Whamcloud - gitweb
Branch HEAD
[fs/lustre-release.git] / lnet / lnet / acceptor.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * Copyright (C) 2005 Cluster File Systems, Inc.
5  *
6  *   This file is part of Lustre, http://www.lustre.org.
7  *
8  *   Lustre is free software; you can redistribute it and/or
9  *   modify it under the terms of version 2 of the GNU General Public
10  *   License as published by the Free Software Foundation.
11  *
12  *   Lustre is distributed in the hope that it will be useful,
13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *   GNU General Public License for more details.
16  *
17  *   You should have received a copy of the GNU General Public License
18  *   along with Lustre; if not, write to the Free Software
19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20  *
21  */
22
23 #define DEBUG_SUBSYSTEM S_LNET
24 #include <lnet/lib-lnet.h>
25
26 #ifdef __KERNEL__
27 static char *accept = "secure";
28 CFS_MODULE_PARM(accept, "s", charp, 0444,
29                 "Accept connections (secure|all|none)");
30
31 static int accept_port = 988;
32 CFS_MODULE_PARM(accept_port, "i", int, 0444,
33                 "Acceptor's port (same on all nodes)");
34
35 static int accept_backlog = 127;
36 CFS_MODULE_PARM(accept_backlog, "i", int, 0444,
37                 "Acceptor's listen backlog");
38
39 static int accept_timeout = 5;
40 CFS_MODULE_PARM(accept_timeout, "i", int, 0644,
41                 "Acceptor's timeout (seconds)");
42
43 struct {
44         int               pta_shutdown;
45         cfs_socket_t     *pta_sock;
46         struct semaphore  pta_signal;
47 } lnet_acceptor_state;
48
49 int
50 lnet_acceptor_timeout(void)
51 {
52         return accept_timeout;
53 }
54 EXPORT_SYMBOL(lnet_acceptor_timeout);
55
56 int
57 lnet_acceptor_port(void)
58 {
59         return accept_port;
60 }
61 EXPORT_SYMBOL(lnet_acceptor_port);
62
63 void
64 lnet_connect_console_error (int rc, lnet_nid_t peer_nid, 
65                            __u32 peer_ip, int peer_port)
66 {
67         switch (rc) {
68         /* "normal" errors */
69         case -ECONNREFUSED:
70                 CDEBUG(D_NETERROR, "Connection to %s at host %u.%u.%u.%u "
71                        "on port %d was refused: "
72                        "check that Lustre is running on that node.\n",
73                        libcfs_nid2str(peer_nid),
74                        HIPQUAD(peer_ip), peer_port);
75                 break;
76         case -EHOSTUNREACH:
77         case -ENETUNREACH:
78                 CDEBUG(D_NETERROR, "Connection to %s at host %u.%u.%u.%u "
79                        "was unreachable: the network or that node may "
80                        "be down, or Lustre may be misconfigured.\n",
81                        libcfs_nid2str(peer_nid), HIPQUAD(peer_ip));
82                 break;
83         case -ETIMEDOUT:
84                 CDEBUG(D_NETERROR, "Connection to %s at host %u.%u.%u.%u on "
85                        "port %d took too long: that node may be hung "
86                        "or experiencing high load.\n",
87                        libcfs_nid2str(peer_nid),
88                        HIPQUAD(peer_ip), peer_port);
89                 break;
90         case -ECONNRESET:
91                 LCONSOLE_ERROR_MSG(0x11b, "Connection to %s at host %u.%u.%u.%u"
92                                    " on port %d was reset: "
93                                    "is it running a compatible version of "
94                                    "Lustre and is %s one of its NIDs?\n",
95                                    libcfs_nid2str(peer_nid),
96                                    HIPQUAD(peer_ip), peer_port,
97                                    libcfs_nid2str(peer_nid));
98                 break;
99         case -EPROTO:
100                 LCONSOLE_ERROR_MSG(0x11c, "Protocol error connecting to %s at "
101                                    "host %u.%u.%u.%u on port %d: is it running "
102                                    "a compatible version of Lustre?\n",
103                                    libcfs_nid2str(peer_nid),
104                                    HIPQUAD(peer_ip), peer_port);
105                 break;
106         case -EADDRINUSE:
107                 LCONSOLE_ERROR_MSG(0x11d, "No privileged ports available to "
108                                    "connect to %s at host %u.%u.%u.%u on port "
109                                    "%d\n", libcfs_nid2str(peer_nid),
110                                    HIPQUAD(peer_ip), peer_port);
111                 break;
112         default:
113                 LCONSOLE_ERROR_MSG(0x11e, "Unexpected error %d connecting to %s"
