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
  32 afs_int32 cryptall = 0;
  33
  34 void cm_PutConn(cm_conn_t *connp)
  35 {
  36         lock_ObtainWrite(&cm_connLock);
  37         osi_assert(connp->refCount-- > 0);
  38         lock_ReleaseWrite(&cm_connLock);
  39 }
  40
  41 void cm_InitConn(void)
  42 {
  43         static osi_once_t once;
  44
  45         if (osi_Once(&once)) {
  46                 lock_InitializeRWLock(&cm_connLock, "connection global lock");
  47                 osi_EndOnce(&once);
  48         }
  49 }
  50
  51 void cm_InitReq(cm_req_t *reqp)
  52 {
  53         memset((char *)reqp, 0, sizeof(cm_req_t));
  54 #ifndef DJGPP
  55         reqp->startTime = GetCurrentTime();
  56 #else
  57         gettimeofday(&reqp->startTime, NULL);
  58 #endif
  59
  60 }
  61
  62 long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
  63         struct cm_req *reqp, cm_serverRef_t **serverspp)
  64 {
  65         long code;
  66         cm_volume_t *volp = NULL;
  67         cm_serverRef_t *serversp = NULL;
  68         cm_cell_t *cellp = NULL;
  69
  70         if (!fidp) {
  71                 *serverspp = NULL;
  72                 return 0;
  73         }
  74
  75         cellp = cm_FindCellByID(fidp->cell);
  76         if (!cellp) return CM_ERROR_NOSUCHCELL;
  77
  78         code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp, &volp);
  79         if (code) return code;
  80
  81         if (fidp->volume == volp->rwID)
  82                 serversp = volp->rwServersp;
  83         else if (fidp->volume == volp->roID)
  84                 serversp = volp->roServersp;
  85         else if (fidp->volume == volp->bkID)
  86                 serversp = volp->bkServersp;
  87         else
  88                 serversp = NULL;
  89
  90         cm_PutVolume(volp);
  91         *serverspp = serversp;
  92         return 0;
  93 }
  94
  95 /*
  96  * Analyze the error return from an RPC.  Determine whether or not to retry,
  97  * and if we're going to retry, determine whether failover is appropriate,
  98  * and whether timed backoff is appropriate.
  99  *
 100  * If the error code is from cm_Conn() or friends, it will be a CM_ERROR code.
 101  * Otherwise it will be an RPC code.  This may be a UNIX code (e.g. EDQUOT), or
 102  * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
 103  * be a security code (e.g. RXKADEXPIRED).
 104  *
 105  * If the error code is from cm_Conn() or friends, connp will be NULL.
 106  *
 107  * For VLDB calls, fidp will be NULL.
 108  *
 109  * volSyncp and/or cbrp may also be NULL.
 110  */
 111 cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
 112         struct cm_fid *fidp,
 113         AFSVolSync *volSyncp, cm_callbackRequest_t *cbrp, long errorCode)
 114 {
 115         cm_server_t *serverp;
 116         cm_serverRef_t *serversp, *tsrp;
 117         cm_ucell_t *ucellp;
 118         int retry = 0;
 119         int dead_session;
 120
 121         osi_Log2(afsd_logp, "cm_Analyze connp 0x%x, code %d",
 122                  (long) connp, errorCode);
 123
 124         /* no locking required, since connp->serverp never changes after
 125          * creation */
 126         dead_session = (userp->cellInfop == NULL);
 127         if (connp)
 128                 serverp = connp->serverp;
 129
 130         /* Update callback pointer */
 131         if (cbrp && errorCode == 0) cbrp->serverp = connp->serverp;
 132
 133         /* If not allowed to retry, don't */
 134         if (reqp->flags & CM_REQ_NORETRY)
 135                 goto out;
 136
 137         /* if all servers are busy, mark them non-busy and start over */
 138         if (errorCode == CM_ERROR_ALLBUSY) {
 139                 cm_GetServerList(fidp, userp, reqp, &serversp);
 140                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 141                         if (tsrp->status == busy)
 142                                 tsrp->status = not_busy;
 143                 }
 144                 thrd_Sleep(5000);
 145                 retry = 1;
 146         }
 147
 148         /* special codes:  VBUSY and VRESTARTING */
 149         if (errorCode == VBUSY || errorCode == VRESTARTING) {
 150                 cm_GetServerList(fidp, userp, reqp, &serversp);
 151                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 152                         if (tsrp->server == serverp
 153                             && tsrp->status == not_busy) {
 154                                 tsrp->status = busy;
 155                                 break;
 156                         }
 157                 }
 158                 retry = 1;
 159         }
 160
 161         /* special codes:  missing volumes */
 162         if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE
 163             || errorCode == VSALVAGE || errorCode == VNOSERVICE) {
 164                 long oldSum, newSum;
 165                 int same;
 166
 167                 /* Back off to allow move to complete */
 168                 thrd_Sleep(2000);
 169
 170                 /* Update the volume location and see if it changed */
 171                 cm_GetServerList(fidp, userp, reqp, &serversp);
 172                 oldSum = cm_ChecksumServerList(serversp);
 173                 cm_ForceUpdateVolume(fidp, userp, reqp);
 174                 cm_GetServerList(fidp, userp, reqp, &serversp);
 175                 newSum = cm_ChecksumServerList(serversp);
 176                 same = (oldSum == newSum);
 177
