src/WINNT/afsd/cm_conn.c

   1 /*
   2  * Copyright 2000, International Business Machines Corporation and others.
   3  * All Rights Reserved.
   4  *
   5  * This software has been released under the terms of the IBM Public
   6  * License.  For details, see the LICENSE file in the top-level source
   7  * directory or online at http://www.openafs.org/dl/license10.html
   8  */
   9
  10 #include <afs/param.h>
  11 #include <afs/stds.h>
  12
  13 #ifndef DJGPP
  14 #include <windows.h>
  15 #endif /* !DJGPP */
  16 #include <string.h>
  17 #include <malloc.h>
  18 #include <osi.h>
  19 #include <rx/rx.h>
  20 #ifndef DJGPP
  21 #include <rx/rxkad.h>
  22 #else
  23 #include <rx/rxkad.h>
  24 #endif
  25
  26 #include "afsd.h"
  27
  28 osi_rwlock_t cm_connLock;
  29
  30 long RDRtimeout = CM_CONN_DEFAULTRDRTIMEOUT;
  31 long ConnDeadtimeout = CM_CONN_CONNDEADTIME;
  32 long HardDeadtimeout = CM_CONN_HARDDEADTIME;
  33
  34 #define LANMAN_WKS_PARAM_KEY "SYSTEM\\CurrentControlSet\\Services\\lanmanworkstation\\parameters"
  35 #define LANMAN_WKS_SESSION_TIMEOUT "SessTimeout"
  36
  37 afs_int32 cryptall = 0;
  38
  39 void cm_PutConn(cm_conn_t *connp)
  40 {
  41         lock_ObtainWrite(&cm_connLock);
  42         osi_assert(connp->refCount-- > 0);
  43         lock_ReleaseWrite(&cm_connLock);
  44 }
  45
  46 void cm_InitConn(void)
  47 {
  48         static osi_once_t once;
  49         long code;
  50         DWORD sessTimeout;
  51         HKEY parmKey;
  52
  53     if (osi_Once(&once)) {
  54                 lock_InitializeRWLock(&cm_connLock, "connection global lock");
  55
  56         /* keisa - read timeout value for lanmanworkstation  service.
  57          * jaltman - as per
  58          *   http://support.microsoft.com:80/support/kb/articles/Q102/0/67.asp&NoWebContent=1
  59          * the SessTimeout is a minimum timeout not a maximum timeout.  Therefore,
  60          * I believe that the default should not be short.  Instead, we should wait until
  61          * RX times out before reporting a timeout to the SMB client.
  62          */
  63                 code = RegOpenKeyEx(HKEY_LOCAL_MACHINE, LANMAN_WKS_PARAM_KEY,
  64                             0, KEY_QUERY_VALUE, &parmKey);
  65                 if (code == ERROR_SUCCESS)
  66         {
  67                     DWORD dummyLen = sizeof(sessTimeout);
  68                     code = RegQueryValueEx(parmKey, LANMAN_WKS_SESSION_TIMEOUT, NULL, NULL,
  69                                    (BYTE *) &sessTimeout, &dummyLen);
  70                     if (code == ERROR_SUCCESS)
  71             {
  72                 afsi_log("lanmanworkstation : SessTimeout %d", sessTimeout);
  73                 RDRtimeout = sessTimeout;
  74                 if ( ConnDeadtimeout < RDRtimeout + 15 ) {
  75                     ConnDeadtimeout = RDRtimeout + 15;
  76                     afsi_log("ConnDeadTimeout increased to %d", ConnDeadtimeout);
  77                 }
  78                 if ( HardDeadtimeout < 2 * ConnDeadtimeout ) {
  79                     HardDeadtimeout = 2 * ConnDeadtimeout;
  80                     afsi_log("HardDeadTimeout increased to %d", HardDeadtimeout);
  81                 }
  82             }
  83         }
  84
  85         osi_EndOnce(&once);
  86     }
  87 }
  88
  89 void cm_InitReq(cm_req_t *reqp)
  90 {
  91         memset((char *)reqp, 0, sizeof(cm_req_t));
  92 #ifndef DJGPP
  93         reqp->startTime = GetCurrentTime();
  94 #else
  95         gettimeofday(&reqp->startTime, NULL);
  96 #endif
  97
  98 }
  99
 100 long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
 101         struct cm_req *reqp, cm_serverRef_t **serverspp)
 102 {
 103         long code;
 104         cm_volume_t *volp = NULL;
 105         cm_serverRef_t *serversp = NULL;
 106     cm_serverRef_t *newServersp = NULL;
 107         cm_cell_t *cellp = NULL;
 108
 109         if (!fidp) {
 110                 *serverspp = NULL;
 111                 return 0;
 112         }
 113
 114         cellp = cm_FindCellByID(fidp->cell);
 115         if (!cellp) return CM_ERROR_NOSUCHCELL;
 116
 117         code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp, &volp);
 118         if (code) return code;
 119
 120     lock_ObtainMutex(&volp->mx);
 121         if (fidp->volume == volp->rwID)
 122                 serversp = volp->rwServersp;
 123         else if (fidp->volume == volp->roID)
 124                 serversp = volp->roServersp;
 125         else if (fidp->volume == volp->bkID)
 126                 serversp = volp->bkServersp;
 127         else
 128                 serversp = NULL;
 129
 130     /* make a copy of the server list because by the time the
 131     caller tries to use it, it might have been freed.
