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 <linux/completion.h>
39 #include <lnet/lib-lnet.h>
41 static int accept_port = 988;
42 static int accept_backlog = 127;
43 static int accept_timeout = 5;
47 struct socket *pta_sock;
48 struct completion pta_signal;
49 } lnet_acceptor_state = {
54 lnet_acceptor_port(void)
60 lnet_accept_magic(__u32 magic, __u32 constant)
62 return (magic == constant ||
63 magic == __swab32(constant));
66 EXPORT_SYMBOL(lnet_acceptor_port);
68 static char *accept = "secure";
70 CFS_MODULE_PARM(accept, "s", charp, 0444,
71 "Accept connections (secure|all|none)");
72 CFS_MODULE_PARM(accept_port, "i", int, 0444,
73 "Acceptor's port (same on all nodes)");
74 CFS_MODULE_PARM(accept_backlog, "i", int, 0444,
75 "Acceptor's listen backlog");
76 CFS_MODULE_PARM(accept_timeout, "i", int, 0644,
77 "Acceptor's timeout (seconds)");
79 static char *accept_type = NULL;
82 lnet_acceptor_get_tunables(void)
84 /* Userland acceptor uses 'accept_type' instead of 'accept', due to
85 * conflict with 'accept(2)', but kernel acceptor still uses 'accept'
86 * for compatibility. Hence the trick. */
92 lnet_acceptor_timeout(void)
94 return accept_timeout;
96 EXPORT_SYMBOL(lnet_acceptor_timeout);
99 lnet_connect_console_error (int rc, lnet_nid_t peer_nid,
100 __u32 peer_ip, int peer_port)
103 /* "normal" errors */
105 CNETERR("Connection to %s at host %pI4h on port %d was "
106 "refused: check that Lustre is running on that node.\n",
107 libcfs_nid2str(peer_nid), &peer_ip, peer_port);
111 CNETERR("Connection to %s at host %pI4h "
112 "was unreachable: the network or that node may "
113 "be down, or Lustre may be misconfigured.\n",
114 libcfs_nid2str(peer_nid), &peer_ip);
117 CNETERR("Connection to %s at host %pI4h on "
118 "port %d took too long: that node may be hung "
119 "or experiencing high load.\n",
120 libcfs_nid2str(peer_nid), &peer_ip, peer_port);
123 LCONSOLE_ERROR_MSG(0x11b, "Connection to %s at host %pI4h"
124 " on port %d was reset: "
125 "is it running a compatible version of "
126 "Lustre and is %s one of its NIDs?\n",
127 libcfs_nid2str(peer_nid), &peer_ip,
128 peer_port, libcfs_nid2str(peer_nid));
131 LCONSOLE_ERROR_MSG(0x11c, "Protocol error connecting to %s at "
132 "host %pI4h on port %d: is it running "
133 "a compatible version of Lustre?\n",
134 libcfs_nid2str(peer_nid), &peer_ip,
138 LCONSOLE_ERROR_MSG(0x11d, "No privileged ports available to "
139 "connect to %s at host %pI4h on port "
140 "%d\n", libcfs_nid2str(peer_nid),
141 &peer_ip, peer_port);
144 LCONSOLE_ERROR_MSG(0x11e, "Unexpected error %d connecting to %s"
145 " at host %pI4h on port %d\n", rc,
146 libcfs_nid2str(peer_nid),
147 &peer_ip, peer_port);
151 EXPORT_SYMBOL(lnet_connect_console_error);
154 lnet_connect(struct socket **sockp, lnet_nid_t peer_nid,
155 __u32 local_ip, __u32 peer_ip, int peer_port)
157 lnet_acceptor_connreq_t cr;
163 CLASSERT (sizeof(cr) <= 16); /* not too big to be on the stack */
165 for (port = LNET_ACCEPTOR_MAX_RESERVED_PORT;
166 port >= LNET_ACCEPTOR_MIN_RESERVED_PORT;
168 /* Iterate through reserved ports. */
170 rc = lnet_sock_connect(&sock, &fatal,
179 CLASSERT (LNET_PROTO_ACCEPTOR_VERSION == 1);
181 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
182 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
