lnet/lnet/acceptor.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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 <lnet/lib-lnet.h>
  39
  40 #ifdef __KERNEL__
  41 static char *accept = "secure";
  42 CFS_MODULE_PARM(accept, "s", charp, 0444,
  43                 "Accept connections (secure|all|none)");
  44
  45 static int accept_port = 988;
  46 CFS_MODULE_PARM(accept_port, "i", int, 0444,
  47                 "Acceptor's port (same on all nodes)");
  48
  49 static int accept_backlog = 127;
  50 CFS_MODULE_PARM(accept_backlog, "i", int, 0444,
  51                 "Acceptor's listen backlog");
  52
  53 static int accept_timeout = 5;
  54 CFS_MODULE_PARM(accept_timeout, "i", int, 0644,
  55                 "Acceptor's timeout (seconds)");
  56
  57 struct {
  58         int               pta_shutdown;
  59         cfs_socket_t     *pta_sock;
  60         struct semaphore  pta_signal;
  61 } lnet_acceptor_state;
  62
  63 int
  64 lnet_acceptor_timeout(void)
  65 {
  66         return accept_timeout;
  67 }
  68 EXPORT_SYMBOL(lnet_acceptor_timeout);
  69
  70 int
  71 lnet_acceptor_port(void)
  72 {
  73         return accept_port;
  74 }
  75 EXPORT_SYMBOL(lnet_acceptor_port);
  76
  77 void
  78 lnet_connect_console_error (int rc, lnet_nid_t peer_nid,
  79                            __u32 peer_ip, int peer_port)
  80 {
  81         switch (rc) {
  82         /* "normal" errors */
  83         case -ECONNREFUSED:
  84                 CDEBUG(D_NETERROR, "Connection to %s at host %u.%u.%u.%u "
  85                        "on port %d was refused: "
  86                        "check that Lustre is running on that node.\n",
  87                        libcfs_nid2str(peer_nid),
  88                        HIPQUAD(peer_ip), peer_port);
  89                 break;
  90         case -EHOSTUNREACH:
  91         case -ENETUNREACH:
  92                 CDEBUG(D_NETERROR, "Connection to %s at host %u.%u.%u.%u "
  93                        "was unreachable: the network or that node may "
  94                        "be down, or Lustre may be misconfigured.\n",
  95                        libcfs_nid2str(peer_nid), HIPQUAD(peer_ip));
  96                 break;
  97         case -ETIMEDOUT:
  98                 CDEBUG(D_NETERROR, "Connection to %s at host %u.%u.%u.%u on "
  99                        "port %d took too long: that node may be hung "
 100                        "or experiencing high load.\n",
 101                        libcfs_nid2str(peer_nid),
 102                        HIPQUAD(peer_ip), peer_port);
 103                 break;
 104         case -ECONNRESET:
 105                 LCONSOLE_ERROR_MSG(0x11b, "Connection to %s at host %u.%u.%u.%u"
 106                                    " on port %d was reset: "
 107                                    "is it running a compatible version of "
 108                                    "Lustre and is %s one of its NIDs?\n",
 109                                    libcfs_nid2str(peer_nid),
 110                                    HIPQUAD(peer_ip), peer_port,
 111                                    libcfs_nid2str(peer_nid));
 112                 break;
 113         case -EPROTO:
 114                 LCONSOLE_ERROR_MSG(0x11c, "Protocol error connecting to %s at "
 115                                    "host %u.%u.%u.%u on port %d: is it running "
 116                                    "a compatible version of Lustre?\n",
 117                                    libcfs_nid2str(peer_nid),
 118                                    HIPQUAD(peer_ip), peer_port);
 119                 break;
 120         case -EADDRINUSE:
 121                 LCONSOLE_ERROR_MSG(0x11d, "No privileged ports available to "
 122                                    "connect to %s at host %u.%u.%u.%u on port "
 123                                    "%d\n", libcfs_nid2str(peer_nid),
 124                                    HIPQUAD(peer_ip), peer_port);
 125                 break;
 126         default:
