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, 2014, 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
38 #include <lnet/lib-lnet.h>
40 static int accept_port = 988;
41 static int accept_backlog = 127;
42 static int accept_timeout = 5;
46 cfs_socket_t *pta_sock;
47 struct completion pta_signal;
48 } lnet_acceptor_state = {
53 lnet_acceptor_port(void)
59 lnet_accept_magic(__u32 magic, __u32 constant)
61 return (magic == constant ||
62 magic == __swab32(constant));
65 EXPORT_SYMBOL(lnet_acceptor_port);
67 static char *accept = "secure";
69 CFS_MODULE_PARM(accept, "s", charp, 0444,
70 "Accept connections (secure|all|none)");
71 CFS_MODULE_PARM(accept_port, "i", int, 0444,
72 "Acceptor's port (same on all nodes)");
73 CFS_MODULE_PARM(accept_backlog, "i", int, 0444,
74 "Acceptor's listen backlog");
75 CFS_MODULE_PARM(accept_timeout, "i", int, 0644,
76 "Acceptor's timeout (seconds)");
78 static char *accept_type = NULL;
81 lnet_acceptor_get_tunables(void)
83 /* Userland acceptor uses 'accept_type' instead of 'accept', due to
84 * conflict with 'accept(2)', but kernel acceptor still uses 'accept'
85 * for compatibility. Hence the trick. */
91 lnet_acceptor_timeout(void)
93 return accept_timeout;
95 EXPORT_SYMBOL(lnet_acceptor_timeout);
98 lnet_connect_console_error (int rc, lnet_nid_t peer_nid,
99 __u32 peer_ip, int peer_port)
102 /* "normal" errors */
104 CNETERR("Connection to %s at host %u.%u.%u.%u on port %d was "
105 "refused: check that Lustre is running on that node.\n",
106 libcfs_nid2str(peer_nid),
107 HIPQUAD(peer_ip), peer_port);
111 CNETERR("Connection to %s at host %u.%u.%u.%u "
112 "was unreachable: the network or that node may "
113 "be down, or Lustre may be misconfigured.\n",
114 libcfs_nid2str(peer_nid), HIPQUAD(peer_ip));
117 CNETERR("Connection to %s at host %u.%u.%u.%u on "
118 "port %d took too long: that node may be hung "
119 "or experiencing high load.\n",
120 libcfs_nid2str(peer_nid),
121 HIPQUAD(peer_ip), peer_port);
124 LCONSOLE_ERROR_MSG(0x11b, "Connection to %s at host %u.%u.%u.%u"
125 " on port %d was reset: "
126 "is it running a compatible version of "
127 "Lustre and is %s one of its NIDs?\n",
128 libcfs_nid2str(peer_nid),
129 HIPQUAD(peer_ip), peer_port,
130 libcfs_nid2str(peer_nid));
133 LCONSOLE_ERROR_MSG(0x11c, "Protocol error connecting to %s at "
134 "host %u.%u.%u.%u on port %d: is it running "
135 "a compatible version of Lustre?\n",
136 libcfs_nid2str(peer_nid),
137 HIPQUAD(peer_ip), peer_port);
140 LCONSOLE_ERROR_MSG(0x11d, "No privileged ports available to "
141 "connect to %s at host %u.%u.%u.%u on port "
142 "%d\n", libcfs_nid2str(peer_nid),
143 HIPQUAD(peer_ip), peer_port);
146 LCONSOLE_ERROR_MSG(0x11e, "Unexpected error %d connecting to %s"
147 " at host %u.%u.%u.%u on port %d\n", rc,
148 libcfs_nid2str(peer_nid),
149 HIPQUAD(peer_ip), peer_port);
153 EXPORT_SYMBOL(lnet_connect_console_error);
156 lnet_connect(cfs_socket_t **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 = libcfs_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 = libcfs_sock_write(sock, &cr, sizeof(cr),
214 libcfs_sock_release(sock);
216 lnet_connect_console_error(rc, peer_nid, peer_ip, peer_port);
219 EXPORT_SYMBOL(lnet_connect);
222 lnet_accept(cfs_socket_t *sock, __u32 magic)
224 lnet_acceptor_connreq_t cr;
232 LASSERT (sizeof(cr) <= 16); /* not too big for the stack */
234 rc = libcfs_sock_getaddr(sock, 1, &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 = libcfs_sock_write(sock, &cr, sizeof(cr),
252 CERROR("Error sending magic+version in response"
253 "to LNET magic from %u.%u.%u.%u: %d\n",
254 HIPQUAD(peer_ip), rc);
258 if (magic == le32_to_cpu(LNET_PROTO_TCP_MAGIC))
259 str = "'old' socknal/tcpnal";
260 else if (lnet_accept_magic(magic, LNET_PROTO_RA_MAGIC))
263 str = "unrecognised";
265 LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %u.%u.%u.%u"
266 " magic %08x: %s acceptor protocol\n",
267 HIPQUAD(peer_ip), magic, str);
271 flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC);
273 rc = libcfs_sock_read(sock, &cr.acr_version,
274 sizeof(cr.acr_version),
277 CERROR("Error %d reading connection request version from "
278 "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
283 __swab32s(&cr.acr_version);
285 if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION) {
286 /* future version compatibility!
