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 offline, mark them non-busy and start over */
 138         if (errorCode == CM_ERROR_ALLOFFLINE) {
 139             osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE.");
 140             thrd_Sleep(5000);
 141             /* cm_ForceUpdateVolume marks all servers as non_busy */
 142             cm_ForceUpdateVolume(fidp, userp, reqp);
 143             retry = 1;
 144         }
 145
 146         /* if all servers are busy, mark them non-busy and start over */
 147         if (errorCode == CM_ERROR_ALLBUSY) {
 148                 cm_GetServerList(fidp, userp, reqp, &serversp);
 149                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 150                         if (tsrp->status == busy)
 151                                 tsrp->status = not_busy;
 152                 }
 153                 thrd_Sleep(5000);
 154                 retry = 1;
 155         }
 156
 157         /* special codes:  VBUSY and VRESTARTING */
 158         if (errorCode == VBUSY || errorCode == VRESTARTING) {
 159                 cm_GetServerList(fidp, userp, reqp, &serversp);
 160                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 161                         if (tsrp->server == serverp
 162                             && tsrp->status == not_busy) {
 163                                 tsrp->status = busy;
 164                                 break;
 165                         }
 166                 }
 167                 retry = 1;
 168         }
 169
 170         /* special codes:  missing volumes */
 171         if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE
 172             || errorCode == VSALVAGE || errorCode == VNOSERVICE) {
 173                 long oldSum, newSum;
 174                 int same;
 175
 176                 /* Log server being offline for this volume */
 177                 osi_Log4(afsd_logp, "cm_Analyze found server %d.%d.%d.%d marked offline for a volume",
 178                          ((serverp->addr.sin_addr.s_addr & 0xff)),
 179                          ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
 180                          ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
 181                          ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
 182                 /* Create Event Log message */
 183                 {
 184                     HANDLE h;
 185                     char *ptbuf[1];
 186                     char s[100];
 187                     h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
 188                     sprintf(s, "cm_Analyze: Server %d.%d.%d.%d reported volume %d as missing.",
 189                             ((serverp->addr.sin_addr.s_addr & 0xff)),
 190                             ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
 191                             ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
 192                             ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24),
 193                             fidp->volume);
 194                     ptbuf[0] = s;
 195                     ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
 196                                 1, 0, ptbuf, NULL);
 197                     DeregisterEventSource(h);
 198                 }
 199
 200                 /* Mark server offline for this volume */
 201                 cm_GetServerList(fidp, userp, reqp, &serversp);
 202
 203                 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
 204                         if (tsrp->server == serverp)
 205                                 tsrp->status = offline;
 206                 }
 207                 retry = 1;
 208         }
 209
 210         /* RX codes */
 211         if (errorCode == RX_CALL_TIMEOUT) {
 212                 /* server took longer than hardDeadTime
 213                  * don't mark server as down but don't retry
 214                  * this is to prevent the SMB session from timing out
 215                  * In addition, we log an event to the event log
 216                  */
 217 #ifndef DJGPP
 218                 HANDLE h;
 219                 char *ptbuf[1];
 220                 char s[100];
 221                 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
 222                 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
 223                 ptbuf[0] = s;
 224                 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
 225                         1, 0, ptbuf, NULL);
 226                 DeregisterEventSource(h);
 227 #endif /* !DJGPP */
 228
 229                 retry = 0;
 230                 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
 231         }
 232         else if (errorCode >= -64 && errorCode < 0) {
 233                 /* mark server as down */
