_nid_t peer_nid,
102 __u32 peer_ip, int peer_port)
105 /* "normal" errors */
107 CNETERR("Connection to %s at host %pI4h on port %d was "
108 "refused: check that Lustre is running on that node.\n",
109 libcfs_nid2str(peer_nid), &peer_ip, peer_port);
113 CNETERR("Connection to %s at host %pI4h "
114 "was unreachable: the network or that node may "
115 "be down, or Lustre may be misconfigured.\n",
116 libcfs_nid2str(peer_nid), &peer_ip);
119 CNETERR("Connection to %s at host %pI4h on "
120 "port %d took too long: that node may be hung "
121 "or experiencing high load.\n",
122 libcfs_nid2str(peer_nid), &peer_ip, peer_port);
125 LCONSOLE_ERROR_MSG(0x11b, "Connection to %s at host %pI4h"
126 " on port %d was reset: "
127 "is it running a compatible version of "
128 "Lustre and is %s one of its NIDs?\n",
129 libcfs_nid2str(peer_nid), &peer_ip,
130 peer_port, libcfs_nid2str(peer_nid));
133 LCONSOLE_ERROR_MSG(0x11c, "Protocol error connecting to %s at "
134 "host %pI4h on port %d: is it running "
135 "a compatible version of Lustre?\n",
136 libcfs_nid2str(peer_nid), &peer_ip,
140 LCONSOLE_ERROR_MSG(0x11d, "No privileged ports available to "
141 "connect to %s at host %pI4h on port "
142 "%d\n", libcfs_nid2str(peer_nid),
143 &peer_ip, peer_port);
146 LCONSOLE_ERROR_MSG(0x11e, "Unexpected error %d connecting to %s"
147 " at host %pI4h on port %d\n", rc,
148 libcfs_nid2str(peer_nid),
149 &peer_ip, peer_port);
153 EXPORT_SYMBOL(lnet_connect_console_error);
156 lnet_connect(struct socket **sockp, lnet_nid_t peer_nid,
157 __u32 local_ip, __u32 peer_ip, int peer_port)
159 lnet_acceptor_connreq_t cr;
165 CLASSERT (sizeof(cr) <= 16); /* not too big to be on the stack */
167 for (port = LNET_ACCEPTOR_MAX_RESERVED_PORT;
168 port >= LNET_ACCEPTOR_MIN_RESERVED_PORT;
170 /* Iterate through reserved ports. */
172 rc = lnet_sock_connect(&sock, &fatal,
181 CLASSERT (LNET_PROTO_ACCEPTOR_VERSION == 1);
183 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
184 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
185 cr.acr_nid = peer_nid;
187 if (the_lnet.ln_testprotocompat != 0) {
188 /* single-shot proto check */
189 lnet_net_lock(LNET_LOCK_EX);
190 if ((the_lnet.ln_testprotocompat & 4) != 0) {
192 the_lnet.ln_testprotocompat &= ~4;
194 if ((the_lnet.ln_testprotocompat & 8) != 0) {
195 cr.acr_magic = LNET_PROTO_MAGIC;
196 the_lnet.ln_testprotocompat &= ~8;
198 lnet_net_unlock(LNET_LOCK_EX);
201 rc = lnet_sock_write(sock, &cr, sizeof(cr),
216 lnet_connect_console_error(rc, peer_nid, peer_ip, peer_port);
219 EXPORT_SYMBOL(lnet_connect);
222 lnet_accept(struct socket *sock, __u32 magic)
224 lnet_acceptor_connreq_t cr;
232 LASSERT (sizeof(cr) <= 16); /* not too big for the stack */
234 rc = lnet_sock_getaddr(sock, true, &peer_ip, &peer_port);
235 LASSERT (rc == 0); /* we succeeded before */
237 if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) {
239 if (lnet_accept_magic(magic, LNET_PROTO_MAGIC)) {
240 /* future version compatibility!
241 * When LNET unifies protocols over all LNDs, the first
242 * thing sent will be a version query. I send back
243 * LNET_PROTO_ACCEPTOR_MAGIC to tell her I'm "old" */
245 memset (&cr, 0, sizeof(cr));
246 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
247 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
248 rc = lnet_sock_write(sock, &cr, sizeof(cr),
252 CERROR("Error sending magic+version in response"
253 "to LNET magic from %pI4h: %d\n",
258 if (magic == le32_to_cpu(LNET_PROTO_TCP_MAGIC))
259 str = "'old' socknal/tcpnal";
261 str = "unrecognised";
263 LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %pI4h"
264 " magic %08x: %s acceptor protocol\n",
265 &peer_ip, magic, str);
269 flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC);
271 rc = lnet_sock_read(sock, &cr.acr_version,
272 sizeof(cr.acr_version),
275 CERROR("Error %d reading connection request version from "
276 "%pI4h\n", rc, &peer_ip);
281 __swab32s(&cr.acr_version);
283 if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION) {
284 /* future version compatibility!
