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_cell_t *cellp = NULL;
 107
 108         if (!fidp) {
 109                 *serverspp = NULL;
 110                 return 0;
 111         }
 112
 113         cellp = cm_FindCellByID(fidp->cell);
 114         if (!cellp) return CM_ERROR_NOSUCHCELL;
 115
 116         code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp, &volp);
 117         if (code) return code;
 118
 119         if (fidp->volume == volp->rwID)
 120                 serversp = volp->rwServersp;
 121         else if (fidp->volume == volp->roID)
 122                 serversp = volp->roServersp;
 123         else if (fidp->volume == volp->bkID)
 124                 serversp = volp->bkServersp;
 125         else
 126                 serversp = NULL;
 127
 128         cm_PutVolume(volp);
 129         *serverspp = serversp;
 130         return 0;
 131 }
 132
 133 /*
 134  * Analyze the error return from an RPC.  Determine whether or not to retry,
 135  * and if we're going to retry, determine whether failover is appropriate,
 136  * and whether timed backoff is appropriate.
 137  *
 138  * If the error code is from cm_Conn() or friends, it will be a CM_ERROR code.
 139  * Otherwise it will be an RPC code.  This may be a UNIX code (e.g. EDQUOT), or
 140  * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
 141  * be a security code (e.g. RXKADEXPIRED).
 142  *
 143  * If the error code is from cm_Conn() or friends, connp will be NULL.
 144  *
 145  * For VLDB calls, fidp will be NULL.
 146  *
 147  * volSyncp and/or cbrp may also be NULL.
 148  */
 149 int
 150 cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
 151         struct cm_fid *fidp,
 152         AFSVolSync *volSyncp, cm_callbackRequest_t *cbrp, long errorCode)
 153 {
 154         cm_server_t *serverp;
 155         cm_serverRef_t *serversp, *tsrp;
 156         cm_ucell_t *ucellp;
 157         int retry = 0;
 158         int dead_session;
 159
 160         osi_Log2(afsd_logp, "cm_Analyze connp 0x%x, code %d",
 161                  (long) connp, errorCode);
 162
 163         /* no locking required, since connp->serverp never changes after
 164          * creation */
 165         dead_session = (userp->cellInfop == NULL);
 166         if (connp)
 167                 serverp = connp->serverp;
 168
 169         /* Update callback pointer */
 170         if (cbrp && errorCode == 0) cbrp->serverp = connp->serverp;
 171
 172         /* If not allowed to retry, don't */
 173         if (reqp->flags & CM_REQ_NORETRY)
 174                 goto out;
 175
 176         /* if timeout - check that it did not exceed the SMB timeout
 177            and retry */
 178         if (errorCode == CM_ERROR_TIMEDOUT)
 179     {
 180             long timeUsed, timeLeft;
 181             /* timeleft - get if from reqp the same way as cmXonnByMServers does */
 182 #ifndef DJGPP
 183             timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
 184 #else
 185             gettimeofday(&now, NULL);
 186             timeUsed = sub_time(now, reqp->startTime) / 1000;
 187 #endif
 188
 189             /* leave 5 seconds margin for sleep */
 190             timeLeft = RDRtimeout - timeUsed;
 191             if (timeLeft > 5)
 192         {
 193             thrd_Sleep(3000);
 194             cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, NULL);
 195             retry = 1;
 196         }
 197     }
 198
 199     /* if all servers are offline, mark them non-busy and start over */
 200         if (errorCode == CM_ERROR_ALLOFFLINE) {
 201             osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE.");
 202             thrd_Sleep(5000);
 203             /* cm_ForceUpdateVolume marks all servers as non_busy */
 204                 /* No it doesn't and it won't do anything if all of the
 205                  * the servers are marked as DOWN.  So clear the DOWN
 206                  * flag and reset the busy state as well.