114                                    " at host %u.%u.%u.%u on port %d\n", rc,
115                                    libcfs_nid2str(peer_nid),
116                                    HIPQUAD(peer_ip), peer_port);
117                 break;
118         }
119 }
120 EXPORT_SYMBOL(lnet_connect_console_error);
121
122 int
123 lnet_connect(cfs_socket_t **sockp, lnet_nid_t peer_nid,
124             __u32 local_ip, __u32 peer_ip, int peer_port)
125 {
126         lnet_acceptor_connreq_t cr;
127         cfs_socket_t           *sock;
128         int                     rc;
129         int                     port;
130         int                     fatal;
131
132         CLASSERT (sizeof(cr) <= 16);            /* not too big to be on the stack */
133
134         for (port = LNET_ACCEPTOR_MAX_RESERVED_PORT; 
135              port >= LNET_ACCEPTOR_MIN_RESERVED_PORT; 
136              --port) {
137                 /* Iterate through reserved ports. */
138
139                 rc = libcfs_sock_connect(&sock, &fatal, 
140                                          local_ip, port, 
141                                          peer_ip, peer_port);
142                 if (rc != 0) {
143                         if (fatal)
144                                 goto failed;
145                         continue;
146                 }
147
148                 CLASSERT (LNET_PROTO_ACCEPTOR_VERSION == 1);
149
150                 if (the_lnet.ln_ptlcompat != 2) {
151                         /* When portals compatibility is "strong", simply
152                          * connect (i.e. send no acceptor connection request).
153                          * Othewise send an acceptor connection request. I can
154                          * have no portals peers so everyone else should
155                          * understand my protocol. */
156                         cr.acr_magic   = LNET_PROTO_ACCEPTOR_MAGIC;
157                         cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
158                         cr.acr_nid     = peer_nid;
159
160                         if (the_lnet.ln_testprotocompat != 0) {
161                                 /* single-shot proto check */
162                                 LNET_LOCK();
163                                 if ((the_lnet.ln_testprotocompat & 4) != 0) {
164                                         cr.acr_version++;
165                                         the_lnet.ln_testprotocompat &= ~4;
166                                 }
167                                 if ((the_lnet.ln_testprotocompat & 8) != 0) {
168                                         cr.acr_magic = LNET_PROTO_MAGIC;
169                                         the_lnet.ln_testprotocompat &= ~8;
170                                 }
171                                 LNET_UNLOCK();
172                         }
173
174                         rc = libcfs_sock_write(sock, &cr, sizeof(cr),
175                                                accept_timeout);
176                         if (rc != 0)
177                                 goto failed_sock;
178                 }
179                 
180                 *sockp = sock;
181                 return 0;
182         }
183
184         rc = -EADDRINUSE;
185         goto failed;
186         
187  failed_sock:
188         libcfs_sock_release(sock);
189  failed:
190         lnet_connect_console_error(rc, peer_nid, peer_ip, peer_port);
191         return rc;
192 }
193 EXPORT_SYMBOL(lnet_connect);
194
195 static inline int
196 lnet_accept_magic(__u32 magic, __u32 constant)
197 {
198         return (magic == constant ||
199                 magic == __swab32(constant));
200 }
201
202 int
203 lnet_accept(lnet_ni_t *blind_ni, cfs_socket_t *sock, __u32 magic)
204 {
205         lnet_acceptor_connreq_t cr;
206         __u32                   peer_ip;
207         int                     peer_port;
208         int                     rc;
209         int                     flip;
210         lnet_ni_t              *ni;
211         char                   *str;
212
213         /* CAVEAT EMPTOR: I may be called by an LND in any thread's context if
214          * I passed the new socket "blindly" to the single NI that needed an
215          * acceptor.  If so, blind_ni != NULL... */
216
217         LASSERT (sizeof(cr) <= 16);             /* not too big for the stack */
218         
219         rc = libcfs_sock_getaddr(sock, 1, &peer_ip, &peer_port);
220         LASSERT (rc == 0);                      /* we succeeded before */
221
222         if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) {
223
224                 if (lnet_accept_magic(magic, LNET_PROTO_MAGIC)) {
225                         /* future version compatibility!