183 cr.acr_nid = peer_nid;
185 if (the_lnet.ln_testprotocompat != 0) {
186 /* single-shot proto check */
187 lnet_net_lock(LNET_LOCK_EX);
188 if ((the_lnet.ln_testprotocompat & 4) != 0) {
190 the_lnet.ln_testprotocompat &= ~4;
192 if ((the_lnet.ln_testprotocompat & 8) != 0) {
193 cr.acr_magic = LNET_PROTO_MAGIC;
194 the_lnet.ln_testprotocompat &= ~8;
196 lnet_net_unlock(LNET_LOCK_EX);
199 rc = lnet_sock_write(sock, &cr, sizeof(cr),
214 lnet_connect_console_error(rc, peer_nid, peer_ip, peer_port);
217 EXPORT_SYMBOL(lnet_connect);
220 lnet_accept(struct socket *sock, __u32 magic)
222 lnet_acceptor_connreq_t cr;
230 LASSERT (sizeof(cr) <= 16); /* not too big for the stack */
232 rc = lnet_sock_getaddr(sock, true, &peer_ip, &peer_port);
233 LASSERT (rc == 0); /* we succeeded before */
235 if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) {
237 if (lnet_accept_magic(magic, LNET_PROTO_MAGIC)) {
238 /* future version compatibility!
239 * When LNET unifies protocols over all LNDs, the first
240 * thing sent will be a version query. I send back
241 * LNET_PROTO_ACCEPTOR_MAGIC to tell her I'm "old" */
243 memset (&cr, 0, sizeof(cr));
244 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
245 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
246 rc = lnet_sock_write(sock, &cr, sizeof(cr),
250 CERROR("Error sending magic+version in response"
251 "to LNET magic from %pI4h: %d\n",
256 if (magic == le32_to_cpu(LNET_PROTO_TCP_MAGIC))
257 str = "'old' socknal/tcpnal";
259 str = "unrecognised";
261 LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %pI4h"
262 " magic %08x: %s acceptor protocol\n",
263 &peer_ip, magic, str);
267 flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC);
269 rc = lnet_sock_read(sock, &cr.acr_version,
270 sizeof(cr.acr_version),
273 CERROR("Error %d reading connection request version from "
274 "%pI4h\n", rc, &peer_ip);
279 __swab32s(&cr.acr_version);
281 if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION) {
282 /* future version compatibility!
283 * An acceptor-specific protocol rev will first send a version
284 * query. I send back my current version to tell her I'm
286 int peer_version = cr.acr_version;
288 memset (&cr, 0, sizeof(cr));
289 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
290 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
292 rc = lnet_sock_write(sock, &cr, sizeof(cr),
296 CERROR("Error sending magic+version in response"
297 "to version %d from %pI4h: %d\n",
298 peer_version, &peer_ip, rc);
302 rc = lnet_sock_read(sock, &cr.acr_nid,
304 offsetof(lnet_acceptor_connreq_t, acr_nid),
307 CERROR("Error %d reading connection request from "
308 "%pI4h\n", rc, &peer_ip);
313 __swab64s(&cr.acr_nid);
315 ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid));
316 if (ni == NULL || /* no matching net */
317 ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */
320 LCONSOLE_ERROR_MSG(0x120, "Refusing connection from %pI4h "
321 "for %s: No matching NI\n",
322 &peer_ip, libcfs_nid2str(cr.acr_nid));
326 if (ni->ni_lnd->lnd_accept == NULL) {
327 /* This catches a request for the loopback LND */
329 LCONSOLE_ERROR_MSG(0x121, "Refusing connection from %pI4h "
330 "for %s: NI doesn not accept IP connections\n",
331 &peer_ip, libcfs_nid2str(cr.acr_nid));