 127                 LCONSOLE_ERROR_MSG(0x11e, "Unexpected error %d connecting to %s"
 128                                    " at host %u.%u.%u.%u on port %d\n", rc,
 129                                    libcfs_nid2str(peer_nid),
 130                                    HIPQUAD(peer_ip), peer_port);
 131                 break;
 132         }
 133 }
 134 EXPORT_SYMBOL(lnet_connect_console_error);
 135
 136 int
 137 lnet_connect(cfs_socket_t **sockp, lnet_nid_t peer_nid,
 138             __u32 local_ip, __u32 peer_ip, int peer_port)
 139 {
 140         lnet_acceptor_connreq_t cr;
 141         cfs_socket_t           *sock;
 142         int                     rc;
 143         int                     port;
 144         int                     fatal;
 145
 146         CLASSERT (sizeof(cr) <= 16);            /* not too big to be on the stack */
 147
 148         for (port = LNET_ACCEPTOR_MAX_RESERVED_PORT;
 149              port >= LNET_ACCEPTOR_MIN_RESERVED_PORT;
 150              --port) {
 151                 /* Iterate through reserved ports. */
 152
 153                 rc = libcfs_sock_connect(&sock, &fatal,
 154                                          local_ip, port,
 155                                          peer_ip, peer_port);
 156                 if (rc != 0) {
 157                         if (fatal)
 158                                 goto failed;
 159                         continue;
 160                 }
 161
 162                 CLASSERT (LNET_PROTO_ACCEPTOR_VERSION == 1);
 163
 164                 cr.acr_magic   = LNET_PROTO_ACCEPTOR_MAGIC;
 165                 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
 166                 cr.acr_nid     = peer_nid;
 167
 168                 if (the_lnet.ln_testprotocompat != 0) {
 169                         /* single-shot proto check */
 170                         LNET_LOCK();
 171                         if ((the_lnet.ln_testprotocompat & 4) != 0) {
 172                                 cr.acr_version++;
 173                                 the_lnet.ln_testprotocompat &= ~4;
 174                         }
 175                         if ((the_lnet.ln_testprotocompat & 8) != 0) {
 176                                 cr.acr_magic = LNET_PROTO_MAGIC;
 177                                 the_lnet.ln_testprotocompat &= ~8;
 178                         }
 179                         LNET_UNLOCK();
 180                 }
 181
 182                 rc = libcfs_sock_write(sock, &cr, sizeof(cr),
 183                                        accept_timeout);
 184                 if (rc != 0)
 185                         goto failed_sock;
 186
 187                 *sockp = sock;
 188                 return 0;
 189         }
 190
 191         rc = -EADDRINUSE;
 192         goto failed;
 193
 194  failed_sock:
 195         libcfs_sock_release(sock);
 196  failed:
 197         lnet_connect_console_error(rc, peer_nid, peer_ip, peer_port);
 198         return rc;
 199 }
 200 EXPORT_SYMBOL(lnet_connect);
 201
 202 static inline int
 203 lnet_accept_magic(__u32 magic, __u32 constant)
 204 {
 205         return (magic == constant ||
 206                 magic == __swab32(constant));
 207 }
 208
 209 int
 210 lnet_accept(cfs_socket_t *sock, __u32 magic)
 211 {
 212         lnet_acceptor_connreq_t cr;
 213         __u32                   peer_ip;
 214         int                     peer_port;
 215         int                     rc;
 216         int                     flip;
 217         lnet_ni_t              *ni;
 218         char                   *str;
 219
 220         LASSERT (sizeof(cr) <= 16);             /* not too big for the stack */
 221
 222         rc = libcfs_sock_getaddr(sock, 1, &peer_ip, &peer_port);
 223         LASSERT (rc == 0);                      /* we succeeded before */
 224
 225         if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) {
 226
 227                 if (lnet_accept_magic(magic, LNET_PROTO_MAGIC)) {
 228                         /* future version compatibility!
