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 /* No it doesn't. It won't do anything if all of the
205 * the servers are marked as DOWN. So clear the DOWN
206 * flag and reset the busy state as well.
208 cm_GetServerList(fidp, userp, reqp, &serversp);
209 lock_ObtainWrite(&cm_serverLock);
210 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
211 tsrp->server->flags &= ~CM_SERVERFLAG_DOWN;
212 if (tsrp->status == busy)
213 tsrp->status = not_busy;
215 lock_ReleaseWrite(&cm_serverLock);
218 cm_ForceUpdateVolume(fidp, userp, reqp);
222 /* if all servers are busy, mark them non-busy and start over */
223 if (errorCode == CM_ERROR_ALLBUSY) {
224 cm_GetServerList(fidp, userp, reqp, &serversp);
225 lock_ObtainWrite(&cm_serverLock);
226 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
227 if (tsrp->status == busy)
228 tsrp->status = not_busy;
230 lock_ReleaseWrite(&cm_serverLock);
235 /* special codes: VBUSY and VRESTARTING */
236 if (errorCode == VBUSY || errorCode == VRESTARTING) {
237 cm_GetServerList(fidp, userp, reqp, &serversp);
238 lock_ObtainWrite(&cm_serverLock);
239 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
240 if (tsrp->server == serverp
241 && tsrp->status == not_busy) {
246 lock_ReleaseWrite(&cm_serverLock);
250 /* special codes: missing volumes */
251 if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE
252 || errorCode == VSALVAGE || errorCode == VNOSERVICE) {
253 /* Log server being offline for this volume */
254 osi_Log4(afsd_logp, "cm_Analyze found server %d.%d.%d.%d marked offline for a volume",
255 ((serverp->addr.sin_addr.s_addr & 0xff)),
256 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
257 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
258 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
259 /* Create Event Log message */
264 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
265 sprintf(s, "cm_Analyze: Server %d.%d.%d.%d reported volume %d as missing.",
266 ((serverp->addr.sin_addr.s_addr & 0xff)),
267 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
268 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
269 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24),
272 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
274 DeregisterEventSource(h);
277 /* Mark server offline for this volume */
278 cm_GetServerList(fidp, userp, reqp, &serversp);
280 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
281 if (tsrp->server == serverp)
282 tsrp->status = offline;
288 if (errorCode == RX_CALL_TIMEOUT) {
289 /* server took longer than hardDeadTime
290 * don't mark server as down but don't retry
291 * this is to prevent the SMB session from timing out
292 * In addition, we log an event to the event log
298 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
299 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
301 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
303 DeregisterEventSource(h);
307 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
309 else if (errorCode >= -64 && errorCode < 0) {
310 /* mark server as down */
311 lock_ObtainMutex(&serverp->mx);
312 serverp->flags |= CM_SERVERFLAG_DOWN;
313 lock_ReleaseMutex(&serverp->mx);
317 if (errorCode == RXKADEXPIRED && !dead_session) {
318 lock_ObtainMutex(&userp->mx);
319 ucellp = cm_GetUCell(userp, serverp->cellp);
320 if (ucellp->ticketp) {
321 free(ucellp->ticketp);
322 ucellp->ticketp = NULL;
324 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
326 lock_ReleaseMutex(&userp->mx);
330 if (retry && dead_session)
334 /* drop this on the way out */
338 /* retry until we fail to find a connection */
342 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
343 cm_req_t *reqp, cm_conn_t **connpp)
346 cm_serverRef_t *tsrp;
349 int someBusy = 0, someOffline = 0, allBusy = 1, allDown = 1;
350 long timeUsed, timeLeft, hardTimeLeft;
