2 * Copyright 2000, International Business Machines Corporation and others.
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
10 #include <afs/param.h>
28 osi_rwlock_t cm_connLock;
30 long RDRtimeout = CM_CONN_DEFAULTRDRTIMEOUT;
31 long ConnDeadtimeout = CM_CONN_CONNDEADTIME;
32 long HardDeadtimeout = CM_CONN_HARDDEADTIME;
34 #define LANMAN_WKS_PARAM_KEY "SYSTEM\\CurrentControlSet\\Services\\lanmanworkstation\\parameters"
35 #define LANMAN_WKS_SESSION_TIMEOUT "SessTimeout"
37 afs_int32 cryptall = 0;
39 void cm_PutConn(cm_conn_t *connp)
41 lock_ObtainWrite(&cm_connLock);
42 osi_assert(connp->refCount-- > 0);
43 lock_ReleaseWrite(&cm_connLock);
46 void cm_InitConn(void)
48 static osi_once_t once;
53 if (osi_Once(&once)) {
54 lock_InitializeRWLock(&cm_connLock, "connection global lock");
56 /* keisa - read timeout value for lanmanworkstation service.
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.
63 code = RegOpenKeyEx(HKEY_LOCAL_MACHINE, LANMAN_WKS_PARAM_KEY,
64 0, KEY_QUERY_VALUE, &parmKey);
65 if (code == ERROR_SUCCESS)
67 DWORD dummyLen = sizeof(sessTimeout);
68 code = RegQueryValueEx(parmKey, LANMAN_WKS_SESSION_TIMEOUT, NULL, NULL,
69 (BYTE *) &sessTimeout, &dummyLen);
70 if (code == ERROR_SUCCESS)
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);
78 if ( HardDeadtimeout < 2 * ConnDeadtimeout ) {
79 HardDeadtimeout = 2 * ConnDeadtimeout;
80 afsi_log("HardDeadTimeout increased to %d", HardDeadtimeout);
89 void cm_InitReq(cm_req_t *reqp)
91 memset((char *)reqp, 0, sizeof(cm_req_t));
93 reqp->startTime = GetCurrentTime();
95 gettimeofday(&reqp->startTime, NULL);
99 static long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
100 struct cm_req *reqp, cm_serverRef_t ***serversppp)
103 cm_volume_t *volp = NULL;
104 cm_cell_t *cellp = NULL;
111 cellp = cm_FindCellByID(fidp->cell);
112 if (!cellp) return CM_ERROR_NOSUCHCELL;
114 code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp, &volp);
115 if (code) return code;
117 *serversppp = cm_GetVolServers(volp, fidp->volume);
124 * Analyze the error return from an RPC. Determine whether or not to retry,
125 * and if we're going to retry, determine whether failover is appropriate,
126 * and whether timed backoff is appropriate.
128 * If the error code is from cm_Conn() or friends, it will be a CM_ERROR code.
129 * Otherwise it will be an RPC code. This may be a UNIX code (e.g. EDQUOT), or
130 * it may be an RX code, or it may be a special code (e.g. VNOVOL), or it may
131 * be a security code (e.g. RXKADEXPIRED).
133 * If the error code is from cm_Conn() or friends, connp will be NULL.
135 * For VLDB calls, fidp will be NULL.
137 * volSyncp and/or cbrp may also be NULL.