 229                          * When LNET unifies protocols over all LNDs, the first
 230                          * thing sent will be a version query.  I send back
 231                          * LNET_PROTO_ACCEPTOR_MAGIC to tell her I'm "old" */
 232
 233                         memset (&cr, 0, sizeof(cr));
 234                         cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
 235                         cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
 236                         rc = libcfs_sock_write(sock, &cr, sizeof(cr),
 237                                                accept_timeout);
 238
 239                         if (rc != 0)
 240                                 CERROR("Error sending magic+version in response"
 241                                        "to LNET magic from %u.%u.%u.%u: %d\n",
 242                                        HIPQUAD(peer_ip), rc);
 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 if (lnet_accept_magic(magic, LNET_PROTO_OPENIB_MAGIC))
 251                         str = "'old' openibnal";
 252                 else
 253                         str = "unrecognised";
 254
 255                 LCONSOLE_ERROR_MSG(0x11f, "Refusing connection from %u.%u.%u.%u"
 256                                    " magic %08x: %s acceptor protocol\n",
 257                                    HIPQUAD(peer_ip), magic, str);
 258                 return -EPROTO;
 259         }
 260
 261         flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC);
 262
 263         rc = libcfs_sock_read(sock, &cr.acr_version,
 264                               sizeof(cr.acr_version),
 265                               accept_timeout);
 266         if (rc != 0) {
 267                 CERROR("Error %d reading connection request version from "
 268                        "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
 269                 return -EIO;
 270         }
 271
 272         if (flip)
 273                 __swab32s(&cr.acr_version);
 274
 275         if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION) {
 276                 /* future version compatibility!
 277                  * An acceptor-specific protocol rev will first send a version
 278                  * query.  I send back my current version to tell her I'm
 279                  * "old". */
 280                 int peer_version = cr.acr_version;
 281
 282                 memset (&cr, 0, sizeof(cr));
 283                 cr.acr_magic = LNET_PROTO_ACCEPTOR_MAGIC;
 284                 cr.acr_version = LNET_PROTO_ACCEPTOR_VERSION;
 285
 286                 rc = libcfs_sock_write(sock, &cr, sizeof(cr),
 287                                        accept_timeout);
 288
 289                 if (rc != 0)
 290                         CERROR("Error sending magic+version in response"
 291                                "to version %d from %u.%u.%u.%u: %d\n",
 292                                peer_version, HIPQUAD(peer_ip), rc);
 293                 return -EPROTO;
 294         }
 295
 296         rc = libcfs_sock_read(sock, &cr.acr_nid,
 297                               sizeof(cr) -
 298                               offsetof(lnet_acceptor_connreq_t, acr_nid),
 299                               accept_timeout);
 300         if (rc != 0) {
 301                 CERROR("Error %d reading connection request from "
 302                        "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
 303                 return -EIO;
 304         }
 305
 306         if (flip)
 307                 __swab64s(&cr.acr_nid);
 308
 309         ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid));
 310         if (ni == NULL ||               /* no matching net */
 311             ni->ni_nid != cr.acr_nid) { /* right NET, wrong NID! */
 312                 if (ni != NULL)
 313                         lnet_ni_decref(ni);
 314                 LCONSOLE_ERROR_MSG(0x120, "Refusing connection from %u.%u.%u.%u"
 315                                    " for %s: No matching NI\n",
 316                                    HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
 317                 return -EPERM;
 318         }
 319
 320         if (ni->ni_lnd->lnd_accept == NULL) {
 321                 /* This catches a request for the loopback LND */
 322                 lnet_ni_decref(ni);
 323                 LCONSOLE_ERROR_MSG(0x121, "Refusing connection from %u.%u.%u.%u"