 132     Preserve server order. */
 133     if(serversp) {
 134         cm_serverRef_t ** nl;
 135         cm_serverRef_t * tref;
 136
 137         nl = &newServersp;
 138
 139         lock_ObtainWrite(&cm_serverLock);
 140         while(serversp) {
 141
 142             tref = malloc(sizeof(cm_serverRef_t));
 143             tref->next = NULL;
 144             tref->server = serversp->server;
 145             tref->status = serversp->status;
 146
 147             tref->server->refCount++;
 148
 149             *nl = tref;
 150             nl = &tref->next;
 151
 152             serversp = serversp->next;
 153         }
 154         lock_ReleaseWrite(&cm_serverLock);
 155     }
 156     lock_ReleaseMutex(&volp->mx);
 157
 158         cm_PutVolume(volp);
 159     *serverspp = newServersp;
 160         return 0;
 161 }
 162
 163 /*
 164  * Analyze the error return from an RPC.  Determine whether or not to retry,
 165  * and if we're going to retry, determine whether failover is appropriate,
 166  * and whether timed backoff is appropriate.
 167  *
 168  * If the error code is from cm_Conn() or friends, it will be a CM_ERROR code.
 169  * Otherwise it will be an RPC code.  This may be a UNIX code (e.g. EDQUOT), or
 170  * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
 171  * be a security code (e.g. RXKADEXPIRED).
 172  *
 173  * If the error code is from cm_Conn() or friends, connp will be NULL.
 174  *
 175  * For VLDB calls, fidp will be NULL.
 176  *
 177  * volSyncp and/or cbrp may also be NULL.
 178  */
 179 int
 180 cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
 181         struct cm_fid *fidp,
 182         AFSVolSync *volSyncp, cm_callbackRequest_t *cbrp, long errorCode)
 183 {
 184         cm_server_t *serverp;
 185         cm_serverRef_t *serversp, *tsrp;
 186         cm_ucell_t *ucellp;
 187         int retry = 0;
 188         int dead_session;
 189
 190         osi_Log2(afsd_logp, "cm_Analyze connp 0x%x, code %d",
 191                  (long) connp, errorCode);
 192
 193         /* no locking required, since connp->serverp never changes after
 194          * creation */
 195         dead_session = (userp->cellInfop == NULL);
 196         if (connp)
 197                 serverp = connp->serverp;
 198
 199         /* Update callback pointer */
 200         if (cbrp && errorCode == 0) cbrp->serverp = connp->serverp;
 201
 202         /* If not allowed to retry, don't */
 203         if (reqp->flags & CM_REQ_NORETRY)
 204                 goto out;
 205
 206         /* if timeout - check that it did not exceed the SMB timeout
 207            and retry */
 208         if (errorCode == CM_ERROR_TIMEDOUT)
 209     {
 210             long timeUsed, timeLeft;
 211             /* timeleft - get if from reqp the same way as cmXonnByMServers does */
 212 #ifndef DJGPP
 213             timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
 214 #else
 215             gettimeofday(&now, NULL);
 216             timeUsed = sub_time(now, reqp->startTime) / 1000;
 217 #endif
 218
 219             /* leave 5 seconds margin for sleep */
 220             timeLeft = RDRtimeout - timeUsed;
 221             if (timeLeft > 5)
 222         {
 223             thrd_Sleep(3000);
 224             cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, NULL);
 225             retry = 1;
 226         }
 227     }
 228
 229     /* if all servers are offline, mark them non-busy and start over */
 230         if (errorCode == CM_ERROR_ALLOFFLINE) {
 231             osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE.");
 232             thrd_Sleep(5000);
 233             /* cm_ForceUpdateVolume marks all servers as non_busy */
 234                 /* No it doesn't and it won't do anything if all of the
 235                  * the servers are marked as DOWN.  So clear the DOWN
 236                  * flag and reset the busy state as well.
