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 all servers are offline, mark them non-busy and start over */
208 else if (errorCode == CM_ERROR_ALLOFFLINE) {
210 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE.");
212 /* cm_ForceUpdateVolume marks all servers as non_busy */
213 /* No it doesn't and it won't do anything if all of the
214 * the servers are marked as DOWN. So clear the DOWN
215 * flag and reset the busy state as well.
218 cm_GetServerList(fidp, userp, reqp, &serverspp);
219 serversp = *serverspp;
223 lock_ObtainWrite(&cm_serverLock);
224 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
225 tsrp->server->flags &= ~CM_SERVERFLAG_DOWN;
226 if (tsrp->status == busy)
227 tsrp->status = not_busy;
229 lock_ReleaseWrite(&cm_serverLock);
231 cm_FreeServerList(&serversp);
232 *serverspp = serversp;
237 if (fidp != NULL) /* Not a VLDB call */
238 cm_ForceUpdateVolume(fidp, userp, reqp);
242 /* if all servers are busy, mark them non-busy and start over */
243 else if (errorCode == CM_ERROR_ALLBUSY) {
247 cm_GetServerList(fidp, userp, reqp, &serverspp);
248 serversp = *serverspp;
251 lock_ObtainWrite(&cm_serverLock);
252 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
253 if (tsrp->status == busy)
254 tsrp->status = not_busy;
256 lock_ReleaseWrite(&cm_serverLock);
258 cm_FreeServerList(&serversp);
259 *serverspp = serversp;
265 /* special codes: VBUSY and VRESTARTING */
266 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
268 cm_GetServerList(fidp, userp, reqp, &serverspp);
269 serversp = *serverspp;
272 lock_ObtainWrite(&cm_serverLock);
273 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
274 if (tsrp->server == serverp
275 && tsrp->status == not_busy) {
280 lock_ReleaseWrite(&cm_serverLock);
282 cm_FreeServerList(&serversp);
283 *serverspp = serversp;
288 /* special codes: missing volumes */
289 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE
290 || errorCode == VSALVAGE || errorCode == VNOSERVICE)
292 /* Log server being offline for this volume */
293 osi_Log4(afsd_logp, "cm_Analyze found server %d.%d.%d.%d marked offline for a volume",
294 ((serverp->addr.sin_addr.s_addr & 0xff)),
295 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
296 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
297 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
298 /* Create Event Log message */
303 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
304 sprintf(s, "cm_Analyze: Server %d.%d.%d.%d reported volume %d as missing.",
305 ((serverp->addr.sin_addr.s_addr & 0xff)),
306 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
307 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
308 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24),
311 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
313 DeregisterEventSource(h);
316 /* Mark server offline for this volume */
318 cm_GetServerList(fidp, userp, reqp, &serverspp);
319 serversp = *serverspp;
322 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
323 if (tsrp->server == serverp)
324 tsrp->status = offline;
327 cm_FreeServerList(&serversp);
328 *serverspp = serversp;
335 else if (errorCode == RX_CALL_TIMEOUT) {
336 /* server took longer than hardDeadTime
337 * don't mark server as down but don't retry
338 * this is to prevent the SMB session from timing out
339 * In addition, we log an event to the event log
345 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
346 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
348 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
350 DeregisterEventSource(h);
354 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
356 else if (errorCode >= -64 && errorCode < 0) {
357 /* mark server as down */
358 lock_ObtainMutex(&serverp->mx);
359 serverp->flags |= CM_SERVERFLAG_DOWN;
360 lock_ReleaseMutex(&serverp->mx);
364 else if (errorCode == RXKADEXPIRED) {
366 lock_ObtainMutex(&userp->mx);
367 ucellp = cm_GetUCell(userp, serverp->cellp);
368 if (ucellp->ticketp) {
369 free(ucellp->ticketp);
370 ucellp->ticketp = NULL;
372 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
374 lock_ReleaseMutex(&userp->mx);
380 osi_Log1(afsd_logp, "cm_Analyze: ignoring error code 0x%x", errorCode);
383 if (retry && dead_session)
387 /* drop this on the way out */
391 /* retry until we fail to find a connection */
395 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
396 cm_req_t *reqp, cm_conn_t **connpp)
399 cm_serverRef_t *tsrp;
402 int someBusy = 0, someOffline = 0, allBusy = 1, allDown = 1;
403 long timeUsed, timeLeft, hardTimeLeft;
411 timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
