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);
100 long cm_GetServerList(struct cm_fid *fidp, struct cm_user *userp,
101 struct cm_req *reqp, cm_serverRef_t **serverspp)
104 cm_volume_t *volp = NULL;
105 cm_serverRef_t *serversp = NULL;
106 cm_cell_t *cellp = NULL;
113 cellp = cm_FindCellByID(fidp->cell);
114 if (!cellp) return CM_ERROR_NOSUCHCELL;
116 code = cm_GetVolumeByID(cellp, fidp->volume, userp, reqp, &volp);
117 if (code) return code;
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;
129 *serverspp = serversp;
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.
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).
143 * If the error code is from cm_Conn() or friends, connp will be NULL.
145 * For VLDB calls, fidp will be NULL.
147 * volSyncp and/or cbrp may also be NULL.
150 cm_Analyze(cm_conn_t *connp, cm_user_t *userp, cm_req_t *reqp,
152 AFSVolSync *volSyncp, cm_callbackRequest_t *cbrp, long errorCode)
154 cm_server_t *serverp;
155 cm_serverRef_t *serversp, *tsrp;
160 osi_Log2(afsd_logp, "cm_Analyze connp 0x%x, code %d",
161 (long) connp, errorCode);
163 /* no locking required, since connp->serverp never changes after
165 dead_session = (userp->cellInfop == NULL);
167 serverp = connp->serverp;
169 /* Update callback pointer */
170 if (cbrp && errorCode == 0) cbrp->serverp = connp->serverp;
172 /* If not allowed to retry, don't */
173 if (reqp->flags & CM_REQ_NORETRY)
176 /* if timeout - check that it did not exceed the SMB timeout
178 if (errorCode == CM_ERROR_TIMEDOUT)
180 long timeUsed, timeLeft;
181 /* timeleft - get if from reqp the same way as cmXonnByMServers does */
183 timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
185 gettimeofday(&now, NULL);
186 timeUsed = sub_time(now, reqp->startTime) / 1000;
189 /* leave 5 seconds margin for sleep */
190 timeLeft = RDRtimeout - timeUsed;
194 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, NULL);
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.");
203 /* cm_ForceUpdateVolume marks all servers as non_busy */
204 cm_ForceUpdateVolume(fidp, userp, reqp);
208 /* if all servers are busy, mark them non-busy and start over */
209 if (errorCode == CM_ERROR_ALLBUSY) {
210 cm_GetServerList(fidp, userp, reqp, &serversp);
211 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
212 if (tsrp->status == busy)
213 tsrp->status = not_busy;
219 /* special codes: VBUSY and VRESTARTING */
220 if (errorCode == VBUSY || errorCode == VRESTARTING) {
221 cm_GetServerList(fidp, userp, reqp, &serversp);
222 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
223 if (tsrp->server == serverp
224 && tsrp->status == not_busy) {
232 /* special codes: missing volumes */
233 if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE
234 || errorCode == VSALVAGE || errorCode == VNOSERVICE) {
235 /* Log server being offline for this volume */
236 osi_Log4(afsd_logp, "cm_Analyze found server %d.%d.%d.%d marked offline for a volume",
237 ((serverp->addr.sin_addr.s_addr & 0xff)),
238 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
239 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
240 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
241 /* Create Event Log message */
246 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
247 sprintf(s, "cm_Analyze: Server %d.%d.%d.%d reported volume %d as missing.",
248 ((serverp->addr.sin_addr.s_addr & 0xff)),
249 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
250 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
251 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24),
254 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
256 DeregisterEventSource(h);
259 /* Mark server offline for this volume */
260 cm_GetServerList(fidp, userp, reqp, &serversp);
262 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
263 if (tsrp->server == serverp)
264 tsrp->status = offline;
270 if (errorCode == RX_CALL_TIMEOUT) {
271 /* server took longer than hardDeadTime
272 * don't mark server as down but don't retry
273 * this is to prevent the SMB session from timing out
274 * In addition, we log an event to the event log
280 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
281 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
283 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
285 DeregisterEventSource(h);
289 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
291 else if (errorCode >= -64 && errorCode < 0) {
292 /* mark server as down */
293 lock_ObtainMutex(&serverp->mx);
294 serverp->flags |= CM_SERVERFLAG_DOWN;
295 lock_ReleaseMutex(&serverp->mx);