226                          * When LNET unifies protocols over all LNDs, the first
227                          * thing sent will be a version query.  I send back
228                          * LNET_PROTO_ACCEPTOR_MAGIC to tell her I'm "old" */
229
230                         memset (&cr, 0, sizeof(cr));
231                         cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
232                         cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
233                         rc = libcfs_sock_write(sock, &cr, sizeof(cr),
234                                                accept_timeout);
235
236                         if (rc != 0)
237                                 CERROR("Error sending magic+version in response"
238                                        "to LNET magic from %u.%u.%u.%u: %d\n",
239                                        HIPQUAD(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 if (lnet_accept_magic(magic, LNET_PROTO_RA_MAGIC))
246                         str = "'old' ranal";
247                 else if (lnet_accept_magic(magic, LNET_PROTO_OPENIB_MAGIC))
248                         str = "'old' openibnal";
249                 else
250                         str = "unrecognised";
251             
252                 LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %u.%u.%u.%u"
253                                    " magic %08x: %s acceptor protocol\n",
254                                    HIPQUAD(peer_ip), magic, str);
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) {
264                 CERROR("Error %d reading connection request version from "
265                        "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
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
279                 memset (&cr, 0, sizeof(cr));
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)
287                         CERROR("Error sending magic+version in response"
288                                "to version %d from %u.%u.%u.%u: %d\n",
289                                peer_version, HIPQUAD(peer_ip), rc);
290                 return -EPROTO;
291         }
292
293         rc = libcfs_sock_read(sock, &cr.acr_nid,
294                               sizeof(cr) -
295                               offsetof(lnet_acceptor_connreq_t, acr_nid),
296                               accept_timeout);
297         if (rc != 0) {
298                 CERROR("Error %d reading connection request from "
299                        "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
300                 return -EIO;
301         }
302
303         if (flip)
304                 __swab64s(&cr.acr_nid);
305
306         ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid));
307         if (ni == NULL ||               /* no matching net */
308             ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */
309                 if (ni != NULL)
310                         lnet_ni_decref(ni);
311                 LCONSOLE_ERROR_MSG(0x120, "Refusing connection from %u.%u.%u.%u"
312                                    " for %s: No matching NI\n",
313                                    HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
314                 return -EPERM;
315         }
316
317         if (ni->ni_lnd->lnd_accept == NULL) {
318                 /* This catches a request for the loopback LND */
319                 lnet_ni_decref(ni);
320                 LCONSOLE_ERROR_MSG(0x121, "Refusing connection from %u.%u.%u.%u"
321                                   " for %s: NI doesn not accept IP connections\n",
322                                   HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
323                 return -EPERM;
324         }
325
326         CDEBUG(D_NET, "Accept %s from %u.%u.%u.%u%s\n",
327                libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip),
328                blind_ni == NULL ? "" : " (blind)");
329
330         if (blind_ni == NULL) {
331                 /* called by the acceptor: call into the requested NI... */
332                 rc = ni->ni_lnd->lnd_accept(ni, sock);
333         } else {
334                 /* portals_compatible set and the (only) NI called me to verify
335                  * and skip the connection request... */
336                 LASSERT (the_lnet.ln_ptlcompat != 0);
337                 LASSERT (ni == blind_ni);
338                 rc = 0;
339         }
340
341         lnet_ni_decref(ni);
342         return rc;
343 }
344 EXPORT_SYMBOL(lnet_accept);
345         
346 int
347 lnet_acceptor(void *arg)
348 {
349         char           name[16];