 324                                   " for %s: NI doesn not accept IP connections\n",
 325                                   HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
 326                 return -EPERM;
 327         }
 328
 329         CDEBUG(D_NET, "Accept %s from %u.%u.%u.%u\n",
 330                libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip));
 331
 332         rc = ni->ni_lnd->lnd_accept(ni, sock);
 333         lnet_ni_decref(ni);
 334         return rc;
 335 }
 336
 337 int
 338 lnet_acceptor(void *arg)
 339 {
 340         char           name[16];
 341         cfs_socket_t  *newsock;
 342         int            rc;
 343         __u32          magic;
 344         __u32          peer_ip;
 345         int            peer_port;
 346         int            secure = (int)((long_ptr_t)arg);
 347
 348         LASSERT (lnet_acceptor_state.pta_sock == NULL);
 349
 350         snprintf(name, sizeof(name), "acceptor_%03d", accept_port);
 351         cfs_daemonize(name);
 352         cfs_block_allsigs();
 353
 354         rc = libcfs_sock_listen(&lnet_acceptor_state.pta_sock,
 355                                 0, accept_port, accept_backlog);
 356         if (rc != 0) {
 357                 if (rc == -EADDRINUSE)
 358                         LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port"
 359                                            " %d: port already in use\n",
 360                                            accept_port);
 361                 else
 362                         LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port "
 363                                            "%d: unexpected error %d\n",
 364                                            accept_port, rc);
 365
 366                 lnet_acceptor_state.pta_sock = NULL;
 367         } else {
 368                 LCONSOLE(0, "Accept %s, port %d\n", accept, accept_port);
 369         }
 370
 371         /* set init status and unblock parent */
 372         lnet_acceptor_state.pta_shutdown = rc;
 373         mutex_up(&lnet_acceptor_state.pta_signal);
 374
 375         if (rc != 0)
 376                 return rc;
 377
 378         while (!lnet_acceptor_state.pta_shutdown) {
 379
 380                 rc = libcfs_sock_accept(&newsock, lnet_acceptor_state.pta_sock);
 381                 if (rc != 0) {
 382                         if (rc != -EAGAIN) {
 383                                 CWARN("Accept error %d: pausing...\n", rc);
 384                                 cfs_pause(cfs_time_seconds(1));
 385                         }
 386                         continue;
 387                 }
 388
 389                 rc = libcfs_sock_getaddr(newsock, 1, &peer_ip, &peer_port);
 390                 if (rc != 0) {
 391                         CERROR("Can't determine new connection's address\n");
 392                         goto failed;
 393                 }
 394
 395                 if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
 396                         CERROR("Refusing connection from %u.%u.%u.%u: "
 397                                "insecure port %d\n",
 398                                HIPQUAD(peer_ip), peer_port);
 399                         goto failed;
 400                 }
 401
 402                 rc = libcfs_sock_read(newsock, &magic, sizeof(magic),
 403                                       accept_timeout);
 404                 if (rc != 0) {
 405                         CERROR("Error %d reading connection request from "
 406                                "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
 407                         goto failed;
 408                 }
 409
 410                 rc = lnet_accept(newsock, magic);
 411                 if (rc != 0)
 412                         goto failed;
 413
 414                 continue;
 415
 416         failed:
 417                 libcfs_sock_release(newsock);
 418         }
 419
 420         libcfs_sock_release(lnet_acceptor_state.pta_sock);
 421         lnet_acceptor_state.pta_sock = NULL;
 422
 423         LCONSOLE(0, "Acceptor stopping\n");
 424
 425         /* unblock lnet_acceptor_stop() */
 426         mutex_up(&lnet_acceptor_state.pta_signal);
 427         return 0;
 428 }
 429
 430 int
 431 lnet_acceptor_start(void)
 432 {
 433         long   pid;
 434         long   secure;
 435
 436         LASSERT (lnet_acceptor_state.pta_sock == NULL);