 178                 /* mark servers as appropriate */
 179                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 180                         if (tsrp->server == serverp)
 181                                 tsrp->status = offline;
 182                         else if (!same)
 183                                 tsrp->status = not_busy;
 184                 }
 185                 retry = 1;
 186         }
 187
 188         /* RX codes */
 189         if (errorCode == RX_CALL_TIMEOUT) {
 190                 /* server took longer than hardDeadTime
 191                  * don't mark server as down but don't retry
 192                  * this is to prevent the SMB session from timing out
 193                  * In addition, we log an event to the event log
 194                  */
 195 #ifndef DJGPP
 196                 HANDLE h;
 197                 char *ptbuf[1];
 198                 char s[100];
 199                 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
 200                 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
 201                 ptbuf[0] = s;
 202                 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
 203                         1, 0, ptbuf, NULL);
 204                 DeregisterEventSource(h);
 205 #endif /* !DJGPP */
 206
 207                 retry = 0;
 208                 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
 209         }
 210         else if (errorCode >= -64 && errorCode < 0) {
 211                 /* mark server as down */
 212                 lock_ObtainMutex(&serverp->mx);
 213                 serverp->flags |= CM_SERVERFLAG_DOWN;
 214                 lock_ReleaseMutex(&serverp->mx);
 215                 retry = 1;
 216         }
 217
 218         if (errorCode == RXKADEXPIRED && !dead_session) {
 219                 lock_ObtainMutex(&userp->mx);
 220                 ucellp = cm_GetUCell(userp, serverp->cellp);
 221                 if (ucellp->ticketp) {
 222                         free(ucellp->ticketp);
 223                         ucellp->ticketp = NULL;
 224                 }
 225                 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
 226                 ucellp->gen++;
 227                 lock_ReleaseMutex(&userp->mx);
 228                 retry = 1;
 229         }
 230
 231         if (retry && dead_session)
 232                 retry = 0;
 233
 234 out:
 235         /* drop this on the way out */
 236         if (connp)
 237                 cm_PutConn(connp);
 238
 239         /* retry until we fail to find a connection */
 240         return retry;
 241 }
 242
 243 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
 244         cm_req_t *reqp, cm_conn_t **connpp)
 245 {
 246         long code;
 247         cm_serverRef_t *tsrp;
 248         cm_server_t *tsp;
 249         long firstError = 0;
 250         int someBusy = 0, someOffline = 0;
 251         long timeUsed, timeLeft, hardTimeLeft;
 252 #ifdef DJGPP
 253         struct timeval now;
 254 #endif /* DJGPP */
 255
 256         *connpp = NULL;
 257
 258 #ifndef DJGPP
 259         timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
 260 #else
 261         gettimeofday(&now, NULL);
 262         timeUsed = sub_time(now, reqp->startTime) / 1000;
 263 #endif
 264
 265         /* leave 5 seconds margin of safety */
 266         timeLeft = RDRtimeout - timeUsed - 5;
 267         hardTimeLeft = timeLeft;
 268
 269         /* Time enough to do an RPC? */
 270         if (timeLeft < 1) {
 271                 return CM_ERROR_TIMEDOUT;
 272         }
 273
 274         lock_ObtainWrite(&cm_serverLock);
 275
 276         for(tsrp = serversp; tsrp; tsrp=tsrp->next) {
 277                 tsp = tsrp->server;
 278                 tsp->refCount++;
 279                 lock_ReleaseWrite(&cm_serverLock);
 280                 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
 281                         if (tsrp->status == busy)
 282                                 someBusy = 1;
 283                         else if (tsrp->status == offline)
 284                                 someOffline = 1;
 285                         else {
 286                                 code = cm_ConnByServer(tsp, usersp, connpp);
 287                                 if (code == 0) {
 288                                         cm_PutServer(tsp);
 289                                         /* Set RPC timeout */
 290                                         if (timeLeft > CM_CONN_CONNDEADTIME)
 291                                                 timeLeft = CM_CONN_CONNDEADTIME;
 292
 293                                         if (hardTimeLeft > CM_CONN_HARDDEADTIME)
 294                                                 hardTimeLeft = CM_CONN_HARDDEADTIME;
 295
 296                                         lock_ObtainMutex(&(*connpp)->mx);
 297                                         rx_SetConnDeadTime((*connpp)->callp,
 298                                                            timeLeft);
 299                                         rx_SetConnHardDeadTime((*connpp)->callp,
 300                                                                (u_short) hardTimeLeft);
 301                                         lock_ReleaseMutex(&(*connpp)->mx);
 302
 303                                         return 0;
 304                                 }
 305                                 if (firstError == 0) firstError = code;
 306                         }
 307                 }