350         cfs_socket_t  *newsock;
351         int            rc;
352         int            n_acceptor_nis;
353         __u32          magic;
354         __u32          peer_ip;
355         int            peer_port;
356         lnet_ni_t     *blind_ni = NULL;
357         int            secure = (int)((unsigned long)arg);
358
359         LASSERT (lnet_acceptor_state.pta_sock == NULL);
360
361         if (the_lnet.ln_ptlcompat != 0) {
362                 /* When portals_compatibility is enabled, peers may connect
363                  * without sending an acceptor connection request.  There is no
364                  * ambiguity about which network the peer wants to connect to
365                  * since there can only be 1 network, so I pass connections
366                  * "blindly" to it. */
367                 n_acceptor_nis = lnet_count_acceptor_nis(&blind_ni);
368                 LASSERT (n_acceptor_nis == 1);
369                 LASSERT (blind_ni != NULL);
370         }
371
372         snprintf(name, sizeof(name), "acceptor_%03d", accept_port);
373         cfs_daemonize(name);
374         cfs_block_allsigs();
375
376         rc = libcfs_sock_listen(&lnet_acceptor_state.pta_sock,
377                                 0, accept_port, accept_backlog);
378         if (rc != 0) {
379                 if (rc == -EADDRINUSE)
380                         LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port"
381                                            " %d: port already in use\n",
382                                            accept_port);
383                 else
384                         LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port "
385                                            "%d: unexpected error %d\n",
386                                            accept_port, rc);
387
388                 lnet_acceptor_state.pta_sock = NULL;
389         } else {
390                 LCONSOLE(0, "Accept %s, port %d%s\n", 
391                          accept, accept_port,
392                          blind_ni == NULL ? "" : " (proto compatible)");
393         }
394         
395         /* set init status and unblock parent */
396         lnet_acceptor_state.pta_shutdown = rc;
397         mutex_up(&lnet_acceptor_state.pta_signal);
398         
399         if (rc != 0)
400                 return rc;
401
402         while (!lnet_acceptor_state.pta_shutdown) {
403                 
404                 rc = libcfs_sock_accept(&newsock, lnet_acceptor_state.pta_sock);
405                 if (rc != 0) {
406                         if (rc != -EAGAIN) {
407                                 CWARN("Accept error %d: pausing...\n", rc);
408                                 cfs_pause(cfs_time_seconds(1));
409                         }
410                         continue;
411                 }
412
413                 rc = libcfs_sock_getaddr(newsock, 1, &peer_ip, &peer_port);
414                 if (rc != 0) {
415                         CERROR("Can't determine new connection's address\n");
416                         goto failed;
417                 }
418
419                 if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
420                         CERROR("Refusing connection from %u.%u.%u.%u: "
421                                "insecure port %d\n",
422                                HIPQUAD(peer_ip), peer_port);
423                         goto failed;
424                 }
425
426                 if (blind_ni != NULL) {
427                         rc = blind_ni->ni_lnd->lnd_accept(blind_ni, newsock);
428                         if (rc != 0) {
429                                 CERROR("NI %s refused 'blind' connection from "
430                                        "%u.%u.%u.%u\n", 
431                                        libcfs_nid2str(blind_ni->ni_nid), 
432                                        HIPQUAD(peer_ip));
433                                 goto failed;
434                         }
435                         continue;
436                 }
437                 
438                 rc = libcfs_sock_read(newsock, &magic, sizeof(magic),
439                                       accept_timeout);
440                 if (rc != 0) {
441                         CERROR("Error %d reading connection request from "
442                                "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
443                         goto failed;
444                 }
445
446                 rc = lnet_accept(NULL, newsock, magic);
447                 if (rc != 0)
448                         goto failed;
449                 
450                 continue;