285 * An acceptor-specific protocol rev will first send a version
286 * query. I send back my current version to tell her I'm
288 int peer_version = cr.acr_version;
290 memset (&cr, 0, sizeof(cr));
291 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
292 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
294 rc = lnet_sock_write(sock, &cr, sizeof(cr),
298 CERROR("Error sending magic+version in response"
299 "to version %d from %pI4h: %d\n",
300 peer_version, &peer_ip, rc);
304 rc = lnet_sock_read(sock, &cr.acr_nid,
306 offsetof(lnet_acceptor_connreq_t, acr_nid),
309 CERROR("Error %d reading connection request from "
310 "%pI4h\n", rc, &peer_ip);
315 __swab64s(&cr.acr_nid);
317 ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid));
318 if (ni == NULL || /* no matching net */
319 ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */
322 LCONSOLE_ERROR_MSG(0x120, "Refusing connection from %pI4h "
323 "for %s: No matching NI\n",
324 &peer_ip, libcfs_nid2str(cr.acr_nid));
328 if (ni->ni_lnd->lnd_accept == NULL) {
329 /* This catches a request for the loopback LND */
331 LCONSOLE_ERROR_MSG(0x121, "Refusing connection from %pI4h "
332 "for %s: NI doesn not accept IP connections\n",
333 &peer_ip, libcfs_nid2str(cr.acr_nid));
337 CDEBUG(D_NET, "Accept %s from %pI4h\n",
338 libcfs_nid2str(cr.acr_nid), &peer_ip);
340 rc = ni->ni_lnd->lnd_accept(ni, sock);
347 lnet_acceptor(void *arg)
349 struct socket *newsock;
354 int secure = (int)((uintptr_t)arg);
356 LASSERT (lnet_acceptor_state.pta_sock == NULL);
360 rc = lnet_sock_listen(&lnet_acceptor_state.pta_sock,
361 0, accept_port, accept_backlog);
363 if (rc == -EADDRINUSE)
364 LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port"
365 " %d: port already in use\n",
368 LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port "
369 "%d: unexpected error %d\n",
372 lnet_acceptor_state.pta_sock = NULL;
374 LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port);
377 /* set init status and unblock parent */
378 lnet_acceptor_state.pta_shutdown = rc;
379 complete(&lnet_acceptor_state.pta_signal);
384 while (!lnet_acceptor_state.pta_shutdown) {
386 rc = lnet_sock_accept(&newsock, lnet_acceptor_state.pta_sock);
389 CWARN("Accept error %d: pausing...\n", rc);
390 set_current_state(TASK_UNINTERRUPTIBLE);
391 schedule_timeout(cfs_time_seconds(1));
396 /* maybe we're waken up with lnet_sock_abort_accept() */
397 if (lnet_acceptor_state.pta_shutdown) {
398 sock_release(newsock);
402 rc = lnet_sock_getaddr(newsock, true, &peer_ip, &peer_port);
404 CERROR("Can't determine new connection's address\n");
408 if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
409 CERROR("Refusing connection from %pI4h: "
410 "insecure port %d\n", &peer_ip, peer_port);
414 rc = lnet_sock_read(newsock, &magic, sizeof(magic),
417 CERROR("Error %d reading connection request from "
418 "%pI4h\n", rc, &peer_ip);
422 rc = lnet_accept(newsock, magic);
429 sock_release(newsock);
432 sock_release(lnet_acceptor_state.pta_sock);
433 lnet_acceptor_state.pta_sock = NULL;
435 CDEBUG(D_NET, "Acceptor stopping\n");
437 /* unblock lnet_acceptor_stop() */
438 complete(&lnet_acceptor_state.pta_signal);
443 accept2secure(const char *acc, long *sec)
445 if (!strcmp(acc, "secure")) {
448 } else if (!strcmp(acc, "all")) {
451 } else if (!strcmp(acc, "none")) {
454 LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n",
461 lnet_acceptor_start(void)
463 struct task_struct *task;
468 /* if acceptor is already running return immediately */
469 if (!lnet_acceptor_state.pta_shutdown)
472 LASSERT (lnet_acceptor_state.pta_sock == NULL);
474 rc = lnet_acceptor_get_tunables();
478 init_completion(&lnet_acceptor_state.pta_signal);
479 rc = accept2secure(accept_type, &secure);
483 if (lnet_count_acceptor_nis() == 0) /* not required */
486 task = kthread_run(lnet_acceptor, (void *)(uintptr_t)secure,
487 "acceptor_%03ld", secure);
490 CERROR("Can't start acceptor thread: %ld\n", rc2);
495 /* wait for acceptor to startup */
496 wait_for_completion(&lnet_acceptor_state.pta_signal);
498 if (!lnet_acceptor_state.pta_shutdown) {
500 LASSERT(lnet_acceptor_state.pta_sock != NULL);
504 LASSERT(lnet_acceptor_state.pta_sock == NULL);
510 lnet_acceptor_stop(void)
514 if (lnet_acceptor_state.pta_shutdown) /* not running */
517 lnet_acceptor_state.pta_shutdown = 1;
519 sk = lnet_acceptor_state.pta_sock->sk;
521 /* awake any sleepers using safe method */
522 sk->sk_state_change(sk);
524 /* block until acceptor signals exit */
525 wait_for_completion(&lnet_acceptor_state.pta_signal);