 237                  */
 238                 cm_GetServerList(fidp, userp, reqp, &serversp);
 239         if (serversp) {
 240             lock_ObtainWrite(&cm_serverLock);
 241             for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 242                 tsrp->server->flags &= ~CM_SERVERFLAG_DOWN;
 243                 if (tsrp->status == busy)
 244                     tsrp->status = not_busy;
 245             }
 246             lock_ReleaseWrite(&cm_serverLock);
 247
 248             retry = 1;
 249         }
 250         cm_FreeServerList(&serversp);
 251
 252         if (fidp != NULL)   /* Not a VLDB call */
 253             cm_ForceUpdateVolume(fidp, userp, reqp);
 254         }
 255
 256         /* if all servers are busy, mark them non-busy and start over */
 257         if (errorCode == CM_ERROR_ALLBUSY) {
 258                 cm_GetServerList(fidp, userp, reqp, &serversp);
 259                 lock_ObtainWrite(&cm_serverLock);
 260                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 261                         if (tsrp->status == busy)
 262                                 tsrp->status = not_busy;
 263                 }
 264         lock_ReleaseWrite(&cm_serverLock);
 265         cm_FreeServerList(&serversp);
 266                 thrd_Sleep(5000);
 267                 retry = 1;
 268         }
 269
 270         /* special codes:  VBUSY and VRESTARTING */
 271         if (errorCode == VBUSY || errorCode == VRESTARTING) {
 272                 cm_GetServerList(fidp, userp, reqp, &serversp);
 273                 lock_ObtainWrite(&cm_serverLock);
 274                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 275                         if (tsrp->server == serverp
 276                             && tsrp->status == not_busy) {
 277                                 tsrp->status = busy;
 278                                 break;
 279                         }
 280                 }
 281         lock_ReleaseWrite(&cm_serverLock);
 282         cm_FreeServerList(&serversp);
 283                 retry = 1;
 284         }
 285
 286         /* special codes:  missing volumes */
 287         if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE
 288             || errorCode == VSALVAGE || errorCode == VNOSERVICE) {
 289                 /* Log server being offline for this volume */
 290                 osi_Log4(afsd_logp, "cm_Analyze found server %d.%d.%d.%d marked offline for a volume",
 291                          ((serverp->addr.sin_addr.s_addr & 0xff)),
 292                          ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
 293                          ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
 294                          ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
 295                 /* Create Event Log message */
 296                 {
 297                     HANDLE h;
 298                     char *ptbuf[1];
 299                     char s[100];
 300                     h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
 301                     sprintf(s, "cm_Analyze: Server %d.%d.%d.%d reported volume %d as missing.",
 302                             ((serverp->addr.sin_addr.s_addr & 0xff)),
 303                             ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
 304                             ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
 305                             ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24),
 306                             fidp->volume);
 307                     ptbuf[0] = s;
 308                     ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
 309                                 1, 0, ptbuf, NULL);
 310                     DeregisterEventSource(h);
 311                 }
 312
 313                 /* Mark server offline for this volume */
 314                 cm_GetServerList(fidp, userp, reqp, &serversp);
 315                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 316                         if (tsrp->server == serverp)
 317                                 tsrp->status = offline;
 318                 }
 319         cm_FreeServerList(&serversp);
 320                 retry = 1;
 321         }
 322
 323         /* RX codes */
 324         if (errorCode == RX_CALL_TIMEOUT) {
 325                 /* server took longer than hardDeadTime
 326                  * don't mark server as down but don't retry
 327                  * this is to prevent the SMB session from timing out
 328                  * In addition, we log an event to the event log
 329                  */
 330 #ifndef DJGPP
 331                 HANDLE h;
 332                 char *ptbuf[1];
 333                 char s[100];
 334                 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