 437         init_mutex_locked(&lnet_acceptor_state.pta_signal);
 438
 439         if (!strcmp(accept, "secure")) {
 440                 secure = 1;
 441         } else if (!strcmp(accept, "all")) {
 442                 secure = 0;
 443         } else if (!strcmp(accept, "none")) {
 444                 return 0;
 445         } else {
 446                 LCONSOLE_ERROR_MSG(0x124, "Can't parse 'accept=\"%s\"'\n",
 447                                    accept);
 448                 return -EINVAL;
 449         }
 450
 451         if (lnet_count_acceptor_nis() == 0)  /* not required */
 452                 return 0;
 453
 454         pid = cfs_kernel_thread(lnet_acceptor, (void *)(ulong_ptr_t)secure, 0);
 455         if (pid < 0) {
 456                 CERROR("Can't start acceptor thread: %ld\n", pid);
 457                 return -ESRCH;
 458         }
 459
 460         mutex_down(&lnet_acceptor_state.pta_signal); /* wait for acceptor to startup */
 461
 462         if (!lnet_acceptor_state.pta_shutdown) {
 463                 /* started OK */
 464                 LASSERT (lnet_acceptor_state.pta_sock != NULL);
 465                 return 0;
 466         }
 467
 468         LASSERT (lnet_acceptor_state.pta_sock == NULL);
 469         return -ENETDOWN;
 470 }
 471
 472 void
 473 lnet_acceptor_stop(void)
 474 {
 475         if (lnet_acceptor_state.pta_sock == NULL) /* not running */
 476                 return;
 477
 478         lnet_acceptor_state.pta_shutdown = 1;
 479         libcfs_sock_abort_accept(lnet_acceptor_state.pta_sock);
 480
 481         /* block until acceptor signals exit */
 482         mutex_down(&lnet_acceptor_state.pta_signal);
 483 }
 484
 485 #else /* __KERNEL__ */
 486 #ifdef HAVE_LIBPTHREAD
 487
 488 static char *accept_type;
 489 static int accept_port = 988;
 490 static int accept_backlog;
 491 static int accept_timeout;
 492
 493 struct {
 494         int                   pta_shutdown;
 495         int                   pta_sock;
 496         struct cfs_completion pta_completion;
 497 } lnet_acceptor_state;
 498
 499 int
 500 lnet_acceptor_port(void)
 501 {
 502         return accept_port;
 503 }
 504
 505 int
 506 lnet_parse_int_tunable(int *value, char *name, int dflt)
 507 {
 508         char    *env = getenv(name);
 509         char    *end;
 510
 511         if (env == NULL) {
 512                 *value = dflt;
 513                 return 0;
 514         }
 515
 516         *value = strtoull(env, &end, 0);
 517         if (*end == 0)
 518                 return 0;
 519
 520         CERROR("Can't parse tunable %s=%s\n", name, env);
 521         return -EINVAL;
 522 }
 523
 524 int
 525 lnet_parse_string_tunable(char **value, char *name, char *dflt)
 526 {
 527         char    *env = getenv(name);
 528
 529         if (env == NULL)
 530                 *value = dflt;
 531         else
 532                 *value = env;
 533
 534         return 0;
 535 }
 536
 537 int
 538 lnet_acceptor_get_tunables()
 539 {
 540         int rc;
 541         rc = lnet_parse_string_tunable(&accept_type, "LNET_ACCEPT", "secure");
 542
 543         if (rc != 0)
 544                 return rc;
 545
 546         rc = lnet_parse_int_tunable(&accept_port, "LNET_ACCEPT_PORT", 988);
 547
 548         if (rc != 0)
 549                 return rc;
 550
 551         rc = lnet_parse_int_tunable(&accept_backlog, "LNET_ACCEPT_BACKLOG", 127);
 552
 553         if (rc != 0)
 554                 return rc;
 555
 556         rc = lnet_parse_int_tunable(&accept_timeout, "LNET_ACCEPT_TIMEOUT", 5);
 557
 558         if (rc != 0)
 559                 return rc;
 560
 561         CDEBUG(D_NET, "accept_type     = %s\n", accept_type);
 562         CDEBUG(D_NET, "accept_port     = %d\n", accept_port);
 563         CDEBUG(D_NET, "accept_backlog  = %d\n", accept_backlog);
 564         CDEBUG(D_NET, "accept_timeout  = %d\n", accept_timeout);
 565         return 0;
 566 }
 567
 568 static inline int
 569 lnet_accept_magic(__u32 magic, __u32 constant)
 570 {
 571         return (magic == constant ||
 572                 magic == __swab32(constant));
 573 }
 574
 575 /* user-land lnet_accept() isn't used by any LND's directly. So, we don't
 576  * do it visible outside acceptor.c and we can change its prototype
 577  * freely */
 578 static int
 579 lnet_accept(int sock, __u32 magic, __u32 peer_ip, int peer_port)
 580 {
 581         int rc, flip;
 582         lnet_acceptor_connreq_t cr;
 583         lnet_ni_t *ni;
