1 /*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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
19 *
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
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37 #define DEBUG_SUBSYSTEM S_LNET
38 #include <linux/completion.h>
39 #include "../../include/linux/lnet/lib-lnet.h"
40
41 static int accept_port = 988;
42 static int accept_backlog = 127;
43 static int accept_timeout = 5;
44
45 static struct {
46 int pta_shutdown;
47 struct socket *pta_sock;
48 struct completion pta_signal;
49 } lnet_acceptor_state;
50
51 int
lnet_acceptor_port(void)52 lnet_acceptor_port(void)
53 {
54 return accept_port;
55 }
56 EXPORT_SYMBOL(lnet_acceptor_port);
57
58 static inline int
lnet_accept_magic(__u32 magic,__u32 constant)59 lnet_accept_magic(__u32 magic, __u32 constant)
60 {
61 return (magic == constant ||
62 magic == __swab32(constant));
63 }
64
65 static char *accept = "secure";
66
67 module_param(accept, charp, 0444);
68 MODULE_PARM_DESC(accept, "Accept connections (secure|all|none)");
69 module_param(accept_port, int, 0444);
70 MODULE_PARM_DESC(accept_port, "Acceptor's port (same on all nodes)");
71 module_param(accept_backlog, int, 0444);
72 MODULE_PARM_DESC(accept_backlog, "Acceptor's listen backlog");
73 module_param(accept_timeout, int, 0644);
74 MODULE_PARM_DESC(accept_timeout, "Acceptor's timeout (seconds)");
75
76 static char *accept_type;
77
78 static int
lnet_acceptor_get_tunables(void)79 lnet_acceptor_get_tunables(void)
80 {
81 /* Userland acceptor uses 'accept_type' instead of 'accept', due to
82 * conflict with 'accept(2)', but kernel acceptor still uses 'accept'
83 * for compatibility. Hence the trick. */
84 accept_type = accept;
85 return 0;
86 }
87
88 int
lnet_acceptor_timeout(void)89 lnet_acceptor_timeout(void)
90 {
91 return accept_timeout;
92 }
93 EXPORT_SYMBOL(lnet_acceptor_timeout);
94
95 void
lnet_connect_console_error(int rc,lnet_nid_t peer_nid,__u32 peer_ip,int peer_port)96 lnet_connect_console_error(int rc, lnet_nid_t peer_nid,
97 __u32 peer_ip, int peer_port)
98 {
99 switch (rc) {
100 /* "normal" errors */
101 case -ECONNREFUSED:
102 CNETERR("Connection to %s at host %pI4h on port %d was refused: check that Lustre is running on that node.\n",
103 libcfs_nid2str(peer_nid),
104 &peer_ip, peer_port);
105 break;
106 case -EHOSTUNREACH:
107 case -ENETUNREACH:
108 CNETERR("Connection to %s at host %pI4h was unreachable: the network or that node may be down, or Lustre may be misconfigured.\n",
109 libcfs_nid2str(peer_nid), &peer_ip);
110 break;
111 case -ETIMEDOUT:
112 CNETERR("Connection to %s at host %pI4h on port %d took too long: that node may be hung or experiencing high load.\n",
113 libcfs_nid2str(peer_nid),
114 &peer_ip, peer_port);
115 break;
116 case -ECONNRESET:
117 LCONSOLE_ERROR_MSG(0x11b, "Connection to %s at host %pI4h on port %d was reset: is it running a compatible version of Lustre and is %s one of its NIDs?\n",
118 libcfs_nid2str(peer_nid),
119 &peer_ip, peer_port,
120 libcfs_nid2str(peer_nid));