451                 
452         failed:
453                 libcfs_sock_release(newsock);
454         }
455         
456         libcfs_sock_release(lnet_acceptor_state.pta_sock);
457         lnet_acceptor_state.pta_sock = NULL;
458
459         if (blind_ni != NULL)
460                 lnet_ni_decref(blind_ni);
461
462         LCONSOLE(0,"Acceptor stopping\n");
463         
464         /* unblock lnet_acceptor_stop() */
465         mutex_up(&lnet_acceptor_state.pta_signal);
466         return 0;
467 }
468
469 int
470 lnet_acceptor_start(void)
471 {
472         long   pid;
473         long   secure;
474
475         LASSERT (lnet_acceptor_state.pta_sock == NULL);
476         init_mutex_locked(&lnet_acceptor_state.pta_signal);
477
478         if (!strcmp(accept, "secure")) {
479                 secure = 1;
480         } else if (!strcmp(accept, "all")) {
481                 secure = 0;
482         } else if (!strcmp(accept, "none")) {
483                 return 0;
484         } else {
485                 LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n",
486                                    accept);
487                 return -EINVAL;
488         }
489         
490         if (lnet_count_acceptor_nis(NULL) == 0)  /* not required */
491                 return 0;
492         
493         pid = cfs_kernel_thread(lnet_acceptor, (void *)secure, 0);
494         if (pid < 0) {
495                 CERROR("Can't start acceptor thread: %ld\n", pid);
496                 return -ESRCH;
497         }
498
499         mutex_down(&lnet_acceptor_state.pta_signal); /* wait for acceptor to startup */
500
501         if (!lnet_acceptor_state.pta_shutdown) {
502                 /* started OK */
503                 LASSERT (lnet_acceptor_state.pta_sock != NULL);
504                 return 0;
505         }
506
507         LASSERT (lnet_acceptor_state.pta_sock == NULL);
508         return -ENETDOWN;
509 }
510
511 void
512 lnet_acceptor_stop(void)
513 {
514         if (lnet_acceptor_state.pta_sock == NULL) /* not running */
515                 return;
516         
517         lnet_acceptor_state.pta_shutdown = 1;
518         libcfs_sock_abort_accept(lnet_acceptor_state.pta_sock);
519
520         /* block until acceptor signals exit */
521         mutex_down(&lnet_acceptor_state.pta_signal);
522 }
523
524 #else /* __KERNEL__ */
525 #ifdef HAVE_LIBPTHREAD
526
527 static char *accept_type;
528 static int accept_port = 988;
529 static int accept_backlog;
530 static int accept_timeout;
531
532 struct {
533         int                   pta_shutdown;
534         int                   pta_sock;
535         struct cfs_completion pta_completion;
536 } lnet_acceptor_state;
537
538 int
539 lnet_acceptor_port(void)
540 {
541         return accept_port;
542 }
543
544 int
545 lnet_parse_int_tunable(int *value, char *name, int dflt)
546 {
547         char    *env = getenv(name);
548         char    *end;
549
550         if (env == NULL) {
551                 *value = dflt;
552                 return 0;
553         }
554
555         *value = strtoull(env, &end, 0);
556         if (*end == 0)
557                 return 0;
558
559         CERROR("Can't parse tunable %s=%s\n", name, env);
560         return -EINVAL;
561 }
562
563 int
564 lnet_parse_string_tunable(char **value, char *name, char *dflt)
565 {
566         char    *env = getenv(name);
567
568         if (env == NULL)
569                 *value = dflt;             
570         else
571                 *value = env;
572
573         return 0;
574 }
575
576 int
577 lnet_acceptor_get_tunables()
578 {
579         int rc;
580         rc = lnet_parse_string_tunable(&accept_type, "LNET_ACCEPT", "secure");
581
582         if (rc != 0)
583                 return rc;
584
585         rc = lnet_parse_int_tunable(&accept_port, "LNET_ACCEPT_PORT", 988);
586
587         if (rc != 0)
588                 return rc;
589         
590         rc = lnet_parse_int_tunable(&accept_backlog, "LNET_ACCEPT_BACKLOG", 127);
591
592         if (rc != 0)
593                 return rc;
594
595         rc = lnet_parse_int_tunable(&accept_timeout, "LNET_ACCEPT_TIMEOUT", 5);
596
597         if (rc != 0)
598                 return rc;
599
600         CDEBUG(D_NET, "accept_type     = %s\n", accept_type);
601         CDEBUG(D_NET, "accept_port     = %d\n", accept_port);
602         CDEBUG(D_NET, "accept_backlog  = %d\n", accept_backlog);
603         CDEBUG(D_NET, "accept_timeout  = %d\n", accept_timeout);
604         return 0;
605 }
606
607 static inline int
608 lnet_accept_magic(__u32 magic, __u32 constant)
609 {
610         return (magic == constant ||