413 gettimeofday(&now, NULL);
414 timeUsed = sub_time(now, reqp->startTime) / 1000;
417 /* leave 5 seconds margin of safety */
418 timeLeft = ConnDeadtimeout - timeUsed - 5;
419 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
421 lock_ObtainWrite(&cm_serverLock);
422 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
424 cm_GetServerNoLock(tsp);
425 lock_ReleaseWrite(&cm_serverLock);
426 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
427 if (tsrp->status == busy) {
430 } else if (tsrp->status == offline) {
435 code = cm_ConnByServer(tsp, usersp, connpp);
436 if (code == 0) { /* cm_CBS only returns 0 */
438 /* Set RPC timeout */
439 if (timeLeft > ConnDeadtimeout)
440 timeLeft = ConnDeadtimeout;
442 if (hardTimeLeft > HardDeadtimeout)
443 hardTimeLeft = HardDeadtimeout;
445 lock_ObtainMutex(&(*connpp)->mx);
446 rx_SetConnDeadTime((*connpp)->callp, timeLeft);
447 rx_SetConnHardDeadTime((*connpp)->callp, (u_short) hardTimeLeft);
448 lock_ReleaseMutex(&(*connpp)->mx);
452 /* therefore, this code is never executed */
457 lock_ObtainWrite(&cm_serverLock);
458 cm_PutServerNoLock(tsp);
461 lock_ReleaseWrite(&cm_serverLock);
462 if (firstError == 0) {
463 if (serversp == NULL)
464 firstError = CM_ERROR_NOSUCHVOLUME;
466 firstError = CM_ERROR_ALLOFFLINE;
468 firstError = CM_ERROR_ALLBUSY;
470 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
471 firstError = CM_ERROR_TIMEDOUT;
475 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
479 /* called with a held server to GC all bad connections hanging off of the server */
480 void cm_GCConnections(cm_server_t *serverp)
486 lock_ObtainWrite(&cm_connLock);
487 lcpp = &serverp->connsp;
488 for (tcp = *lcpp; tcp; tcp = *lcpp) {
490 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
491 /* do the deletion of this guy */
492 cm_PutServer(tcp->serverp);
493 cm_ReleaseUser(userp);
495 rx_DestroyConnection(tcp->callp);
496 lock_FinalizeMutex(&tcp->mx);
500 /* just advance to the next */
504 lock_ReleaseWrite(&cm_connLock);
507 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
508 cm_server_t *serverp)
513 struct rx_securityClass *secObjp;
515 if (serverp->type == CM_SERVER_VLDB) {
520 osi_assert(serverp->type == CM_SERVER_FILE);
524 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
527 tcp->cryptlevel = rxkad_crypt;
529 tcp->cryptlevel = rxkad_clear;
531 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
532 &ucellp->sessionKey, ucellp->kvno,
533 ucellp->ticketLen, ucellp->ticketp);
537 secObjp = rxnull_NewClientSecurityObject();
539 osi_assert(secObjp != NULL);
540 tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
545 rx_SetConnDeadTime(tcp->callp, ConnDeadtimeout);
546 rx_SetConnHardDeadTime(tcp->callp, HardDeadtimeout);
547 tcp->ucgen = ucellp->gen;
549 rxs_Release(secObjp); /* Decrement the initial refCount */
552 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
557 lock_ObtainMutex(&userp->mx);
558 lock_ObtainWrite(&cm_connLock);
559 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
560 if (tcp->userp == userp)
564 /* find ucell structure */
565 ucellp = cm_GetUCell(userp, serverp->cellp);
567 cm_GetServer(serverp);
568 tcp = malloc(sizeof(*tcp));
569 memset(tcp, 0, sizeof(*tcp));
570 tcp->nextp = serverp->connsp;
571 serverp->connsp = tcp;
574 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
575 lock_ObtainMutex(&tcp->mx);
576 tcp->serverp = serverp;
577 tcp->cryptlevel = rxkad_clear;
578 cm_NewRXConnection(tcp, ucellp, serverp);
580 lock_ReleaseMutex(&tcp->mx);
582 if ((tcp->ucgen < ucellp->gen) ||
583 (tcp->cryptlevel != (cryptall ? rxkad_crypt : rxkad_clear)))
585 if (tcp->ucgen < ucellp->gen)
586 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
588 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
589 lock_ObtainMutex(&tcp->mx);
590 rx_DestroyConnection(tcp->callp);
591 cm_NewRXConnection(tcp, ucellp, serverp);
592 lock_ReleaseMutex(&tcp->mx);
596 lock_ReleaseWrite(&cm_connLock);
597 lock_ReleaseMutex(&userp->mx);
599 /* return this pointer to our caller */
600 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
606 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
611 cm_serverRef_t **serverspp;
613 code = cm_GetServerList(fidp, userp, reqp, &serverspp);
619 code = cm_ConnByMServers(*serverspp, userp, reqp, connpp);
620 cm_FreeServerList(serverspp);
624 extern struct rx_connection *
625 cm_GetRxConn(cm_conn_t *connp)
627 struct rx_connection * rxconn;
628 lock_ObtainMutex(&connp->mx);
629 rxconn = connp->callp;
630 rx_GetConnection(rxconn);
631 lock_ReleaseMutex(&connp->mx);