Commit | Line | Data |
---|---|---|
d7e09d03 PT |
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 | |
9fdaf8c0 | 38 | #include "../../include/linux/lnet/lib-lnet.h" |
d7e09d03 PT |
39 | |
40 | ||
41 | static int accept_port = 988; | |
42 | static int accept_backlog = 127; | |
43 | static int accept_timeout = 5; | |
44 | ||
c49d4b96 | 45 | static struct { |
d7e09d03 | 46 | int pta_shutdown; |
e327dc88 | 47 | struct socket *pta_sock; |
d7e09d03 PT |
48 | struct completion pta_signal; |
49 | } lnet_acceptor_state; | |
50 | ||
51 | int | |
52 | lnet_acceptor_port(void) | |
53 | { | |
54 | return accept_port; | |
55 | } | |
735c2e26 | 56 | EXPORT_SYMBOL(lnet_acceptor_port); |
d7e09d03 PT |
57 | |
58 | static inline int | |
59 | lnet_accept_magic(__u32 magic, __u32 constant) | |
60 | { | |
61 | return (magic == constant || | |
62 | magic == __swab32(constant)); | |
63 | } | |
64 | ||
d7e09d03 PT |
65 | static char *accept = "secure"; |
66 | ||
8cc7b4b9 PT |
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)"); | |
d7e09d03 | 75 | |
9f100b8b | 76 | static char *accept_type; |
d7e09d03 | 77 | |
c49d4b96 | 78 | static int |
d7e09d03 PT |
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 | |
89 | lnet_acceptor_timeout(void) | |
90 | { | |
91 | return accept_timeout; | |
92 | } | |
93 | EXPORT_SYMBOL(lnet_acceptor_timeout); | |
94 | ||
95 | void | |
f93c0616 | 96 | lnet_connect_console_error(int rc, lnet_nid_t peer_nid, |
d7e09d03 PT |
97 | __u32 peer_ip, int peer_port) |
98 | { | |
99 | switch (rc) { | |
100 | /* "normal" errors */ | |
101 | case -ECONNREFUSED: | |
5129097b | 102 | CNETERR("Connection to %s at host %pI4h on port %d was refused: check that Lustre is running on that node.\n", |
d7e09d03 | 103 | libcfs_nid2str(peer_nid), |
5e8f6920 | 104 | &peer_ip, peer_port); |
d7e09d03 PT |
105 | break; |
106 | case -EHOSTUNREACH: | |
107 | case -ENETUNREACH: | |
5129097b | 108 | CNETERR("Connection to %s at host %pI4h was unreachable: the network or that node may be down, or Lustre may be misconfigured.\n", |
5e8f6920 | 109 | libcfs_nid2str(peer_nid), &peer_ip); |
d7e09d03 PT |
110 | break; |
111 | case -ETIMEDOUT: | |
5129097b | 112 | CNETERR("Connection to %s at host %pI4h on port %d took too long: that node may be hung or experiencing high load.\n", |
d7e09d03 | 113 | libcfs_nid2str(peer_nid), |
5e8f6920 | 114 | &peer_ip, peer_port); |
d7e09d03 PT |
115 | break; |
116 | case -ECONNRESET: | |
5129097b | 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", |
d7e09d03 | 118 | libcfs_nid2str(peer_nid), |
5e8f6920 | 119 | &peer_ip, peer_port, |
d7e09d03 PT |
120 | libcfs_nid2str(peer_nid)); |
121 | break; | |
122 | case -EPROTO: | |
5129097b | 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", |
d7e09d03 | 124 | libcfs_nid2str(peer_nid), |
5e8f6920 | 125 | &peer_ip, peer_port); |
d7e09d03 PT |
126 | break; |
127 | case -EADDRINUSE: | |
5129097b RK |
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), | |
5e8f6920 | 130 | &peer_ip, peer_port); |
d7e09d03 PT |
131 | break; |
132 | default: | |
5129097b RK |