 584
 585         if (!lnet_accept_magic(magic, LNET_PROTO_ACCEPTOR_MAGIC)) {
 586                 LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u magic %08x: "
 587                                "unsupported acceptor protocol\n",
 588                                HIPQUAD(peer_ip), magic);
 589                 return -EPROTO;
 590         }
 591
 592         flip = (magic != LNET_PROTO_ACCEPTOR_MAGIC);
 593
 594         rc = libcfs_sock_read(sock, &cr.acr_version,
 595                               sizeof(cr.acr_version),
 596                               accept_timeout);
 597         if (rc != 0) {
 598                 CERROR("Error %d reading connection request version from "
 599                        "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
 600                 return -EIO;
 601         }
 602
 603         if (flip)
 604                 __swab32s(&cr.acr_version);
 605
 606         if (cr.acr_version != LNET_PROTO_ACCEPTOR_VERSION)
 607                 return -EPROTO;
 608
 609         rc = libcfs_sock_read(sock, &cr.acr_nid,
 610                               sizeof(cr) -
 611                               offsetof(lnet_acceptor_connreq_t, acr_nid),
 612                               accept_timeout);
 613         if (rc != 0) {
 614                 CERROR("Error %d reading connection request from "
 615                        "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
 616                 return -EIO;
 617         }
 618
 619         if (flip)
 620                 __swab64s(&cr.acr_nid);
 621
 622         ni = lnet_net2ni(LNET_NIDNET(cr.acr_nid));
 623
 624         if (ni == NULL ||                    /* no matching net */
 625              ni->ni_nid != cr.acr_nid) {     /* right NET, wrong NID! */
 626                 if (ni != NULL)
 627                         lnet_ni_decref(ni);
 628                 LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u for %s: "
 629                                " No matching NI\n",
 630                                HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
 631                 return -EPERM;
 632         }
 633
 634         if (ni->ni_lnd->lnd_accept == NULL) {
 635                 lnet_ni_decref(ni);
 636                 LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u for %s: "
 637                                " NI doesn not accept IP connections\n",
 638                                HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
 639                 return -EPERM;
 640         }
 641
 642         CDEBUG(D_NET, "Accept %s from %u.%u.%u.%u\n",
 643                libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip));
 644
 645         rc = ni->ni_lnd->lnd_accept(ni, sock);
 646
 647         lnet_ni_decref(ni);
 648         return rc;
 649 }
 650
 651 int
 652 lnet_acceptor(void *arg)
 653 {
 654         char           name[16];
 655         int            secure = (int)((unsigned long)arg);
 656         int            rc;
 657         int            newsock;
 658         __u32          peer_ip;
 659         int            peer_port;
 660         __u32          magic;
 661
 662         snprintf(name, sizeof(name), "acceptor_%03d", accept_port);
 663         cfs_daemonize(name);
 664         cfs_block_allsigs();
 665
 666         rc = libcfs_sock_listen(&lnet_acceptor_state.pta_sock,
 667                                 0, accept_port, accept_backlog);
 668         if (rc != 0) {
 669                 if (rc == -EADDRINUSE)
 670                         LCONSOLE_ERROR("Can't start acceptor on port %d: "
 671                                        "port already in use\n",
 672                                        accept_port);
 673                 else
 674                         LCONSOLE_ERROR("Can't start acceptor on port %d: "
 675                                        "unexpected error %d\n",
 676                                        accept_port, rc);
 677
 678         } else {
 679                 LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port);
 680         }
 681
 682         /* set init status and unblock parent */
 683         lnet_acceptor_state.pta_shutdown = rc;
 684         cfs_complete(&lnet_acceptor_state.pta_completion);
 685
 686         if (rc != 0)
 687                 return rc;
 688
 689         while (!lnet_acceptor_state.pta_shutdown) {
 690