140 cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
142 AFSVolSync *volSyncp,
143 cm_serverRef_t * serversp,
144 cm_callbackRequest_t *cbrp, long errorCode)
146 cm_server_t *serverp = 0;
147 cm_serverRef_t **serverspp = 0;
148 cm_serverRef_t *tsrp;
151 int free_svr_list = 0;
153 long timeUsed, timeLeft;
155 osi_Log2(afsd_logp, "cm_Analyze connp 0x%x, code 0x%x",
156 (long) connp, errorCode);
158 /* no locking required, since connp->serverp never changes after
160 dead_session = (userp->cellInfop == NULL);
162 serverp = connp->serverp;
164 /* Update callback pointer */
165 if (cbrp && serverp && errorCode == 0) {
167 if ( cbrp->serverp != serverp ) {
168 lock_ObtainWrite(&cm_serverLock);
169 cm_PutServerNoLock(cbrp->serverp);
170 cm_GetServerNoLock(serverp);
171 lock_ReleaseWrite(&cm_serverLock);
174 cm_GetServer(serverp);
176 lock_ObtainWrite(&cm_callbackLock);
177 cbrp->serverp = serverp;
178 lock_ReleaseWrite(&cm_callbackLock);
181 /* If not allowed to retry, don't */
182 if (reqp->flags & CM_REQ_NORETRY)
185 /* if timeout - check that it did not exceed the SMB timeout
188 /* timeleft - get if from reqp the same way as cmXonnByMServers does */
190 timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
192 gettimeofday(&now, NULL);
193 timeUsed = sub_time(now, reqp->startTime) / 1000;
196 /* leave 5 seconds margin for sleep */
197 timeLeft = RDRtimeout - timeUsed;
199 if (errorCode == CM_ERROR_TIMEDOUT) {
202 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, NULL);
207 /* if there is nosuchvolume, then we have a situation in which a
208 * previously known volume no longer has a set of servers
209 * associated with it. Either the volume has moved or
210 * the volume has been deleted. Try to find a new server list
211 * until the timeout period expires.
213 else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
215 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
220 if (fidp != NULL) /* Not a VLDB call */
221 cm_ForceUpdateVolume(fidp, userp, reqp);
225 else if (errorCode == CM_ERROR_ALLOFFLINE) {
227 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE.");
229 /* cm_ForceUpdateVolume marks all servers as non_busy */
230 /* No it doesn't and it won't do anything if all of the
231 * the servers are marked as DOWN. So clear the DOWN
232 * flag and reset the busy state as well.
235 cm_GetServerList(fidp, userp, reqp, &serverspp);
236 serversp = *serverspp;
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;
246 lock_ReleaseWrite(&cm_serverLock);
248 cm_FreeServerList(&serversp);
249 *serverspp = serversp;
254 if (fidp != NULL) /* Not a VLDB call */
255 cm_ForceUpdateVolume(fidp, userp, reqp);
259 /* if all servers are busy, mark them non-busy and start over */
260 else if (errorCode == CM_ERROR_ALLBUSY) {
264 cm_GetServerList(fidp, userp, reqp, &serverspp);
265 serversp = *serverspp;
268 lock_ObtainWrite(&cm_serverLock);
269 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
270 if (tsrp->status == busy)
271 tsrp->status = not_busy;
273 lock_ReleaseWrite(&cm_serverLock);
275 cm_FreeServerList(&serversp);
276 *serverspp = serversp;
282 /* special codes: VBUSY and VRESTARTING */
283 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
285 cm_GetServerList(fidp, userp, reqp, &serverspp);
286 serversp = *serverspp;
289 lock_ObtainWrite(&cm_serverLock);
290 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
291 if (tsrp->server == serverp
292 && tsrp->status == not_busy) {
297 lock_ReleaseWrite(&cm_serverLock);
299 cm_FreeServerList(&serversp);
300 *serverspp = serversp;
305 /* special codes: missing volumes */
306 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE
307 || errorCode == VSALVAGE || errorCode == VNOSERVICE)
309 /* Log server being offline for this volume */
310 osi_Log4(afsd_logp, "cm_Analyze found server %d.%d.%d.%d marked offline for a volume",
311 ((serverp->addr.sin_addr.s_addr & 0xff)),
312 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
