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 = NULL;
147 cm_serverRef_t **serverspp = NULL;
148 cm_serverRef_t *tsrp;
151 int free_svr_list = 0;
153 long timeUsed, timeLeft;
156 osi_Log2(afsd_logp, "cm_Analyze connp 0x%x, code 0x%x",
157 (long) connp, errorCode);
159 /* no locking required, since connp->serverp never changes after
161 dead_session = (userp->cellInfop == NULL);
163 serverp = connp->serverp;
165 /* Update callback pointer */
166 if (cbrp && serverp && errorCode == 0) {
168 if ( cbrp->serverp != serverp ) {
169 lock_ObtainWrite(&cm_serverLock);
170 cm_PutServerNoLock(cbrp->serverp);
171 cm_GetServerNoLock(serverp);
172 lock_ReleaseWrite(&cm_serverLock);
175 cm_GetServer(serverp);
177 lock_ObtainWrite(&cm_callbackLock);
178 cbrp->serverp = serverp;
179 lock_ReleaseWrite(&cm_callbackLock);
182 /* If not allowed to retry, don't */
183 if (reqp->flags & CM_REQ_NORETRY)
186 /* if timeout - check that it did not exceed the SMB timeout
189 /* timeleft - get if from reqp the same way as cmXonnByMServers does */
191 timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
193 gettimeofday(&now, NULL);
194 timeUsed = sub_time(now, reqp->startTime) / 1000;
197 /* leave 5 seconds margin for sleep */
198 timeLeft = RDRtimeout - timeUsed;
200 if (errorCode == CM_ERROR_TIMEDOUT) {
203 cm_CheckServers(CM_FLAG_CHECKDOWNSERVERS, NULL);
208 /* if there is nosuchvolume, then we have a situation in which a
209 * previously known volume no longer has a set of servers
210 * associated with it. Either the volume has moved or
211 * the volume has been deleted. Try to find a new server list
212 * until the timeout period expires.
214 else if (errorCode == CM_ERROR_NOSUCHVOLUME) {
216 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_NOSUCHVOLUME.");
221 if (fidp != NULL) /* Not a VLDB call */
222 cm_ForceUpdateVolume(fidp, userp, reqp);
226 else if (errorCode == CM_ERROR_ALLOFFLINE) {
228 osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE.");
231 /* cm_ForceUpdateVolume marks all servers as non_busy */
232 /* No it doesn't and it won't do anything if all of the
233 * the servers are marked as DOWN. So clear the DOWN
234 * flag and reset the busy state as well.
237 code = cm_GetServerList(fidp, userp, reqp, &serverspp);
239 serversp = *serverspp;
244 lock_ObtainWrite(&cm_serverLock);
245 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
246 tsrp->server->flags &= ~CM_SERVERFLAG_DOWN;
247 if (tsrp->status == busy)
248 tsrp->status = not_busy;
250 lock_ReleaseWrite(&cm_serverLock);
252 cm_FreeServerList(&serversp);
253 *serverspp = serversp;
258 if (fidp != NULL) /* Not a VLDB call */
259 cm_ForceUpdateVolume(fidp, userp, reqp);
265 /* if all servers are busy, mark them non-busy and start over */
266 else if (errorCode == CM_ERROR_ALLBUSY) {
270 code = cm_GetServerList(fidp, userp, reqp, &serverspp);
272 serversp = *serverspp;
276 lock_ObtainWrite(&cm_serverLock);
277 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
278 if (tsrp->status == busy)
279 tsrp->status = not_busy;
281 lock_ReleaseWrite(&cm_serverLock);
283 cm_FreeServerList(&serversp);
284 *serverspp = serversp;