121 break;
122 case -EPROTO:
123 LCONSOLE_ERROR_MSG(0x11c, "Protocol error connecting to %s at host %pI4h on port %d: is it running a compatible version of Lustre?\n",
124 libcfs_nid2str(peer_nid),
125 &peer_ip, peer_port);
126 break;
127 case -EADDRINUSE:
128 LCONSOLE_ERROR_MSG(0x11d, "No privileged ports available to connect to %s at host %pI4h on port %d\n",
129 libcfs_nid2str(peer_nid),
130 &peer_ip, peer_port);
131 break;
132 default:
133 LCONSOLE_ERROR_MSG(0x11e, "Unexpected error %d connecting to %s at host %pI4h on port %d\n",
134 rc, libcfs_nid2str(peer_nid),
135 &peer_ip, peer_port);
136 break;
137 }
138 }
139 EXPORT_SYMBOL(lnet_connect_console_error);
140
141 int
lnet_connect(struct socket ** sockp,lnet_nid_t peer_nid,__u32 local_ip,__u32 peer_ip,int peer_port)142 lnet_connect(struct socket **sockp, lnet_nid_t peer_nid,
143 __u32 local_ip, __u32 peer_ip, int peer_port)
144 {
145 lnet_acceptor_connreq_t cr;
146 struct socket *sock;
147 int rc;
148 int port;
149 int fatal;
150
151 CLASSERT(sizeof(cr) <= 16); /* not too big to be on the stack */
152
153 for (port = LNET_ACCEPTOR_MAX_RESERVED_PORT;
154 port >= LNET_ACCEPTOR_MIN_RESERVED_PORT;
155 --port) {
156 /* Iterate through reserved ports. */
157
158 rc = lnet_sock_connect(&sock, &fatal, local_ip, port, peer_ip,
159 peer_port);
160 if (rc != 0) {
161 if (fatal)
162 goto failed;
163 continue;
164 }
165
166 CLASSERT(LNET_PROTO_ACCEPTOR_VERSION == 1);
167
168 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
169 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
170 cr.acr_nid = peer_nid;
171
172 if (the_lnet.ln_testprotocompat != 0) {
173 /* single-shot proto check */
174 lnet_net_lock(LNET_LOCK_EX);
175 if ((the_lnet.ln_testprotocompat & 4) != 0) {
176 cr.acr_version++;
177 the_lnet.ln_testprotocompat &= ~4;
178 }
179 if ((the_lnet.ln_testprotocompat & 8) != 0) {
180 cr.acr_magic = LNET_PROTO_MAGIC;
181 the_lnet.ln_testprotocompat &= ~8;
182 }
183 lnet_net_unlock(LNET_LOCK_EX);
184 }
185
186 rc = lnet_sock_write(sock, &cr, sizeof(cr), accept_timeout);
187 if (rc != 0)
188 goto failed_sock;
189
190 *sockp = sock;
191 return 0;
192 }
193
194 rc = -EADDRINUSE;
195 goto failed;
196
197 failed_sock:
198 sock_release(sock);
199 failed:
200 lnet_connect_console_error(rc, peer_nid, peer_ip, peer_port);
201 return rc;
202 }
203 EXPORT_SYMBOL(lnet_connect);
204
205 /* Below is the code common for both kernel and MT user-space */
206
207 static int
lnet_accept(struct socket * sock,__u32 magic)208 lnet_accept(struct socket *sock, __u32 magic)
209 {
210 lnet_acceptor_connreq_t cr;
211 __u32 peer_ip;
212 int peer_port;
213 int rc;
214 int flip;
215 lnet_ni_t *ni;
216 char *str;
217
218 LASSERT(sizeof(cr) <= 16); /* not too big for the stack */
219
220 rc = lnet_sock_getaddr(sock, 1, &peer_ip, &peer_port);
221 LASSERT(rc == 0); /* we succeeded before */
222
223 if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) {
224
225 if (lnet_accept_magic(magic, LNET_PROTO_MAGIC)) {
226 /* future version compatibility!