313 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
314 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
315 /* Create Event Log message */
320 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
321 sprintf(s, "cm_Analyze: Server %d.%d.%d.%d reported volume %d as missing.",
322 ((serverp->addr.sin_addr.s_addr & 0xff)),
323 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
324 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
325 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24),
328 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
330 DeregisterEventSource(h);
333 /* Mark server offline for this volume */
335 cm_GetServerList(fidp, userp, reqp, &serverspp);
336 serversp = *serverspp;
339 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
340 if (tsrp->server == serverp)
341 tsrp->status = offline;
344 cm_FreeServerList(&serversp);
345 *serverspp = serversp;
352 else if (errorCode == RX_CALL_TIMEOUT) {
353 /* server took longer than hardDeadTime
354 * don't mark server as down but don't retry
355 * this is to prevent the SMB session from timing out
356 * In addition, we log an event to the event log
362 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
363 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
365 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
367 DeregisterEventSource(h);
371 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
373 else if (errorCode >= -64 && errorCode < 0) {
374 /* mark server as down */
375 lock_ObtainMutex(&serverp->mx);
376 serverp->flags |= CM_SERVERFLAG_DOWN;
377 lock_ReleaseMutex(&serverp->mx);
381 else if (errorCode == RXKADEXPIRED ||
382 errorCode == RXKADBADTICKET) {
384 lock_ObtainMutex(&userp->mx);
385 ucellp = cm_GetUCell(userp, serverp->cellp);
386 if (ucellp->ticketp) {
387 free(ucellp->ticketp);
388 ucellp->ticketp = NULL;
390 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
392 lock_ReleaseMutex(&userp->mx);
398 char * s = "unknown error";
399 switch ( errorCode ) {
400 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
401 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT "; break;
402 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL "; break;
403 case RXKADTICKETLEN : s = "RXKADTICKETLEN "; break;
404 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
405 case RXKADNOAUTH : s = "RXKADNOAUTH "; break;
406 case RXKADBADKEY : s = "RXKADBADKEY "; break;
407 case RXKADBADTICKET : s = "RXKADBADTICKET "; break;
408 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY "; break;
409 case RXKADEXPIRED : s = "RXKADEXPIRED "; break;
410 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON "; break;
411 case RXKADDATALEN : s = "RXKADDATALEN "; break;
412 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL "; break;
414 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
419 if (retry && dead_session)
423 /* drop this on the way out */
427 /* retry until we fail to find a connection */
431 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
432 cm_req_t *reqp, cm_conn_t **connpp)
435 cm_serverRef_t *tsrp;
438 int someBusy = 0, someOffline = 0, allBusy = 1, allDown = 1;
439 long timeUsed, timeLeft, hardTimeLeft;
447 timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
449 gettimeofday(&now, NULL);
450 timeUsed = sub_time(now, reqp->startTime) / 1000;
453 /* leave 5 seconds margin of safety */
454 timeLeft = ConnDeadtimeout - timeUsed - 5;
455 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
457 lock_ObtainWrite(&cm_serverLock);
458 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
460 cm_GetServerNoLock(tsp);
461 lock_ReleaseWrite(&cm_serverLock);
462 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
463 if (tsrp->status == busy) {
466 } else if (tsrp->status == offline) {
471 code = cm_ConnByServer(tsp, usersp, connpp);
472 if (code == 0) { /* cm_CBS only returns 0 */
474 /* Set RPC timeout */
475 if (timeLeft > ConnDeadtimeout)