287 * An acceptor-specific protocol rev will first send a version
288 * query. I send back my current version to tell her I'm
290 int peer_version = cr.acr_version;
292 memset (&cr, 0, sizeof(cr));
293 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
294 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
296 rc = libcfs_sock_write(sock, &cr, sizeof(cr),
300 CERROR("Error sending magic+version in response"
301 "to version %d from %u.%u.%u.%u: %d\n",
302 peer_version, HIPQUAD(peer_ip), rc);
306 rc = libcfs_sock_read(sock, &cr.acr_nid,
308 offsetof(lnet_acceptor_connreq_t, acr_nid),
311 CERROR("Error %d reading connection request from "
312 "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
317 __swab64s(&cr.acr_nid);
319 ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid));
320 if (ni == NULL || /* no matching net */
321 ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */
324 LCONSOLE_ERROR_MSG(0x120, "Refusing connection from %u.%u.%u.%u"
325 " for %s: No matching NI\n",
326 HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
330 if (ni->ni_lnd->lnd_accept == NULL) {
331 /* This catches a request for the loopback LND */
333 LCONSOLE_ERROR_MSG(0x121, "Refusing connection from %u.%u.%u.%u"
334 " for %s: NI doesn not accept IP connections\n",
335 HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
339 CDEBUG(D_NET, "Accept %s from %u.%u.%u.%u\n",
340 libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip));
342 rc = ni->ni_lnd->lnd_accept(ni, sock);
349 lnet_acceptor(void *arg)
351 cfs_socket_t *newsock;
356 int secure = (int)((long_ptr_t)arg);
358 LASSERT (lnet_acceptor_state.pta_sock == NULL);
362 rc = libcfs_sock_listen(&lnet_acceptor_state.pta_sock,
363 0, accept_port, accept_backlog);
365 if (rc == -EADDRINUSE)
366 LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port"
367 " %d: port already in use\n",
370 LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port "
371 "%d: unexpected error %d\n",
374 lnet_acceptor_state.pta_sock = NULL;
376 LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port);
379 /* set init status and unblock parent */
380 lnet_acceptor_state.pta_shutdown = rc;
381 complete(&lnet_acceptor_state.pta_signal);
386 while (!lnet_acceptor_state.pta_shutdown) {
388 rc = libcfs_sock_accept(&newsock, lnet_acceptor_state.pta_sock);
391 CWARN("Accept error %d: pausing...\n", rc);
392 cfs_pause(cfs_time_seconds(1));
397 /* maybe we're waken up with libcfs_sock_abort_accept() */
398 if (lnet_acceptor_state.pta_shutdown) {
399 libcfs_sock_release(newsock);
403 rc = libcfs_sock_getaddr(newsock, 1, &peer_ip, &peer_port);
405 CERROR("Can't determine new connection's address\n");
409 if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
410 CERROR("Refusing connection from %u.%u.%u.%u: "
411 "insecure port %d\n",
412 HIPQUAD(peer_ip), peer_port);
416 rc = libcfs_sock_read(newsock, &magic, sizeof(magic),
419 CERROR("Error %d reading connection request from "
420 "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
424 rc = lnet_accept(newsock, magic);
431 libcfs_sock_release(newsock);
434 libcfs_sock_release(lnet_acceptor_state.pta_sock);
435 lnet_acceptor_state.pta_sock = NULL;
437 CDEBUG(D_NET, "Acceptor stopping\n");
439 /* unblock lnet_acceptor_stop() */
440 complete(&lnet_acceptor_state.pta_signal);
445 accept2secure(const char *acc, long *sec)
447 if (!strcmp(acc, "secure")) {
450 } else if (!strcmp(acc, "all")) {
453 } else if (!strcmp(acc, "none")) {
456 LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n",
463 lnet_acceptor_start(void)
465 struct task_struct *task;
470 /* if acceptor is already running return immediately */
471 if (!lnet_acceptor_state.pta_shutdown)
474 LASSERT (lnet_acceptor_state.pta_sock == NULL);
476 rc = lnet_acceptor_get_tunables();
480 init_completion(&lnet_acceptor_state.pta_signal);
481 rc = accept2secure(accept_type, &secure);
483 fini_completion(&lnet_acceptor_state.pta_signal);
487 if (lnet_count_acceptor_nis() == 0) /* not required */
490 task = kthread_run(lnet_acceptor, (void *)(ulong_ptr_t)secure,
491 "acceptor_%03ld", secure);
494 CERROR("Can't start acceptor thread: %ld\n", rc2);
495 fini_completion(&lnet_acceptor_state.pta_signal);
500 /* wait for acceptor to startup */
501 wait_for_completion(&lnet_acceptor_state.pta_signal);
503 if (!lnet_acceptor_state.pta_shutdown) {
505 LASSERT(lnet_acceptor_state.pta_sock != NULL);
509 LASSERT(lnet_acceptor_state.pta_sock == NULL);
510 fini_completion(&lnet_acceptor_state.pta_signal);
516 lnet_acceptor_stop(void)
518 if (lnet_acceptor_state.pta_shutdown) /* not running */
521 lnet_acceptor_state.pta_shutdown = 1;
522 libcfs_sock_abort_accept(lnet_acceptor_state.pta_sock);
524 /* block until acceptor signals exit */
525 wait_for_completion(&lnet_acceptor_state.pta_signal);
527 fini_completion(&lnet_acceptor_state.pta_signal);