227 * When LNET unifies protocols over all LNDs, the first
228 * thing sent will be a version query. I send back
229 * LNET_PROTO_ACCEPTOR_MAGIC to tell her I'm "old" */
230
231 memset(&cr, 0, sizeof(cr));
232 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
233 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
234 rc = lnet_sock_write(sock, &cr, sizeof(cr),
235 accept_timeout);
236
237 if (rc != 0)
238 CERROR("Error sending magic+version in response to LNET magic from %pI4h: %d\n",
239 &peer_ip, rc);
240 return -EPROTO;
241 }
242
243 if (magic == le32_to_cpu(LNET_PROTO_TCP_MAGIC))
244 str = "'old' socknal/tcpnal";
245 else
246 str = "unrecognised";
247
248 LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %pI4h magic %08x: %s acceptor protocol\n",
249 &peer_ip, magic, str);
250 return -EPROTO;
251 }
252
253 flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC);
254
255 rc = lnet_sock_read(sock, &cr.acr_version, sizeof(cr.acr_version),
256 accept_timeout);
257 if (rc != 0) {
258 CERROR("Error %d reading connection request version from %pI4h\n",
259 rc, &peer_ip);
260 return -EIO;
261 }
262
263 if (flip)
264 __swab32s(&cr.acr_version);
265
266 if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION) {
267 /* future version compatibility!
268 * An acceptor-specific protocol rev will first send a version
269 * query. I send back my current version to tell her I'm
270 * "old". */
271 int peer_version = cr.acr_version;
272
273 memset(&cr, 0, sizeof(cr));
274 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
275 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
276
277 rc = lnet_sock_write(sock, &cr, sizeof(cr), accept_timeout);
278 if (rc != 0)
279 CERROR("Error sending magic+version in response to version %d from %pI4h: %d\n",
280 peer_version, &peer_ip, rc);
281 return -EPROTO;
282 }
283
284 rc = lnet_sock_read(sock, &cr.acr_nid,
285 sizeof(cr) -
286 offsetof(lnet_acceptor_connreq_t, acr_nid),
287 accept_timeout);
288 if (rc != 0) {
289 CERROR("Error %d reading connection request from %pI4h\n",
290 rc, &peer_ip);
291 return -EIO;
292 }
293
294 if (flip)
295 __swab64s(&cr.acr_nid);
296
297 ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid));
298 if (ni == NULL || /* no matching net */
299 ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */
300 if (ni != NULL)
301 lnet_ni_decref(ni);
302 LCONSOLE_ERROR_MSG(0x120, "Refusing connection from %pI4h for %s: No matching NI\n",
303 &peer_ip, libcfs_nid2str(cr.acr_nid));
304 return -EPERM;
305 }
306
307 if (ni->ni_lnd->lnd_accept == NULL) {
308 /* This catches a request for the loopback LND */
309 lnet_ni_decref(ni);
310 LCONSOLE_ERROR_MSG(0x121, "Refusing connection from %pI4h for %s: NI doesn not accept IP connections\n",
311 &peer_ip, libcfs_nid2str(cr.acr_nid));
312 return -EPERM;
313 }
314
315 CDEBUG(D_NET, "Accept %s from %pI4h\n",
316 libcfs_nid2str(cr.acr_nid), &peer_ip);
317
318 rc = ni->ni_lnd->lnd_accept(ni, sock);
319
320 lnet_ni_decref(ni);
321 return rc;
322 }
323
324 static int
lnet_acceptor(void * arg)325 lnet_acceptor(void *arg)
326 {
327 struct socket *newsock;
328 int rc;
329 __u32 magic;
330 __u32 peer_ip;
331 int peer_port;
332 int secure = (int)((long_ptr_t)arg);
333
334 LASSERT(lnet_acceptor_state.pta_sock == NULL);
335
336 cfs_block_allsigs();
337
338 rc = lnet_sock_listen(&lnet_acceptor_state.pta_sock, 0, accept_port,
339 accept_backlog);
340 if (rc != 0) {
341 if (rc == -EADDRINUSE)
342 LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port %d: port already in use\n",
343 accept_port);
344 else