476 timeLeft = ConnDeadtimeout;
478 if (hardTimeLeft > HardDeadtimeout)
479 hardTimeLeft = HardDeadtimeout;
481 lock_ObtainMutex(&(*connpp)->mx);
482 rx_SetConnDeadTime((*connpp)->callp, timeLeft);
483 rx_SetConnHardDeadTime((*connpp)->callp, (u_short) hardTimeLeft);
484 lock_ReleaseMutex(&(*connpp)->mx);
488 /* therefore, this code is never executed */
493 lock_ObtainWrite(&cm_serverLock);
494 cm_PutServerNoLock(tsp);
497 lock_ReleaseWrite(&cm_serverLock);
498 if (firstError == 0) {
499 if (serversp == NULL)
500 firstError = CM_ERROR_NOSUCHVOLUME;
502 firstError = CM_ERROR_ALLOFFLINE;
504 firstError = CM_ERROR_ALLBUSY;
506 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
507 firstError = CM_ERROR_TIMEDOUT;
511 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
515 /* called with a held server to GC all bad connections hanging off of the server */
516 void cm_GCConnections(cm_server_t *serverp)
522 lock_ObtainWrite(&cm_connLock);
523 lcpp = &serverp->connsp;
524 for (tcp = *lcpp; tcp; tcp = *lcpp) {
526 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
527 /* do the deletion of this guy */
528 cm_PutServer(tcp->serverp);
529 cm_ReleaseUser(userp);
531 rx_DestroyConnection(tcp->callp);
532 lock_FinalizeMutex(&tcp->mx);
536 /* just advance to the next */
540 lock_ReleaseWrite(&cm_connLock);
543 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
544 cm_server_t *serverp)
549 struct rx_securityClass *secObjp;
551 if (serverp->type == CM_SERVER_VLDB) {
556 osi_assert(serverp->type == CM_SERVER_FILE);
560 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
563 tcp->cryptlevel = rxkad_crypt;
565 tcp->cryptlevel = rxkad_clear;
567 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
568 &ucellp->sessionKey, ucellp->kvno,
569 ucellp->ticketLen, ucellp->ticketp);
573 tcp->cryptlevel = rxkad_clear;
574 secObjp = rxnull_NewClientSecurityObject();
576 osi_assert(secObjp != NULL);
577 tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
582 rx_SetConnDeadTime(tcp->callp, ConnDeadtimeout);
583 rx_SetConnHardDeadTime(tcp->callp, HardDeadtimeout);
584 tcp->ucgen = ucellp->gen;
586 rxs_Release(secObjp); /* Decrement the initial refCount */
589 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
594 lock_ObtainMutex(&userp->mx);
595 lock_ObtainWrite(&cm_connLock);
596 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
597 if (tcp->userp == userp)
601 /* find ucell structure */
602 ucellp = cm_GetUCell(userp, serverp->cellp);
604 cm_GetServer(serverp);
605 tcp = malloc(sizeof(*tcp));
606 memset(tcp, 0, sizeof(*tcp));
607 tcp->nextp = serverp->connsp;
608 serverp->connsp = tcp;
611 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
612 lock_ObtainMutex(&tcp->mx);
613 tcp->serverp = serverp;
614 tcp->cryptlevel = rxkad_clear;
615 cm_NewRXConnection(tcp, ucellp, serverp);
617 lock_ReleaseMutex(&tcp->mx);
619 if ((tcp->ucgen < ucellp->gen) ||
620 (tcp->cryptlevel != (cryptall ? (ucellp->flags & CM_UCELLFLAG_RXKAD ? rxkad_crypt : rxkad_clear) : rxkad_clear)))
622 if (tcp->ucgen < ucellp->gen)
623 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
625 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
626 lock_ObtainMutex(&tcp->mx);
627 rx_DestroyConnection(tcp->callp);
628 cm_NewRXConnection(tcp, ucellp, serverp);
629 lock_ReleaseMutex(&tcp->mx);
633 lock_ReleaseWrite(&cm_connLock);
634 lock_ReleaseMutex(&userp->mx);
636 /* return this pointer to our caller */
637 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
643 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
648 cm_serverRef_t **serverspp;
650 code = cm_GetServerList(fidp, userp, reqp, &serverspp);
656 code = cm_ConnByMServers(*serverspp, userp, reqp, connpp);
657 cm_FreeServerList(serverspp);
661 extern struct rx_connection *
662 cm_GetRxConn(cm_conn_t *connp)
664 struct rx_connection * rxconn;
665 lock_ObtainMutex(&connp->mx);
666 rxconn = connp->callp;
667 rx_GetConnection(rxconn);
668 lock_ReleaseMutex(&connp->mx);