290 /* special codes: VBUSY and VRESTARTING */
291 else if (errorCode == VBUSY || errorCode == VRESTARTING) {
293 code = cm_GetServerList(fidp, userp, reqp, &serverspp);
295 serversp = *serverspp;
299 lock_ObtainWrite(&cm_serverLock);
300 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
301 if (tsrp->server == serverp
302 && tsrp->status == not_busy) {
307 lock_ReleaseWrite(&cm_serverLock);
309 cm_FreeServerList(&serversp);
310 *serverspp = serversp;
315 /* special codes: missing volumes */
316 else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
317 errorCode == VSALVAGE || errorCode == VNOSERVICE || errorCode == VIO)
319 /* Log server being offline for this volume */
320 osi_Log4(afsd_logp, "cm_Analyze found server %d.%d.%d.%d marked offline for a volume",
321 ((serverp->addr.sin_addr.s_addr & 0xff)),
322 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
323 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
324 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
325 /* Create Event Log message */
330 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
331 sprintf(s, "cm_Analyze: Server %d.%d.%d.%d reported volume %d as missing.",
332 ((serverp->addr.sin_addr.s_addr & 0xff)),
333 ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
334 ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
335 ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24),
338 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
340 DeregisterEventSource(h);
343 /* Mark server offline for this volume */
345 code = cm_GetServerList(fidp, userp, reqp, &serverspp);
347 serversp = *serverspp;
351 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
352 if (tsrp->server == serverp)
353 tsrp->status = offline;
356 cm_FreeServerList(&serversp);
357 *serverspp = serversp;
364 else if (errorCode == RX_CALL_TIMEOUT) {
365 /* server took longer than hardDeadTime
366 * don't mark server as down but don't retry
367 * this is to prevent the SMB session from timing out
368 * In addition, we log an event to the event log
374 h = RegisterEventSource(NULL, AFS_DAEMON_EVENT_NAME);
375 sprintf(s, "cm_Analyze: HardDeadTime exceeded.");
377 ReportEvent(h, EVENTLOG_WARNING_TYPE, 0, 1009, NULL,
379 DeregisterEventSource(h);
383 osi_Log0(afsd_logp, "cm_Analyze: hardDeadTime exceeded");
385 else if (errorCode >= -64 && errorCode < 0) {
386 /* mark server as down */
387 lock_ObtainMutex(&serverp->mx);
388 serverp->flags |= CM_SERVERFLAG_DOWN;
389 lock_ReleaseMutex(&serverp->mx);
393 else if (errorCode == RXKADEXPIRED ||
394 errorCode == RXKADBADTICKET) {
396 lock_ObtainMutex(&userp->mx);
397 ucellp = cm_GetUCell(userp, serverp->cellp);
398 if (ucellp->ticketp) {
399 free(ucellp->ticketp);
400 ucellp->ticketp = NULL;
402 ucellp->flags &= ~CM_UCELLFLAG_RXKAD;
404 lock_ReleaseMutex(&userp->mx);
410 char * s = "unknown error";
411 switch ( errorCode ) {
412 case RXKADINCONSISTENCY: s = "RXKADINCONSISTENCY"; break;
413 case RXKADPACKETSHORT : s = "RXKADPACKETSHORT"; break;
414 case RXKADLEVELFAIL : s = "RXKADLEVELFAIL"; break;
415 case RXKADTICKETLEN : s = "RXKADTICKETLEN"; break;
416 case RXKADOUTOFSEQUENCE: s = "RXKADOUTOFSEQUENCE"; break;