358 timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
360 gettimeofday(&now, NULL);
361 timeUsed = sub_time(now, reqp->startTime) / 1000;
364 /* leave 5 seconds margin of safety */
365 timeLeft = ConnDeadtimeout - timeUsed - 5;
366 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
368 lock_ObtainWrite(&cm_serverLock);
370 for(tsrp = serversp; tsrp; tsrp=tsrp->next) {
373 lock_ReleaseWrite(&cm_serverLock);
374 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
376 if (tsrp->status == busy)
378 else if (tsrp->status == offline)
382 code = cm_ConnByServer(tsp, usersp, connpp);
385 /* Set RPC timeout */
386 if (timeLeft > ConnDeadtimeout)
387 timeLeft = ConnDeadtimeout;
389 if (hardTimeLeft > HardDeadtimeout)
390 hardTimeLeft = HardDeadtimeout;
392 lock_ObtainMutex(&(*connpp)->mx);
393 rx_SetConnDeadTime((*connpp)->callp,
395 rx_SetConnHardDeadTime((*connpp)->callp,
396 (u_short) hardTimeLeft);
397 lock_ReleaseMutex(&(*connpp)->mx);
405 lock_ObtainWrite(&cm_serverLock);
406 osi_assert(tsp->refCount-- > 0);
409 lock_ReleaseWrite(&cm_serverLock);
410 if (firstError == 0) {
412 firstError = CM_ERROR_ALLOFFLINE;
414 firstError = CM_ERROR_ALLBUSY;
415 else if (serversp == NULL)
416 /* Only return CM_ERROR_NOSUCHVOLUME if there are no
417 * servers for this volume
419 firstError = CM_ERROR_NOSUCHVOLUME;
421 firstError = CM_ERROR_TIMEDOUT;
423 osi_Log1(afsd_logp, "cm_ConnByMServers returning %x", firstError);
427 /* called with a held server to GC all bad connections hanging off of the server */
428 void cm_GCConnections(cm_server_t *serverp)
434 lock_ObtainWrite(&cm_connLock);
435 lcpp = &serverp->connsp;
436 for(tcp = *lcpp; tcp; tcp = *lcpp) {
438 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
439 /* do the deletion of this guy */
440 cm_ReleaseUser(userp);
442 rx_DestroyConnection(tcp->callp);
443 lock_FinalizeMutex(&tcp->mx);
447 /* just advance to the next */
451 lock_ReleaseWrite(&cm_connLock);
454 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
455 cm_server_t *serverp)
460 struct rx_securityClass *secObjp;
463 if (serverp->type == CM_SERVER_VLDB) {
468 osi_assert(serverp->type == CM_SERVER_FILE);
472 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
476 tcp->cryptlevel = rxkad_crypt;
480 secObjp = rxkad_NewClientSecurityObject(level,
481 &ucellp->sessionKey, ucellp->kvno,
482 ucellp->ticketLen, ucellp->ticketp);
487 secObjp = rxnull_NewClientSecurityObject();
489 osi_assert(secObjp != NULL);
490 tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
495 rx_SetConnDeadTime(tcp->callp, ConnDeadtimeout);
496 rx_SetConnHardDeadTime(tcp->callp, HardDeadtimeout);
497 tcp->ucgen = ucellp->gen;
499 rxs_Release(secObjp); /* Decrement the initial refCount */
502 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
507 lock_ObtainMutex(&userp->mx);
508 lock_ObtainWrite(&cm_connLock);
509 for(tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
510 if (tcp->userp == userp) break;
512 /* find ucell structure */
513 ucellp = cm_GetUCell(userp, serverp->cellp);
515 tcp = malloc(sizeof(*tcp));
516 memset(tcp, 0, sizeof(*tcp));
517 tcp->nextp = serverp->connsp;
518 serverp->connsp = tcp;
521 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
522 tcp->serverp = serverp;
523 tcp->cryptlevel = rxkad_clear;
524 cm_NewRXConnection(tcp, ucellp, serverp);
528 if ((tcp->ucgen < ucellp->gen) || (tcp->cryptlevel != cryptall))
530 rx_DestroyConnection(tcp->callp);
531 cm_NewRXConnection(tcp, ucellp, serverp);
535 lock_ReleaseWrite(&cm_connLock);
536 lock_ReleaseMutex(&userp->mx);
538 /* return this pointer to our caller */
539 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
545 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
550 cm_serverRef_t *serversp;
552 code = cm_GetServerList(fidp, userp, reqp, &serversp);
558 code = cm_ConnByMServers(serversp, userp, reqp, connpp);