 234                 lock_ObtainMutex(&serverp->mx);
 235                 serverp->flags |= CM_SERVERFLAG_DOWN;
 236                 lock_ReleaseMutex(&serverp->mx);
 237                 retry = 1;
 238         }
 239
 240         if (errorCode == RXKADEXPIRED && !dead_session) {
 241                 lock_ObtainMutex(&userp->mx);
 242                 ucellp = cm_GetUCell(userp, serverp->cellp);
 243                 if (ucellp->ticketp) {
 244                         free(ucellp->ticketp);
 245                         ucellp->ticketp = NULL;
 246                 }
 247                 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
 248                 ucellp->gen++;
 249                 lock_ReleaseMutex(&userp->mx);
 250                 retry = 1;
 251         }
 252
 253         if (retry && dead_session)
 254                 retry = 0;
 255
 256 out:
 257         /* drop this on the way out */
 258         if (connp)
 259                 cm_PutConn(connp);
 260
 261         /* retry until we fail to find a connection */
 262         return retry;
 263 }
 264
 265 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
 266         cm_req_t *reqp, cm_conn_t **connpp)
 267 {
 268         long code;
 269         cm_serverRef_t *tsrp;
 270         cm_server_t *tsp;
 271         long firstError = 0;
 272         int someBusy = 0, someOffline = 0;
 273         long timeUsed, timeLeft, hardTimeLeft;
 274 #ifdef DJGPP
 275         struct timeval now;
 276 #endif /* DJGPP */
 277
 278         *connpp = NULL;
 279
 280 #ifndef DJGPP
 281         timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
 282 #else
 283         gettimeofday(&now, NULL);
 284         timeUsed = sub_time(now, reqp->startTime) / 1000;
 285 #endif
 286
 287         /* leave 5 seconds margin of safety */
 288         timeLeft = RDRtimeout - timeUsed - 5;
 289         hardTimeLeft = timeLeft;
 290
 291         /* Time enough to do an RPC? */
 292         if (timeLeft < 1) {
 293                 return CM_ERROR_TIMEDOUT;
 294         }
 295
 296         lock_ObtainWrite(&cm_serverLock);
 297
 298         for(tsrp = serversp; tsrp; tsrp=tsrp->next) {
 299                 tsp = tsrp->server;
 300                 tsp->refCount++;
 301                 lock_ReleaseWrite(&cm_serverLock);
 302                 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
 303                         if (tsrp->status == busy)
 304                                 someBusy = 1;
 305                         else if (tsrp->status == offline)
 306                                 someOffline = 1;
 307                         else {
 308                                 code = cm_ConnByServer(tsp, usersp, connpp);
 309                                 if (code == 0) {
 310                                         cm_PutServer(tsp);
 311                                         /* Set RPC timeout */
 312                                         if (timeLeft > CM_CONN_CONNDEADTIME)
 313                                                 timeLeft = CM_CONN_CONNDEADTIME;
 314
 315                                         if (hardTimeLeft > CM_CONN_HARDDEADTIME)
 316                                                 hardTimeLeft = CM_CONN_HARDDEADTIME;
 317
 318                                         lock_ObtainMutex(&(*connpp)->mx);
 319                                         rx_SetConnDeadTime((*connpp)->callp,
 320                                                            timeLeft);
 321                                         rx_SetConnHardDeadTime((*connpp)->callp,
 322                                                                (u_short) hardTimeLeft);
 323                                         lock_ReleaseMutex(&(*connpp)->mx);
 324
 325                                         return 0;
 326                                 }
 327                                 if (firstError == 0) firstError = code;
 328                         }
 329                 }
 330                 lock_ObtainWrite(&cm_serverLock);
 331                 osi_assert(tsp->refCount-- > 0);
 332         }
 333
 334         lock_ReleaseWrite(&cm_serverLock);
 335         if (firstError == 0) {
 336                 if (someBusy) firstError = CM_ERROR_ALLBUSY;
 337                 else if (someOffline) firstError = CM_ERROR_ALLOFFLINE;
 338                 else if (serversp) firstError = CM_ERROR_TIMEDOUT;
 339                 /* Only return CM_ERROR_NOSUCHVOLUME if there are no
 340                    servers for this volume */
 341                 else firstError = CM_ERROR_NOSUCHVOLUME;
 342         }
 343         osi_Log1(afsd_logp, "cm_ConnByMServers returning %x", firstError);
 344         return firstError;
 345 }
 346
 347 /* called with a held server to GC all bad connections hanging off of the server */
 348 void cm_GCConnections(cm_server_t *serverp)