335 CDEBUG(D_NET, "Accept %s from %pI4h\n",
336 libcfs_nid2str(cr.acr_nid), &peer_ip);
338 rc = ni->ni_lnd->lnd_accept(ni, sock);
345 lnet_acceptor(void *arg)
347 struct socket *newsock;
352 int secure = (int)((long_ptr_t)arg);
354 LASSERT (lnet_acceptor_state.pta_sock == NULL);
358 rc = lnet_sock_listen(&lnet_acceptor_state.pta_sock,
359 0, accept_port, accept_backlog);
361 if (rc == -EADDRINUSE)
362 LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port"
363 " %d: port already in use\n",
366 LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port "
367 "%d: unexpected error %d\n",
370 lnet_acceptor_state.pta_sock = NULL;
372 LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port);
375 /* set init status and unblock parent */
376 lnet_acceptor_state.pta_shutdown = rc;
377 complete(&lnet_acceptor_state.pta_signal);
382 while (!lnet_acceptor_state.pta_shutdown) {
384 rc = lnet_sock_accept(&newsock, lnet_acceptor_state.pta_sock);
387 CWARN("Accept error %d: pausing...\n", rc);
388 set_current_state(TASK_UNINTERRUPTIBLE);
389 schedule_timeout(cfs_time_seconds(1));
394 /* maybe we're waken up with lnet_sock_abort_accept() */
395 if (lnet_acceptor_state.pta_shutdown) {
396 sock_release(newsock);
400 rc = lnet_sock_getaddr(newsock, true, &peer_ip, &peer_port);
402 CERROR("Can't determine new connection's address\n");
406 if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
407 CERROR("Refusing connection from %pI4h: "
408 "insecure port %d\n", &peer_ip, peer_port);
412 rc = lnet_sock_read(newsock, &magic, sizeof(magic),
415 CERROR("Error %d reading connection request from "
416 "%pI4h\n", rc, &peer_ip);
420 rc = lnet_accept(newsock, magic);
427 sock_release(newsock);
430 sock_release(lnet_acceptor_state.pta_sock);
431 lnet_acceptor_state.pta_sock = NULL;
433 CDEBUG(D_NET, "Acceptor stopping\n");
435 /* unblock lnet_acceptor_stop() */
436 complete(&lnet_acceptor_state.pta_signal);
441 accept2secure(const char *acc, long *sec)
443 if (!strcmp(acc, "secure")) {
446 } else if (!strcmp(acc, "all")) {
449 } else if (!strcmp(acc, "none")) {
452 LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n",
459 lnet_acceptor_start(void)
461 struct task_struct *task;
466 /* if acceptor is already running return immediately */
467 if (!lnet_acceptor_state.pta_shutdown)
470 LASSERT (lnet_acceptor_state.pta_sock == NULL);
472 rc = lnet_acceptor_get_tunables();
476 init_completion(&lnet_acceptor_state.pta_signal);
477 rc = accept2secure(accept_type, &secure);
481 if (lnet_count_acceptor_nis() == 0) /* not required */
484 task = kthread_run(lnet_acceptor, (void *)(ulong_ptr_t)secure,
485 "acceptor_%03ld", secure);
488 CERROR("Can't start acceptor thread: %ld\n", rc2);
493 /* wait for acceptor to startup */
494 wait_for_completion(&lnet_acceptor_state.pta_signal);
496 if (!lnet_acceptor_state.pta_shutdown) {
498 LASSERT(lnet_acceptor_state.pta_sock != NULL);
502 LASSERT(lnet_acceptor_state.pta_sock == NULL);
508 lnet_acceptor_stop(void)
512 if (lnet_acceptor_state.pta_shutdown) /* not running */
515 lnet_acceptor_state.pta_shutdown = 1;
517 sk = lnet_acceptor_state.pta_sock->sk;
519 /* awake any sleepers using safe method */
520 sk->sk_state_change(sk);
522 /* block until acceptor signals exit */
523 wait_for_completion(&lnet_acceptor_state.pta_signal);