133 | LCONSOLE_ERROR_MSG(0x11e, "Unexpected error %d connecting to %s at host %pI4h on port %d\n", |
134 | rc, libcfs_nid2str(peer_nid), | |
5e8f6920 | 135 | &peer_ip, peer_port); |
d7e09d03 PT |
136 | break; |
137 | } | |
138 | } | |
139 | EXPORT_SYMBOL(lnet_connect_console_error); | |
140 | ||
141 | int | |
e327dc88 | 142 | lnet_connect(struct socket **sockp, lnet_nid_t peer_nid, |
d7e09d03 PT |
143 | __u32 local_ip, __u32 peer_ip, int peer_port) |
144 | { | |
145 | lnet_acceptor_connreq_t cr; | |
e327dc88 | 146 | struct socket *sock; |
d7e09d03 PT |
147 | int rc; |
148 | int port; | |
149 | int fatal; | |
150 | ||
f93c0616 | 151 | CLASSERT(sizeof(cr) <= 16); /* not too big to be on the stack */ |
d7e09d03 PT |
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 = libcfs_sock_connect(&sock, &fatal, | |
159 | local_ip, port, | |
160 | peer_ip, peer_port); | |
161 | if (rc != 0) { | |
162 | if (fatal) | |
163 | goto failed; | |
164 | continue; | |
165 | } | |
166 | ||
f93c0616 | 167 | CLASSERT(LNET_PROTO_ACCEPTOR_VERSION == 1); |
d7e09d03 PT |
168 | |
169 | cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; | |
170 | cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; | |
171 | cr.acr_nid = peer_nid; | |
172 | ||
173 | if (the_lnet.ln_testprotocompat != 0) { | |
174 | /* single-shot proto check */ | |
175 | lnet_net_lock(LNET_LOCK_EX); | |
176 | if ((the_lnet.ln_testprotocompat & 4) != 0) { | |
177 | cr.acr_version++; | |
178 | the_lnet.ln_testprotocompat &= ~4; | |
179 | } | |
180 | if ((the_lnet.ln_testprotocompat & 8) != 0) { | |
181 | cr.acr_magic = LNET_PROTO_MAGIC; | |
182 | the_lnet.ln_testprotocompat &= ~8; | |
183 | } | |
184 | lnet_net_unlock(LNET_LOCK_EX); | |
185 | } | |
186 | ||
187 | rc = libcfs_sock_write(sock, &cr, sizeof(cr), | |
188 | accept_timeout); | |
189 | if (rc != 0) | |
190 | goto failed_sock; | |
191 | ||
192 | *sockp = sock; | |
193 | return 0; | |
194 | } | |
195 | ||
196 | rc = -EADDRINUSE; | |
197 | goto failed; | |
198 | ||
199 | failed_sock: | |
200 | libcfs_sock_release(sock); | |
201 | failed: | |
202 | lnet_connect_console_error(rc, peer_nid, peer_ip, peer_port); | |
203 | return rc; | |
204 | } | |
205 | EXPORT_SYMBOL(lnet_connect); | |
206 | ||
207 | ||
208 | /* Below is the code common for both kernel and MT user-space */ | |
209 | ||
c49d4b96 | 210 | static int |
e327dc88 | 211 | lnet_accept(struct socket *sock, __u32 magic) |
d7e09d03 PT |
212 | { |
213 | lnet_acceptor_connreq_t cr; | |
214 | __u32 peer_ip; | |
215 | int peer_port; | |
216 | int rc; | |
217 | int flip; | |
218 | lnet_ni_t *ni; | |
219 | char *str; | |
220 | ||
f93c0616 | 221 | LASSERT(sizeof(cr) <= 16); /* not too big for the stack */ |
d7e09d03 PT |
222 | |
223 | rc = libcfs_sock_getaddr(sock, 1, &peer_ip, &peer_port); | |
f93c0616 | 224 | LASSERT(rc == 0); /* we succeeded before */ |
d7e09d03 PT |
225 | |
226 | if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) { | |
227 | ||
228 | if (lnet_accept_magic(magic, LNET_PROTO_MAGIC)) { | |
229 | /* future version compatibility! | |
230 | * When LNET unifies protocols over all LNDs, the first | |