417 case RXKADNOAUTH : s = "RXKADNOAUTH"; break;
418 case RXKADBADKEY : s = "RXKADBADKEY"; break;
419 case RXKADBADTICKET : s = "RXKADBADTICKET"; break;
420 case RXKADUNKNOWNKEY : s = "RXKADUNKNOWNKEY"; break;
421 case RXKADEXPIRED : s = "RXKADEXPIRED"; break;
422 case RXKADSEALEDINCON : s = "RXKADSEALEDINCON"; break;
423 case RXKADDATALEN : s = "RXKADDATALEN"; break;
424 case RXKADILLEGALLEVEL : s = "RXKADILLEGALLEVEL"; break;
425 case VSALVAGE : s = "VSALVAGE"; break;
426 case VNOVNODE : s = "VNOVNODE"; break;
427 case VNOVOL : s = "VNOVOL"; break;
428 case VVOLEXISTS : s = "VVOLEXISTS"; break;
429 case VNOSERVICE : s = "VNOSERVICE"; break;
430 case VOFFLINE : s = "VOFFLINE"; break;
431 case VONLINE : s = "VONLINE"; break;
432 case VDISKFULL : s = "VDISKFULL"; break;
433 case VOVERQUOTA : s = "VOVERQUOTA"; break;
434 case VBUSY : s = "VBUSY"; break;
435 case VMOVED : s = "VMOVED"; break;
436 case VIO : s = "VIO"; break;
437 case VRESTRICTED : s = "VRESTRICTED"; break;
438 case VRESTARTING : s = "VRESTARTING"; break;
439 case VREADONLY : s = "VREADONLY"; break;
441 osi_Log2(afsd_logp, "cm_Analyze: ignoring error code 0x%x (%s)",
447 if (retry && dead_session)
451 /* drop this on the way out */
455 /* retry until we fail to find a connection */
459 long cm_ConnByMServers(cm_serverRef_t *serversp, cm_user_t *usersp,
460 cm_req_t *reqp, cm_conn_t **connpp)
463 cm_serverRef_t *tsrp;
466 int someBusy = 0, someOffline = 0, allBusy = 1, allDown = 1;
467 long timeUsed, timeLeft, hardTimeLeft;
475 timeUsed = (GetCurrentTime() - reqp->startTime) / 1000;
477 gettimeofday(&now, NULL);
478 timeUsed = sub_time(now, reqp->startTime) / 1000;
481 /* leave 5 seconds margin of safety */
482 timeLeft = ConnDeadtimeout - timeUsed - 5;
483 hardTimeLeft = HardDeadtimeout - timeUsed - 5;
485 lock_ObtainWrite(&cm_serverLock);
486 for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
488 cm_GetServerNoLock(tsp);
489 lock_ReleaseWrite(&cm_serverLock);
490 if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
491 if (tsrp->status == busy) {
494 } else if (tsrp->status == offline) {
499 code = cm_ConnByServer(tsp, usersp, connpp);
500 if (code == 0) { /* cm_CBS only returns 0 */
502 /* Set RPC timeout */
503 if (timeLeft > ConnDeadtimeout)
504 timeLeft = ConnDeadtimeout;
506 if (hardTimeLeft > HardDeadtimeout)
507 hardTimeLeft = HardDeadtimeout;
509 lock_ObtainMutex(&(*connpp)->mx);
510 rx_SetConnDeadTime((*connpp)->callp, timeLeft);
511 rx_SetConnHardDeadTime((*connpp)->callp, (u_short) hardTimeLeft);
512 lock_ReleaseMutex(&(*connpp)->mx);
516 /* therefore, this code is never executed */
521 lock_ObtainWrite(&cm_serverLock);
522 cm_PutServerNoLock(tsp);
525 lock_ReleaseWrite(&cm_serverLock);
526 if (firstError == 0) {
527 if (serversp == NULL)
528 firstError = CM_ERROR_NOSUCHVOLUME;
530 firstError = CM_ERROR_ALLOFFLINE;
532 firstError = CM_ERROR_ALLBUSY;
534 osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
535 firstError = CM_ERROR_TIMEDOUT;
539 osi_Log1(afsd_logp, "cm_ConnByMServers returning 0x%x", firstError);
543 /* called with a held server to GC all bad connections hanging off of the server */
544 void cm_GCConnections(cm_server_t *serverp)