 308                 lock_ObtainWrite(&cm_serverLock);
 309                 osi_assert(tsp->refCount-- > 0);
 310         }
 311
 312         lock_ReleaseWrite(&cm_serverLock);
 313         if (firstError == 0) {
 314                 if (someBusy) firstError = CM_ERROR_ALLBUSY;
 315                 else if (someOffline) firstError = CM_ERROR_NOSUCHVOLUME;
 316                 else firstError = CM_ERROR_TIMEDOUT;
 317         }
 318         osi_Log1(afsd_logp, "cm_ConnByMServers returning %x", firstError);
 319         return firstError;
 320 }
 321
 322 /* called with a held server to GC all bad connections hanging off of the server */
 323 void cm_GCConnections(cm_server_t *serverp)
 324 {
 325         cm_conn_t *tcp;
 326         cm_conn_t **lcpp;
 327         cm_user_t *userp;
 328
 329         lock_ObtainWrite(&cm_connLock);
 330         lcpp = &serverp->connsp;
 331         for(tcp = *lcpp; tcp; tcp = *lcpp) {
 332                 userp = tcp->userp;
 333                 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
 334                         /* do the deletion of this guy */
 335                         cm_ReleaseUser(userp);
 336                         *lcpp = tcp->nextp;
 337                         rx_DestroyConnection(tcp->callp);
 338                         lock_FinalizeMutex(&tcp->mx);
 339                         free(tcp);
 340                 }
 341                 else {
 342                         /* just advance to the next */
 343                         lcpp = &tcp->nextp;
 344                 }
 345         }
 346         lock_ReleaseWrite(&cm_connLock);
 347 }
 348
 349 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
 350         cm_server_t *serverp)
 351 {
 352         unsigned short port;
 353         int serviceID;
 354         int secIndex;
 355         struct rx_securityClass *secObjp;
 356         afs_int32 level;
 357
 358         if (serverp->type == CM_SERVER_VLDB) {
 359                 port = htons(7003);
 360                 serviceID = 52;
 361         }
 362         else {
 363                 osi_assert(serverp->type == CM_SERVER_FILE);
 364                 port = htons(7000);
 365                 serviceID = 1;
 366         }
 367         if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
 368                 secIndex = 2;
 369                 if (cryptall) {
 370                         level = rxkad_crypt;
 371                         tcp->cryptlevel = rxkad_crypt;
 372                 } else {
 373                         level = rxkad_clear;
 374                 }
 375                 secObjp = rxkad_NewClientSecurityObject(level,
 376                         &ucellp->sessionKey, ucellp->kvno,
 377                         ucellp->ticketLen, ucellp->ticketp);
 378         }
 379         else {
 380                 /* normal auth */
 381                 secIndex = 0;
 382                 secObjp = rxnull_NewClientSecurityObject();
 383         }
 384         osi_assert(secObjp != NULL);
 385         tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
 386                 port,
 387                 serviceID,
 388                 secObjp,
 389                 secIndex);
 390         rx_SetConnDeadTime(tcp->callp, CM_CONN_CONNDEADTIME);
 391         rx_SetConnHardDeadTime(tcp->callp, CM_CONN_HARDDEADTIME);
 392         tcp->ucgen = ucellp->gen;
 393 }
 394
 395 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
 396 {
 397         cm_conn_t *tcp;
 398         cm_ucell_t *ucellp;
 399
 400         lock_ObtainMutex(&userp->mx);
 401         lock_ObtainWrite(&cm_connLock);
 402         for(tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
 403                 if (tcp->userp == userp) break;
 404         }
 405         /* find ucell structure */
 406         ucellp = cm_GetUCell(userp, serverp->cellp);
 407         if (!tcp) {
 408                 tcp = malloc(sizeof(*tcp));
 409                 memset(tcp, 0, sizeof(*tcp));
 410                 tcp->nextp = serverp->connsp;
 411                 serverp->connsp = tcp;
 412                 tcp->userp = userp;
 413                 cm_HoldUser(userp);
 414                 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
 415                 tcp->serverp = serverp;
 416                 tcp->cryptlevel = rxkad_clear;
 417                 cm_NewRXConnection(tcp, ucellp, serverp);
 418                 tcp->refCount = 1;
 419         }
 420         else {
 421                 if ((tcp->ucgen < ucellp->gen) || (tcp->cryptlevel != cryptall))
 422                 {
 423                         rx_DestroyConnection(tcp->callp);
 424                         cm_NewRXConnection(tcp, ucellp, serverp);
 425                 }
 426                 tcp->refCount++;
 427         }
 428         lock_ReleaseWrite(&cm_connLock);
 429         lock_ReleaseMutex(&userp->mx);
 430
 431         /* return this pointer to our caller */
 432         osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
 433         *connpp = tcp;
 434
 435         return 0;
 436 }
 437
 438 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
 439         cm_conn_t **connpp)
 440 {
 441         long code;
 442
 443         cm_serverRef_t *serversp;
 444
 445         code = cm_GetServerList(fidp, userp, reqp, &serversp);
 446         if (code) {
 447                 *connpp = NULL;
 448                 return code;
 449         }
 450
 451         code = cm_ConnByMServers(serversp, userp, reqp, connpp);
 452         return code;
 453 }