231 | * thing sent will be a version query. I send back | |
232 | * LNET_PROTO_ACCEPTOR_MAGIC to tell her I'm "old" */ | |
233 | ||
f93c0616 | 234 | memset(&cr, 0, sizeof(cr)); |
d7e09d03 PT |
235 | cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; |
236 | cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; | |
237 | rc = libcfs_sock_write(sock, &cr, sizeof(cr), | |
238 | accept_timeout); | |
239 | ||
240 | if (rc != 0) | |
5129097b | 241 | CERROR("Error sending magic+version in response to LNET magic from %pI4h: %d\n", |
5e8f6920 | 242 | &peer_ip, rc); |
d7e09d03 PT |
243 | return -EPROTO; |
244 | } | |
245 | ||
246 | if (magic == le32_to_cpu(LNET_PROTO_TCP_MAGIC)) | |
247 | str = "'old' socknal/tcpnal"; | |
248 | else if (lnet_accept_magic(magic, LNET_PROTO_RA_MAGIC)) | |
249 | str = "'old' ranal"; | |
250 | else | |
251 | str = "unrecognised"; | |
252 | ||
5129097b | 253 | LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %pI4h magic %08x: %s acceptor protocol\n", |
5e8f6920 | 254 | &peer_ip, magic, str); |
d7e09d03 PT |
255 | return -EPROTO; |
256 | } | |
257 | ||
258 | flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC); | |
259 | ||
260 | rc = libcfs_sock_read(sock, &cr.acr_version, | |
261 | sizeof(cr.acr_version), | |
262 | accept_timeout); | |
263 | if (rc != 0) { | |
5129097b RK |
264 | CERROR("Error %d reading connection request version from %pI4h\n", |
265 | rc, &peer_ip); | |
d7e09d03 PT |
266 | return -EIO; |
267 | } | |
268 | ||
269 | if (flip) | |
270 | __swab32s(&cr.acr_version); | |
271 | ||
272 | if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION) { | |
273 | /* future version compatibility! | |
274 | * An acceptor-specific protocol rev will first send a version | |
275 | * query. I send back my current version to tell her I'm | |
276 | * "old". */ | |
277 | int peer_version = cr.acr_version; | |
278 | ||
f93c0616 | 279 | memset(&cr, 0, sizeof(cr)); |
d7e09d03 PT |
280 | cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC; |
281 | cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION; | |
282 | ||
283 | rc = libcfs_sock_write(sock, &cr, sizeof(cr), | |
284 | accept_timeout); | |
285 | ||
286 | if (rc != 0) | |
5129097b | 287 | CERROR("Error sending magic+version in response to version %d from %pI4h: %d\n", |
5e8f6920 | 288 | peer_version, &peer_ip, rc); |
d7e09d03 PT |
289 | return -EPROTO; |
290 | } | |
291 | ||
292 | rc = libcfs_sock_read(sock, &cr.acr_nid, | |
293 | sizeof(cr) - | |
294 | offsetof(lnet_acceptor_connreq_t, acr_nid), | |
295 | accept_timeout); | |
296 | if (rc != 0) { | |
5129097b RK |
297 | CERROR("Error %d reading connection request from %pI4h\n", |
298 | rc, &peer_ip); | |
d7e09d03 PT |
299 | return -EIO; |
300 | } | |
301 | ||
302 | if (flip) | |
303 | __swab64s(&cr.acr_nid); | |
304 | ||
305 | ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid)); | |
306 | if (ni == NULL || /* no matching net */ | |
307 | ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */ | |
308 | if (ni != NULL) | |
309 | lnet_ni_decref(ni); | |
5129097b | 310 | LCONSOLE_ERROR_MSG(0x120, "Refusing connection from %pI4h for %s: No matching NI\n", |