 691                 rc = libcfs_sock_accept(&newsock, lnet_acceptor_state.pta_sock,
 692                                         &peer_ip, &peer_port);
 693                 if (rc != 0)
 694                         continue;
 695
 696                 /* maybe we're waken up with libcfs_sock_abort_accept() */
 697                 if (lnet_acceptor_state.pta_shutdown) {
 698                         libcfs_sock_release(newsock);
 699                         break;
 700                 }
 701
 702                 if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
 703                         CERROR("Refusing connection from %u.%u.%u.%u: "
 704                                "insecure port %d\n",
 705                                HIPQUAD(peer_ip), peer_port);
 706                         goto failed;
 707                 }
 708
 709                 rc = libcfs_sock_read(newsock, &magic, sizeof(magic),
 710                                       accept_timeout);
 711                 if (rc != 0) {
 712                         CERROR("Error %d reading connection request from "
 713                                "%u.%u.%u.%u\n", rc, HIPQUAD(peer_ip));
 714                         goto failed;
 715                 }
 716
 717                 rc = lnet_accept(newsock, magic, peer_ip, peer_port);
 718                 if (rc != 0)
 719                         goto failed;
 720
 721                 continue;
 722
 723           failed:
 724                 libcfs_sock_release(newsock);
 725         }
 726
 727         libcfs_sock_release(lnet_acceptor_state.pta_sock);
 728         LCONSOLE(0,"Acceptor stopping\n");
 729
 730         /* unblock lnet_acceptor_stop() */
 731         cfs_complete(&lnet_acceptor_state.pta_completion);
 732
 733         return 0;
 734 }
 735
 736 static int skip_waiting_for_completion;
 737
 738 int
 739 lnet_acceptor_start(void)
 740 {
 741         long   secure;
 742         int rc;
 743
 744         rc = lnet_acceptor_get_tunables();
 745         if (rc != 0)
 746                 return rc;
 747
 748         /* Do nothing if we're liblustre clients */
 749         if ((the_lnet.ln_pid & LNET_PID_USERFLAG) != 0)
 750                 return 0;
 751
 752         cfs_init_completion(&lnet_acceptor_state.pta_completion);
 753
 754         if (!strcmp(accept_type, "secure")) {
 755                 secure = 1;
 756         } else if (!strcmp(accept_type, "all")) {
 757                 secure = 0;
 758         } else if (!strcmp(accept_type, "none")) {
 759                 skip_waiting_for_completion = 1;
 760                 return 0;
 761         } else {
 762                 LCONSOLE_ERROR ("Can't parse 'accept_type=\"%s\"'\n", accept_type);
 763                 cfs_fini_completion(&lnet_acceptor_state.pta_completion);
 764                 return -EINVAL;
 765         }
 766
 767         if (lnet_count_acceptor_nis() == 0) { /* not required */
 768                 skip_waiting_for_completion = 1;
 769                 return 0;
 770         }
 771
 772         rc = cfs_create_thread(lnet_acceptor, (void *)secure);
 773         if (rc != 0) {
 774                 CERROR("Can't start acceptor thread: %d\n", rc);
 775                 cfs_fini_completion(&lnet_acceptor_state.pta_completion);
 776                 return rc;
 777         }
 778
 779         /* wait for acceptor to startup */
 780         cfs_wait_for_completion(&lnet_acceptor_state.pta_completion);
 781
 782         if (!lnet_acceptor_state.pta_shutdown)
 783                 return 0;
 784
 785         cfs_fini_completion(&lnet_acceptor_state.pta_completion);
 786         return -ENETDOWN;
 787 }
 788
 789 void
 790 lnet_acceptor_stop(void)
 791 {
 792         /* Do nothing if we're liblustre clients */
 793         if ((the_lnet.ln_pid & LNET_PID_USERFLAG) != 0)
 794                 return;
 795
 796         if (!skip_waiting_for_completion) {
 797                 lnet_acceptor_state.pta_shutdown = 1;
 798                 libcfs_sock_abort_accept(accept_port);
 799
 800                 /* block until acceptor signals exit */
 801                 cfs_wait_for_completion(&lnet_acceptor_state.pta_completion);
 802         }
 803
 804         cfs_fini_completion(&lnet_acceptor_state.pta_completion);
 805 }
 806 #else
 807 int
 808 lnet_acceptor_start(void)
 809 {
 810         return 0;
 811 }
 812
 813 void
 814 lnet_acceptor_stop(void)
 815 {
 816 }
 817 #endif /* !HAVE_LIBPTHREAD */
 818 #endif /* !__KERNEL__ */