 207                  */
 208                 cm_GetServerList(fidp, userp, reqp, &serversp);
 209         if (serversp) {
 210             lock_ObtainWrite(&cm_serverLock);
 211             for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 212                 tsrp->server->flags &= ~CM_SERVERFLAG_DOWN;
 213                 if (tsrp->status == busy)
 214                     tsrp->status = not_busy;
 215             }
 216             lock_ReleaseWrite(&cm_serverLock);
 217
 218             retry = 1;
 219         }
 220
 221         if (fidp != NULL)   /* Not a VLDB call */
 222             cm_ForceUpdateVolume(fidp, userp, reqp);
 223         }
 224
 225         /* if all servers are busy, mark them non-busy and start over */
 226         if (errorCode == CM_ERROR_ALLBUSY) {
 227                 cm_GetServerList(fidp, userp, reqp, &serversp);
 228                 lock_ObtainWrite(&cm_serverLock);
 229                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 230                         if (tsrp->status == busy)
 231                                 tsrp->status = not_busy;
 232                 }
 233         lock_ReleaseWrite(&cm_serverLock);
 234                 thrd_Sleep(5000);
 235                 retry = 1;
 236         }
 237
 238         /* special codes:  VBUSY and VRESTARTING */
 239         if (errorCode == VBUSY || errorCode == VRESTARTING) {
 240                 cm_GetServerList(fidp, userp, reqp, &serversp);
 241                 lock_ObtainWrite(&cm_serverLock);
 242                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 243                         if (tsrp->server == serverp
 244                             && tsrp->status == not_busy) {
 245                                 tsrp->status = busy;
 246                                 break;
 247                         }
 248                 }
 249         lock_ReleaseWrite(&cm_serverLock);
 250                 retry = 1;
 251         }
 252
 253         /* special codes:  missing volumes */
 254         if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE
 255             || errorCode == VSALVAGE || errorCode == VNOSERVICE) {
 256                 /* Log server being offline for this volume */
 257                 osi_Log4(afsd_logp, "cm_Analyze found server %d.%d.%d.%d marked offline for a volume",
 258                          ((serverp->addr.sin_addr.s_addr & 0xff)),
 259                          ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
 260                          ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
 261                          ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
 262                 /* Create Event Log message */
 263                 {
 264                     HANDLE h;
 265                     char *ptbuf[1];
 266                     char s[100];
 267                     h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
 268                     sprintf(s, "cm_Analyze: Server %d.%d.%d.%d reported volume %d as missing.",
 269                             ((serverp->addr.sin_addr.s_addr & 0xff)),
 270                             ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
 271                             ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
 272                             ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24),
 273                             fidp->volume);
 274                     ptbuf[0] = s;
 275                     ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
 276                                 1, 0, ptbuf, NULL);
 277                     DeregisterEventSource(h);
 278                 }
 279
 280                 /* Mark server offline for this volume */
 281                 cm_GetServerList(fidp, userp, reqp, &serversp);
 282
 283                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 284                         if (tsrp->server == serverp)
 285                                 tsrp->status = offline;
 286                 }
 287                 retry = 1;
 288         }
 289
 290         /* RX codes */
 291         if (errorCode == RX_CALL_TIMEOUT) {
 292                 /* server took longer than hardDeadTime
 293                  * don't mark server as down but don't retry
 294                  * this is to prevent the SMB session from timing out
 295                  * In addition, we log an event to the event log
 296                  */
 297 #ifndef DJGPP
 298                 HANDLE h;
 299                 char *ptbuf[1];
 300                 char s[100];
 301                 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
 302                 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
 303                 ptbuf[0] = s;