345 LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port %d: unexpected error %d\n",
346 accept_port, rc);
347
348 lnet_acceptor_state.pta_sock = NULL;
349 } else {
350 LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port);
351 }
352
353 /* set init status and unblock parent */
354 lnet_acceptor_state.pta_shutdown = rc;
355 complete(&lnet_acceptor_state.pta_signal);
356
357 if (rc != 0)
358 return rc;
359
360 while (!lnet_acceptor_state.pta_shutdown) {
361
362 rc = lnet_sock_accept(&newsock, lnet_acceptor_state.pta_sock);
363 if (rc != 0) {
364 if (rc != -EAGAIN) {
365 CWARN("Accept error %d: pausing...\n", rc);
366 set_current_state(TASK_UNINTERRUPTIBLE);
367 schedule_timeout(cfs_time_seconds(1));
368 }
369 continue;
370 }
371
372 /* maybe the LNet acceptor thread has been waken */
373 if (lnet_acceptor_state.pta_shutdown) {
374 sock_release(newsock);
375 break;
376 }
377
378 rc = lnet_sock_getaddr(newsock, 1, &peer_ip, &peer_port);
379 if (rc != 0) {
380 CERROR("Can't determine new connection's address\n");
381 goto failed;
382 }
383
384 if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
385 CERROR("Refusing connection from %pI4h: insecure port %d\n",
386 &peer_ip, peer_port);
387 goto failed;
388 }
389
390 rc = lnet_sock_read(newsock, &magic, sizeof(magic),
391 accept_timeout);
392 if (rc != 0) {
393 CERROR("Error %d reading connection request from %pI4h\n",
394 rc, &peer_ip);
395 goto failed;
396 }
397
398 rc = lnet_accept(newsock, magic);
399 if (rc != 0)
400 goto failed;
401
402 continue;
403
404 failed:
405 sock_release(newsock);
406 }
407
408 sock_release(lnet_acceptor_state.pta_sock);
409 lnet_acceptor_state.pta_sock = NULL;
410
411 CDEBUG(D_NET, "Acceptor stopping\n");
412
413 /* unblock lnet_acceptor_stop() */
414 complete(&lnet_acceptor_state.pta_signal);
415 return 0;
416 }
417
418 static inline int
accept2secure(const char * acc,long * sec)419 accept2secure(const char *acc, long *sec)
420 {
421 if (!strcmp(acc, "secure")) {
422 *sec = 1;
423 return 1;
424 } else if (!strcmp(acc, "all")) {
425 *sec = 0;
426 return 1;
427 } else if (!strcmp(acc, "none")) {
428 return 0;
429 }
430
431 LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n",
432 acc);
433 return -EINVAL;
434 }
435
436 int
lnet_acceptor_start(void)437 lnet_acceptor_start(void)
438 {
439 int rc;
440 long rc2;
441 long secure;
442
443 LASSERT(lnet_acceptor_state.pta_sock == NULL);
444
445 rc = lnet_acceptor_get_tunables();
446 if (rc != 0)
447 return rc;
448
449 init_completion(&lnet_acceptor_state.pta_signal);
450 rc = accept2secure(accept_type, &secure);
451 if (rc <= 0)
452 return rc;
453
454 if (lnet_count_acceptor_nis() == 0) /* not required */
455 return 0;
456
457 rc2 = PTR_ERR(kthread_run(lnet_acceptor,
458 (void *)(ulong_ptr_t)secure,
459 "acceptor_%03ld", secure));
460 if (IS_ERR_VALUE(rc2)) {
461 CERROR("Can't start acceptor thread: %ld\n", rc2);
462
463 return -ESRCH;
464 }
465
466 /* wait for acceptor to startup */
467 wait_for_completion(&lnet_acceptor_state.pta_signal);
468
469 if (!lnet_acceptor_state.pta_shutdown) {
470 /* started OK */
471 LASSERT(lnet_acceptor_state.pta_sock != NULL);
472 return 0;
473 }
474
475 LASSERT(lnet_acceptor_state.pta_sock == NULL);
476
477 return -ENETDOWN;
478 }
479
480 void
lnet_acceptor_stop(void)481 lnet_acceptor_stop(void)
482 {
483 if (lnet_acceptor_state.pta_sock == NULL) /* not running */
484 return;
485
486 lnet_acceptor_state.pta_shutdown = 1;
487 wake_up_all(sk_sleep(lnet_acceptor_state.pta_sock->sk));
488
489 /* block until acceptor signals exit */
490 wait_for_completion(&lnet_acceptor_state.pta_signal);
491 }
492