5e8f6920 | 311 | &peer_ip, libcfs_nid2str(cr.acr_nid)); |
d7e09d03 PT |
312 | return -EPERM; |
313 | } | |
314 | ||
315 | if (ni->ni_lnd->lnd_accept == NULL) { | |
316 | /* This catches a request for the loopback LND */ | |
317 | lnet_ni_decref(ni); | |
5129097b | 318 | LCONSOLE_ERROR_MSG(0x121, "Refusing connection from %pI4h for %s: NI doesn not accept IP connections\n", |
5e8f6920 | 319 | &peer_ip, libcfs_nid2str(cr.acr_nid)); |
d7e09d03 PT |
320 | return -EPERM; |
321 | } | |
322 | ||
5e8f6920 PT |
323 | CDEBUG(D_NET, "Accept %s from %pI4h\n", |
324 | libcfs_nid2str(cr.acr_nid), &peer_ip); | |
d7e09d03 PT |
325 | |
326 | rc = ni->ni_lnd->lnd_accept(ni, sock); | |
327 | ||
328 | lnet_ni_decref(ni); | |
329 | return rc; | |
330 | } | |
331 | ||
c49d4b96 | 332 | static int |
d7e09d03 PT |
333 | lnet_acceptor(void *arg) |
334 | { | |
e327dc88 | 335 | struct socket *newsock; |
d7e09d03 PT |
336 | int rc; |
337 | __u32 magic; | |
338 | __u32 peer_ip; | |
339 | int peer_port; | |
340 | int secure = (int)((long_ptr_t)arg); | |
341 | ||
f93c0616 | 342 | LASSERT(lnet_acceptor_state.pta_sock == NULL); |
d7e09d03 PT |
343 | |
344 | cfs_block_allsigs(); | |
345 | ||
346 | rc = libcfs_sock_listen(&lnet_acceptor_state.pta_sock, | |
347 | 0, accept_port, accept_backlog); | |
348 | if (rc != 0) { | |
349 | if (rc == -EADDRINUSE) | |
5129097b | 350 | LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port %d: port already in use\n", |
d7e09d03 PT |
351 | accept_port); |
352 | else | |
5129097b | 353 | LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port %d: unexpected error %d\n", |
d7e09d03 PT |
354 | accept_port, rc); |
355 | ||
356 | lnet_acceptor_state.pta_sock = NULL; | |
357 | } else { | |
358 | LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port); | |
359 | } | |
360 | ||
361 | /* set init status and unblock parent */ | |
362 | lnet_acceptor_state.pta_shutdown = rc; | |
363 | complete(&lnet_acceptor_state.pta_signal); | |
364 | ||
365 | if (rc != 0) | |
366 | return rc; | |
367 | ||
368 | while (!lnet_acceptor_state.pta_shutdown) { | |
369 | ||
370 | rc = libcfs_sock_accept(&newsock, lnet_acceptor_state.pta_sock); | |
371 | if (rc != 0) { | |
372 | if (rc != -EAGAIN) { | |
373 | CWARN("Accept error %d: pausing...\n", rc); | |
d3caf4d5 PT |
374 | set_current_state(TASK_UNINTERRUPTIBLE); |
375 | schedule_timeout(cfs_time_seconds(1)); | |
d7e09d03 PT |
376 | } |
377 | continue; | |
378 | } | |
379 | ||
380 | /* maybe we're waken up with libcfs_sock_abort_accept() */ | |
381 | if (lnet_acceptor_state.pta_shutdown) { | |
382 | libcfs_sock_release(newsock); | |
383 | break; | |
384 | } | |
385 | ||
386 | rc = libcfs_sock_getaddr(newsock, 1, &peer_ip, &peer_port); | |
387 | if (rc != 0) { | |
388 | CERROR("Can't determine new connection's address\n"); | |
389 | goto failed; | |
390 | } | |
391 | ||
392 | if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) { | |
5129097b | 393 | CERROR("Refusing connection from %pI4h: insecure port %d\n", |
5e8f6920 | 394 | &peer_ip, peer_port); |
d7e09d03 PT |