 304                 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
 305                         1, 0, ptbuf, NULL);
 306                 DeregisterEventSource(h);
 307 #endif /* !DJGPP */
 308
 309                 retry = 0;
 310                 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
 311         }
 312         else if (errorCode >= -64 && errorCode < 0) {
 313                 /* mark server as down */
 314                 lock_ObtainMutex(&serverp->mx);
 315                 serverp->flags |= CM_SERVERFLAG_DOWN;
 316                 lock_ReleaseMutex(&serverp->mx);
 317                 retry = 1;
 318         }
 319
 320         if (errorCode == RXKADEXPIRED && !dead_session) {
 321                 lock_ObtainMutex(&userp->mx);
 322                 ucellp = cm_GetUCell(userp, serverp->cellp);
 323                 if (ucellp->ticketp) {
 324                         free(ucellp->ticketp);
 325                         ucellp->ticketp = NULL;
 326                 }
 327                 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
 328                 ucellp->gen++;
 329                 lock_ReleaseMutex(&userp->mx);
 330                 retry = 1;
 331         }
 332
 333         if (retry && dead_session)
 334                 retry = 0;
 335
 336 out:
 337         /* drop this on the way out */
 338         if (connp)
 339                 cm_PutConn(connp);
 340
 341         /* retry until we fail to find a connection */
 342         return retry;
 343 }
 344
 345 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
 346         cm_req_t *reqp, cm_conn_t **connpp)
 347 {
 348         long code;
 349         cm_serverRef_t *tsrp;
 350         cm_server_t *tsp;
 351         long firstError = 0;
 352         int someBusy = 0, someOffline = 0, allBusy = 1, allDown = 1;
 353         long timeUsed, timeLeft, hardTimeLeft;
 354 #ifdef DJGPP
 355         struct timeval now;
 356 #endif /* DJGPP */
 357
 358         *connpp = NULL;
 359
 360 #ifndef DJGPP
 361         timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
 362 #else
 363         gettimeofday(&now, NULL);
 364         timeUsed = sub_time(now, reqp->startTime) / 1000;
 365 #endif
 366
 367         /* leave 5 seconds margin of safety */
 368         timeLeft =  ConnDeadtimeout - timeUsed - 5;
 369         hardTimeLeft = HardDeadtimeout - timeUsed - 5;
 370
 371         lock_ObtainWrite(&cm_serverLock);
 372
 373     for(tsrp = serversp; tsrp; tsrp=tsrp->next) {
 374         tsp = tsrp->server;
 375         tsp->refCount++;
 376         lock_ReleaseWrite(&cm_serverLock);
 377         if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
 378             allDown = 0;
 379             if (tsrp->status == busy)
 380                 someBusy = 1;
 381             else if (tsrp->status == offline)
 382                 someOffline = 1;
 383             else {
 384                                 allBusy = 0;
 385                 code = cm_ConnByServer(tsp, usersp, connpp);
 386                 if (code == 0) {
 387                     cm_PutServer(tsp);
 388                     /* Set RPC timeout */
 389                     if (timeLeft > ConnDeadtimeout)
 390                         timeLeft = ConnDeadtimeout;
 391
 392                     if (hardTimeLeft > HardDeadtimeout)
 393                         hardTimeLeft = HardDeadtimeout;
 394
 395                     lock_ObtainMutex(&(*connpp)->mx);
 396                     rx_SetConnDeadTime((*connpp)->callp,
 397                                         timeLeft);
 398                     rx_SetConnHardDeadTime((*connpp)->callp,
 399                                             (u_short) hardTimeLeft);
 400                     lock_ReleaseMutex(&(*connpp)->mx);
 401
 402                     return 0;
 403                 }
 404                 if (firstError == 0)
 405                     firstError = code;
 406             }
 407                 }
 408         lock_ObtainWrite(&cm_serverLock);
 409         osi_assert(tsp->refCount-- > 0);
 410     }
 411
 412         lock_ReleaseWrite(&cm_serverLock);
 413         if (firstError == 0) {
 414         if (serversp == NULL)
 415                         firstError = CM_ERROR_NOSUCHVOLUME;
 416         else if (allDown)
 417                         firstError = CM_ERROR_ALLOFFLINE;
 418                 else if (allBusy)
 419                         firstError = CM_ERROR_ALLBUSY;
 420                 else
 421                         firstError = CM_ERROR_TIMEDOUT;
 422         }
 423         osi_Log1(afsd_logp, "cm_ConnByMServers returning %x", firstError);
 424     return firstError;
 425 }
 426
 427 /* called with a held server to GC all bad connections hanging off of the server */
 428 void cm_GCConnections(cm_server_t *serverp)