611                 magic == __swab32(constant));
612 }
613
614 /* user-land lnet_accept() isn't used by any LND's directly. So, we don't
615  * do it visible outside acceptor.c and we can change its prototype
616  * freely */
617 static int
618 lnet_accept(int sock, __u32 magic, __u32 peer_ip, int peer_port)
619 {
620         int rc, flip;
621         lnet_acceptor_connreq_t cr;
622         lnet_ni_t *ni;
623         
624         if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) {
625                 LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u magic %08x: "
626                                "unsupported acceptor protocol\n",
627                                HIPQUAD(peer_ip), magic);
628                 return -EPROTO;
629         }
630
631         flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC);
632         
633         rc = libcfs_sock_read(sock, &cr.acr_version, 
634                               sizeof(cr.acr_version),
635                               accept_timeout);
636         if (rc != 0) {
637                 CERROR("Error %d reading connection request version from "
638                        "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
639                 return -EIO;
640         }
641
642         if (flip)
643                 __swab32s(&cr.acr_version);
644
645         if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION)
646                 return -EPROTO;
647                 
648         rc = libcfs_sock_read(sock, &cr.acr_nid,
649                               sizeof(cr) -
650                               offsetof(lnet_acceptor_connreq_t, acr_nid),
651                               accept_timeout);
652         if (rc != 0) {
653                 CERROR("Error %d reading connection request from "
654                        "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
655                 return -EIO;
656         }
657
658         if (flip)
659                 __swab64s(&cr.acr_nid);
660
661         ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid));
662                        
663         if (ni == NULL ||                    /* no matching net */
664              ni->ni_nid != cr.acr_nid) {     /* right NET, wrong NID! */
665                 if (ni != NULL)
666                         lnet_ni_decref(ni);
667                 LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u for %s: "
668                                " No matching NI\n",
669                                HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
670                 return -EPERM;
671         }
672
673         if (ni->ni_lnd->lnd_accept == NULL) {
674                 lnet_ni_decref(ni);
675                 LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u for %s: "
676                                " NI doesn not accept IP connections\n",
677                                HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
678                 return -EPERM;
679         }
680         
681         CDEBUG(D_NET, "Accept %s from %u.%u.%u.%u\n",
682                libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip));
683
684         rc = ni->ni_lnd->lnd_accept(ni, sock);
685         
686         lnet_ni_decref(ni);
687         return rc;
688 }
689
690 int
691 lnet_acceptor(void *arg)
692 {
693         char           name[16];
694         int            secure = (int)((unsigned long)arg);
695         int            rc;
696         int            newsock;
697         __u32          peer_ip;
698         int            peer_port;
699         __u32          magic;
700
701         snprintf(name, sizeof(name), "acceptor_%03d", accept_port);
702         cfs_daemonize(name);
703         cfs_block_allsigs();
704
705         rc = libcfs_sock_listen(&lnet_acceptor_state.pta_sock,
706                                 0, accept_port, accept_backlog);
707         if (rc != 0) {
708                 if (rc == -EADDRINUSE)
709                         LCONSOLE_ERROR("Can't start acceptor on port %d: "
710                                        "port already in use\n",
711                                        accept_port);
712                 else
713                         LCONSOLE_ERROR("Can't start acceptor on port %d: "
714                                        "unexpected error %d\n",
715                                        accept_port, rc);
716
717         } else {
718                 LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port);
719         }
720         
721         /* set init status and unblock parent */
722         lnet_acceptor_state.pta_shutdown = rc;