299 if (errorCode == RXKADEXPIRED && !dead_session) {
300 lock_ObtainMutex(&userp->mx);
301 ucellp = cm_GetUCell(userp, serverp->cellp);
302 if (ucellp->ticketp) {
303 free(ucellp->ticketp);
304 ucellp->ticketp = NULL;
306 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
308 lock_ReleaseMutex(&userp->mx);
312 if (retry && dead_session)
316 /* drop this on the way out */
320 /* retry until we fail to find a connection */
324 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
325 cm_req_t *reqp, cm_conn_t **connpp)
328 cm_serverRef_t *tsrp;
331 int someBusy = 0, someOffline = 0, allDown = 1;
332 long timeUsed, timeLeft, hardTimeLeft;
340 timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
342 gettimeofday(&now, NULL);
343 timeUsed = sub_time(now, reqp->startTime) / 1000;
346 /* leave 5 seconds margin of safety */
347 timeLeft = ConnDeadtimeout - timeUsed - 5;
348 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
350 lock_ObtainWrite(&cm_serverLock);
352 for(tsrp = serversp; tsrp; tsrp=tsrp->next) {
355 lock_ReleaseWrite(&cm_serverLock);
356 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
358 if (tsrp->status == busy)
360 else if (tsrp->status == offline)
363 code = cm_ConnByServer(tsp, usersp, connpp);
366 /* Set RPC timeout */
367 if (timeLeft > ConnDeadtimeout)
368 timeLeft = ConnDeadtimeout;
370 if (hardTimeLeft > HardDeadtimeout)
371 hardTimeLeft = HardDeadtimeout;
373 lock_ObtainMutex(&(*connpp)->mx);
374 rx_SetConnDeadTime((*connpp)->callp,
376 rx_SetConnHardDeadTime((*connpp)->callp,
377 (u_short) hardTimeLeft);
378 lock_ReleaseMutex(&(*connpp)->mx);
386 lock_ObtainWrite(&cm_serverLock);
387 osi_assert(tsp->refCount-- > 0);
390 lock_ReleaseWrite(&cm_serverLock);
391 if (firstError == 0) {
393 firstError = CM_ERROR_ALLBUSY;
394 else if (someOffline)
395 firstError = CM_ERROR_ALLOFFLINE;
396 else if (!allDown && serversp)
397 firstError = CM_ERROR_TIMEDOUT;
398 /* Only return CM_ERROR_NOSUCHVOLUME if there are no
399 servers for this volume */
401 firstError = CM_ERROR_NOSUCHVOLUME;
403 osi_Log1(afsd_logp, "cm_ConnByMServers returning %x", firstError);
407 /* called with a held server to GC all bad connections hanging off of the server */
408 void cm_GCConnections(cm_server_t *serverp)
414 lock_ObtainWrite(&cm_connLock);
415 lcpp = &serverp->connsp;
416 for(tcp = *lcpp; tcp; tcp = *lcpp) {
418 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
419 /* do the deletion of this guy */
420 cm_ReleaseUser(userp);
422 rx_DestroyConnection(tcp->callp);
423 lock_FinalizeMutex(&tcp->mx);
427 /* just advance to the next */
431 lock_ReleaseWrite(&cm_connLock);
434 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
435 cm_server_t *serverp)
440 struct rx_securityClass *secObjp;
443 if (serverp->type == CM_SERVER_VLDB) {
448 osi_assert(serverp->type == CM_SERVER_FILE);
452 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
456 tcp->cryptlevel = rxkad_crypt;
460 secObjp = rxkad_NewClientSecurityObject(level,
461 &ucellp->sessionKey, ucellp->kvno,
462 ucellp->ticketLen, ucellp->ticketp);
467 secObjp = rxnull_NewClientSecurityObject();
469 osi_assert(secObjp != NULL);
470 tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
475 rx_SetConnDeadTime(tcp->callp, ConnDeadtimeout);
476 rx_SetConnHardDeadTime(tcp->callp, HardDeadtimeout);
477 tcp->ucgen = ucellp->gen;
479 rxs_Release(secObjp); /* Decrement the initial refCount */
482 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
487 lock_ObtainMutex(&userp->mx);
488 lock_ObtainWrite(&cm_connLock);
489 for(tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
490 if (tcp->userp == userp) break;
492 /* find ucell structure */
493 ucellp = cm_GetUCell(userp, serverp->cellp);
495 tcp = malloc(sizeof(*tcp));
496 memset(tcp, 0, sizeof(*tcp));
497 tcp->nextp = serverp->connsp;
498 serverp->connsp = tcp;
501 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
502 tcp->serverp = serverp;
503 tcp->cryptlevel = rxkad_clear;
504 cm_NewRXConnection(tcp, ucellp, serverp);
508 if ((tcp->ucgen < ucellp->gen) || (tcp->cryptlevel != cryptall))
510 rx_DestroyConnection(tcp->callp);
511 cm_NewRXConnection(tcp, ucellp, serverp);
515 lock_ReleaseWrite(&cm_connLock);
516 lock_ReleaseMutex(&userp->mx);
518 /* return this pointer to our caller */
519 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
525 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
530 cm_serverRef_t *serversp;
532 code = cm_GetServerList(fidp, userp, reqp, &serversp);
538 code = cm_ConnByMServers(serversp, userp, reqp, connpp);