 349 {
 350         cm_conn_t *tcp;
 351         cm_conn_t **lcpp;
 352         cm_user_t *userp;
 353
 354         lock_ObtainWrite(&cm_connLock);
 355         lcpp = &serverp->connsp;
 356         for(tcp = *lcpp; tcp; tcp = *lcpp) {
 357                 userp = tcp->userp;
 358                 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
 359                         /* do the deletion of this guy */
 360                         cm_ReleaseUser(userp);
 361                         *lcpp = tcp->nextp;
 362                         rx_DestroyConnection(tcp->callp);
 363                         lock_FinalizeMutex(&tcp->mx);
 364                         free(tcp);
 365                 }
 366                 else {
 367                         /* just advance to the next */
 368                         lcpp = &tcp->nextp;
 369                 }
 370         }
 371         lock_ReleaseWrite(&cm_connLock);
 372 }
 373
 374 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
 375         cm_server_t *serverp)
 376 {
 377         unsigned short port;
 378         int serviceID;
 379         int secIndex;
 380         struct rx_securityClass *secObjp;
 381         afs_int32 level;
 382
 383         if (serverp->type == CM_SERVER_VLDB) {
 384                 port = htons(7003);
 385                 serviceID = 52;
 386         }
 387         else {
 388                 osi_assert(serverp->type == CM_SERVER_FILE);
 389                 port = htons(7000);
 390                 serviceID = 1;
 391         }
 392         if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
 393                 secIndex = 2;
 394                 if (cryptall) {
 395                         level = rxkad_crypt;
 396                         tcp->cryptlevel = rxkad_crypt;
 397                 } else {
 398                         level = rxkad_clear;
 399                 }
 400                 secObjp = rxkad_NewClientSecurityObject(level,
 401                         &ucellp->sessionKey, ucellp->kvno,
 402                         ucellp->ticketLen, ucellp->ticketp);
 403         }
 404         else {
 405                 /* normal auth */
 406                 secIndex = 0;
 407                 secObjp = rxnull_NewClientSecurityObject();
 408         }
 409         osi_assert(secObjp != NULL);
 410         tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
 411                 port,
 412                 serviceID,
 413                 secObjp,
 414                 secIndex);
 415         rx_SetConnDeadTime(tcp->callp, CM_CONN_CONNDEADTIME);
 416         rx_SetConnHardDeadTime(tcp->callp, CM_CONN_HARDDEADTIME);
 417         tcp->ucgen = ucellp->gen;
 418 }
 419
 420 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
 421 {
 422         cm_conn_t *tcp;
 423         cm_ucell_t *ucellp;
 424
 425         lock_ObtainMutex(&userp->mx);
 426         lock_ObtainWrite(&cm_connLock);
 427         for(tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
 428                 if (tcp->userp == userp) break;
 429         }
 430         /* find ucell structure */
 431         ucellp = cm_GetUCell(userp, serverp->cellp);
 432         if (!tcp) {
 433                 tcp = malloc(sizeof(*tcp));
 434                 memset(tcp, 0, sizeof(*tcp));
 435                 tcp->nextp = serverp->connsp;
 436                 serverp->connsp = tcp;
 437                 tcp->userp = userp;
 438                 cm_HoldUser(userp);
 439                 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
 440                 tcp->serverp = serverp;
 441                 tcp->cryptlevel = rxkad_clear;
 442                 cm_NewRXConnection(tcp, ucellp, serverp);
 443                 tcp->refCount = 1;
 444         }
 445         else {
 446                 if ((tcp->ucgen < ucellp->gen) || (tcp->cryptlevel != cryptall))
 447                 {
 448                         rx_DestroyConnection(tcp->callp);
 449                         cm_NewRXConnection(tcp, ucellp, serverp);
 450                 }
 451                 tcp->refCount++;
 452         }
 453         lock_ReleaseWrite(&cm_connLock);
 454         lock_ReleaseMutex(&userp->mx);
 455
 456         /* return this pointer to our caller */
 457         osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
 458         *connpp = tcp;
 459
 460         return 0;
 461 }
 462
 463 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
 464         cm_conn_t **connpp)
 465 {
 466         long code;
 467
 468         cm_serverRef_t *serversp;
 469
 470         code = cm_GetServerList(fidp, userp, reqp, &serversp);
 471         if (code) {
 472                 *connpp = NULL;
 473                 return code;
 474         }
 475
 476         code = cm_ConnByMServers(serversp, userp, reqp, connpp);
 477         return code;
 478 }