723         cfs_complete(&lnet_acceptor_state.pta_completion);
724
725         if (rc != 0)
726                 return rc;
727
728         while (!lnet_acceptor_state.pta_shutdown) {
729
730                 rc = libcfs_sock_accept(&newsock, lnet_acceptor_state.pta_sock,
731                                         &peer_ip, &peer_port);
732                 if (rc != 0)
733                         continue;
734
735                 /* maybe we're waken up with libcfs_sock_abort_accept() */
736                 if (lnet_acceptor_state.pta_shutdown) {
737                         close(newsock);
738                         break;
739                 }
740
741                 if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
742                         CERROR("Refusing connection from %u.%u.%u.%u: "
743                                "insecure port %d\n",
744                                HIPQUAD(peer_ip), peer_port);
745                         goto failed;
746                 }
747
748                 rc = libcfs_sock_read(newsock, &magic, sizeof(magic),
749                                       accept_timeout);
750                 if (rc != 0) {
751                         CERROR("Error %d reading connection request from "
752                                "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
753                         goto failed;
754                 }
755
756                 rc = lnet_accept(newsock, magic, peer_ip, peer_port);
757                 if (rc != 0)
758                         goto failed;
759                 
760                 continue;
761                 
762           failed:
763                 close(newsock);
764         }
765         
766         close(lnet_acceptor_state.pta_sock);
767         LCONSOLE(0,"Acceptor stopping\n");
768
769         /* unblock lnet_acceptor_stop() */
770         cfs_complete(&lnet_acceptor_state.pta_completion);        
771
772         return 0;
773 }
774
775 static int skip_waiting_for_completion;
776
777 int
778 lnet_acceptor_start(void)
779 {
780         long   secure;
781         int rc;
782
783         rc = lnet_acceptor_get_tunables();
784         if (rc != 0)
785                 return rc;
786
787         /* Do nothing if we're liblustre clients */
788         if ((the_lnet.ln_pid & LNET_PID_USERFLAG) != 0)
789                 return 0;
790                         
791         cfs_init_completion(&lnet_acceptor_state.pta_completion);
792
793         if (!strcmp(accept_type, "secure")) {
794                 secure = 1;
795         } else if (!strcmp(accept_type, "all")) {
796                 secure = 0;
797         } else if (!strcmp(accept_type, "none")) {
798                 skip_waiting_for_completion = 1;
799                 return 0;
800         } else {
801                 LCONSOLE_ERROR ("Can't parse 'accept_type=\"%s\"'\n", accept_type);
802                 cfs_fini_completion(&lnet_acceptor_state.pta_completion);
803                 return -EINVAL;
804         }
805
806         if (lnet_count_acceptor_nis(NULL) == 0) { /* not required */
807                 skip_waiting_for_completion = 1;
808                 return 0;
809         }
810
811         rc = cfs_create_thread(lnet_acceptor, (void *)secure);
812         if (rc != 0) {
813                 CERROR("Can't start acceptor thread: %d\n", rc);
814                 cfs_fini_completion(&lnet_acceptor_state.pta_completion);
815                 return rc;
816         }
817
818         /* wait for acceptor to startup */
819         cfs_wait_for_completion(&lnet_acceptor_state.pta_completion);
820
821         if (!lnet_acceptor_state.pta_shutdown)
822                 return 0;
823         
824         cfs_fini_completion(&lnet_acceptor_state.pta_completion);
825         return -ENETDOWN;
826 }
827
828 void
829 lnet_acceptor_stop(void)
830 {
831         /* Do nothing if we're liblustre clients */
832         if ((the_lnet.ln_pid & LNET_PID_USERFLAG) != 0)
833                 return;
834
835         if (!skip_waiting_for_completion) {
836                 lnet_acceptor_state.pta_shutdown = 1;
837                 libcfs_sock_abort_accept(accept_port);
838                 
839                 /* block until acceptor signals exit */
840                 cfs_wait_for_completion(&lnet_acceptor_state.pta_completion);
841         }
842         
843         cfs_fini_completion(&lnet_acceptor_state.pta_completion);
844 }
845 #else
846 int
847 lnet_acceptor_start(void)
848 {
849         return 0;
850 }
851
852 void
853 lnet_acceptor_stop(void)
854 {
855 }
856 #endif /* !HAVE_LIBPTHREAD */
857 #endif /* !__KERNEL__ */