 335                 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
 336                 ptbuf[0] = s;
 337                 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
 338                         1, 0, ptbuf, NULL);
 339                 DeregisterEventSource(h);
 340 #endif /* !DJGPP */
 341
 342                 retry = 0;
 343                 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
 344         }
 345         else if (errorCode >= -64 && errorCode < 0) {
 346                 /* mark server as down */
 347                 lock_ObtainMutex(&serverp->mx);
 348                 serverp->flags |= CM_SERVERFLAG_DOWN;
 349                 lock_ReleaseMutex(&serverp->mx);
 350                 retry = 1;
 351         }
 352
 353         if (errorCode == RXKADEXPIRED && !dead_session) {
 354                 lock_ObtainMutex(&userp->mx);
 355                 ucellp = cm_GetUCell(userp, serverp->cellp);
 356                 if (ucellp->ticketp) {
 357                         free(ucellp->ticketp);
 358                         ucellp->ticketp = NULL;
 359                 }
 360                 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
 361                 ucellp->gen++;
 362                 lock_ReleaseMutex(&userp->mx);
 363                 retry = 1;
 364         }
 365
 366         if (retry && dead_session)
 367                 retry = 0;
 368
 369 out:
 370         /* drop this on the way out */
 371         if (connp)
 372                 cm_PutConn(connp);
 373
 374         /* retry until we fail to find a connection */
 375         return retry;
 376 }
 377
 378 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
 379         cm_req_t *reqp, cm_conn_t **connpp)
 380 {
 381         long code;
 382         cm_serverRef_t *tsrp;
 383         cm_server_t *tsp;
 384         long firstError = 0;
 385         int someBusy = 0, someOffline = 0, allBusy = 1, allDown = 1;
 386         long timeUsed, timeLeft, hardTimeLeft;
 387 #ifdef DJGPP
 388         struct timeval now;
 389 #endif /* DJGPP */
 390
 391         *connpp = NULL;
 392
 393 #ifndef DJGPP
 394         timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
 395 #else
 396         gettimeofday(&now, NULL);
 397         timeUsed = sub_time(now, reqp->startTime) / 1000;
 398 #endif
 399
 400         /* leave 5 seconds margin of safety */
 401         timeLeft =  ConnDeadtimeout - timeUsed - 5;
 402         hardTimeLeft = HardDeadtimeout - timeUsed - 5;
 403
 404         lock_ObtainWrite(&cm_serverLock);
 405
 406     for(tsrp = serversp; tsrp; tsrp=tsrp->next) {
 407         tsp = tsrp->server;
 408         tsp->refCount++;
 409         lock_ReleaseWrite(&cm_serverLock);
 410         if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
 411             allDown = 0;
 412             if (tsrp->status == busy)
 413                 someBusy = 1;
 414             else if (tsrp->status == offline)
 415                 someOffline = 1;
 416             else {
 417                                 allBusy = 0;
 418                 code = cm_ConnByServer(tsp, usersp, connpp);
 419                 if (code == 0) {
 420                     cm_PutServer(tsp);
 421                     /* Set RPC timeout */
 422                     if (timeLeft > ConnDeadtimeout)
 423                         timeLeft = ConnDeadtimeout;
 424
 425                     if (hardTimeLeft > HardDeadtimeout)
 426                         hardTimeLeft = HardDeadtimeout;
 427
 428                     lock_ObtainMutex(&(*connpp)->mx);
 429                     rx_SetConnDeadTime((*connpp)->callp,
 430                                         timeLeft);
 431                     rx_SetConnHardDeadTime((*connpp)->callp,
 432                                             (u_short) hardTimeLeft);
 433                     lock_ReleaseMutex(&(*connpp)->mx);
 434
 435                     return 0;
 436                 }
 437                 if (firstError == 0)
 438                     firstError = code;
 439             }
 440                 }
 441         lock_ObtainWrite(&cm_serverLock);
 442         osi_assert(tsp->refCount-- > 0);
 443     }
 444
 445         lock_ReleaseWrite(&cm_serverLock);
 446         if (firstError == 0) {
 447         if (serversp == NULL)
 448                         firstError = CM_ERROR_NOSUCHVOLUME;
 449         else if (allDown)
 450                         firstError = CM_ERROR_ALLOFFLINE;
 451                 else if (allBusy)
 452                         firstError = CM_ERROR_ALLBUSY;
 453                 else
 454                         firstError = CM_ERROR_TIMEDOUT;
 455         }
 456         osi_Log1(afsd_logp, "cm_ConnByMServers returning %x", firstError);
 457     return firstError;
 458 }
 459
 460 /* called with a held server to GC all bad connections hanging off of the server */