395 | goto failed; |
396 | } | |
397 | ||
398 | rc = libcfs_sock_read(newsock, &magic, sizeof(magic), | |
399 | accept_timeout); | |
400 | if (rc != 0) { | |
5129097b RK |
401 | CERROR("Error %d reading connection request from %pI4h\n", |
402 | rc, &peer_ip); | |
d7e09d03 PT |
403 | goto failed; |
404 | } | |
405 | ||
406 | rc = lnet_accept(newsock, magic); | |
407 | if (rc != 0) | |
408 | goto failed; | |
409 | ||
410 | continue; | |
411 | ||
b5154614 | 412 | failed: |
d7e09d03 PT |
413 | libcfs_sock_release(newsock); |
414 | } | |
415 | ||
416 | libcfs_sock_release(lnet_acceptor_state.pta_sock); | |
417 | lnet_acceptor_state.pta_sock = NULL; | |
418 | ||
419 | CDEBUG(D_NET, "Acceptor stopping\n"); | |
420 | ||
421 | /* unblock lnet_acceptor_stop() */ | |
422 | complete(&lnet_acceptor_state.pta_signal); | |
423 | return 0; | |
424 | } | |
425 | ||
426 | static inline int | |
427 | accept2secure(const char *acc, long *sec) | |
428 | { | |
429 | if (!strcmp(acc, "secure")) { | |
430 | *sec = 1; | |
431 | return 1; | |
432 | } else if (!strcmp(acc, "all")) { | |
433 | *sec = 0; | |
434 | return 1; | |
435 | } else if (!strcmp(acc, "none")) { | |
436 | return 0; | |
d7e09d03 | 437 | } |
2c1d2535 TP |
438 | |
439 | LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n", | |
440 | acc); | |
441 | return -EINVAL; | |
d7e09d03 PT |
442 | } |
443 | ||
444 | int | |
445 | lnet_acceptor_start(void) | |
446 | { | |
447 | int rc; | |
448 | long rc2; | |
449 | long secure; | |
450 | ||
f93c0616 | 451 | LASSERT(lnet_acceptor_state.pta_sock == NULL); |
d7e09d03 PT |
452 | |
453 | rc = lnet_acceptor_get_tunables(); | |
454 | if (rc != 0) | |
455 | return rc; | |
456 | ||
457 | ||
458 | init_completion(&lnet_acceptor_state.pta_signal); | |
459 | rc = accept2secure(accept_type, &secure); | |
6de9710b | 460 | if (rc <= 0) |
d7e09d03 | 461 | return rc; |
d7e09d03 PT |
462 | |
463 | if (lnet_count_acceptor_nis() == 0) /* not required */ | |
464 | return 0; | |
465 | ||
466 | rc2 = PTR_ERR(kthread_run(lnet_acceptor, | |
467 | (void *)(ulong_ptr_t)secure, | |
468 | "acceptor_%03ld", secure)); | |
469 | if (IS_ERR_VALUE(rc2)) { | |
470 | CERROR("Can't start acceptor thread: %ld\n", rc2); | |
d7e09d03 PT |
471 | |
472 | return -ESRCH; | |
473 | } | |
474 | ||
475 | /* wait for acceptor to startup */ | |
476 | wait_for_completion(&lnet_acceptor_state.pta_signal); | |
477 | ||
478 | if (!lnet_acceptor_state.pta_shutdown) { | |
479 | /* started OK */ | |
480 | LASSERT(lnet_acceptor_state.pta_sock != NULL); | |
481 | return 0; | |
482 | } | |
483 | ||
484 | LASSERT(lnet_acceptor_state.pta_sock == NULL); | |
d7e09d03 PT |
485 | |
486 | return -ENETDOWN; | |
487 | } | |
488 | ||
489 | void | |
490 | lnet_acceptor_stop(void) | |
491 | { | |
492 | if (lnet_acceptor_state.pta_sock == NULL) /* not running */ | |
493 | return; | |
494 | ||
495 | lnet_acceptor_state.pta_shutdown = 1; | |
496 | libcfs_sock_abort_accept(lnet_acceptor_state.pta_sock); | |
497 | ||
498 | /* block until acceptor signals exit */ | |
499 | wait_for_completion(&lnet_acceptor_state.pta_signal); | |
d7e09d03 | 500 | } |