 429 {
 430         cm_conn_t *tcp;
 431     cm_conn_t **lcpp;
 432     cm_user_t *userp;
 433
 434         lock_ObtainWrite(&cm_connLock);
 435         lcpp = &serverp->connsp;
 436         for(tcp = *lcpp; tcp; tcp = *lcpp) {
 437                 userp = tcp->userp;
 438                 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
 439                         /* do the deletion of this guy */
 440             cm_ReleaseUser(userp);
 441             *lcpp = tcp->nextp;
 442                         rx_DestroyConnection(tcp->callp);
 443             lock_FinalizeMutex(&tcp->mx);
 444             free(tcp);
 445         }
 446         else {
 447                         /* just advance to the next */
 448             lcpp = &tcp->nextp;
 449         }
 450     }
 451         lock_ReleaseWrite(&cm_connLock);
 452 }
 453
 454 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
 455         cm_server_t *serverp)
 456 {
 457     unsigned short port;
 458     int serviceID;
 459     int secIndex;
 460     struct rx_securityClass *secObjp;
 461         afs_int32 level;
 462
 463         if (serverp->type == CM_SERVER_VLDB) {
 464                 port = htons(7003);
 465         serviceID = 52;
 466     }
 467     else {
 468                 osi_assert(serverp->type == CM_SERVER_FILE);
 469         port = htons(7000);
 470         serviceID = 1;
 471     }
 472         if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
 473                 secIndex = 2;
 474                 if (cryptall) {
 475                         level = rxkad_crypt;
 476                         tcp->cryptlevel = rxkad_crypt;
 477                 } else {
 478                         level = rxkad_clear;
 479                 }
 480         secObjp = rxkad_NewClientSecurityObject(level,
 481                                                 &ucellp->sessionKey, ucellp->kvno,
 482                                                 ucellp->ticketLen, ucellp->ticketp);
 483     }
 484     else {
 485         /* normal auth */
 486         secIndex = 0;
 487         secObjp = rxnull_NewClientSecurityObject();
 488     }
 489         osi_assert(secObjp != NULL);
 490     tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
 491                                   port,
 492                                   serviceID,
 493                                   secObjp,
 494                                   secIndex);
 495         rx_SetConnDeadTime(tcp->callp, ConnDeadtimeout);
 496         rx_SetConnHardDeadTime(tcp->callp, HardDeadtimeout);
 497         tcp->ucgen = ucellp->gen;
 498     if (secObjp)
 499         rxs_Release(secObjp);   /* Decrement the initial refCount */
 500 }
 501
 502 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
 503 {
 504         cm_conn_t *tcp;
 505     cm_ucell_t *ucellp;
 506
 507         lock_ObtainMutex(&userp->mx);
 508         lock_ObtainWrite(&cm_connLock);
 509         for(tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
 510                 if (tcp->userp == userp) break;
 511     }
 512         /* find ucell structure */
 513     ucellp = cm_GetUCell(userp, serverp->cellp);
 514         if (!tcp) {
 515                 tcp = malloc(sizeof(*tcp));
 516         memset(tcp, 0, sizeof(*tcp));
 517         tcp->nextp = serverp->connsp;
 518         serverp->connsp = tcp;
 519         cm_HoldUser(userp);
 520         tcp->userp = userp;
 521         lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
 522         tcp->serverp = serverp;
 523                 tcp->cryptlevel = rxkad_clear;
 524                 cm_NewRXConnection(tcp, ucellp, serverp);
 525                 tcp->refCount = 1;
 526     }
 527         else {
 528                 if ((tcp->ucgen < ucellp->gen) || (tcp->cryptlevel != cryptall))
 529                 {
 530                         rx_DestroyConnection(tcp->callp);
 531                         cm_NewRXConnection(tcp, ucellp, serverp);
 532                 }
 533         tcp->refCount++;
 534         }
 535         lock_ReleaseWrite(&cm_connLock);
 536     lock_ReleaseMutex(&userp->mx);
 537
 538         /* return this pointer to our caller */
 539     osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
 540         *connpp = tcp;
 541
 542     return 0;
 543 }
 544
 545 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
 546         cm_conn_t **connpp)
 547 {
 548         long code;
 549
 550         cm_serverRef_t *serversp;
 551
 552         code = cm_GetServerList(fidp, userp, reqp, &serversp);
 553         if (code) {
 554                 *connpp = NULL;
 555                 return code;
 556         }
 557
 558         code = cm_ConnByMServers(serversp, userp, reqp, connpp);
 559         return code;
 560 }