550 lock_ObtainWrite(&cm_connLock);
551 lcpp = &serverp->connsp;
552 for (tcp = *lcpp; tcp; tcp = *lcpp) {
554 if (userp && tcp->refCount == 0 && (userp->vcRefs == 0)) {
555 /* do the deletion of this guy */
556 cm_PutServer(tcp->serverp);
557 cm_ReleaseUser(userp);
559 rx_DestroyConnection(tcp->callp);
560 lock_FinalizeMutex(&tcp->mx);
564 /* just advance to the next */
568 lock_ReleaseWrite(&cm_connLock);
571 static void cm_NewRXConnection(cm_conn_t *tcp, cm_ucell_t *ucellp,
572 cm_server_t *serverp)
577 struct rx_securityClass *secObjp;
579 if (serverp->type == CM_SERVER_VLDB) {
584 osi_assert(serverp->type == CM_SERVER_FILE);
588 if (ucellp->flags & CM_UCELLFLAG_RXKAD) {
591 tcp->cryptlevel = rxkad_crypt;
593 tcp->cryptlevel = rxkad_clear;
595 secObjp = rxkad_NewClientSecurityObject(tcp->cryptlevel,
596 &ucellp->sessionKey, ucellp->kvno,
597 ucellp->ticketLen, ucellp->ticketp);
601 tcp->cryptlevel = rxkad_clear;
602 secObjp = rxnull_NewClientSecurityObject();
604 osi_assert(secObjp != NULL);
605 tcp->callp = rx_NewConnection(serverp->addr.sin_addr.s_addr,
610 rx_SetConnDeadTime(tcp->callp, ConnDeadtimeout);
611 rx_SetConnHardDeadTime(tcp->callp, HardDeadtimeout);
612 tcp->ucgen = ucellp->gen;
614 rxs_Release(secObjp); /* Decrement the initial refCount */
617 long cm_ConnByServer(cm_server_t *serverp, cm_user_t *userp, cm_conn_t **connpp)
622 lock_ObtainMutex(&userp->mx);
623 lock_ObtainWrite(&cm_connLock);
624 for (tcp = serverp->connsp; tcp; tcp=tcp->nextp) {
625 if (tcp->userp == userp)
629 /* find ucell structure */
630 ucellp = cm_GetUCell(userp, serverp->cellp);
632 cm_GetServer(serverp);
633 tcp = malloc(sizeof(*tcp));
634 memset(tcp, 0, sizeof(*tcp));
635 tcp->nextp = serverp->connsp;
636 serverp->connsp = tcp;
639 lock_InitializeMutex(&tcp->mx, "cm_conn_t mutex");
640 lock_ObtainMutex(&tcp->mx);
641 tcp->serverp = serverp;
642 tcp->cryptlevel = rxkad_clear;
643 cm_NewRXConnection(tcp, ucellp, serverp);
645 lock_ReleaseMutex(&tcp->mx);
647 if ((tcp->ucgen < ucellp->gen) ||
648 (tcp->cryptlevel != (cryptall ? (ucellp->flags & CM_UCELLFLAG_RXKAD ? rxkad_crypt : rxkad_clear) : rxkad_clear)))
650 if (tcp->ucgen < ucellp->gen)
651 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to token update");
653 osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
654 lock_ObtainMutex(&tcp->mx);
655 rx_DestroyConnection(tcp->callp);
656 cm_NewRXConnection(tcp, ucellp, serverp);
657 lock_ReleaseMutex(&tcp->mx);
661 lock_ReleaseWrite(&cm_connLock);
662 lock_ReleaseMutex(&userp->mx);
664 /* return this pointer to our caller */
665 osi_Log1(afsd_logp, "cm_ConnByServer returning conn 0x%x", (long) tcp);
671 long cm_Conn(struct cm_fid *fidp, struct cm_user *userp, cm_req_t *reqp,
676 cm_serverRef_t **serverspp;
678 code = cm_GetServerList(fidp, userp, reqp, &serverspp);
684 code = cm_ConnByMServers(*serverspp, userp, reqp, connpp);
685 cm_FreeServerList(serverspp);
689 extern struct rx_connection *
690 cm_GetRxConn(cm_conn_t *connp)
692 struct rx_connection * rxconn;
693 lock_ObtainMutex(&connp->mx);
694 rxconn = connp->callp;
695 rx_GetConnection(rxconn);
696 lock_ReleaseMutex(&connp->mx);