 461 void cm_GCConnections(cm_server_t *serverp)
 462 {
 463         cm_conn_t *tcp;
 464     cm_conn_t **lcpp;
 465     cm_user_t *userp;
 466
 467         lock_ObtainWrite(&cm_connLock);
 468         lcpp = &serverp->connsp;
 469         for(tcp = *lcpp; tcp; tcp = *lcpp) {
 470                 userp = tcp->userp;
 471                 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
 472                         /* do the deletion of this guy */
 473             cm_ReleaseUser(userp);
 474             *lcpp = tcp->nextp;
 475                         rx_DestroyConnection(tcp->callp);
 476             lock_FinalizeMutex(&tcp->mx);
 477             free(tcp);
 478         }
 479         else {
 480                         /* just advance to the next */
 481             lcpp = &tcp->nextp;
 482         }
 483     }
 484         lock_ReleaseWrite(&cm_connLock);
 485 }
 486
 487 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
 488         cm_server_t *serverp)
 489 {
 490     unsigned short port;
 491     int serviceID;
 492     int secIndex;
 493     struct rx_securityClass *secObjp;
 494         afs_int32 level;
 495
 496         if (serverp->type == CM_SERVER_VLDB) {
 497                 port = htons(7003);
 498         serviceID = 52;
 499     }
 500     else {
 501                 osi_assert(serverp->type == CM_SERVER_FILE);
 502         port = htons(7000);
 503         serviceID = 1;
 504     }
 505         if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
 506                 secIndex = 2;
 507                 if (cryptall) {
 508                         level = rxkad_crypt;
 509                         tcp->cryptlevel = rxkad_crypt;
 510                 } else {
 511                         level = rxkad_clear;
 512                 }
 513         secObjp = rxkad_NewClientSecurityObject(level,
 514                                                 &ucellp->sessionKey, ucellp->kvno,
 515                                                 ucellp->ticketLen, ucellp->ticketp);
 516     }
 517     else {
 518         /* normal auth */
 519         secIndex = 0;
 520         secObjp = rxnull_NewClientSecurityObject();
 521     }
 522         osi_assert(secObjp != NULL);
 523     tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
 524                                   port,
 525                                   serviceID,
 526                                   secObjp,
 527                                   secIndex);
 528         rx_SetConnDeadTime(tcp->callp, ConnDeadtimeout);
 529         rx_SetConnHardDeadTime(tcp->callp, HardDeadtimeout);
 530         tcp->ucgen = ucellp->gen;
 531     if (secObjp)
 532         rxs_Release(secObjp);   /* Decrement the initial refCount */
 533 }
 534
 535 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
 536 {
 537         cm_conn_t *tcp;
 538     cm_ucell_t *ucellp;
 539
 540         lock_ObtainMutex(&userp->mx);
 541         lock_ObtainWrite(&cm_connLock);
 542         for(tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
 543                 if (tcp->userp == userp) break;
 544     }
 545         /* find ucell structure */
 546     ucellp = cm_GetUCell(userp, serverp->cellp);
 547         if (!tcp) {
 548                 tcp = malloc(sizeof(*tcp));
 549         memset(tcp, 0, sizeof(*tcp));
 550         tcp->nextp = serverp->connsp;
 551         serverp->connsp = tcp;
 552         cm_HoldUser(userp);
 553         tcp->userp = userp;
 554         lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
 555         tcp->serverp = serverp;
 556                 tcp->cryptlevel = rxkad_clear;
 557                 cm_NewRXConnection(tcp, ucellp, serverp);
 558                 tcp->refCount = 1;
 559     }
 560         else {
 561                 if ((tcp->ucgen < ucellp->gen) || (tcp->cryptlevel != cryptall))
 562                 {
 563                         rx_DestroyConnection(tcp->callp);
 564                         cm_NewRXConnection(tcp, ucellp, serverp);
 565                 }
 566         tcp->refCount++;
 567         }
 568         lock_ReleaseWrite(&cm_connLock);
 569     lock_ReleaseMutex(&userp->mx);
 570
 571         /* return this pointer to our caller */
 572     osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
 573         *connpp = tcp;
 574
 575     return 0;
 576 }
 577
 578 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
 579         cm_conn_t **connpp)
 580 {
 581         long code;
 582
 583         cm_serverRef_t *serversp;
 584
 585         code = cm_GetServerList(fidp, userp, reqp, &serversp);
 586         if (code) {
 587                 *connpp = NULL;
 588                 return code;
 589         }
 590
 591         code = cm_ConnByMServers(serversp, userp, reqp, connpp);
 592     cm_FreeServerList(&serversp);
 593         return code;
 594 }